HOBBES ON DRUGS

UNDERSTANDING DRUG VIOLENCE IN MEXICO

A Dissertation

Submitted to the Graduate School

of the University of Notre Dame
in Partial Fulfillment of the Requirements
for the Degree of

Doctor of Philosophy

by
Javier Osorio

Christian Davenport, Director

Graduate Program in Political Science

Notre Dame, Indiana
July 2013
 ProQuest Number: 3738644

All rights reserved

INFORMATION TO ALL USERS

The quality of this reproduction is dependent upon the quality of the copy submitted.

In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.

ProQuest 3738644

Published by ProQuest LLC (2015). Copyright of the Dissertation is held by the Author.

All rights reserved.

This work is protected against unauthorized copying under Title 17, United States Code
Microform Edition © ProQuest LLC.

ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
 
c Copyright by

Javier Osorio
2013
All Rights Reserved
 HOBBES ON DRUGS
UNDERSTANDING DRUG VIOLENCE IN MEXICO

Abstract
by
Javier Osorio

This dissertation analyzes the unprecedented eruption of organized criminal vi-

olence in Mexico. To understand the dynamics of drug violence, this dissertation
addresses three questions. What explains the onset of the war on drugs in Mexico?
Once the conflict starts, why does drug violence escalate so rapidly? And lastly, why
is there subnational variation in the concentration of violence?
Based on a game theoretic model, the central argument indicates that democrati-
zation erodes the peaceful configurations between the state and criminal organizations
and motivates authorities to fight crime, thus triggering a wave of violence between
the state and organized criminals and among rival criminal groups fighting to con-
trol strategic territories. In this account, state action is not neutral: law enforcement
against a criminal group generates the opportunity for a rival criminal organization to
invade its territory, thus leading to violent interactions among rival criminal groups.
These dynamics of violence tend to concentrate in territories favorable for the re-
ception, production and distribution of drugs. In this way, the disrupting effect of
law enforcement unleashes a massive wave of violence of all-against-all resembling a
Hobbesian state of war.
To test the observable implications of the theory, the empirical assessment relies
on a novel database of geo-referenced daily event data at municipal level providing
 Javier Osorio

detailed information on who did what to whom, when and where in the Mexican war
on drugs. This database covers all municipalities of the country between 2000 and
2010, thus comprising about 9.8 million observations. The creation of this fine-grained
database required the development of Eventus ID, a novel software for automated
coding of event data from text in Spanish. The statistical assessment relies on quasi-
experimental identification strategies and time-series analysis to overcome problems
of causal inference associated with analyzing the distinct - yet overlapping - processes
of violence between government authorities and organized criminals and among rival
criminal groups. In addition, the statistical analysis is complemented with insights
from fieldwork and historical process tracing. Results provide strong support for the
empirical implications derived from the theoretical model.
 A mamá y papá,
quienes me enseñaron el valor del trabajo
y la responsabilidad de no ser indiferente.

ii
 CONTENTS

FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii

CHAPTER 1: INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 The Main Argument . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 The Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.2 Methodological Strategy . . . . . . . . . . . . . . . . . . . . . 8
1.1.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Redefining the Role of Organized Criminal Violence in Conflict Research 13
1.3 Key Conceptual Definitions . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 The Road Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

CHAPTER 2: AN INTEGRATIVE THEORY OF ORGANIZED CRIMINAL

VIOLENCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 A Fragmented Set of Alternative Explanations . . . . . . . . . . . . . 33
2.3 An Integrative Theory of Organized Crime Violence . . . . . . . . . . 50
2.3.1 Players and Actions . . . . . . . . . . . . . . . . . . . . . . . 51
2.3.2 Fighting Crime in New Democracies . . . . . . . . . . . . . . 55
2.3.3 Shifting Military Capabilities . . . . . . . . . . . . . . . . . . 58
2.3.4 Payoffs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.3.5 Equilibrium Conditions . . . . . . . . . . . . . . . . . . . . . . 70
2.3.5.1 Conditions for the Target to Fight the Challenger . . 70
2.3.5.2 Conditions for the Challenger to Invade the Target . 73
2.3.5.3 Conditions for the Target to Retaliate Against the State 74
2.3.5.4 Conditions for the State to Enforce the Law . . . . . 75
2.3.5.5 Summary of Equilibrium Conditions . . . . . . . . . 76
2.4 Empirical Implications . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.4.1 War of All Against All . . . . . . . . . . . . . . . . . . . . . . 81
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

iii
CHAPTER 3: AUTOMATED CODING OF EVENT DATA USING EVEN-
TUS ID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.2 Historical and Technological Developments of Automated Coding . . 91
3.2.1 A Brief History of Event Data . . . . . . . . . . . . . . . . . . 91
3.2.2 Trade-offs Between Manual and Machine Generated Event Data 97
3.3 Coding Event Data Using Eventus ID . . . . . . . . . . . . . . . . . . 104
3.3.1 Eventus ID coding process . . . . . . . . . . . . . . . . . . . . 106
3.4 Stage 1. Input Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.4.1 Step 1.a. Information Sources, Coverage and Selection Criteria 107
3.4.1.1 Manual gathering and selection criteria . . . . . . . . 111
3.4.2 Step 1.b. Collecting Individual Input Files . . . . . . . . . . . 115
3.4.3 Step 1.c. Eventus ID Input Format . . . . . . . . . . . . . . . 116
3.4.4 Step 1.d. Corpus of Text . . . . . . . . . . . . . . . . . . . . . 120
3.5 Stage 2. Event Coding . . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.5.1 Step 2.a. Actor Dictionary . . . . . . . . . . . . . . . . . . . . 121
3.5.2 Step 2.b. Verb Dictionary . . . . . . . . . . . . . . . . . . . . 123
3.5.3 Step 2.c. Event Coding Using Eventus ID . . . . . . . . . . . 127
3.5.3.1 General sequence algorithm . . . . . . . . . . . . . . 129
3.5.3.2 Partial sequence algorithm . . . . . . . . . . . . . . . 131
3.5.4 Step 2.d. Event Database . . . . . . . . . . . . . . . . . . . . 133
3.6 Stage 3. Event Location . . . . . . . . . . . . . . . . . . . . . . . . . 134
3.6.1 Step 3.a. Location Dictionaries . . . . . . . . . . . . . . . . . 135
3.6.2 Step 3.b. Event Location Using Eventus ID . . . . . . . . . . 136
3.6.3 Step 3.c. Georeferenced Event Data . . . . . . . . . . . . . . . 139
3.7 Stage 4. Validation and Recoding . . . . . . . . . . . . . . . . . . . . 139
3.7.1 Step 4.a. Validation . . . . . . . . . . . . . . . . . . . . . . . 141
3.7.2 Step 4.b. Recoding . . . . . . . . . . . . . . . . . . . . . . . . 142
3.7.3 Duplicates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
3.8 Stage 5. Validated Event Database . . . . . . . . . . . . . . . . . . . 153
3.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

CHAPTER 4: MAIN TRENDS OF ORGANIZED CRIMINAL VIOLENCE . 156

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.2 Limitations of Body Counts . . . . . . . . . . . . . . . . . . . . . . . 158
4.3 Main Trends of Organized Criminal Violence . . . . . . . . . . . . . . 167
4.3.1 General Description of Event Data . . . . . . . . . . . . . . . 167
4.3.2 Temporal Trends of Organized Criminal Violence . . . . . . . 173
4.3.3 Disaggregating the Trends of Violence at State Level . . . . . 179
4.3.4 Main Trends of Violence at Municipal Level on a Daily Basis . 186
4.3.5 Spatial Trends of Violence Between Criminal Organizations . . 187

iv
CHAPTER 5: THE EMERGENCE, CONSOLIDATION AND COLLAPSE
OF ORDER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
5.2 Explicit Assumptions about Democratization and Law Enforcement . 195
5.3 The Emergence of Order out of the Mexican Revolution . . . . . . . . 198
5.3.1 Early Days of Drug Traffic in Mexico . . . . . . . . . . . . . . 200
5.3.2 The Post Revolutionary Era . . . . . . . . . . . . . . . . . . . 205
5.4 The Consolidation of Order Under Authoritarianism . . . . . . . . . . 210
5.4.1 Cold War Politics and its Repressive Instrumentation . . . . . 212
5.4.2 Political Threats and Authoritarian Reaction . . . . . . . . . 217
5.4.3 Counter Narcotic Efforts During the Dirty War . . . . . . . . 221
5.4.4 Maintaining Peace in Drug Markets . . . . . . . . . . . . . . . 225
5.5 The Erosion of the Preexisting Order in a Democratic Context . . . . 232
5.5.1 Increasing Strength of Drug Trafficking Organizations . . . . . 233
5.5.2 Dismantling the Political Security Apparatus . . . . . . . . . . 240
5.5.3 The Process of Democratization in Mexico . . . . . . . . . . . 244
5.5.4 Not All Good Things Come Together . . . . . . . . . . . . . . 258
5.5.5 The Collapse of Order . . . . . . . . . . . . . . . . . . . . . . 267

CHAPTER 6: UNDERSTANDING DRUG VIOLENCE IN MEXICO . . . . . 281

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
6.2 Structural Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
6.2.1 Hypotheses and Operationalization . . . . . . . . . . . . . . . 283
6.2.2 Model Specification . . . . . . . . . . . . . . . . . . . . . . . . 296
6.2.3 Statistical Results of the Structural Model . . . . . . . . . . . 297
6.2.4 Overall Model Fit . . . . . . . . . . . . . . . . . . . . . . . . . 307
6.3 Interactive Model of Drug Violence . . . . . . . . . . . . . . . . . . . 308
6.3.1 Hypotheses to Evaluate . . . . . . . . . . . . . . . . . . . . . 310
6.3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
6.3.3 Model Specification . . . . . . . . . . . . . . . . . . . . . . . . 315
6.3.4 Effect of Violent Enforcement on Violence Among DTOs . . . 320
6.3.5 Effect of Violent and Non-Violent Enforcement on Violence
Among DTOs . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
6.3.6 Overall Model Fit . . . . . . . . . . . . . . . . . . . . . . . . . 342
6.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344

CHAPTER 7: DYNAMICS OF VIOLENCE IN THE MEXICAN WAR ON

DRUGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
7.2 Hypotheses and Research Design . . . . . . . . . . . . . . . . . . . . 351
7.3 Time Series of Organized Criminal Violence . . . . . . . . . . . . . . 353
7.3.1 Time Series Identification . . . . . . . . . . . . . . . . . . . . 353
7.3.2 Time Series Estimation and Diagnosis . . . . . . . . . . . . . 357

v
 7.3.3 Model Specification of Time Series Processes . . . . . . . . . . 360
7.4 Dynamic Analysis of Drug-Related Violence . . . . . . . . . . . . . . 363
7.4.1 Base-Line Vector Autoregressive Model for the Dynamics of
Drug Violence . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
7.4.2 Full Specification of the Vector Autoregressive Model for the
Dynamic Analysis of Drug Violence . . . . . . . . . . . . . . . 376
7.5 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 390

CHAPTER 8: CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . 399

8.1 Why do Politicians Decide to Fight Criminal Organizations After Hav-
ing Peacefully Coexisted With Them? . . . . . . . . . . . . . . . . . . 400
8.2
Why
Does
Drug-Related
Violence
Escalate
so
Rapidly?


.
.
.
.
.
.
.
406
8.3 Why is Violence More Concentrated in Some Areas Than in Others? 412
8.4 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
8.5 Looking Ahead . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417

APPENDIX
A


.










.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
420
A.1 Infolatina Query . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
A.2 Number of Information Sources by Year . . . . . . . . . . . . . . . . 422
A.3 List of Information Sources . . . . . . . . . . . . . . . . . . . . . . . . 423
A.4 List of Mexican States . . . . . . . . . . . . . . . . . . . . . . . . . . 427
A.5 Map of Mexican States . . . . . . . . . . . . . . . . . . . . . . . . . . 428
A.6 Descriptive Statistics of Data at Municipal Level on a Daily Basis . . 429
A.7 Descriptive Statistics of Data at the National Level on a Daily Basis . 431
A.8 Hospitalizations by Drug Intoxication . . . . . . . . . . . . . . . . . . 432

BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433

vi
 FIGURES

1.1 Types of organized violence . . . . . . . . . . . . . . . . . . . . . . . 15

2.1 The Game of Drug Violence . . . . . . . . . . . . . . . . . . . . . . . 52

2.2 Shifting military capabilities . . . . . . . . . . . . . . . . . . . . . . . 62
1−σ
2.3 Military damage and recovery capabilities in γ . . . . . . . . . . . 63
2−2σ
2.4 Military damage and recovery capabilities in γ . . . . . . . . . . . 65
2.5 Military damage and recovery capabilities in γ 1−σ and γ 2−2σ . . . . . 66
2.6 Military capabilities recovered or gained through fighting . . . . . . . 71
2.7 War of all against all . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.1 Continuum of Textual Annotation Methods . . . . . . . . . . . . . . 98

3.2 Eventus ID process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

4.1 Frequency of events associated with drug-related violence by type of

event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
4.2 Frequency plots of the main variables of drug-related violence . . . . 173
4.3 Main trends of drug-related violence in Mexico . . . . . . . . . . . . . 175
4.4 Violent and non-violent law enforcement . . . . . . . . . . . . . . . . 176
4.5 Violent competition among criminal organizations at state level on a
monthly basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.6 Criminal retaliation against government authorities at state level on a
monthly basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
4.7 Violent law enforcement at state level on a monthly basis . . . . . . . 185
4.8 Main trends of violence at municipal level on a daily basis . . . . . . 186
4.9 Spatial trends of violence among criminal organizations . . . . . . . . 189

5.1 Number of seats in Congress by type of electoral rule . . . . . . . . . 247

5.2 Percentage of votes for Congressional elections . . . . . . . . . . . . . 248
5.3 Percentage of votes for Presidential elections . . . . . . . . . . . . . . 251
5.4 Effective number of parties (ENP) at the federal level . . . . . . . . . 257
5.5 Effective number of parties (ENP) at the local level . . . . . . . . . . 258

vii
5.6 Annual number of homicides in Mexico 1938-2011 . . . . . . . . . . . 266
5.7 Annual budget for security agencies in Mexico 2000-2010 . . . . . . . 275

6.1 Drug markets in Mexico: hospital discharges diagnosed with intoxica-

tion by narcotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
6.2 Drug valuable territories . . . . . . . . . . . . . . . . . . . . . . . . . 293
6.3 Annual production of assault weapons in the U.S. . . . . . . . . . . . 295
6.4 National average of unemployment rate on a quarterly basis . . . . . 296
6.5 Price per pure gram of cocaine in the U.S. . . . . . . . . . . . . . . . 304
6.6 Goodness of fit of the structural model . . . . . . . . . . . . . . . . . 309
6.7 Effect of political competition on violent enforcement . . . . . . . . . 323
6.8 Effect of divided government on violent enforcement . . . . . . . . . . 324
6.9 Effect of political strain on violent enforcement . . . . . . . . . . . . 325
6.10 Effect of violent enforcement on violent competition among DTOs . . 328
6.11 Effect of violent law enforcement on different territories . . . . . . . . 330
6.12 Effect of political competition on violent and non-violent enforcement 335
6.13 Effect of divided government on violent and non-violent enforcement . 336
6.14 Effect of political strain on violent and non-violent enforcement . . . 337
6.15 Effect of violent and non-violent enforcement on violent competition
among DTOs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
6.16 Goodness of fit of the interactive model . . . . . . . . . . . . . . . . . 343
6.17 Comparative model fit between structural and dynamic models . . . . 344

7.1 Daily time series of violent competition among DTOs at the national
level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
7.2 Autocorrelation (AC) and Partial Autocorrelation (PAC) of violent
competition among DTOs . . . . . . . . . . . . . . . . . . . . . . . . 356
7.3 Autocorrelation (AC) and Partial Autocorrelation (PAC) functions of
residuals from the ARIMA AR(37) model . . . . . . . . . . . . . . . 359
7.4 General intuition of Vector Autoregressive (VAR) models . . . . . . . 365
7.5 Orthogonal Impulse Response Functions (OIRF) of violence in the
Mexican war on drugs . . . . . . . . . . . . . . . . . . . . . . . . . . 371
7.6 Cumulative Orthogonal Impulse Response Functions (COIRF) of vio-
lence in the Mexican war on drugs . . . . . . . . . . . . . . . . . . . . 375
7.7 Cumulative Orthogonal Impulse Response Functions (COIRF) for the
effect of violent and non-violent enforcement on competition among
DTOs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383

viii
7.8 Cumulative Orthogonal Impulse Response Functions (COIRF) for the
effect of violent and non-violent enforcement on criminal retaliation . 385
7.9 Cumulative Orthogonal Impulse Response Functions (COIRF) for the
effect of competition among DTOs on violent and non-violent enforce-
ment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
7.10 Cumulative Orthogonal Impulse Response Functions (COIRF) for the
effect of criminal retaliation on violent and non-violent enforcement . 388
7.11 Cumulative Orthogonal Impulse Response Functions (COIRF) of sus-
tained campaigns of violence during six months . . . . . . . . . . . . 389

A.1 Number of information sources by year . . . . . . . . . . . . . . . . . 422

A.2 Map of Mexican states . . . . . . . . . . . . . . . . . . . . . . . . . . 428
A.3 Hospitalizations by drug intoxication at state level . . . . . . . . . . . 432

ix
 TABLES

2.1 ALTERNATIVE EXPLANATIONS . . . . . . . . . . . . . . . . . . . 34

2.2 SEVERITY OF MILITARY DAMAGE AND RECOVERY CAPA-
BILITY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.3 EQUILIBRIUM CONDITIONS FOR THE USE OF VIOLENCE . . 77
2.4 SUMMARY OF EMPIRICAL IMPLICATIONS . . . . . . . . . . . . 82

3.1 COMPARATIVE ADVANTAGES OF MANUAL AND SUPERVISED

ANNOTATION METHODS . . . . . . . . . . . . . . . . . . . . . . . 101
3.2 NUMBER OF INFORMATION SOURCES . . . . . . . . . . . . . . 109
3.3 AVAILABILITY OF INFORMATION SOURCES . . . . . . . . . . . 110
3.4 NUMBER OF NEWS REPORTS . . . . . . . . . . . . . . . . . . . . 115
3.5 CORPUS OF TEXT FILES BY INFORMATION SOURCE . . . . . 120
3.6 ACTORS CATEGORIES . . . . . . . . . . . . . . . . . . . . . . . . 122
3.7 VERB CATEGORIES . . . . . . . . . . . . . . . . . . . . . . . . . . 123
3.8 VERB TENSES IN ENGLISH AND SPANISH . . . . . . . . . . . . 126
3.9 EVENTUS ID CODING ALGORITHMS . . . . . . . . . . . . . . . . 128
3.10 EVENTUS ID LOCATION ALGORITHM . . . . . . . . . . . . . . . 138
3.11 DIRECT CRITERIA FOR DEFAULT ACTOR ASSIGNMENT . . . 146
3.12 CONDITIONAL CRITERIA FOR DEFAULT ACTOR ASSIGNMENT148

4.1 DATABASES OF DRUG-RELATED VIOLENCE IN MEXICO . . . 161

4.2 DESCRIPTIVE STATISTICS OF THE MAIN VARIABLES OF OR-
GANIZED CRIMINAL VIOLENCE AT THE MUNICIPAL LEVEL
ON A DAILY BASIS . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

5.1 MILITARY COUNTER-NARCOTIC OPERATIONS CONDUCTED

BY THE MEXICAN ARMY BETWEEN DECEMBER 2006 AND
DECEMBER 2010 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274

6.1 STRUCTURAL DETERMINANTS OF VIOLENT COMPETITION

AMONG DTOs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300

x
6.2 FIRST STAGE: BASE-LINE AND FULL MODEL OF VIOLENT
ENFORCEMENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
6.3 SECOND STAGE: BASE-LINE AND FULL MODEL OF VIOLENT
COMPETITION AMONG DOTs . . . . . . . . . . . . . . . . . . . . 326
6.4 FIRST STAGE: POLITICAL DETERMINANTS OF VIOLENT AND
NON-VIOLENT ENFORCEMENT . . . . . . . . . . . . . . . . . . . 333
6.5 SECOND STAGE: VIOLENT COMPETITION AMONG DTOs . . . 338

7.1 AUGMENTED DICKEY-FULLER TEST FOR DIFFERENT PRO-

CESSES OF VIOLENCE . . . . . . . . . . . . . . . . . . . . . . . . 357
7.2 PORTMANTEAU TEST FOR WHITE NOISE IN DIFFERENT PRO-
CESSES OF VIOLENCE . . . . . . . . . . . . . . . . . . . . . . . . 360
7.3 EXPLANATORY POWER OF VAR MODEL 1 . . . . . . . . . . . . 367
7.4 GRANGER CAUSALITY TEST FOR VAR MODEL 1 . . . . . . . . 369
7.5 EXPLANATORY POWER FOR VAR MODEL 2 . . . . . . . . . . . 380
7.6 GRANGER CAUSALITY TEST FOR VAR MODEL 2 . . . . . . . . 382

A.1 LIST OF INFORMATION SOURCES . . . . . . . . . . . . . . . . . 423

A.2 LIST OF MEXICAN STATES . . . . . . . . . . . . . . . . . . . . . . 427
A.3 DESCRIPTIVE STATISTICS OF DATA AT MUNICIPAL LEVEL
ON A DAILY BASIS . . . . . . . . . . . . . . . . . . . . . . . . . . 429
A.4 DESCRIPTIVE STATISTICS OF DATA AT NATIONAL LEVEL ON
A DAILY BASIS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431

xi
 ACKNOWLEDGMENTS

I am deeply indebted to my advisor Christian Davenport for his crucial role in my

doctoral research. Christian, who has long been an inspiring figure for me, provided
priceless expertise, guidance and assistance for transitioning from a graduate student
into a researcher. His example and mentorship was fundamental for embarking in
this ambitious project. Christian gave me the freedom that I needed to conduct my
research, yet always gave me his sharp feedback, strategic advice and encouragement
to confront the multiple challenges that emerged during the course of this research.
I cannot thank him enough.
I am also very grateful to Michael Coppedge for the substantial influence he had
on my graduate education. Michael always encouraged me to think creatively and
rigorously on big substantive questions as well as on different ways of measuring tough
concepts. I am also very thankful to Monika Nalepa for her support as a student and
her mentoring in the theoretical development of my dissertation.
I express my gratitude to Naunihal Singh for his support and input as a graduate
student and in my dissertation research. I also owe a great debt of gratitude to
David Nickerson for the opportunity to participate in the “vote buying” project and
the fantastic learning associated with it. During my time at the University of Notre
Dame, I was incredibly fortunate of having other fantastic professors who contributed
to my graduate education: Guillermo O’Donnell, Scott Mainwaring, Fran Hagopian,
Timothy Scully, Debra Javeline, Christina Wolbrecht and Darren Davis.

xii
 This research would not have been possible without the generous financial sup-
port, opportunities, space and time provided by several institutions during the dif-
ferent stages of this project. To conduct my dissertation research, I received support
from the Doctoral Dissertation Research Improvement Grant of the National Sci-
ence Foundation; the Program on Order, Conflict and Violence at Yale University;
the Dissertation Fellowship of the Harry Frank Guggenheim Foundation; the Drugs,
Security and Democracy Fellowship of the Social Science Research Council and the
Open Society Foundations; the Jennings Randolph Peace Scholarship Dissertation
Program of United States Institute of Peace; and the Graduate Research Grant of
the Kellogg Institute for International Studies at the University of Notre Dame. At
the heart of these prestigious institutions I found fantastic individuals who gave me
their support and immensely valuable opportunities. In am particularly thankful to
Stathis Kalyvas, Joel Wallman, David Holiday, Elizabeth Cole, Scott Mainwaring,
Denise Wright and Sharon Schierling.
Special thanks to Alejandro Reyes for all his work and creativity in developing
Eventus ID. I thank him for introducing me to the fantastic world of natural language
processing. I found in him a super smart researcher, a great co-author and, most
importantly, a fantastic friend. Without his knowledge and experience on computer
programing, Eventus ID would have never said “Hello World!”
My gratitude also goes on the excellent team of research assistants who spent
countless hours scraping news reports, gathering data and contributing to different
pieces of this overambitious research project: Fátima Ávila, Eréndira González, Eli
Granados, Citlali Hernández, Kelly Lima, Nayelli Lima, Mauricio Ochoa, Frank Orta,
Manuel Rı́os, Lucı́a Rodrı́guez and Melisa Pineda. I also want to thank Margaret
Schroeder for her help in editing my dissertation.
I want to thank all the people who devoted significant amounts of time and at-
tention to share with me their knowledge, opinion, experience, frustration, suffering

xiii
and endurance during the several interviews conducted as part of this research. Your
testimonies did not only provide key insights for my research but, most importantly,
they served as a constant reminder of the humanitarian drama associated with the
wave of violence ravaging Mexico. As promised, your names are preserved in confi-
dentiality. You have my deepest gratitude, respect and admiration.
I owe an immense debt of gratitude to my “family” at Notre Dame: Esteban
Manteca, Chad Kiewiet de Jonge, Ezequiel González, Carlos Meléndez, Craig Garcı́a,
Fernando Galáz, Daniel Cibotaru, Nara Pavao, Laura Gamboa, Lucı́a Tiscornia, Keti
Nozadze, Sofı́a Galván, Anne Baker, Mark Driessen, Vı́ctor Huerta and Rodrigo
Castro. Thanks to Sandra Botero for helping me to learn from my academic and
personal shortcomings. You guys were an endless source of fun, love, support, caring
and endurance. Thanks a lot for making graduate school such an enriching part of
my life. I also want to thank Naomi Echeandı́a who taught me how to find magic in
every detail. In Mexico City, I am particularly thankful to Iliana Haro, who opened
her house and soul to me. Roomy, thanks a lot for all your support, advice and
complicity. I am also thankful to Leobardo Morales for always keeping an eye on
me regardless the time and distance. I am also grateful for finding great friends and
fantastic scholars among the DSD fellows: Angélica Durán-Martı́nez, Susan Brewer,
Eduardo Mocada, Reynaldo Rojo and Yanilda González. I would also like to thank
my brothers in arms, Luca Falciola and Henning Tamm, for all the learning and great
times we had at Yale. I am deeply grateful to Ángeles Villanueva, whose love, support,
caring, patience and understanding helped me to find the long needed balance.
At least, but not at last, I am extremely thankful to my family for their infinite
love, tireless support, priceless example and innumerable sacrifices. You taught me
to work hard, live with passion, be responsible and follow my dreams. This is what
I do, this is what I am.

xiv
 CHAPTER
1

INTRODUCTION

By the early years of the twenty-first century, the pattern of political violence
that had marked a lengthy part of Mexican history in the previous century had dis-
sipated. The long revolutionary war was settled in the 1920s, the repression of social
movements and guerrillas in the 1960s and 1970s had calmed and, in the early 1990s,
the resonating words of the Zapatista movement shook the country more than its
rifle-shaped sticks. The old México bronco (untamed Mexico) seemed a figure of the
past. However, since the turn of the century, Mexico has begun to experience a
massive escalation of violence of unprecedented characteristics. Instead of agrarian
revolutionaries and politically motivated protesters or insurgents, the central actors
of this wave of violence are ruthless criminals fighting for drug-related turf. Reversing
a downward trend of homicides, the dramatic eruption of violence associated with
organized crime generated more than 50,000 casualties between 2006 and 2011 (Sis-
tema Nacional de Seguridad Pública, 2011a). The lethality of this surge of criminal
violence is comparable to the onset of 50 civil wars in only six years (Osorio, 2012).
The drama not only relates the increasing number of homicides, but also on their
nature. Numerous times, convoys of heavily armed sicarios (hitmen) carrying high-

1
caliber weapons battled members of rival criminal groups or government authorities
in the midst of crowded public areas. Bodies were dumped on the streets showing
signs of horrific brutality that often included torture and mutilation. Authorities
began discovering mass graves, and criminals arranged the bodies of their victims in
public places in lurid tableaus: decapitated bodies appeared hanging from bridges
and severed heads were left at the doors of government buildings, tossed into bars, or
placed openly in public squares. What caused this unprecedented wave of violence?
The recent escalation of organized criminal violence in Mexico is puzzling for
several reasons. Criminal organizations in Mexico have existed since at since the
late nineteenth century (Astorga, 2005). For several decades they conducted their
activities of drug cultivation and trafficking into the U.S. without relying primarily
on the use of violence. During this period, criminals enjoyed protection – or at least
lack of law enforcement –f rom corrupt government authorities. Of course there
were some isolated and incidental events of violence. However, this illegal sector
was largely characterized by a long absence of overt and sustained hostilities. The
relations between criminals and the state as well as among criminal groups were
characterized by order and peace. Then, why did criminal organizations suddenly
become so violent? Moreover, why were they not violent before and what explains
their current aggressiveness? In addition, why and how did state officials coexist
peacefully with criminal groups for such a long time? Why are government authorities
now waging battles against criminal organizations?
Paradoxically, the eruption of organized criminal violence is occurring at a time
when the Mexican political system is consolidating its recently-won democracy. To-
wards the end of the twentieth century, Mexico underwent a gradual process of de-
mocratization that generated a competitive party system, effective division of powers,
and several other characteristics associated with a well functioning – although not
fully consolidated – democracy (Coppedge, 2012; Held, 2006). Based on the theo-

2
retical foundations of this domestic democratic peace, we could expect democratic
regimes to be associated with low levels of intra-state violence (Davenport, 2009b).
Then why has such a dramatic outbreak of violence occurred in a recent, yet solid
democratic system?
To address the puzzling escalation of organized criminal violence, this research
focuses on analyzing three different aspects of the ongoing wave organized criminal
violence in Mexico. First, this study examines the onset of violence by asking why
politicians have decided to fight criminal organizations after having peacefully coex-
isted with them for several decades? Second, the research aims to understand the
temporal dynamics of violence by asking why drug-related violence has escalated so
rapidly? The third aspect of interest refers to the spatial variation of violence, asking
why violence is concentrated more in some areas than in others? These three re-
search questions guide a comprehensive inquiry which seeks to understand the onset,
escalation, and geographic concentration of organized criminal violence.

1.1
The
Main
Argument

At its core, the endurance of a political community depends on the ability of

institutions and mechanisms to maintain order by coping with internal tensions and
adapting to external pressures. As indicated by Huntington (1968), the relative
power of the various elements of a community tends to change over time, and in
order to maintain its cohesiveness, power must be exercised and channeled through
institutions so as to make force compatible with the elements of the community. The
more complex and heterogeneous the community, the more it will require an effective
system for maintaining order. Since Hobbes (1651), the central role of the state
is to impose order. The survival of the state as a cohesive community depends on

3
the ability of its institutions to impose and maintain order among its constituent
elements.
Historically, the process of state formation has been inherently connected to vi-
olence as states try to draw to themselves the means of coercion until they prevail
in holding the legitimate monopoly of the use of force (Tilly, 1985, 1992; Weber,
1978). However, not all states succeed in securing the monopoly on violence by neu-
tralizing all other actors with the capability and willingness to conduct organized
violence. Weak states usually have limited presence and domain over the entire ter-
ritory contained within their borders, and often lack absolute control of the means of
coercion (Bates, Greif and Singh, 2002; Bates, 2008; O’Donnell, 1993; Scott, 2009).
In these contexts, state authorities usually coexist with a myriad of non-state actors
actively exercising active on parts of the territory and capable of resorting to vio-
lence to impose, maintain, and defend their position of power. The concurrence of
government authorities and non-state actors makes weak states particularly prone
to violence. This propensity for conflict increases as the relative power of non-state
actors increases with respect to the government authorities and as the state lacks the
institutions or the means to temper their threat.
However, the failure to monopolize coercive means in the hands of the state and
the coexistence of government authorities with other non-state actors capable of
conducting violence does not necessarily mean the absence of order and the presence
of violence. Peace and order can exist between government authorities and non-
state actors based on a system of institutions and mechanisms capable of forcing the
compatibility of their interests in a regularized, stable, sustained manner. In this
context, order emerges as the behavioral manifestation of the mutual interests of
the members of the community in not using their means of violence on one another.
The stability of order and the prospects of peace depend on the strength and scope

4
of political institutions capable of implementing an effective system of incentives,
procedures and organizational devices for resolving tensions within the community.
As illustrated by the Mexican case, government authorities at the federal and
local levels have coexisted peacefully with criminal organizations for decades. Order
was the defining characteristic of the relations between the state and criminals as well
as among different criminal organizations operating within the country. Peaceful co-
existence was possible thanks to the enduring political hegemony of the Institutional
Revolutionary Party (PRI) that held power for more than 70 years without interrup-
tion. During this period, government authorities maintained order partly because of
a combination of corruption and selective enforcement, but most importantly thanks
to a system of political incentives that instilled discipline throughout the government
structure without the need to rely on violence. Why did this order fall apart? Why
did violence emerge?
The central argument of this research is that democratization erodes the peaceful
configurations between the state and criminal organizations and motivates politicians
to fight crime, thus triggering a wave of violence between the state and organized crim-
inals and among rival criminal groups fighting to control valuable territories. This
explanation is rooted in the Hobbesian tradition of conflict and argues that violence
emerges as the consequence of the collapse of order. The gradual process of democ-
ratization undermines the strength and scope of authoritarian institutions, relations
and mechanisms that induce order between the state and criminal groups as well
as among criminal organization through a system of political incentives. Increasing
political competition in a democratic regime motivates government authorities to en-
force the law against criminals, thus setting in motion a chain of actions and reactions
that generate an aggregated escalation of violence. In the absence of authoritarian
institutions capable of regulating tensions through political means, the use of law
enforcement to fight crime motivates a violent reaction from criminal organizations

5
against the state. In addition, law enforcement against a criminal group generates
the opportunity for a rival criminal organization to invade its territory, thus leading
to violent interactions between rival criminal groups. These different, but overlap-
ping dynamics of violence tend to concentrate in strategic territories favorable for
conducting criminal activities. In short, in the absence of regulation mechanisms,
the disrupting effect of law enforcement unleashes a massive wave of violence of all-
against-all resembling a Hobbesian state of war.
In this account, violence is endogenous. Considered in general terms, violence is
the cause as well as the consequence. However, in order to explain the “bulk” of vio-
lence it is necessary to understand its external factors as well as its internal dynamics.
As indicated by Kalyvas (2006), only by disentangling the different components of
violence it is possible to appreciate its transformative force. In this respect, state
violence motivates criminals to conduct violent retaliation against authorities and
generates opportunities for criminal organizations to engage in violent competition
against their rivals. This research is thus located within the research agenda study-
ing the microdynamics of conflict. Studying violence from this perspective requires
a considerable effort of theoretical development capable of identifying the various
components of violence as well as defining its causal mechanisms. In addition, the
dynamic and interactive characteristics of conflict represent a challenge for empirical
identification. This demands the use of fine-grained data capable of revealing dis-
tinct behavioral manifestations of violence and the design of adequate identification
strategies to test the theoretical implications against the empirical evidence.

1.1.1
The
Mechanisms

The theoretical explanation relies on the analytic leverage of a formal model to

elucidate the different dynamics operating behind the onset, escalation and the geo-
graphic concentration of violence. The model departs from a set of explicit assump-

6
tions and propositions about the different actors operating within structural factors.
In addition, the analytical framework integrates these elements into a coherent set of
systematic mechanisms that allow understanding the strategic interactions between
the state and organized criminals, as well as among rival criminal groups.
To explain the onset of the war on drugs, the theoretical model claims that de-
mocratization disrupts the peaceful configurations that enable coexistence between
corrupt government authorities and criminal organizations in contexts of authori-
tarian rule. By increasing the number of relevant political actors at different levels
of government and by favoring the effective circulation of political elites, democra-
tization erodes the political structures that enable the existence of non-aggressive
agreements between authorities and criminals. Increased democratization thus alters
the system of political incentives for enforcing the law, motivating government au-
thorities to fight crime. In addition, the model argues that government authorities
obtain political benefits from implementing harsh security policies when their legiti-
macy is threatened by periods of political strain, thus reinforcing their incentives to
fight crime.
To explain the escalation of violence, the theory relies on a contest success model
for territorial competition. According to the argument, increased levels of law en-
forcement are likely to trigger an escalation of conflict between the state and criminal
organizations, and violence among rival criminal groups. In this way, the action of
the state is non-neutral and has a highly disruptive effect on the relative military bal-
ance of criminal organizations. Law enforcement weakens the capability of a criminal
group to protect its territory, thus motivating an invasion from a competing criminal
group that now faces a weaker rival. The equilibrium conditions of the model indi-
cate that violence committed by criminal organizations–either against the state or
their rivals–is a function of the severity of military damage inflicted on them, their
capability of recovering from it, and the value of the territory. This implies that

7
organized criminals are likely to use violence if the net military strength recovered
through fighting back after being attacked is larger than the cost of fighting, given
the value of a territory.
Finally, to explain the geographic distribution and concentration of violence, the
model explicitly incorporates the importance of territorial value as a key determinant
of conflict. Departing from the assumption that criminals are primarily motivated by
economic gains, the model indicates that criminal organizations are willing to engage
in violent confrontations to capture or defend strategic territory that give them access
to profitable illicit activities. In consequence, violence tends to concentrate around
valuable territories.

1.1.2
Methodological
Strategy

The central object of analysis in this research is the variation in levels of violent
competition among rival criminal organizations across time and space. Based on
the theoretical framework, an understanding of violence among criminal groups re-
quires analyzing this phenomenon in relation to violence conducted by the state and
violence perpetrated by criminals against government authorities. In consequence,
when analyzing violent competition among criminal organizations the related pro-
cesses of violence between the state and criminals are also relevant. Analyzing the
temporal and spatial variation of violence requires an eclectic methodology. This re-
search primarily relies on a quantitative analysis of the external and internal factors
determining different manifestations of violent behavior. The empirical basis of the
analysis is a large database of events of drug-related violence covering all municipal-
ities in Mexico on a daily basis from 2000 to 2010. Chapters 3 and 4 present the
methodology used to build this database and describe the main characteristics of the
data. Chapters 6 and 7 present the result of the quantitative analysis.

8
 Besides the quantitative analysis that constitutes the core of this dissertation,
the research also benefited enormously from a long stay and several trips to Mexico
where I conducted fieldwork for a total of 14 months. The witness accounts and in-
sights gathered during semi-structured interviews conducted in Mexico City, Juárez,
Culiacán and Tijuana provided crucial contributions to understanding the causes
and processes of violence. Following the methodological recommendations for nested
analysis of Lieberman (2005), the combination of quantitative examination of the
data and qualitative assessment of interviews and documents constituted an interac-
tive analytical process. Based on the implications of an early theoretical conception
of the work, I conducted some preliminary large-N analyses; these initial results were
then used to select locations for fieldwork and to inform the structure of the inter-
views. Qualitative information gathered during fieldwork contributed to redefining
the theoretical explanation and to providing more insightful interpretations of the
statistical results. These developments served in turn to dig deeper in subsequent
qualitative analyses. The result of this theoretical–quantitative–qualitative process
proved crucial for generating the theoretical explanation presented in Chapter 2. The
insights from the qualitative analysis also influenced the core argument presented in
Chapter 5 discussing the onset of the war on drugs. The qualitative analysis fur-
ther served to suggest variables for inclusion and inform the interpretation of the
quantitative results in Chapters 6 and 7.

1.1.3
Contributions

This research presents several theoretical, empirical and substantive contributions

to the study of organized criminal violence and more broadly to the understanding
of the microdynamics of conflict.
Theoretically, this study redefines our understanding of large-scale organized crim-
inal violence as a phenomenon largely neglected by conflict scholars in political sci-

9
ence despite the lethality and pervasiveness of this form of violence in the developing
world. By analyzing the behavior of the state and criminal organizations, this re-
search bridges ideas from the literature on political violence, criminal sociology, eco-
nomics of crime and democratization to provide new insights that improve our under-
standing of this complex and understudied phenomenon. For the reasons explained
below, political science mistakenly abandoned an early research agenda initiated by
Graham and Gurr (1969) and Gurr (1989) which combined insights from criminal
sociology and political science to study riots, protests, crime and assassinations. The
present study constitutes an effort to revive this research agenda.
Another theoretical contribution of this research is the development of a formal
model to understand different aspects of organized criminal violence contained within
an integrative analytical framework. This model is based on a set of explicit assump-
tions, and develops a simple set of propositions to provide a general explanation for
the onset, escalation and geographic concentration of violence. The emphasis on the-
oretical desegregating makes it possible to specify the conditions of why, when, by
whom and where violence emerges as a behavioral manifestation of assorted actors
operating within structural factors. The theoretical explanation thus contributes to
the research agenda of the microdynamics of conflict. In contrast to the analysis
of violence against civilians in civil wars dominating this research agenda (Kalyvas,
2006, 2012), this dissertation directs the focus of the micro-level agenda of conflict
towards a study the dynamics of organized criminal violence.
The analytical leverage of the formal model is used to derive a set of empiri-
cal implications which are then subject to empirical scrutiny. The analysis tests the
observational extensions of the model by focusing on Mexico and conducting compar-
ative analyses of all its sub-national units for a time span of eleven years. Evaluating
the scope conditions, explanatory power, and external validity of the theory outside
the Mexican case is a pending task for future work. If this theoretical explanation

10
contributes to an understanding of organized criminal violence or other processes of
violence in other latitudes, a major – as yet undeserved – compliment will be served
to this research.
Empirically, this research presents a novel data base of event data on organized
criminal violence called “Organized Criminal Violence Event Data in Mexico 2000–
2010” (OCVED). The data provide information on the violent and non-violent law
enforcement actions conducted by government authorities against criminal groups,
as well as events of violence perpetrated by criminals against the state, and violent
confrontations between rival criminal organizations. OCVED include information
from all Mexican municipalities on a daily basis from January 1, 2000, to December
31, 2010, comprising more than 9.8 million observations. This database of geo-
referenced daily event data at the municipal level provides detailed information on
who did what to whom, when and where in the Mexican war on drugs. The empirical
support in this research also joins a recent trend in political science relying on “big
data” to conduct large-data set analysis of conflict (Leetaru and Schrodt, 2013).
The use of micro-level data enables claims about the relevance of macro-structural
factors often used to explain political violence and criminal behavior to be evaluated.
As discussed in the empirical chapters of this dissertation, the empirical results based
on fine-grained data challenge the explanatory power of structural factors and reveal
the highly dynamic and interactive character of violence at the micro level.
Extant methodological strategies for collecting and systematizing data on violence
proved prohibitively expensive or largely inaccurate for coding information from origi-
nal sources that was necessary to generate the type of data required to test the theory.
To overcome this limitation, this study developed Eventus ID, a novel software for
automated coding of event data from text written in Spanish. The components of an
event are defined as someone (the source), doing something (the action), to someone
else (the target). By bringing together cutting-edge advances in computer science

11
with recent methodological developments in conflict research, this software will also
serve as a public good for other scholars to make advances in the creation of new
event data sets in their own fields and topics. In addition, the ability of this soft-
ware to code text written in Spanish enables researchers to analyze timely, detailed
information written in this language that often is not translated into English.
Substantively, the results of this research question the international paradigm
encouraging the implementation of punitive strategies to fight drugs. In particular,
the contribution of this study helps reinforce the outcry by victims, public opinion,
activists, analysts and scholars denouncing the escalation of the war on drugs in Mex-
ico. Beyond the astonishing scarcity of rigorous, evidence-based analysis that rarely
rises above well-intentioned opinion-making, this research offers a sound explanation
and solid evidence about the causes, mechanisms, magnitude, scope and length of
the wave of criminal violence generated by the implementation of a massive quasi-
military strategy to fight organized crime in Mexico. By focusing on this country, this
research is also a response to the imperative need to understand and curb large-scale
organized crime violence, an understudied yet pervasive threat to political stability in
Latin America (e.g., Colombia, Brazil, El Salvador, Guatemala and Honduras) and
in fragile states worldwide (e.g., Afghanistan, Tajikistan, Myanmar and Somalia).
Beginning with the Shanghai Opium Commission in 1909, international efforts
led by the United States and held at several conventions of the United Nations (UN)
since 19611 have adopted a global paradigm largely criminalizing the production,
transportation, and consumption of narcotics. The criminalization approach relies
on a punitive perspective primarily focused on combating the supply of narcotics.
The basic intuition behind this policy is that criminalization is expected to increase

1
These efforts are based on a set of international instruments including the Single Convention on
Narcotic Drugs in 1961 (amended by the 1972 Protocol), the Convention on Psychotropic Substances
in 1971, the Convention Against Illicit Traffic in Narcotic Drugs in 1988, the Convention Against
Transnational Organized Crime in 2003 and the United Nations Convention Against Corruption in
2005 (United Nations, 1961, 1971, 1988, 2000, 2002).

12
the prices of illicit drugs to the point of making them prohibitively expensive for con-
sumers, thus reducing the demand for drugs. To achieve this goal, the international
paradigm encourages the implementation of harsh law enforcement that has evolved
over the last few decades into a generalized use of quasi-military strategies to fight
drugs. However, after a century of implementation, this approach shows no signs of
success. The policy has largely failed to reduce the demand for drugs or to increase
drug prices. In 2010, there were about 300 million people worldwide who used an
illicit substance at least once in the previous year, among which 38.6 million were
drug dependent (United Nations Office on Drugs and Crime, 2013b). In addition, the
price of all drugs has been continuously declining for several decades while the purity
has remained stable or increased (Office of National Drug Control Policy, 2004). For-
tunately, a group of world leaders, international civil society organizations, scholars,
and the media have denounced the pernicious consequences of this policy in terms
of the violence, corruption, public insecurity, health epidemics and poverty traps it
has contributed to causing in developing countries (Crook, 2009; Global Commission
on Drug Policy, 2012, 2013; Latin American Commission on Drugs and Democracy,
2009; Reuter, 2008; The Vienna Declaration, 2010; Zedillo and Wheeler, 2012). Un-
fortunately, revisionist efforts have lacked the necessary coordination and analytical
rigor to generate solid results. This research contributes to these international efforts
by offering a robust explanation supported by rigorous empirical evidence about the
deleterious effects of implementing quasi-military strategies to fight drugs.

1.2
Redefining
the
Role
of
Organized
Criminal
Violence
in
Conflict
Research

Despite the pervasiveness of organized criminal violence in developing countries

and the lethality of this type of violence, conflict scholars in political science have

13
largely neglected the study of sustained campaigns of violence perpetrated by criminal
organizations against government authorities or against rival criminal associations.
This research seeks to fill the gap in conflict literature by attracting the attention of
political scientists to the micro-dynamics of organized criminal violence as a relevant
type of conflict.
Figure 1.1 helps to elucidate the gap in the literature by presenting a typology
of different forms of violence along three dimensions. First, violence can be situated
along the horizontal axis according to the degree of political motivation held by the
perpetrators of violence. The second dimension maps the extent to which violence
is motivated by economic objectives. The vertical axis represents the orientation
towards the status quo showing to what extent violence is being employed for changing
the status quo or for preserving it. Status quo is here broadly understood as the
prevailing power structures defining the political and economic conditions of a polity.
This tridimensional space helps to picture the combination of political and economic
motivations of violence that aims to challenge or preserve the status quo. The clusters
of spheres shown in Figure 1.1 do not represent real cases of conflict; they are mere
abstractions for illustrative purposes.
One type of violence referred to here as political preservation represents cases
where violence is used mainly for political reasons in order to maintain the status
quo. This type of violence is represented by the green cluster, and includes in-
stances in which government authorities use the coercive apparatus of the state to
stay in power. Different types of coercive behavior fit in this cluster of politically
motivated violence for preserving the status quo, including state repression (Daven-
port, 2009b; Goldstein, 1978; Regan, 2009), covert political surveillance (Cunning-
ham, 2003; Davenport, 2005), protest policing (Schwedler, 2005), counter-insurgency
strategies (Berman, Shapiro and Felter, 2011; Petraeus, 2006), death squads (Mason
and Krane, 1989) and even genocide (Harff, 2003; Valentino, 2000).

14
(Skocpol, 1979), insurgencies (Wickham-Crowley, 1992) and terrorism (Enders and
Sandler, 2006; Weinberg, Pedahzur and Hirsch-Hoefler, 2004).
The third group, referred to here as economic and political challenge and repre-
sented by the gray cluster, comprises cases in which rebels use violence to change the
status quo for a combination of political and economic reasons. These conflicts are of-
ten depicted as predatory conflicts in which insurgents use violent means to attempt
to obtain political power and seize economic opportunities for looting. There is a
large literature on civil wars arguing that rebels in civil wars are mostly motivated
by “greed” (Collier, 2000; Collier and Hoeffler, 1998) and there are several studies
linking natural resources such as oil, diamonds and other primary commodity exports
with conflict as key determinants (Fearon, 2005; Humphreys, 2005; Ross, 2006).
The fourth group refers to economic preservation and challenge and includes cases
in which violence is mostly used for economic purposes in an attempt to maintain
or disrupt the status quo. This largely overlooked type of violence is the focus of
analysis of this study and is represented by the blue cluster. Some cases in the lower
part of this cluster could illustrate criminal organizations using violence to resist law
enforcement efforts. Violence is used in such cases to maintain the power structures
that allow criminals to extract economic rents from illicit markets. The upper part of
the blue cluster refers to criminal organizations using violence to disrupt the status
quo that allows their rivals to exist and for pressing for a redistribution of power. A
central characteristic of the blue cluster is that the actors are not motivated mainly
by political ends. A few recent studies explore the types of violence corresponding
to this cluster such as warlords, pirates and organized criminals (Bailey and Taylor,
2009; Blattman and Miguel, 2010; Davis, 2009; Koonings and Kruijt, 2004; Skaperdas,
2002; Thomas, Kiser and Casebeer, 2005; Thomson, 1996; Tulchinm, Frühling and
Golding, 2003). However, this type of violence remains largely underdeveloped in
theoretical and empirical terms.

16
 As illustrated in Figure 1.1, there is an area where the blue cluster of violence
for economic preservation and challenge overlaps with the gray cluster of violence for
economic and political challenge. These are instances in which rebels in civil wars
seem to behave mostly like rent-seeking actors primarily motivated by the opportunity
to loot. Predatory behavior by some rebels rests at the core argument of several
authors who dismiss grievances and political objectives, claiming that civil wars are
largely conducted by kleptocrats (Brito and Intriligator, 1992; Collier, 2000; Collier
and Hoeffler, 2004; Grossman, 1999; Hirshleifer, 1991). A clear example is Grossman
(1999, 269) arguing that “insurgents are indistinguishable from bandits or pirates.”
There are certainly some aspects of organized criminal violence that are similar
to the characteristics of conflict in civil wars. The resemblance is clear when the
key elements included in the definition of civil wars are considered. According to
Sambanis (2004), there is a broad consensus among conflict scholars that a civil war
is defined as a conflict characterized by: (i) reciprocal and sustained armed action
(ii) between the state’s armed forces operating within its own territory and (iii) one
or many organized armed actors, (iv) capable of mounting effective resistance and
(v) generating a toll of at least 1,000 people killed as the product of armed clashes
between warring parties. Although this study does not claim to define organized
criminal violence as a type of civil war, it is useful to draw parallels between these
two types of violence.

• Reciprocal and sustained armed action. Organized crime violence often

includes several armed clashes between criminal groups and the state security
forces or violent confrontations between rival criminal groups. These armed
encounters are not isolated or sporadic events generated by random circum-
stances, rather they are part of systematic and sustained campaigns of violence.
These violent interactions between state security forces and insurgents are also
a central characteristic of civil wars.

• Use of the armed forces operating within the state’s territory. Due to
the endemic weakness of police forces and the high levels of corruption affect-
ing local security forces in developing countries, anti-criminal strategies often

17
 rely on the deployment of military personnel for conducting policing activities.
Armed forces operate within the territorial boundaries of the state to deal with
domestic security concerns. The reason for using the army is also justified by
the considerable firepower capability of criminal organizations, which can eas-
ily outgun a poorly equipped local police forces. Similarly, civil wars involve
the direct participation of military personnel to fight rebels within the state’s
territory.

• Organized armed actors. The term “organized” in “organized crime” refers

to the ability of individual criminals to overcome dilemmas of collective action
and coordinate efforts to conduct violence and engage in criminal activities.
As insurgents do in civil wars, criminal groups use a variety of mechanisms to
organize collective efforts of violence against state authorities or rival criminal
associations. Following Lichbach’s (1998) mechanisms for solving dilemmas of
collective violence, it is possible to identify how criminal organizations rely
strongly on economic incentives from illicit markets to motivate their members
(market mechanism) and use hierarchical structures for imposing, monitoring
and enforcing agreements on their lower ranks (hierarchical mechanism).

• Effective resistance. Due to the substantial financial resources extracted

from illicit markets, some criminal organizations are capable of recruiting a
large number of members and acquiring the necessary material and military
supplies for mounting an effective resistance against state authorities or rival
criminal organizations. The availability of resources characteristic of criminal
organizations contrasts with the scarcity of financial and material resources
often limiting the military capability of insurgent groups for effectively resisting
state actions in civil wars.

• Lethality of violence. The standard threshold of defining an intra-state

conflict as a civil war is at least 1,000 casualties in a year. The death toll
of organized crime violence is similar to that of civil wars. For example, the
Mexican war on drugs killed about 35,000 people between 2007 and 2010. As
noted by Osorio (2012), this death toll comparable to the onset of some 35 civil
wars in only four years.

Collier (2000) considers that rebellion is a quasi-criminal activity. According

to this author, rebellion and organized crime are largely analogous, although not
identical. However, the distinction is based on four “stylized facts” (Collier, 2000,
840). The first difference refers to levels of poverty. Richer societies are far less likely
to experience civil wars. However, improved economic conditions do not seem to be
associated with less violent crime. Second, income inequality is not related to a higher

18
risk of civil war. In contrast, increased income inequality does seem to be a related
to higher crime rates. Third, natural resources increase the likelihood of rebellion.
However, Collier considers that natural resources do not not have such a large impact
on criminal behavior. Finally, according to this author, although insurgencies are not
caused by identities (ethnic, religious or class), rebels usually adopt the discourse of
cleavages as part of the rebellion. In contrast, organised criminal behavior typically
does not occur along identity lines.
Unfortunately, Collier misses a crucial distinction that differentiates loot-seeking
behavior in civil wars from violence perpetrated by criminal organizations: the ori-
entation of violence towards the status quo. In some civil wars, rebels use violence
against the state in order to overthrow government authorities and control economic
resources for private benefit. In civil conflict, capturing power is a way of gaining
access to economic gain, and the only way to do it is by challenging the status quo. In
contrast, organized criminal violence comprises two distinct types of behavior. Orga-
nized criminals may use violence against state authorities as an attempt to maintain
the power structures that permit them to extract rents from illicit markets. Criminal
violence used in this way is a reaction against law enforcement and is used for preserv-
ing the status quo. The other type of behaviour in which criminal organizations may
engage is to commit violence against other criminal groups in an effort to challenge
the status quo in which their rivals enjoy a privileged position for extracting rents
from illicit markets. Analyzing the orientation towards the status quo shows clearly
that despite the behavioral similarities between rent-seeking rebels and criminals,
they are fundamentally different in nature: rebels use violence against the state to
challenge the existing power structures, while criminals use violence against the state
to preserve the status quo and against other criminals to challenge it.
One of the main reasons why political scientists have paid scant attention to the
study of organized criminal violence may be that this type of violence is often disre-

19
garded as non-political in nature or not politically relevant. However, by neglecting
the study of organized criminal violence, political scientists overlook a central prob-
lem in political theory: the existence and collapse of order. The central role of the
state is to impose order. Since Thomas Hobbes (1651), the foundation of the state
and the social condition has depended on the sovereign’s control of the means of co-
ercion. The legitimate monopoly of violence is also a distinctive characteristic of the
state according to Max Weber (1978). But order is not only a theoretical construct.
Historically, the process of state formation has been inherently connected to violence
as states try to monopolize the accumulation and concentration of coercive means to
impose order (Bates, 2008; Scott, 2009; Tilly, 1985, 1992, 1975). According to Tilly,
coercion defines the extent of domination and order: those who apply coercion on
others obtain their compliance, and from that compliance draw multiple political and
economic advantages not available to those who are less powerful Tilly (1992, 70).
However, as noted by Kalyvas, Shapiro and Masoud (2008), violence is not only used
by those trying to dominate but also by those who want to overthrow the existing
order. The violent confrontations between organized criminals and government au-
thorities, as well as between rival criminal groups fighting each other, thus reflect the
efforts of different actors using force for maintaining or contesting the existing order.
By focusing on organized criminal violence, this research addresses central concerns
of political science on the existence and collapse of order.

1.3
Key
Conceptual
Definitions

To guide the rest of this study, this section discusses the main conceptualizations
of organized crime, drug-trafficking organizations and violence used throughout this
dissertation.

Organized crime

20
 There is no consensus among scholars about what constitutes a criminal orga-
nization. For example, von Lampe (2013) presents a collection of more than 170
theoretical, legal and operational definitions of organized crime. Their nuances are
probably as diverse as the authors and their cases of interest. Despite this diver-
sity, there are four elements that often appear as key components of many different
conceptualizations of organized crime: (i) organizational structures; (ii) the use of
violence or the threat of violence; (iii) the centrality of territorial control and (iv) the
economic benefits derived from illicit markets. Following Reuter (2008), I refer to
organized crime as a set of stable, hierarchically organized groups of criminals with
the ability to use violence – or the threat of it – for acquiring or defending control of
illegal markets in order to extract economic benefits from them.
According to this definition, criminals are organized according to hierarchical and
stable structures. This organizational characteristic distinguishes organized crime
from individuals engaging in criminal behavior in a sporadic and isolated way. The
term organization implies the cohesion of the group over a period of time. Criminal
organizations conduct their illicit endeavors in a sustained manner rather than sim-
ply getting together to commit a few crimes and then disbanding. Organizational
characteristics also imply the implementation of structures and mechanisms favor-
ing the resolution of collective action dilemmas, thus allowing criminals to achieve
shared economic interests. The solution of collective action may include mechanisms
of selective incentives, monitoring, and sanctions to prevent free-riding or defection.
Organizations also imply a certain degree of specialization in which different members
or groups within the organization have the means and skills to conduct an array of
specific activities. One basic distinction is between members specialized in produc-
tive activities and others specialized in coercion. This differentiation is often referred
as “bread or bullets” in the literature on the economics of crime (Skaperdas and
Syropoulos, 1995) and political violence (Bates, Greif and Singh, 2002). Having spe-

21
cialists in violence is crucial for criminal organizations because, as noted by Reuter
(1989), criminals do not have the legal and institutional protection to enforce their
agreements that licit markets have. Therefore, the only way to enforce contracts is
through a group of members that can use – or threaten to use – violence.
The hierarchical characteristic is also a distinguishing feature of organized crime.
Criminal organizations are often led by an individual or small group of leaders who
acquire decision-level positions due to their family ties, strategic skills, violent reputa-
tion, ethnic base or a combination of any of these factors.2 The hierarchical structure
allows leaders to impose centralized control over the members of the organization with
the objective of coordinating and monitoring their productive or coercive activities.
The definition also requires these hierarchical to be stable over time and not the
result of ephemeral opportunities.
Due to their hierarchical structure and organizational complexity, large criminal
groups such as drug-trafficking organizations can be distinguished from gangs operat-
ing at street level. The behavior of gangs can be similar to the behavior displayed by
large criminal organizations. Gang members may engage in some or any of drug traf-
ficking, racketeering, kidnapping, crime, extortion, or violence against government
authorities and rival gangs. However, gangs tend to operate in smaller geographic

2
It is hard to describe a stereotypical structure of a criminal group. For example, consider the or-
ganizational characteristics of the Chinese, Colombian, Italian and Japanese criminal organizations
presented by Mandel (2011). The Chinese mafia is composed of seven major groups divided along
ethnic lines: the Sun Yee On, the Wo Group, the 14K, the Luen, the Big Circle Gang, the United
Bamboo, and the Four Seas Gang. These groups were characterized by a tight hierarchy and rigid
structure, but in recent decades have adopted a more flexible cellular organization. In Colombia, the
Medellı́n and Cali cartels were the two most prominent drug-trafficking organizations that emerged
in the 1970s. Pablo Escobar was the head of the Medellı́n cartel and the Rodrı́guez Orejuela broth-
ers led the Cali organization. After their leaders were captured and killed in the mid-1990s, these
large organizations eroded into smaller cells. Eventually, the Norte del Valle cartel emerged as a
confederation of drug-trafficking families. The Italian Mafia is composed of three loosely connected
groups known as the Camorra (in Naples), la Cosa Nostra (in Sicily) and the ’Ndrangheta (in Cal-
abria). The Italian Mafia is structured around family and loyalty ties and operates with long-rooted
cultural codes of morality and behavior. A final example, the Japanese Yakuza is composed of three
large organizations: the Yamaguchi-gumi, the Ingawaka-kai, and the Sumiyoshi-kai. The Yakuza
are notorious for their strict hierarchy and rigid codes of conduct regulating justice and duty.

22
areas than large criminal organizations and the menu of criminal activities and in-
ternal areas of specialization are more limited in gangs than in large criminal groups.
Another central difference is that gangs do not seem to follow the command of a cen-
tralized authority. Instead, they operate in a horizontal and decentralized manner in
small cells with fluid leadership.

Drug trafficking organizations

Organized criminals engage in a wide variety of activities ranging from production
and trafficking of illegal drugs, human and arms smuggling, bookmaking, money laun-
dering, extortion, and more. This research focuses on drug trafficking organizations
(DTOs), a type of criminal organization primarily focused on growing, manufactur-
ing, transporting, distributing and marketing substances which are subject to drug
prohibition laws (United Nations Office on Drugs and Crime, 2013a). The main focus
of DTOs on drug-trafficking activities does not exclude other illegal activities such
as extortion, money laundering, kidnapping, or weapons smuggling. However, these
activities are considered secondary as they are undertaken to facilitate, reinforce or
complement drug-trafficking.
Due to the variety of activities associated with markets of illicit substances, drug-
trafficking organizations tend to have large structures comprising different groups spe-
cializing in different aspects of the production chain of illicit drugs. Drug-trafficking
organizations also tend to have a sector specialized in money laundering. In addition,
these organizations generally have a strong security apparatus for protecting valu-
ables at the various stages of the production process and for enforcing agreements.
Due to the global characteristics of drug markets, drug-trafficking organizations also
tend to extend their operations internationally by bridging drug-producing regions
and drug-consuming regions.

23
 In this research, the terms organized crime, criminal organizations, criminal groups,
DTOs, and drug cartels are used indistinctively for referring to drug-trafficking or-
ganizations.

Violence
As mentioned above, one of the key characteristics of criminal organizations is
their ability to use violence, or the threat of using it, in order to secure access to the
economic benefits of illicit markets. Due to the centrality of violence, it is important
to offer a definition of violent behavior. Following Kalyvas (2006), violence refers to
the deliberate infliction of physical harm on people or damage on their property that
can be inflicted for either tactical reasons – to eliminate a specific target – or strategic
motivations – to prevent a certain behavior. In this sense, violence is defined from a
rationalist perspective and is the result of strategic calculations by those perpetrating
acts of violence.
The rationalist approach is certainly a minimalist point of view that does not con-
sider the mental and physical stress related to killing and learning to kill as discussed
by Grossman (1995). Nevertheless, the analytical leverage of analyzing violence from
a strategic perspective is used for understanding the strategic interactions of con-
flict rather than explaining violence as the product of psychological disturbances or
deviant behavior.3
This study utilizes broad definition of drug-related violence as the violence perpe-
trated either by drug-trafficking organizations or the state’s security personnel when
conducting anti-criminal activities. The general category of drug-related violence is
further disaggregated into three distinct types of violence classified by the dynamic
interactions between different actors. First, enforcement refers to the coercive ac-
tions conducted by the state against drug-trafficking organizations in an attempt to
enforce the law. Government authorities have a broad menu of security tactics for

3
For an excellent review of psychological explanations of violence see Gurr (1968).

24
fighting crime which includes the use of violent enforcement – referring to the use of
lethal force against criminals – and non-violent enforcement – referring to arrests and
seizure of criminal properties and goods. The second type of violence is retaliation,
which includes the deliberate use of violence by members of drug-trafficking organi-
zations against state security forces or government authorities. Finally, competition
refers to acts of violence perpetrated by members of a criminal organization against
rival criminal groups.
Returning to the analytical framework illustrated in Figure 1.1, law enforcement
refers to violence in which state authorities have primarily political motivations to
fight criminals in an attempt to change the status quo that allows those criminal
groups to exist. Retaliation refers to violence perpetrated by criminal organizations
against the state in an effort to preserve the power structures that enable them to
extract economic benefits from illicit markets. Competition refers to violent inter-
action between rival criminal organizations fighting to preserve or change the status
quo that generates more favorable economic conditions for some criminal groups over
their competitors.
Finally, this research considers large-scale drug violence as the aggregated pro-
cesses of violence caused by a systematic series of struggles between the state and
drug-trafficking organizations, as well as between rival criminal groups. With respect
to the state, this definition of violence implies the deployment of the armed forces for
fighting criminal organizations and for conducting policing activities. For organized
criminals, it implies the groups of heavily armed men conducting sustained cam-
paigns of violence against government authorities and state security forces or against
members of rival criminal organizations.

25
1.4
The
Road
Map

As aptly pointed out by Escalante Gonzalbo (2012), most of what we know, or

think we know, about drug trafficking and organized criminal violence is often a jum-
ble of stereotypes, fears, folklore, scattered facts, social reconstruction and abysmal
ignorance. Categorical statements are usually informed by loaded ideological stands
or laden with stigma. The secret and clandestine character of organized criminal
activities make evidence elusive and increase the difficulty of analyzing it systemati-
cally. In consequence, it is hard to confirm or discard theoretical efforts. In addition,
some of those who have ventured in the criminal underworld with investigative im-
petus found that the lethality of organized criminals is not simply a myth. Our
“common wisdom” about organized criminal violence is rarely based on theoretical
foundations, sound evidence or rigorous empirical analysis. This project seeks to
contribute to correcting this state of affairs.
The research begins by laying out the theoretical foundations for the rest of the
study. Chapter 2 presents a game theoretic model to understand the onset, escalation
and geographic concentration and distribution of organized criminal violence. Based
on a set of explicit assumptions and basic propositions, the model helps identify the
conditions under which the various actors are likely to engage in violence. The general
argument is that democratization erodes the peaceful configurations between the
state and criminals and motivates politicians to fight crime, thus triggering a wave of
violence between the state and organized criminals and between rival criminal groups
fighting to control valuable territories. The micro-level predictions indicate that state
action has a non-neutral effect among criminals: by weakening one criminal group
though law enforcement, the state indirectly motivates a rival criminal organization
to invade the weakened group. The analytical framework is used for generating
a set of hypotheses about the micro-mechanisms inherent to drug-related violence.

26
These empirical implications are subject to scrutiny in the empirical chapters of this
dissertation.
In order to test the empirical implications of a formal model emphasizing theo-
retical disaggregation, it is necessary to rely on fine-grained evidence reflecting the
dynamic interactions among the different actors required by the theory. Chapter 3
presents the strategy for generating data that meets the standards of the theoretical
explanation. This chapter describes the features and capabilities of Eventus ID, the
software developed as part of this research to code event data from text sources in
Spanish. After locating this data generation strategy within a broader family conflict
databases, the chapter evaluates the strengths and shortcomings of computerized and
manual annotation strategies. The chapter then discusses the massive data-gathering
strategy that was used to collect thousands of news reports from 105 different infor-
mation sources at the national and local level. For the sake of transparency, the
discussion makes explicit the coding rules used in this research. Finally, the chapter
discusses in detail the technological innovations that enable the software to cope with
the challenges of Spanish grammar and complexities of geo-referencing event data.
Chapter 4 presents the database of event data on drug-related violence generated
by the automated coding protocol. This chapter first discusses the conceptual and
methodological limitations of extant databases of homicides in Mexico. Instead of
taking a body-count approach, this research relies on an event data approach, which
provides more detailed information on who did what to whom, when and where.
This chapter discusses the characteristics of the data and describes the main trends
of violence at the national, state and municipal level using different units of temporal
aggregation. The “bulk” of violence is disentangled into its different components
and the temporal and spatial variation of violent competition between rival crimi-
nal organizations: violent retaliation perpetrated by criminals against government
authorities; violent law enforcement tactics used by the state to fight criminal or-

27
ganizations; and a broad menu of and non-violent law enforcement tactics including
arrests, confiscation of criminal assets, drug interdiction and seizures of weapons.
The data base thus comprises fine-grained information of different manifestations of
violence perpetrated by different actors. The data covers all municipalities of Mexico
on a daily basis from January 1, 2000, to December 31, 2010; a massive time-series
cross-sectional array of more than 9.8 million observations.
Chapter 5 analyzes the onset of the Mexican war on drugs by tracing the his-
torical process that led to the emergence, consolidation, and collapse of order. The
analytical narrative presented in this chapter is based on a set of explicit assump-
tions presented in the theoretical model to explain how democratization undermines
peaceful agreements between government authorities and criminal organizations and
motivates politicians to fight crime. This chapter tells how political order emerged
after the Mexican revolution with the formation of the PRI as a political agreement
to peacefully regulate access to power. The pact granted relative autonomy to lo-
cal leaders on the periphery, who benefited from the economic opportunities created
by Prohibition in the U.S. The second section analyzes the consolidation of order
during the period of political hegemony imposed by the PRI. Although corruption
and selective enforcement served to maintain order among criminal organizations, the
primary instrument used by the PRI to instill and impose order was a hierarchical,
centralized system of political incentives aligned with the overall government, and
party structures characteristic of the authoritarian regime. Finally, the third section
analyzes how the process of democratization eroded the feasibility of establishing
corrupt agreements. By increasing the number of political parties and favoring the
effective circulation of political elites, democracy subverted the system of political
incentives that allowed order and discipline to be imposed on criminal organizations.
In addition, increased political competition generated incentives for political actors
to enforce the law in an effort to gain popular support.

28
 Chapter 6 provides the main quantitative assessment of the expectations derived
from the theoretical model. The quantitative analysis is based on a careful exam-
ination of the hypotheses about the escalation and geographic concentration and
distribution of violence among rival criminal organizations. The central dependent
variable of this section is the quantity of violent competition among criminal orga-
nizations. The first section of the chapter estimates the analytical leverage of extant
explanations based on macro-structural factors. The analysis reveals their limited
explanatory power for understanding the broad and rapid variation in violent com-
petition among criminals at the micro-level. The second section evaluates the main
hypotheses derived from the theoretical model. To do so, the analysis builds on
the set of structural explanations and incorporates an interactive approach to ac-
count for the effect of law enforcement on violence among criminal groups. This
perspective reflects the conflict interactions between the state and criminal organiza-
tions specified in the theoretical framework, thus receiving the name of “interactive
approach.” To overcome the challenges of endogeneity generated by distinct but over-
lapping types of violence, the identification strategy uses an instrumental variables
(IV) method. This approach permits reliable estimates to be made of the effect of law
enforcement on violence between criminals, while avoiding the problem of reciprocal
causation. The research design uses measures of democratization and political strain
as instrumental variables to generate plausibly exogenous variation in the levels of
law enforcement, which then has an effect on levels of violent competition between
criminals. This quasi-experimental research design not only represents a plausible
identification strategy, but also conforms to the process of conflict specified by the
theoretical framework, thus enabling a the ontology of the theory and the method-
ology used for testing it to be aligned. The results indicate that democratization
and political strain increase law enforcement, which then has a profound disrupting
effect and triggers waves of violent competition among rival criminal groups. These

29
results are consistent across the different model specifications evaluating the effect of
violent and non-violent tactics on the levels of violence among criminals. The sta-
tistical analysis also reveals that criminal violence tends to cluster around strategic
territories favorable for the reception, production and international distribution of
illicit drugs.
Chapter 7 evaluates the dynamic interactions between law enforcement, criminal
retaliation, and violent competition among criminal groups. The empirical analysis
uses various time-series analysis strategies that allow the temporal inertia of each
type of violence and its interaction with other forces of violence to be incorporated.
The research design uses vector autoregressive (VAR) models and impulse response
functions (IRF) to assess how changes in one process of violence affect the dynamic
behavior of other processes of violence. These estimation techniques allow the mag-
nitude and the duration of dynamic effects to be estimated. The results reveal that
increasing the levels of law enforcement generates a massive escalation of violence
among criminal groups and a substantial increase in criminal retaliation against the
state. In addition, the dynamic analysis indicates that intensifying the levels of vio-
lent competition among criminals has a modest effect of increasing law enforcement.
The results show that direct attacks perpetrated by criminals against government
authorities, do however, generate more marked reactions from the state.
Finally, Chapter 8 concludes with a general discussion of the findings of this
research and suggests future areas of development.

30
 CHAPTER
2

AN
INTEGRATIVE
THEORY

OF
ORGANIZED
CRIMINAL
VIOLENCE

2.1
Introduction

This research studies the dynamics of the ongoing wave of drug violence in Mex-
ico by addressing three key questions. Why do politicians decide to fight criminal
organizations after having peacefully coexisted with them for several decades? Once
the conflict starts, why has drug-related violence escalated so rapidly? And lastly,
why is violence more concentrated in some areas than in others? In contrast to uni-
directional accounts of conflict primarily emphasizing the role of structural factors, I
advance a dynamic and interactive explanation of violence driven by territorial com-
petition. This chapter presents a formal model offering an integrative explanation
for the onset, escalation and concentration of organized criminal violence. Based on
a set of explicit assumptions and basic propositions, the general argument holds that
democratization erodes the peaceful configurations between the state and criminals
and motivates politicians to fight crime, thus triggering a wave of violence between

31
the state and organized criminals and among rival criminal groups, which tends to
concentrate around valuable territories.
To explain the onset of the war on drugs, the theoretical model claims that de-
mocratization disrupts the peaceful configurations that enable coexistence between
corrupt government authorities and criminal organizations in contexts of low demo-
cratic development. Increased democratization thus alters the system of political
incentives for enforcing the law, motivating government authorities to fight crime. In
addition, the model argues that government authorities obtain political benefits from
implementing harsh security policies when their legitimacy is threatened by periods
of political strain.
To explain the escalation of violence, the theory applies a contest success model
for territorial competition. According to the model, increased levels of law enforce-
ment are likely to trigger an escalation of conflict between the state and criminal
organizations, and violence among rival criminal groups. In this way, the action of
the state has a disruptive effect on the relative military balance of criminal orga-
nizations. Law enforcement weakens the capability of a criminal group to protect
its territory, thus motivating an invasion from a competing criminal group that now
faces a weaker rival. The equilibrium conditions of the model indicate that violence
committed by criminal organizations is a function of the severity of military damage,
their capability of recovering from it, and the value of the territory. As the elabo-
ration of the model will show, the equilibrium implies that organized criminals are
likely to use violence if the net military strength recovered through fighting back after
being attacked is larger than the cost of fighting, given the value of a territory.
Finally, to explain the geographic distribution and concentration of violence, the
model explicitly incorporates territorial value as a key determinant of conflict. Ac-
cording to the model, criminal organizations are willing to engage in violent con-
frontations to capture or defend valuable territory.

32
 This chapter is divided into four sections. The first part surveys the literature
on conflict research, criminology, and economics to assess a broad set of alternative
explanations. The second section presents the theoretical model proposed in this
research to explain the onset, escalation and concentration of large-scale organized
criminal violence. The third section derives a set of observable implications based on
the analytical leverage of the formal model; these implications are tested in the em-
pirical chapters of this study. Finally, the conclusion summarizes the characteristics
of the model and discusses its theoretical contributions.

2.2
A
Fragmented
Set
of
Alternative
Explanations

This research draws on different bodies of literature including criminology, eco-

nomics, and political violence to evaluate various arguments that seek to explain the
dynamics of drug-related violence. Extant explanations of criminal violence can be
classified into four main groups: economic benefits, sanctions against crime, social
propensity and political factors. Table 2.1 compares these four types of explanations.
Although some of them provide valuable insights, these explanations have two main
limitations. First, extant theories offer a set of fragmented propositions of isolated
aspects of criminal violence and there is no encompassing theory seeking to provide
a comprehensive account of the onset, escalation and geographic concentration of
criminal violence. The second shortcoming is that most of these explanations em-
phasize the importance of large-structural factors or processes that do not account for
the rapid variations in violence across time and space nor the dynamic interactions
among different actors. In consequence, these structural accounts leave a substantial
amount of variation unexplained.

The theoretical model advanced in this research addresses the limitations of these
approaches by offering two main contributions. First, the theory disaggregates dif-

33
 TABLE 2.1

ALTERNATIVE EXPLANATIONS

Type Concept Expected effect on

criminal violence
Poverty
Benefits Drug markets Increase
Territorial value
Law enforcement
Sanctions Decrease
State repression
Propensity Social structure Increase
Corruption and protection
Political Decrease
Coercive reaction

ferent mechanisms to explain the onset, escalation and diffusion of violence, and
integrates them into a unifying analytical framework. Second, the theoretical model
outlines a set of micro-mechanisms for the dynamic interactions of violence between
the various actors and integrates these rapidly changing actions and reactions within
larger structural factors. In this way, the theoretical explanation brings together
structural and dynamic mechanisms of conflict. The remainder of this section out-
lines the explanations offered by other authors and evaluates their theoretical and
empirical strengths and limitations.

Economic benefits Explanations focused on economic benefits refer to elements

that increase the economic utility derived from engaging in violent criminal activities.
These arguments emphasize the role of three central factors: poverty, the high value
of drug markets, and the availability of natural resources. Poverty is one of the most
common arguments invoked to explain criminal and violent behavior. Since poverty
is a multidimensional factor, numerous disciplines have analyzed its relationship with
crime and violence from diverse perspectives. Some authors argue that the benefits of

34
engaging in criminal behavior depend on the level of poverty. This approach claims
that individuals living in lower economic strata derive relatively larger economic
benefits from engaging in criminal activities than individuals with higher levels of
income. From a sociological perspective, (Merton, 1938, 1957) proposed a seminal
theory of social anomie arguing that crime is the product of a gap between the set
of economic aspirations and the possibility of realizing them through licit means.1
However, critics of this theory argue that societies do not have a set of homogeneous
values equally shared across the population Miller (1958) and that social anomie
does not explain why some individuals conform to the rules of society despite living
in harsh economic conditions Hirschi (1969). Despite the theoretical shortcomings
of sociological explanations linking poverty and crime, economists have identified a
positive correlation between high levels of poverty and income inequality with high
levels of criminal violence (Fajnzylber, Lederman and Loayza, 2000a, 2002a,b).
Conflict scholars in political science have also identified a strong association be-
tween poverty and political violence, but they offer at least two different mecha-
nisms to explain this relationship. As mentioned before, Collier and Hoeffler (2004);
Grossman (1999) argue that members of armed groups are largely characterised as
loot-seeking actors using violence to appropriate sources of income. This being so,
poverty intensifies motivations of greed and increases the propensity to use violence as
a way to obtain personal economic benefits. An alternative explanation argues that
poverty is indicative of state weakness. Rebels can easily prey on weak states because
the latter usually lack the institutional capability, economic resources, and military
strength to successfully control insurgents (Fearon, 2005; Fearon and Laitin, 2003;
Herbst, 2004). Although explanations based on greed and state weakness provide

1
According to this theory, society homogeneously imposes a set of aspirations to economic well-
being on all its members, yet the legitimate means of realizing such goals are not evenly distributed.
If the social structure is not capable of allowing its members to achieve their economic aspirations,
individual frustration motivates them to use illicit ways to fulfill their aspirations.

35
valuable analytical leverage, their applicability to an understanding of the Mexican
war on drugs may be limited for two empirical reasons. First, Mexico is a middle in-
come country with an annual per capita income of $15,150 dollars (The World Bank,
2010), thus located well above the low levels of economic development associated
with higher risks of intra-state violence. More importantly, even when considering
the subnational variation of violence and poverty in Mexico, the relationship does
not hold. States in the south and south-east of the country report the highest levels
of poverty, illiteracy, malnutrition and lack of infrastructure, yet they have substan-
tially lower levels of violence than the wealthy industrialized states in the north of
the country.
Economists specializing in the study of organized crime argue that there is a
strong relationship between the value of drug markets and the use of criminal vio-
lence (Kilmer et al., 2010; Levitt and Dubner, 2005; Reuter, 2008, 2009). According
to this perspective, the high price of illicit drugs promises substantial economic ben-
efits to those engaging in drug trafficking and motivates criminals to use violence for
capturing those rents. The importance of illicit drugs fits within a broader consensus
among scholars about the economic salience of illicit markets for explaining criminal
activity (Gambetta, 1993; Guerrero, 2009b; Maltz, 1976; National Advisory Commit-
tee on Criminal Justice Standards and Goals, 1976; Reuter, 1989; Schelling, 1967a,
1971; Volkov, 2002) and is consistent with the argument of greed commonly used
in civil war research. Based on this approach, it could be expected that increased
drug prices are associated with higher levels of drug violence. However, according
to the Office of National Drug Control Policy (2004), the prices of almost all drugs
have been systematically declining during the past three decades. For example, the
average price of a gram of pure cocaine powder in the United States dropped about
70 percent in the 1980s, then experienced an additional decline of 30–40 percent in
the 1990s and a further reduction of 12–21 percent in the 2000s. Other drugs show

36
similar downwards trends. If the positive association between drug prices and crimi-
nal violence were true, we would expect lower levels of violence as drug prices decline.
Nevertheless, in spite of the continued drop in drug prices, violence has recently es-
calated in Mexico. In consequence, there are reasons to be skeptical about a positive
relationship between drug prices and violence.
Research on political violence has identified the relevance of natural resources
to explanations of domestic conflict. Several studies on civil wars show that the
production of oil, diamonds, minerals or other primary commodities is strongly asso-
ciated with the risk of intra-state violence (Dunning and Wirpsa, 2004; Fearon, 2005;
Humphreys, 2005; Ross, 2006; Snyder and Bhavnani, 2005). There are two mech-
anisms linking the availability of valuable commodities with conflict. Rent-seeking
insurgents may use violence to seize control of sources of income (Collier and Ho-
effler, 2004). In addition, rebels may use natural resources as valuable sources of
funding that enable them to sustain prolonged campaigns of violence (Fearon, 2004).
It is not only important to determine whether a country has natural resources or
not, but also where those resources are located within the country. Research on the
sub-national determinants of conflict find that the spatial variation of violence is
largely influenced by geographic characteristics, including the spatial distribution of
natural resources (Buhaug and Ketil Rod, 2006; Buhaug, Gates and Lujala, 2009).
In consequence, the strategic value of particular regions is crucial for understanding
the spatial distribution of violence.
The importance of territorial value is not only recognized by conflict researchers
but has also been identified by organized crime scholars (Gambetta, 1993; Reuter,
2009; Schelling, 1971; Varese, 2013; Volkov, 2002). The boundaries of criminal ter-
ritories, like any other property rights in illicit markets, are not protected by legal
institutions or processes of conflict resolution. Therefore, criminals rely on violence
or the threat of violence to protect the borders of their territory, and tend to allocate

37
more resources and effort to protecting the more valuable territories (Reuter, 1989).
Of course, the menu of illicit activities is very broad, and not all illegal sectors rely
equally on the importance of territorial value.2 Among illicit activities, drug markets
are particularly sensitive to the importance of territorial control. Effective control
over specific areas is crucial for a wide range of drug-related activities such as cul-
tivation of land for illicit crops, securing transportation routes or controlling street
corners for retail sales.
In this sense, studies from political violence and organized crime address the the
relevance of strategic territories as a valuable insight for understanding the structural
determinants of organized criminal violence. However, the strategic location of some
areas or the economic value associated with them are large structural factors that do
not change over time (e.g. their geographic position) or if they change, they usually
do so slowly and gradually (e.g. growing illicit crops). In consequence, they have
limited explanatory power to account for rapidly changing actions and reactions of
actors involved in violent conflicts. In any case, valuable territories can serve as
containers of several micro-mechanisms of conflict concentrated around the specific
characteristics that make them valuable.

Sanctions against crime and violence

The second group of explanations refers to societal and government actions aimed
at inhibiting criminal and violent behavior by increasing the cost associated with en-
gaging in this type of activities. According to Becker (1968), criminal behavior can
be explained from an economic point of view by a cost–benefit analysis. Criminals
assess the net benefit derived from committing a crime and the costs associated with

2
For example, Schelling (1967b, 1971) argues that criminal activities such as gambling do not
depend much on controlling a territory because illicit transactions can happen anywhere at any time.
In contrast, other activities such as racketeering or drug distribution are highly territorial. In those
markets, criminals do not want competition because allowing rivals to extract rents would reduce
their own income. In consequence, criminals in protection and drug markets will try to monopolize
the territorial control of those markets.

38
it. If the benefits outweigh the costs, then they are more likely to commit the crime;
but if the costs are larger than the benefits, then it is less likely that they will engage
in the behavior. This point of view is compatible with the theory of social control pro-
moted by (Hirschi, 1969) and further developed by (Kornhauser, 1978). This account
argues that societies play a central role in imposing sanctions (control) and instilling
internalized codes of conduct (self-control) that reduce the propensity for criminal
behavior. If formal or informal measures of control fail, then higher rates of crime
are likely. In addition to society, government authorities count on the criminal jus-
tice system to enforce the law. States have police forces, prosecutors, judges, courts,
correctional institutions, and laws to control and sanction criminal behavior. The
traditional approach to fighting crime promotes the deployment of punitive actions
under the expectation that increased probability and severity of sanctions will inhibit
criminal behavior through deterrence, incapacitation, rehabilitation, imprisonment,
and moral education (Tonry, 2009). Unfortunately, empirical research on the effec-
tiveness of law enforcement is inconclusive and often plagued by anecdotal evidence
lacking systematic scrutiny. For example, some studies show that incarceration is ef-
fective in reducing crime rates (Levitt, 1996; Marvell and Moody, 1994) while others
argue that imprisonment is not an effective policy for reducing crime because some
criminals such as drug traffickers and gang members are quickly replaced (Nagin,
1998).
Political scientists have also devoted considerable attention to studying why and
how political authorities use coercive power domestically against potential and exist-
ing challengers (Davenport, 2007). According to Goldstein (1978), repression involves
the threat or actual use of physical harm, instrumented by the state coercive appa-
ratus within its territorial jurisdiction against individuals or organizations, for the
purpose of imposing a cost on the target, as well as deterring specific activities and/or
beliefs perceived to be challenging to government personnel, practices or institutions.

39
 Early research on state repression understood government coercion as a behavior
influenced by systemic factors such as the level of economic development or political
characteristics (Dallin and Breslauer, 1970). Later work explained repression as a
pathological predisposition of specific leaders who were unable or unwilling to rule
by non-coercive means (Walker, 1969). The current theoretical paradigm explains
state repression as a rational decision-making process based on cost–benefit calcula-
tions (Davenport, Johnston and Mueller, 2005; Duvall and Stohl, 1988; Gartner and
Regan, 1996; Hoover and Kowalewski, 1992; Lichbach, 1984; Moore, 2000; Simon,
2008). According to this approach, if the expected benefits from successfully using
coercive actions against potential or existing challengers outweigh the costs of using
such tactics, then state repression is likely. In contrast, if the costs of coercion are
larger than the potential benefits, government authorities may prefer less costly al-
ternatives. Empirical analyses of the effectiveness of state repression in inhibiting
or suppressing challengers show mixed results. Some authors argue that govern-
ment repression is capable of deterring violent behavior by dissident organizations
(Luttwak, 1999; Petraeus, 2006; Tilly, 1978; Wagner, 1993). Others argue that the
use of force by the state’s coercive apparatus intensifies hostilities perpetrated by
dissidents against government authorities (Gurr, 1970; Hibbs, 1973). As mentioned
by Davenport (2007) and Lichbach (1987), there is no conclusive evidence nor pre-
vailing theoretical explanation on whether state repression leads to the escalation or
deterrence of contentious behavior by dissidents.

Social structure and the propensity for criminal behavior

The third block of explanations refers to social factors increasing the propensity
for criminal violence. One of these explanations invokes the relevance of changes in
traditional family structures. According to Hirschi (1969), families perform the func-
tion of instilling a sense of self-control in their members, thus reducing the propensity
to engage in criminal or aggressive behavior. On a similar note, Sampson, Laub and

40
Wimer (2006) argue that marriage defines responsibilities between spouses and de-
velops the sense of mutual support and self-discipline, which in turn reduces the
propensity for engaging in criminal activities. In consequence, divorce dismantles the
traditional family structure and disrupts these behavioral patterns, thus increasing
the risk of criminal behavior. Following this argument, empirical studies have found
that increasing divorce rates are associated with higher crime incidence (Cáceres-
Delpiano and Giolito, 2008) and monoparental families increase the propensity for
criminal behavior (Comanor and Phillips, 2002), especially in female-headed house-
holds (Glaeser and Sacerdote, 1999).
Research on criminology has also identified that adolescent motherhood increases
the incidence of crime. For example, Nagin, Pogarsky and Farrington (1997) argue
that children from adolescent mothers have higher propensity for engaging in crimi-
nal activities. However, in contrast to the theory of self-control proposed by Hirschi
(1969), this propensity is not due to the lack of mother’s emotional maturity nor to
a dysfunctional family environment but to the lack of financial resources. According
to this perspective, adolescent mothers are not fully incorporated into the labor force
and often lack economic support from their partner or extended family. Following this
argument, the relationship between adolescent motherhood and crime is not caused
by deficient socialization within the family but by adverse economic conditions. As
a counterexample to the positive association between adolescent motherhood and
higher crime rates, Donohue and Levitt (2001) argue that legal reforms decriminal-
izing abortion are a crucial factor to explain the decline of crime rates in the United
States. According to these authors, access to abortion allows women to postpone the
age of motherhood until they consider their economic conditions are appropriate for
having children.
Political scientists have also identified a relationship between sociological factors
and criminal behavior. In his seminal work, Putnam (1993) defined social capital

41
as the features of social organization such as trust, norms, and networks that can
improve the efficiency of society by facilitating coordinated actions. Subsequently,
he found that crime rates in the United States began to rise sharply at about the
time that social capital began a downturn (Putnam, 2000). The inverse relationship
between social capital and crime has been broadly confirmed in several developed
countries such as Finland (Salmi and Kivivuori, 2006), the Netherlands (Akçomak
and ter Weel, 2008), Italy (Buonanno, Montolio and Vanin, 2009), and in developing
countries including Mexico (Paras, 2007). In addition, Mexican government author-
ities largely relied on the argument of the deteriorating social fabric to justify the
crusade against drugs (Presidencia de la República, 2012). However, when analyzing
the relationship between specific indicators of social capital (including trust, religios-
ity, membership and participation rate in organizations) and crime across several
countries, Lederman, Loayza and Menendez (2002) find that the association only
holds for the indicator of trust, but is absent for the other measures. Moreover,
(Glaeser, Sacerdote and Scheinkman, 1996) argue that increasing social interaction
may lead to higher rates of crime. One possible causal mechanism suggests that
denser social interaction may also include “perverse social capital,” understood as
the networks, contacts, power relations and informal norms of behavior that reward
and motivate rent-seeking or criminal behavior (Rubio, 1997).

Political determinants
According to Davenport (2007), one of the most long-standing and stable find-
ings in the literature on political repression is that government authorities generally
respond with some form of repressive action to behavior deemed to be a threat to the
political or economic system. The consistency of this effect is known as the “law of
coercive responsiveness.” State repression is generally observed as a reaction against
challengers trying to subvert the political or economic status quo. Repressive action
invariably occurs in contexts of civil wars, insurgencies, revolutions or protest move-

42
ments. However, there are several reasons why the applicability of this argument is
limited for understanding aspects of large-scale organized criminal violence. First, the
law of coercive responsiveness assumes that government authorities invariably repress
challengers. However, in the Mexican case, as in many other scenarios, criminal or-
ganizations have the economic power to corrupt the state in order to deter or prevent
it from using repression against them. These corrupt agreements allowed the peace-
ful coexistence of criminal organizations and the state for several decades. Second,
criminal organizations are capable of neutralizing the law of coercive responsiveness
not only because they have the economic means to co-opt government authorities
but also, and perhaps most importantly, because criminals do not usually represent
a political threat to the state. As mentioned in Section 1.2, criminal organizations
function like firms primarily motivated by economic goals. They are not politically
motivated to overthrow the state and impose a new regime aligned with their ideo-
logical preferences. In consequence, they might be considered not to be a threat to
the political status quo, especially if they do not use violence against the state or
against other groups. Third, once criminal organizations resort to violent means, the
law of coercive responsiveness says that the state will react by employing repressive
actions to counter or eliminate the threat. However, this interaction becomes highly
endogenous as the state reacts in response to the violent threat of criminal groups.
Another political factor often associated with criminal violence refers to high levels
of corruption. Dozens of authors have analyzed the relationship between corruption
and drug violence in Mexico from a variety of approaches including history (Andreas,
1998; Astorga, 2005, 2010), journalism (Campbell, 2009; Osorno, 2009; Ravelo, 2007a,
2009) and political science (Bailey and Taylor, 2009; Bailey and Godson, 2000; Garay
Salamanca and Salcedo-Albarán, 2012; Lessing, 2012; Morris, 2012, 2013; Snyder,
2006; Snyder and Duran-Martinez, 2009). However, the direction of the relationship
is not clear. Most authors argue that criminal organizations use corruption as a

43
way to prevent law enforcement against them, thus suggesting that high levels of
corruption are associated with low levels of enforcement and violence among criminal
groups. In contrast, others argue that corruption is associated with weak states that
allow criminal organizations to conduct their illegal activities using high levels of
violence without incurring sanctions.
The debate linking corruption and violence is not only about the direction of the
relationship, but also about who is the agent and who is the principal in this ex-
change. On one hand, Lupsha (1995) argues that the Mexican case is characterized
by an “elite-exploitation model” in which politicians manipulate and exploit criminal
organizations, which serve as “cash cows” that provide funding and illicit enrichment
for the political elite. On the other hand, Lessing (2012) argues that criminals have
the upper hand with respect to government authorities as they use violence to in-
timidate law enforcers and lower the price of bribes, exemplified by the infamous
law of “plata o plomo” (silver or lead). In addition, Kenny and Serrano (2012a)
argue that both models are right, but they correspond to different stages of Mexi-
can history. During the period of PRI hegemony, the relationship between political
elites and DTOs was characterized by criminal subordination to the state, but the
power relationship inverted as the PRI’s dominance eroded. In any case, there is no
theoretical agreement nor homogeneous empirical results regarding the relationship
between corruption and violence.
Conflict scholars have also analyzed the link between democratization and polit-
ical violence. Several studies argue that semi-democracies are more prone to civil
war than either stable democratic or authoritarian regimes (Ellingsen and Gleditsch,
1997; Hegre et al., 2001). However, not only is the level of democracy important
for assessing the risk of conflict, but the rate of political change characteristic of
democratization processes may also trigger episodes of violence (Cederman, Hug and
Krebs, 2010; Hegre et al., 2001; Mansfield and Snyder, 2005; Snyder, 2000). These

44
explanations, located in a Huntingtonian tradition (1968), generally agree that the
failure of political elites to mobilize citizens massively franchised through democrati-
zation is likely to generate violent outcomes. However, these arguments are limited
for providing an understanding large-scale organized criminal violence, as they fail
to consider a key difference between political and criminal violence as discussed in
Section 1.2. In political conflict, rebels, motivated by political or economic goals, use
violence to challenge the status quo. In contrast, criminals do not seek to overthrow
state authorities and impose their own government agenda. If criminals exercise vi-
olence, they do it in the course of resisting law enforcement that operates against
their economic interests. Thus organized criminals use violence for preserving the
status quo that allows them to extract economic benefits from illegal markets. This
distinction makes it difficult to extend the argument of mass political mobilization
to an explanation of large-scale organized criminal violence.
Recent research specifically focused on violence in Mexico provides a more illumi-
nating account of the relationship between democratization and violence. Snyder and
Duran-Martinez (2009) offer the dominant explanation to account for the increase of
drug-related violence in Mexico. These authors propose a theory of state-sponsored
protection rackets understood as a set of informal institutions though which law en-
forcers refrain from sanctioning criminal organizations in exchange for benefits. These
state-sponsored protection rackets enable peaceful agreement between criminal and
government authorities that help maintain low levels of violence. Peace is enforced
through the expectation of selective enforcement in which the state will credibly pun-
ish criminals who are not capable of refraining from using violence against their rivals.
But during the process of democratization in Mexico, the state-sponsored protection
agreements collapsed and led to increasing levels of violence. A set of factors such
as increasing political competition, anti-corruption reforms, and lack of coordination
across levels of government eroded the state’s capacity for protecting some criminals

45
and selectively enforcing the law against their rivals. In consequence, the lack of
credible punishment allowed criminal organizations to engage in violent behavior.
Rios (2012a) provides a similar argument which states that greater decentraliza-
tion of formal and informal political institutions increases the propensity of criminal
groups to employ violence because criminal organizations are less likely to be pun-
ished. She refines the core argument, proposing that under decentralization there
are several government agents that can protect criminals within small jurisdictions
under their control, but they cannot guarantee protection outside their jurisdiction.
If a criminal group protected by one government agent uses violence in the jurisdic-
tion of another government agent, its behavior may go unpunished. In other words,
decentralization reduces the likelihood of punishment.
The theory of state-sponsored protection rackets is further developed by Duran-
Martinez (2012), who argues that the interaction between the cohesiveness of the
state security apparatus and the distribution of drug markets determines the fre-
quency and visibility3 of criminal violence. According to her argument, a shift from
monopolistic to competitive drug markets increases the frequency of violence as crim-
inal organizations use force to drive out their competitors. In addition, the transition
of the state coercive apparatus from a cohesive to a fragmented condition increases
the visibility of violence because fragmented states are less capable of protecting or
punishing criminals.
The family of explanations around the concept of state-sponsored protection rack-
ets offers some valuable insights. It is particularly useful for understanding variations
in criminal violence caused by the collapse of protection rackets. However, the causal
mechanism assumed by this theory is highly problematic. At its core, the theory
of state-sponsored protection rackets argues that criminal organizations engage in

3
Visibility is defined as the overt display of violence or criminal evidence and claims of respon-
sibility for criminal attacks.

46
violence because government authorities are not capable of punishing criminal orga-
nizations. As the coercive capability of the state decreases, criminal groups face fewer
restrictions or sanctions on violence against their rivals. Unfortunately, this depiction
of the state as passive and incompetent does not correspond with the ongoing efforts
of the Mexican government to fight crime. Empirical evidence shows unprecedented
levels of law enforcement, substantial development of security institutions, and the
massive deployment of the state coercive apparatus.
As will be discussed in Chapter 4, the punitive strategy launched by the Mexi-
can government to fight criminal organizations is characterized by exceptionally high
levels of both violent law enforcement and non-violent tactics such as arrests and
seizures of drugs, assets and weapons across the country. The actions of the Mexican
state include deployment of the Army and Navy for policing activities, substantial
increases in personnel, equipment and intelligence capabilities of the Federal Police,
implementation of several joint enforcement operations between federal forces and
state and municipal security forces, unprecedented budgetary allocations for security
forces at all levels of government, and a set of constitutional reforms to facilitate
law enforcement. In consequence, the probability and severity of punishment against
criminal organizations has never been so high in Mexico. The proactive and highly
aggressive behavior of the state coercive apparatus does not correspond to the depic-
tion of incompetent, passive security forces assumed by the theory of state-sponsored
protection rackets. In summary, the central claim of the state-sponsored protection
racket theory is based on the assumption that low expectations of punishment against
criminal organizations motivates the use of violence. However, this assumption is not
fulfilled by the empirical evidence. In consequence, the observed escalation of crim-
inal violence may be caused by a different mechanism than that proposed by this
theory.

47
 In contrast to the passive role of the state proposed in the theory of state-
sponsored protection rackets, the theoretical explanation advanced in this research
emphasizes the proactive role of the state in the fight against crime as a key de-
terminant of variation in the levels of organized criminal violence. The details of
the theoretical argument will be discussed in the following section. The effect of
law enforcement as a disturbance generating criminal violence has been identified in
other recent studies. Varioius policy analysts, journalists and scholars argue that
the punitive anti-crime campaign launched by President Calderón in December 2006
triggered the escalation of drug violence in Mexico (Calderón et al., 2012; Castañeda
and Aguilar, 2010; Escalante Gonzalbo, 2011; Escalante Gonzalbo et al., 2011; Guer-
rero, 2010a,b, 2011b,c; Merino, 2011). However, with the exception of Dell (2011),
most arguments lack a robust theoretical explanation, fail to empirically and system-
atically assess alternative explanations, and largely ignore problems of selection and
endogeneity in their empirical evaluations. Dell’s research is perhaps one of the most
theoretically and empirically sophisticated studies on drug violence in Mexico. She
argues that narrow election victories by the president’s political party, the Partido
Acción Nacional (PAN), at the municipal level are associated with an immediate
increase in drug-related violence. Based on an innovative identification strategy, she
finds that government crackdowns on drug-trafficking routes followed by close PAN
victories diverted criminal activities to neighboring municipalities, thus generating a
spillover effect and increasing the levels of drug violence in these municipalities.
Despite its valuable contributions, Dell’s study has some limitations. The dispo-
sition to enforce the law is exogenous to the model: there is no discussion about the
political determinants that generated the demise of state-sponsored protection rack-
ets nor about the incentives for government authorities to enforce the law. Failing
to provide a sound explanation for the collapse of the pre-existing order structures
misses a central part of the explanation for the escalation of drug-related violence.

48
This theoretical limitation is also reflected in the empirical analysis, as it fails to as-
sess the relationship between enforcement and violence before Calderón launched the
war on drugs in December 2006. The empirical analysis has some further limitations.
The measure of law enforcement used by Dell is based on a confidential database of
unknown precedence containing data on high-level drug arrests. Besides problems of
transparency and validity, focusing exclusively on high-level arrests ignores a broad
menu of violent and non-violent security tactics used by law enforcers to fight crime.
In addition, as will be discussed in Chapter 4, the use of homicide count as the de-
pendent variable is problematic because it does not provide information about either
the perpetrators or the profiles of the victim. This rough dependent variable obscures
crucial information about the interactive dynamics of conflict between the state and
criminal organizations, as well as among rival criminal groups.

General assessment of the literature

As discussed in this encompassing literature review, the extant accounts provide
valuable insights about the economic benefits motivating criminal violence, the imple-
mentation of sanctions aiming to suppress crime, the set of societal factors increasing
the propensity for delinquent behavior, and a group of political factors containing
and triggering criminal violence. Despite their contributions, the extant theories of-
fer only fragmented explanations of isolated aspects of criminal violence. There is
no unifying theory attempting to provide an integrated explanation for the onset,
escalation, and concentration of large-scale organized criminal violence. Another im-
portant limitation of extant accounts is that they rely largely on macro-structural
factors to explain the rapid variations in violence. However, these structural fac-
tors provide limited analytical leverage for understanding the micro-dynamic and
interactive characteristics of violence.

49
2.3
An
Integrative
Theory
of
Organized
Crime
Vi olence

The theoretical model advanced in this research provides an integrative frame-

work to explain different yet interrelated aspects of organized criminal violence. The
model disentangles the dynamic interactions among different actors operating within
structural factors, and integrates these elements into a coherent set of systematic
propositions. This analytical framework both disaggregates and integrates to provide
an encompassing account for the onset, escalation, and concentration of large-scale
criminal violence.
In general, the theoretical model provides an explanation of how democratization
disrupts the political conditions that motivate politicians to fight crime, thus trig-
gering an escalation of drug violence between the state and criminal organizations,
and conflict among rival criminal groups. At its core, this model is driven by a con-
test success function in which players compete for territorial control. The analytical
leverage of the formal model helps to elucidate how changes in the political structure
affect the system of incentives for politicians and motivate law enforcement. Accord-
ing to this account, state action is not neutral: law enforcement has a disruptive
effect on the relative military balance of criminal organizations. As government au-
thorities fight crime, law enforcement against a target criminal group weakens the
capability of this group to protect its territory and indirectly empowers the relative
military position of a rival criminal organization, which may motivate an invasion
into the territory of the weakened group. Depending on the strategic value of a ter-
ritory, criminal organizations may be willing and capable of engaging in sustained
territorial struggles against the state or their rivals if they have the capability to
recover from the damage created by law enforcement or attacks from rival groups.
The model indicates the conditions under which democratization motivates law en-
forcement and how fighting crime affects the relative military balance among rival

50
criminal organizations and generates a turf war for controlling valuable territories.
Finally, extending the argument of the model reveals that when the state launches a
general campaign against all criminal organizations within its territory, it generates
a wave of conflict of all against all resembling a Hobbesian state of war.
The explanation of the theoretical model is divided into six sections. The first part
presents a set of conceptual definitions of organized crime, drug-trafficking organiza-
tions and violence. The players and their actions in the model are then introduced.
The third section explains how democratization changes the political incentives for
fighting crime. The fourth part illustrates the micro-mechanisms of violence among
criminal organizations caused by the disruptive effect of law enforcement. The fifth
section presents the payoffs. Finally, the equilibrium conditions for the use of violence
are analyzed in the sixth segment.

2.3.1
Players
and
Actions

The formal model consists of a sequential game of complete information. The

model analyzes the interactions between three players: the State (S), the Target DTO
(T ) and the Challenger DTO (C). The game is played in a sequence of five steps.
First, the degree of democracy of the political system is set by Nature (N ) along
a continuum ranging from low to high levels of democracy (D), where D ∈ [0, 1].
Second, the State chooses to enforce the law against the Target DTO (E) or not
to enforce the law (∼ E). Third, if the State enforces the law, the Target criminal
organization decides whether to retaliate against the State (R) or not (∼ R). Fourth,
a Challenger criminal group decides whether to invade (I) the Target’s territory or
not (∼ I). Finally, if the Challenger invades, the Target decides whether to fight
back (F ) against the Challenger or not (∼ F ).
Figure 2.1 presents the extensive form of the game illustrating the sequence of
each player’s possible moves at every decision point.

51
 N
∼D D
S

∼E E

∼R R

C C C

∼I ∼I ∼I
I I I

52
[1] [4] [7]
B, ΩG − KST , ΩG − λ − KST ,
T T T
Mτ, M γτ , M γ 1−σ τ − KT S ,
(1 − M )τ (1 − M γ)τ (1 − M γ 1−σ )τ

∼F F ∼F F ∼F F

[2] [3] [5] [6] [8] [9]

B, B, ΩG − KST , ΩG − KST , ΩG − λ − KST , ΩG − λ − KST ,
M γτ , M γ 1−σ τ − KT C , M γ 2τ , M γ 2−σ τ − KT C , M γ 2−σ τ − KT S , M γ 2−2σ τ − KT S − KT C ,
(1 − M γ)τ − KCT (1 − M γ 1−σ )τ − KCT (1 − M γ 2 )τ − KCT (1 − M γ 2−σ )τ − KCT (1 − M γ 2−σ )τ − KCT (1 − M γ 2−2σ )τ − KCT

Figure 2.1. The Game of Drug Violence

 The model is based on two key assumptions: (i) criminal organizations are pri-
marily motivated by economic benefits, and (ii) before the breakout of hostilities,
government authorities and organized criminals coexist in a peaceful arrangement.
The scope conditions of this theory are valid at least in those cases where these
assumptions hold.

Assumption 1: Criminal organizations as profit maximizers

The model assumes that DTOs primarily behave like profit-maximizing firms ori-
ented towards extracting economic rents from illegal markets. This assumption is
broadly supported by the literature on organized crime advanced from criminological
(Maltz, 1976; National Advisory Committee, 1976), sociological (Gambetta, 1993;
Volkov, 2002), policy (Guerrero, 2009b) and economic perspectives (Reuter, 1989;
Schelling, 1967a, 1971). This assumption also implies that criminal organizations are
not mainly driven by political goals. Although DTOs may use violence to resist or
inhibit law enforcement, criminal organizations are not ideologically-oriented insur-
gents aiming to overthrow the government.4 In consequence, criminal organizations
are not expected to unilaterally initiate an attack against the state.

Assumption 2: Peaceful coexistence in the status quo

The model also assumes that prior to the beginning of a state campaign against
crime, government authorities and organized criminals coexist in a peaceful arrange-
ment: the status quo is defined by a non-hostile coexistence between the state and
DTOs. According to this perspective, the coexistence of state authorities with crim-
inal organizations capable of committing violence does not mean the various parties

4
There may be cases in which criminal organizations may seek to seize control over political insti-
tutions. For example, the prominent leader of the Medellin drug cartel in Colombia, Pablo Escobar,
was elected to the House of Representatives of Colombia’s Congress in 1982 (Guillén Jiménez, 2007).
In addition, some authors argue that right-wing paramilitiaries in Colombia have “captured” state
institutions at various levels of government (Garay Salamanca and Salcedo-Albarán, 2012; Garay
Salamanca et al., 2008; López, 2010). However, even in those cases, the penetration of criminal
organizations into the political scene seeks to inhibit law enforcement or other kinds of policies that
may affect or jeopardize criminal opportunities to extract rents from illegal markets. Therefore,
their infiltration is mainly motivated by economic reasons, not political or ideological goals.

53
constantly engage in violent confrontations. Rather, this peaceful coexistence indi-
cates that there is order, a set of norms regulating peaceful interaction among different
actors. This assumption is similar to what Snyder and Duran-Martinez (2009) call
“state-sponsored protection rackets,” in which public officials refrain from enforcing
the law in exchange for a share of the profits generated by criminal organizations.
The premise of peaceful coexistence diverges from the Weberian assumption that the
state has the legitimate monopoly on violence and allows for coexistence of both state
authorities and parallel power structures. As stated by O’Donnell (1993) and con-
firmed by others (Astorga, 2010; Obert, 2011), ineffective states in new democracies
often coexist with autonomous, also territorially-based, spheres of power.
As part of the premise of peaceful coexistence between the state and criminal
groups, the model also assumes that there are at least two DTOs existing within
the state’s territory. By assuming the existence of multiple (at least two) criminal
organizations, the model considers the origin of criminal groups as an exogenous
process. Other authors have analyzed the historical and political conditions favor-
able to the emergence of criminal organizations. Anderson (2005) argues that mafias
emerge due to the abdication of legitimate government power, excessive bureaucratic
power and the potential for illegal markets. Olson (2000) addresses the importance
of economic conditions leading to the origins of stationary bandits. Finally, Tilly
(1985) and Skaperdas and Syropoulos (1995) draw parallels between the process of
state formation and the emergence of criminal organizations based on their compar-
ative advantage of coercion in a context of anarchy. The assumption of a peaceful
configuration is also plausible for the Mexican case in which criminal organizations
peacefully coexisted with government authorities for several decades and conducted
their illegal activities without committing systematic violence (Astorga, 2005).

54
2.3.2
Fighting
Crime
in
New
Democracies

The link between democratization and drug violence is located within the Hobbe-
sian tradition of conflict research, in which violence emerges as the product of the
breakdown of political order (Hobbes, 1651; Kalyvas, Shapiro and Masoud, 2008;
Olson, 2000; Tilly, 1985). The model assumes that at low levels of democratic de-
velopment, state authorities coexist with criminals on a basis of corruption. The
parameter B > 0 represents bribes received by government officials in exchange for
not enforcing the law under a non-democratic regime. At higher levels of democratic
development, government authorities are motivated instead to provide public goods,
such as public security. The benefits of enforcing the law in a democratic setting are
expressed by the parameter G > 0. One of the key assumptions of the model is that
the political benefits of providing public security are larger than the benefits from
corruption (G > B) at high levels of democratic development. In an authoritarian
context, the political benefits that government authorities receive from enforcing the
law are smaller than the benefits of corruption (G < B).
Peaceful configurations do not need to be explicit “pacts” between the state and
DTOs, but can be achieved as a behavioral equilibrium. At low levels of democracy,
organized criminals may bribe government officials as a way of preventing prosecution
or even of receiving protection from the state (Bailey and Taylor, 2009; Guerrero,
2009b). The small number of relevant political actors characteristic of authoritarian
regimes favors peaceful arrangements between the state and DTOs in several ways.
Having a small number of political actors makes it easier for criminals to bargain with
government officials and makes it cheaper to bribe them. The small number of key
political actors also facilitates coordination among corrupt political elites and reduces
the risk of defection (Olson, 1965). In addition, the hierarchical chain of command
increases compliance by the lower ranks within the government structure, thus adding
stability to the pacts. Finally, the lack of effective elite circulation through electoral

55
means in non-democratic settings favors credible expectations about the stability
of these pacts. The coexistence of state authority and criminal power structures
is consistent with the view of O’Donnell (1993) on the challenges that many new
democracies face.
The process of democratization is considered as an external force gradually al-
tering the peaceful arrangement between government officials and criminals. De-
mocratization increases the number of relevant political actors at multiple levels of
government. Increasing competition motivates politicians to provide goods such as
public security. For organized criminals, a larger number of political actors increases
the difficulty of bargaining with state officials and the costs of bribing them. For
corrupt authorities, the entrance of new political actors makes coordination more
difficult and reduces their capacity for detecting and sanctioning those not comply-
ing with the pact (Olson, 1965). Even if a non-aggression agreement is achieved,
the diversity of party labels at different levels of government would break the chain
of command and would make compliance with the pact very difficult. In addition,
elections tending to increase the effective circulation of elites reduce the duration of
any non-aggression arrangement and increase the uncertainty about the possibility of
future arrangements (Przeworski, 1991). Under democracy, political actors also have
direct incentives to enforce the law as they seek to gain citizen support by deliberately
breaking corrupt agreements and framing themselves as honest politicians. Finally,
elections leading to increases in the effective circulation of political elites reduce the
duration of any non-aggression arrangement.
Democratization provides the basic conditions leading to higher levels of law en-
forcement. However, a full-fledged campaign against crime requires a trigger to shift
from implicit peaceful coexistence to active belligerency. Goldstein (1978) provides
a theory for understanding the decision of political leaders to use repressive tactics.
He argues that increasing levels of tension in the political arena are perceived by

56
authorities as threats to their legitimacy, thus increasing their disposition to repress.
According to this author, “a high level of strain and dissent will tend to increase
the anxiety of political authorities and incline them towards a policy of repression”
(Goldstein, 1978, p. 559). In addition, the adoption of repressive policies is facilitated
by the presence of suitable target groups in society which can readily be made into
scapegoats.5 Muller (1970) also argues that in periods of crisis, politicians usually
reap political benefits from deploying aggressive policies and displaying an image of
strong leadership. This is the well-known “rally-round-the-flag” effect.
The arguments advanced by Goldstein and Muller can be used as a meso-theory
to explain the motivation of political elites towards aggressive security policies in
periods of political crisis. As levels of strain and tension in the political arena in-
tensify, government authorities can be expected to deploy increasingly aggressive
security policies to fight crime. When political actors perceive that their legitimacy
is threatened by episodes of political strain, they have greater incentives to use puni-
tive security policies in an attempt to boost their levels of popular support. The
provision of heavy-handed security policies is particularly attractive for reaping po-
litical benefits because public security is a highly valuable public good across partisan
labels, thus making fighting crime very appealing to broad sectors of the population.
In addition, criminal organizations are suitable scapegoats for repressive policies that
are not likely to receive objections from other political elites. To express this idea in
formal terms, let Ω be the level of political strain, such that Ω > 1. Now, assume
that raising political strain increases the political benefits of fighting organized crime,
such that ΩHigh G > ΩLow G, when ΩHigh > ΩLow and G is held constant.

5
Scapegoats are individuals or groups in the society who are the focus of feelings of aggression and
hostility or receive negative treatment from the political elites or other groups in society. Repressive
policies against scapegoat groups are often feasible due to the lack of opposition to repression from
political elites. Goldstein developed his theory to explain intense political repression in the United
States against some political groups such as workers’ unions, communist groups and civil rights
movements, which have variously served as scapegoat groups.

57
2.3.3
Shifting
Military
Capabilities

Criminal organizations use their military strength to control specific territories.

The territory has a value of τ > 0. In a simplistic perspective, the value of the
territory is defined by the objective characteristics of a geographic space that can be
translated into economic benefits for criminal organizations. A more sophisticated
approach assigns territorial value as a function of the objective characteristics of the
area and the specific production functions of criminal organizations. In this version of
the model, criminal groups with different production functions could assign different
values to the same objective features of a given territory.6 However, for the sake
of simplicity, in this work the model assumes that all drug-trafficking organizations
value the territory in the same way.
The military strength of the Target is denoted by M ∈ [0, 1]. The Target’s relative
military capacity can be considered as a contest success function for the probability
of winning a turf war against a Challenger over a disputed territory. The military
power of the Challenger is defined relative to the Target’s strength as (1 − M ).
The Target enjoys the share of the territory it manages to control as measured by
M τ , and the Challenger controls the rest of the territory; (1 − M )τ . One of the
central characteristics of contest success functions is that the efforts of the players
are adversarially combined so that increased probabilities of winning for one actor
imply decreased probabilities for all adversaries (Jia, Skaperdas and Vaidya, 2012).
Contest success functions therefore reflect the substantive claim made by Tilly (1992,
71), “coercion is always relative; anyone who controls concentrated means of coercion
runs the risk of losing advantages when a neighbor builds up his means.”

6
For example, suppose one criminal group specialises in aerial transport of illicit drugs and
another in growing marijuana. Given their different production functions, these criminal groups
would assign different values to a mountainous terrain. The mountain may represent low value for
the group specialized in aerial transportation because airplanes cannot land on rough terrain, while
the drug farming organization may assign a high value to the mountain because its sheltered recesses
and high elevations are favorable for growing illegal crops.

58
 The most common functional form of the contest success function is that pro-
posed by Tullock (1980), sometimes called the “power” form, or also referred to as
the “ratio” form. In its most basic version, the contest success function considers two
adversaries, players i and j. Each contestant devotes some effort (e) to controling
its territory, their respective efforts denoted by ei and ej . For any given combination
of efforts, each player has a probability of winning and a probability of losing a con-
frontation. Let player i be the Target DTO with a probability of winning denoted
by Mi (ei ; ej ) and let player j be the Challenger DTO with probability of winning
Mj (ei ; ej ). In order to define the player’s military capabilities as a reciprocal proba-
bility, Mi and Mj must take values between zero and one, and add to one such that
Mj (ei ; ej ) = 1 − Mi (ei ; ej ) ≥ 0. In consequence, we can express the players’ contest
success functions as the ratio of their military efforts:
Target DTO (player i)
ei
Mi (ei , ej ) =
ei + ej

Challenger DTO (player j)

ej
Mj (ej , ei ) =
ei + ej

For notational simplicity, from here on M = Mj (ei ; ej ) will refer to the Target
DTO and (1 − M ) = Mj (ei ; ej ) to the Challenger.
The model also considers that an attack on the Target (delivered by either the
State or the Challenger) damages its military capabilities by a factor of γ ∈ [0, 1],
such that γM < M . If γ has values close to 1, then the degree of damage is minor
and M is barely affected. In contrast, values of γ close to 0 indicate a substantial
damage on the Target, severely affecting its military capabilities. The severity of
military damage can sequentially increase or decrease depending on the actions of
each player such that γ V , where V = (E − Rσ + I − F σ). Parameters E, R, I and

59
F correspond to the set of violent actions sequentially available to each player. If
any player opts to use violence, its respective action (E, R, I, F ) takes the value of
1, or 0 otherwise. However, if the Target fights back, it may neutralize some of the
damage caused by the attacker and reestablish part of the relative military balance
by a factor of σ ∈ [0, 1], which represents the Target’s recovery capability. The
Target can reestablish the relative military balance in two ways: it can increase its
own military capabilities (e.g. recruiting more hit men, increasing the cruelty of its
tactics or using more powerful weapons) or it can reduce the military capabilities
of its opponent (e.g. killing a rival). Values of σ close to 1 indicate a strong Target
capable of reestablishing the relative military balance in either of these ways. In
contrast, if σ is close to 0, it reflects a weak Target incapable of recovering its relative
military position.
Table 2.2 shows various scenarios of military damage and recovery based on the
different values taken by V depending on the actions taken by each player. As the
first row shows, if no one uses violence, then V = 0, γ 0 = 1, and the military balance
is not altered. In the second and third rows, if either the State or the Challenger
attack the Target and latter does not fight back, the attack diminishes the Target’s
military capabilities by γM . In the fourth and fifth rows, if the Target fights back
against the attacker, the reaction offsets part of the damage and the net power balance
is γ (1−σ) M . In the sixth row, the Target’s military strength is severely damaged by
γ 2 M if both the State and the Challenger sequentially attack and the Target does not
respond to either of them. In the seventh and eight rows, the Target is attacked by
the State and the Challenger but only fights back against one of them, thus leading
to a net damage of γ 2−σ M on the Target. In the last row, if the Target is attacked
twice and fights both aggressors, the Target neutralizes part of the damage and the
net military balance becomes γ 2−2σ M .

60
 TABLE 2.2

SEVERITY OF MILITARY DAMAGE AND RECOVERY CAPABILITY

Target State Target Challenger Target

Attacked Responds Enforces Retaliates Invades Fights V γV
No No 0 0 0 0 0 1
Once No 1 0 0 0 1 γ
Once No 0 0 1 0 1 γ
Once Once 1 1 0 0 1−σ γ (1−σ)
Once Once 0 0 1 1 1−σ γ (1−σ)
Twice No 1 0 1 0 2 γ2
Twice Once 1 1 1 0 2−σ γ (2−σ)
Twice Once 1 0 1 1 2−σ γ (2−σ)
Twice Twice 1 1 1 1 2 − 2σ γ (2−2σ)

Figure 2.2 provides an example of the shifting military capabilities caused by

violent interactions between different players. For the three different scenarios in this
figure, let M = 0.8 for the Target’s initial military strength, γ = 0.5 for the severity
of damage inflicted by the aggressor and σ = 0.7 for the Target’s effectiveness in
recovering from the attack. In the first scenario (A), the Target is attacked by either
the State or the Challenger, and fights back. The attack initially reduces the Target’s
military position from M to M γ. The reaction from the Target against its aggressor
allows it to recover some of the military loss by shifting M γ to its new position
at M γ 1−σ . Although the Target is better off by retaliating against its aggressor
(M γ 1−σ > M γ), there is some fraction of military strength that the Target is not
able to recover (M > M γ 1−σ ).
The second scenario (B) in Figure 2.2 shows the Target being attacked by both the
State and the Challenger and retaliating only against one aggressor. In this scenario

61
close to 1). As mentioned before, the Target can shift back the relative military
balance by reducing the military capabilities of its rivals through a counter-attack or
by increasing its own military capabilities by hiring more soldiers or improving its
tactics. If the Target fights back and readjusts the military balance with respect to its
adversary, the Target’s net military strength will be located around the right corner of
the surface in Figure 2.3. After this violent interaction, the relative military position
of the Target is not that much different from the initial point at the top corner of
the surface, but the process has generated two episodes of violence that altered and
restored the balance of power.
In addition, the dark patterned surface in Figure 2.4 represents the Target’s net
military strength after being attacked both by the State and a rival criminal group
and fighting back against both (γ 2−2σ ). The net military balance in this surface
is governed by the same relationship between the severity of military damage (γ)
and the effectiveness of recovery (σ) as in the previous example when there is only
one aggressor. Consider the case when the military balance of the Target is located
around the top corner of Figure 2.4. If the attacks on the Target by the State and a
challenging DTO cause minimum damage, its net military capabilities will be located
at the left corner of the surface. However, if both attacks cause severe damage on
the Target and its retaliation is not effective, then the Target’s net position will be
around the lower edge of the surface. Finally, if the Target has the capability to
reestablish the military balance by effectively fighting back against both the State
and the challenger DTO, then its net position will be around the right corner of
the surface. Regardless of the number of aggressors, the logic of violence remains
the same: if the attacks on the Target are severe enough, violence may end through
extermination. However, if the Target is capable of fighting back and effectively
recovering its military position, violence is likely to escalate.

64
more weapons). In consequence, launching the invasion generates a cost of KCT for
the Challenger and leads to a payoff of (1 − M γ)τ − KCT .
In Payoff 3 the absence of law enforcement gives the State the benefits of corrup-
tion (B). The Challenger launches an invasion and the Target resists the attack. After
the violent interaction between criminal groups, the Target enjoys the part of the ter-
ritory that it managed to recover after the invasion at some cost (M γ 1−σ )τ − KT C
and the Challenger gains a fraction of the territory after facing some resistance from
the Target ((1 − M γ 1−σ )τ − KCT ). In this scenario, the Target and the Challenger
engage in a confrontation that generates some episodes of violence in their respective
efforts to shift the military balance and control part of the territory.
In scenarios 4–9, the state enforces the law against criminals. In Payoff 4, the
State fights the Target but there is no retaliation against law enforcement nor violence
among rival DTOs. Enforcing the law gives the State a political benefit for providing
public security at some cost of enforcement (ΩG−KST ). For now, Ω is held constant.
In this case, the cost to the government (KST ) may involve increasing the budget for
security agencies or any other expense for spending financial, material or human
resources to fight criminal organizations. In this case, the actions undertaken by
the State’s coercive apparatus against a criminal organization weaken the Target’s
military capabilities for defending its territory by M γτ and improve the relative
power of the Challenger by (1 − M γ)τ . This implies that law enforcement has a non-
neutral effect on rival criminal organizations that may disrupt the relative military
balance existing between them. By enforcing the law, the State debilitates one group
and indirectly benefits its rivals.
Payoff 5 shows the scenario in which the State enforces the law and the Challenger
launches an invasion, but the Target does not retaliate against either the State or the
Challenger. In this case the State receives the political benefit of providing public
security at a certain cost (ΩG − KST ). Government action undermines the Target’s

67
military capabilities. The damage caused by law enforcement on the Target indirectly
increases the Challenger’s relative strength. If the Challenger decides to launch an
invasion, it will incur a cost and further improve its position by (1 − M γ 2 )τ − KCT .
If the Target does not defend itself from either the State or the Challenger, the
sequential attacks from these two actors will reduce the Target’s military capacity to
control its territory by γ 2 M τ . This scenario represents the worst situation for the
Target, as it may be seriously damaged by the attacks from both fronts.
In Payoff 6 both the State and the Challenger fight the Target criminal organi-
zation, and the latter fights the invaders but not the government security forces. In
this case, the payoff for the State is ΩG − KST for enforcing the law against the
Target. The sequential attacks from the State and the Challenger undermine the
Target’s ability to secure its territory, but the Target’s reaction against the Chal-
lenger helps it to recover part of the military loss caused by the invader’s incursion
((M γ 2−σ )τ − KT C ). After facing resistance from the Target, the Challenger enjoys

a relative military position indirectly improved by the State’s actions and further
increased by the invasion ((1 − M γ 2−σ )τ − KCT ).
Payoff 7 represents the situation in which the State enforces the law and the Target
retaliates against the State but there is no violent interaction between competing
criminal groups. In this scenario the State receives some benefit for enforcing the
law. However, criminal retaliation against law enforcement diminishes the State’s
benefits for providing public security by a factor of λ > 1, thus leaving a payoff for
the State defined by ΩG − λ − KST . After being damaged by law enforcement and
retaliating against the State, the Target’s payoff is defined by the net military ability
to defend its territory and the costs of fighting the State, as (M γ 1−σ )τ − KT S . In this
scenario the Challenger refrains from invading against the Target. In consequence,
the Challenger only receives the benefits of an improved relative military position
caused by State enforcement against the Target, (1 − M γ 1−σ )τ .

68
 In Payoff 8, both the State and the Challenger attack the Target, but the lat-
ter only resists law enforcement, not the invasion. The confrontation between the
State and the Target diminishes the benefits received by the State for providing
public security (ΩG − λ − KST ) and undermines the Target’s military strength. In
addition, if the Challenger decides to launch an invasion and the Target does not
expel the trespasser, the invasion would further reduce the Target’s military power
by M γ 2−σ τ − KT S and substantially improve the Challenger’s ability to control the
territory ((1 − M γ 2−σ )τ − KCT ).
Finally, Payoff 9 represents the situation of a war of all against all. The State
receives the political benefits of fighting criminals, even when facing resistance from
the Target and incurring the cost of enforcing the law (ΩG − λ − KST ). Government
actions undermine the Target’s strength and indirectly improve the relative position
of the Challenger. After retaliating against the State, the Target recovers some of its
military position damaged by law enforcement. If the Challenger decides to launch
an invasion, it will further weaken the Target and improve its own military strength.
However, if the Target resists the invasion, it will recover some of its relative power
position. These violent interactions will shift the military balance back and forth
between the Target and the Challenger and generate some costs, thus yielding a
payoff of M γ 2−2σ τ − KT S − KT C for the Target and (1 − M γ 2−2σ )τ − KCT for the
Challenger.
This payoff represents the most violent scenario involving actions and reactions
between the State and the Target and between rival criminal organizations. As
mentioned in Section 2.3.3, depending on the severity of military damage and the
ability of criminal organizations to recover from an attack, confrontations between
different players may lead to the extinction of some criminal groups and the end
of violence if they suffer serious damage and are incapable of recovering from it.
However, this interaction may alternatively lead to sustained campaigns of violence

69
if the military damage caused to each other is not sufficiently severe and if actors
manage to recover effectively after being attacked.

2.3.5
Equilibrium
Conditions

I use backward induction to find the sub-game perfect equilibrium of this sequen-
tial game of complete information. I start at the bottom of the game tree in order to
identify the conditions for the Target fighting against the Challenger. I then analyze
the conditions for the Challenger to invade knowing that the Target will resist the
invasion. At the next level, I identify the circumstances under which the Target will
retaliate against law enforcement. Finally, I analyze the conditions that would moti-
vate the State to enforce the law against organized crime. The equilibrium analysis
reveals the conditions under which each actor has incentives to use violence against
other players, given the benefits and costs associated with their actions.

2.3.5.1 Conditions for the Target to Fight the Challenger

A comparison of payoffs 2 and 3 for the Target shows that if the State does not
enforce the law, the Target DTO will fight the Challenger if the following condition
KT C
holds: (γ 1−σ − γ) > Mτ
. Parameter γ represents the extent of military damage
caused by an attack on the Target and γ 1−σ accounts for the Target’s capacity to re-
KT C
cover from an attack by fighting back. In addition, Mτ
represents the attractiveness
of engaging in a confrontation depending on the costs of fighting and the value of the
territory that the Target manages to control given its military capabilities.
In order to present a more intuitive interpretation of the equilibrium condition, we
can define θ = (γ 1−σ − γ) as the Target’s net military strength recovered by fighting
back after being attacked. Figure 2.6 offers a visual representation of parameter θ.
Consider any unidimensional space representing the military strength of the Target
(M ) and assume that this actor has full control of its territory (M = 1). If the

70
 The comparison of payoffs 8 and 9 in which the State enforces the law and the
Target retaliates against the State indicates that the Target will fight the invader if
KT C
the following condition holds: θ > γ 1−σ M τ
. Parameter γ 1−σ on the right-hand side
represents the Target’s military strength recovered by retaliating against law enforce-
ment. In this case, the Target will resist the invasion if the proportion of military
strength recovered by fighting the Challenger is larger than the relative attractive-
ness of battling over the disputed territory, even after the Target has suffered law
enforcement and retaliated against it.
In general, the analysis of equilibrium conditions indicates the same underlying
logic for the Target: if the proportion of relative military capacity recovered by force
is larger than the attractiveness of fighting for a valuable territory, then the Target
will fight.
A more nuanced analysis of the equilibrium condition reveals that this is not a
tit-for-tat model, since not every attack on the Target delivered by the Challenger
is immediately reciprocated by a violent reaction from the Target. The same “toler-
ance” applies to law enforcement actions conducted by the State against the Target.
The interaction among criminal organizations and between the state and DTOs can
tolerate “minor errors.” A member of the Challenger DTO could unintentionally
cross into the Target’s territory or government authorities might occasionally arrest
a member of the Target DTO. These minor events might not necessarily provoke a
violent reaction from the Target. The equilibrium condition reveals that the amount
of military damage has to be sufficiently large to motivate the Target to weigh the
benefits of waging war despite the costs associated with using violence. In particular,
the Target will fight if the proportion of military strength recovered though violence
is larger than the relative attractiveness of fighting for a territory of a certain value.
This condition might not be fulfilled by minor events such as inadvertently crossing
over to the rival’s territory or sporadic arrests. However, if the frequency and in-

72
tensity of minor events become part of a sustained campaign of hostilities against
the Target, then this condition is more likely to be fulfilled, thus triggering a violent
reaction from the Target against the instigators.

2.3.5.2 Conditions for the Challenger to Invade the Target

The second level of the model helps identify the conditions under which the Chal-
lenger will invade even knowing that the Target will fight back. The comparison of
the Challenger’s payoffs 1 and 3 indicates that in the absence of law enforcement,
the Challenger is likely to launch an invasion against the Target under the following
KCT
condition: 1 − γ 1−σ > Mτ
. Parameter 1 − γ 1−σ represents the Challenger’s net gain
in military strength after launching an invasion and facing resistance from the Target.
KCT
In addition, parameter Mτ
represents the costs of invading given the value of the
territory and the Target’s military strength.
KCT
We can define π = (1 − γ 1−σ ) and rewrite the equilibrium condition as π > Mτ
.
This indicates that the Challenger will invade if the net military gain is larger than the
attractiveness of invading, knowing that the Target will fight. Figure 2.6 also offers
a visual representation of parameter π. The invasion will improve the Challenger’s
military capabilities by a factor of 1 − γ) However, if the Target fights back against
the invader it will recover some of the military loss by the amount γ 1−σ . Therefore,

the net military gain for the Challenger launching an invasion and facing resistance
from the Target is defined by the space between 1 and γ 1−σ .
Consider payoffs 4 and 6, in which the State enforces the law and the Target does
not retaliate against the government. The equilibrium indicates that the Challenger
KCT
will launch an invasion if π > γM τ
. Parameter γ on the right-hand side represents
the additional damage caused by law enforcement on the Target. This indicates
that the Challenger will invade if the net military gain of doing so is larger than the

73
attractiveness of invading, even when the Target is likely to fight the Challenger after
being weakened by the State.
Finally, consider payoffs 7 and 9, where the State enforces the law and the Target
retaliates against the government. In this situation, the Challenger will invade if the
utility of doing so is larger than the utility of not invading, even after knowing that
the Target will resist the invasion. In this case, the Challenger will fight under the
KCT
following condition: π > γ 1−σ M τ
. Parameter γ 1−σ on the right-hand side accounts for
the Target’s damage and recovery caused by the violent interaction between the State
and the Target. The model indicates that the Challenger will carry out an invasion
if the net military gain of doing so is larger than the attractiveness of invading, even
knowing that the Target will fight the Challenger after retaliating against the State.
The model consistently indicates that law enforcement weakens the Target crim-
inal organization and improves the relative military position of the Challenger. This
suggests another important intuition about the non-neutral effect of State actions
on the relative military balance among criminals: regardless of the Target’s reaction
against its aggressors, the Challenger has more incentives to invade when the State
enforces the law against the Target than when it does not.

2.3.5.3 Conditions for the Target to Retaliate Against the State

Now, consider the Target’s decision to retaliate against the State by comparing
payoffs 6 and 9. If there is law enforcement, the Target will retaliate against gov-
ernment authorities if the utility of doing so is larger than the utility of not fighting
the state’s coercive apparatus. The equilibrium analysis indicates that the Target
KT S
will retaliate against law enforcement under the following condition: θ > γ 1−σ M T
.
As defined before, θ represents the Target’s net military capacity recovered through
fighting after being attacked, in this case by the State. In addition, parameter γ 1−σ
on the right-hand side represents the Target’s military strength recovered after re-

74
sisting the invasion launched by the Challenger. In consequence, the equilibrium
condition indicates that the Target will retaliate against law enforcement if military
strength recovered through fighting is larger than the attractiveness of engaging in a
confrontation with the State after the Target and the Challenger have already clashed
over a disputed territory.

2.3.5.4 Conditions for the State to Enforce the Law

Finally, comparing payoffs 3 and 9 reveals the conditions under which the State
will enforce the law. Knowing that the Target and the Challenger will engage in
territorial conflict and the Target will retaliate against law enforcement, the State
will launch a campaign against criminal organizations if the benefits of fighting crime
are greater than the benefits of not doing so. The equilibrium analysis indicates that
the State will enforce the law under the following condition: ΩG > B +λ+KST . Even

incurring the costs of law enforcement and the damage created by criminal retaliation
against government authorities, the State will enforce the law if the political benefits
of providing security as a public good are larger than the benefits of corruption from
not enforcing the law.
As mentioned before, while holding the levels of political strain (Ω) constant, the
model assumes that the political benefits of enforcing the law are larger than the
benefits from bribes (G > B) at high levels of democratic development, whereas the
relationship is the opposite (G < B) at low levels of democracy. In consequence, as
democratization increases, politicians are likely to intensify law enforcement activi-
ties against criminal organizations. As mentioned in Section 2.3.2, democratization
undermines the feasibility of peaceful configurations between politicians and crimi-
nals and affects the system of incentives for authorities, thus motivating politicians
to fight crime.

75
 Now, if we allow variation in the levels of political strain, the equilibrium condi-
tions for State action indicates that periods of strain in the political arena will lead
to more aggressive security policies against criminal organizations. As suggested by
Goldstein (1978) and Muller (1970), increased political tension facilitates repressive
behavior from government authorities and allows them to reap political benefits from
harsh security policies.

2.3.5.5 Summary of Equilibrium Conditions

Table 2.3 presents the summary of the equilibrium conditions for different pro-
cesses of violence inherent to the war on drugs. The structure of the equilibrium
conditions in columns 1–3 shows that criminal organizations follow the same under-
lying logic of violence whether fighting each other or against the state. According to
the analysis, if the proportion of relative military capacity recovered by force is larger
than the attractiveness of fighting for a valuable territory, then criminal organiza-
tions will fight. In addition, the table shows that the state will enforce the law if the
political benefits derived from enforcing the law exceed the benefits from corruption.

2.4
Empirical
Implications

Based on the set of conditions reported in Table 2.3, it is possible to derive

some observable implications. Consider first the equilibrium condition for the use of
state violence shown in the fourth column of Table 2.3. The equilibrium indicates
that government authorities will intensify law enforcement against criminals if the
political benefits from fighting crime (G) and levels of political strain (Ω) are larger
than the benefits from corruption (B), the damage caused by criminal action (λ)
and the cost of using force against criminals (KST ). As discussed before, the model

76
 TABLE 2.3

EQUILIBRIUM CONDITIONS FOR THE USE OF VIOLENCE

Target fights Challenger invades Target retaliates State enforces

the Challenger the Target against the State the law

KT C KCT
θ> Mτ
π> Mτ

KT C KCT
θ> γM τ
π> γM τ

KT C KCT KT S
θ> γ 1−σ M τ
π> γ 1−σ M τ
θ> γ 1−σ M τ
ΩG > B + λ + KST

assumes that the process of democratization and periods of political strain are key
external factors affecting the political incentives of politicians to enforce the law.
Based on these equilibrium conditions, the following hypotheses can be derived:

(H1 ) Increased democratization is associated with high levels of law enforce-

ment.

(H2 ) Increased political strain is associated with high levels of law enforcement.

(H3 ) Increased levels of corruption are associated with low levels of law enforce-
ment.

Hypothesis (H2 ) stating the relationship between political strain and law enforce-
ment can be further refined for different types of enforcement tactics. Based on
the “rally-round-the-flag” effect (Muller, 1970), assume that using violent enforce-
ment to fight crime (Gv ) provides greater political benefits than non-violent tactics
(G∼v ), such that Gv > G∼v . Also assume that the cost of using violent tactics
(KST v ) is higher than that of using non-violent law enforcement (KST ∼v ), such that
KST v > KST ∼v . In contexts of political strain, the net benefits of using violent tac-

77
tics are larger than those of non-violent enforcement, such that (Ωhigh Gv − KST v ) >
(Ωhigh G∼v − KST ∼v ) and (Ωlow Gv − KST v ) < (Ωlow G∼v − KST ∼v ). This implies that
violent law enforcement will only be used under high levels of political strain. This
refinement can be stated in terms of the following hypothesis:

(H2.1 ) Increased political strain is associated with high levels of violent enforce-
ment than non-violent enforcement tactics.

Now consider the equilibrium conditions for violence perpetrated by criminal or-
ganizations. Table 2.3 indicates that the Target and the Challenger have the same
underlying incentives for committing violence. The equilibrium conditions contained
in the first three columns of the table can be expressed in a more general set of
conditions:

Kij
δ> (2.1)
Mτ
Kij
δ> (2.2)
γM τ
Kij
δ > 1−σ (2.3)
γ Mτ

where the Target’s or Challenger’s benefits from using violence, δ, can take values of
{θ,π}. Parameter Kij represents the costs of using violence and can take any com-
bination of pairs {T C, CT , T S}. These conditions enable some general observable
implications to be identified.
First consider the disruptive effect of law enforcement. As mentioned in Section
2.3.3, parameter γ denotes the Target’s loss of military strength caused by an attack
either from the state or a rival criminal organization. For illustrative purposes in
this example, consider that γ is exclusively caused by law enforcement from the
state. Holding everything constant in equations (2.1), (2.2) and (2.3), comparing the
denominators on the right side of the three equations indicates that M τ > γ 1−σ M τ >
γ 1−σ M τ . This suggests that based on a given level of military capabilities (M τ ) and

78
in the context of a valuable territory, criminal organizations control a larger share of
their territory than when the state cracks down on them and reduces their military
capabilities (M τ or γ 1−σ M τ ). Still, if the territory is valuable enough, criminal
groups are better off retaliating against law enforcement in order to recover part of the
lost territory than not contesting law enforcement (γ 1−σ M τ > M τ ). In consequence,
if the model is correct, we should expect that criminals will use violence to fight back
the state. This suggests the following hypothesis:

(H4 ) Increased law enforcement is associated with high levels of criminal retal-
iation against the state.

According to the equilibrium conditions, the actions of the state also have a dis-
ruptive effect on the relative military balance among criminal organizations and, if the
territory is valuable enough, law enforcement may trigger violence among criminal or-
ganizations. Consider that a Target criminal group has a certain military strength to
control a territory (M τ ); in this case the relative military strength of the Challenger
DTO is denoted by (1 − M )τ . If the state enforces the law against the Target, the
crackdown will cause a certain amount of damage and reduce the military capabilities
of this group by a factor of γ, such that M γτ < M τ . By weakening the capability of
the Target to defend its territory, the State indirectly improves the relative military
strength of the Challenger criminal group, such that (1 − M )τ > γ(1 − M )τ . The
Challenger is thus better off after the State has enforced the law against the Target.
In consequence, the improved relative military position of the Challenger generates
an opportunity in this actor’s favor and may motivate an invasion against the al-
ready weakened Target criminal group. In addition, following the same logic of the
Target’s retaliation against the State, the Target is better off by fighting back the
Challenger’s invasion if the territory is worth the battle. Of course, it is difficult to
observe the operation of these micro-mechanisms of violence, but if the model is cor-
rect, we should observe higher levels of violence among criminal organizations after

79
government crackdowns. This violent interaction can be expressed in the following
hypothesis:

(H5 ) Increased law enforcement is associated with high levels of violent compe-
tition among criminal organizations.

The equilibrium conditions also suggest a set of empirical implications for the
degree of military damage and recovery capabilities of criminal organizations. As
discussed in Section 2.3.3, rival criminal groups may engage in violent interactions
trying to shift the relative military balance between them. Criminals can do so by
increasing their capability to inflict damage against their rivals (high values of γ),
or by increasing their own ability to recover from an attack (high values of σ). If
criminals become more able to inflict damage on their rivals, it is expected that
they will use that military strength to undertake violent actions. Therefore, violence
is likely to increase. In addition, if criminal organizations increase their capacity
to effectively recover from damage inflicted by their rivals, they are likely to keep
fighting for control of valuable territory. In consequence, violence is also likely to
rise. These relationships can be formulated according to the following hypotheses:

(H6 ) Increased capability to inflict damage is associated with high levels of

violent competition among criminal organizations.

(H7 ) Increased capability to recover from an attack is associated with high

levels of violent competition among criminal organizations.

Now, consider the importance of territorial value. As indicated by parameter τ in

equations (2.1), (2.2) and (2.3), increasing the strategic value of a territory (higher
values of τ ) is associated with a higher propensity to use violence for capturing or
defending those territories. In this way, the model incorporates the importance of
structural factors such as geographic conditions that may determine the concentration
of different dynamics of violence. This relationship can be formulated in the following
hypothesis:

80
 (H8 ) Increased territorial value is associated with high levels of violent compe-
tition among criminal organizations.

Finally, Table 2.4 summarizes all the hypotheses that have been derived from
the formal model. These empirical implications are organized according to their cor-
respondence to the main research questions motivating the theoretical model: the
onset, escalation and concentration of large-scale drug-related violence. The first set
of hypotheses refers to the effect of democratization and political strain on increasing
levels of law enforcement. This group also includes the negative relationship between
corruption and enforcement. The second group of empirical implications relates to
the diverse mechanisms operating in the escalation of violence. These include the ef-
fect of state actions on levels of criminal violence against government authorities and
the disrupting effect of enforcement on violent competition among criminals. This
group also includes a set of hypotheses for the relationship between the extent of
military damage and recovery capability on violence among criminal organizations.
Finally, to explain the geographic concentration of violence, the equilibrium condi-
tions suggest the hypotheses linking the strategic value of certain territories with the
level of conflict among criminal organizations. In general, the set of hypotheses re-
veals how the theoretical model is capable of integrating distinct explanations for the
onset, escalation and concentration of violence into a unified analytical framework.

2.4.1
War
of
All
Against
All

The formal model presented in this research analyzes interactions between the
state and two criminal organizations. By focusing on a small number of actors, the
theoretical explanations allow the different micro-mechanisms of violence between the
state and criminal organizations to be disentangled, as well as conflict between rival
criminal groups. In addition, the model helps in understanding the effect of law en-

81
 TABLE 2.4

SUMMARY OF EMPIRICAL IMPLICATIONS

Question Hypothesis Independent var. Relationship Outcome var.

H1 democratization increase law enforcement

Onset H2 political strain increase law enforcement
H3 corruption decrease law enforcement
H4 law enforcement increase violence against the state
H5 law enforcement increase violence among DTOs
Escalation
H6 damage capability increase violence among DTOs
H7 recovery capability increase violence among DTOs
Concentration H8 territorial value increase violence among DTOs

forcement in disturbing the relative military balance between criminal organizations,

which is capable of triggering a turf war between two criminal groups. The basic
premises of this model can be generalized to more than two criminal organizations.
Generalizing the formal model to any number n of DTOs would make the math-
ematical argument too complex. Instead, consider Figure 2.7, which shows the ex-
tension of the argument visually. Assume that there are several drug-trafficking
organizations operating within the jurisdiction of the state, denoted by DTO 1–6
in the figure. Panel (a) shows the interaction between the state and DTO 1. In
this scenario, the use of law enforcement against DTO 1 generates criminal retalia-
tion against government authorities but there is no violence between rival criminal
groups. In Panel (b), state law enforcement against DTO 1 has disturbed the rela-
tionship between DTO 1 and DTO 2, triggering an invasion by DTO 2 against DTO
1. In this scenario, DTO 1 retaliates against the state and fights back against DTO 2
to repel the invaders. In Panel (c), state action not only triggers conflict between the
state and DTO 1 but also a turf war between DTO 1 and DTO 2. Weakening DTO 1

82
may also signal to DTO 6 that it has an opportunity to invade the territory of DTO
1, thus opening an additional front of territorial competition between DTO 1 and
DTO 6. This scenario shows that law enforcement can generate a wave of violence
among neighboring criminal organizations.

(a) (b) (c)

DTO 1 DTO 1 DTO 1

DTO 6 DTO 2 DTO 6 DTO 2 DTO 6 DTO 2

State State State

DTO 5 DTO 3 DTO 5 DTO 3 DTO 5 DTO 3

DTO 4 DTO 4 DTO 4

(d) (e)

Legend
DTO 1 DTO 1
Enforce
DTO 6 DTO 2 DTO 6 DTO 2
Retaliate
State State Invade

DTO 5 DTO 3 DTO 5 DTO 3 Fight

DTO 4 DTO 4

Figure 2.7. War of all against all

Panel (c) in Figure 2.7 also reveals an interesting characteristic of the strategic
interaction between the state and drug trafficking organizations. In some circum-
stances, the state can threaten criminal organizations with the deployment of selec-

83
tive punishment. Consider a context where there is a general agreement between
government authorities and DTOs capable of maintaining a peaceful configuration.
In this scenario, the state might threaten to deliver selective punishment against a
specific criminal group and not prevent predation by its neighbors. Law enforce-
ment would signal that the targeted criminal group is no longer protected by a peace
agreement with the state and might trigger an invasion from all other criminal or-
ganizations operating around that group. The threat of severe damage caused by
law enforcement and simultaneous invasion by competing groups could deter crimi-
nal organizations from breaching the terms of the agreement. The prospects of law
enforcement affecting the relative military balance among criminals against the tar-
geted DTO should be especially worrisome if the criminal group has low military and
recovery capability. The expectation of selective punishment might help the state to
maintain criminal violence at bay in contexts characterized by non-aggression pacts.
Panel (d) in Figure 2.7 depicts the scenario in which government authorities launch
a full-fledged campaign against all criminal organizations operating within the state’s
territory. This situation is likely to happen when there is a general decision from gov-
ernment authorities to end corrupt agreements and fight criminal groups. According
to the theoretical model, this kind of political motivation to enforce the law is likely
to emerge from democratization. As discussed in Section 2.3.2, improving democratic
conditions is likely to erode peaceful configurations between corrupt government of-
ficials and criminal organizations, thus favoring the collapse of the preexisting order
based on corruption. Improving democratic conditions increases the number of politi-
cal actors and fosters political competition for the provision of public goods, including
public security. Having a large number of political actors makes it more difficult for
politicians to bargain with criminals and solve problems of collective action. Even
if a corrupt pact is achieved, the diversity of party labels across levels of govern-
ment would break the chain of command, thus compromising the feasibility of the

84
agreement, and effective elections would reduce its duration. Democratization also
provides personal incentives for politicians to enforce the law as they seek to gain
citizen support. In particular, new politicians might enforce the law as an effort to
distinguish themselves from corrupt former officials. In addition, according to the
model, periods of political strain tend to increase the willingness of government au-
thorities to use harsh security policies as an effort to boost approval rankings and
increase their legitimacy.
Finally, Panel (e) in Figure 2.7 represents a situation of war of all against all. In
this bellum omnium contra omnes scenario, there is no order or peace. This situation
resembles a Hobbesian state of war where life is “poor, nasty, brutish, and short”
(Hobbes, 1651). According to the basic implications of the formal model, there are
certain conditions under which law enforcement is likely to generate criminal retal-
iation against the state and territorial conflict between two criminal organizations.
Panel (e) shows the extension of the implications of this argument to a generalized
and sustained crime-fighting effort by the state. If government authorities enforce
the law against several criminal organizations operating within their territory, these
punitive efforts are likely to trigger a massive wave of violence where the targeted
criminal groups retaliate against the state and fight each other in territorial conflicts.
Thus generalized law enforcement is likely to open several battle fronts at the same
time between the state and criminals and among competing criminal groups.
When the state tries to deliberately subvert the conditions that allow these crim-
inal organizations to exist and to extract valuable rents from illicit markets, a violent
reaction against the state can be expected. However, as discussed in Section 1.2, this
is an effort by criminal organizations to maintain the status quo for economic pur-
poses, not an attempt to overthrow the government for political reasons. In addition,
Panel (e) shows how generalized law enforcement is likely to trigger a massive wave of
violence between rival criminal organizations competing for valuable territories. The

85
disruptive effect of state action may affect the relative military balance across all
criminal organizations, thus generating multiple opportunities for criminal groups to
seize control over strategic areas using violence whether as instigators or defenders.
If the model is correct, we could expect to observe that law enforcement stimulates
more events of violent competition among rival criminal organizations than criminal
violence against the state. In addition, we should be able to observe higher levels of
violence in territories holding a larger number of criminal organizations, which are
likely to be areas of high strategic value for drug-related activities.
In the scenario depicted in Panel (e), no actor has enough military power to
monopolize the means of coercion. This implies that the state is not sufficiently strong
to be able to suppress all criminal organizations. In Hobbesian terms, the state is not
a Leviathan capable of imposing order by force on all the criminal groups. This also
implies that criminal organizations have some capability for using violence to either
fight the state or their rivals. However, it does not mean that all players (authorities or
criminals) have equal military capabilities. It merely assumes that actors have enough
military power to protect themselves or to employ violence against other actors if the
conditions are favorable and the situation demands it. As stated by Hobbes, in times
of war of all against all, actors have no other security than what their own strength
can provide them. In this sense, the term “war on drugs” may not be rhetoric, but
actually imply a generalized state of hostility between government authorities and
criminal groups, as well as violence between rival criminal organizations. This state of
violence of all against all in the context of the Mexican war on drugs is what motivates
the title of this dissertation, “Hobbes on Drugs: Understanding Drug Violence in
Mexico.”

86
2.5
Conclusion

Based on the analytical leverage of formal modeling, this research advances an

integrated explanation for the onset, escalation and concentration of organized crim-
inal violence. The model emphasizes both theoretical disaggregation and integration
to account for the dynamic interactions between the actors – between the state and
criminals and among rival criminal organizations – operating under the influence
of structural factors – such as changing democratic conditions and varying levels
of territorial value. In this way, the model allows for the interconnection between
micro-mechanisms of violence and structural factors. The general argument of the
model states that democratization motivates politicians to enforce the law against
criminals, thus triggering a wave of violence between the state and criminal organiza-
tions and between rival criminal groups, which tends to concentrate around strategic
territories.
The theoretical model helps create an understanding of how democratization dis-
rupts the peaceful configurations allowing coexistence between corrupt government
authorities and criminal organizations in contexts of low democratic development.
Democratization erodes these peaceful configurations and alters the system of polit-
ical incentives for government authorities, thus motivating politicians to fight crime.
In addition, the model incorporates the effect of periods of political strain that in-
crease the political benefits of government authorities using forceful security tactics
to fight crime.
Based on a contest success model for territorial control, the theoretical explanation
identifies the conditions under which increased levels of law enforcement are likely
to trigger an escalation of conflict between the state and criminal organizations, and
violence between rival criminal groups. In the model, the action of the state has
a disruptive effect on the relative military balance of criminal organizations. Law
enforcement weakens the capability of a criminal group to protect its territory, thus

87
motivating an invasion from a competing criminal group that now faces a weaker
rival. The equilibrium conditions of the model indicate that criminal violence is a
function of the severity of military damage, capability of recovering from damage,
and the value of the territory. Organized criminals are thus likely to use violence if
the net military strength recovered through fighting back when attacked is greater
than the attractiveness of engaging in a confrontation given the value of a territory.
The model uses a set of clearly defined propositions and certain assumptions to
build an integrative analytical framework to account for the onset, escalation and
concentration of organized criminal violence in Mexico. The basic propositions of the
model allow clear empirical implications to be derived, which will be evaluated in the
empirical chapters of this dissertation.

88
 CHAPTER
3

AUTOMATED
CODING
OF
EVENT
DATA

USING
EVENTUS
ID

3.1
Introduction

The previous chapter presented the formal model of how political factors and
structural determinants affect the violent interactions between the state and crim-
inal organizations, and conflict between rival criminal groups. In order to test the
empirical implications of a formal model emphasizing theoretical disaggregation, it is
necessary to have fine-grained evidence reflecting the dynamic interactions among dif-
ferent actors. Unfortunately, extant measures of drug-related violence in Mexico are
exclusively focused on counting homicides and are too approximate or aggregated
to provide sufficient analytical leverage to test the implications of the theoretical
model. In contrast to the usual body count approach, this research relies on event
data to analyze the actions and reactions between the different actors involved in the
Mexican war on drugs. Building this database of event data required bringing to-
gether cutting-edge advances in computer science with research on political violence.
The product of this multi-disciplinary approach is the development of Eventus ID,

89
a novel software for automated coding of event data from text written in Spanish.
Eventus ID is useful for extracting detailed data on who did what to whom, when
and where from news reports. This software allows fine-grained information to be
generated on the actions undertaken by government authorities against criminal or-
ganizations, and violent actions perpetrated by criminal groups against the state and
against other criminal organizations. In consequence, Eventus ID provides detailed
information that can be used to analyze the micro-dynamics of conflict stipulated in
the theoretical model.
This chapter is focused on describing the features and capabilities of Eventus ID.
In addition, it presents the coding strategy implemented in this research for building
the database on large-scale organized criminal violence that supports the empirical
chapters of this dissertation. In general terms, Eventus ID is an automated protocol
capable of reading news reports in order to identify event data from text written in
Spanish. Eventus ID relies on the core coding technique used in Tabari, another
automated protocol developed by (Schrodt, 2009) for coding event data from text
written in English. However, Tabari performed poorly when coding text written
in Spanish. Eventus ID provides several technological innovations that overcome the
limitations of Tabari. Eventus ID is the first software for event coding capable of
processing text written in Spanish. This is possible thanks to the development of
a more sophisticated and flexible coding algorithm that adapts to the grammatical
features of Spanish. In addition, Eventus ID has an ancillary application capable
of identifying the geographic location of events at municipal level, thus enabling the
possibility of georeferencing event data.
This chapter is structured in two parts. The first section locates the automated
coding protocol used in this dissertation within a broader context of databases of
event data in conflict research. This section also discusses the methodological advan-

90
tages and limitations of different computer-based coding protocols vis-à-vis coding
projects relying on manual annotation.
The second section presents Eventus ID, the automated textual annotation proto-
col used to build the database on large-scale organized criminal violence used in this
research. This segment provides a technical discussion of the five stages implemented
for generating the database. The first stage explains the strategy for gathering a
massive collection of news reports from 105 sources of information at the national
and local level. This section makes explicit the criteria for selecting news reports that
meet the inclusion requirements and explains how individual reports are reformatted
for Eventus ID. The second stage describes the characteristics of the actors and verb
dictionaries used by Eventus ID as searching categories for identifying events from
news reports. This section also provides a technical discussion of the algorithms that
enable Eventus ID to accurately identify events in the text. The flexibility of these
algorithms enable Eventus ID to adjust the coding protocol to the grammatical fea-
tures of Spanish. The third section illustrates the procedure for georeferencing events
already identified from news reports. The fourth section provides an assessment of
the accuracy of the automated coding protocol compared to manual annotation and
discusses some recoding rules to improve the coding precision. Finally, the last stage
discusses the characteristics of the validated database of event data.

3.2
Historical
and
Technological
Developments
of
"VUPNBUFE
$PEJOH

Automated
Coding
3.2.1
A
Brief
History
of
Event
Data

Political scientists have relied on event data to conduct quantitative analysis of

conflict in international relations and internal violence for several decades. In the
beginning, researchers relied on large teams of coders who would read newspapers or

91
other information sources in order to manually code event data. These projects were
highly demanding in terms of labor and financial resources. Eventually, high costs
made it difficult to update or expand most of those projects. With the advent of
artificial intelligence and increasing availability of information in the Internet, quan-
titative conflict research received a new impulse with machine-generated databases.
Scholars in political science took advantage of developments in a branch of computer
science called natural language processing and developed protocols for extracting
textual information from written sources and building new databases.
Some of the earliest projects to systematically study international crises began in
the 1960s. McClelland (1978) developed the World Event Interaction Survey (WEIS),
a coding scheme for studying conflict and cooperation between states in a political
crisis. WEIS used information from The New York Times and organized the flow of
action and response between countries into 22 broad categories comprising 63 types
of events ranging from diplomatic cooperation to conventional warfare. The data
and coding scheme of this seminal project have been broadly used by international
relations scholars for several decades (see Azar and Ben-Dak, 1975; Dixon, 1981;
Goldstein, 1992; Howell, 1983; McClelland and Hoggard, 1969; Tanter, 1972; Volgy
and Quistgard, 1974).
WEIS motivated other efforts to quantify conflict and cooperation in international
relations such as the Dimensionality of Nations Project (DON) which analyzed con-
flict interactions of 1,557 nation pairs between 1950 and 1965 (Rummel, 1966, 1976a,b,
1979). DON codes the actor and target involved in a conflict, the date of conflict, and
a dichotomous variable for the presence or absence of violent conflicts. Another study
was the Comparative Research on the Events of Nations (CREON) (Hermann, 1975;
Hermann et al., 1977, 1973). CREON contains conflict data for 36 countries between
1959 and 1968 from the Deadline Data on World Affairs, a publication containing in-
formation on historical, political and economic matters from 46 international sources.

92
This project provided a more detailed list of actors and sources, and categorized ac-
tion variables into verbal and nonverbal behavior, which can be further divided into
conflictual and cooperative actions.
Perhaps one of the largest projects influenced by WEIS is the Conflict and
Peace Data Bank (COPDAB) directed by (Azar, 1970, 1975, 1980) at the University
of North Carolina. This database contains international interactions and domes-
tic events for 135 countries from 1948 to 1978 coded from more than 70 sources.
COPDAB includes a detailed description of each event and records the date of the
episode, the actor initiating the event, the target of the action and other information
about the scale and severity of the event. COPDAB uses this information to create
16 ordinal categories of conflict and cooperation.
The enthusiasm for generating conflict databases in international relations ex-
tended to studies of intra-state violence. In the late 1960s, Arthur Banks launched
the Cross-National Time-Series Data Archive at the Center of Comparative Research
in State University of New York at Binghamton (Banks, 1971). The Bank’s dataset
contains conflict event data on a variety of actions such as general strikes, guerrilla
warfare, purges, riots, revolutions and anti-government demonstrations. In addition,
Gurr (1968) led a large data collection project named “A Causal Model of Civil
Strive” coding data for 114 nations between 1961 and 1965. This project included
data on conspiracy (e.g. internal war, turmoil and total strife), deprivation (e.g. eco-
nomic and political deprivation in the short and long term) and mediating variables
(e.g. legitimacy, coercion, institutions and previous levels of strife).
In general, the 1960s and 1970s saw the production of large event data sets of
international and domestic conflict. These quantitative efforts relied on a large num-
ber of coders to process vast amounts of information. Due to their nature, these
coding projects were highly intensive in terms of time and labor, and demanded
large amounts of financial resources. According to Schrodt (2006), the production of

93
large data sets slowed down as funding by the U.S. Department of Defense Advanced
Research Projects Agency (DARPA) was discontinued. With resources scarce, re-
searchers struggled to find the funding necessary for updating existing projects or
producing new databases that relied on legions of coders flipping newspaper pages.
In consequence, many of the data sets generated during these two decades were not
even updated due to the lack of funding.1
In the late 1980s, the generation of conflict databases received a new impulse
thanks to the support of the National Science Foundation (NSF) through the “Data
Development in International Relation” (DDIR) project (Merritt, Muncaster and
Zinne, 1993). In contrast to previous coding projects in the 1960s and 1970s, several
new projects began incorporating machine coding instead of human coders to reduce
the labor and financial costs of generating and updating data sets. One of these
developments was the Global Events Data Set (GEDS) project led by Davies (1998),
which incorporates computer-assisted coding to identify daily international and intra-
national events from Reuters reports.
At the end of the 1980s, Philip Schrodt started incorporating developments from
artificial intelligence for building data sets as an effort to overcome some of the
limitations and costs of generating databases using human coders. Schrodt received
funding from the NSF to develop the Kansas Event Data System (KEDS) (http:
//web.ku.edu/~keds/). KEDS is an automated protocol for coding news reports
to generate event data on international conflicts in the Middle East, the Balkans
and West Africa (Schrodt, Davis and Weddle, 1994). KEDS uses the lead sentence
of English-language news wires generated by the international news agency Reuters
and gathered through LexisNexis.

1
Arthur Banks was one of the few authors who managed to update his data set, doing so by
creating a private firm and selling the data. The Cross-National Time-Series Data Archive is current
up to 2011 and can be found at http://www.databanksinternational.com/

94
 KEDS represented a break with the past in the generation of databases of political
behavior because of its sparse parsing technique. Sparse parsing requires only some
parts of a sentence (subject, verb and object) in order to identify an event. These
elements are determined by dictionaries providing patterns of pronouns and verb
phrases. Because of this, sparse parsing only needs to focus on the basic subject-
verb-object structure to code an event. This strategy allows news reports to be coded
much faster than human coders with similar levels of accuracy can, thus substantially
reducing the costs of generating event databases (Schrodt, 2001; Schrodt, Davis and
Weddle, 1994).
KEDS strongly influenced other machine-coding projects such as the Protocol for
the Analysis of Nonviolent Direct Action (PANDA) led by Doug Bond at the Center
for International Affairs at Harvard (Bond et al., 1997). PANDA eventually turned
into a commercial risk management enterprise named Virtual Research Associates,
Inc. (VRA) (http://vranet.com/). The PANDA protocol also served as the basis
for the Integrated Data for Event Analysis (IDEA) (Bond et al., 2003).
Based on the experienced gained with KEDS over several years, Schrodt devel-
oped another software program called Textual Analysis by Augmented Replacement
Instructions (Tabari) (Schrodt, 2009). Tabari uses the same sparse parsing prin-
ciples originally developed for KEDS, but in contrast to its predecessor written in
Pascal, Tabari is written in C/C++ programming language, which makes it faster,
more efficient and readily usable in Macintosh and Linux environments. Data gen-
erated by Tabari is widely used for studying international conflict and the software
itself is used in other projects generating new datasets such as The Rice Events Data
Extractor (REDE) (Subramanian and Stoll, 2006a,b) and the Project Civil Strife
(PCS) (Shellman, 2008, 2013). Schrodt and his team later began working on the
Political Instability Task Force and eventually developed the Conflict and Mediation
Event Observations (CAMEO) scheme (Gerner et al., 2002).

95
 Several technological innovations such as the Internet enable massive amounts
of machine-readable text to be generated that greatly facilitate creating computer-
generated databases. Readily available information also helps to substantially reduce
the costs associated with launching new coding projects and updating existing ones.
According to Lazer et al. (2009), the emergence of computational social science is
based on the unprecedented capacity to collect and analyze data with an exceptional
breadth, depth and scale. Based on these technological possibilities, political science
is joining a recent surge on “big data” using vast amounts of data to analyze complex
behavior. The most important development made in this area is the Global Data on
Events, Location and Tone (GDELT) database developed by Leetaru and Schrodt
(2013). Based on the initial developments of CAMEO, GDELT relies on Tabari
to code event data from 1979 to the present covering all countries in the world to
generate more than 200-million political events. This unprecedented amount of data
created vivid enthusiasm in the academic community as “big data” opens new ways
to analyze political events.
Despite the boom in computerized coding, conflict scholars still rely on human
coding for studying conflict processes. For instance, the Uppsala Conflict Data Pro-
gram (UCDP) started by Wallensteen (1981) comprises several data sets on global
armed conflicts since the 1970s and is currently one of the most used databases in
conflict research http://www.pcr.uu.se/research/ucdp/. However, the generation
and sustainability of ambitious data projects such as UCDP are still highly dependent
on substantial financial resources. In particular, UCDP receives funds from Sida (the
Swedish International Development Cooperation Agency), the Bank of Sweden Ter-
centenary Foundation, the Swedish Research Council, Uppsala University and other
contributors.
Whether automated or human-based strategies are used for building or updating
databases, the central challenge of building databases of any sort is that raw infor-

96
mation must be organized and classified so that researchers can use it for analytical
purposes (Cardie and Wilkerson, 2008). As discussed in the following section, there
are trade-offs between manual and computer-assisted protocols. Manual approaches
can be challenging for large tasks of information gathering and data coding, espe-
cially when human and financial resources are scarce. Humans, although slower, are
substantially better than computers at comprehending abstract and complex infor-
mation. Automated methods, on the other hand, have the advantage of coding at
a greater speed than human coders. However, machine coding is usually not as ac-
curate as manual methods for coding complex information. In any case, researchers
should analyze which coding scheme represents a better strategy for the needs of any
specific project.

3.2.2
Trade-offs
Between
Manual
and
Machine
Generated
Event
Data

According to Cardie and Wilkerson (2008), textual annotation methods can be

categorized according to the extent to which humans intervene in the coding process.
Figure 3.1 shows how annotation strategies can be located in a continuum of textual
annotation methods. At one extreme, manual methods refer to coding schemes in
which humans perform all the activities associated with analyzing and classifying
raw information according to specific categories. At the other extreme, unsupervised
learning algorithms refer to computer-based protocols used for identifying patterns
in text without any human intervention. Between these two extremes, supervised
learning algorithms refer to coding protocols in which the computer uses annotation
instructions developed by the researcher to replicate human annotation decisions.
Human-based methods are the best for generating accurate interpretation of com-
plex texts. However, manual annotation is time and resource intensive. Baumgartner,
Jones and MacLeod (1998) describe the difficulties of building large data sets using

97
 A wide range of supervised learning algorithms are used for assigning texts to pre-
determined categories. This is why these coding schemes are the most common type
of content analysis used in political science (Grimmer and Stewart, 2013). Super-
vised methods usually require the researcher to develop dictionaries containing key
words that represent categories of interest. Automated-coding protocols use these
dictionaries as input provided by humans to classify text according to the set of
categories. The combination of dictionaries and the coding algorithm aims to repli-
cate the decision process that humans would use for classifying text according to the
predetermined categories.
Although machine coding offers some advantages over manual methods, comput-
erized annotation is not a silver bullet. Moreover, as mentioned by Grimmer and
Stewart (2013), the first principle of automated text analysis is that all quantitative
models of language are wrong, yet some are useful. A language is a system – a set of
ordered rules – that allows its users to structure symbols for reference or represen-
tation purposes. Different languages (e.g. English, Spanish, Chinese, mathematical,
chemical, gesture, chromatic, etc.) use different sets of rules and symbols to represent
their objects of interest. In natural language (phonetic and written), words constitute
these symbols of representation. Words are abstractions of the things they refer to.
These abstractions, represented by symbols, are the building blocks of language and
the key elements used for reasoning and knowledge. Language is thus not only useful
for communication purposes, it also constitutes an instrument. Natural language
is often highly complex, and computerized methods of textual analysis fall short in
accurately capturing the abstractions represented through language. However, de-
spite this limitation, automated-coding protocols can be highly valuable for specific
research objectives.
Based on the analysis made by Schrodt and Gerner (2012) on the advantages and
shortcomings of human and automated coding approaches, Table 3.1 compares the

99
trade-offs between manual and supervised methods across four main issues. The first
group refers to trade-offs concerning the characteristics of the coding project. Super-
vised textual annotation is better suited for processing large volumes of documents,
whereas human coding is more appropriate for small scale projects. Automated
coding has the advantage of allowing researchers to recode the same documents in
repeated coding periods. Supervised machine coding also allows researchers to eas-
ily modify or update dictionaries and recode the entire set of documents with the
new dictionaries. This also allows for easily updating or expanding the project by
processing new information. In contrast, coding projects relying on humans usually
carry out the coding stage only once because recoding would require substantial re-
sources in terms of time and labor. Sometimes researchers using manual methods
discover limitations or problems in their dictionaries when the coding project is at an
intermediate or advanced stage. In such cases, modifying the dictionaries for the rest
of the project would cause problems of internal consistency, but not modifying them
would mean carrying the dictionary problems or limitations through to the end of the
project. Another option would be to modify the dictionaries and restart the coding
from the beginning. However, as discussed in Section 3.2.1, scarcity of resources often
makes it impractical to recode or update projects using manual methods.

100
 TABLE 3.1

COMPARATIVE ADVANTAGES OF MANUAL AND SUPERVISED

ANNOTATION METHODS
Criteria Manual coding Supervised coding
Coding project
Volume of documents Small Large
Coding period Once Repeated or continuous
Recoding possibility No Yes
Updating possibility Difficult Easy
Dictionary modification Not recommended Easy
Content of interest
Coding unit Entire document Sentence or paragraph
Syntax characteristics Complex Simple
Content of interest Metaphoric or idiomatic Literal
Bias concerns
Sources of coder bias Several Unique
Concern of inter-coder reliability Considerable Non-existent
Bias caused by coder fatigue Considerable Non-existent
Possibility of information bias Considerable Minimum
Feasibility of the coding project
Coding time Slow Fast
Labor and financial demands High Low

The second group of trade-offs refers to the specific content of interest in a coding
project. Automated coding is more appropriate when the coding unit consists of sen-
tences or short paragraphs with simple syntax, and when the researcher is interested
in the literal content of the text. In contrast, manual methods are more appropriate
when researchers are interested in analyzing the overall content of entire documents.

101
In particular, human coding is more suitable when the coding process requires ana-
lytical abstraction from the text or when the researcher is interested in the figurative
or metaphorical use of language.
The third group refers to concerns of bias by coders or in the information sources
that may affect the quality of the database. According to Adcock and Collier (2001),
measurement validity is a central concern in both quantitative and qualitative re-
search and the application of coding protocols may sometimes yield to some incon-
sistencies. If these inconsistencies are non-systematic, then the measure may have
some problems of random error which could affect the reliability of the results but
are not likely to generate misleading results. However, if the error is systematic in a
specific direction or form, the measure may suffer problems of bias which could lead
to erroneous conclusions (Collier and Brady, 2004; Geddes, 2003; King, Keohane and
Verba, 1994). Coding projects based on manual methods usually require training
human coders to understand and apply the rules of the coding protocol. However,
as Tversky and Kahneman (1974) show, these efforts are not sufficient to guarantee
compliance with the coding rules since human coders often apply their own subjec-
tive assessment and heuristic principles to classify information. In consequence, as
the complexity of the coding scheme increases, interpretation and judgment become
more important (Harvey, 2008; Sipes, 1976).
If the project uses human coders, each individual person involved in the coding
process is a potential source of coder bias. There is a broad range of subjective
influences that can generate inconsistencies between coders. Unfortunately, despite a
few exceptions (Coppedge and Reinicke, 1990; Rohner and Katz, 1970), researchers
do not usually pay much attention to inter-rater reliability of coders in projects
based on manual coding methods. Systematic repetition of the same coding scheme
in automated methods eliminates the problem of inter-coder reliability; however, the

102
dictionary developed by the researcher remains a potential source of coder bias in
automated coding.
As noted by Baumgartner, Jones and MacLeod (1998), coder fatigue is an im-
portant source of bias in manual coding. Large projects usually involve teams of
coders slowly reading vast volumes of information and often individual motivation
and attention diminish while boredom increases over time. When coders are fatigued
they may simply skim through the information they are supposed to read carefully,
thus missing some important pieces of information and introducing type II error into
the database. Tired coders also tend to be less meticulous in the application of com-
plex coding rules, thus leading to more type I error. In machine-generated databases
tiredness and boredom are not a concern as the machine never tires.
Another source of bias may come from the information sources used in the coding
project. Due to limited resources and time constraints, researchers using manual
methods tend to rely on a handful of information sources. In conflict research, news-
papers remain the dominant source of information for studying violence.2 However,
as noted by Davenport and Ball (2002) and Davenport (2009a), different sources of
information may cover the same events from very different perspectives, thus gener-
ating important consequences for the inferences drawn from the evidence reported in
those sources. Automated coding reduces the effect of specific newspapers by simul-
taneously processing a large number of information sources. Increasing the number
of information sources reduces concerns of under-reporting due to coverage bias and
helps minimize ideological bias caused by specific political views.
Finally, the fourth group of trade-offs refers to the resources required by different
coding strategies. Manual coding usually requires a substantial investment in terms

2
For example, news report databases are prevalent in research on contentious politics (Hibbs,
1973; McAdam, 1982), as well as in social movements (Almeida, 2008; Tarrow, 1998; Trejo, 2009),
state repression and dissent (Davenport, 1995; Francisco, 1996; Moore, 1998), and international
conflict (Bond et al., 1997; Davis and Moore, 1995; Shellman, Hatfield and Mills, 2010).

103
of time, labor and financial resources. Unfortunately, research projects often face
significant constraints that have to be weighed when assessing the overall feasibility
of the project. Machine-based protocols offer an alternative that substantially reduces
the time and financial demands for some types of coding projects that may increase
the feasibility of research endeavors when there are budgetary constraints. However,
as reflected in Table 3.1, automated coding may not be the best strategy for all
projects and researchers have to carefully evaluate the trade-offs between manual
and machine-assisted methods.

3.3
Coding
Event
Data
Using
Eventus
ID

Initially, I attempted to use Tabari (Schrodt, 2009) for building the database of
events of drug-related violence in Mexico. As mentioned in Section 3.2.1, Tabari is
a widely used software for coding event data which has stimulated a broad spectrum
of research in international relations and intra-state conflict.3 Tabari has some nice
features that allow it to accurately code events from news reports written in English.
Most research projects using Tabari rely on headlines of news reports generated by
Reuters. These headlines are usually crafted by professional journalists who manage
to provide a succinct yet accurate description of an event in a well-written sentence.
Reuters headlines are the main coding unit for most projects using Tabari, which
facilitates the task of event coding using high quality text. In addition, English is
a highly structured language with clear sentence patterns and simple, yet general,
grammatical rules that enable clear, direct communication of ideas (Stockwell, Bowen
and Martin, 1965). As mentioned by Schrodt (2009), the quality of text is crucial for

3
For a non-exhaustive list of research using Tabari-generated data, see http://eventdata.
psu.edu/papers.html.

104
the accuracy of Tabari, and the carefully crafted Reuters reports written in English
are key elements for accurate performance of this software.
Unfortunately, several tests showed that Tabari performed poorly for coding
event data from text written in Spanish. Of course it is not reasonable to ask a
natural language processing protocol originally designed to code in one language to
perform well when processing information in a different language. There are two
elements that undermine the performance of Tabari for coding event data from
text in Spanish. First, the grammatical structure of Spanish is different than that
of English. In addition, journalists in the Mexican media tend to write sentences
in passive voice, which increases the complexity of the text. However, the limited
performance of Tabari was not only caused by the need to process text in a language
for which the software was not designed. The other limitation is directly related to
the lack of flexibility of Tabari’s coding algorithm. Tabari looks for three elements
in the text: source, action and target. In order to code an event, all three elements
have to appear in the text in that specific order. If one of those elements is missing
or the three of them do not appear in the required sequence, Tabari does not report
an event. Spanish word order is more flexible than English, and the subject of a
sentence can be unstated and implicit; therefore the rigidity of Tabari’s coding
algorithm generated substantial error and under-reporting when attempting to code
events from text in Spanish. The lack of software capable of accurately coding events
from text written in Spanish motivated the development of Eventus ID.4

4
The development of Eventus ID was possible thanks to Phillip Schrodt who kindly shared with
me the core event-coding algorithm of Tabari, which served as the cornerstone for independently
developing Eventus ID. The development of Eventus ID was possible thanks to the collaboration of
Alejandro Reyes, a computer scientist of extraordinary talent, who guided me through the fascinat-
ing discipline of natural language processing.

105
3.3.1
Eventus
ID
coding
process

Eventus ID is a supervised coding protocol for automated identification of event

data based on pattern recognition of text written in Spanish. The empirical evidence
that supports this research is produced using Eventus ID according to a coding scheme
consisting of five stages represented in Figure 3.2. The first stage refers to the process
of gathering information from news reports based on specific inclusion and exclusion
criteria; the compilation of individual files containing relevant reports; reformatting
the files for making them machine-readable; and compiling individual reports into a
general text corpus to be used as the input file for automated coding. The second
stage consists of running Eventus ID’s coding algorithm to identify event data. This
task requires the development of actor and verb dictionaries for identifying events in
the text. The output of this stage is a file containing event data. The third stage
refers to the process of georeferencing the data set. In this stage, Eventus ID takes
the database generated in the previous step and identifies the location of the events
mentioned in the original reports. This stage requires the development of a location
dictionary and produces a database of georeferenced event data. The fourth stage
consists of validating the data set by comparing a sample of the machine-generated
event data to a database manually coded by humans which serves as the “gold stan-
dard” for validation purposes. Validation often requires the modification of actor,
verb and location dictionaries as well as recoding of some events to improve the
accuracy of the machine-generated database. The validation and recoding process
is repeated several times to increase the convergence between human and machine
coding. Finally, the fifth stage generates and outputs a validated database of geo-
referenced event data. The following sections of this chapter describe the processes
implemented in each stage.

106
labor demands of manually coding newspaper databases, large coding efforts rarely
include subnational level sources.
Although news reports are a primary source of information for conflict databases,
the use of newspapers does not come without problems. According to (Davenport
and Ball, 2002), news coverage of contentious politics is affected by three elements:
event intensity, media location and media sensitivity. Media tends to devote more
attention to large, violent or bizarre events. Newspapers are also more likely to report
events occurring close to their location or within their area of coverage. In addition,
the political orientation of the newspaper also imprints a specific bias on the events
reported. These elements may combine to generate two main types of systematic
bias in newspaper-generated data. One source of bias is selection bias, which refers
to the media bias in determining what is published or not among the many possible
events they can observe and report. Another source refers to description bias in
the events they do select to report, which includes aspects such as emphasizing the
source’s political orientation, highlighting specific aspects of the event, and including
different subsets of the report.
Coverage bias is an important concern, but the magnitude of the problem is not
clear.6 Since there is no consensus on the magnitude of bias caused by differences
in media coverage, the most appropriate strategy is to take a conservative approach.
As Davenport (2009a) argues, probably the best way of minimizing the influence of
specific coverage bias from individual information sources is to build the data sets
using multiple sources.

6
Some argue that coverage among national newspapers is fairly similar. For example, Jenkins
and Perrow (1977) found little difference in how The New York Times, The Chicago Tribune and Los
Angeles Times covered farm worker insurgencies in California over a twenty-seven–year period. In
contrast, others argue that coverage differences are substantial between national and local newspa-
pers. Snyder and Kelly (1977) found large discrepancies in national and local newspaper accounts of
racial disturbances across 673 U.S. cities between 1965 and 1969, and nonviolent protests occurring
in 43 U.S. cities in 1968.

108
 To minimize concerns about coverage bias, this research relies on 105 information
sources that issued news reports written in Spanish between 2000 and 2010. Table 3.2
reports the main types of information sources, which includes four federal government
agencies, 32 local government agencies (one per each state), 11 national newspapers,
and 58 local newspapers (at least one per each state). This combination of official
and public sources at the national and local level minimizes the risk of coverage and
description bias in the database.

TABLE 3.2

NUMBER OF INFORMATION SOURCES

Source Type Number of sources
Federal government agencies 4
Local government agencies 32
National newspapers and magazines 11
Local newspapers 58
Total 105

Government information sources at the federal level include the Federal Security
Ministry, Secretarı́a de Seguridad Pública (SSP); the Army, Secretarı́a de la Defensa
Nacional (SEDENA); the Navy, Secretarı́a de Marina Armada de México (SEMAR);
and the Office of the Attorney General, Procuradurı́a General de la República (PGR).
Official sources at the local level came from the offices of State Attorney Generals,
Procuradurı́as de Justicia Estatales (PJEs) for each of the 32 states.7 Table 3.3

7
Press releases from the different government agencies can be found at the follow-
ing links; Federal Police (http://www.ssp.gob.mx/portalWebApp/wlp.c?__c=85c), the
Army (http://www.sedena.gob.mx/index.php/sala-de-prensa/comunicados-de-prensa

109
shows that, with the exception of the Federal Police and the Navy, all other sources
of information have available information throughout the research period from 2000
to 2010, thus minimizing concerns of temporal bias in news coverage.8

TABLE 3.3

AVAILABILITY OF INFORMATION SOURCES

Source 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Police
Navy
Attorney General
State Attorneys
Army
National press
Local press
Dark gray is for gathered reports and light gray is for not available reports.

The list of national newspapers includes Servicio Universal de Noticias (also

known as El Universal ), El Economista, El Financiero, Excélsior, Notimex–Nacional
(the state news agency), Reforma, La Jornada, El Sol de México, Milenio Diario,
Revista Proceso, and La Crónica de Hoy. It is known that different national level

for central headquarters and http://www.sedena.gob.mx/index.php/sala-de-prensa/

comunicados-de-prensa-de-los-mandos-territoriales for the different territorial divi-
sions), the Navy (http://www.semar.gob.mx/s/sala-prensa/comunicados-2013.html), Office
of the Attorney General(http://www.pgr.gob.mx/Prensa/prensa.asp), and the offices of State
Attorney Generals (http://www.pgr.gob.mx/prensa/prensa.asp?tRcd=2).
8
I issued a couple information access requests asking the Police (request number 0002200060811)
and the Navy (request number 0001300022711) to give me copy of their press releases between 2000
and 2006. However, these to agencies responded that all the press releases for those years have been
destroyed and are no longer available. In Mexico there is a Bill allowing government agencies to
destroy all official information, archives and records older than five years (see Diario Oficial de la
Federación, 1998).

110
newspapers in Mexico have different coverage and ideological orientations. For exam-
ple, Reforma has better coverage of the north of the country and is usually considered
to be a conservative newspaper. In contrast, La Jornada has better coverage of the
south of Mexico and often takes a left-wing view in its reports. Having several na-
tional newspapers reduces the coverage and ideological limitations of each individual
source.
Although illustrative, national newspapers may not tell the full story. News that
are important at the local level often do not find their way up to national newspa-
pers. Space limitations and editorial decisions often prevent a large number of local
news stories from appearing in national newspapers. In order to minimize problems
of media under-reporting between the national and local levels, this research also
collected data from 58 local newspapers. Appendix A.2 presents the number of infor-
mation sources per year and shows that there is at least one local newspaper per state
in the entire research period. In addition, Appendix A.3 reports the complete list
of sources. This research paid special care to attempting to minimize coverage and
under-reporting of events by using a large number of official and private sources. Such
a large number of information sources may raise concerns of multiple counting when
more than one newspaper reports the same event. For this reason, as discussed in
Section 3.7.3, a procedure for detecting and eliminating duplicates was implemented.

3.4.1.1 Manual gathering and selection criteria

Information gathering consists of a hybrid strategy combining information tech-

nologies for retrieving reports and a team of research assistants for identifying relevant
news reports from the large number of information sources.
For national and local newspapers, the information-gathering strategy uses In-
folatina (http://www.infolatina.com.mx/), a system providing access to a large
collection of news sources in Mexico. A query entered into Infolatina’s search en-

111
gine is used for systematically searching within the large number of newspapers for
any reports associated with drug violence. Appendix A.1 shows the query used in
Infolatina’s system.9 The result of the search is a list of news reports that meet the
criteria specified in the query. The list includes the headline of each report and a link
to its entire content. After Infolatina has presented the results of the search, a team
of human coders reads all the headlines and selects the set of news reports according
to a specific selection criteria.
Infolatina was only used for identifying reports from newspapers. However, this
collection of sources does not include press releases from government agencies. To
overcome this limitation, the team of human coders searchesd the websites of the
government agencies associated with the war on drugs and identified any press releases
relevant to this study. In most cases, state agencies make their press releases available
online, especially those issued since 2007. Press releases issued by the Army before
December 2006 are not available on the Internet and were obtained in hard copy.
Research assistants reviewed the Army press releases and selected those relevant to
this research.
To select the appropriate news reports, research assistants were trained on the
key conceptual definitions guiding this research that are mentioned in sections 1.2
and 1.3. In addition, they were instructed to apply the following criteria for inclusion
and exclusion of news reports:

Inclusion criteria: The main objective of the selection criteria is to select reports
providing information about events of organized criminal violence. The instructions
for selection were:

9
The query consists of two parts. The first section, presented in typewriter font in Ap-
pendix A.1, refers to all the words used in the search for selecting news reports that contain any of
the specified words. However, violence tends to receive intense coverage from the media and often
generates a large volume of reports that are not directly relevant to this research. For that reason,
the second part of the query refers to the exclusion criteria which is presented in typewriter font
in Italics in the Appendix. The elements of this exclusion criteria are used for filtering out
reports that are not directly relevant to the objectives of the research.

112
1. Include reports of events associated with violent actions such as armed clashes,
murders, killings, shootings, ambushes, attacks, assassination attempts, wound-
ing, kidnapping, torture or mutilation that involve the participation of presumed
members of criminal organizations as perpetrators or victims.

2. Some reports of violent actions may not explicitly mention the participation of
organized criminals as perpetrators or victims, but these events should be included
if their modus operandi involves one or more of the following characteristics:10

• Use of assault weapons.

• Two or more victims.
• Signs of mutilation or torture in the victims.
• Execution-style killings (coup de grâce, known in Spanish as “tiro de gracia”).
• Participation of at least one group of armed men (commandos or death
squads).
• Participants traveling in convoys of vehicles (usually SUVs, pick-up trucks or
luxury vehicles).
• Presence of signs, marks or messages (“narcomensajes”) associated with or-
ganized crime.
• Bodies discovered in clandestine mass graves, wrapped in blankets (“ensara-
pados”), found inside abandoned vehicles (“encajuelados”) or in containers
(“entambados”).

3. Include reports of events associated with kidnapping, extortion or money launder-

ing even though organized crime or drug trafficking organizations are not explicitly
mentioned in the report.

4. Include reports of events associated with law enforcement actions by state security
forces when conducting operations against criminal organizations. Law enforce-
ment actions can be violent or non-violent:

• Violent law enforcement refers to events in which the state’s coercive appara-
tus uses force to deliberately inflict physical damage on suspected members
of criminal organizations. These actions may include events in which the
state attacked suspected organized criminals or repelled a criminal act of ag-
gression, or events in which security forces wounded or killed one or more
suspected organized criminals.
• Non-violent law enforcement refers to state actions that resulted in the arrest
of suspected members of criminal organizations or the confiscation of drugs,

10
This set of characteristics is based on the criteria used by the Mexican government for classifying
a homicide as being presumed to be associated with organized crime. See Sistema Nacional de
Seguridad Pública (2011b).

113
 weapons or assets (e.g. money, real estate, vehicles, items) used for their
illegal activities.

The elements of the inclusion criteria provided the main guidelines for selecting
relevant reports associated with drug violence. However, the exclusion criteria was
equally important to guarantee the quality of information included in the database.

Exclusion criteria: This research is exclusively focused on events associated with

organized crime violence. In consequence, the process of information gathering should
exclude reports containing the following kind of information:

1. Exclude reports about speeches, declarations, discourses, opinions, newspaper ed-

itorials or public statements made about events associated with organized crime
violence.

• This criterion is crucial. Drug violence receives a lot of attention from the
media and lots of people talk about it. However, this research is strictly
focused on events (facts, episodes, things that happened), not what people
have said about those events.
2. Exclude reports associated with deaths, injuries or material damage caused by any
of the following:

• Attacks by insurgent groups (e.g. Ejercito Zapatista de Liberación Nacional

(EZLN) or Ejercito Popular Revolucionario (EPR)) or any other radical
group with political motivations.
• Accidents, natural disasters, diseases or attacks by animals.
• Crimes of passion.
• Street level crime and minor offenses (e.g. burglary, theft, simple assault,
robbery).
• Violent criminal behavior that does not have the characteristics mentioned
in the inclusion criteria (e.g. a police officer shot during a robbery of a con-
venience store).
• Events of organized crime violence occurring outside Mexico (e.g. clashes
between Zetas and local drug dealers in Guatemala).
• Violence associated with protests, riots or contentious tactics undertaken by
groups with social, political, economic or environmental demands against the
government.
3. Exclude government reports summarizing the activities conducted or results achieved
over aggregated periods of time (e.g. monthly or annual reports).

114
 Systematic application of the inclusion and exclusion criteria help identifying valid
events of organized criminal violence that correspond to the conceptual definitions
presented in Section 1.3 of Chapter 1. These news reports constitute the raw ma-
terial for building the database and are crucial for validation of the coding protocol
discussed in Section 3.7.1.

3.4.2
Step
1.b.
Collecting
Individual
Input
Files

Application of the selection criteria identified 41,838 reports of events associated

with organized crime violence. Table 3.4 presents the distribution of the number of
reports across different sources of information. The largest number of news reports
were from the offices of State Attorney Generals, national news sources and local
newspapers. These three sources of information provided 80.05% of all reports.

TABLE 3.4

NUMBER OF NEWS REPORTS

Source type Num. of documents Percentage by type Gathering goal
Police 1,572 3.76% 100%
Navy 421 1.01% 100%
Attorney General 3,328 7.95% 100%
State Attorney General 12,728 30.424% 100%
Army 3,026 7.23% 100%
National press 10,849 25.93% 100%
Local press 9,914 23.7% 100%
Total 41,838 100%

115
 The content of all individual reports were extracted from their original formats
and stored in plain text files. Due to the diversity of information sources, original files
come in a variety of formats such as Hypertext Markup Language (.html), Microsoft
Word (.doc) and Portable Document Format (.pdf). The set of press releases issued
by the Army which were available in hard copy were scanned and converted to .pdf
format. The content of individual .pdf files was converted to plain text format (.txt)
using the ReadIris 12 Optical Character Recognition (OCR) program.11 After being
processed, all .html, .doc and .pdf files were saved in plain text .txt with UTF-8
econding.
The process of transforming different file formats into plain text files includes
assigning a unique name to each file. The nomenclature standard consists on three
elements: the date of the event; an acronym indicating the type of information source;
and a counter for unique identification. The name of each file is structured as:

YYYYMMDDccc SRC.txt

The first eight digits (YYYYMMDD) represent the date of the report in year (YYYY),
month (DD), day (DD) order. The second element, (ccc), is a counter starting from
001 up to 999 to distinguish different reports issued by the same source on the same
day. The third element, (SRC), represents the acronym identifying the information
source. In this way, the nomenclature allows news reports to be identified uniquely,
a key element for finding and eliminating duplicates (see Section 3.7.3).

3.4.3
Step
1.c.
Eventus
ID
Input
Format

Lessons from natural language processing of event data show that it is easier to
process text using smaller coding units such as sentences or paragraphs rather than
entire reports. The idea is straightforward: simpler coding units reduce the amount of

11
The software and its documentation can be found at http://www.irislink.com/.

116
error. For this reason, the basic coding units analyzed by Eventus ID are paragraphs.
In general, Eventus ID is capable of processing any UTF-8 plain text file with the
following structure:

date doc id
text line 1 no longer than 80 characters
text line 2 no longer than 80 characters
..
.
text line n no longer than 80 characters
blank line

However, news reports do not originally come in this format. In order to make
news reports readable for Eventus ID at the paragraph level, it is necessary to use
Doc2Eventus.pl. This is an ancillary software developed for this research that auto-
matically reformats the content of individual news reports into Eventus ID–readable
text. The program breaks the entire content of a report into paragraphs and creates
a unique identifier for each paragraph by executing the following tasks:

1. Take the entire content of each report and break it into individual paragraphs.

• The content of the paragraph must be broken down into 80-character lines.

2. Extract the date from the file name and write it to the top of each paragraph.

3. Write the rest of the file name next to the date at the top of each paragraph.

4. Add a sequential counter for each paragraph of the report next to the file name
at the top of each paragraph.

5. Add a global sequential counter for all paragraphs of all the reports already pro-
cessed.

6. After the end of the paragraph add a blank line.

For example, consider a sample press release named 20100908001 SEDENA 001.txt
which was issued by the Army on September 8, 2010:

117
 20100908001 SEDENA 001.txt
Lomas de Sotelo, D.F. 8 Sep. 2010.
Ejercito mexicano asegura mas de dos toneladas de mariguana en
Sonora.
La Secretaria de la Defensa Nacional, informa a la opinion
publica, que dentro del marco del combate integral del estado
mexicano contra el narcotrafico y delincuencia organizada,
tropas jurisdiccionadas a la 45/a. Zona Militar, establecidas
en el municipio de San Luis Rio Colorado, Son., aseguraron 227
paquetes de mariguana con un peso total de dos toneladas 250
kilogramos.
Este aseguramiento se logro durante la revision de un tracto
camion que remolcaba un tanque que contenia melaza, entre la
cual se localizaron los paquetes del citado enervante; por lo
que el personal castrense procedio a la detencion del conductor,
asi como al aseguramiento de la droga y el automotor, mismos que
fueron puestos a disposicion de la autoridad correspondiente.

Doc2Eventus.pl separates the content of the document into individual paragraphs

and assigns a unique identifier to each paragraph. The identifier beginning with
the date of the event (which is extracted from the filename assigned in Step 1.b),
followed by the name of the file, and ends with a counter for the paragraph within
the document and a global counter for all paragraphs already processed from other
reports. The Eventus ID–readable output from the document shown above is:

118
 100908 20100908001 SEDENA P0 P631.
Lomas de Sotelo, D.F. 8 Sep. 2010.

100908 20100908001 SEDENA P1 P632.

Ejercito mexicano asegura mas de dos toneladas de mariguana en
Sonora.

100908 20100908001 SEDENA P2 P633.

La Secretaria de la Defensa Nacional, informa a la opinion
publica, que dentro del marco del combate integral del estado
mexicano contra el narcotrafico y delincuencia organizada,
tropas jurisdiccionadas a la 45/a. Zona Militar, establecidas
en el municipio de San Luis Rio Colorado, Son., aseguraron 227
paquetes de mariguana con un peso total de dos toneladas 250
kilogramos.

100908 20100908001 SEDENA P3 P634.

Este aseguramiento se logro durante la revision de un tracto
camion que remolcaba un tanque que contenia melaza, entre la
cual se localizaron los paquetes del citado enervante; por lo
que el personal castrense procedio a la detencion del conductor,
asi como al aseguramiento de la droga y el automotor, mismos que
fueron puestos a disposicion de la autoridad correspondiente.

In this example, the nomenclature produced by Doc2Eventus.pl enables the date

to be identified in every paragraph, as 100908. It also identifies each paragraph with
the document 20100908001 SEDENA from which they were all extracted. In addition,
there is an identifier for each paragraph in the document, denoted by the first counter
P0-P3, and a global identifier for each paragraph in the complete corpus, which in
this example ranges from P631 to P634.

119
3.4.4
Step
1.d.
Corpus
of
Text

All 41,838 documents were processed with Doc2Eventus.pl, breaking them into
paragraphs and assigning unique identifiers. All files were grouped according to the
information source from which they were extracted. All the documents of each group
were then compiled into one large text file. These files constitute the text corpus used
as input information for Eventus ID.

TABLE 3.5

CORPUS OF TEXT FILES BY INFORMATION SOURCE

Source type Document size in KB Number of text lines in each file
Police 5,256 134,200
Navy 1,001 22,929
Attorney General 8,299 213,388
State Attorney General 21,310 575,129
Army 7,236 168579
National press 26,432 789,020
Local press 24,517 791,784
Total 94,051 2,695,029

3.5
Stage
2.
Event
Coding

Eventus ID relies on pattern recognition to identify events from text. The intuitive
concept of an event is that it provides information on someone doing something to
someone else. Events are composed of three key elements:

120
Source: Refers to the actor or perpetrator of the action. Actors are identified by
Eventus ID as proper nouns in the text.

Action: Indicates the specific action carried out by the source. Actions are identified
by the system as verb phrases in the text.

Target: Refers to the actor towards or upon whom the perpetrator carried out an
action.

Eventus ID uses large dictionaries of proper nouns and verbs for identifying events
in the corpus. While reading the text, Eventus ID uses the categories provided by
the dictionaries to recognize actors and actions. Once these elements are detected,
Eventus ID puts the textual information from the elements in numeric format into
a database. Identifying actors and verbs in the text enables information on who
(source) did what (action) to whom (target) to be extracted. This section describes
the development of actor and verb dictionaries used to identify the key components
of an event. It also describes the coding algorithm used by Eventus ID for event
coding. Finally, it provides an example of the output database.

3.5.1
Step
2.a.
Actor
Dictionary

Eventus ID uses the actor dictionary to identify both the source and target of an
event. The actor dictionary consists of a list of proper nouns related to perpetrators
and victims of various types of organized crime violence. This dictionary contains
2,277 actors grouped into nine categories presented in Table 3.6. These categories
enable identification of a wide variety of political actors and security forces at federal
and local levels, members of different criminal organizations, victims, and a wide
variety of criminal assets, drugs and weapons. As reflected in Table 3.6, each category
is assigned a specific code number. These categories are further divided into sub-
types with their corresponding sub-codes. Disaggregating groups in this way provides
the system with a detailed coding criterion while maintaining consistency within
categories.

121
 TABLE 3.6

ACTORS CATEGORIES
Actor category Main group code
Federal government 1
Coercive apparatus 2
Local government 3
Individuals 4
Victims 5
Criminal organizations 6
Criminal assets 7
Drugs 8
Weapons 9

Each actor in the dictionary is associated with a numeric code that corresponds
to the actor’s main group and subgroup. Words in the dictionary are separated by
an underscore “ ” to help Eventus ID searching for the words in the text12 . The
following list presents an example of the actor dictionary:

army troops [202051]

police officer [204021]
member of a criminal organization [601060]
cocaine [801022]
AK 47 [901013]

Developing the actor and verbs dictionaries required a gradual process of learn-
ing, refinement, knowledge accumulation, detailed reading and, most importantly,
feedback from the validation process detailed in Section 3.7.1. This iterative pro-
cess of coding and validation enabled the dictionaries to be fine-tuned by adding

12
Eventus ID does not require words in the source text to be separated by an underscore. The
software “reads” the corpus containing words separated by blank spaces as in any regular text, but
it uses the concatenated words of the dictionaries to identify patterns in the corpus.

122
actors and verbs or modifying existing ones. Although the actor dictionary already
includes an exhaustive list of 2,277 actors, this list may not be “perfect” in the sense
of including “absolutely all” possible actors. However, as discussed in Section 3.2.2,
supervised learning the dictionaries to be updated and further refined to improve
coding accuracy.

3.5.2
Step
2.b.
Verb
Dictionary

The verb dictionary consists of a list of 1,755 verb phrases referring to a set of
violent and non-violent actions grouped into eleven categories. Table 3.7 presents the
list of categories used to classify the actions analyzed in this research.

TABLE 3.7

VERB CATEGORIES
Violent actions Non-violent actions
Action category Group code Action category Group code
Attack 101 Detect 201
Shoot 102 Seize 202
Clash 103 Eradicate 203
Arrest 104
Kidnap 105
Raid 107
Burn 108
Wound 177
Torture and mutilate 188
Kill 199

123
 As presented below, the verb dictionary consists of a list of verbs followed by a
numeric code for its corresponding action category. As shown in the example, the
verbs “attack” and “attacked” have the same code 88101. Some verbs are followed
by a set of associated words that refine the meaning of the verb in its context. In
these cases, the “*” indicates where the verb itself should appear in the phrase. In
the example below, Eventus ID automatically inserts the verb `‘textttattacked” into
the item were * and looks for the verb phrase “were attacked” in the text. The
code for “attack” is slightly different than for “were attacked.” The root code is
the same in both cases (101) but they differ in the prefix (88) for the former and (99)
for the latter. The prefix (99) indicates a verb conjugated in passive voice and serves
as a hint for the recoding process. This is important for coding in Spanish because,
in contrast to writing recommendations in English, the use of passive voice is highly
common in journalistic reports written in Spanish. The following list is an example
of the verb dictionary:

attack [88101]
attacked [88101]
- were * [99101]
- was * [99101]
arrest [88104]
- under * [88104]
fight [88101]
- strengthen the * against [- - -]

As mentioned earlier, Eventus ID uses a pattern recognition coding scheme sim-

ilar to that implemented by Tabari. However, a key difference between these two
protocols lies in the way they handle verb phrases. Tabari is designed for coding
in English, whose grammatical rules allow for simple general ways of combining the
subject, verb and object in a sentence. For example, most regular verbs in English
are easily conjugated by adding “s,” “ed,” or “ing” at the end of the infinitive verb
form. Another characteristic of English is that the gender and number of the noun
generally do not affect the conjugation of verbs. This means that “you,” “he,” “she,”

124
“we” or “they” can be combined with the different verb forms almost indistinctly.
Based on these simple general grammatical rules, Tabari incorporates a stemming
algorithm to automatically identify all the different forms from a verb stem. Using
the verb “to arrest” as an example, Table 3.8 shows how Tabari’s stemming algo-
rithm can easily identify a variety of verb tenses. In this case, the verb dictionary
would only require the stem “arrest” and the software takes care of detecting the
different verb forms by adding “s,” “ed,” or “ing” at the end of the stem.
Although Tabari’s stemming facility is very convenient for coding in English,
this feature caused a substantial amount of error when coding in Spanish. Using an
English-based stemming algorithm is not appropriate for event coding in Spanish be-
cause verb tenses in the latter do not end with “s,” “ed” or “ing.” Verb conjugation in
Spanish is much more varied, and using an English-based stemming algorithm would
require more than simply “tweaking” the algorithm. Table 3.8 shows that the ending
part of the verb “arrestar ” (to arrest) is very different across the various combinations
of verb tenses, number and gender. Given the complexity of conjugation in Spanish,
Eventus ID does not include a stemming algorithm. The “shortcuts” that might be
useful for coding in English would be counterproductive in Spanish. Unfortunately,
reducing the propensity to generate error through a stemming process comes at a
cost. Eventus ID demands the development of large, detailed verb dictionaries, thus
putting a heavier burden on the researcher to develop detailed dictionaries.
Some items in the verb and actor dictionaries are coded as [- - -]. This is
an instruction for Eventus ID to ignore that word in order to avoid confusion. For
example, the verb “fight” is coded as [88101] because it refers to a specific action.
However, the phrase “strengthen the fight against” is identified as a common
closing formula in government press releases, which has a metaphoric meaning but
does not refer to a specific event. In such cases, the code [- - -] indicates Eventus
ID that the phrase should be ignored.

125
 TABLE 3.8

VERB TENSES IN ENGLISH AND SPANISH

Indicative Subjuntive Past
Person Present Past Present Imperfect Gerund passive voice
English
I arrest arrested arrest arrested arresting was arrested
You arrest arrested arrest arrested arresting were arrested
He, she arrests arrested arrests arrested arresting were arrested
We arrest arrested arrest arrested arresting were arrested
You arrest arrested arrest arrested arresting were arrested

126
They arrest arrested arrest arrested arresting were arrested
Spanish
Yo arresto arresté arreste arrestara o arrestase arrestando fui arrestado
Tú arrestas arrestaste arrestes arrestaras o arrestases arrestando fuiste arrestado
Ella, él, usted arresta arrestó arreste arrestara o arrestase arrestando fue arrestada, fue
arrestado
Nosotros arrestamos arrestamos arrestemos arrestáramos o arrestásemos arrestando fuimos arrestados
Vosotros arrestáis arrestasteis arrestéis arrestarais o arrestaseis arrestando fuisteis arrestados
Ellas, ellos, uds. arrestan arrestaron arresten arrestaran o arrestasen arrestando fueron arrestadas,
fueron arrestados
Vos arrestás arrestaste arrestes arrestaras o arrestases arrestando fuisteis arrestado
3.5.3
Step
2.c.
Event
Coding
Using
Eventus
ID

Eventus ID requires three input files for event coding:

1. Text corpus: source of the text to be analyzed.

2. Actor dictionary: provides the list of actors and their numeric codes.

3. Verb dictionary: provides the list of verbs and their codes.

In order to code events from the source text, Eventus ID uses two pattern recogni-
tion algorithms: the general sequence algorithm which focuses on the source–action–
target structure and the partial sequence algorithm which focuses on the verb–target
structure. Both algorithms use the principles of the sparse parsing technique origi-
nally developed by Shrodt in KEDS and later refined in Tabari. The sparse parsing
method uses the actor and verb dictionaries as searching criteria to identify only the
relevant parts of the text that correspond to an event, while the rest is ignored for
coding purposes.
Both coding algorithms first identify the date of the event (date) and the docu-
ment identification label (docid) from the top of each paragraph (the input format
for Eventus ID is discussed in Section 3.4.3). Each algorithm then uses its own
scheme to recognize the elements of the event contained in the paragraph. Table 3.9
describes the steps undertaken by each algorithm for event coding. Eventus ID then
saves the outcome of the coding process in a plain text file. Each line of the outcome
file contains the set elements corresponding to each coded event, separating those
elements by tabs and ordering them as follows:

date doc id [actor1] [verb] [actor2]

127
 TABLE 3.9

EVENTUS ID CODING ALGORITHMS

General sequence algorithm Partial sequence algorithm
1) Search for the actor 1) Search for the verb
- Load the actor dictionary - Load the verb dictionary
- Start reading the text - Start reading the text
- Search for the longest actor first - Search for the longest verb first
- When an actor is found, store as actor1 - When a verb is found, store as verb
- Pause the search where the actor is - Pause the search where the verb is
found found
2) Search for the verb 2) Search for the actor
- Load the verb dictionary - Load the actor dictionary
- Resume reading from previous pause - Resume reading from previous pause
- Keep reading the text - Keep reading the text
- Search for the longest verb first - Search for the longest actor first
- When a verb is found, store as verb - When an actor is found, store as actor2
- Pause the search where the verb is - Pause the search where the actor is
found found
- If no verb is found, go to Step 4 - If no actor is found, go to Step 3
3) Search for the actor 3) Save the event
- Reload the actor dictionary - Save [- - -] [verb] [actor2] in
database
- Resume reading from previous pause If no actor is found, save the event as [-
- -] [verb] [- - -] in the database
- Keep reading the text - Start again from Step 1
- Search for the longest actor first
- When an actor is found, store as actor2
- Pause the search where the actor is
found
4) Save the event
- Save [actor1] [verb] [actor2] in
database
- If no verb is found, save the event
as [actor1] [- - -] [- - -] in the
database
- Start again from Step 1

128
3.5.3.1 General sequence algorithm

The general sequence algorithm is useful for identifying events that follow the
source–action–target structure. In order to code an event with this algorithm, all
three elements must appear in a sentence in the required order. Consider the following
sentence:

Army troops arrested a member of a criminal group.

In this example, all the three elements of the event are present in the sentence in
the required order. In consequence, the general sequence algorithm identifies “Army
troops” as the source, ““arrested” as the action and “member of a criminal
group” as the target. Eventus ID then codes the event in numeric format in the
database as:

[202051] [88104] [601060]

Since sparse parsing only focuses on the relevant parts of the text based on the
words provided by the actor and verb dictionaries, the text could be more verbose
without affecting the result of the coding. Consider the following sentence:

In a press release, the Mexican government informed the public that

troops posted in the municipality of San Luis Rio Colorado, Son. seized
227 packages of marijuana with a total weight of two tonnes, 250 kilograms
while patrolling rural roads in the area.

Despite the wordiness of the paragraph, sparse parsing allows Eventus ID to

recognize the key components of the event and code the “troops” as the source,
“seized” as the action and “packages of marijuana” as the target.
As mentioned above, journalists in Mexico often use passive voice to write their
news reports. Passive voice increases the grammatical complexity of a sentence by
inverting the order of the subject and object. Consider the following line:

A member of a criminal organization was arrested by Army troops.

129
 In this example, all the three elements of the event are present in the sentence.
However the subject and object appear in reverse order. According to the general
sequence algorithm, Eventus ID identifies “member of a criminal group” as the
first actor, the verb phrase “was arrested” as the action, and “Army troops” as
the second actor. The output of this event after coding is:

[601060] [99104] [202051]

However, this could be interpreted as “a member of a criminal group arrested

elements of the Army,” which does not correspond to the idea presented in the text.
As discussed in Section 3.5.2, the verb dictionary adds prefix [99] to codes corre-
sponding to verb tenses in passive voice. Later, the recoding process (discussed in
Section 3.7.2) uses this prefix as a cue to correct the erroneous coding directionality
caused by use of the passive voice. In this way, dictionary development, the cod-
ing algorithm and the recoding scheme work together to disentangle more complex
grammatical structures and reduce coding error in the database. When the recoding
rules have been applied, the event is correctly coded as:

[202051] [88104] [601060]

As indicated in Table 3.9, the general sequence algorithm begins by searching for
the first actor in the sentence. Once it is found, it switches to the verb dictionary
and looks for the action. However, sometimes news reports mention a series of items
which are not followed by a verb. This particularly common in government press
releases including a list of items. For example consider the following paragraph:

Army troops arrested a member of a criminal group. Troops seized 6

kilograms of cocaine, 372 packages of Clindamycin phosphate, two AK-47s
and an R-15 assault rifle with the respective ammunition.

Applying the general coding algorithm, Eventus ID recognizes “Army troops,”

“arrested” and “member of a criminal group” as the first set of source, action

130
and target. Next, it identifies “troops” “seized” “cocaine” as the second set of
source, action and target in the paragraph. Then, continues reading and identifies
“Clindamycin phosphate,” “AK-47,” “R-15 assault rifle” and “ammunition” as
independent items not followed by a corresponding action. The output of this para-
graph is:

[202051] [88104] [601060]

[202051] [88202] [801022]
[604021] [- - -] [- - -]
[901013] [- - -] [- - -]
[901014] [- - -] [- - -]
[901021] [- - -] [- - -]
As discussed in Section 3.7.2, the recoding scheme uses the information from the
complete coding lines to rearrange and fill in the information in the incomplete coding
lines. In this way, Eventus ID allows more detailed information to be extracted from
complex events, thus better reflecting the multidimensionality of conflict.

3.5.3.2 Partial sequence algorithm

The partial sequence algorithm is useful for identifying more complex grammatical
structures such as sentences using present indicative tenses. In general, the present
indicative is used similarly in English and in Spanish. However, the key difference
is that in English the present progressive is more commonly used than the present
indicative, while Spanish uses the present indicative tense. In English, the present
progressive tense is a finite form of the verb that has the mood, tense, and person
clearly defined. For example, in the sentence “they are arresting a criminal” the verb
“to arrest” is conjugated in the indicative mood, present progressive tense, third
person plural.
The present progressive sentence that literally corresponds to this example in
Spanish is “ellos están arrestando a un criminal ”. However, in Spanish one would
simply use the present indicative tense; thus the sentence would read “arrestan a

131
un criminal.” As shown in Table 3.8, this conjugation of the verb “arrestar ” (to
arrest) corresponds to the third person of the present indicative. What makes present
indicative in Spanish more complex is that the conjugation already gives information
about the person as part of the verb, and in consequence, the subject of the action
is often omitted from the sentence. To make things even more complex, the present
indicative is often used for referring to events that occurred in the past (historical
present). Thus while the sentence “arrestan a un criminal ” might refer to an action
carried out in the present, it might equally refer to a past event.
The use of present indicative is a very common grammatical structure in Spanish
media. There might be several reasons why journalists use it so often. The first might
be editorial. Since the present indicative tense usually omits the subject, sentences
using this conjugation tend to be shorter than those using the present progressive
form. This feature makes the present indicative more efficient in terms of printing
space. Since newspapers have strict space limitations – determined by the size of the
paper they use for printing – editors may favor the use of present indicative tenses for
making news reports shorter, which might allow for more reports to be included. The
second reason may be related to a higher impact on readers. Sentences written in
present indicative usually start with the verb, which draws the reader’s attention to
the action that took place. In this way, editors and journalists often use sensationalist
– sometimes lurid – verbs to craft headlines to hook the readers.
The partial sequence algorithm helps code sentences in which the verb is conju-
gated in present indicative tense. This feature is particularly useful for event coding
from text written in Spanish because this verb tense is very common in Latin Amer-
ican media. Since the translation of present indicative from Spanish into English
obscures the nuances of this verb form, the next example is presented in Spanish.
Consider the following sentence:

Arrestan a un criminal

132
 In this sentence, the subject of the action is omitted because of the conjugation of
the verb in present indicative tense. In the absence of a first actor, Eventus ID uses
the partial sequence algorithm to identify “arrestan ” (to arrest) as the action and
“a un criminal ” (a criminal) as the second actor. In consequence, the algorithm
generates the following event coding:

[- - -] [88104] [601060]

The recoding process described in Section 3.7.2 then uses Rule 4 to complete
the missing part of the event by assigning a default source actor to specific types of
actions. In the example used above, the recoding process assigns a default source
code for the state as the source of the event because it is the only kind of actor
capable of making an arrest.

3.5.4
Step
2.d.
Event
Database

The coding process used by Eventus ID generates an output database containing

the numeric codes of the events identified through both coding algorithms. The
output file is stored in a plain text UTF-8 file (.txt) referred as the “Event Database”
(event database file.txt). Each row in this file contains information about the date
of the event (date), the specific document and paragraph (doc id) from which the
event was extracted, and the codes for the source (actor1), action (verb) and target
(actor2). Based on the criteria provided by the dictionaries, the output also includes
the words recognized in the text as the source (srctxt), action (vrbtxt) and target
(trgtxt) of the event. In this way, the output database indicates when the event took
place and provides information about who did what to whom in both numeric and
textual format. The following example illustrates the basic structure of the output
file:

133
 date1 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt
date1 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt
date1 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt
date1 doc id3 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt
date2 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt
..
.
In this example, the first three lines correspond to three events identified on the
same date (date1) from the same document (doc id1). The fourth line corresponds
to an event that took place on the same date (date1) but was extracted from a
different document (doc id3). The remaining lines correspond to other events taking
place on different dates and extracted from different sources.

3.6
Stage
3.
Event
Location

The features of Eventus ID described above enable pieces of information that

describe when the event took place and who did what to whom to be extracted.
However, in order to provide a complete account of the event, it is also crucial to
know where the episode took place. Eventus ID has the capability to identify the
location of an event at state and municipal level when the information is provided
in the source text. Having georeferenced event data allows the interaction dynamics
between a variety of actors across time and space to be analysed. The event location
function is another feature that distinguishes Eventus ID from Tabari.
In general terms, Eventus ID uses events identified by event coding protocol and
goes back to the original source of information to find the location where the event
took place. The software uses detailed dictionaries of states and municipalities in
order to identify the locality mentioned in the original sources. In addition, due
to the complexities of this specific research, the location protocol includes a filter
dictionary that prevents certain words or phrases from being wrongly classified as
geographic locations.

134
3.6.1
Step
3.a.
Location
Dictionaries

Georeferencing event data using Eventus ID requires the development of location

dictionaries at the state and municipal level. Eventus ID uses these dictionaries as
categories for pattern recognition of event locations in the source text. The list of
states and municipalities and their corresponding codes comes from the National Geo-
statistical Framework generated by the Mexican census bureau, Instituto Nacional
de Estadı́stica y Geografı́a (INEGI) (2011b). Using official locality codes enables
the event data generated in this research to be readily cross-referenced with other
databases using the same codes for describing demographic, economic or geographic
attributes of each location.
The following list shows an example extracted from the state dictionary:

1 Aguascalientes
2 Baja California
3 Baja California Sur

The following list presents an example extracted from the municipality dictionary:

1002 Asientos
2004 Tijuana
3001 Comondu

Feedback from the validation check discussed in Section 3.7.1 revealed the need to
include a set of filters to prevent the event location algorithm from identifying false
positives. These problems emerged from the fact that some criminal organizations
are named after the states or cities where they operate. That is the case of “El Cartel
de Sinaloa,” “El Cartel de Tijuana,” “El Cartel de Juárez ” and “El Cartel de Jalisco
Nueva Generación,” among others. Eventus ID uses a filter dictionary to reduce
the risk of the algorithm confusing the names of criminal groups such as these with
locations.

135
 The filter dictionary also performs other nuanced filtering tasks. For example,
there is a municipality in the state of Tabasco named “Centro,” which means “center”
or “downtown.” The validation process revealed that some reports describing armed
clashes occurring in the center of the respective city (“centro de la ciudad ”) or dead
bodies left with a criminal mark in the center of their chest (“centro del pecho”) were
mistakenly coded as taking place in the municipality Centro, Tabasco. Other sources
of location error are reports mentioning the registration of vehicle license plates when
they are seized by the authoritiesstates included within military regions, and local
newspapers that have the state or municipality as part of their name (e.g. “El Diario
de Juárez ”). The Eventus ID location filter dictionary helps to minimize the risk of
coding error caused by these types of reports.
The following list shows an example extracted from the filter dictionary:

0 Cartel de Sinaloa
0 3/a Zona Militar La Paz BCS
0 Operativo Conjunto Michoacan

3.6.2
Step
3.b.
Event
Location
Using
Eventus
ID

To find the location of an event, Eventus ID combines four different types of files:

1. Text corpus: source of text to be analyzed.

2. Event database: containing the event data identified in Stage 2.
3. Location dictionaries: provides the list of locations and their codes.

• State dictionary: provides the list of states

• Municipality dictionary: provides the list of municipalities and towns
4. Location filter: provides a set of items that should not be confused with geographic
locations.
The event location coding procedure is outlined in Table 3.10. In general, the
location algorithm uses the event database generated by the event coding algorithms

136
and identifies the source paragraph from which each event was extracted. Then reads
the entire text corpus in order to identify the specific paragraph containing that
event. Once the paragraph is identified, the algorithm uses the information provided
by the location dictionaries to search for the name of a state or municipality in the
paragraph. If a location is identified, the protocol uses the filter dictionary to verify
whether the location should be assigned or discarded. If the location is not filtered,
the algorithm saves the location code next to the corresponding event in the event
dataset. If no location is found in the paragraph, the algorithm expands the search
to the rest of the document (Section 3.4.3 discusses the nomenclature and formatting
characteristics of the text corpus that enable all paragraphs that constitute a single
document to be identified). If a state or municipality is recognized in the document,
the protocol checks whether it should be filtered or not. If it passes the filter, the
algorithm saves the code of the location next to its corresponding event in the event
dataset. If no location is identified in the document, the protocol stops searching for
the location of this event and moves to the next event in the event database.

137
 TABLE 3.10

EVENTUS ID LOCATION ALGORITHM

1) Identify an event
- Load the event database
- Select an event
- Identify the paragraph name (doc id) from which the event was extracted
- Use the entire paragraph name as searching criteria
2) Identify the document in the source text
- Load the text corpus
- Search for the paragraph from which the event was extracted
3) Search for the location of the event in the paragraph
- Load the location directory
- Use the items of the location directory as searching criteria
- Start searching for the location in the source paragraph
- If the location is found, pause the search, store the code and go to Step 5
- If the location is not found in the paragraph, go to Step 4
4) Expand the search to the rest of the document
- Select the remaining paragraphs belonging to the same document
- Search for the location in all paragraphs of the document
- Begin searching in the first paragraph
- If the location is found in the document, store the location code and go to Step 5
- If the location is not found in the document, stop searching and go to Step 1
5) Filter the location
- Load the location filter
- Verify that the location identified does not match any item in the location filter
- If the location matches a filter, go back to Step 3
- If the location does not match a filter, go to Step 6
6) Save the location
- Save the location at the end of the coded event line in the event database
- Start again from Step 1

138
3.6.3
Step
3.c.
Georeferenced
Event
Data

The location algorithm takes the information generated at the event coding stage
and if a location is assigned, adds the coded location information. The output of
this procedure is a plain text UTF-8 file called the “Georeferenced Event Database”
(georeferenced event database file.txt) with information on who did what to whom,
when and where. The following example illustrates the output of the georeferenced
file indicating the state (state1) and municipality (mun1) of each event:

date1 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1
date1 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1
date1 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1
date1 doc id3 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1
date2 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1

3.7
Stage
4.
Validation
and
Recoding

Measurement validity is a central concern for social scientists using both quan-
titative or qualitative methods (Adcock and Collier, 2001; Bollen, 1989; Collier and
Brady, 2004; Goertz, 2005; King, Keohane and Verba, 1994). Measurement validity
is achieved when the scores meaningfully capture the ideas contained in the corre-
sponding concepts. Simply stated by Bollen (1989, 184), a score is valid if “a variable
measures what it is supposed to measure.” According to Adcock and Collier (2001),
measurement validity should be understood in relation to the congruency between
concepts and observations. These authors propose an analytical framework for eval-
uating measurement validity in terms of the degree of congruency across four levels:

1. Background concept. This level encompasses the constellation of diverse meanings

associated with a given concept.

2. Systematized concept. This level contains the specific formulation or definition of

a concept adopted by a particular researcher.

139
3. Indicators. This level refers to the procedure used for systematically building the
measures associated with the definition.
4. Scores for cases. This level refers to the numerical scores or qualitative classifica-
tion assigned as values for each measure.

In general terms, the framework indicates that a measure is valid to the extent that
the scores (level 4) correspond to a set of indicators (level 3), that can be meaningfully
interpreted in terms of the definition (level 2) used to represent a broader concept
(level 1).
As discussed in Section 3.4.1.1, the criteria for including or excluding relevant
notes on drug-related violence correspond to the relationship between the background
concept (level 2) and the systematized concept (level 3) of the validation framework.
This reflects the extent to which the indicators are congruent to the definition. The
systematic application of these criteria helps to identify valid news reports that corre-
spond to the conceptual components of organized crime violence defined in Section 1.3
of Chapter 1 and for excluding reports that do not meet the required criteria.
In order to validate the measures used in this research, it is necessary to assess
the congruence between levels 3 and 4, referring to the match between indicators and
scores. As discussed in Section 3.2.2, computerized textual annotation is not a silver
bullet. For some types of research projects manual coding may be a more useful
and feasible strategy, whereas in other cases machine coding my be a better fit for
the research objectives and resources. For this reason, it is necessary to validate the
output of an automated coding protocol. According to Grimmer and Stewart (2013),
the complexity of natural language implies that no computerized method is capable
of providing an exact representation of the content of texts. Although automated
text analysis may substantially reduce the amount of time and financial resources
required in manual annotation projects, computer-based coding schemes are likely to
generate some error. For this reason, researchers have the responsibility of validating
the coding result.

140
3.7.1
Step
4.a.
Validation

The approach used in this research is an iterative process consisting of automated

coding, validation and recoding. Validation is conducted by comparing a sample
of the machine-generated database with a manually coded database. Discrepancies
between human and machine coding are used to inform modifications in the dictio-
naries, coding algorithm or reformatting protocol. Repeated iterations of automated
coding, human-based validation and recoding increase the validity of the coding out-
put. This validation process is illustrated by the dotted line looping back from stage
4 to stages 2 and 3 in Figure 3.2 on page 107.
The validation check generated several contributions that proved crucial for the
development of this coding project and the validity of the output:

1. Software development: As mentioned previously, this project initially expected

to use Tabari as a tool for event coding. However, initial validation checks
revealed the poor performance of Tabari for coding in Spanish. This was the
first step towards the development of Eventus ID to code event data from text
written in Spanish.

2. Development of flexible event coding algorithms: Initial validation checks

helped to identify the limitations of a rigid algorithm only capable of coding events
with the subject–verb–object structure. This motivated development of the gen-
eral sequence algorithm (described in Section 3.5.3.1) to allow greater flexibility
in the coding process. In addition, the validation process was crucial for identify-
ing the coding errors caused by the substantive grammatical differences between
English and Spanish regarding the use of the present indicative tense. This dif-
ference motivated the development of the partial sequence algorithm discussed in
Section 3.5.3.2.

3. Actors and verb dictionary development: The validation process and reading
of several press reports helped to enhance and refine the nouns, pronouns, verbs
and verb phrases used in the actor and verb dictionaries.

4. Location filter: The validation check also helped identify complex reports in
which the event location process was identifying the location of an event erro-
neously. As discussed in Section 3.6.1, the detection of these situations motivated
the inclusion of a filter as part of the location algorithm in order to reduce coding
error.

141
5. Recoding: The repeated iterations between automated coding and validation
checks helped develop a set of rules used for recoding certain types of events using
statistical software. These rules are discussed in the next section.

The output validation process consists of comparing the events generated by au-
tomated coding with a human-generated database. The comparison is based on a
sample of news reports. These documents constitute the text corpus for the valida-
tion check. A team of trained research assistants manually coded relevant events of
drug-related violence from the sample corpus while keeping in mind the stages and
steps of the automated coding algorithm. In addition, the sample text corpus was
used for automated event coding using Eventus ID. The comparison between human
and automated annotation resulted in an 82 percent rate of accuracy. This indicates
that the dictionaries and recoding rules used in this project enable Eventus ID to be
right most of the time when coding event data from text in Spanish. However, as in
any other computer-based protocol, sometimes the text is so complex that there is a
small margin of error, which is estimated to be 18 percent in this project.

3.7.2
Step
4.b.
Recoding

Eventus ID combines the use of detailed actor and verb dictionaries, the applica-
tion of flexible event coding algorithms that consider grammatical characteristics of
the Spanish language, location recognition and filtering processes, and an iterative
validation coding process for generating accurate event data from text written in
Spanish. However, sometimes the characteristics of the source text and codification
strategies require additional recoding to improve the accuracy of the coding output.
This section outlines the main recoding rules used for refining the quality of event
data. Due to their complexity, these procedures are not integrated into Eventus ID
and are implemented using statistical software.

142
Recoding Rule 1: Passive voice. As discussed in Section 3.5.2, sentences using the
passive voice invert the order of the subject and object in the grammatical structure.
For example, consider the following sentence presented first in past indicative and
then in passive voice:

Army troops arrested a member of a criminal organization.

A member of a criminal organization was arrested by Army troops.

The passive voice sentence is erroneously coded as:

[601060] [99104] [202051]

But the recoding process corrects the directionality of the event as:

[202051] [88104] [601060]

Passive voice is a common grammatical structure in written news media in Spanish

and it generates error in the directionality of the event. To deal with those cases,
the dictionary development included a prefix 99 in the codes corresponding to verb
phrases conjugated in passive voice. The recoding recognizes the 99 prefix of verbs
conjugated in passive voice and inverts the position of the first and second actor so it
accurately reflects the directionality of the source conducting an action towards the
target.

Recoding Rule 2: Single actor. In some instances, the event identification pro-
tocol only generated an event code for the first actor ([actor1]) and put missing
codes for the verb ([- - -]) and second actor ([- - -]). This is usually the case
of government press releases about seizures as presented in the following sentence:

Military troops seized 6 kilograms of cocaine, two AK-47 rifles and

ammunition.

This sentence is coded as:

143
 [202051] [88202] [801022]
[901013] [- - -] [- - -]
[901015] [- - -] [- - -]

In this type of situation, the recoding protocol first whether the coded element
in [actor1] [- - -] [- - -] correspond to the categories of criminal assets, drugs
or weapons (see Table 3.6 for actor categories). If [actor1] meets the requirement,
then the recoding rule moves the code from the column of [actor2] and uses the
[actor1] and [verb] of the previous cell to fill in the empty spaces.13 The output
of this recoding process is:

[202051] [88202] [801022]

[202051] [88202] [901013]
[202051] [88202] [901015]

Depending on the paragraph structure, news reports sometimes mention the list
of seized items in separate lines. For example, consider the following sentence:

Military personnel seized the following items:

- 6 kilograms of cocaine,
- two AK-47 rifles,
- ammunition caliber 5.56×45

This sentence is initially coded as:

[202051] [88202] [- - -]
[801022] [- - -] [- - -]
[901013] [- - -] [- - -]
[901015] [- - -] [- - -]

In those cases, the recoding protocol applies the same rule to generate the follow-
ing event codes:

[202051] [88202] [801022]

[202051] [88202] [901013]
[202051] [88202] [901015]

13
If the [actor1] code does not meet the criteria of criminal assets, drugs or weapons, it means
that the code corresponds to a state actor, victim or criminal organization. In those cases, the event
coding provides no information about the action related to an actor of this type. In these cases the
event code is discarded from the database.

144
Recoding Rule 3: Authorities arrest an unidentified person. Eventus ID
uses the nouns and pronouns provided by the actor dictionary to identify the source
and target of an action. However, the software only recognizes an actor if it is already
loaded in the dictionary. There are some instances in which government press releases
inform about the arrest of a person, identifying the detainee by name. With the
exception of prominent criminal leaders, the actor dictionary used in this research
does not include a detailed list of names and last names of all people arrested.14 The
following sentence constitutes an example of this situation:

Army troops arrested Javier Osorio.

Since the name “Javier Osorio” is not part of the actor dictionary, the coding
scheme does not recognize it in the text sourcem and it generates the following
event:

[202051] [88104] [- - -]

In these cases, the recoding protocol verifes whether the code in [actor1] cor-
responds to the categories of any state actor (see Table 3.6) and if the [verb] code
coincides with the action category of arrests (see Table 3.7). If both elements meet
the criteria, the recoding protocol assigns a default code for “member of criminal
group” and generates the following event code:

[202051] [88104] [601060]

Recoding Rule 4: Sentences in present indicative. As discussed in Sec-

tion 3.5.3.2, sentences in which the verb is conjugated in the present indicative tend

14
Building such a detailed list of names would require a large number of coders reading all the
reports in order to identify the names of all individual and putting them into the actor dictionary.
The labor required for such a task would defeat the entire purpose of automated coding. Even if
the technological means were developed for gathering all names, building such a database would
contradict ethical principles required by the Institutional Review Board (IRB) that supports this
research and could even imply security concerns for the researcher.

145
to omit the subject of the action. The partial sequence algorithm enables events to
be identified from text in which the first relevant element of the sentence is a verb in
present indicative form. In those cases, the recoding rule assigns a default code for
[actor1] that corresponds to the type of action indicated by the verb.
The recoding rule for sentences in present indicative considers two types of default
actors; state [99999] and criminals [66666]. As shown in Table 3.11, some kind of
actions can only be undertaken by either one or the other of these actors. In these
cases, the recoding rule assigns the corresponding default code for [actor1]. If the
action refers to arrests, seizures, eradication, detection or raids, the recoding rule
uses the default code of the state. If the action refers to kidnapping or torture and
mutilation, the rule inserts the default code for criminals.15

TABLE 3.11

DIRECT CRITERIA FOR DEFAULT ACTOR ASSIGNMENT

Default state code [99999] Default DTO code [66666]
Arrest Kidnap
Seize Torture and mutilate
Eradicate
Detect
Raid

15
Assigning a default criminal code for perpetrators of torture and mutilation does not necessarily
mean that state authorities are not capable of conducting this kind of action. In fact, there have
been news reports indicating that the police or the army have used torture and similar abuses
against presumed members of criminal organizations. When these events are reported in the media,
they usually are presented in such a way that explicitly indicate the state security forces as the
perpetrators of such actions. The main purpose of those reports is to clearly denounce this type of
behavior. In such cases, reports usually do not omit the subject of the action and explicitly state
the subject, verb and object of the sentence. In consequence, this type of report is more likely to
be fully coded by the complete sequence algorithm than the partial sequence algorithm where the
subject is implicit. Therefore, it is highly unlikely that the recoding rule would erroneously assign
a default criminal code for a torturing event perpetrated by the state.

146
 For example, the state is the only actor that can arrest a person. Consider the
following sentence:

Arrestan a un criminal

The coding algorithm generates an event where [actor1] is missing:

[- - -] [88104] [601060]

The recoding protocol then identifies that the verb corresponds to the action “to
arrest” and fills the missing subject with the corresponding default, which is the state
code.

[99999] [88104] [601060]

Unfortunately, not all actions are mutually exclusive. Some actions, specially
violent ones, can be perpetrated by either the state security forces or members of
criminal organizations. In those cases, the recoding rule uses a more sophisticated
identification process based on the wording of the action and the type of target
identified in the sentence. Table 3.12 shows the conditional criteria used in this rule
for assigning default actors. Some actions are easily identifiable. For example, if
the sentence uses the verb “to burn” in present indicative to report that drugs were
burned, the recoding rule assigns the state as the default actor. The reason is that
security forces often burn drugs after seizures or crop eradication. In contrast, if
the report indicates that a body was burned, the rule assigns the default code for
criminals. The reason is that organized criminals sometimes burn the body of their
victim to prevent identification of the body.
Some other cases are more complicated. For example, consider a sentence using
present indicative to describe a situation in which a man was shot. In Spanish, the
report would say:

Le disparan a un hombre

147
 TABLE 3.12

CONDITIONAL CRITERIA FOR DEFAULT ACTOR ASSIGNMENT

Action in [verb] Wording in [verb] Target in [actor2] Assigned source
Drugs
State
Crops
Burn Body
Person Criminal
Property
“repelen” Criminal State
“agreden”
Attack
“atacan” State or person Criminal
“emboscan”
“disparan”
Criminal State
“abren fuego”
“ametrallan”
Shoot
“balacean”
State or person Criminal
“rafaguean”
“disparan”
Criminal State
Wound
State Criminal
“abaten” Criminal State
“asesinan”
“matan”
Kill “acribillan”
State or person Criminal
“masacran”
“encajuelan”
“entamban”

The coding algorithm generates the following event without specifying the perpe-
trator of the action:

148
 [- - -] [88104] [601060]

Fortunately, the writing style of journalists and the language of press releases
issued by government agencies provide valuable hints for identifying the most likely
perpetrators of particular types of actions. Based on the content analysis of several
reports, it was clear that reports use different words for referring to violent acts
conducted by criminal organizations that those carried out by the state. For example,
journalists tend to use words such as “ametrallan” (shoot with a machine gun),
“rafaguean” (spray with bullets) or “acribillan” (riddle with bullets) for referring to
shootings perpetrated by criminal organizations. In contrast, they use more formal
terms, such as “abren fuego” (open fire), for referring to situations when the state
security forces shoot at presumed criminals.
For example, if the sentence says that a man was “riddled with bullets,”

Acribillan a un hombre

the coding algorithm attributes the event to criminal organizations:

[66666] [88104] [601060]

In contrast, if the sentence suggests that some one “opened fire” against a man,

Abren fuego contra un hombre

the recoding rule attributes the event to the state:

[99999] [88104] [601060]

In this sense, the conditional criteria for default actor assignment uses the nuances
of language and journalistic style to identify the perpetrator of certain types of actions
expressed in present indicative.

Recoding Rule 5: Discovery of a dead body. Although fatal victims are often
the result of overt armed clashes between rival criminal groups or between the state

149
and a criminal organization, a large number of casualties are killed out of sight and
their bodies are abandoned. In those cases, news reports usually describe the event
as:

The police found a dead body

In this case, the event coding algorithm identifies the police as the source, found
as the action and a dead body as the target, and generates this event code:

[99999] [88104] [601060]

Of course, this does not mean that the police killed that person. The sentence
indicates that a person died (or was killed) and the body was discovered by the
police. However, Eventus ID does not know that. It is simply an algorithm following
coding instructions and is not capable of abstracting the action represented in the
sentence. For this reason, the recoding rule codes the event as a murder perpetrated
by a criminal organization and generates the following event:

[99999] [88104] [601060]

Recoding Rule 6: Armed clashes. There are several instances in which news
reports describe confrontations between government authorities and criminal organi-
zations. Since the event-based approach used in this research emphasizes the inter-
active characteristics of violence, the recoding rule considers armed clashes as dyads
between the actors involved in the confrontation. An armed clash is usually described
as:

Troops of the Fifth Infantry Battalion clashed with a group of hired

assassins.

In this case, the event coding algorithm identifies Infantry Battalion as the
source, clashed as the action and a group of hired assassins as the target, and
generates this event code:

150
 [202031] [88103] [501050]

In order to provide an interactive measure of conflict, the recoding rule gener-

ates the following dyadic event of the Army conducting violence against criminal
organizations and the latter responding in kind to the former:

[202031] [88103] [501050]

[501050] [88103] [202031]

3.7.3
Duplicates

There are several instances in the data gathering process and the coding protocol
used in this research that could lead to multiple coding of events. As discussed in
Section 3.4.1, 105 information sources are used in an effort to minimize coverage bias
by having multiple national and local sources of news reports and government press
releases. However, using multiple sources increases the risk of artificially inflating the
number of reports. This can happen when several newspapers report a prominent
event. In addition, as discussed in Section 3.4.3, the coding protocol extracts infor-
mation from the entire content of news reports. This generates the risk of multiple
event coding when the news report mentions the same episode several times as part
of the narrative.
To avoid artificially augmenting the number of events, this research pays detailed
attention on identifying and excluding multiple events. I use the term “duplicates”
for referring to events counted multiple times (two or more times). Technically, the
detection of duplicate observations is implemented through the duplicates command
in Stata, which makes it possible to report, give examples of, list, browse, tag, and
delete duplicates. This research relies on a highly conservative approach for detecting
and eliminating duplicates.

151
 The first step of the protocol operates at the paragraph level and identifies multi-
ple events occurring in the same municipality–day and mentioned in the same para-
graph. If a single paragraph contains two or more events of the same kind taking
place in the same municipality–day (e.g./ the police seized some drugs), then one of
those events is excluded from the database. In contrast, if the paragraph contains
multiple events in the same municipality–day indicating different types of actions
(e.g./ the army seized drugs and arrested a criminal), then these different events are
not considered duplicates at this stage.
This is a conservative approach for eliminating duplicates especially for events
related to drug interdiction and gun seizures. Sometimes reports describe episodes
where government authorities seize drugs detected in different bundles or arsenals
containing a variety of weapons and calibers. For example, a news report could
indicate that the army seized “25 packages of marijuana and a bag with 12 kg.
of marijuana seed.” Initially, Eventus ID would code this report as two events of
drug seizure, but one of those events would be excluded in the duplicates detection
protocol.
After excluding duplicates at the paragraph level, the protocol focuses on the
document as the unit of analysis. The second step compares the content of para-
graphs within the same document. If there are multiple events of the same kind
occurring in the same municipality–day in the document, the protocol keeps only
one of these events and excludes the rest. This avoids multiple counting of the same
event mentioned in different paragraphs of the same news report.
Finally, the third step of the duplicates detection protocol analyzes the different
documents contained in the text corpus. Having already excluded duplicates at the
paragraph and document levels, the protocol detects and excludes multiple events
of the same kind occurring in the same municipality–day that might get reported

152
by several information sources. This prevents inflation of the event count of highly
visible episodes of violence that are reported by multiple newspapers.
It is important to note that the detection of duplicates focuses on events identified
at the municipal level. This implies that multiple events reported exclusively at
the state level are excluded from the database. National newspapers sometimes
issue news reports providing very general accounts of drug-related violence without
specifying the facts. These reports usually say something like: “Yesterday, the wave
of violence left 10 people dead. Drug cartels killed five men in Chihuahua, two
in Durango and three more in Sinaloa.” Reports of this kind do not indicate the
municipalities where the events took place. In consequence, they are excluded from
the database. In this way, the database contains exclusively those events that are
reported only once at the municipal level on a daily basis and excludes possible
duplicates and those reported at the state level.

3.8
Stage
5.
Validated
Event
Database

The final stage of the coding protocol is the release of the validated event database
containing unique events which are components of large-scale organized criminal vi-
olence at municipal level on a daily basis. The content of the output database has
the following structure:

date1 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1
date1 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1
date1 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1
date1 doc id3 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1
date2 doc id1 [actor1] [verb] [actor2] srctxt vrbtxt trgtxt state1 mun1

In order to reduce the complexity in the myriad of combinations among perpe-

trators, actions and targets, the information in the database is aggregated into three
main types of events; law enforcement activities conducted by the state against orga-
nized criminals, violence perpetrated by criminal organizations against government

153
authorities, and violent events between rival criminal organizations. In addition, de-
pending on the analytical purposes of the empirical chapters of this dissertation, the
information of the validated event data set can be aggregated at different units of
analysis such as municipal, state or national level, or by daily, monthly or yearly time
periods. Disaggregated events can also be used as data points for spatial analysis
using geographic information systems.

3.9
Conclusion

This chapter focuses on describing the main features and operational capabilities
of Eventus ID, a novel software for automated textual annotation of event data
from text written in Spanish. The software uses a sparse parsing algorithm capable
of extracting event data information from news reports, thus providing a detailed
account on who did what to whom, when and where in the Mexican war on drugs.
The data collection strategy used in this research relies on a massive collection
of news reports from 105 information sources including press releases from the main
federal and local government agencies, several national newspapers and dozens of
local newspapers. The selection of reports is conducted by a team of human coders
carefully applying explicit criteria for inclusion and exclusion of relevant news reports,
thus contributing to the validity of the database. Eventus ID processes thousands
of news reports related to the different components of organized criminal violence.
In contrast to the poor performance of Tabari for coding text written Spanish,
Eventus ID provides an accurate and proficient way of identifying event data from
news reports in Spanish. This is possible thanks to the development of a coding
algorithm capable of adapting to grammatical features of the Spanish language. In
addition, Eventus ID is capable of georeferencing event data at municipal level. The

154
next chapter presents the database on drug-related violence generated using Eventus
ID.
The validation assessment shows that Eventus ID reached a level of accuracy of
82 percent in this research project, when compared to the coding results generated
by manual annotation methods. This suggest that Eventus ID has a high level of
accuracy when coding event data, despite the grammatical complexities of Spanish
and given the quality of news reports. Although no computerized annotation method
is perfectly accurate, the detailed methodological discussion presented in this chap-
ter provides a transparent and feasible strategy for coding event data from text in
Spanish.
Based on the combination of recent technological advances in natural language
processing and quantitative research on conflict, Eventus ID constitutes a key con-
tribution for generating fine-grained event data to analyze the micro-dynamics of
conflict. Previous efforts to implement automated coding protocols have relied ex-
clusively on news reports written in English. However, this has limited the amount
and kind of information that researchers can use for building databases on conflict.
In contrast, Eventus ID opens up the possibility of accurately coding events from
non-English information sources, thus allowing researchers to obtain more detailed
information from original sources.
Finally, Eventus ID constitutes a public good that could help other researchers
to conduct their own event coding projects in other Spanish-speaking countries. Ap-
propriate modifications to the coding protocol used in this research could this coding
strategy to be replicated in other Latin American countries, thus facilitating the
generation of comparative data on organized criminal violence in the region.

155
 CHAPTER
4

MAIN
TRENDS
OF
ORGANIZED
CRIMINAL
VIOLENCE

4.1
Introduction

The previous chapter discusses the empirical strategy used for measuring processes
of conflict inherent to the Mexican war on drugs. The measurement strategy is based
on collecting data from a large number of news reports and relies on Eventus ID for
automated-coding of event data. This chapter presents the database generated by
the Eventus ID coding protocol. The central empirical support used in this research
comes from a novel database called “Organized Criminal Violence Event Data in Mex-
ico 2000–2010” (OCVED). This data set comprises daily event data geo-referenced
at municipal level on the law enforcement actions conducted by government authori-
ties against criminals, violent actions perpetrated by criminals against the state, and
events of violence between different criminal organizations. OCVED includes infor-
mation from all Mexican municipalities on a daily basis between January 1, 2000 and
December 3, 2010, thus comprising more than 9.8 million observations. In general

156
terms, the database used in this research provides fine-grained information on who
did what to whom, when and where in the Mexican war on drugs.
The analysis of the data indicates varying trends in the components of violence
in the Mexican war on drugs. The data reveals that violent competition between
rival criminal organizations constitutes the most prominent type of violence. After
operating non-violently for several years, criminal organizations engaged in an un-
precedented wave of violence against their rivals. The sustained increase of criminal
conflict is evident at the national level and shows varying levels of intensity at the
sub-national level. In addition, the spatial analysis of violence among criminals shows
different patterns of intensification and diffusion across time and space.
Violence perpetrated by criminal organizations against government authorities
constitutes the second most common type of violence. Criminal hostilities against
the state also show a sustained increase after 2007. Violent law enforcement actions
constitute the smallest share of violent actions, thus suggesting that government
authorities do not generally resort to the use of violence to fight crime. Nevertheless,
the trend of violent enforcement shows a substantial increase in the recent years.
However, the state has a broader menu of security tactics to fight crime beyond violent
law enforcement actions. The data reveals that beyond violence, the government
mostly employs drug interdiction and arrests to fight criminal organizations. Other
non-violent tactics such as seizures of criminal assets and weapons complement the
menu of policy options available to the state, but these are used less frequently.
This chapter is divided in two sections. The first part discusses the limitations of
current efforts for measuring drug-related violence in Mexico by counting homicides,
also known as the body-count approach. This section also discusses how an event-
based perspective of OCVED overcomes some of the shortcomings of homicide data.
In addition, it discusses the analytical leverage gained from disentangling the different

157
types of actors and actions inherent to the war on drugs, and from emphasizing
temporal and spatial disaggregation of data.
The second section describes the main trends of violence in the Mexican war on
drugs. The first part of this section provides a general description of the frequency
and distribution of the different types of law enforcement tactics, and of violence
perpetrated by criminals against the state or against rival criminal organizations.
The next part provides an aggregated overview of temporal trends of violence. This
is followed by a disaggregation of the analysis focusing on the temporal variation of
violence across states. The next part discusses the main trends of violence at the
municipal level on a daily basis. Finally, the last segment gives an overview of the
spatial trends of intensification and diffusion of violence.

4.2
Limitations
of
Body
Counts

One of the central challenges of studying drug violence from a quantitative per-
spective is the scarcity of systematic, reliable, good quality, available data. In Mex-
ico, there are four main databases measuring drug-related violence. One of these
databases was generated by the Mexican government, two were created by national
newspapers and the fourth belongs to a policy analyst. In 2010, the Mexican Security
Council, Sistema Nacional de Seguridad Pública, SNSP (2011a) released a database
of drug-related homicides only after receiving severe pressure from civil society orga-
nizations, public opinion, policy analysts and victims to provide information about
the casualties associated with the war on drugs. Initially, the government released a
database with homicide data presumably related to rivalries among criminal organi-
zations between 2007 and 2010. This data was compiled at the municipal level on a
monthly basis. After the release of the data, public opinion and analysts used the
information to criticize the lack of effectiveness of the government’s security strategy

158
in reducing violence. In an effort to reduce criticism, the Mexican government delib-
erately reduced the quality of the data by aggregating the information on a quarterly
basis and eventually stopped updating the database of drug-related homicides. In
addition to official counts, two newspapers, Reforma (http://www.reforma.com/)
and Milenio (http://www.milenio.com/), and a prominent security analyst, Ed-
uardo Guerrero (2010a), have undertaken the only private efforts to quantify the
wave of drug violence in Mexico. Unfortunately, none of these private sources release
their information for public scrutiny. Researchers interested on conducting quantita-
tive analysis of drug-related violence in Mexico are thus restricted to using the data
generated by the government.
There are two other government databases that researchers sometimes use to
analyze violence in Mexico. One is the number of homicides counted by the Mexican
Census Authority, Instituto Nacional de Estadı́stica y Geografı́a, INEGI (2013b) and
the health statistics of the Mexican Health Ministry Sistema Nacional de Información
en Salud, SINAIS (2013). These databases have the advantage of having annual time
series going back to year 2000. This is a characteristic that some researchers find
useful for exploring the conditions prior to the onset of the Mexican war on drugs
in 2007. However, these measures capture homicides of all kinds, including those
which could be caused by crimes of passion, vehicle accidents, armed robbery, and all
other causes. Therefore, these measures are too broad to accurately capture drug-
related violence, and thus raise considerable concerns of measurement validity and
measurement error. For this reason, these measures are not discussed in this research.
One of the empirical contributions of this research is to present the database
“Organized Criminal Violence Event Data in Mexico 2000–2010.” This database
overcomes several conceptual, methodological and scope limitations of previous ef-
forts to quantify large-scale organized criminal violence. Conceptually, instead of
using a body count approach as the other databases do, OCVED relies on an event

159
data approach to classify the interactions among and between a variety of actors
carrying out diverse types of violent actions against each other. Methodologically,
OCVED is based on a rigorous, transparent process of information gathering and au-
tomated event coding, discussed in Chapter 3. In addition, OCVED overcomes the
temporal and spatial limitations of the other databases by systematically covering all
the municipalities of the country on a daily basis between 2000 and 2010. OCVED
thus constitutes an unprecedented effort for measuring and analyzing the wave of
drug violence in Mexico from a quantitative perspective.
Before presenting the main trends of the different types of violence contained in
OCVED, this section discusses the characteristics and main limitations of alternative
measures of drug-related violence. As Table 4.1 shows, extant databases have sev-
eral methodological problems and limitations with respect to the definition of their
object of study, selection criteria, coding methodology, unit of analysis, spatial and
temporal coverage, and information sources. All these efforts to quantify the wave of
violence measure exclusively the number of homicides related to the war on drugs.
This approach has several limitations. Death is certainly the ultimate expression of
violence. However, it is not the only one. There is a wide range of violent actions
that do not necessarily lead to a person being killed. For example, if the government
security forces engage in an armed confrontation with organized criminals and the
clash only leads to the detention of those criminals, but does not generate casualties
on either side, such an event would not be coded in a homicide-based approach. In
consequence, there much information on violence that is not captured by a body
count database.

160
 TABLE 4.1

DATABASES OF DRUG-RELATED VIOLENCE IN MEXICO

SNSP Reforma Milenio Guerrero Osorio

Object Homicides Homicides Homicides Homicides Event Data

Selection criteria Explicit Not explicit Not explicit Not explicit Explicit

Methodology Explicit Not explicit Not explicit Not explicit Explicit

Spatial unit Municipal Municipala Municipala Municipal Municipal

Spatial coverage Most municipalities Not explicit Not explicit All municipalities All municipalities

161
Temporal unit Quarterlyb Monthly Monthly Monthly Daily

Temporal coverage 12/2006 - 12/2010 12/2006 - 12/2012 12/2006 - 12/2012 12/2006 - 12/2012 1/2000 - 12/2010

Observations 57,183 Unknown Unknown Unknown 9,860,840

Information sources Explicit Not explicit Not explicit Explicit Explicit

Coding Manual Manual Manual Manual Automated

Publicly available Yes No No No Yesc

a
Not explicitly mentioned by these sources but it can be inferred from their reports.
b
Data initially released on a monthly basis but later on it became only available quarterly.
c
The database will be available after the its first peer-reviewed publication.
 By focusing exclusively on homicides, body counts also miss a crucial character-
istic of violence: its interactive character. After all, counting a body does not give
any information on who killed the person, who was the target, or how the person
died. Using a homicide-based approach to measure violence neglects key pieces of
information such as who is using violence, what type of tactics are being used to con-
duct violence, or who is the target of violence. Neglecting this information severely
compromises our efforts to understand the characteristics and dynamics of conflict
processes. Drawing a simple analogy between sports and the measurement of vi-
olence, it is not possible to understand the outcome of a boxing match by merely
seeing one boxer knocked out on the floor. In order to give an accurate account of
the boxing match, an observer must carefully analyze the exchange of attack and
defense moves made by both contenders. Similarly, simply counting homicides gives
an overly limited view of the complexities of large-scale organized criminal violence.
Related to the homicide-based approach, extant databases do not provide a clear
definition of the type of homicide being coded. Among the body count databases,
only the government has an explicit definition of what is considered a drug-related
homicide. The official database comprises information on “executions” and “deaths
resulting from clashes and aggressions” (Sistema Nacional de Seguridad Pública,
2011b). Executions are defined by the SNSP as violent deaths in which the victim or
murderer is presumably a member of a criminal organization and which are not the
result of clashes or aggression. Official data also comprises counts of deaths caused
by clashes between criminal organizations or attacks against government authorities.
In contrast, none of the newspaper databases nor the database created by Guerrero
explicitly define their object of study. This does not mean that they do not follow
any coding scheme, it simply is not made public. Therefore, it is not possible to know
what are they counting as a drug-related homicide, nor if the measure is valid for the
object of study.

162
 The government briefly explains its methodology for gathering data. The process
consists of five steps: (i) the Army, Navy, Federal Police and the Mexican intelligence
agency each supply information about drug-related homicides; (ii) specialists from
each security agency verify whether homicides meet the criteria for being categorized
as executions or homicides from clashes or aggressions; (iii) the Office of the Attorney
General collects all the information into a database; (iv) the Attorney General checks
for duplicates and updates; and (v) representatives of each agency meet periodically
to validate the database. Unfortunately, due to the confidentiality of these official
sources, it is not possible to replicate or validate this database, nor to evaluate
its strengths, weaknesses or biases. Therefore, researchers using official data have
no other option than to trust the government reports. Despite its methodological
limitations, the government database is at least more transparent than the private
databases, which do not make their methodology explicit or publicly available.
Government data is available at the municipal level until 2010 and covers most
municipalities. A careful review of the official database showed that the it contains
information for only 1,167 out of the total of 2,456 municipalities in the country. Un-
fortunately, the methodology is not explicit whether there have been no drug-related
homicides in those municipalities or if there is missing data in those locations. A sys-
tematic database should contain records for all municipalities, reporting zero counts
for those with no homicides. Although Guerrero provides no explicit methodology
or coverage for his database, he reported in a personal interview that his database
comprises information at municipal level and covers all municipalities in the country.
it is not clear whether Reforma and Milenio collect their data at municipal or state
level, nor the scope of their coverage.
As mentioned above, the Mexican government initially released a database report-
ing events on a monthly basis. However, without proper methodological justification,
state officials decided to aggregate the data on a quarterly basis. As mentioned by

163
Shellman (2004), aggregating data into larger time periods creates methodological
problems, as it reduces the efficiency of estimates and is likely to induce bias. In
consequence, the decision to aggregate the official reports severely undermined the
quality of the data. The databases generated by Reforma, Milenio and Guerrero con-
tain monthly data. Although the monthly periodicity of these databases represents
an improvement with respect to the dominant country–year approach in the quanti-
tative study of political conflict, this temporal aggregation still is inappropriate for
analyzing the rapidly-changing dynamics of violence. Since monthly data compresses
several actions and reactions among different parties in conflict into a single aggre-
gated measure, using a month-by-month aggregation will not throw light on those
interactions. Bundling a variety of different events into an aggregated monthly unit
increases the chances of generating misleading conclusions and inefficient estimates.
This problem is even worse for quarterly data, as all the interactions that took place
within three months are lumped into a single aggregated measure.
All these databases start the count of drug-related homicides in December 2006
when President Calderón took office and declared a nation-wide crusade against crimi-
nal organizations. In total, the official database contains 57,183 municipality-months,
while the number of observations of the private databases remains unknown. Unfor-
tunately, none of these databases provides information about drug violence earlier
than that date. The lack of data prior to the onset of the war on drugs creates
a truncated data problem that is likely to generate bias in quantitative estimation
(Geddes, 2003; King, Keohane and Verba, 1994). Without information about the
trends and characteristics of violence before December 2006, it is not possible to
analyze the structural causes and dynamic factors that preceded the escalation of
violence. Moreover, if there is no information about the characteristics of drug vio-
lence before the onset of the conflict it is not possible to provide an evidence-based
assessment of the official discourse arguing that drug violence was on the upsurge

164
before Calderón took office, nor the argument advanced by his critics who claim that
the government crusade against drugs triggered the escalation of violence.
Finally, the methodology of the official database explicitly states that the in-
formation comes from reports gathered by the Army, Navy, Federal Police and the
Mexican intelligence agency. Presumably these federal agencies have a presence in
the entire country and the risk of coverage bias of these information sources is min-
imal. However, due to the confidentiality of their reports, their validity cannot be
confirmed. Guerrero does not explicitly state the information sources used to build
his database, mentioning only that the data are extracted from “19 national and local
newspapers” (see footnote in Guerrero (Figure 1 2010a)). In consequence, it is not
possible to assess coverage bias that might affect his database. Unfortunately, Re-
forma and Milenio do not report their sources of information. It is not clear whether
they rely exclusively on their own journalists to provide information and coding data,
or whether they systematically use other sources to complement their lack of coverage
in certain regions of the country. Despite being national newspapers, Reforma and
Milenio have a stronger presence in the north and central regions of the country than
in other areas of the country. In consequence, there is a basis for suspecting coverage
bias.
To overcome the limitations of extant databases of organized criminal violence
in Mexico, this research introduces OCVED, a novel machine-generated database of
daily event data on drug-related violence at the municipal level between 2000 and
2010. This database comprises detailed information on who did what to whom, when
and where in the Mexican war on drugs. The last column of Table 4.1 shows the
characteristics of OCVED that overcome the methodological shortcomings of previ-
ous efforts to quantify the wave of drug violence in Mexico. Instead of focusing on
a narrow operationalization of violence based on homicides, this database relies on
a more sophisticated conceptualization of violence based on the interaction among

165
actors conducting different types of actions against each other. Section 1.3 in Chap-
ter 1 discusses the conceptualization of violent and non-violent actions by the state
against criminal organizations, as well as violent retaliation from criminals against
government authorities and violent competition between criminal organizations. This
interactive approach of conflict is better reflected by event data than by victim counts.
An event is composed by three elements: the source, which is the actor perpetrating
an action; the action being carried out; and the target of such action. In other words,
an event comprises information on someone doing something to someone else. This
database considers actions undertaken by the state against criminal organizations,
actions perpetrated by criminal groups against the state, and actions by criminal
organisations against other criminal organizations. In addition, the menu of actions
analyzed in this database goes beyond homicides and includes a broad range of vi-
olent actions with varying levels of intensity. Sections 3.5.1 and 3.5.2 in Chapter 3
report the list of actors and actions used for event coding.
This database provides additional improvements in terms of spatial and temporal
coverage. The research covers all 2,456 municipalities of the country.1 In contrast
to monthly or quarterly aggregated information used in other databases, OCVED
reports drug violence events on a daily basis. In addition, it covers a larger tem-
poral horizon, as it starts on January 1, 2000 and runs to December 31, 2010, thus
overcoming the problem of data truncation prior to December 2006 shared by the
other databases. The spatial coverage (N=2,456 municipalities) and time coverage
(T=4,017 days) leads to a large time series cross-sectional dataset (N×T) of 9,860,840
observations. As reported in Section 3.4.1 in Chapter 3, this database is built using
reports from 105 sources which include both press releases from federal and state gov-
ernment agencies and news reports from national and local newspapers. Finally, in

1
Between 2000 and 2010, the federal government created 14 new municipalities. In order to
keep the consistency of the panel database before the creation of those municipalities, I imputed
information from the municipalities out of which the new ones were created.

166
contrast to the other databases, which relied on human coders to systematize informa-
tion, this database has been created using a computer-assisted approach employing
Eventus ID, the software for automated annotation of event data from text written
in Spanish described in Chapter 3. In consequence, this unprecedented quantitative
effort provides the first machine-generated database comprising detailed information
on who did what to whom in the Mexican war on drugs on a daily basis in each
municipality over an eleven-year period. The rest of this chapter describes the main
characteristics of the components of drug-related violence contained in OCVED.

4.3
Main
Trends
of
Organized
Criminal
Violence

This section presents the main temporal and spatial characteristics of the variables
of violent and non-violent enforcement, criminal retaliation from drug trafficking
organizations against the state, and violent competition among rival criminal groups.
To discuss the data, this section is divided into five parts. The first subsection
discusses the type of data contained in OCVED and presents the main descriptive
statistics of the database. The second part takes a first look at the data at the national
level and discusses the different components of violence on a monthly basis. The next
segment provides a more disaggregated point of view by describing the different trends
of violence at the state level. The following section discusses the characteristics of
violence at municipal level on a daily basis. Finally, the last part relies on Geographic
Information Systems (GIS) to show the spatio-temporal variation of violence.

4.3.1
General
Description
of
Event
Data

Based on the conceptual definitions of the components of large-scale organized

criminal violence discussed in Section 1.3 in Chapter 1, this research uses the coding
protocol of Eventus ID discussed in Chapter 3 to identify a set of law enforcement

167
activities conducted by the state against organized criminals, violence perpetrated
by criminal organizations against government authorities, and violent events between
rival criminal organizations. These different violent events constitute the core com-
ponents of OCVED, the database used in this research for analyzing trends of drug-
related violence in all Mexican municipalities on a daily basis between 2000 and
2010.
The set of law enforcement tactics used by government authorities can be di-
vided into violent and non-violent enforcement. Violent enforcement includes events
in which government authorities attacked, wounded or killed suspected members of
criminal organizations or repelled an attack from them. Non-violent law enforce-
ment includes four different types of events; arrests, seizure of criminal assets, drug
interdiction, and seizure of weapons. Arrests refers to events in which government
authorities detained suspected members of criminal organizations. To avoid dou-
ble counting, the measure of arrests does not include events in which criminals are
sentenced to prison; it only counts the event where they were arrested. Seizure of
assets includes events in which government authorities seized criminal assets such
as properties (e.g. mansions, safe houses, storage facilities) or any type of vehicle
for land, water or air transportation (e.g. armored SUVs, speed boats, airplanes).
Seizure of drugs includes events in which government authorities confiscated illegal
drugs. The list of illicit substances considered for coding events of drug interdiction
comprises an exhaustive inventory of drugs including cannabis, opiates (e.g. cocaine,
crack, heroin) and non-natural hallucinogens (e.g. LSD, methadone).2 Finally, the
set of non-violent law enforcement actions includes seizures of weapons referring to
events in which government authorities confiscated guns from members of criminal
organizations. The list of weaponry covered in event coding ranges from pistols and

2
For a list of illicit drugs and their scientific definitions see UNODC (2003)

168
semi-automatic weapons (e.g. AK-47 or R-15 rifles), to explosives (e.g. hand grenades)
and anti-materiel rifles capable of piercing armored vehicles (e.g. Barret M82 rifle).
The measure of criminal retaliation consists of events in which members of crim-
inal organizations perpetrated violence against government authorities. The list of
actions included in criminal retaliation include attacks, ambushes, shootings, kid-
napping, wounding, killing, torture and mutilation where the specific target of those
actions is a government official or member of government security forces. The mea-
sure of violent competition between criminal organizations considers any of the actions
listed above in which violence is perpetrated by members of a criminal organization
against another criminal group.
Figure 4.1 shows the frequency of events of violent and non-violent law enforce-
ment from the state against DTOs, events of violent retaliation from criminals against
government authorities, and event of violent competition between rival criminal or-
ganizations. In total, the database contains 251,167 events of drug-related violence.
The figure reveals that the state relies mostly on drug interdiction and arrests to fight
criminals. These two types of non-violent events constitute 32.6 percent and 25.2 per-
cent of the total number of events coded in the OCVED database. Other non-violent
actions, such as seizure of criminal assets and weapons, comprise 8.1 percent and 6.8
percent, respectively. The figure also shows that violent law enforcement by the state
is rarely used, constituting only 2.4 percent of the total events coded in the database.
Violent competition among rival criminal organizations, in contrast, represents a sub-
stantial share of the number of events. Confrontations between criminals represent
about 18.9 percent of all events coded in the data set. Finally, the figure shows that
criminal retaliation against government authorities represents about 6 percent of the
total events contained in OCVED.
The type of information contained in each variable in OCVED is event count data.
Event counts are defined as non-negative integers I such that I ≥ 0, representing the

169
 Figure 4.1. Frequency of events associated with drug-related violence by
type of event

number of times an event occurs within the time span of observation (one day) and
in the spatial unit of analysis (one municipality). Table 4.2 presents the descriptive
statistics of the different components of violence counted in the Mexican war on
drugs at the municipal level on a daily basis. The mean and standard deviation of
each variable seem very small. This could suggest that these events are extremely
rare events. However, readers must keep in mind that these descriptive statistics are
calculated with respect to the total number of 9.8 million observations contained in
the entire database.
The descriptive statistics in Table 4.2 show that violent law enforcement has a
mean of 0.0006 events per municipality–day and ranges from 0 to 9 events of gov-
ernment violence against criminals. Putting the mean of violent enforcement in
perspective, this average corresponds to 1.53 events of state violence each day in the
entire country. The Table reports the following distributions for non-violent enforce-
ment: arrests have a mean of 0.0064; the average number of seizures of assets is
0.0021 events; the mean number of drug interdiction events is 0.0083; and the aver-

170
age number of gun seizures is 0.0017 events. With respect to retaliation, the table
indicates that the average number of criminal attacks against government authori-
ties per municipality–day is 0.0015 events, and ranges from 0 to 11 attacks a day.
The mean retaliation on a municipality–daymunicipality–day level is equivalent of
observing 3.74 criminal attacks against government authorities every day across the
country. Finally, the descriptive statistics show that, on average, there are 0.0048
violent events among rival criminal organizations on a municipal daily basis, and the
number ranges from 0 to 36 events of violence a day. This corresponds to observing
11.83 events of violence among criminal organizations every day at the national level.

TABLE 4.2

DESCRIPTIVE STATISTICS OF THE MAIN VARIABLES OF

ORGANIZED CRIMINAL VIOLENCE AT THE MUNICIPAL LEVEL
ON A DAILY BASIS

Variable Observations Mean Std. Dev. Min. Max.

Violent enforcement 9,868,208 0.0006 0.0327 0 9

Arrests 9,868,208 0.0064 0.1210 0 60

Seizure of assets 9,868,208 0.0021 0.0561 0 16

Seizure of drugs 9,868,208 0.0083 0.1436 0 26

Seizure of weapons 9,868,208 0.0017 0.0582 0 20

Violent competition 9,868,208 0.0048 0.1141 0 36

Criminal retaliation 9,868,208 0.0015 0.0517 0 11

171
 The graphs in Figure 4.2 show that all variables contained in OCVED are dis-
tributed according to negative binomial functions. This type of distribution is typical
of event count data in which the vast majority of the data is concentrated around 0
and low event counts, with only a few observations with a high count. The negative
binomial distribution is a special type of Poisson distribution, with the characteristic
of being highly skewed towards the left. This high degree of skewness is known as
“hyper-dispersion” and occurs when there is a small number of cases containing a
substantially large number of events, thus making the distribution heavily skewed.
This characteristic is clear in all graphs in Figure 4.2. The vertical axis in these
graphs represents the frequency by municipality–days and the horizontal axis repre-
sents the number of events counted in each municipality according to every type of
action included in OCVED. In all these graphs, the largest frequency of municipalities
is concentrated at or near zero and there are a few cases with a large number of event
counts. This indicates that there were no events of violence for most municipality–
days, although there are a few municipality–days where a varying count of events were
observed. These characteristics describe a long tail towards the right in each graph.
In addition to visual analysis, the most straightforward quantitative diagnostic of
hyper-dispersion is to compare the mean and the standard deviation of each variable.
If the estimate of the standard deviation is larger than the mean, it is evidence of
hyper-dispersion. This condition is confirmed in the descriptive statistics of all the
variables reported in Table 4.2, where it can be seen that all standard deviations are
larger than the mean of the respective variable.

172
a monthly basis. The trends of conflict show that clashes between criminal organiza-
tions contribute the largest share of violence. This time series is represented by the
dashed line in Figure 4.3. The process of violence among DTOs can be divided into
three periods. The first stage, between 2000 and 2004, is characterized by a “Pax
Mafiosa” in which criminal organizations conducted their illegal activities without
using significant violence against each other nor against the state. The second stage,
denoted here as the period of “unrest,” indicates early signs of conflict between crim-
inal organizations. This period lasts from 2005 to 2006, when violence among DTOs
broke the long, stable trend of peace that had characterized previous years. The third
stage, referred to as the “escalation,” is characterised by a generalized and sustained
increase of violence among rival criminal organizations at unprecedented levels. This
period lasts from 2007 to at least the end of data collection in 2010. In general,
violent competition among criminal organizations constitute the largest proportion
of violence when compared with the other time series. This indicates that the war
on drugs in Mexico is mostly a war between rival criminal organizations.
The dotted line in Figure 4.3 represents the time series of criminal retaliation
against government authorities or security personnel. The trend in this series can be
divided into two stages. The first period goes from 2000 to 2006 and is termed the
“acquiescence” stage. During this long time span, organized criminals did not commit
systematic violence against state authorities. There is a slight increase in the number
of violent events against the state in 2005 and 2006. However, these were not part
of a sustained challenge against the state. A qualitative assessment of news reports
during this period suggest that violent events against authorities were related to the
period of “unrest.” The early signs of conflict between rival DTOs, were that some
attacks against government authorities or security personnel were directed towards
corrupt officials presumably collaborating with criminal organizations. The second
period, termed “retaliation”, covers from 2007 to the end of data gathering, and is

174
 2,000
Number of monthly events
500 1,000
0 1,500
2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010
Enforcement Competiton Retaliation

Figure 4.3. Main trends of drug-related violence in Mexico

termed a stage of “overt contestation” by criminal organizations against government

authorities. During this period a sustained campaign of criminal violence against the
state can be identified.
Finally, the solid line in Figure 4.3 shows the trend of violent law enforcement
by the state against organized criminals. The time series of state violence can be di-
vided into two stages. The first period goes from 2000 to 2007 and is termed “peaceful
coexistence” between the state and DTOs. Although government authorities used
certain other non-violent tactics to fight criminal organizations, violence was largely
not commonly employed as part of the menu of tactics to fight crime. The second
period ranges from 2007 the end of data gathering and is characterized by “overt
belligerence” on the part of the state toward criminal organizations. In this stage
the use of state violence against organized criminals shows a sustained and increasing
trend. As Figure 4.3 shows, violent law enforcement constitutes the smallest share of
violence when compared to the time series of retaliation and criminal competition. In

175
addition, the figure shows that criminal retaliation exceeds violent law enforcement.
Members of criminal organizations perpetrate approximately 2.5 attacks against gov-
ernment authorities for every event of violent law enforcement conducted by the state
against criminals.
Government authorities have a broad menu of security policies to fight crime.
Figure 4.4 presents the trends of violent and non-violent law enforcement tactics
used by government agencies between 2000 and 2010. The time series of violent
law enforcement is the same as that presented in Figure 4.3 and is included here for
illustrative purposes. When compared to non-violent tactics used against criminals, it
is clear that violent law enforcement constitutes a small proportion (only 3.3 percent)
of the actions used by government authorities to fight crime.

Violent enforcement Arrests Seizure of Assets Seizure of Drugs Seizure of Guns

2000

2000

2000

2000

2000
1500

1500

1500

1500

1500
1000

1000

1000

1000

1000
500

500

500

500

500
0

0
2000m1

2002m1

2004m1

2006m1

2008m1

2010m1

2000m1

2002m1

2004m1

2006m1

2008m1

2010m1

2000m1

2002m1

2004m1

2006m1

2008m1

2010m1

2000m1

2002m1

2004m1

2006m1

2008m1

2010m1

2000m1

2002m1

2004m1

2006m1

2008m1

2010m1

Figure 4.4. Violent and non-violent law enforcement

The analysis of counts of non-violent law enforcement tactics shows that arrests
of suspected members of criminal organizations and interdiction of drugs are the

176
actions most commonly used to fight crime. Arrests represent 43.4 percent of the
total number of state actions, and drug seizures account for 33.5 percent. These
anti-criminal tactics were used extensively even before the onset of the war on drugs
in December 2006, and remain the most used tactics in the war on drugs.
The time series of arrests in Figure 4.4 can be divided into two stages. In the first
period, the administration of President Fox used arrests as a systematic tactic to fight
crime. Between 2000 and 2006, the average number of arrests was 0.004 events per
municipality–day. During this period, there was an increase in the number of arrests
between 2001 and 2004. In 2003, the Mexican army arrested Osiel Cárdenas Guillén,
the leader of the Cartel del Golfo, the most prominent drug trafficking organization
at the time. As will be discussed later in Chapterchapter:zetas, this arrest is crucial
for understanding the Mexican war on drugs. The detention of Cárdenas Guillén
marked the split between the cartel’s armed wing, Los Zetas, a group of deserters
from an elite Army special forces unit, and the rest of the Cartel del Golfo. The wave
of arrests declined between the second half of 2004 and during 2005. Then there was
another increase of arrests during the first half of 2006. This spike in arrests coincides
with the increase of violence among criminal organizations identified in the period of
“unrest” discussed in Figure 4.3.
The second period in the time series of arrests is characterized by a substantial and
sustained increase in the number of detentions. Between 2007 and 2010, the average
number of events where authorities arrested a member of a criminal organizations
was 0.0107, which is 2.7 times greater than the mean during the first stage. This
unprecedented wave of arrests had a significant disturbing effect on DTOs, as it
dismantled their organizational structure by removing several of their key members.
The systematic effort of government authorities to arrest members of criminal groups
also signaled that these organizations were no longer protected by the cohesive corrupt
agreements that they had had with the political elites.

177
 The analysis of the time series of drug seizures in Figure 4.4 shows that inter-
diction of illegal drugs is the most frequently used tactic to fight drug trafficking
organizations. The graph reveals a sustained effort by the Mexican government to
stop drug shipments before they enter the U.S. market. During the Fox adminis-
tration, government authorities largely used interdiction to fight DTOs; this tactic
was used more frequently than any of the other violent and non-violent tactics in
their repertoire. Between 2000 and 2006, the average of drug seizures was 0.006 per
municipality–day. The analysis of the time series reveals an increase and later de-
cline of drug interdiction between 2002 and 2005. This trend is comparable to the
trend in arrests during the same period, thus suggesting that arrests and seizures
are used as complementary non-violent tactics. After the announcement of the war
against organized crime declared by President Calderón upon taking office in Decem-
ber 2006, there was a marked surge of drug interdiction. The number of drug seizure
events dramatically increased during 2007 and 2008, with an average of 0.0146 per
municipality–day, which corresponds to 2.5 times as many seizures as in the previ-
ous period. Later, in 2009 and 2010, the rate of drug seizures declined slightly, but
remained relatively high with an average of 0.0103 events per municipality–day. The
sharp increase and subsequent decline in drug interdiction suggests a change in the
strategy implemented by government authorities in the war on drugs.
The time series of seizures of criminal assets and weapons in Figure 4.4 shows
that these strategies are not widely used by the Mexican government to fight crime,
although both series show a sustained increase starting on 2007. There is a clear
change in the trend of seizures of criminal assets after Calderón took office. In
the period 2000–2006, before Calderón was elected, the average number of asset
seizures was 0.0012 events per municipality–day. In contrast, during the tenure of
his administration from 2007 until the end of data collection in 2010, the average
number of seizures of criminal assets was 0.0036 events per municipality–day, 2.86

178
times higher than in the pre-Calderó period. Gun seizures also show a clear increase
after the war on drugs was launched. Between 2000 and 2006, the average number of
gun seizures was 0.0007 events per municipality–day, but this increased about fivefold
in the period 2007–2010 to an average of 0.0036 events per municipality–day.
In September 2004, the Federal Assault Weapons Ban (AWB) expired, ending a
nation-wide restriction on the production and domestic sales of weapons with military
characteristics3 that had been in force in the U.S. since September 1994. After the
end of the AWB, the production of assault weapons in the U.S. increased rapidly.
According to the Bureau of Alchohol, Tobacco and Explosives (2012), the annual
production of assault weapons went from 1.3 million rifles produced in 2004 to 2.3
million in 2009.

4.3.3
Disaggregating
the
Trends
of
Violence
at
State
Level

This section analyzes the main trends of violence at the state level on a monthly
basis. Rather than focusing on all the different components of OCVED, this section is
explicitly focused on analyzing the dynamics of conflict including violent competition
among criminal groups, criminal retaliation, and violent law enforcement. Therefore,
non-violent law enforcement tactics are not discussed. To facilitate the comprehension
of the figures described in this subsection, readers can refer to Appendix A.4 which
gives a list of Mexican states and their abbreviations.
Figure 4.5 reports the trends of violent competition between criminal organiza-
tions at state level on a monthly basis. The first group refers to the states where

3
Assault weapons, also known as semi-automatic firearms, have several military-like character-
istics. When fired, assault weapons automatically eject the spent cartridge casing and load another
cartridge into the chamber for the next shot. However, these weapons do not fire automatically
as machine-guns do. Assault weapons are also characterized by accepting detachable magazines
with multiple rounds. Another set of characteristics for categorizing a firearm as an assault weapon
include features that allow mounting a bayonet, attaching a grenade launcher or a flash suppressor,
among others.

179
violent confrontations between criminal organizations are most intense. This group is
led by the state of Chihuahua, followed by Coahuila, Michoacán, Nuevo León, Sinaloa
and Tamaulipas. Among this group, Chihuahua reports the most dramatic and sus-
tained increase of violence between criminal organizations. During the entire period,
the average month in Chihuahua reports 67 violent events, and the count ranges from
0 to 347 events per month. Between 2000 and 2006, there were a few incidents of
violence between criminal organizations in Chihuahua; the monthly average event
count in this period ranged between 7.1 and 30.3 per month. These relatively low
figures contrast with the rapid escalation of criminal competition between 2007 and
2010; the count ranges from 64.2 to 221.5 events per month in Chihuahua. The other
states in this group also show high levels of violence between criminal organizations,
although not as dramatic as in Chihuahua. The average number of violent events
in these states per month during the entire observation period are Coahuila 24.8,
Michoacán 28.9, Nuevo León 24.7, Sinaloa 35.9 and Tamaulipas 27.2.
The second group contains other states with high levels of violent competition
between criminal organizations, yet not as intense as in the previous group. This set
of states comprises Baja California, Durango, Jalisco, Estado de México and Sonora,
and the trends of violence between criminal groups in all these states intensify after
2007. The average number of monthly events of violence between 2000 and 2010 in
this group are Baja California 12.7, Durango 14.5, Jalisco 11.2, Estado de México
11.4 and Sonora 11.2.
There is a third group of states with low levels of violent confrontation between
criminal organizations. This group includes the Federal District, Chiapas, Guanaju-
ato, Guerrero, Morelos, Veracruz and Zacatecas. Monthly averages of violent events
among criminal organizations between 2000 and 2010 are Mexico City 5.9, Chiapas
12.2, Guanajuato 7.9, Guerrero 8.6, Morelos 5.4, Veracruz 10.7 and Zacatecas 7.9.

180
 0 100200300400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400
2000m1

2002m1

2004m1

DF

SIN
2006m1

PUE
AGS

HGO

MOR

TLAX
COAH

2008m1

2010m1

0 100200300400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400
2000m1

2002m1

2004m1
BC

JAL
2006m1

VER
NAY
COL

SON
DUR

QRO
2008m1

2010m1

181
0 100200300400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400
2000m1

2002m1

2004m1

on a monthly basis
NL
2006m1

TAB
BCS

YUC
MEX
GTO
CHIS

QROO
2008m1

2010m1

0 100200300400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400 0 100 200 300 400
2000m1

2002m1

2004m1
SLP

2006m1

ZAC
TAM
OAX
GRO
CHIH

MICH
CAMP

2008m1

2010m1

Figure 4.5. Violent competition among criminal organizations at state level

 Finally, there is a fourth set of states that have barely been affected by violence
between criminal organizations. This group includes the states of Aguascalientes,
Baja California Sur, Campeche, Colima, Hidalgo, Nayarit, Puebla, Querétaro, Quin-
tana Roo, San Luis Potosı́, Tabasco, Tlaxcala and Yucatán. The average numbers of
events of violence between criminal organizations per month observed in these states
are Aguascalientes 1.7, Baja California Sur 0.7, Campeche 1.3, Colima 2.1, Hidalgo
3.8, Nayarit 3.1, Puebla 3.4, Querétaro 1.5, Quintana Roo 2, San Luis Potosı́ 4.4,
Tabasco 2.4, Tlaxcala 1.3 and Yucatán 2.6.
Figure 4.6 shows the trends of violent retaliation perpetrated by criminal organi-
zations against government authorities at state level on a monthly basis. Based on the
frequency of criminal attacks against the government, it is possible to divide states
into three main groups. The first group comprises six states with the highest levels of
violence against the state and includes Chihuahua, Tamaulipas, Nuevo León, Sinaloa,
Michoacán and Coahuila. Among these states Chihuahua is the most dangerous place
for authorities and security personnel. The average number of retaliation attacks in
Chihuahua is 18.22 events per month in the entire period of observation. This general
average ranges from 0.5 monthly events in 2001 to 57.9 events per month in 2010.
The mean counts of retaliation against government authorities in the other states are
Tamaulipas 11.90, Nuevo León 9.77, Sinaloa 9.45, Michoacán 8.37 and Coahuila 7.42.
Among these sub-national units, Tamaulipas presents an unusual trend of violence
against the state. In 2010 there is a remarkably high spike of aggressions against
government authorities that generated an average of 43.8 monthly events that year.

182
 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150
2000m1

2002m1

2004m1

DF

SIN
2006m1

PUE
AGS

HGO

MOR

TLAX
COAH

2008m1

2010m1

0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150
2000m1

2002m1

2004m1
BC

JAL
2006m1

VER
NAY
COL

SON
DUR

QRO
2008m1

2010m1

183
0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150
2000m1

2002m1

2004m1
NL
2006m1

TAB
BCS

YUC
MEX
GTO
CHIS

QROO

level on a monthly basis

2008m1

2010m1

0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150 0 50 100 150
2000m1

2002m1

2004m1
SLP

2006m1

ZAC
TAM
OAX
GRO
CHIH

MICH
CAMP

2008m1

2010m1

Figure 4.6. Criminal retaliation against government authorities at state

 There is a second group of states which show only a few monthly events of crim-
inal retaliation against government authorities. This group includes the states of
Baja California, Durango, Jalisco, Sonora, Estado de México, Veracruz, Chiapas,
Guanajuato, Zacatecas, Guerrero, Morelos, San Luis Potosı́, Federal District, Aguas-
calientes, Nayarit and Puebla. The average number of attacks against government
forces in this group ranges from 1.1 events in Puebla to 4.6 events in Baja California.
Finally, there is a third group where the monthly average of retaliation events against
the government is less than one per month. This group includes Tabasco, Hidalgo,
Yucatán, Colima, Quintana Roo, Oaxaca, Querétaro, Campeche, Tlaxcala and Baja
California Sur.
Figure 4.7 reports the main trends of violent law enforcement against criminal
organizations at state level on a monthly basis. The first group of states includes
those where the government deploys the most intense use of violence to fight criminals.
This group includes the states of Chihuahua, Tamaulipas, Nuevo León, Michoacán,
Coahuila, Sinaloa, Veracruz and Jalisco. The states of Chihuahua and Tamaulipas
report the largest monthly average of violent law enforcement with 6.1 and 5.7 events
respectively. However, the trends of violence are different in these two states. In
Chihuahua, violent law enforcement shows a progressive intensification of government
use of violence with some fluctuations, which reached a maximum of 37 events per
month in 2010. The use of state violence in Tamaulipas is marked by a notable spike
in 2010 that reached 103 events per month.
The second group is composed of states with 1 to 2 events of violent enforcement
per month. The states in this group are Aguascalientes, Baja California, Chiapas,
Durango, Guanajuato, Estado de México, Sonora and Zacatecas. Finally, the third
group includes the remaining states which average fewer than one monthly event of
violent law enforcement.

184
 0 20406080100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100
2000m1

2002m1

2004m1

DF

SIN
2006m1

PUE
AGS

HGO

MOR

TLAX
COAH

2008m1

2010m1

0 20406080100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100

2000m1

2002m1

2004m1
BC

JAL
2006m1

VER
NAY
COL

SON
DUR

QRO
2008m1

2010m1

185
0 20406080100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100
2000m1

2002m1

2004m1
NL
2006m1

TAB
BCS

YUC
MEX
GTO
CHIS

QROO
2008m1

2010m1

0 20406080100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100

2000m1

2002m1

2004m1
SLP

2006m1

ZAC
TAM
OAX
GRO
CHIH

MICH
CAMP

2008m1

Figure 4.7. Violent law enforcement at state level on a monthly basis

2010m1
4.3.4
Main
Trends
of
Violence
at
Municipal
Level
on
a
Daily
Basis

Figure 4.8 disaggregates the main trends of daily violent events at municipal
level. The first panel shows the scatter plot of violent competition among criminal
organizations by municipality–day. The second panel shows the daily distribution of
criminal retaliation against government authorities at municipal level. Finally, the
third panel presents the deployment of violent law enforcement against crime on a
daily municipal level.

Figure 4.8. Main trends of violence at municipal level on a daily basis

The data illustrated in the figure confirms that violent competition among crim-
inal groups constitutes the most substantial source of violence, in comparison to
criminal retaliation and violent law enforcement. The first panel shows a massive
increase in observations reporting higher numbers of daily events of violence among
criminal organizations. Between 2000 and 2006, there is a slow but sustained incre-

186
ment in violence. During this period, the trend is pulled upward by a few outliers
of more than five events of violent competition per day. However, after 2007 there
is a large cloud of observations spreading towards higher levels of violence among
criminal groups. In the last years of the observation period, the first panel shows
a larger number of values with uncommonly high levels of violence between rival
criminal organizations.
The second panel shows a steady but continuous increment in the number of
municipalities reporting more intense criminal retaliation against government au-
thorities. This panel indicates that there is only a handful of cases with more than
five attacks perpetrated by criminal organizations against the state per municipality–
day. Finally, the third panel shows that the trend of violent law enforcement consists
of gradual slight growth in the number of observations where the state uses force
to fight crime. The third panel also reveals that there is only an small number of
municipalities where government authorities deploy the most intense use of violence
against criminal groups.

4.3.5
Spatial
Trends
of
Violence
Between
Criminal
Organi-

zations

The overview of spatial trends of conflict presented in this section is only focused
on the dynamics of violence between one criminal organisation and another. Figure
4.9 shows the cumulative spatial concentration of violence between criminal organi-
zations at the municipal level in 2000, 2005 and 2010.4 The maps indicate hotspots
of violence identified through kernel density functions based on the concentration of
geo-referenced data points within a range of 50 km. Kernel density functions assign
higher values to areas where there is a higher concentration of data points. The ker-

4
A three-dimensional animation of the spatial concentration of violence among criminal orga-
nizations is available at http://www.youtube.com/watch?v=d8ObscVMJmE. The video shows the
spatial and temporal dynamics of violent competition between rival DTOs from 2000 to 2010.

187
nel density functions are then integrated into a raster plot of continuous information
that can be plotted in a three-dimensional space. A higher elevation in the raster
plot indicates more intense violence between criminal organizations in that specific
area. For visualization purposes, maps in Figure 4.9 show Mexico’s political divisions
at the state level, but the kernel density functions are estimated with data at the
municipal level. To facilitate the identification of each state, readers may refer to
Appendix A.5 showing the sub-national division of the country.
The spatial analysis of violence between criminal organizations in Figure 4.9 re-
veals two distinct but overlapping processes of intensification and spread of violence.
The upper map in the figure shows that in 2000 there are hardly any significant
clusters of conflict between criminal organizations. The map in the middle of the
figure shows that in 2005 there are some areas with early signs of violence between
criminal groups. Finally, the bottom map shows that by 2010 violent competition
between rival criminal organizations had spread across the country at varying levels
of intensity.
In 2005, violent competition between rival criminal organizations is concentrated
in a few isolated hotspots scattered along the U.S.-Mexico border, the Pacific coast
and some areas in the center of the country. Violence along the border seems to cluster
in the northeast and affects the states of Tamaulipas, Nuevo León, and Coahuila.
There are also some scattered spots of violence in the north-central region of the
country affecting the state of Chihuahua. Additionally, there are some early signs
of violence in the northwest, particularly in the state of Baja California. The 2005
map also shows some scattered spots of violence along the Pacific coast affecting the
states of Sinaloa, Jalisco and Michoacán. There is another cluster of violence in the
center of the country covering some areas of Estado de México, the Federal District,
and Morelos.

188
Figure 4.9. Spatial trends of violence among criminal organizations

189
 In 2010, the wave of violence between criminal organizations has spread across
the entire country. The prominent spikes of violence in the 2010 map in Figure 4.9
indicate that there are some areas with particularly high concentrations of conflict
between criminal organizations in the north region and on the northern Pacific coast.
Among those highly conflictive areas, it is possible to identify two trends. In Baja
California and Chihuahua, specifically in Ciudad Juárez, violence intensifies without
spreading to neighboring areas. In contrast, spikes of violence in Sinaloa and the
northeast are characterized by both intensification and a substantial spill-over of
violence to neighboring regions. In addition, the map shows that there are areas in the
south and central regions of Mexico where violence between criminal organisations
is widely diffused throughout the region, although not as intense as in the north.
This spillover affects the states of Jalisco, Michoacán, Guerrero, Estado de México,
Querétaro, Guanajuato, Hidalgo and Mexico City.
The maps presented in Figure 4.9 only show three different times separated by
several years. Moreover, the 50 km. radius of the kernel functions may be useful for
purposes of overall visualization of patterns in violence, but it obscures more specific
variations of violence at a smaller scale. For that reason, these maps are limited for
showing the more nuanced processes of violence. Nevertheless, they provide valuable
insights for identifying broad trends of temporal and spatial dynamics of violent
competition between criminal groups.

190
 CHAPTER
5

THE
EMERGENCE,
CONSOLIDATION

AND
COLLAPSE
OF
ORDER

“Violence is the last refuge of the incompetent”

Isaac Asimov

5.1
Introduction

The formal model presented in Chapter 2 is based on a series of assumptions

about the relationship between political actors and criminals. These assumptions
serve as the building blocks for the mechanisms of the model, and are then used to
derive a set of observable implications from the theoretical explanation, which must
be tested on the basis of empirical evidence. One of the central implications of the
theoretical model is that violence is the consequence of the collapse of order. Located
within the Hobbesian tradition of conflict research, the argument indicates that in
contexts of authoritarian regimes, the relationship between political actors and crim-

191
inals is characterized by peaceful agreements sustained on the basis of corruption
and the stability of the political regime. But as democratization gradually erodes
the foundations of the authoritarian regime, the system of incentives no longer favors
political alliances with criminal organizations and motivates law enforcement against
crime. As a result, the previous order collapses and leads to the escalation of conflict
characterized by the state using violence against crime, criminal organizations re-
sponding with violence against the authorities and, most importantly, the escalation
of a wave of violence among rival criminal organizations. Violence thus emerges from
the collapse of the preexisting order.
The theoretical model analyzes the interactions between the state and criminal
organizations in two different settings. The first scenario is characterized by the lack
of law enforcement in an authoritarian regime and the peaceful coexistence between
criminals and state authorities. The second scenario is defined by proactive law
enforcement efforts against criminals, which trigger the escalation of drug related
violence. The transition from one scenario to the other is represented by the central
hypotheses about the onset of the war on drugs:

(H1 ) Increased democratization is positively associated with higher levels of law

enforcement against organized crime.

Why do politicians display different behavior towards criminals under an author-

itarian rule than in a democratic regime? The central argument is that democratiza-
tion disrupts the structure of political incentives favoring corrupt agreements under
authoritarianism and motivates government authorities to enforce the law in a demo-
cratic setting. The process of democratization alters the the system of incentives by
first eroding the feasibility of corrupt agreements and then by motivating the enforce-
ment of the law against criminals. The erosion of corrupt agreements is caused by two
factors. First, democratization increases the number of relevant political actors at
different levels of government, thus increasing the difficulty of solving problems of col-

192
lective action necessary for establishing corrupt agreements. In a democratic regime,
the plurality of political actors undermines the coordination necessary to establish
pacts with criminals. Second, democratization facilitates the effective circulation of
political elites and introduces uncertainty about the results of the electoral process,
thus reducing the stability of corrupt agreements over time and increasing the costs
of establishing new agreements in the future. In a democratic system, certainty about
the limited tenure of political office and uncertainty about the next person in office
reduce the duration and stability of corrupt agreements. This is particularly impor-
tant for the Mexican case because the national constitution does not allow re-election
for executive offices (president, governors and mayors), nor does it allow consecutive
reelection of legislators.
Besides corroding the feasibility and sustainability of non-aggressive pacts be-
tween corrupt politicians and criminals, the most significant effect of democratization
is to inculcate motivation for government authorities to fight crime. The process of
democratic transition moves the prospects of political survival away from the deci-
sion of authoritarian political elites and places the likelihood of political success on
the favor of broad sectors of the electorate, which depends on the provision of public
goods. Under democracy, political actors have personal incentives to enforce the law
as a form of public good. This incentive is stronger for new politicians who want
to differentiate themselves from former politicians who might have alliances with
criminals.
To test the argument, this chapter examines the Mexican case from a process
tracing perspective. The historical assessment analyzes the process that favored
the emergence of order out of the Revolution, the factors that contributed to its
consolidation under an authoritarian regime, and the collapse of authoritarian order
caused by the process of democratic transition. The chapter is divided into five
segments.

193
 The first section makes explicit the set of assumptions behind the hypothesis of
democratization, thus improving the transparency of the argument and facilitating
the identification of these factors in the subsequent segments. The second section an-
alyzes the conditions that favored the emergence of political order out of the Mexican
Revolution. This segment also identifies the domestic and international factors that
facilitated the engagement of political actors in drug-related activities. The third
section analyzes the consolidation of political order under authoritarianism. This
section challenges mainstream explanations of non-aggressive agreements between
politicians and criminals based primarily on corruption and selective enforcement.
Economic motives and the use of coercion are certainly important but do not con-
stitute a complete explanation. This section argues that political threats against
the state characteristic of Cold War politics justified the involvement of the state
security apparatus in the criminal underworld. Such involvement not only offered
opportunities for personal peculation but, most importantly, it enabled monitoring
of clandestine activities that could threaten the survival of the regime. In addition,
the involvement of political actors in drug trafficking gave them a hold over crimi-
nals through a system of political incentives rather than relying on the use of force.
The fourth segment analyzes the process of democratization based on the interaction
between the electoral system and the party system. The analysis shows how minor
electoral reforms enabled the operation of opposition parties, who in turn pushed
for broader reforms. The protracted process of democratic transition undermined
the hegemony of the authoritarian regime. Democratization increased the number
of relevant political actors at all levels of government, introduced certainty about
the electoral rules and uncertainty about election results. This generalized erosion
disrupted the ability of the authoritarian regime to monitor criminals and to instill
discipline through a system of political incentives. The precarious equilibrium con-
ditions collapsed as the Mexican government intensified law enforcement activities,

194
thus triggering a wave of violence of all-against-all. Finally, the last section presents
the conclusions.

5.2
Explicit
Assumptions
about
Democratization and
Law
Enforcement

The central argument about the onset of the war on drugs indicates that democ-
ratization erodes peaceful configurations between corrupt politicians and criminals
and motivates government authorities to fight crime. This argument is based on the
following set of assumptions about the different behavior of political actors in the
context of authoritarian regimes and under democratic rule.

Order under authoritarian regimes

The construction of order under authoritarian regimes is based on the following
assumptions of the behavior of political actors and criminals:

Assumption 1. In the context of an authoritarian regime, political authorities

coexist with organized criminals in a peaceful configuration.

Assumption 2. The political benefits for enforcing the law under authoritari-
anism are lower than the benefits of corruption obtained by collaborating with
criminals. This is represented by the relationship G < B discussed in Section
2.3.2, where G represents the political benefits of enforcing the law as a public
good and B represents the benefits from bribes received for not enforcing the
law.

Assumption 3. Order is maintained in part by the threat of coercion from an

authoritarian state, but most importantly, by the political incentives generated
throughout the hierarchical and centralized political structure of the regime.

Assumption 4. The small number of relevant political actors characteris-

tic of authoritarian regimes favors peaceful arrangements between government
officials and criminals.

Based on the theory of collective action advanced by Olson (1965), it is

possible to extend the assumptions for a reduced number of political actors
and their relationships with criminals in the following terms:

195
 Assumption 4.a. A reduced number of political actors makes pacts
more feasible because it reduces the transactions and information costs
for bargaining with organized crime and reduces the costs of bribes that
criminals have to pay.
Assumption 4.b. A smaller number of political actors facilitates the
enforcement of the corrupt agreement within the structure of the state.
It makes it easier for corrupt actors to coordinate between each other. It
also makes it easier for them to detect defectors who might take bribes
without the approval of the rest of the group.
Assumption 4.c. The cohesiveness of an authoritarian chain of command
generates behavior similar to that of a reduced number of political actors,
thus facilitating compliance by the lower ranks in the hierarchy to the
terms of the corrupt agreement procured at the top of the hierarchy.

Assumption 5. The lack of effective circulation of the political elite provides

extended time horizons for both criminals and political actors, thus generating
credible expectations about the stability of the pact.

Based on the analysis of strategic cooperation made by Schelling (1960),

this assumption can be further developed in the following terms:
Assumption 5.a. Extended time horizons reduce the costs of bargaining
between criminals and corrupt politicians because once an initial agree-
ment is achieved, actors are likely to keep cooperating if the expiration of
the agreement is not foreseen in the short run.
Assumption 5.b. Extended time horizons increase the incentives for
criminals to cooperate because the advancement of their criminal career
and future economic benefits are likely to be conditional on their compli-
ance with the pact in the present.
Assumption 5.c. Extended time horizons provide incentives for polit-
ical actors at the top of the hierarchy to comply with the terms of the
agreement because of the expectation of receiving economic benefits in
the future without suffering sanctions from the electorate.
Assumption 5.d. Extended time horizons increase the incentives for
political actors in the middle and lower ranks of the political hierarchy to
comply with the commands of the political elite because the expectation of
patronage and their future political career depends on receiving the favor
of the elite.

Democratization and the collapse of the preexisting order

196
 According to the theoretical model, democratization erodes the peaceful arrange-
ments between corrupt politicians and criminals characteristic of authoritarian regimes.
This argument is based on the following assumptions:

Assumption 6. Democratization erodes the non-aggression agreements be-

tween corrupt politicians and organized criminals that are likely to exist under
authoritarianism.

Assumption 7. The political benefits for enforcing the law in a democratic

regime are larger than the benefits of corruption obtained from non-enforcement.
This is represented by the relationship G > B mentioned above and discussed
in Section 2.3.2.

Assumption 8. The decentralized and fragmented structure of political re-

lationships in a democratic system does not favor the imposition of order on
criminals based on political incentives.

Assumption 9. Democratization increases the number of relevant political

actors, thus eroding the feasibility of peaceful arrangements between corrupt
government officials and organized criminals.

This assumption can be further extended in the following terms:

Assumption 9.a. A larger number of political actors makes it more dif-
ficult to overcome problems of collective action by increasing transactions
and information costs for bargaining with criminals and increasing the
amount of bribes that criminals have to pay.
Assumption 9.b. An increased number of political actors makes it more
difficult for corrupt politicians to coordinate between each other. It also
hinders their capability to detect free riders who might take bribes pri-
vately without the approval of the rest of the group.
Assumption 9.c. Democratization fragments the chain of command
across different levels of government, thus reducing the prospects of com-
pliance with a corrupt agreement along the entire chain of command.

Assumption 10. Effective circulation of the political elite through electoral

means increases the uncertainty about the winner of the next election (Prze-
worski, 1991), thus reducing the temporal horizons and stability of non-aggression
agreements between corrupt politicians and organized criminals.

Assumption 10.a. Elite circulation reduces the duration of corrupt

agreements and increase the costs of negotiating new agreements following
the election of the new elite.

197
 Assumption 10.b. Short time horizons reduce the incentives for crimi-
nals to establish corrupt agreements with the state because elite circulation
reduces the certainty about their future economic benefits.

Assumption 11. Increasing political competition motivates politicians to en-

force the law against organized crime.

Assumption 11.a. Increasing political competition motivates govern-

ment authorities to provide public goods, including the provision of public
security, in order to gain electoral support.
Assumption 11.b. Increasing political competition provides personal
incentives for new politicians to fight crime as an effort to distinguish
themselves from previous corrupt politicians.
Assumption 11.c. Increasing political competition generates incentives
for politicians to avoid corrupt agreements with criminals because they
can be sanctioned by the electorate.

5.3
The
Emergence
of
Order
out
of
the
Mexican

Revolution

This section analyzes the historical process that gave rise to the creation of po-
litical order after the prolonged violent struggles of the Mexican Revolution. In this
perspective, order emerged as a political pact between the main actors capable of
conducting organized violence. The main objective of the agreement was to termi-
nate the recurrent bloody battles among revolutionary leaders. This political pact
set the foundations for the creation of the Institutional Revolutionary Party (Partido
Revolucionario Institucional, PRI) as the main mechanism for managing access and
the peaceful exercise of power. To do so, the political pact recognized local strong-
men and allowed them to enjoy a substantial degree of autonomy in their local affairs
as long as they remained loyal to the political elite in the central government. The
relative autonomy of the periphery gave local leaders the opportunity to reap the
economic benefits from illicit markets created by prohibition laws in the U.S. at the
beginning of the twentieth century.

198
 The Mexican Revolution involved the mass mobilization of insurgents fighting for
a variety of political and economic reasons. Starting with the November, 1910 upris-
ing led by Francisco I. Madero against the long-lasting dictatorship of Porfirio Dı́az,
the Mexican Revolution soon combined several insurrections advancing democratic,
socialist, anarchist, liberal, populist and agrarian demands (Knight, 1990a,b). Dur-
ing the Mexican Revolution, the primary way to access power was through the use of
arms. The national political scene witnessed the rise and fall (often by bullets) of sev-
eral presidents and strongmen. Local politics were characterized by similar dynamics.
Revolutionary generals became governors due to their service in battle and loyalty to
those higher up in the military and political structure, but they often fell in disgrace
as fast as they had entered power. A sequence of presidential assassinations, military
coups, and reactionary movements marked the instability of national politics. Such
volatility also extended to the aftermath of the revolutionary war. Alliances emerged
frequently and just as quickly dissolved under gunfire only to reappear in a different
configuration. Due to the instability of politics in the heart of the country, generals
who rose to power in the periphery enjoyed a substantial degree of autonomy from
the central government.
Efforts to regulate the production and consumption of drugs and alcohol in the
U.S. coincided with periods of volatile political conditions in Mexico. Under these
circumstances, some governors, especially those in the northern part of the country,
saw an opportunity of enriching themselves thanks to the profitable illicit markets
opened by U.S. prohibition laws. Governing with the violent skills acquired during
the revolution, these governors imposed political order within their domains and
jealously controlled the main drug and alcohol smuggling networks on the Mexican
side of the border.
Due to the recurrent sequence of violent confrontations among different factions,
President Plutarco Elı́as Calles, a general who had emerged as the “maximum boss of

199
the revolution,” created the National Revolutionary Party, (Partido Nacional Rev-
olucionario, PNR) in 1929. This party eventually became the Institutional Revo-
lutionary Party (PRI), which managed to remain in power without interruption for
more than seven decades. The party gathered together and incorporated all the main
military and political forces of the country, including new and old generals who made
it through the revolution, but also large sectors of peasants and workers that emerged
as key political actors in world politics after the Russian Revolution. All the varied
sectors contained within the party agreed to solve their differences through peaceful
means. The party served as the central structure for regulating the use of and access
to power. The PNR thus became the main political mechanism for halting the pro-
longed era of bloodshed of the Mexican Revolution and setting up the corner-stone
for the process of state-making in modern Mexico.

5.3.1
Early
Days
of
Drug
Traffic
in
Mexico

Marijuana and poppy had been cultivated and trafficked in Mexico since at least
the last quarter of the nineteenth century (1886) (Astorga, 2005). After the U.S.
passed the Chinese Exclusion Act of 1882 prohibiting the legal immigration of Chinese
male workers to the United States, Chinese immigrants went to Mexico in search of
economic opportunities. They quickly settled in the northwest part of the country, in
an area known as the North and South Territories of Baja California1 and in the state
of Sinaloa. There was another wave of Chinese immigrants who settled in Mexicali,
Baja California, after the Great San Francisco earthquake in 1906. By 1910, Chinese
had settled in almost every state but were mainly concentrated in the states located
along the Mexican Pacific coast.

1
Baja California was a territory divided into two jurisdictions: the Northern Territory of Baja
California, which became a state in January 16, 1952 under the name of Baja California, and the
Southern Territory of Baja California, which joined the Mexican federation as a state in October 8,
1974 as Baja California Sur.

200
 The Chinese immigrants were prosperous merchants in legal businesses but they
also thrived as opium producers and traders. Of course, drug trafficking was not an
activity exclusive to these immigrants. The local population also participated in the
production and transportation of poppy and marijuana. Drug trafficking followed
the same commercial corridors as legal trade along the Pacific states up to the border
cities of Nogales, Mexicali and Tijuana into the U.S. According to Astorga (1999),
those who participated in the cultivation and trafficking of marijuana and opium
were known as “gomeros” and were regular members of their societies, usually living
in small villages; they were not considered to be outlaws. Coexistence with drug
producers and traffickers wasx acceptable locally because of the absence of drug use
and abuse in Mexico. Opium and marijuana were produced for the market abroad,
not for local consumption, except perhaps for a few people who may have used
marijuana for medicinal or recreational purposes.
Between 1911 and 1919, in the midst of the Mexican Revolution, there were several
xenophobic movements against the Chinese population in Mexico. The anti-Chinese
campaign escalated in 1930s and led to the expulsion of the Chinese from Sonora and
other states. By 1926, the Chinese comprised the second largest immigrant group
in all of Mexico with more than 24,000 people. Ten years later, there were only
4,856 Chinese in the country (Chao Romero, 2010, 175). With the displacement
of the Chinese, Mexicans took over the production and transportation of opium.
According to Knight (2012, 124), just as with oil and mineral resources, drugs became
an example of the old nationalist adage of “Mexico for the Mexicans.”

Colonel Cantú in Northern Baja California

After the Mexican Revolution broke out in 2010, a group of anarchists led by
the Flores Magón brothers conducted a military campaign to control the Northern
Territory of Baja California. In reaction, Francisco I. Madero sent Colonel Esteban
Cantú to suppress the anarchists and recover control of the peninsula. Cantú secured

201
the territory in June, 2011 and remained there as military commander. Two years
later, Victoriano Huerta led a successful conspiracy to overthrow and kill President
Francisco I. Madero. After the assassination of Madero, Colonel Cantú quickly rec-
ognized the authority of Victoriano Huerta who allowed him to stay in command of
the armed forces in Northern Territory of Baja California and declared him governor
of the territory.
In December, 1914, the U.S. passed the Harrison Narcotics Tax Act. This law
regulated and imposed a tax on opiates and their derivatives. The Harrison Act
was the domestic expression of an international campaign led by the U.S. to control
opium after the Shanghai Commission of 1909 and the Hague International Opium
Convention of 1912 (United Nations Office on Drugs and Crime, 2008a,b). The
prohibition of opiates on the U.S. side of the border created favorable conditions
for illegal trade in drugs, and Colonel Cantú did not miss the opportunity to enrich
himself. Taking advantage of his position of power, Cantú used his good relations
with the Chinese community in Mexicali, Baja California to build a network of opium
producers and traffickers. In this way, Cantú – reputedly an addict – became the
first Mexican drug dealer (Astorga, 2003; Knight, 2012).
After the assassination of Madero, General Venustiano Carranza created the Con-
stitutional Army to oust Huerta. Carranza joined forces with other prominent rev-
olutionary caudillos (leaders) such as Emiliano Zapata, Francisco Villa and Álvaro
Obregón. The Constitutional Army defeated Huerta and sent him to exile in 1914.
To accommodate the new political configuration, Colonel Esteban Cantú recognized
Carranza as president of the country, thus gaining his favor and managing to stay in
his position of governor of Baja California.
The conditions for economic prosperity of Colonel Cantú improved in January,
1920 when the Eighteenth Amendment and the Volstead Act went into effect in the
U.S. These regulations banned the sale, production and transportation of alcohol.

202
To enforce them, the U.S. federal government commissioned 1,520 police officers.
However, Colonel Cantú did not manage to reap the benefits of the Prohibition era
because of the assassination of President Venustiano Carranza in May of the same
year. The incoming president, General Álvaro Obregón, discovered that Cantú was
part of an attempt to detach Baja California as an independent state, and sent a
military expedition led by General Abelardo L. Rodrı́guez to reaffirm the authority
of the central government (Kenny and Serrano, 2012a; Maurer, 2012). After sending
Cantú to exile, General Rodrı́guez expelled the Chinese community that he had
protected from Mexicali.
After the Mexican government banned marijuana in 1920 and opium in 1926, drug
traffickers needed more protection than they had before. This was not a concern
for Abelardo L. Rodrı́guez because his loyalty to the central government won him
the appointment of Military Commander of Northern Baja California in 1921 and
the governorship of the state in 1923. With his position consolidated, Abelardo L.
Rodrı́guez installed himself as a monopolist of the opium and alcohol trade to the U.S
in northwestern Mexico, which enabled him to amass a substantial fortune (Kenny
and Serrano, 2012a). He was later ratified as governor by President Calles and served
as Minister of Industry, Trade and Labor as well as Minister of War for President Ortı́z
Rubio. The combination of wealth, astute political skills, military experience, and
government performance at the local and federal levels led him to become president of
Mexico between 1932 and 1934. Abelardo L. Rodrı́guez is considered the first Mexican
president directly related with drug trafficking (Cruz, 2008; Muedano, 2011).

General Quevedo in Chihuahua

In 1932, General Rodrigo M. Quevedo Moreno, a close friend of President Abelardo
L. Rodrı́guez, became governor of Chihuahua, a position that he held until 1936. Dur-
ing the early days of the revolution, General Quevedo had fought alongside Francisco
I. Madero but later on participated in the coup that killed Madero. He managed

203
to navigate the turbulent waters of revolutionary politics and became the cacique
(strongman) of Chihuahua and the central figure of the Quevedo political camarilla
(clique). As governor, General Rodrigo Quevedo consolidated his family’s position of
power by imposing this brother, Jesús Quevedo, as mayor of Ciudad Juárez in 1932,
where he took control of customs revenues, car theft from the U.S., prostitution and
gambling on the Mexican side, and drug and alcohol smuggling to the U.S. Later, in
1936, General Rodrigo Quevedo installed another member of his family, José Quevedo
Jr., as mayor of Ciudad Juárez.
As part of his administration as governor, General Quevedo appointed Francisco
Rodrı́guez as Minister of the Interior, who also served twice as interim governor
of Chihuahua 1932 and 1933. Before his appointment, Francisco Rodrı́guez had
had a political career as mayor of Juárez between 1921 and 1922. However, he
was forced to resign after a dispute with the governor of Chihuahua at the time
and President Álvaro Obregón over the dispersal of funds from the city’s gambling
concession (Wasserman, 1993, 134).
After leaving the state governor’s office, General Rodrigo Quevedo kept a firm
hand on his nepotistic network and family business while serving as the chief of several
military zones for more than twenty years (1936–1958). As part of his extended
control over Chihuahua, General Quevedo imposed Carlos Villareal Ochoa as mayor of
Juárez in 1947. Villareal was married to the daughter of former mayor José Quevedo,
and he carried on the long-standing Quevedo family business and rule, using his
experience as a federal policeman. As part of his operation, he even established a
secret municipal police force dedicated to drug trafficking to the U.S. (Cruz, 2008;
Kenny and Serrano, 2012a).

204
5.3.2
The
Post
Revolutionary
Era

The cases of Colonel Cantú in Baja California and General Quevedo in Chihuahua
are two examples of a large number of revolutionary leaders who consolidated enough
political power to reap personal economic benefits from their strongholds. However,
there was a key difference between Cantú and Quevedo that largely determined the
diverging trajectories of these two actors. During the revolution, political survival
depended not only on the generals’ military abilities but perhaps even more on their
skill for building and adapting political alliances in highly volatile circumstances.
Colonel Cantú failed to adapt to the rapidly changing political conditions, which
led to the end of his political career. In contrast, General Quevedo coped with
instability and recognised a longer term trend towards the centralization of political
power. Abelardo L. Rodrı́guez was perhaps even more skillful not only in identifying
the trend of centralization but in actively contributing to its consolidation.
Revolutionary leaders defeated the dictatorship of Porfirio Dı́az but they were
trapped in a series of factional conspiracies that killed nearly all the caudillos. Mex-
ico managed to overthrow the preexisting order in a way similar to the way the
Chinese and the Soviet revolutions did. However, in contrast to the these revolution-
ary paradigms, Mexico did not have a preexisting institutional structure capable of
guiding the construction of the new order after the revolution. The breaking point in
the establishment of a new political order in Mexico was the creation of the National
Revolutionary Party (Partido Nacional Revolucionario, PNR) in 1929 (ten years after
the end of the revolution). The central idea was to create a party to unify all mem-
bers of the large “familia revolucionaria” (revolutionary family) in order to regulate
access to power in a stable and peaceful manner.
The creation of the PNR did not follow the decentralized model of U.S. party
conventions but the vertical, centralized national structure of communist parties.
According to Medina Peña (1994), the organic configuration of these parties was

205
more effective for the purpose of unifying and instilling discipline in the unpredictable
familia revolucionaria. The first proponent of the idea of a large unifying party was
General Álvaro Obregón, who held the presidential office from 1920 to 1924, but was
assassinated shortly after winning the presidential election for the second time in
1928. Obregón’s Minister of the Interior, General Plutarco Elı́as Calles, followed the
steps initiated by his predecessor and consolidated the creation of the PNR. Using his
political stature as “maximum boss of the revolution,” Calles convinced the political
class that their interests and ambitions would be better served within a large alliance
that would help curb the destructive tendencies brought about by factious struggles.
The statutes of the PNR centralized authority under the National Executive Com-
mittee (Comité Ejecutivo Naconal, CEN) but recognized State and Territorial Direc-
tive Committees and gave them autonomy “in every issue related to local affairs”
(Medina Peña, 1994, 72). The most important characteristic of the party structure
was the charge given to the National Executive Committee to “serve as harmonizer
and arbitrator of the controversies and difficulties occurring among the organs of the
Party” (Partido Nacional Revolucionario, 1929, Article 45, Fraction VII). This was
the origin of the PNR, a party created as an alliance for peacefully managing and
distributing power quotas at the national and local level. As a unifying structure,
the party attempted to include all relevant political actors. In consequence, the PNR
did not have a clear or concrete ideological platform but an array of broad political
purposes to which all different groups would agree. The inclusiveness and encom-
passing character of the party did not tolerate opposition. If there was any, it was
labeled a reactionary movement contrary to the unifying purpose of the party. This
was sufficient to disqualify the opposition and to defeat it by the use of force.
The main innovation of the PNR was the creation of mechanisms of conflict
resolution that permitted autonomy to local actors but imposed greater discipline and
demanded greater loyalty towards the center. At the local level, these mechanisms

206
were implemented by the sub-national managing committees and at the national level
by the National Executive Committee. However, the central guarantee of stability
relied on a supreme arbitrator with the ability to dampen dissent or to impose consent
if necessary; the president of the country. By this means, the institutional structure
of the party gave the president de facto authority over the entire political system.
The party became the cornerstone of state-building in post-revolutionary Mexico.
In 1938, General Lázaro Cárdenas incorporated the labor and peasant sectors into
the PNR and changed its name to Party of the Mexican Revolution (Partido de la
Revolución Mexicana, PRM). A few years later, his successor, President Manuel Ávila
Camacho, gave the party its current name of Partido Revolucionario Institucional ;
PRI (Institutional Revolutionary Party) in 1946. Between the 1930s and the 1950s,
the “revolutionary family” ruled with absolute hegemony while running a fairly stable
and predictable political system.
Rodrigo Quevedo in Chihuahua and Abelardo L. Rodrı́guez in Baja California
are key examples of politicians who knew how to navigate the waters of the new
political pact. They ruled their respective strongholds with an iron fist and had
sufficient autonomy in their own legal and illegal businesses, but at the same time they
respected the new rules of elite circulation and remained loyal to the “revolutionary
family.” After stepping down from their respective administrations as governors, they
took care to carefully designate their successors and moved up in the political ladder.
As skillful and respected members of the political elite, General Quevedo rotated
positions as chief of several military regions for more than twenty years, and General
Rodrı́guez climbed to the top of the power pyramid to become president.
During the revolution, the war devastated economy and the productive capacity
of the country. The economic situation was further aggravated as the Great De-
pression of 1929 hit Mexico. The following decades were marked by slow economic
reactivation. In this context, the post-revolutionary political arrangements allowed

207
local caciques to extract rents from all sorts of legal businesses such as oil compa-
nies, landowners, merchants, bankers, mining companies, workers unions, peasant
organizations, newspapers, factories, and more. Illicit trades such as drugs, alcohol,
prostitution, and gambling were simply another form of personal peculation avail-
able to strongmen. As stated by Knight (2012, 122), “drug income was there for
the taking.” This source of wealth was particularly attractive in the northern part
of the country due to the illegal markets created by prohibition laws in the U.S.
Rents extracted from legal or illicit sources were also instrumental for politicians as
they served as “cash cows” that helped finance election campaigns and political ac-
tivities (Lupsha, 1995). Although effective competition from opposition parties was
non-existent, PRI politicians still used these funds to mobilize mass support to boost
their strength within the party and to buy the necessary sympathy from members of
the higher ranks. This mass support was also crucial for the survival of the party,
as it deterred prominent leaders from splintering away from the party (Magaloni,
2006). By the time Prohibition ended in December 1933, contraband networks were
well established along the border. However, the illicit contraband business did not
decline after the end of Prohibition because the U.S. passed the Marijuana Tax Act
in 1937. U.S. efforts to control the transportation and selling of the marijuana gave
another boost to drug smuggling from Mexico.
Although violence was ubiquitous during the revolutionary war, drug trafficking
was not a violent activity. Astorga (1999) argues that “in small towns, it was more
difficult, although not impossible, to resort to violence because almost all the inhabi-
tants were related. There was room for everybody in the drug business, so it was not
necessary to fight to death to get a share of the market.” In the post-revolutionary
era, preventing violence was a priority. Caciques strongly enforced the pax PRIı́sta
upon local criminals and drug traffickers. As noted by Knight (2012), the use of law
enforcement was limited and selective, yet exemplary, employed merely to remind

208
drug dealers who was in charge. Sporadic episodes of occasional violence occurred
as new criminals challenged old ones. Those who did not conform to the informal –
yet clear – rules were severely punished. The sanction also served to make clear to
other criminals that they were all “replaceable cogs.” Arrests of traffickers and drug
seizures were also used to ameliorate diplomatic pressure from the U.S. on Mexico’s
anti-drug enforcement. Caciques thus imposed discipline on criminal networks in
the same way political discipline was imposed on them through the party structure.
Every member of the “revolutionary family” had its place in the new political order.
A dark episode in the Quevedo family illustrates the type of selective discipline
imposed by the political elite on non-compliers. After having been governor of Chi-
huahua, General Rodrigo Quevedo operated the political network in the state. As
part of his network of nepotism, he installed Carlos Villareal, who had joined the
Quevedo family by marrying the daughter of José Quevedo, as mayor of Juárez.
Villareal took care of the family businesses of drug trafficking, contraband through
customs, and car theft in the U.S. However, Villareal eventually tried to emulate the
example of General Quevedo by designating his successor in City Hall. The nominee
was Vı́ctor Manuel Ortiz, who was mayor of Juárez from 1950 to 1952, but was not
related to the Quevedo family. Together, Villareal and Ortiz started to take over con-
trol of the drug business. They felt safe because of their close friendship with Gustavo
Dı́az Ordáz, an emerging but strong politician from the state of Puebla. This protec-
tion gave Villareal the aspiration to become governor of Chihuahua. After spending
more than twenty years as chief of several military regions, General Quevedo joined
the Congress in two consecutive periods between 1958 and 1964. During his time in
the lower house of Congress (the Chamber of Deputies), General Quevedo realized
that Villareal and Ortiz were trying to take over the political structure as well as the
customs and illegal business in Chihuahua. In February 1963, a gunman killed both
Carlos Villareal and Vı́ctor Ortiz in a bar in Ciudad Juárez. A year later, in 1964,

209
Gustavo Dı́az Ordáz, the politician who had mentored Villareal, became president of
Mexico. However, he did not take reprisals against General Quevedo, who became
Senator that year. After all, General Quevedo was an old and well respected member
of the “revolutionary family” who imposed discipline on those who did not comply
with the political customs. Four years later, General Quevedo died of natural causes
in El Paso, TX.
After the volatile conditions of the Mexican Revolution, a reduced number of
strongmen created order by forging an alliance to regulate the peaceful use of and
access to power through the mediation of a centralized party, the PRI. By appealing
to their own interests, this agreement became the most effective way to encourage
self-restraint on the part of those capable of organizing violence. The party then
became the expression of the new order that emerged out of the Revolution.

5.4
The
Consolidation
of
Order
Under
Authoritarianism

After the end of the Mexican Revolution, the PRI emerged as the main mechanism
to manage disputes among the political elite in a peaceful manner. The party relied
on local leadership and initially gave them a wide margin to maneuver for their po-
litical affairs as long as they remained loyal to the center. The PRI centralized power
around the figure of the president, who used a set of informal rules to moderate dis-
agreements and impose political discipline. The encompassing character of the party
included a broad range of political groups, all integrated within the large “revolu-
tionary family.” In this way, the party served as the main instrument for maintaining
order. The autonomy of local caciques allowed them to engage in drug smuggling to
the U.S. as a source of personal enrichment. Strongmen could run drug businesses
and this did not represent a national security concern. However, the beginning of the

210
Cold War and the polarization of world politics created unprecedented new threats to
the survival of the political regime. The hegemonic control of the PRI faced numer-
ous challenges from subversive organizations motivated by left-wing ideologies. The
regime immediately reacted by developing a highly centralized and brutally repressive
security apparatus. In order to secure the survival of the regime, the PRI engaged
in harsh repression against students, workers and political opponents. Some of these
dissidents radicalized and engaged in urban or rural guerrilla movements, only to see
the escalation of violent repression deployed by the state against them. In contrast
to the severe treatment that government authorities gave to political dissidents, drug
trafficking organizations enjoyed the complicity of politicians and security agents.
The difference in treatment lay in the distinct political nature of subversive groups
and drug trafficking organizations. Dissidents represented a direct political threat to
the survival of the regime; in consequence, they were severely repressed by govern-
ment authorities. In contrast, drug traffickers not only represented a source of money
for corrupt politicians but, more importantly, they did not represent a political threat
to the regime. In consequence, they were not repressed by the state. Moreover, the
involvement of security agents in the criminal underworld served authorities by en-
abling them to monitor the behavior of clandestine organizations that could attempt
to acquire weapons to challenge the state or to engage in drug smuggling to fund
their struggle.
This section describes how the international and domestic dynamics of the Cold
War favored the development of a strong centralized security apparatus to protect the
integrity of the state against subversive forces. The historical account also surveys
the escalation of state repression of political dissenters over a more than twenty-year
period. In contrast to the harsh repression suffered by political dissidents, efforts
to counter the narcotics trade were modest. There were a few crackdowns, mostly
motivated by U.S. pressure. However, not even the most significant of these anti-

211
drug operations generated a substantial escalation of criminal violence against the
state nor against other drug organizations. This section also describes how the PRI
political hegemony served as a key instrument for maintaining peace in drug markets
without the need of using state coercion.

5.4.1
Cold
War
Politics
and
its
Repressive
Instrumentation

In 1944, during World War II, the Mexican Coast Guard stopped a U.S. private
yacht while patrolling national waters in the Gulf of Mexico. The yacht was discov-
ered to be transporting a shipment of opium and morphine. The drugs and the boat
were taken to the military port of the state of Veracruz. A few hours later, the gov-
ernor Veracruz, Miguel Alemán, showed up at the Coast Guard office and demanded
that the boat and its crew be released. Officers refused the governor’s request but
then they received orders from Mexico City to release the boat (Astorga, 2003, 58).
During his administration as governor, Alemán appointed Carlos Serrano as his chief
of police. Serrano had a long criminal career that began when he was seventeen,
smuggling rum from Cuba. Often relying on violence against minor competitors,
Serrano used his position as chief of police to run an opium and morphine network
to New York. Two years after the yacht incident, in 1946, Miguel Alemán became
president of Mexico. Serrano was appointed leader of the Senate and Alemán kept
him as one of his closest advisers. The president gave him the title of Colonel despite
he never served in the military, and his violent reputation gave him the nickname of
“The President’s gunman” (“El Pistolero del Presidente”) (Cedillo, 2011).
By the time Miguel Alemán became president, the U.S. government already had
reports of his participation in drug trafficking activities. However, with the end of
World War II and the beginning of the Cold War, U.S. authorities were far more
concerned about the communist threat than about drug trafficking. The strategic
position of Mexico as the neighbour of the U.S. was of paramount importance for

212
containing a communist threat in Latin America. Mexico was quickly drawn into the
global realignment of Cold War politics.
In step with aggressive U.S. foreign policy aiming to counter the expansion of
communism in the region, President Alemán created the Federal Office of Security
(Dirección Federal de Seguridad, DFS ) in 1947, the same year that the U.S. cre-
ated the Central Intelligence Agency (CIA). Since its beginning, the DFS received
support, training and intelligence collaboration from the U.S. to assist Mexican au-
thorities in their efforts to suppress subversive organizations. The DFS reported
directly to the president and its main mission was to serve as a political police force
to gather information and neutralize actual and potential subversive threats to the
Mexican government. The first formal director of the DFS was Colonel Marcelino In-
urreta de la Fuente. However, it is broadly recognized among historians that Colonel
Carlos Serrano was the real man in charge of the DFS. The direct antecedent of the
Directorate goes back to the Office of Political Information (Oficina de Información
Polı́tica, OIP) created by president Cárdenas in 1939 at the start of World War II,
which was used to conduct investigations into political opponents. The OIP was
later transformed into the Department of Political and Social Investigations, (De-
partamento de Investigación Polı́tica y Social, DIPS). The same year as the DFS
was founded, the responsibility for drug issues was transferred from the Ministry of
Health to the Office of the General Attorney (Procuradurı́a General de la República,
PGR), which had the Federal Judicial Police (Policı́a Judicial Federal, PJF) as its
enforcement agency.
Due to the background of the people in charge of the DFS, the Mexican secret
police were directly involved in illicit activities since the creation of the agency. Ac-
cording to Scott (2010), a confidential wire from the U.S. Embassy in Mexico sent on
September 4, 1947 identified Senator Carlos Serrano, DFS head Marcelino Inurreta,
and his subordinate Lieutenant Colonel Manuel Magoral as being directly involved in

213
drug trafficking. Serrano also controlled a high-level prostitution network in Mexico
City where DFS agents took pictures their clients, which gave Serrano firm control
over politicians, bankers, and businessmen (Astorga, 2003; Cedillo, 2011). Later, in
1951, a CIA agent based in Mexico filed a secret report indicating that DFS personnel
abused their power to control illegal activities such as narcotics smuggling, and the
directorate of the DFS was unscrupulously involved in these activities (Scott, 2010,
51).
DFS agents were not only involved in illegal activities themselves. Most impor-
tantly, the concentration of enforcement in close association with the highest levels of
political power enabled them to forge connections with criminals all over the country.
The DFS established a pyramidal mode of control with a high degree of concentra-
tion at the top but preserving the structure of the local criminal base (Kenny and
Serrano, 2012a). In doing so, the DFS replicated the hierarchical and centralized
power structure of the PRI in the criminal world. Just as the political elite concen-
trated power and gave autonomy to local caciques in exchange for their loyalty to
the “revolutionary family” and compliance with its informal rules, the DFS managed
an extended network of local drug dealers and concentrated its management at the
top of the pyramid. In doing so, the DFS set the tone for the relationships between
political power, law enforcement and criminal activities from the 1950s to the 1980s.
The close connections between DFS personnel and the political elite gave them
absolute impunity in the event of any judicial accusation. In addition, the DFS
recruited the most violent police officers they could find. As stated by Kenny and
Serrano (2012a, 34), “their brutality was a service ready to be hired out.” The
combination of extreme coercive power and political connections gave DFS agents the
upper hand over common criminals and ensured that local government authorities
would defer to their authority. In 1947, Lieutenant Colonel Maurice Holden, the
Assistant Military Atraché of the U.S. in Mexico, indicated that DFS agents were a

214
law onto themselves with the power of life and death, “they were a GESTAPO by
other name” (Astorga, 2003, 286).
The is no doubt that the direct engagement of the DFS in drug trafficking activ-
ities was a key source of personal enrichment for corrupt politicians and enforcers.
However, historians and journalists studying the early days of drug trafficking in
Mexico (Astorga, 2003, 2005; Cruz, 2008; Knight, 2012) have not sufficiently empha-
sized the importance of the political context to explain the involvement of the DFS
in criminal activities. The strong coercive and political power of the DFS was devel-
oped in the context of the Cold War. The central concern for government authorities
at the time was identification and neutralization of subversive political movements
motivated by leftist ideologies, which were well underway by the 1950s. The develop-
ment of the DFS as a political police force gave the Mexican government the ability
to infiltrate and operate in the social and political underworld, an area certainly
populated by criminals, but also likely to cultivate subversive organizations deemed
as a direct political threat to the state.
The control of the DFS over criminal networks in the entire country not only gave
enforcement agents the opportunity to reap economic benefits from illicit activities
but, more importantly, it gave the political system the ability to monitor subversive
political activities and, if necessary, strike an effective repressive blow. Clandes-
tine political organizations trying to organize an armed struggle against the state
were likely to approach gun traffickers in the criminal world if they needed weapons.
Moreover, drug trafficking could be perceived as a tempting source of funding to
sustain a political struggle, as demonstrated by the Fuerzas Armadas Revolucionar-
ias de Colombia (FARC) in Colombia and other insurgencies worldwide (Bruce and
Hayes, 2010; Fearon, 2004; Felbab-Brown, 2010; Youngers and Rosin, 2005). Due to
the system of incentives implemented by the DFS, criminals had motives to inform
their political bosses about any subversive political organizations they might detect.

215
The DFS could then coordinate with the Federal Judicial Police and the Army to
crush them. In addition, the meeting points between insurgents and criminals in the
clandestine world meant that the state could attempt to infiltrate subversive orga-
nizations using criminals as state-sponsored covert agents. The involvement of the
DFS in criminal activities was not only supported by corrupt politicians for economic
reasons but, more importantly, it was justified and encouraged for political reasons.
As discussed in Chapter 1, the primary economic motivation of criminal organi-
zations does not represent a direct threat to the political survival of the state. For
this reason, the existence of criminal networks was tolerated in the context of Cold
War politics. In contrast, the Mexican government did not spare any effort to repress
any potential political threat that could represent a fundamental challenge to the
existing power structure. As mentioned earlier, the historical processes leading to
the development of PRI’s political hegemony gave no margin of tolerance to any op-
position outside the party. For this reason, leftist movements inspired by communist
or socialist ideologies represented a profound threat to the survival and stability of
the political regime and were severely repressed.
The authoritarian nature of the Mexican state did not only have domestic origins.
The coercive apparatus of the Mexican state developed with the direct support of the
U.S. government. The foreign policy of the U.S. towards Latin America during the
Cold War was marked by proactive and sustained efforts to prevent the expansion
of the communist threat in the region. The U.S. provided both covert and overt
support for authoritarian regimes throughout Latin America, and Mexico was no
exception. U.S. security agencies provided training to Mexican military and law
enforcement agents and facilitated information-sharing regarding political hazards.
The containment of the communist threat in Mexico was of vital importance to the
U.S. and often required that U.S. security agencies turn a blind eye to the involvement
of Mexican political authorities and enforcement agents in drug trafficking.

216
5.4.2
Political
Threats
and
Authoritarian
Reaction

The creation of the DFS marked the beginning of a period of high levels of political
repression and social unrest. During three decades, from the 1950s to the 1970s, the
hegemonic party system systematically engaged in political repression and conducted
anti-drug operations. The brutality of repression against political movements and
dissident organizations seemed to have no limits. This period of Mexican history is
known as the Dirty War (La Guerra Sucia). However, the frequency and intensity
of repression and anti-drug efforts were not on the same scale: the implementation
of anti-drug actions were primarily the consequence of U.S. pressure rather than a
central concern to government authorities.
One of the first important acts of the DFS took place on May 1, 1952, Labor Day
in Mexico, when a group of police agents coordinated by the DFS violently dispersed a
demonstration organized by the Mexican Communist Party (Partido Comunista Mex-
icano, PCM) and charged its leaders with the crime of “social dissolution” (disturbing
social order). After the incident, opposition labor organizations were not allowed to
parade on Labor Day. Encouraged by the victory of the Cuban Revolution in 1992,
the remaining leadership of the PCM organized a major strike of railroad workers
from 1958 to 1959. The government reacted violently by cracking down on the strike,
raiding the headquarters of the Railroad Workers Union (Sindicato de Trabajadores
Ferrocarrileros de la República Mexicana, STFRM) and arresting thousands of work-
ers. About 800 railroad workers spent several years in the infamous Lecumberi prison
used for detaining political prisoners. In the early 1960s, the teachers’ union went on
strike demanding higher wages but was repressed by the Army. Two years later, the
Army repressed members of another teachers’ organization, the Asociación Cı́vica
Guerrerense (ACG) in the state of Guerrero. After the crackdown, Genaro Vázquez
Rojas, a rural teacher and member of the ACG, went underground to start a ru-

217
ral guerrilla movement in Guerrero, the Asociación Cı́vica Nacional Revolucionaria,
(ACNR).
In January, 1965, Fernando Gutiérrez Barrios became the director of the DFS.
One of his first actions was to mount a raid on the headquarters of the Mexican
Communist Party. The same year, the police crushed a demonstration organized by
the Mexican Doctors’ Alliance (Alianza de Médicos Mexicanos, AMM) and took over
several public hospitals and clinics. Hundreds of doctors were fired and replaced with
military doctors.
The revolutionary movement in Latin America gained renewed enthusiasm af-
ter the meeting of the Latin American Solidarity Organization (Organización Lati-
noamericana de Solidaridad, OLAS) in Havana, Cuba. The OLAS had the central
objective of exporting the armed revolution throughout Latin America. In 1966 and
1967, rural and urban guerrilla attacks intensified in Mexico. Guerrilla activities
were mostly concentrated in the states of Guerrero and Chihuahua, although there
were small armed cells in other states. In 1967, the government violently repressed
another group of teachers in Guerrero. Lucio Cabañas, one of the survivors, radical-
ized and founded a guerrilla group named the Peasants’ Execution Brigade (Brigada
Campesina de Ajusticiamiento, BCA) and later on he founded a larger guerrilla or-
ganization known as the Party of the Poor (Partido de los Pobres, PdlP).
Several other non-violent social movements intensified their activities in Mexico
City, Sonora, Puebla, Tabasco, Guanajuato, Querétaro, Tamaulipas, Quintana Roo
and Michoacán. Some of the most prominent groups were Grupo Popular Guer-
rillero (GPG), Movimiento Revolucionario del Pueblo (MRP), Grupo Popular Guer-
rillero “Arturo Gámiz”, Movimiento 23 de Septiembre (M23S), Movimiento Lati-
noamericano de Liberación (MLL), Ejército Revolucionario del Sur (ERS), and the
Movimiento Revolucionario Marxista-Leninista Mexicano (MRMLM), among oth-

218
ers.2 Most of their members were gunned down by government security forces. Sur-
vivors were arrested, often tortured and some of them executed. Those who managed
to escape radicalized further and continued the armed struggle creating new guerrilla
cells or joining other organizations.
In 1968, the student movement in Mexico City took on unprecedented propor-
tions. Led by students from the National Autonomous University of Mexico (Uni-
versidad Nacoinal Autónoma de México, UNAM) and the National Polytechnical
Institute (Instituto Politécnico Nacional, IPN), the student movement gained the
support of several other organizations all over the country and coordinated massive
protests and strikes. In October 2, 1968, some ten thousand students gathered in
the Tlatelolco Square in Mexico City. The Army dispersed the mobilization by indis-
criminately firing over the crowd. The number of casualties of this event, known as
the Tlatelolco massacre, remains undetermined (Poniatowska, 1971, 1980; Revueltas,
1978). In 1971, there was another, probably smaller, student massacre known as “El
Halconazo.”
After the Tlatelolco massacre, several subversive organizations radicalized and
engaged in guerrilla fighting. As stated by Cedillo (2009), these radical organizations
largely relied on kidnapping businessmen, high-level government officials, politicians,
and even foreign diplomats3 to finance their armed struggle. The primary use of
kidnapping as a funding strategy is indicative that subversive organizations were
not relying on drug trafficking to finance their struggle. Part of the reason for not

2
One of the most exhaustive chronologies of the emergence, actions and subsequent repression
of dozens of subversive organizations between 1943 and 1981 is provided by Castellanos (2007). In
addition, Cedillo (2009) offers another succinct, yet detailed list of organizations and their activities.
3
In October 1973, an armed brigade of the Liga Comunista 23 de Septiembre (LC23S) kidnapped
the British Consul in Guadalajara, Anthony Duncan Williams and a local businessman named
Fernando Aranguren in Guadalajara, Jalisco. The government refused to pay ransom and the
LC23S executed Aranguren but released the consul. A few weeks later, the police arrested and
tortured some members of the brigade. By December of the same year, the police had dismantled
that brigade of the LC23S.

219
engaging in drug smuggling may be that this activity was contrary to the ideological
principles of these highly indoctrinated groups, but also because dealing drugs meant
venturing into a sector largely controlled by the DFS. In any case, the government
reacted to increased guerrilla activity by intensifying repression, especially in the
rural areas. In 1972, the Army killed Genaro Vázquez, the leader of the guerrilla
group Asociación Cı́vica Nacional Revolucionaria. The Mexican government also
conducted a massive operation, deploying some 16,000 soldiers in the mountains in
Guerrero to hunt down Lucio Cabañas, leader of the guerrilla organization Partido de
los Pobres. In December, 1974, the Army found the hidden camp of Lucio Cabañas
and killed him thanks to information provided by a local drug producer4 and an
agrarian leader (Guerrero Robledo, 2009; Redacción de El Universal, 2009; Rivera
and Valdez, 2007).
Other non-radical opposition groups were also restricted or coopted by the state.
Opposition parties were weak due to the limited competition opportunities allowed
by the hegemonic party. The right-wing National Action Party (Partido Acción Na-
cional, PAN) founded in 1939, was maintained at the margins of political positions
and elections were largely rigged by the Ministry of the Interior, the institution in
charge of organizing the elections. Another political party, the Popular Socialist
Party (Partido Popular Socialista, PPS) was founded in 1948 by the leader of the
national labor federation most closely associated with the ruling party, the Confeder-
ation of Mexican Workers (Confederación de Trabajadores de México, CTM). In other
cases, the government directly promoted the creation of “facade” parties such as the
Authentic Party of the Mexican Revolution (Partido Auténtico de la Revolución Mex-
icana, PARM) just to keep up the appearance of competitive elections. The electoral

4
The name of the presumed drug producer was Isabel Ramos Ruiz. Although some secondary
sources identify Ramos as a drug producer or trafficker, his participation in criminal activities has
not been historically confirmed. I thank the Mexican historian Ángeles Maldonado for pointing this
out in a personal conversation.

220
system based on majority rule made it extremely difficult for opposition parties to
secure seats in Congress. As indicated by (Medina Peña, 1994, 166), between 1955
and 1964 opposition parties had in total only 20 seats in the Chamber of Deputies
out of 483 seats available in this period, thus holding only 4.14 percent of the seats.
Later, in 1963, PRI granted a small electoral concession by allowing a party–deputy
system (“Diputados de Partido”), a modest formula that slightly increased the space
for opposition parties that received more than 2.5 percent of the vote. During the
entire period between 1964 and 1970, the opposition held only 100 seats while PRI
increased the number of seats in the Chamber of Deputies and controlled the other
557 positions, thus leaving only 15.2 percent to the opposition.

5.4.3
Counter
Narcotic
Efforts
During
the
Dirty
War

In contrast to the brutality of political repression launched by the Mexican state

against politically motivated subversive organizations, the efforts of government au-
thorities against drug trafficking were modest. A handful of significant anti-drug
operations were caused by U.S. diplomatic pressure. But drug traffickers were by no
means hunted down and punished by government authorities as the political opposi-
tion were.
Between 1945 and the end of the 1970s, the business of opium and marijuana
trafficking in Mexico boomed due somewhat to soldiers returning to the U.S. after
the end of World War II in 1945 but mostly during and after the Vietnam War. The
increasing demand for drugs in the U.S. was met with increased opium and marijuana
production in Mexico. The demand for drugs from Mexico increased further after the
1971 dismantling of the “French Connection,” a transnational heroin trade network
from Turkey to France and then to the U.S. However, the Mexican authorities did
not undertake substantial efforts to eradicate crops of illicit drugs nor to reduce their
traffic.

221
 During the 1940s, Harry J. Anslinger, the Commissioner of the U.S. Federal Bu-
reau of Narcotics saw with frustration that drug enforcement agencies in Mexico were
slacking. He managed to convince Mexican authorities to conduct some drug erad-
ication campaigns and drug seizures in Sinaloa, Chihuahua, Sonora and Durango
during the 1940s, 1947 and 1948 being the years with most activity. However, to
Anslinger’s dismay, the 1951 poppy and marijuana harvests were the largest recorded
in Mexico over the previous decades. According to Astorga (2003), that year drug
producers managed to cultivate individual fields of marijuana up to 20 acres (about
8.09 hectares) and the crop could be bought by the ton. The production of raw
and processed opium was also very high. During the 1950s there were some isolated
crackdowns on drug traffickers but no sustained or large scale enforcement efforts like
those conducted against subversive organizations.
In 1969, two months after President Nixon took office, he established the Special
Presidential Task Force Relating to Narcotics, Marijuana and Dangerous Drugs. The
task force produced a report that singled out Mexico as the primary supplier of
marijuana and a source for a large amount of other dangerous drugs, including heroin.
This task force was one of the first steps of Nixon to fulfill his campaign promise to
the “great silent majority” that he would take tough measures against drugs and
other urgent concerns of U.S. society.
On September 21, 1969, President Nixon launched “Operation Intercept” which
consisted of the deployment of about two thousand customs and border-patrol agents
along the Mexican border to conduct detailed inspections of all trucks, cars, planes
and people crossing the border by foot. Vehicles trying to cross the border were de-
layed up to six hours and pedestrians were stripped and body-searched. The amount
of narcotics seized during the operation was minimum. As stated by Doyle (2003),
the main objective of Operation Intercept was not to stop the flow of drugs into the
U.S. but was to cause an unprecedented economic and social disruption on the border

222
that would force the Mexican government to devote more stringent efforts to drug
eradication.
Operation Intercept was particularly shocking to Mexican authorities because it
was planned in secret. Even when President Nixon and President Dı́az Ordáz met
on September 8, 1969, a few days before the operation, at the Friendship Dam on
the border between Texas and Coahuila, Nixon did not tell Dı́az anything about the
plans for the operation. Operation Intercept remained active only for a few weeks. In
October, Intercept was replaced by a new agreement between Mexico and the United
States called “Operation Cooperation,” in which both countries agreed to cooperate
to reduce the production of narcotics within Mexico and its movement across the
border. The U.S. gave Mexico some air crafts, helicopters, remote sensing equipment
as well as financial and intelligence assistance for eradication efforts (?). Operation
Cooperation led to to a few arrests and the destruction of some illicit plantations.
At the time when guerrilla fighting reached its peak during the mid 1970s, the
Mexican government launched “Operation Condor” (Operación Cóndor ). According
to Craig (1980, 347) “the Mexican government had decided to remove the kid gloves
with drug traffickers.” In 1975 and 1976, the Mexican government conducted an un-
precedented military operation consisting in the deployment of ten thousand soldiers
under the orders of Brigade General José Hernández Toledo to eradicate illicit crops
in Sinaloa, Durango and Chihuahua. Toledo was a experienced general who led the
Parachute Battalion that had perpetrated the Tlatelolco massacre in 1968 and partic-
ipated in other operations against students (Astorga, 2003; Department of Defense,
1968). Operation Condor also had the participation of the Office of the Attorney
General, represented by Carlos Aguilar Garza, and the Federal Judicial Police (PJF)
as its enforcement arm. Aguilar Garza eventually became a drug trafficker and ended
up being assassinated in 1993 (Astorga, 1999).

223
 Although the federal authorities initially declared that no chemical defoliants were
going to be used for eradication, Pedro Ojeda Paullada, Mexico’s Attorney General
at the time, had an agreement with Sheldon Vance, Kissinger’s special advisor on
drug issues, on the use of aerial spraying of defoliants. Based on the U.S. experience
in Vietnam, Operation Condor used Paraquat, also known as Agent Orange, a quick-
acting non-selective defoliant already declared by the U.S. Surgeon General in 1971
to be harmful to humans. The large number of troops involved in Operation Condor
was necessary because the program combined aerial and manual eradication. After
plantations were sprayed from the air with defoliants, soldiers went into the fields to
cut down opium and marijuana plants with machetes. The labor-intensive activity of
manual eradication required significant personnel. Although DEA agents occasionally
discovered that the Army sometimes used fertilizer instead of defoliants to spray
over drug plantations (Toro, 1999), authorities declared Operation Condor a success.
Between 1963 and 1970, government authorities had only eradicated 4,370 hectares
of opium and 2,400 hectares of marijuana. In contrast, between 1970 and 1976,
authorities eradicated 25,000 hectares of opium plantations and 13,300 of marijuana
(Astorga, 2005).
Along with the methods used by security agencies during the Dirty War, the
participation of military personnel was accompanied by human rights abuses. The
Lawyers’ Association of Culiacán, Sinaloa, denounced the usual methods of torture
employed by PJF, DFS and the Army for conducting interrogation of presumed drug
cultivators or traffickers which included beatings, electric shocks, burning, rape, and
waterboarding (Astorga, 2005). In addition, the deployment of the Army and aerial
fumigation generated a massive exodus of peasants to the cities. Hundreds of people
were arrested, but not a single big boss was captured. As stated by Astorga (1999),
the most important leaders of drug trafficking organizations moved to safer areas
and continued their illegal activities. For example, Pablo Acosta Villareal, one of the

224
most prominent drug lords of the 1970s, controlled a substantial share of heroin and
marijuana trafficking from Ojinaga, Chihuahua, and would later become a key figure
in cocaine trafficking during the 1980s as the leader of the Juarez Cartel (Cártel de
Juárez ). Another prominent drug lord, Miguel Ángel Félix Gallardo, known as “The
Godfather” (El Padrino), moved from Sinaloa to Guadalajara, Jalisco, and consol-
idated the Pacific Cartel (Cártel del Pacı́fico). Félix Gallardo was a former police
officer in the Sinaloa state judicial police and served as bodyguard for the son of
governor Leopoldo Sánchez Celis, with whom he developed a close friendship. The
governor had a well known reputation of having close links with drug traffickers and
gunmen (Aguilar Camı́n, 2009). By the behavior970s, another drug trafficker, Juan
Garcı́a Ábrego, began his career under the command of his uncle Juan Nepomuceno
Guerra, leader of the Gulf Cartel (Cártel del Golfo), which operated in the north-
eastern state of Tamaulipas. None of these actors were directly affected by Operation
Condor nor by subsequent counter-narcotic efforts because they had the protection
of high-level politicians and enforcers.

5.4.4
Maintaining
Peace
in
Drug
Markets

The previous sections describe the differences between the state’s behavior in its
relationships with the opposition and with criminals during the Dirty War. The
interaction between the state and subversive organizations was marked by highly re-
pressive efforts by government authorities and direct challenges from radical political
organizations who conducted massive protests and guerrilla actions against the state.
In contrast, the interaction between the state and criminals was not characterized
by sustained law enforcement efforts from the government, nor by sustained violence
perpetrated by criminals against the state or against rival criminal groups. As illus-
trated by Operation Condor, there were instances in which government authorities
flexed their muscle against drug traffickers. However, these law enforcement efforts

225
did not trigger sustained waves of violence against the state nor against rival crim-
inal groups. There were some isolated events where criminals used violence against
the state. Nevertheless, state–criminal relations were mostly non-violent. Scattered
violent events between the state and criminals did not constitute sustained efforts
of violence like those characterizing state–opposition interactions during the Dirty
War and by no means resemble the wave of large-scale organized criminal violence
experienced in the second half of the 2000s.
The lack of violence that pervailed for decades in Mexican drug markets is puz-
zling, especially because of the lack of sustained counter-narcotic campaigns con-
ducted by the state that could have been used to suppress drug trafficking organiza-
tions. However, the puzzle is not only empirical, it is also theoretical. As accurately
stated by Reuter (1989), the use of violence is a distinctive feature of illicit mar-
kets. The violent nature of these markets stems from the lack of access to the usual
mechanisms of dispute resolution such as legal agreements and judicial processes. To
enforce property rights, criminals cannot call on the police or go to court. Those en-
gaged in criminal activities can only use violence or intimidation to resolve disputes
or enforce agreements. In contrast to the theoretical expectations, the use of violence
in illicit markets did not come to pass in the Mexican case. The key reason for the
lack of drug violence is the concentration of political power that existed under PRI
hegemony, which served as the key mechanism of dispute resolution and regulation.
As described earlier, the process of state formation in Mexico after the revolution
was based on an agreement to peacefully channel the capacity of different actors
for conducting organized violence. This process led to the concentration of political
power in the president as the grand administrator of power and arbitrator of disputes.
The entire political system grew as a symbiotic network of political relations and
institutional arrangements favoring the concentration of power at the federal level
and granting a relative degree of autonomy to sub-national power structures on their

226
local issues. This system worked on the basis of political incentives instilling discipline
on the members of the “great PRIı́sta family.”
Arguments about state-sponsored protection rackets (Snyder and Duran-Martinez,
2009) and state-led regulation of criminal activities (Kenny and Serrano, 2012a) indi-
cate that Mexican government authorities generally managed to maintain low levels
of violence in illicit markets. These arguments primarily rest on the state’s coercive
capability as a crucial element of their explanations. According to these perspectives,
the certainty and severity of enforcement allows the state to impose order and peace
in illicit markets. State officials refrain from enforcing the law against criminals or,
alternatively, enforce the law in a selective manner in exchange for bribes. In con-
trast, “if the state lacks the power to enforce the law, illicit actors may prefer to
bear the costs of haphazard and weak enforcement to paying off state officials to
refrain from enforcement” (Snyder and Duran-Martinez, 2009, 255). As discussed in
the previous sections, this argument does not meet the historical characteristics of
counter-narcotic efforts in Mexico. For several decades law enforcement against crim-
inal organizations was at best modest. Levels of enforcement could even be classified
as minimal when compared to the severity of repression deployed by government
authorities upon political opponents.
The role and level of the use of force is certainly important for an understanding of
the maintenance of order in illicit markets in Mexico. However, emphasizing the cen-
trality of the use of coercion against criminal organizations is an overstatement that
obscures the complexities of the Mexican political system. State–criminal relations
during this period were not characterized by the submission of criminal organizations
to the mighty coercive power of the state. Instead, low levels of violence were mostly
maintained by political means due to the centralized hierarchical control held by the
PRI across all levels of government. This argument is congruent with the insightful
analysis provided by Rios (2012a).

227
 Rather than submission under force, the nature of state–criminal relations dur-
ing the period of PRI hegemony is better characterized by a symbiotic coexistence
between criminals and politicians. State actors and criminals were not separate or
distinct actors; policemen often became drug traffickers and kept close connections
with the political elite. Peace and order were maintained through political means in
illicit sectors just as they were in other power spheres contained within the encom-
passing PRIı́sta family. Opposition groups outside the political system faced ruthless
repression from the government authorities, but there was no need to use such tactics
against drug traffickers, since they were closely interwoven into the political system.
After all, the party first emerged as an agreement favoring the peaceful resolution of
differences among those with the power and ability to conduct organized violence.
To pursue that goal, the party served the concentrator of an encompassing pyramid
of power and political relations with the president at the top. The direct connec-
tion between the DFS and the president, later through the Ministry of the Interior,
enabled the executive to maintain tight control of criminal networks. The degrees
of access, use and discipline in these networks were managed primarily by political
means rather than through the use of force.
There are several formal and informal characteristics of the political system that
enabled peace and order in state–criminal relationships to be managed politically
rather than by force. Since the positions of head of the executive and party leader
were combined in the president, this gave him the power to designate the next PRI
candidates for state governors, senators, representatives and mayors. Most impor-
tantly, the president had the power to designate his own successor. This procedure,
known as “el dedazo” (the pointing of the big finger), was the key mechanism for man-
aging a complex system of political rewards and sanctions (Langston, 2001, 2006).
The president had the power to use this appointment prerogative at every level of
government and in any area of the party and the public sector. With the approval

228
of the president, those below him were also allowed to use the same mechanism to
appoint their lower ranks. The political system thus constituted a dense, hierarchical
network of political relationships for managing the access and use of power in every
political and bureaucratic structure.
The density and inter-connectivity of the political system during the era of PRI
dominance provided timely, accurate information on the behavior of every politi-
cal actor within the system. Individuals who did not carry out the commands of
those above them in the hierarchy faced the possibility of suffering political sanc-
tions. Punishment included removing the possibility of promotion to higher political
or government spheres, being demoted from their current position, or even being ex-
pelled from the entire political or bureaucratic structure. Of course there was always
the possibility of being arrested or killed as the price for serious mistakes or misbe-
haviour. But the use of force within the party was a rare anomaly. This system of
incentives favoring discipline and compliance with informal rules is well depicted by
the old PRI adage, “él que se mueve no sale en la foto” (the one who moves will not
appear in the picture).
This system of incentives also extended to state–criminal relations. Those oper-
ating illicit activities such as drug production or trafficking would need approval and
support from enforcement agencies and political actors. Obtaining approval – in the
form of non-enforcement or proactive support – depended not only on the amount of
bribe money that criminals could offer to their political bosses but, more importantly,
on the non-aggressive behavior of criminals while conducting their illegal activities. A
serious outbreak of violence among criminals or against the state could have jeopar-
dized the political survival of those providing protection and support. Criminals thus
had incentive to maintain low levels of violence in order not to lose their networks of
protection. Those providing political support similarly had incentives to ensure low

229
levels of violence so they could continue extracting economic benefits from criminal
groups while advancing their political careers.
As former governor of the state of Zacatecas Ricardo Monreal (2008) acknowl-
edged, the structure of political incentives required criminals to adhere to a set of
basic rules: (1) there should be no bodies left in the streets; (2) criminals were not
allowed to sell drugs in the schools; (3) there should be no media scandals; (4) traf-
fickers should allow periodic seizure of drugs and arrests of lower level traffickers;
(5) drug traffickers must generate economic revenues for their communities; (6) there
should be no proliferation of gangs; (7) traffickers should not make direct deals with
formal branches of government (especially not with the police or the judicial bureau-
cracy); (8) mistakes are to be punished with imprisonment by the authorities, not
with execution by rivals; (9) criminal groups must respect territorial boundaries; and
(10) profits from illicit markets should be “reinvested” in Mexico. The operation
of this informal code of conduct known as “The Decalogue” is recognized by other
prominent analysts such as Guerrero (2009b).
The lack of effective political competition was crucial in the functioning of state–
criminal relations. For decades, the PRI managed without interruption to hold the
executive, impose hegemonic control over the Senate and the Chamber of Deputies,
and secure all governorships and most municipalities. Although there were elections
at these different levels of government, the electoral system imposed stringent con-
ditions on opposition parties. In addition, the PRI perpetrated its rule by often
engaging in vote buying, patronage, voter intimidation and election fraud (Greene,
2010; Magaloni, 2006; Schedler, 2002, 2006). Elections were thus not an effective
mechanism for elite circulation, at least not outside the party.
The PRI hegemony had two important consequences for the maintenance order
and peace in illicit markets. First, PRI dominance across the various levels of govern-
ment made corrupt agreements easy to achieve and feasible to implement. To secure

230
protection, criminals only had to cut deals with a small number of political actors,
usually among the political elite. Having support at high levels of power facilitated
compliance by the lower ranks in the government structure with the terms of the
agreement. In this way, the unified hierarchical chain of command characteristic of
the era of PRI dominance served as a top-down mechanism to enforce the agreement.
An order issued by the president, a general or a governor would mean that their sub-
ordinates must immediately comply with the instruction. This power structure also
served as a bottom-up mechanism in which valuable and timely information about
the behavior of criminal groups and their networks of support could be received and
passed upward. If there was a problem to be solved, those at the lower levels would
know whom to reach in the hierarchy. Such information was crucial for assessing
the need to apply political sanctions or law enforcement without disrupting other
agreements secured at higher levels in the political sphere. The DFS was well aware
of this characteristic of the political structure and took advantage of it with infamous
mastery. If a decision was made to apply a sanction, the order would come from the
top and those below would be compelled to carry out the instruction. In the same
way, if the decision was against applying any sanction, those daring to take the issue
into their own hands would face the political consequences of disobeying orders from
above.
Second, over time, PRI dominance also meant extended temporal horizons for
both criminals and politicians. In contrast to the uncertainty about election results
characteristic of well-functioning democratic systems (Przeworski, 1991), the designa-
tion of PRI candidates by dedazo during the period of party hegemony gave absolute
certainty about the winner of each election at every level of government. The faith of
political actors thus depended on their compliance with the party and not on retain-
ing the favor of the electorate. The extended temporal horizons of the political elite
favored the stability of corrupt agreements between criminals and politicians. The

231
expectation of a stable, long-lasting opportunity to extract economic benefits from
illicit markets without being disturbed gave criminals incentive to comply with terms
of the agreement, including refraining from the use of violence against their rivals or
the state. The long term horizon of economic benefits also motivated authorities to
maintain low levels of violence. Politicians aiming to advance their political career
had to make sure not to upset those higher in the political hierarchy. Doing so would
have meant the end of their aspirations. Those who failed to secure the favor of their
superiors could expect to be cut off from access to the spoils of profitable government
positions. It was therefore crucial to avoid violence because it represented a central
concern for the political elite. The old generation of PRI politicians forged in the
revolution was well aware of the deleterious consequences of violence. In addition, the
new generation of PRIı́stas was already alert to threats of violence from politically
motivated subversive organizations. Neither the old or the new leaders of the PRI
family would tolerate violence within their ranks. Maintaining criminal violence at
its minimum was thus a key requirement for the survival of political actors within
the political system.

5.5
The
Erosion
of
the
Preexisting
Order
in
a
Democratic
Context

The previous sections analyzed the emergence of order out of the violent struggles
of the Mexican Revolution and told how the PRI consolidated its political hegemony
through the use of a highly repressive security apparatus encouraged by the interna-
tional and domestic pressures of the Cold War. The dominance of the PRI in the
political arena also extended to the criminal underworld. The state security appa-
ratus assembled a network of criminal organizations and used political pressure to
instill discipline in illicit markets. However, the ability of the PRI to impose order

232
both on the political and the criminal spheres began to erode due to a convergence
of three processes. The first is the increased strength of Mexican drug trafficking
organizations, primarily caused by the surge in cocaine demand in the U.S. during
the 1980s and the subsequent decline of Colombian cartels in the 1990s. The second
process was the dismantlement of the political police after the end of the Cold War,
which undermined the state’s ability to monitor and control criminal organizations.
However, the most important factor is the advent of democracy which unfolded after
a protracted process of thirty years of gradual electoral reforms that strengthened
the opposition and eroded the hegemony of the PRI. Democratization substantially
altered the system of political incentives that had so long permitted and enabled cor-
rupt agreements between government authorities and criminals. The disruption of
the preexisting order was caused by the entrance of the political opposition at various
levels of government as well as the recurrent and effective elite circulation by means
of elections. The increased number of relevant political actors made it more difficult
to achieve and sustain corrupt agreements. The diversity of party labels at the fed-
eral, state and municipal levels also broke the homogeneous chain of command that
made them feasible. In addition, elections reduced the temporal horizons of pacts
with criminals and introduced uncertainty about establishing such agreements with
the next politicians in office. Most importantly, increasing political competition in-
fused by democratization introduced personal incentives for government authorities
to fight criminal organizations in an effort to gain citizen support.

5.5.1
Increasing
Strength
of
Drug
Trafficking
Organizations

Beginning in the 1980s and extending through the 1990s, a series of changes took
place in the international structure of drug markets, as well as a severe economic crisis
and the liberalization of international trade between the U.S. and Mexico. These
factors largely contributed to strengthening Mexican drug trafficking organizations.

233
 The first and most important factor contributing to the consolidation of Mexi-
can drug traffickers was the surge in cocaine demand in the U.S. As described by
Gootenberg (2011), by the mid-1980s there were about twenty-two million cocaine
users in the U.S. consuming various opium derivatives such as cocaine, freebase co-
caine, heroin, crystal meth, and crack cocaine. The vast demand for drugs accom-
panied by a sharp decline in prices and increased availability turned this period into
the “American crack epidemic.” The immediate beneficiaries of the cocaine boom
in the U.S. were Colombian drug organizations based in Medellı́n, Bogotá and Cali.
Colombian organizations initially relied on island-hopping through the Caribbean to
send wholesale aerial and maritime shipments from South America to Miami. But
the U.S. demand was so large that it also benefited Mexican drug organizations. Al-
though it is hard to estimate the share of cocaine being transported from Mexico to
the U.S., some estimates indicate that in 1989 one third of the cocaine for the U.S.
market entered from Mexico, rising to one half by 1992 and reaching 75–85 percent
by the late 1990s (Andreas, 2009; Astorga, 1995; Gootenberg, 2011).
The boom in drug production in Mexico at the time is well exemplified by the con-
centration of agricultural workers participating in drug activities in Sinaloa. Astorga
(2005, 138) describes that during the 1983 crop season, vehicles with loudspeakers
advertised to recruit peasants for “apple picking” in the Sinaloan highlands. The
usual daily wage of agricultural workers was 600 pesos, but peasants were offered
between 4,000 and 5,000 pesos per day for working on drug plantations. The exodus
of peasants to the highlands led to a scarcity of agricultural workers in the valleys
and landowners had to hire peasants from other states to work on regular farms.
In November 1984, Mexican authorities discovered a large facility for marijuana
growing and processing in Chihuahua. The complex known as “Rancho el Búfalo”
(Buffalo Ranch) located in the municipality of Allende, Chihuahua, was the prop-
erty of Rafael Caro Quintero, a co-founder of the Guadalajara Cartel. The ranch

234
covered an area of about 12 square kilometers (7.4 square miles) and hosted more
than 12,000 workers working on a plantation of 1,000 hectares of marijuana. The
group of 450 soldiers who participated in the operation destroyed some 11,000 tons
of marijuana and arrested 500 workers, yet no leader was captured (Astorga, 2005;
Herrera, 2013). According to some accounts, peasants were held at the plantation as
forced labor coerced by armed men, so they were later released.5 The size of El Búfalo
is a clear example of the scale of operations conducted by Mexican drug trafficking
organizations at the time.
One of the most prominent drug traffickers at the time was Amado Carrillo
Fuentes, a former police officer. Carrillo served as the chief of security of Pablo
Acosta Villareal, leader of the Juárez Cartel. After Acosta Villareal was killed in a
helicopter operation conducted by the Federal Police, his second-in-command, former
DFS agent Rafael Aguilar Guajardo, took over the organization. However, Aguilar
was later assassinated by Carrillo Fuentes, which enabled him to take control of the
Juárez Cartel. In any case, Carrillo Fuentes was not only famous for the violent way
he worked his way up but also for operating a business of massive aerial transporta-
tion of cocaine into the U.S., which gave him the title of “Señor de los Cielos (Lord of
the Heavens). In a Congressional hearing, DEA head Thomas A. Constantine stated
that Carrillo Fuentes owned several airline companies with a fleet of 727 airplanes
and an assortment of jets (Drug Enforcement Administration, 1995).
This period also witnessed the rise and consolidation of other criminal organi-
zations, such as the Tijuana Cartel (Cártel de Tijuana) led by the Arellano Félix
brothers in Baja California. During this period, the Tijuana Cartel had a close re-
lationship with the Guadalajara Cartel (Cártel de Guadalajara) founded by Miguel
Ángel Félix Gallardo, Rafael Caro Quintero and Ernesto Fonseca Carrillo. For several

5
An excellent visual documentation of the El Búfalo counter-narcotics operation and the living
conditions of peasants working in the plantation is available at: http://www.marcoacruz.com/
esclavos/portada_imagenes.html.

235
years, Félix Gallardo served as a mediator between the main drug trafficking orga-
nizations in Mexico, which gave him the nickname “El Padrino” (The Godfather).
Finally, another important criminal group that gained strength was the Gulf Cartel
(Cártel del Golfo). Initially created in the 1930s by Juan Nepomuceno Guerra, the
cartel consolidated under the leadership of his nephew, Juan Garcı́a Ábrego. This
drug trafficking organization dominated the northeastern part of the country and had
its headquarters in the state of Tamaulipas.
This was the period when numerous small, geographically concentrated groups
of drug producers and traffickers grew strong and consolidated into large “drug car-
tels”. Astorga (2003) points out that the term “cartels” does not correspond to the
concept used in economics to define a market characterized by a reduced number of
producers making agreements to restrict the supply or fix the price of a particular
good. However, the term “cartels,” coined in this period, is indicative of the growth
and consolidation that DTOs underwent during the 1980s and 1990s.
The enormous income derived from illicit markets gave Mexican drug trafficking
organizations unprecedented power to corrupt law enforcers and government officials.
Although is hard to find solid evidence on corruption activities, estimates suggest that
the Tijuana Cartel was spending $ 1 million dollars per week on bribes, and criminal
groups in Mexico spending overall between $260 and $460 million dollars a year on
corruption fees, twice the Attorney General Office’s budget (González Ruiz, López
Portillo and Yáñez, 1994; Kenny and Serrano, 2012a; Serrano, 2007).
The relative position of Mexican drug trafficking organizations further improved
towards the end of the 1990s with the crackdown on the Medellı́n and Cali Colombian
cartels. After an unprecedented escalation of drug-related violence in Colombia,
government authorities dismantled the Medellı́n Cartel when the Colombian National
Police shot down the head of the organization, the legendary drug trafficker Pablo
Escobar. In addition, by the mid-1990s, six of the seven leaders of the Cali Cartel

236
had been arrested and were later extradited to the U.S. After the demise of the two
most prominent Colombian cartels, U.S. efforts focused on increasing surveillance of
drug distribution routes in the Caribbean by means of the Joint Interagency Task
Force South (JIATF). Mexican authorities indicate that U.S. monitoring of aerial
and maritime activity in the Caribbean forced a shift of drug trafficking routes away
from the water onto land – Mexican territory.
The first manifestation of the strength of Mexican drug trafficking organizations
became evident on February 7, 1985, when DEA special agent Enrique Camarena
Salazar and Mexican pilot Alfredo Zavala Avelar were kidnapped in Guadalajara. A
few days later, Mexican authorities revealed that a group of gunmen commanded by
Caro Quintero and Félix Gallardo were responsible for their disappearance. But then,
afew days later, the chair of the DEA, Francis Mullen, declared that DFS agents had
been directly involved, covering Caro Quintero as he fled from the Guadalajara Air-
port showing a DFS badge. In March of the same year, authorities found Camarena’s
and Zavala’s bodies, bearing signs of torture. After the event, the DEA launched Op-
eration Leyenda, the largest homicide investigation that DEA had ever undertaken
abroad. DEA investigations in Mexico and other Latin American countries led to the
arrest of high-level members of drug trafficking organizations involved in Camarena’s
assasination, including Rafael Caro Quintero, Ruebén Zuno Arce, Humberto Álvarez
Machaı́n, Mario Verdugo and Ernesto Fonseca Carrillo (Drug Enforcement Adminis-
tration, 2013a,b). However, Miguel Ángel Félix Gallardo, the most important leader
of the Guadalajara Cartel remained at large. In one of the most detailed descriptions
of the Camarena case and its aftermath, Astorga (2005, 139–150) states that the as-
sassination of Camarena was committed in retaliation for his providing information
that led to the dismantling of “El Búfalo”. This version is broadly supported in other
studies (Astorga, 2003; Kenny and Serrano, 2012a; Knight, 2012).

237
 Another important factor for understanding the rise of Mexican criminal orga-
nizations during the 1980s and 1990s is the series of deep crises Mexico’s economy
suffered in 1982, 1987 and 1995. From its origins in the late 1920s to the early
1980s, the PRI was successful in generating economic growth through a number of
development strategies; industrialization, import substitution models, and stabiliz-
ing development (see Medina Peña, 1994). Between 1978 and 1981, Mexico achieved
unprecedentedly high rates of economic development with an annual average GDP
of 8.5 percent. However, during the 1980s and 1990s a series of liberalization policies
and macroeconomic mismanagement, along with recurrent election–budget cycles led
to rampant hyperinflation, the government’s declaration of financial insolvency, and
a sequence of severe economic crises. According to Lustig (1990), the 1982 crisis
severely impoverished the Mexican population by causing a 30 percent drop in real
wages, a contraction of 10 percent in wage share, and a reduction of 19 percent in
social expenditures. During the “lost decade” of economic hardship in Mexico, it is
plausible to expect that some sectors of the population came to view drug traffick-
ing organizations as an alternative option for employment in a collapsed economy;
a profitable way to climb out of poverty. As documented by historians and journal-
ists, drug lords were not only notoriously generous in distributing their wealth to
individuals, but also provided public goods; they constructed schools, installed street
lighting, built parks and churches, and held large parties for entire villages (Astorga,
1995; Campbell, 2009; Knight, 2012; Osorno, 2009; Ravelo, 2007a, 2009).
A third factor that contributed increasing the strength of Mexican drug trafficking
organizations was the liberalization of trade between Mexico and the U.S. as part of
the North American Free Trade Agreement (NAFTA). This trade agreement played
a crucial role in dramatically increasing the flow of commercial exchange between the
two countries, but the flow of Mexican goods to the U.S. became substantially larger
than the flow of U.S. products entering Mexico. According to the United States

238
International Trade Commission, before NAFTA entered into effect in 1994, the U.S.
trade balance with respect to Mexico (measured as the value of U.S. exports minus
the value of U.S. imports) was $1.3 billion dollars, meaning that the U.S. exported
more goods to Mexico than the amount imported from Mexico. Two years after
the beginning of the agreement, the U.S. trade balance had turned into a deficit
of $19.5 billion dollars; the U.S. was now importing more goods from Mexico it
exported to Mexico. This trade deficit has continued, and by 2008 it had reached
$84.8 billion dollars (Villareal, 2009). Most of this massive commercial exchange is
carried out by land transportation across the U.S.–Mexico border. The opening of the
border for trade furnished Mexican drug trafficking organizations with unprecedented
opportunities for smuggling drugs into the U.S. and weapons into Mexico.
In summary, during the 1980s and 1990s, the separate process of changes in the
international drug market structure, recurring collapses of the Mexican economy, and
trade liberalization converged to contribute to the consolidation and strengthening of
Mexican drug trafficking organizations. The increasing demand created by growing
drug consumption in the U.S. gave Mexican cartels the opportunity to obtain enor-
mous economic rents from drug markets, especially after the collapse of Colombian
cartels. The series of economic crises increased levels of poverty and inequality, thus
generating a vast human reserve of cheap, available potential workers, many of whom
saw drug trafficking activities as the only way out of poverty in an economic sys-
tem that systematically frustrated social mobility though legal means. In addition,
the reduction of trade barriers between Mexico and the U.S. and the liberalization
of commercial exchange offered prime opportunities for smuggling drugs across the
border to meet the U.S. demand. The structure of criminal organizations in Mexico
was thus no longer populated by a plethora of small atomized groups engaging in
small-scale production or transport of drugs. Instead, by the end of the 1990s, the

239
structure of the Mexican drug sector consisted of strong criminal organizations with
the capability of producing and transporting drugs into the U.S. on a large scale.

5.5.2
Dismantling
the
Political
Security
Apparatus

The second important process that took place during the 1980s and 1990s was the
dismantling of the political security apparatus that the government had employed in
previous decades to monitor and repress the political opposition. The erosion of the
coercive instruments of the state is part of a broader process of political liberalization
discussed in the next section, but it is relevant enough to deserve a separate discussion
here.
As mentioned above, the 1960s and 1970s were marked by harsh political pros-
ecution by the state against any organization or individual considered a potential
or actual political threat to the PRI hegemony. The state created the DFS as a
sophisticated instrument of political repression and social control. The DFS exten-
sively infiltrated almost every open or clandestine power structure, thus centralizing
a vast network of information and political relations. The close connection between
the DFS and the highest levels of the political elite enabled timely, precise informa-
tion to be provided about potential threats and an effective repressive reaction to be
immediately delivered by the state. Under the protection of political immunity, the
DFS relied on a broad menu of coercive tactics including arbitrary detentions, forced
disappearances, torture, mutilation, extortion and rape, among others (Castellanos,
2007; Doyle, 2006a,b; Montemayor, 2007, 2009, 2010; Poniatowska, 1980). In coordi-
nation with federal and local police forces as well as the Army, the DFS orchestrated
vigilance, control and coercion of students, left-wing political organizations, teachers,
labor unions, peasants and indigenous organizations, religious leaders, independent
media and opposition parties, among others.

240
 Following the sustained wave of political repression that was carried out through-
out the 1960s and 1970s, the government had crushed most radical organizations;
arrested, killed or coopted their leadership; scattered their members and supporters;
neutralized urban guerrilla cells; and eliminated insurgents in the mountains. By the
end of the 1970s, the remaining dissidents were too few and dispersed to represent a
political threat to the regime. The coercive apparatus of the state had prevailed af-
ter twenty years of struggle against political dissidents. During this period, the DFS
became deeply interwoven into criminal networks and directly participated in drug
trafficking and other criminal activities. In contrast to the harsh repression suffered
by political dissidents, criminals enjoyed the protection and often the direct support
of the state security apparatus and the political elite.
When the PRI entered the 1980s, there were no remaining radical political threats
capable of jeopardizing the hegemony or survival of the regime. The demise of domes-
tic political threats was accompanied by the end of the Cold War. In the international
arena, the political and economic liberalization carried out by Mikhail Gorbachev in
Russia and the subsequent fall of the Berlin Wall marked the end of an era of in-
ternational polarization. At the end of the 1980s, the termination of the Cold War
also relaxed the pressure which the U.S. foreign policy had put on Mexico to prevent
expansion of the communist threat. International factors and domestic politics con-
verged during the 1980s to alleviate any concerns the PRI might have for its political
survival.
In this context, President Miguel de la Madrid used his executive powers to dis-
band the DFS in 1985. The dismantlement of Mexico’s most feared political police
initiated the crumbling of the state’s security apparatus. Several authors indicate
that one of the central causes of the end of the DFS was the deep involvement of
its agents in the murder of DEA agent Camarena (Borjón Nieto, 2008; González
Ruiz, López Portillo and Yáñez, 1994; Serrano, 2007). In addition, the last direc-

241
tor of the DFS, José Antonio Zorrilla Pérez, was arrested for the assassination of
Manuel Buendı́a, a prominent journalist who conducted several investigations about
the relationships between the DFS and drug trafficking organizations until his death
in 1984. The purge of the DFS immediately resulted in the dismissal of 427 agents
(Vasquez, 1985). The house-cleaning rapidly extended to the Federal Judicial Police
and the Office of the Attorney General. According to Serrano (2007, 272), between
1986 and 1996, about 7,000 members of the security forces were fired from their re-
spective agencies. Subsequently, 1,200 Federal Judicial Police officers and about 30
percent of the agents Attorney General’s office were discharged between 1994 and
2000. The Federal Judicial Police was effectively dismantled in 1998 and substituted
by the Federal Preventive Police (PFP). A few years after the closure of the DFS,
in 1989, President Carlos Salinas created a new intelligence agency called Centro de
Investigación y Seguridad Nacional (CISEN, Center of Investigations and National
Security) and commissioned General Jorge Carrillo Olea to be the director of this
new agency. At the time, the U.S. Drug Enforcement Administration became the
most valuable source of information for Mexican authorities about the activities of
their drug trafficking organizations.
The dismantling of the DFS and other federal security forces had two important
consequences. First, after thousand of security agents had been discharged, a large
proportion of them joined existing criminal organizations or created new ones; others
joined local police forces in other states and kept doing business as usual; and others
created their own private security firms. Those who joined criminal organizations
took with them the deep and specific knowledge of the structure, operations, tactics
and personnel of security agencies that they had acquired during their years of service.
Their “know-how” and experience of state security operations became an asset for
criminal organizations, thus improving their ability to elude law enforcement.

242
 The second and most important consequence of the massive dismissal of security
agents was the loss of a crucial instrument for controlling the criminal network. For
four decades, the DFS had served as a mechanism of social and political containment
and had held a strong grip on a complex system of criminal and political relation-
ships. There is no doubt that the DFS abused its position of political impunity and
engaged in criminal activities as licensed predators. However, they also served as a
bridge that connected the political world with the criminal underworld and provided
valuable information about criminal networks. Purging security agents who had close
connections with criminal groups might have served to ameliorate the pervasive cor-
ruption in Mexican security forces but it also eroded the ability of state authorities
to monitor and control criminal groups.
As discussed in the previous sections, government authorities did not primarily
rely on the use of force to contain the inherently violent behavior of criminal groups.
Coercion was not largely necessary because the state kept a close eye on criminal
behavior and used timely, precise information to activate political mechanisms of
discipline. However, by disbanding its security agencies and scattering their person-
nel, the political elite lost control of the large dense network of criminal–political
connections. The dismantling of the DFS effected a huge loss in the state’s abil-
ity to impose political control on criminals. The strong U.S. pressure that followed
Camarena’s assassination probably contributed to the President’s decision to purge
the state’s coercive apparatus. In addition, due to the extermination of subversive
organizations and the weight of the Cold War off his shoulders, he probably faced
no direct political concerns or pressures that could justify maintaining the DFS or
preventing its elimination.

243
5.5.3
The
Process
of
Democratization
in
Mexico

During the 1980s and 1990s, Mexican drug trafficking organizations acquired sub-
stantial power due to the surge in drug consumption in the U.S. During this period,
the removal of domestic and international Cold War pressures contributed to the dis-
mantling of the political police. In consequence, the state lost a crucial instrument for
monitoring criminals and instilling discipline through political incentives. However,
the most important factor that eroded the preexisting order in drug markets is the
advent of democracy in Mexico.
According to Brachet-Marquez (1992), the debate about the process of democratic
transition in Mexico centres on two different processes: one citing democratization
from top-down and the other one arguing that political change developed as a bottom-
up process. As accurately claimed by Merino (2003), the best way to understand the
process of democratic transition in Mexico is by analyzing the interaction between
the electoral system and the party system. Both sides of the debate to be drawn
together in order to provide an integrated explanation of bottom-up democratization
pressures and top-down electoral concessions to facilitate political competition. In
contrast to processes of democratization undertaken by political pacts at the elite level
in other countries (Linz and Stepan, 1996; O’Donnell, Schmitter and Whitehead,
1986), the Mexican democratic transition is characterized by a gradual process of
political liberalization. Despite the constricting political conditions imposed by PRI
hegemony, opposition parties pushed for small electoral reforms that allowed them
to secure some political positions, which in turn helped reconfigure the party system
to allow opposition parties to push for further electoral reforms.
In the 1958 presidential election, the PAN accused the PRI of rigging the election
results, and refused to take the seats that it had won in Congress. This opposition
boycott called the legitimacy of the electoral process into question and forced PRI
to make some small, but important electoral concessions. The result was the 1963

244
electoral reform that introduced the party–deputy system (“Diputados de Partido”),
which awarded five seats to minority parties that reached a threshold of 2.5 percent
of the vote. However, the reform included a safeguard for the PRI by stipulating that
minority parties could not hold more than twenty seats. This might seem a small
concession, especially for a Congress that had 210 seats overwhelmingly controlled by
the PRI. However, it was an important first step for opening up the political arena
that had been dominated by a single party since 1929. The party–deputy system re-
mained in effect for ten years and, in 1973, conditions of electoral competition further
improved with the reduction of the threshold of votes required from 2.5 percent to 1.5
percent. In that year, the maximum number of minority party seats also increased
from twenty to twenty-five seats.
As discussed above, the 1960s and 1970s were marked by inexorable repression
of leftist organizations and political parties. Student and labor movements were vi-
olently suppressed, as were left-wing parties such as the Mexican Communist Party.
Those who radicalized and engaged in urban and rural guerrilla operations were ul-
timately neutralized by the mid-1970s. After twenty years of repression, the left was
devastated. On the other side of the political spectrum, the right-wing opposition
represented by the conservative PAN had not been severely repressed, but was never-
theless marginalized from power by systematic electoral fraud. Instead of employing
radical contestation, the PAN decided not to run a presidential candidate in the 1976
election. It was not a deliberate decision to boycott the election; rather, internal
struggles prevented the PAN from nominating a candidate. In any case, the lack of
contenders exposed the unbearable political conditions imposed by the PRI, which
presented José López Portillo as the only candidate for the presidential election. As
indicated by Merino (2003, 23), “the election campaign of López Portillo was an
absolute contradiction: the PRI was at its peak of hegemonic power, but had no
electoral legitimacy.”

245
 Violent contestation from the left and the electoral boycott from the right forced
the PRI to improve the conditions for electoral competition. The result of this process
was the 1977 electoral reform, which introduced proportional representation (PR) in
Congress. The reform increased the number of seats in Congress to 400, opening 100
seats for plurality rule and the remaining 300 for majority rule. As shown in Figure
5.1, increasing the number of seats, especially the proportional rule seats, increased
the availability of positions for minority parties.6 Another important characteristic of
the reform was the reduction of barriers for forming and registering political parties.
The barriers to obtaining “definitive” registration were lowered, and small parties
were allowed to obtain “conditional” registration that would become “definitive”
upon their receiving at least 1.5 percent of the national vote. As stated by Molinar
Horcasitas and Weldon (2003), this concession allowed the Mexican Communist Party
to return from the shadows and gain legal registration. The electoral reform was
part of a broader effort to bring the leftist opposition back to the electoral arena
as the government passed a Federal Amnesty Law in 1978 to exonerate members
of subversive and guerrilla organizations who were in prison (Congreso de la Unión,
1978).
The 1977 reform was the turning point that marked the opening of the political
system. The electoral concessions did not break the hegemony of the PRI; the party
was still winning with large margins of victory. But the reform allowed the opposition
the possibility of amplifying their representation in Congress by winning access to
more seats. More importantly, the system of proportional representation created
incentives for opposition parties to build local constituencies at the state level so they
could access PR seats. In addition, the electoral reform allowed county councillors
(regidores municipales) to be elected by proportional representation, also reinforcing

6
Data from 1951 to 1997 is from Lujambio (2000), the remainder is from Instituto Federal
Electoral (2013).

246
 550
500
450

Number of seats in Congress

400
350
300
250
200
150
100
50
0
1961
1964
1967
1970
1973
1976
1979
1982
1985
1988
1991
1994
1997
2000
2003
2006
2009
2012
Proportional representation Majority rule Total

Source: Made by the author with data from Lujambio (2000)

Figure 5.1. Number of seats in Congress by type of electoral rule

the incentives to develop electoral support at the local level. The incentives towards
the development of decentralized electoral support planted the seeds for a process of
“democratization from below,” through which opposition political parties gradually
increased their strength on the periphery (Lujambio, 2000; Merino, 2003). The effect
of the 1977 electoral reform is clear in Figure 5.2, which shows the share of votes
per party in Congressional elections for the Chamber of Deputies.7 The figure shows
that after the 1977 reform, the PRI’s share of votes in Congress declined while the
opposition parties blossomed. This trend of declining PRI control over the Congress
continued in subsequent elections.
The electoral reform and its consequences opened up the first cracks in the strong
centralist control that the PRI exercised over the periphery. As the opposition started
building support at the local level, the cracks grew deeper and eroded the PRI’s abil-

7
Data for this figure comes from Instituto Federal Electoral (2013) and Shirk (2005).

247
 100
90
80
70

Percentage of votes
60
50
40
30
20
10
0
1952
1955
1958
1961
1964
1967
1970
1973
1976
1979
1982
1985
1988
1991
1994
1997
2000
2003
2006
2009
2012
PRI PAN PRD/Left Other

Figure 5.2. Percentage of votes for Congressional elections

ity to monitor local politicians and enforce discipline on them. The conservative
PAN began gaining increasing support in the north. Mexico’s northern states were
more industrialized than the rest of the country and they also had a politically active
Catholic community – features that coincided with the ideological profile of the PAN
and characterized the party’s core constituency (Loaeza, 1994). The northern states
already had a long presence of drug trafficking organizations, which benefited from
the increasing strength of the opposition to the detriment of the PRI as it meant
a looser grip from the centralized control of federal authorities. An example of this
political decentralization process was the PAN victory in the Ciudad Juárez, Chi-
huahua municipal election in 1983, which marked the end of the PRI’s protracted
political control in that municipality. Another example is Baja California, which
became the first state governed by an opposition party when PAN candidate Ernesto
Ruffo Appel defeated the PRI candidate in the election for governor in 1989. There

248
is no evidence of the Juárez Cartel being involved in the PAN victory in Chihuahua
nor the Tijuana Cartel in Baja California. However, after the opposition victory in
their local governments, these drug organizations certainly benefited from a less strict
control by the PRI in the center of the country.
The opposition began gaining strength at the local level. Increasing competition
led to a series of electoral conflicts at the state level after the PRI rigged the results
of local elections in the early 1980s. The most competitive states were in the north,
including Sonora, Baja California, Chihuahua and Nuevo León, the most distant from
the centralized PRI control in Mexico City. Civil resistance in Chihuahua escalated
after the overt electoral fraud committed in the 1986 election. In addition, social
unrest went widespread in the country due to the discontent caused by the severe
economic crisis of 1982 and the feeble government reaction to aid the population after
the earthquake that devastated Mexico City in 1985. In response, President de la
Madrid advanced a new electoral reform in 1986, increasing proportional representa-
tion in Congress from 100 to 200 seats while keeping the number of seats allocated
by majority rule at 300. The increase in PR seats in 1986 is clear in Figure 5.2.
However, in order to prevent losing control of Congress, the PRI allowed itself to
also benefit from the increased number of PR seats, which were no longer reserved
for the opposition. Besides increasing the potential for opposition representation, the
reform allowed the creation of electoral coalitions among parties for the first time.
The possibility of creating electoral alliances proved crucial for making inroads on
the PRI’s dominance in the 1988 presidential election.
In 1987 a group of PRI politicians led by Cuauhtémoc Cárdenas, son of Gen-
eral Lázaro Cárdenas who was president from 1934 to 1940, challenged the PRI’s
traditional selection of the presidential candidate by “dedazo,” and proposed a demo-
cratic selection mechanism based on direct voting of party members. President de
la Madrid interpreted this as a blatant challenge to the presidential powers that had

249
long been mandated by the unwritten party rules and imposed Carlos Salinas de
Gortari as the official PRI candidate. The decision caused a split between Cárdenas
and other prominent PRI members from the party. Based on the recently passed
electoral reform that allowed the creation of electoral coalitions, a group of leftist
minority parties coalesced into the National Democratic Front, (Frente Democrático
Nacional, FDN) and proposed Cárdenas as the leftist candidate. Six months after
the election, the FDN coalition became the Democratic Revolutionary Party (Partido
de la Revolución Democrátic, PRD). Cárdenas was put forth as a serious contender
to run against the PRI in the 1988 presidential election. In a matter of few months,
his candidacy provoked vibrant enthusiasm and won massive electoral support from
broad sectors of the population. The left managed to channel social discontent with
the political system, critiques of a series of neoliberal policies that had started in
the 1980s, and generalized frustration with the fall in living standards caused by the
severe economic crisis of 1982 (Bruhn, 1996).
On election day on July 6, 1988, the first results favored the FDN; as the day drew
on, the distance between the PRI and FDN became narrower. Panic seized the PRI
political elite and the Minister of the Interior – who also headed the Federal Electoral
Commission – ordered the computer system used for counting the votes to be shut
down. The PRI claimed that the system had inexplicably collapsed, and declared
a victory for PRI candidate Carlos Salinas with 50.7 percent of the vote, awarding
second place to the FDN candidate, Cuauhtémoc Cárdenas, with 31.1 percent of
the votes. The “collapse of the system” raised serious doubts as to the legitimacy
of the election and unleashed massive protests and demonstrations on the streets
(Bruhn, 1996; Loaeza, 1999). As illustrated by Figure 5.3, the 1988 election marked
a dramatic increase of votes for the opposition and accentuated the decline of the
PRI vote share.8

8
Data for this Figure comes from Instituto Federal Electoral (2013).

250
 100
90
80
70

Percentage of votes
60
50
40
30
20
10
0
1952

1958

1964

1970

1976

1982

1988

1994

2000

2006

2012
PRI PAN PRD/Left Other

Figure 5.3. Percentage of votes for Presidential elections

One of the most insightful analysis of Mexican politics under authoritarianism is

provided by Magaloni (2006). She accurately argues that the PRI suffered splits in
1940, 1946, 1952 and 1988, all of which occurred as prominent politicians objected
to the presidential nomination process.9 The case of Cárdenas in 1988 was an excep-
tion; the PRI managed to recover from all other previous splits because it enjoyed
huge margins of victory. The PRI placed a great deal of emphasis on mobilizing
voters to the polling booths as a key strategy to deter elites from splitting off, even
when elections were not competitive. In contrast to the first three splits, Cárdenas
managed to split off and successfully challenge the hegemony of the PRI because the

9
According to (Magaloni, 2006), the most important splits were those initiated by General Juan
Andreu Almazán in 1940, who opposed Manuel Ávila Camacho as the presidential nominee; Ezequiel
Padilla Peñaloza in 1946, who rejected the nomination of Miguel Alemán Valdés; Miguel Henrı́quez
Guzmán in 1952, who stood against the appointment of Adolfo Ruiz Cortı́nez as presidential can-
didate; and Cuauhtémoc Cárdenas in 1988, who protested against Carlos Salinas as nominee.

251
electoral reforms in 1977 and 1986 created favorable political conditions for winning
substantive electoral support outside the PRI.
In December, 1988, Salinas de Gortari took office as president. One month later,
in January, 1989, the Army arrested Joaquı́n Hernández, also known as “La Quina”,
the leader of the the state oil company Petróleos Mexicanos (PEMEX) labor union.
This was a crucial power move because La Quina had begun to challenge the neolib-
eral policies favored by Salinas, and leftist candidate Cárdenas won a considerable
vote share in districts populated by PEMEX workers. The event known as “El
Quinazo” was hailed as the end of the old-time union bossism. In addition, it was
broadly recognized that La Quina was a scapegoat sacrificed to legitimize Salinas’s
government following the electoral fraud. “El Quinazo” came to symbolize an iconic
power move in Mexican politics that signalled a rupture with past corrupt agreements
tolerated by previous administrations. Another important power move to legitimize
his government was the arrest of Miguel Ángel Félix Gallardo in April, 1989, only
four months after Salinas took office. The long-standing leader of the Guadalajara
Cartel was directly associated with the murder of DEA agent Camarena and had
managed to elude Mexican and U.S. efforts to capture him for more than a decade,
until arrested by Salinas. The press celebrated the capture of Félix Gallardo as “a
major political accomplishment for President Carlos Salinas” (Rother, 1989).
The social unrest and political polarization caused by the electoral fraud in 1988
forced President Salinas to make further electoral concessions in 1990. The most im-
portant element of this electoral reform was an effort to dissipate doubts about the
legitimacy of future elections through the creation of the Federal Electoral Institute
(Instituto Federal Electoral, IFE), an autonomous institution in charge of organizing
elections. The reform also included the creation of a professional bureaucracy with
the mandate of implementing clean and transparent elections. Although the electoral
administration was no longer under the jurisdiction of the Ministry of the Interior

252
but of the newly created IFE, the PRI did not lose absolute control over the electoral
bureaucracy, as the Minister of Interior was still the head of the IFE electoral com-
mission. Later, another electoral reform in 1993 gave the IFE the power to limiting
campaign expenditures. This placed an important constraint on the PRI’s use of
public funding and government infrastructure at the federal and local level.
1994 was a turbulent year for Mexico marked by the return of political violence.
On January 1, 1994 – the very day the NAFTA agreement went into effect – the Zap-
atista National Liberation Army (Ejército Zapatista de Liberación Nacional, EZLN)
burst onto the political scene in the southern state of Chiapas. The Zapatista up-
rising was dampened by the Mexican Army in a matter of a few days. However,
their ideological stand found deep resonance in broad sectors of the population and
triggered massive non-violent demonstrations in support of the EZLN (Hayden, 2002;
Womack, 1999). A few months later, in March, 1994, the PRI presidential candidate
was assassinated during a campaign rally in Tijuana, Baja California. The assassi-
nation was a major shock for the Mexican political elite because no major political
figure had been killed since the aftermath of the revolution and the creation of the
party in 1929 (Aguilar Camı́n, 2004, 2006). Later that year, José Francisco Ruiz
Massieu, the president of the PRI was assassinated in Mexico City. In the voice of
the media, Mexico was “bordering on chaos” (Oppenheimer, 1996).
The reaction of the political elite to the violent events was the approval of a major
electoral reform in May, 1994, barely two months before the presidential election. The
reform created the figure of Citizen Counselors (Consejeros Ciudadanos) who were
selected by Congress from a pool of citizens without partisan affiliation. Citizen
Counselors were part of a collective decision-making board chaired by the Minister
of the Interior. This was the first step in the creation of a truly independent electoral
authority outside the political control of the government apparatus and the PRI
(Peschard, 1995). The second and definitive step came in 1996.

253
 The 1996 electoral reform finally removed the Minister of the Interior from the
board of directors of the IFE and gave complete administrative and decision-making
autonomy to the board controlled by non-partisan citizens, now called Electoral
Counselors (Consejeros Electorales). The reform also included an increase of seats in
the Senate, which grew from 62 to 128 seats. There were several other changes that
improved the conditions for electoral competition under fair and transparent rules.
The 1996 electoral reform represented the culmination of a protracted process of in-
stitutional development. It took three decades to completely erode the longstanding
de facto control held by the PRI over nominations at the federal and local levels; the
result was an independent, solid institute in charge of regulating electoral competition
(Becerra, Salazar and Woldenberg, 1997, 2011). The creation of an autonomous elec-
toral institute supported by a strong body of laws and regulations finally consolidated
the two most important elements of a democratic political system: the certainty of
electoral rules and the uncertainty of electoral results (Przeworski, 1991).
The 1996 electoral reform had a profound impact on the reconfiguration of the
political scene in Mexico. As reflected in Figure 5.2 reporting the vote share for
Congressional elections, for the first time since its foundation the PRI lost its majority
in the Chamber of Deputies in the 1997 election. The reform also contributed to the
momentum of political decentralization at the sub-national level. By 1998, seven
states had already had governors from a party other than the PRI; Baja California,
Chihuahua, Guanajuato, Jalisco, Nuevo León, Guerétaro and the Federal District
(Mexico City). According to Lujambio (2000), the distribution of power in the local
legislative assemblies also underwent a substantial process of political pluralization:
in 1974, the PRI controlled 97.8 percent of all 369 legislative seats available at the
local level; by the end of 1999 the PRI was left with only 49.6 percent of the 1,108
local legislative seats in the country. The number of divided governments – in which
the governor is from a different party than the one controlling the majority of the local

254
legislative assembly – increased from one (the first) divided government in 1989 to
fifteen experiences of divided government by 1999. The diversification of the political
scene also extended to the municipal level. By 1998, more than half of the state’s
capitals were governed by opposition parties: 15 state capitals belonged to the PAN;
3 were controlled by the PRD (including Mexico City); one more was controlled by
the PT; and the remaining 13 state capitals remained under PRI control (Merino,
2003, 40). Between 1978 and 1980, the opposition controlled only 34 municipalities.
In contrast, between 1996 and 1998, the number of municipalities governed by the
opposition increased to 510 (Eisenstadt, 2003). Most of these changes at municipal
level occurred in urban areas concentrating a large proportion of the population.
Finally, the most important effect of the 1996 electoral reform is that it paved the
way for the 2000 electoral process in which the PRI lost the presidency for the first
time. July 2, 2000, is marked as a critical inflection point in Mexican politics when
a majority of voters brought to an end the worlds oldest hegemonic party regime,
which had remained in power for more than 70 uninterrupted years. On that day,
the PAN candidate, Vicente Fox, won the election by a comfortable margin of more
than 6 percent over the PRI candidate, Francisco Labastida. As reflected in Figure
5.3 showing the percentage of presidential votes obtained by each political party, the
share of votes for the PRI had been declining since 1982 until the dominant party
was finally defeated in 2000.
There is a broad range of explanations attempting to account for why the PRI
lost the presidential election in 2000 (see Brachet-Marquez, 1992). The most convinc-
ing explanation is offered by Magaloni (2006) who argues that the PRI lost power
because of two factors: it failed to offer sufficient material rewards and access to
government office to the many ambitious politicians within the party; and it failed
to mobilize voters in sufficient numbers to win with the wide margins of victory that
could have generated a strong message of invincibility. Greene (2007) offers a similar

255
view, also based on a twofold argument. One factor in the defeat of the PRI is that
the party lost the resource advantage of limitless spending on campaigns, in which
it used to supplement policy proposals with patronage and vote buying. The other
factor was the PRI’s diminished ability to impose participation costs on the opposi-
tion by increasing opportunity costs or employing violent intimidation. Additionally,
Dominguez and Lawson (2003) offer another explanation, namely that the PRI lost
the election because of the effectiveness of short-term effects of the 2000 election
campaign.
Figures 5.2 and 5.3 presented above show that after the 2000 election, political
competition increased in the Chamber of Deputies and at the presidential level. In
addition, as reflected in Figure 5.4, after the 2000 presidential election, political
diversity in Mexico became well rooted in the national political scene.10 The Figure
reports the effective number of political parties (ENP) at the federal level calculated
according to the formula provided by Laakso and Taagepera (1979). The most diverse
political sphere was the Chamber of Deputies, thus reflecting the plurality of political
parties with different strongholds at the sub-national level that had been increasing
between 2000 and 2010. The figure also shows that there was a marked increase in
the effective number of parties in the Senate and in the 2006 presidential election. In
general, federal political spheres are characterized by the presence of at least three
effective political parties, the PRI, PAN and PRD. The ENP in the lower house of
Congress has a larger value due to the diversity of minority parties with seats in the
Chamber.
Finally, Figure 5.5 reports the effective number of political parties at the local level
from 2000 to 2010.11 The figure shows that the PRI did not have hegemonic control

10
Electoral data comes from Instituto Federal Electoral (2013) and the ENP is calculated by the
author using the Laakso and Taagepera (1979) formula.
11
Election data for both types of elections come from Instituto Federal Electoral (2012) and the
ENP is calculated by the author using the Laakso and Taagepera (1979) formula.

256
 Figure 5.4. Effective number of parties (ENP) at the federal level

over political positions at either the state or municipal level. Panel (a) reports the
box plots of ENP for governorships for all 32 states. The figure shows that the yearly
averages of ENP for governors range from 2.2 to 2.6 effective parties. This indicates
that state politics were mostly defined by bipartisan competition, usually PRI vs PAN
or PRI vs PRD. In addition, Panel (b) reports the ENP for municipal elections. The
average effective number of political parties for the election of mayors ranges from
2.8 to 3.6 political parties. In addition, the figure shows that some municipalities had
higher than average ENP. This indicates that municipal elections are substantially
more competitive that the state or federal political spaces. In general, these plots
show that political diversity was well entrenched at the sub-national level.
As discussed above, the sustained interaction between the electoral system and the
party system over a period of thirty years led to a gradual process of democratization
that eroded the PRI’s hegemonic control at all three levels of government. By the
beginning of the 2000s, the Mexican political scene was characterized by a plurality

257
 (a) ENP for Governor (b) ENP for Major

10

10
8

8
Effective number of parties
Effective number of parties
6

6
4

4
2

2
0

0
2000

2001

2003

2004

2005

2006

2007

2009

2010

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010
Figure 5.5. Effective number of parties (ENP) at the local level

of actors with strong roots at the sub-national level and with an active role at the
federal level. In 2000, the hegemonic party that had ruled Mexico for seven decades
finally lost the presidency for the first time, thus ending a long period of political
dominance. But even well before the presidential election of 2000, the map of political
diversity gave a picture of firmly rooted plurality. The presidential alternation of
2000 was the climax of thirty years of effort to open up political competition in
Mexico. This process of democratic transition was characterized by the gradual
evolution of electoral reforms that managed to increase the certainty of the rules of
electoral competition while generating democratic uncertainty about the winner of
the competition.

5.5.4
Not
All
Good
Things
Come
Together

The gradual process of democratization in Mexico dismantled the hegemonic-

party autocracy imposed by the PRI for seven decades. Although there are certainly

258
key aspects of the Mexican democratic system that need to be improved, there is no
doubt that the democratic transition brought substantial opportunities for political
liberalization and economic development. Several of these pending tasks constitute
what researchers have called an “incomplete,” “diminished,” or “delegative” democ-
racy (Bartra, 2013; Felsen, 2009; Laurell, 1992; O’Donnell, 1994; Rodrı́guez Araujo,
2009; Rubio and Kaufman, 2006; Tuckman, 2012). Regardless of the “adjectives” used
to describe the performance of the Mexican democracy (Collier and Levitsky, 1997),
the process of democratic transition enabled voters to “kick the rascals out” not only
from the presidency but from several states and municipalities. Democratization en-
tailed a fundamental realignment of political forces. It replaced a hegemonic political
system rooted on a centralised, hierarchical political structure, where the only way to
access power was through the party, with a system that allows political diversity on
the basis of solid electoral rules. Democratization thus subverted the lack of division
of powers imposed de facto by the informal rules of the PRI hegemony and activated
a genuine system of checks and balances in which the PRI was reduced to being one
of three major parties on the political scene.
With the collapse of PRI hegemony, democratization also eroded the feasibility
and sustainability of corrupt pacts between political actors and criminal organiza-
tions. The PRI no longer held the reins of a hierarchical chain of command that
reached from the office of the president to the most remote municipality in the pe-
riphery. In this way, the president lost its capability to impose political discipline
on the bureaucracy and party members at lower levels. The political capacity to en-
force discipline also broke down in the criminal sphere. This argument is consistent
with the effect of democratization on the war on drugs described by Knight (2012)
and Rios (2012a). However, based on the premises of the theoretical model, this
chapter offers a more detailed discussion on the mechanisms operating behind this
relationship.

259
 The erosion of the political capacity to impose discipline on criminals was caused
by three essential characteristics of democracy: an increased number of political
parties, effective elite circulation though electoral means, and political incentives
for enforcing the law in order to gain popular support. The first two factors erode
corrupt pacts between criminals and political actors, but do not necessarily lead to
the escalation of violence. In contrast, the third factor, referring to the incentives to
enforce the law, is a crucial factor for the escalation of drug-related violence.
The increased number of political parties is particularly important in the Mex-
ican case due to the single-party system that had characterized the political scene
for many decades. The diversity of political actors that accompanied democracy in-
creased the difficulty in overcoming problems of collective action that could facilitate
the implementation of corrupt agreements. During the era of PRI hegemony it was
enough to establish an agreement with a few members of the political elite to im-
plement a pact. Under democracy, such attempt would require bargaining with a
diverse array of actors probably including the president, a few generals, and several
governors down to a multitude of mayors, all of which may have different party affil-
iations. Even if an agreement is achieved, actors from different political parties have
incentives to denounce the agreement in order to expose the other parties, making
the agreement inoperable.
The entrance of new political actors at every level of government also disrupted
the hierarchical chain of command that had facilitated the implementation of corrupt
agreements. The advent of democracy augmented the number of cases of governments
divided across political branches or levels. In several instances, the party of the pres-
ident was no longer the same as the party holding the majority in the Senate or
the Chamber of Deputies. Often, neither the party of the governor nor the mayor
coincided with the party of the president. This rupture prevented the top-down in-
strumentation of presidential orders. The political survival of government officials or

260
party members thus no longer depended on compliance with the will of the president
or other members of the political elite. Actors from different political denomina-
tions had their own incentives to align with the will of those above them in their
own party structures or with the preferences of their electoral constituencies. This
meant a fundamental change in the structure of political incentives now operating in
a democratic setting.
In addition, the rupture in the hierarchical chain of command broke the mecha-
nism enabling the bottom-up information flow. This problem became more salient
after the dismantling of the DFS and the purge of other security forces. The state
could no longer count on having its political police agents deeply infiltrated into the
criminal underground and keeping a close eye on criminals as they had during the
heyday of PRI dominance. This severely undermined government authorities’ ability
to monitor the behavior of criminal organizations and to impose political discipline.
This situation is clearly illustrated by Knight (2012, 129), who argues that “a call
from Los Pinos, the presidential palace, could no longer help settle political disputes
or narco turf wars; Los Pinos no longer called, or, if it did, there was no answer.”
Arguments about state-sponsored protection rackets (Duran-Martinez, 2012; Sny-
der and Duran-Martinez, 2009) claim that violence in drug markets emerges when
political actors providing protection or selective enforcement can no longer mount
a credible coercive threat to impose discipline. Along with this argument, a broken
chain of command undermines the ability to coordinate law enforcement between fed-
eral (e.g. Army and Federal Police) and local security forces (e.g. state and municipal
police). Although state coercion might be relevant as an method for deterring violent
criminal behavior, the most important consequence of a broken chain of command is
the erosion of the state’s capability to impose discipline through political sanctions
and to collect precise and timely information about criminal operations.

261
 The second factor eroding the sustainability of corrupt agreements is the effective
circulation of political elites through elections. During the era of PRI hegemony,
elections were not only tainted by electoral fraud but, as accurately stated by Mag-
aloni (2006), mass mobilization of voters served as a key mechanism to deter splits
from the party. Both electoral fraud and voter mobilization served to send a clear
message that the prospects of political success outside the PRI were non-existent.
This system helped the PRI to perpetuate itself in power at all levels of govern-
ment for seven decades. Elections were a democratic facade that only facilitated elite
circulation within the PRI. Once the presidential nominee was selected by dedazo,
there was absolute certainty among citizens and politicians that the winner of the
election was going to be the designated PRI candidate. The same system operated in
the other tiers of government. This characteristic motivated Nobel Laureate Mario
Vargas Llosa (1990) to characterize PRI-ruled Mexico as “the perfect dictatorship.”
And then, with the advent of democracy, the system collapsed.
As discussed above, the process of democratization in Mexico evolved by means of
a sequence of electoral reforms that gradually made room for opposition and increased
the legitimacy of the electoral authority. The main achievement of the opposition
was to extract the organization of elections out from under the PRI’s control and to
create an autonomous, trustworthy institution capable of organizing fair, transparent,
competitive elections.
Under democracy, elections serve as a key mechanism for the effective circulation
of political elites between the different parties and across all levels of government.
The certainty about a PRI victory in the next election broke down. In consequence,
any longterm expectations politicians might hold for political careers under the PRI
also evaporated. Today, political survival and prospects for political ascent no longer
depend on elite decisions in favor of “well-behaved” party members. Elections provide
no certainty that those on top will remain there in the future to deliver any promised

262
benefits or to apply guaranteed sanctions. The system of incentives that allowed the
PRI to instill and impose discipline is no longer in operation. Just as democratization
eroded political discipline within the PRI, the relationships between criminals and
political actors collapsed in disarray.
Elections place an expiration date on agreements between criminals and corrupt
political actors. There is no longer certainty that the pacts will be maintained after
an election for the simple reason that democracy introduced uncertainty about the
election result. The system of incentives no longer rewards drug trafficking organiza-
tions for complying with the code of conduct imposed by the government. Political
protection for “well-behaved” criminal groups is irrelevant in the long run. Corrupt
political actors only guarantee protection or non-enforcement from their own secu-
rity forces during the tenure of their administration. Even if they do so, they cannot
guarantee that the security forces of other government authorities operating at dif-
ferent levels of government will adhere to the agreement. Criminal groups seeking
protection from the authorities now have to bargain with a plurality of political ac-
tors with diverse party labels across different tiers of government. Even if criminals
manage to obtain a non-enforcement agreement, political actors operate according
to the electoral clock and such an agreement is likely to expire. Moreover, the sta-
bility of such a complex agreement – if ever achieved – is contingent on the absence
of rival criminal organizations trying to operate in their area. The introduction of
competitors trying to secure agreements with government authorities increases the
volatility of an already unstable and uncertain situation. Devoting time, effort and
resources to trying to secure an overarching and stable non-aggression has become a
futile exercise.
The direct implication for the uncertainty of receiving protection from government
authorities is that criminal organizations will try to find the way to ensure their own
security. As also indicated by Rios (2012a) and Snyder and Duran-Martinez (2009),

263
drug trafficking organizations could try to acquire the necessary weaponry and men
to protect their drug shipments from government crackdowns and to secure their
borders against rival criminal groups. This situation could easily lead to a security
dilemma and a subsequent arms race among rival criminal organizations (Posen,
1993). Moreover, as indicated by Lessing (2012), criminals could use the threat of
violence against government authorities to ensure protection by offering them the
choice between “plata o plomo” (silver or lead).
These incentives can certainly lead to a precarious equilibrium in which rival
criminal organizations would increase their military capabilities to protect their ter-
ritories. These mechanisms are plausible due to the high stakes of drug trafficking
since the 1980s and the availability of military-style weaponry since the expiration
of the U.S. Federal Ban of Assault Weapons in 2004. However, this does not mean
that criminal organizations will unilaterally engage in violence against the state or
against other criminal groups. In this respect, the initial upgrade in military capa-
bility serves a defensive purpose – not an offensive one – with the main objective of
protecting their drug shipments and areas of operation. In other words, the original
reason to enhance their own security is to maintain the status quo, not to challenge
it for expansionist purposes.
The third effect of democratization on eroding the political-criminal relationships
refers to the system of incentives it creates for politicians to enforce the law in order
to gain popular support. Under democracy, the prospects of a successful political
career no longer depend endorsing the political elite in control of the party but on
securing the favor of the electorate. Democratization improves the responsiveness
of political actors by giving them motivation to satisfy the demands raised by their
constituency, thus favoring the delivery of public goods.
Augmented responsiveness from Mexican government officials motivated an in-
crease in law enforcement to meet citizen demands for public security during the

264
1980s and 1990s. At the time, crime was becoming an increasingly important concern
for the population. There is no systematic public opinion data about the magnitude
of this concern during those years, but the official mortality statistics show a substan-
tial increase in homicides over the 1980s and 1990s. Figure 5.6 reports the annual
number of homicides at the national level from 1938 to 2011.12 Although these data
include all types of homicides13 – thus, they should not be interpreted exclusively
as the outcome of criminal behavior – it is possible to argue that a substantial por-
tion of the number of homicides during these two decades were caused by criminal
violence. Many of these homicides were related to armed robberies, and there are
indications that a wave of kidnappings that often ended in the murder of the victim
were conducted by former police agents disbanded from their organizations in the
early 1980s (Kenny and Serrano, 2012b). However, the modus operandi of criminal
violence in the 1980s and 1990s was not characterized by the brutality and overtness
of organized criminal violence that would be seen a decade later (see the large spike
of homicides in Figure 5.6 towards the end of the time series). In any case, the in-
crease of violence generated a substantial demand for public security that motivated
government authorities to enforce the law.

12
Homicide data are from mortality statistics issued by the Mexican census bureau and the Min-
istry of Health. Data from 1938 to 1975 are from several issues of the Anuario estadı́stico de los
Estados Unidos Mexicanos (Annual Yearbook of Mexican Statistics) (Instituto Nacional de Es-
tadı́stica y Geografı́a, 2013a). Data from 1975 to 1989 are from Secretarı́a de Salud (2013). Data
from 1990 to 2011 are from Instituto Nacional de Estadı́stica y Geografı́a (2013b). The methodology
for collecting information on homicides from 1939 to 1975 is not clear from the official reports issued
by the government. Therefore their validity might be questionable. However, these are the only
data available for the period. Fortunately, the data and data collection since 1975 are reliable as
they conforms with the standards of the International Classification of Diseases (ICD) issued by
the World Health Organization (World Health Organization, 2010a). A recollection of this data is
available at http://www.mexicomaxico.org/Voto/Homicidios100M.htm
13
The Instituto Nacional de Estadı́stica y Geografı́a (2008) provides the following clasification
of homicides: homicidio calificado (first degree murder) refers to any murder that is willful and
premeditated; homicidio culposo (second degree murder or involuntary manslaughter) is a murder
that is not willful or premeditated; homicidio por riña (third degree murder) is an intentional killing
that involved no prior intent or premeditation, also known as a “crime of passion;” homicidio por
razones de piedad refers to a mercy killing, such as euthanasia.

265
 30,000

25,000

Total homicides
20,000

15,000

10,000

5,000
1935
1940
1945
1950
1955
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
Total homicides

Figure 5.6. Annual number of homicides in Mexico 1938-2011

The obscure shadow of corruption and links between PRI members and criminal
organizations provided enough material for new political actors to launch accusa-
tions against the PRI. Electoral competition motivates new politicians to break with
the corrupt practices of the past and to differentiate themselves from old politicians
deeply involved in corruption and criminal activities. New politicians advance with
an anti-corruption discourse that appeals to the social interest of having honest gov-
ernment authorities. This does not mean that new politicians will be absolutely clean
and free from corruption or relations with criminal organizations, but democratiza-
tion gives them incentives to publicly denounce competitors engaged in these kinds
of practices, as they will benefit from the electorate’s disapproval and sanction of the
corrupt old politicians. Public denunciation can also put pressure on judicial institu-
tions to arrest corrupt politicians, thus clearing the way for their rivals. In this way,

266
democratization reconfigured the incentives for a new generation of political actors
to enforce the law to signal their distinctiveness from previous corrupt politicians.

5.5.5
The
Collapse
of
Order

By the beginning of the 2000s, three different processes that evolved during the
1980s and 1990s finally converged to shape a new political scenario. First, Mexi-
can drug trafficking organizations underwent a process of growth and consolidation
thanks to the surge in drug consumption in the U.S. and the subsequent decline of the
Colombian cartels. Second, the highly repressive security apparatus of the Mexican
government was largely dismantled after the domestic and international pressures of
the Cold War had waned. In consequence, the state lost control of the connections
between the secret police and the criminal underworld, thus inhibiting its capabil-
ity to control criminal behavior and to instrument political discipline. The third and
most important factor is that democratization subverted the hegemonic dominance of
the PRI, thus overturning the political structures that had facilitated the imposition
of political control over criminal organizations and generating a system of incentives
favoring law enforcement. The convergence of these three factors generated a pre-
carious equilibrium in which government authorities coexisted with strong criminal
organizations already showing signs of animosity, and the state did not have the
mechanisms for monitoring and imposing political discipline on them nor the ability
to effectively repress them.
This precarious balance broke down when the government intensified its effort
to enforce the law. The last high level protector of drug organizations was General
Jesús Gutiérrez Rebollo, ironically the main commander of the National Institute to
Combat Drugs (Instituto Nacional para el Combate a las Drogas, INCD), who was
arrested in 1997. Gutiérrez Rebollo was in charge of leading the national counter-
narcotic strategy and he had the reputation of being tough on drug organizations,

267
but the investigations revealed that he was tough only on the rivals of his boss
Amado Carrillo Fuentes, El Señor de los Cielos, leader of the Juárez Cartel. After
the Gutiérrez Rebollo scandal broke, the U.S. reinforced its pressure on Mexican
authorities to capture Carrillo Fuentes. However, El Señor de los Cielos never went
to prison, as he died in hospital during plastic surgery to change his appearance in
July, 1997. The arrest of General Gutiérrez Rebollo and the death of Carrillo Fuentes
were serious blows for the Juárez Cartel, which suddenly lost both its government
protection and its infamous leader. Although Vicente Carrillo Fuentes, the brother
of El Señor de los Cielos, took over the organization, the weakening of the Juárez
Cartel increased the animosity the Sinaloa Cartel and its ambition to move in on
Juárez territory.
After the PRI lost the presidency in 2000, newly elected president Vicente Fox
ordered high-profile apprehensions of certain prominent drug leaders. However, these
arrests destabilized the delicately balanced geographic distribution of criminal orga-
nizations in Mexico. The first blow was against the Juárez Cartel with the arrest of
Mario Villanueva Madrid, the governor of the state of Quintana Roo, in May 2001.
According to Ravelo (2007b), Villanueva Madrid later confessed that his arrest had
been prepared by the last PRI president, Ernesto Zedillo, because of his insubordina-
tion towards the party and it was left for Fox to finally effect the detention. “Fox was
determined to bury that party and Mario Villanueva, with his obscure background,
was a good instrument for further discrediting the PRI and the priı́stas” (Ravelo,
2007b, 198). In 2001, Fox also increased law enforcement efforts in Tijuana, which
was already the scenario of confrontations between the Arellano Félix brothers, lead-
ers of the Tijuana Cartel, and Ismael “El Mayo” Zambada, one of the leaders of the
Sinaloa Cartel. According to Blancornelas (2002), early eruptions of violence in Ti-
juana took place after a group of sicarios (hitmen) sent by the Arellano Félix brothers
failed to assassinate “El Mayo” Zambada for refusing to pay a toll for transporting

268
drugs through their territory. In February 2002, Ramón Arellano Félix, the chief
enforcer of the Tijuana Cartel, was killed in a shootout with authorities in Mazat-
lan, Sinaloa. One month later, the Army arrested Benjamı́n Arellano Félix who was
hiding in the state of Puebla, one of the remaining strongholds of the PRI at the
subnational level. The third brother, Eduardo Arellano Félix, remained the leader
of the weakened Tijuana Cartel. Another important blow against Mexican DTOs
was delivered though the apprehension of Osiel Cárdenas Guillén, leader of the Gulf
Cartel, who dominated drug trafficking operations in the northeast of the country
until his arrest in March, 2003. One of the key strategic moves made by Cárdenas
Guillén was to recruit a group of elite forces from the Grupo Aeromóvil de Fuerzas
Especiales (GAFES) of the Mexican Army (Osorno, 2012; Ravelo, 2009). Initially
deployed to hunt down criminals, this group of elite military personnel switched sides
and became the feared enforcers of the Gulf Cartel known as Los Zetas. The arrest
of Cárdenas Guillén in 2003 caused a rupture between Los Zetas and the rest of the
Gulf Cartel.
Initial attempts by President Fox to enforce the law against drug trafficking orga-
nizations were obscured by a major scandal. On January 19, 2001, Joaquı́n Guzmán
Loera, known as El Chapo, escaped from a high security prison where he had been
held since 1993 on a twenty-year sentence. The official version of his escape was that
El Chapo received assistance from two prison guards who helped him to escape hid-
den in a laundry basket (Osorno, 2009, 193–198). Further investigations led to the
interrogation of 71 officers who had collaborated in the escape (Procuradurı́a General
de la República, 2001a,b). However, the veracity of this account is questionable. A
different version claims that the escape occurred during a cursory visit to the prison
by Jorge Tello Peón, a high level security officer, to investigate allegations of cor-
rupt security guards. During the visit, Tello Peón ordered El Chapo and two other
inmates to be moved to different cells which generated a disordered mobilization of

269
various police officers inside the penitentiary. Later that night, El Chapo walked out
of the maximum security prison wearing a police uniform (Hernández, 2011, 290-293,
321-322). Following the scandal, no major authorities were arrested. Eight years
later, El Chapo appeared in the Forbes list of billionaires (Bogan, 2009).
After the escape, El Chapo consolidated his leadership in the Sinaloa Cartel.
Following the arrest or killing of several leaders of the other drug trafficking organi-
zations, El Chapo took note of the weakened position of his main competitors and
began a large scale operation to dominate drug trafficking routes in Mexico. The
Sinaloa Cartel had already had run-ins with the Tijuana Cartel due to the rivalries
between “El Mayo” Zambada and Ramón Arellano Félix. The Sinaloa Cartel inten-
sified their trespasses into Tijuana Cartel territory. El Chapo also sent one of his
chief enforcers, Arturo Beltrán Leyva, and a group of sicarios to conduct some initial
incursions into the territory of the Gulf Cartel in the states of Tamaulipas and Nuevo
León. The armed branch of the Gulf Cartel, Los Zetas, responded in kind and began
conducting operations in the state of Michoacán, south of Sinaloa. Later, El Chapo
also aimed his guns against the Juárez Cartel. The first hot-spots of violence between
criminal organizations emerged in mid-2005 in the northeast part of the country and
in the state of Michoacán (refer to the second map in Figure 4.9 in Chapter 4). In
June of the same year, president Fox launched Operation “Safe Mexico” (Operativo
México Seguro) to fight criminal organizations (Presidencia de la República, 2005).
During the first half of 2006, media attention was distracted from the early in-
dications of violence emerging between criminal organizations and focused on the
presidential election campaign. Competition was tight between the PRD candidate,
Andrés Manuel López Obrador, who was mayor of Mexico City and the PAN can-
didate, Felipe Calderón. The PRI candidate remained in third position. The failed
legal attempt of president Fox to topple López Obrador from his position as mayor
of Mexico City set the tone for the electoral race. The media and the PRD made

270
repeated accusations against the overt support of Fox in favor of Calderón that re-
sembled the old PRIı́sta selection by dedazo (Espinosa, 2011; Martı́nez et al., 2007;
Notimex, 2006; Zárraga, 2006). López Obrador campaigned by advocating for redis-
tributive policies while Calderón campaigned by presenting himself as “the president
of employment”; The debate focused mainly on economic issues and security was
barely mentioned in the agenda. The political atmosphere became more polarized
after López Obrador raised the tone of this economic statements and conservative
business sectors moved in to conduct negative campaigning against him.
On July 2, 2006, Felipe Calderón won the election by an unprecedentedly close
margin of victory of 0.56 percent. Due to the narrow margin of victory López Obrador
launched fierce allegations of election rigging. The ghost of electoral fraud charac-
teristic of the PRI era hovered over the 2006 election. The doubt about the election
jeopardized the achievements of the thirty-year struggle for democratization. This
election represented the most profound political crisis in the short democratic history
of Mexico. Followed by thousands of supporters, López Obrador immediately went
to the central square of Mexico City, where he encouraged the crowd to defend his
victory and called for a recount of the entire result under the slogan “voto por voto,
casilla por casilla” (vote by vote, ballot box by ballot box) to clear away the doubts
that clouded the election result. López Obrador launched a campaign of civil resis-
tance and thousands of protesters rallied on the streets in several parts of the country
to demonstrate against the electoral result. As a sustained effort of protest, López
Obrador and his supporters closed the main avenue of Mexico City for a three-month
period with a sit-in (British Broadcasting Corporation, 2006; McKenley, 2006).
The crisis reached its peak on December 1, 2006, when the PRD took over the
stage of the Chamber of Deputies to prevent Calderón from taking the oath of pres-
idential office. Outside the legislative building, about a thousand security forces,
primarily from the Army and the secret service, placed barricades on the streets to

271
prevent protesters from entering Congress (Osorno, 2009, 301). Secret service offers
had to punch through the crowd to rush Calderón onto the stage so he could take
the oath. The entire “ceremony” lasted 10 minutes. There was no inaugurational
speech, no applause, no celebration; only condemnation, catcalls, and clamor. The
electoral crisis of 2006 is broadly identified by political analysts as severely damaging
the legitimacy of President Calderón from the beginning of his administration (e.g.
Álvarez Béjar, 2007; Castañeda, 2012; Castañeda and Aguilar, 2010; Pacheco, 2006).
A few days after taking office, on December 11, Calderón declared a full-scale
offensive against organized crime and launched Operativo Conjunto Michoacán (Joint
Operation Michoacan) consisting of the deployment of about 7,000 soldiers to conduct
counter-narcotic activities in his native state of Michoacán. The next day, Calderón
launched another operation for the states of Nuevo León and Tamaulipas on the
northeast of the country. As indicated by several journalists, scholars and political
analysis (e.g. Castañeda and Aguilar, 2010; Guerrero, 2010a,b, 2011b; Kenny and
Serrano, 2012b; Osorno, 2009, 2012), during the campaign, Calderón had dropped
no hint that security would be his top priority; his decision was forged in the heat
of fierce allegations of election fraud and showed a narrow understanding of the
precarious equilibrium among drug trafficking organizations. In the several interviews
conducted in Mexico as part of this research, testimonies gathered from members
of security forces, political actors, scholars, activists and political analysis broadly
agreed that Calderón launched the war on drugs without a careful assessment of
the capabilities of criminal organizations to retaliate; nor the judicial, military and
intelligence capabilities of the Mexican security forces to meet the challenge; nor
an attentive review of successful counter-narcotic strategies in other countries; and
most importantly, there are no indications of a rigorous assessment of the different
consequences that a full-fledged military campaign would have in terms of violence
and insecurity for the population. Instead of a careful analysis based on accurate

272
information, it seems that the decisions was taken without necessary planning and
in the midst of a political crisis.
Table 5.1 reports the total number of military operations conducted by the Mexi-
can Army between December 2006 and December 2010 on the basis of data provided
by (Secretarı́a de la Defensa Nacional, 2012). The Table indicates that Calderón
deployed an unprecedented number of interdiction and harassment operations (IHO)
to fight criminal organizations as well as several eradication operations (EO) to de-
stroy plantations of illicit drugs. During this period, the Army conducted a total
of 173 IHO and 91 EO operations, many of which were implemented during sev-
eral months over various states. Calderón had declared a generalized war against
all criminal organizations in the country (Presidencia De La República, 2010). This
military mobilization had no precedent in modern Mexican history. The Army went
out the barracks to conduct policing activities all over the country to fight an elusive,
heavily armed, financially well-supplied and largely unknown enemy. The Mexican
government had entered into a non-conventional war.

In addition, Figure 5.7 presents the annual budget for the different security agen-
cies and programs used to fight organized crime in Mexico between 2000 and 2010.
As the figure shows, the Army and the Ministry of Public Security (Secretarı́a de
Seguridad Pública, SSP ) which comprises the Federal Police experienced an unprece-
dented budget increase in 2007. In the period prior to the Mexican war on drugs
between 2000 and 2006, the Army received an average annual budget increase of 4.2
percent and the SSP had an average increase of 8.8 percent in its annual budget. In
contrast, after the onset of the war on drugs, the average budget increase between
2007 and 2010 for the Army was 14.3 percent and the SSP had a substantial increase
of 39.3 percent in its average annual budget. The figure also shows that the Navy
had an important increase in its budget, although it is less marked than the one of

273
 TABLE 5.1

MILITARY COUNTER-NARCOTIC OPERATIONS CONDUCTED BY

THE MEXICAN ARMY BETWEEN DECEMBER 2006 AND
DECEMBER 2010
Type of military operation
Starting year Interdiction and Harassment Eradication
2006 2 0
2007 7 21
2008 26 32
2009 72 24
2010 66 14
Total 173 91

the Army and the SSP. Between 2000 and 2006, the average rate of annual budget
increase for the Navy was 2.5 percent. In contrast, between 2007 and 2010, the aver-
age annual budget increase of the Navy by a rate of 15.3 percent. This indicate that
the war on drugs launched by president Calderón in December 2006, largely relied in
federal security forces, specially on the Army and the Federal Police. In contrast, a
program to provide financial aid for security forces at state level known as Fondo de
Aportaciones para la Seguridad Pública (FASP) reveals that the role of local security
forces was marginal in the war on drugs. The time series of this program presented in
the figure shows that FASP received the lowest allocation of financial resources. The
average annual increase of FASP between 200 and 2006 was 7.4 percent and between
2007 and 2010 it had an annual increase rate of 8.8 poercent.
Perhaps, one of the clearest indicatives of the punitive strategy launched by
Calderón is the lack of allocation of financial resources to the Office of the Attorney
General (Procuradurı́a General de la República, PGR), responsible for the investiga-
tion and prosecution of federal crimes. As indicated in Figure 5.7, between 2000 and

274
nizations. Similarly, it is hard to sustain the claim raised by some critics (Castañeda
and Aguilar, 2010; Osorno, 2009) arguing that the paramount reason for launching
the war on drugs was the legitimacy crisis at the outset of the administration. In
any case, due to the magnitude of such decision, it is surprising that there are no
reliable and robust documents, reports, analysis, evidence or testimonials justifying
the characteristics, scale and scope for the security policies implemented by Calderón.
However, there are good theoretical and empirical reasons to claim that the elec-
toral crisis of 2006 played a role – probably an important one – in motivating the
massive deployment of the Army to fight crime. As discussed in Section 2.3.2 of Chap-
ter 2, Goldstein (1978) provides a theory for understanding the decision of political
leaders to engage in coercive behavior during political crisis challenges. According
to his argument, increasing levels of political strain and social dissent are perceived
by authorities as threats to their legitimacy, thus increasing their disposition to re-
press. The adoption of repressive policies is facilitated by the presence of scapegoats
as suitable target groups who can be readily repressed due to lack of opposition from
political elites. In addition, the well-known “rally-around-the-flag” effect proposed
by Muller (1970) argues that politicians usually reap political benefits from deploying
aggressive policies and displaying an image of strong leadership and resolve, which
helps to boost their approval ratings. The theoretical insights proposed by Goldstein
and Muller can provide an understanding of why Calderón reacted with an excep-
tional display of force in the midst of an exceptional legitimacy crisis. Certainly drug
trafficking organizations showed early signs of violence and public insecurity was be-
coming an important concern for the population. In this context, they could readily
be used as scapegoats, as fighting crime would be approved by the public and there
would be no opposition from other political parties. Moreover, public security was
not an issue that was polarized along party lines. Combating criminal organizations
would thus provide a public good highly rewarded by the population and would have

276
the advantage of not further polarizing the already critical political scenario in the
aftermath of the electoral crisis. Launching a crusade against crime “like never be-
fore” would also improve his approval rankings by enhancing his image as a clean
political actor. Calderón immediately saw the political benefits of deploying the mil-
itary in Operation Michoacán as the public opinion polls revealed that 80 percent of
respondents supported the move (Kenny and Serrano, 2012b, 73). Calderón repeat-
edly portrayed himself as a committed leader taking a moral stand in a matter that
previous administrations had not. In addition, by deploying the Army, Calderón
would be using one of the most respected and trusted political institutions of the
country, thus reaping some benefits from the good image of the Army.14 In any case,
the deployment of the Army was also indicative of the failure of considering other
strategies to control criminal organizations. As elaborated earlier, the alternative of
imposing order and discipline on criminals through political incentives was no longer
feasible in a democratic context.
The unprecedented military mobilization ordered by Calderón had a massive dis-
turbing effect on the already precarious equilibrium among criminal groups. The
recurrent attacks from the state security forces against criminal organizations trig-
gered a wave of violence against government forces. Groups of heavily armed men
attacked municipal and state police officers, as well as Federal Police and Army
troops. Mayors were also targeted (Rios, 2012b) and sicarios also killed the leading
candidate for governor of Tamaulipas (Camarena, 2010).
The most dramatic eruption of violence took place in battles between rival crim-
inal groups. The trends of drug-related violence described in Chapter 4 snow the
massive escalation and diffusion of violence between drug trafficking organizations.
In congruence with the theoretical expectations discussed in Chapter 2, the disrupt-

14
Public opinion studies consistently report that the Army is the second most trusted institution
in Mexico, just after the Church (Mitofsky, 2012; Moreno, 2010; Parametrı́a, 2012).

277
ing effect of counter-narcotic operations weakened some criminal organizations and
indirectly empowered their rivals, thus motivating the invasion of their territories.
The full-fledged military campaign launched by Calderón against all criminal groups
generated an enormous wave of violent competition among criminal organizations
fighting to control strategic territories. During this period, the number of criminal
organizations also increased. Guerrero (2011b) states that before Calderón launched
the crusade against drugs there were only six main cartels and about 30 minor crim-
inal groups in the country; by the end of 2010, the number of cartels had risen to 12
and the number of local organisations proliferated to 114. Confrontations between
criminal groups not only became more frequent; the brutality of their tactics also
increased. As stated by Duran-Martinez (2012), the old modus operandi of drug-
related killings consisting of discrete executions and silent disposal of the bodies was
replaced by an overt flaunting of brutality relying on torture, mutilation, decapitation
and public display of human remains.
The arrests and killings of prominent drug lords destabilized the structure of
criminal groups, thus generating internal struggles. Lower ranks tried to shoot their
way up as those above them also employed violence to impose discipline. Victory by
either side often triggered bloodshed within the organization. If no group managed to
impose on its rivals, the situation usually led to the split of the organization. Those
splits generated overt confrontations in which some criminals forged alliances with
their previous rivals to fight against their former partners. Following the old proverb
“the enemy of my enemy is my friend,” the map of criminal organizations became
highly unstable with the emergence and fracture of criminal groups that expanded
and contracted their territories while creating and breaking alliances.
The escalation of violence also generated a terrible social cost. Armed clashes oc-
curred in public streets during peak hours, in shopping centers, in front of schools and
universities, in bars and music festivals. Innocent bystanders were killed in the cross-

278
fire between security forces and criminals as well as in confrontations among criminal
groups. But civilians were not only killed randomly. The wave of violence came
accompanied with a wave of kidnappings as criminal organizations took advantage of
the general mayhem to collect ransom. Kidnappings were initially targeted against
prominent businessmen (Duarte, 2008) but later extended to the middle class. Even
illegal immigrants from Central America trying to make their way up to the U.S.
became targets of mass kidnapping and extortion. The large sums of ransom income
were also complemented by money from day-to-day racketeering affecting up to 36
percent of the economy (American Chamber, 2013). The escalation of drug violence
showed other features characteristic of large humanitarian crises such as the forced
displacement of about 160 thousand people from their communities due to violent
confrontations between criminal groups (CNN-Editor, 2012; Notimex, 2012). Jour-
nalists conducting investigations into drug-related violence were severely targeted by
criminal organizations, putting Mexico above Afghanistan at the top of the list of
the world’s most dangerous countries for journalists (Article 19, 2013). However,
civilians not only suffered by the hand of criminal organizations. The deployment
of the Army for conducting policing activities also opened the door for systematic
violations of human rights that remain largely in impunity (Daly, Heinle and Shirk,
2012) and leading to accusations of negligence against government authorities for not
taking action on this issue (Human Rights Watch, 2011). An obscure consequence of
the escalation of violence against civilians is the dramatic number of disappearances,
which finally attracted the attention of the United Nations Human Rights (2011)
and Human Rights Watch (2013), which exercised pressure on the Mexican govern-
ment until government authorities finally acknowledged the disappearance of up to
25,000 people during the Calderón administration (Torres, 2013). It is still unknown
what proportion of these disappearances were conducted by members of criminal or-
ganizations or by government security forces. What is known is that this figure of

279
disappearances is comparable to the number of forced disappearances undertaken by
military dictatorships in South America in the 1970s, yet there are no precedents of
such horrific figures in contemporary democratic regimes in the region.
The historical process tracing analysis conducted in this chapter shows that the
long decades of political control imposed by an hegemonic party capable of maintain-
ing order and peace among criminal organizations are gone. The failure to impose
order through political mechanisms and the deleterious consequences of generalized,
punitive operations against drug trafficking organizations led to a Hobbesian state of
war of all-against-all.

280
 CHAPTER
6

UNDERSTANDING
DRUG
VIOLENCE
IN
MEXICO

6.1
Introduction

The previous chapter analyzed the historical process that favored the emergence,
consolidation, erosion and collapse of order and specifies the circumstances that led
to the onset of the Mexican war on drugs. This chapter provides the main quanti-
tative assessment for the expectations derived from the theoretical model about the
escalation and geographic concentration of violence among rival criminal organiza-
tions. The central dependent variable of this section refers to the levels of violent
competition among criminal organizations.
The first section of the chapter surveys extant explanations of organized criminal
violence mostly emphasizing the relevance of macro-structural factors. The statistical
assessment evaluates the analytical leverage offered by these structural determinants.
The results reveal that with the exception of territorial variables and economic factors,
other macro-structural explanations offer limited explanatory power or are wrong. In

281
general, the structural model of conflict provides limited insights to understand the
wide and rapid variation of violent competition among criminals at the micro level.
The second section evaluates the main hypotheses derived from the theoretical
framework. Based on the conflict interactions between the state and criminal organi-
zations specified by the theory, the empirical assessment incorporates an “interactive
approach” to reflect these relationships. To do so, the statistical analysis builds on
the set of structural explanations and includes additional variables of law enforcement
and criminal retaliation in the model of violent competition among criminal groups.
To overcome the challenges of endogeneity generated by distinct, yet overlapping
types of violence, the identification strategy relies on an Instrumental Variables (IV).
This approach allows to generate reliable estimates of the effect of law enforcement
on violence among criminals that overcome the problem of reciprocal causation. The
research design considers measures of democratization and political strain as instru-
mental variables capable of generating a plausibly exogenous variation of the levels of
law enforcement, which then generates an effect on the levels of violent competition
among criminals. This quasi-experimental research design not only represents a plau-
sible identification strategy, but also conforms with the process of conflict specified
by the theoretical framework, thus favoring the alignment between the ontology of
the theory and the methodology used for testing it.
Results indicate that the measures of democratization and political strain increase
the levels of law enforcement, which then have a profound disrupting effect and trigger
waves of violent competition among rival criminal groups. These results are consistent
across different model specifications evaluating the effect of violent and non-violent
tactics on the levels of violence among criminals. The statistical analysis also reveals
that criminal violence tends to cluster around strategic territories favorable for the
reception, production and international distribution of illicit drugs.

282
6.2
Structural
Factors

This first section analyzes the explanatory power of structural factors for under-
standing the escalation of violent competition among criminal organizations. As dis-
cussed in Section 2.2, the literature on political violence, criminology and economics
has identified several structural variables as determinants of violence, including de-
gree of economic development, levels of corruption, characteristics of drug markets,
and social structures. In addition, the theoretical model presented in Chapter 2
emphasizes the relevance of drug-valuable territories as key structural factors for
understanding the dynamics of violence among DTOs. The theoretical explanation
also indicates the relevance of criminals’ ability to inflict military damage on their
rivals and their ability of recovering from it. These factors are also incorporated in a
baseline model to analyze the effect of structural factors on the dynamics of violent
competition among criminal organizations.

6.2.1
Hypotheses
and
Operationalization

Perhaps one of the most important factors associated with criminal behavior and
political violence is poverty. There is a broad consensus between rationalist and
sociological theories of crime, as well as among conflict scholars that low levels of
economic development are associated with higher benefits for engaging in crime or
political violence (Becker, 1968; Collier, 2000; Collier and Hoeffler, 2004; Fajnzyl-
ber, Lederman and Loayza, 2000b; Fearon and Laitin, 2003; Hirschi, 1969; Merton,
1938, 1957; Weinstein, 2007). There are competing explanations about how and why
poverty is associated with criminal and violent behavior, but the empirical relation-
ship is remarkably robust. This can be stated in the following hypotheses:

(H9 ) Low levels of economic development are positively associated with violent
competition between criminal organizations.

283
 The operationalization of poverty is based on two measures. The variable poverty
measures the level of economic development at the municipal level on a yearly basis
as reported by Consejo Nacional de Evaluación de la Polı́tica de Desarrollo Social
(2012). This measure is a multidimensional index of thirteen indicators including
income, education, health, social security, living environment, service provision, food
access, and other multidimensional measures of poverty.1 This poverty indicator
is the most robust and fine-grained variable used to measure the level of economic
development in Mexico. High values of the variable indicate high levels of poverty
at the municipal level. The second variable used for measuring poverty is the yearly
Gross Domestic Product (GDP) at state level measured in Mexican pesos as reported
by Instituto Nacional de Estadı́stica y Geografı́a (2011c). In order to improve the fit
of the model, the statistical analysis uses the natural logarithm of state GDP. High
values of GDP indicate higher levels of economic development at the state level.
Corruption is another important factor associated with violence. However, as
discussed in the literature review, the relationship between corruption and violence
between criminal organizations is not clear. Some argue that high levels of corrup-
tion are associated with low levels of violence among DTOs because it means the
organisations have a peace agreement with the state. Others argue that corruption
leads to criminal violence because it allows DTOs to operate with impunity. These
two arguments can be stated the following terms:

(H10.1 ) High levels of corruption are positively associated with violent compe-
tition between criminal organizations.
(H10.2 ) High levels of corruption are negatively associated with violent compe-
tition between criminal organizations.

Since corruption takes place in secrecy, it is difficult to observe it in a systematic

way. In order to measure the levels of corruption, the empirical strategy relies on

1
For a detailed description of the indicators and the methodology of aggregation, see Consejo
Nacional de Evaluación de la Polı́tica de Desarrollo Social (2010).

284
the National Index of Corruption and Good Governance provided by Transparencia
Mexicana (2012), the Mexican chapter of Transparency International. This index
of corruption is based on a large national survey, representative at the state level,
asking respondents about their experience of corruption victimizationin a wide variety
of public services.2 The specific variable of corruption used in this research is the
estimated percentage of the population that reported being asked for a bribe by police
in order to prevent being arrested.3 Higher values of this variable are associated with
high levels of corruption.
Researchers on the economics of crime broadly agree that criminal organizations
are mostly driven by the enormous economic benefits of illicit markets (Gambetta,
1993; Reuter, 1989; Schelling, 1971). The profits are so large that criminal organi-
zations are willing to use violence to protect or expand their access to those rents.
In particular, Reuter (1989) argues that the threat of violence is a key element for
enforcing agreements in illegal markets because, by definition, these economic sectors
are not protected by legal institutions of conflict resolution. This relationship can be
expressed in the following way:

(H11 ) The high value of drug markets is positively associated with violent com-
petition between criminal organizations.

Although the intuition behind the economic motivation of using violence is straight-
forward, it is not easy to measure the size or profitability of illegal drug markets. The
two key ingredients for determining the size of a market are the quantity of goods
being traded and their price. Some efforts have been made to measure the size of the
illegal drug market in the U.S. to collect information about the value of this market.

2
This survey was conducted in 2001, 2003, 2005, 2007 and 2010 in each state of Mexico. I
interpolated each of the measures to give year-by-year values for the complete 2000–2010 time range
analyzed in this research.
3
For a robustness check, I used other measures of police corruption related to avoiding getting
a ticket for transit violations or preventing a car from being towed. These alternative questions led
to similar results as that used in the rest of the chapter.

285
Unfortunately, there are no reliable measures of the amount of drugs being smuggled
into the U.S. from Mexico. Official estimates suggest that drug revenues of Mexi-
can DTOs range from $18 billion to $39 billion dollars (U.S. Department of Justice,
2008). However, as indicated by Kilmer et al. (2010) and Kilmer and Liccardo Pac-
ula (2009), these figures should not be taken seriously because they do not survive
basic methodological scrutiny or validation checks. Perhaps, the most transparent
and accurate estimate of the profits made by Mexican DTOs ranges from $1.5 billion
to $2 billion dollars (Kilmer et al., 2010). However, despite the methodological rigor
used in constructing this estimate, the calculation is based on too many assumptions
to be reliable. In addition, it is not possible to use this estimate in the context of
this research because the estimation is not available for the time span analyzed in
this research.
In order to assess the value of drug markets in a more objective and reliable way,
this research uses two variables. One measure is based on the price of drugs in the
U.S. market collected by the U.S. government. The Office of National Drug Control
Policy has been measuring the retail price and purity of various drugs since the 1980s.
In general terms, the methodology consists of using covert agents to buy drugs such
as marijuana, cocaine, crack cocaine, heroin, and methamphetamine in several U.S.
cities. After measuring the quality of each drug in specialized laboratories, the data
is used to generate an estimate of the average price per pure gram (Office of National
Drug Control Policy, 2004). This research uses the most up-to-date time series of
the price of cocaine reported by the Drug Enforcement Administration (2012). The
variable cocaine provides information about the price of a gram of pure cocaine in
U.S. dollars between 2000 and 2010.4

4
The measure of cocaine is preferred over other drugs because according Kilmer et al. (2010) the
largest share of revenues of Mexican DTOs come from cocaine, and not from marijuana. According
to these authors, about 80 percent of the cocaine consumed in the U.S. enters the country from
Mexico, which represents an estimated export revenue of about $3.4 billion dollars for Mexican
DTOs.

286
 A second measure of the value of drug markets is focused on the levels of local drug
consumption. Assessing drug consumption in Mexico is crucial because President
Calderón broadly justified launching the war on drugs indicating that there were
worrying levels of drug prevalence in Mexico. This official discourse was reflected
in a national awareness campaign using the slogan “To prevent drugs from reaching
your children” (“Para que las drogas no lleguen a tus hijos”) (Secretarı́a de Salud
y Secretarı́a de Desarrollo Social, 2009). The core message of the campaign is that
Mexico has such a severe problem of drug consumption that it causes of the wave
of violence between criminal organizations and erodes the social fabric. However,
the available data does not support this claim. According to the latest National
Survey on Addictions conducted by the Health Ministry in 2011, only 1.2 percent
of the population reported using marijuana at least once in the previous year, 0.5
percent reported using cocaine, 0.1 percent crack and 0.2 percent methamphetamine
(Secretarı́a de Salud, 2012b). These estimates are so small that they fall completely
within the overall statistical error of the survey, calculated at 3.27 percent. Moreover,
the methodological report of the National Survey on Addictions warns that “. . . this
study is not designed for estimating low prevalence, which can occur when measuring
very rare events.” Farther on, the document recommends that readers should be
cautious about estimates smaller than 2 percent, because they are not distinguishable
from the design error of the survey. Therefore, based on the government’s own data,
it is not possible to conclude that Mexico has a significant domestic problem of drug
consumption.5 Besides the 2011 survey, there are other two other National Surveys
on Addictions conducted in 2002 and 2008. With data from only three years, there
are too few data points and their estimates are so small that it is not possible to use

5
Despite the official discourse justifying the war on drugs by claiming a substantial increase in
drug consumption, the evidence does not support the claim. In fact, the media often criticized the
government for delaying and even hiding the results of the National Survey on Addictions 2011
(Service, 2012).

287
these measures to assess levels of drug consumption in Mexico in a systematic and
reliable way.
This research thus relies on an alternative measure based on the records of hos-
pitalizations due to consumption of illegal drugs. The intuition behind this measure
assumes that higher levels of drug consumption should generate more cases of drug in-
toxication. Of course we cannot observe individual drug consumption systematically,
but we can measure the number of hospitalization cases caused by drug poisoning.
Using this proxy relies on the plausible assumption that observed cases of intoxication
and unobserved drug consumption are equally distributed. This data is reported by
the National Health System (Secretarı́a de Salud, 2012a) and has been used by other
authors as the most reliable proxy of drug consumption in Mexico (Madrazo and
Guerrero, 2012; Rios, 2012a). The variable drug markets measures the total number
of hospital discharges from the morbidity statistics in which the patient was diag-
nosed with intoxication by narcotics.6 The data is collected by all state clinics and
hospitals across the country and reported monthly at the municipal level. The diag-
nostic is based on the International Classification of Diseases (ICD-10) of the World
Health Organization (2010a) and uses the F10–F19 codes for mental and behavioral
disorders due to the use of psychoactive substance including opioids, cannabinoids,
cocaine, hallucinogens or volatile solvents (World Health Organization, 2010b). Fig-
ure 6.1 shows the total number of hospital discharges by type of drug between 2000
and 2010 at the national level. The measure of drug markets used in this research
aggregates the number of hospital discharges caused by all drugs at the municipal
level on a monthly basis. This figure only presents the disaggregated data by drug
type for illustrative purposes. The trends indicate that the use of drugs in Mexico is

6
Hospitalization discharges occur when the patient leaves the hospital dead or alive. If the patient
is still alive at the time of leaving the facility, the event is recorded under the morbidity statistics.
If the patient is dead when leaving the hospital, the case is recorded under the mortality statistics.
The measure used in this research uses only morbidity statistics.

288
remarkably stable. Ironically, as the wave of violence increased in 2007, the number
of hospitalization cases due to cocaine use declined. There is a stable trend at low
levels in intoxication caused by opium, cannabis, hallucinogens and solvents. The
only time series showing a marked spike in 2010 is the use of multiple or other drugs.
Appendix A.8 shows that this is spike is concentrated in Mexico City, but does not
constitute part of a generalized trend of increased drug consumption in the country.
Based on the most reliable indicators of local drug markets, there is no evidence of a
substantial increase of drug consumption in Mexico as the official discourse claims.
500 1000 1500 2000 2500
Number of drug morbility cases
0

2000 2002 2004 2006 2008 2010

Year

Cocaine Cannabis
Opium Hallucinogens
Dissolvents Multiple/other

Figure 6.1. Drug markets in Mexico: hospital discharges diagnosed with

intoxication by narcotics

The erosion of the social fabric has been proposed as another factor leading to
criminal behavior. According to Putnam (1993), social capital is defined as the system

289
of trust, norms and networks that improve the efficiency of society by facilitating
coordinated actions. As discussed in the literature review, there is a broad consensus
that low levels of social capital are associated with higher rates of crime. The following
hypothesis formulates this relationship:

(H12 ) The erosion of social structures is positively associated with violent com-
petition between criminal organizations.

Despite the appealing intuition behind social capital, it is difficult to measure

such a broad concept in a systematic and reliable way. Putnam himself uses dozens
of structural and behavioral indicators to capture his concept (Putnam, 1993, 2000).
For example, in order to assess the relationship between crime and social capital,
Paras (2007) uses public opinion data to measure the social capital based on questions
about individual trust, trust in government institutions, organizational membership,
changes in behavioral patters, and attitudes towards democracy. Unfortunately, in
Mexico there are no surveys measuring these concepts in a comparable and system-
atic way between 2000 and 2010 with data representative at the state level. Instead
of focusing on a broad concept such as social capital, the operationalization of this
hypothesis relies on two narrower aspects of social capital associated with changes in
the traditional family structure. Variable divorces measures the annual number of di-
vorces registered at the municipal level between 2000 and 2010 reported by Instituto
Nacional de Estadı́stica y Geografı́a (2011c). According to the conservative perspec-
tive proposed by Hirschi (1969), divorce constitutes a change in the traditional family
structure that might hinder the instilling of traditional norms of self-control in their
members, which may be conducive to engagement in criminal behavior. The second
variable used to operationalize this hypothesis is young motherhood. This variable
measures the proportion of females between the age of 12 and 19 years old who are
mothers of at least one child. The measure is constructed from census data at the mu-
nicipal level reported by Instituto Nacional de Estadı́stica y Geografı́a (2011c). The

290
link between the erosion of the traditional family structure and organized criminal
violence was also essential for Calderón’s justification for the war on drugs as reflected
in the National Security Strategy. One of the three core objectives of the strategy
is “to rebuild the social fabric eroded by the lack of opportunities for young people
and family, and social disintegration, as well as the loss of values.” (Presidencia de
la República, 2012).
The theoretical model discussed in Chapter 2 also indicates the relevance of the
strategic value of certain territories as key structural determinants for understanding
violence between criminal organizations. The central intuition of the contest success
model of territorial competition is that DTOs will devote more resources to defending
and capturing strategically valuable territories. Some areas are so valuable that
criminals would be willing to kill and die to controlling them. Section 2.4 presented
this hypothesis in the following terms:

(H8 ) Higher territorial value is positively associated with higher levels of violent
competition between criminal organizations.

The strategic importance of a territory is determined both by the objective char-

acteristics of the territory and its subjective value as assigned by each DTO. This
subjective appreciation might depend on the production function, organizational ca-
pability and area of specialization of each criminal group. Unfortunately, it is not
possible to measure the subjective value assigned by organized criminals to each spe-
cific territory. For that reason, the operationalization of this hypothesis is limited
to a number of objective characteristics that might influence the strategic value of
a territory. These measures focus on the geographic characteristics of some areas
that might be favorable for the production, reception and distribution of illicit drugs.
Variable drug production measures the production of illegal marijuana and poppy
crops on a four-level scale (none, low, medium and high) detected by the Mexican
Army at the municipal level (Secretarı́a de la Defensa Nacional, 2011). Variables

291
Gulf and Pacific identify municipalities located along the Gulf of Mexico and the
Pacific coasts. These areas are favorable for the reception of maritime and aerial
drug shipments from Central and South America. Variable Gulf takes the value of 1
for the strip of three adjacent municipalities located along the Gulf of Mexico and 0
otherwise. Variable Pacific is also a dichotomous measure for municipalities located
along the Pacific coastline. Variable North identifies territories favorable for inter-
national distribution of drugs. Municipalities located along the Northern border are
key points of entry to the U.S. drug market. For this reason, these areas are highly
valuable. This variable takes the value of 1 for the strip of three contiguous mu-
nicipalities located along the Mexico–U.S. border. Figure 6.2 maps these territories
favorable for the production, reception and distribution of illicit drugs.
Interviews conducted in the cities of Juarez, Chihuahua and Tijuana revealed the
need to include one additional territorial variable to incorporate the impact of the
9/11 terrorist attacks on border security. After the 9/11 attacks, tightened border
security measures implemented by the U.S. government made drug trafficking more
difficult. Reducing the flow of drugs may have increased the price of the drugs suc-
cessfully smuggled though the border, thus generating more incentives to fight for
those sources of revenue. Another mechanism suggests that the September 11 terror-
ist attacks particularly increased the difficulty of smuggling drugs by air and sea, thus
increasing the strategic importance of land transportation. This improved the rela-
tive leverage of Mexican DTOs vis-à-vis Colombian cartels, who were charged larger
fees for smuggling through Mexican territory. As a consequence, the value of the en-
tire Mexican territory increased because controlling the reception, transportation and
distribution points became more profitable for Mexican DTOs. Another mechanism
suggest that the slow-down in drug smuggling into the U.S. after the 9/11 attacks
turned several northern municipalities into storage regions. The accumulation of
drugs waiting to be smuggled increased the strategic importance of controlling these

292
 (H6 ) Greater ability to inflict damage is positively associated with higher levels
of violent competition between criminal organizations.
(H7 ) Greater capability of recovering from an attack is positively associated
with higher levels of violent competition between criminal organizations.

Measuring the potential of criminal organizations to inflict military damage is

difficult. This research relies on the variable rifles measuring the number of assault
rifles produced in the U.S. Data for this variable comes from the Bureau of Alchohol,
Tobacco and Explosives (2012) and it is measured in terms of million rifles annually
produced. Although it is not the ideal measure, this variable serves as a proxy
of the firepower of criminal organizations. The increasing availability of weapons
with military characteristics such as automatic chargers, detachable magazines with
multiple rounds, grenade launchers, and high-power rifles capable of piercing armored
vehicles represent a substantial increase in the firepower of criminal groups. These
types of weapons available to criminal organizations largely outstrip the weapons
that municipal and state police forces are generally equipped with. In addition, the
acquisition of this type of weaponry by one criminal group is likely to generate an
arms race against their rivals. As indicated by Dube et al. (2013), the role of assault
weapons in drug violence in Mexico became particularly important following the end
of the Federal Assault Weapons Ban in 2004. The expiration of the ban increased the
production of rifles and facilitated gun smuggling into Mexico, which substantially
increased criminal firepower and led to higher murder rates after 2004. As Figure 6.3
shows, the production of assault weapons in the U.S. substantially increased after
the expiration of the ban.
According to the theoretical model, criminal organizations can alter the relative
military balance by causing damage to their rival’s military capabilities or by effec-
tively recovering from an attack. The empirical assessment in this research uses a
measure of unemployment as a proxy for assessing criminals’ recovery capabilities.
The intuition is that high unemployment rates increase the human reserve available to

294
 2.2
2
Rifles (million)
1.8
1.6
1.4
1.2

2000m1 2002m1 2004m1 2006m1 2008m1 2010m1

Time (Year Month)

Figure 6.3. Annual production of assault weapons in the U.S.

criminal organizations, thus facilitating the replacement of casualties. This intuition

is supported by research on criminal behavior and rebel recruitment indicating that
higher levels of unemployment are associated with higher crime rates and participa-
tion in political violence (Fajnzylber, Lederman and Loayza, 2002b; Weinstein, 2007).
The variable unemployment measures the percentage of unemployed population at
state level on a quarterly basis with data from Instituto Nacional de Estadı́stica y
Geografı́a (2011a). As shown in Figure 6.4, Mexico experienced a substantial in-
crease in unemployment in 2009 when the U.S. economic crisis affected the Mexican
economy. In March of the same year, Mexico also suffered a severe economic shock
caused by the outbreak of the AH1N1 Influenza crisis which exacerbated already high
rates of unemployment. These high unemployment levels might be associated with
the dramatic increase in levels of violence between criminal organizations in 2009 and
2010.

295
 4.5
4
Unemployment
3 3.5
2.5
2

2000m1 2002m1 2004m1 2006m1 2008m1 2010m1

Time (Year Month)

Figure 6.4. National average of unemployment rate on a quarterly basis

6.2.2
Model
Specification

The statistical analysis evaluates the effect of these structural factors on violence
between criminal organizations. The dependent variable used in the empirical assess-
ment corresponds to the variable competition discussed in Chapter 4. This variable
comprises the number of violent events between rival criminal organizations. The
unit of analysis is at the municipality–day level and the model considers all munici-
palities of the country on a daily basis from January 1, 2000 to December 31, 2010.
As shown in Figure 4.2, the dependent variable is distributed in a negative binomial
function with hyper-dispersion. This indicates that in most municipality–days there
are no events of violence between criminal organizations, but there is a handful of
cases with high levels of violent competition. For that reason, the model consists
of a negative binomial model for panel data with random effects and errors clus-
tered by municipality. The decision to use random effects (RE) over fixed effects
(FE) is informed by the theory, according to which violence is concentrated more

296
in strategic territories. Since the geographic location is temporally invariant, an FE
model would exclude territorial variables from the estimation, thus eliminating the
counterfactual and generating omitted variable bias. In contrast, an RE model incor-
porates territorial variation across units to estimate theoretically relevant variables.
The following equation specifies the structural model of violent competition between
criminal groups:

Yit = α + d1 pit + d2 τ it + d3 rit + d4 uit + d5 cit + d6 dit + d7 sit + μit . (6.1)

where Yit represents violent competition among DTOs in municipality i on day t;

pit is a vector of poverty variables including poverty and State GDP ; τ it is a vector
of territorial variables including drug production, Gulf, Pacific, North and 9/11 ; rit
is a measure of the production of rifles; uit is the measure of unemployment; cit is a
measure levels of corruption; dit a vector that includes the price of cocaine and drug
markets measured by hospital discharges; and sit measures social fabric by a vector
including variables of divorce and young mothers. Scalars d1−7 are the coefficients
associated with each covariate, and μit contains the error terms.
For the sake of simplicity, let Xit be a vector of all exogenous covariates and let δ 
be a vector of scalars with their corresponding coefficients. In consequence, equation
(6.1) can be expressed as:
Yit = α + δ  Xit + μit . (6.2)

6.2.3
Statistical
Results
of
the
Structural
Model

This section presents the results of the statistical model assessing the effect of
structural factors on violent competition between criminal groups. Models 1–5 in
Table 6.1 sequentially introduce the set of variables corresponding to each group of

297
covariates. For the sake of simplicity, the discussion of results is focused on Model 5
presenting the full model specification. It is important to note that the coefficients
in the regression tables of this chapter are expressed in terms of the log of expected
counts, while the interpretation of results and the graphs are expressed in terms of the
number of expected events times the mean.7 Also, note that the sign and magnitude
of coefficients are remarkably stable across models. Another characteristic of the
results is the high statistical significance of the estimates. With “big data” (very
large data sets), it is not unusual to have highly significant estimates. This is due
to one of the most essential assumptions of statistical analysis: the central limit
theorem. This theorem states that if we have a sufficiently large population sample
from a true population with limited variance, the mean of all population samples
will be approximately equal to the mean of the true population and, as the number
of observations increase, the distribution of the sample means will approximate a
normal distribution. One of the implications of the central limit theorem is that
if the population sample is extracted from several other uncorrelated draws, all of
them “contaminated” with random error, the error of the population sample tends
to be normally distributed as the number of draws increase. This indicates that
as the number of observations in the population sample increase, the mean of the
population sample approaches the mean of the true population. In other words, there
is less difference or “error” between the mean of the sample and the true population.
Thus, having a large number of observations usually enables high levels of certainty
about the results without inducing bias in the estimates.

7
The description of the data presented in Chapter 4 shows that the events of violence are anoma-
lous episodes. Since the database has 9.8 million observations and there are no events of violence
for most municipality–days, the average or expected number of events of violence among DTOs is
very low; 0.005 events per municipality–day. In order to provide a more intuitive interpretation of
the results, the discussion of findings is expressed in terms of the number of expected events times
the mean. For example, if the mean of Y is 0.005 and a given change in X generates a predicted
increase of 0.015 in Y , this can be expressed as: the given change in X is associated with a threefold
predicted increase in Y over the average number of events.

298
 The findings resulting from Model 5 shown in Table 6.1 provide valuable insight
that contradicts some of the key explanations of violence and help put in perspective
other factors often associated with high levels of criminal behavior. The most striking
finding of the structural model is that higher levels of economic development at
the municipal and state levels are associated with more intense violence between
DTOs. This finding contradicts dominant theories of large-scale political violence
emphasizing the role of poverty to explain violent behavior, thus suggesting the need
to further analyze the distinction between rebels and criminals suggested in Section
1.2. The statistical analysis also provides strong support for the relevance of territorial
value suggested by the theoretical model. In general, violence between DTOs is more
intense in areas favorable for the production, reception and distribution of drugs.
The results also provide support for the relevance of military damage and recovery
capability suggested by the theoretical explanation. Another surprising finding is
that a reduction in drug prices is associated with higher levels of violence between
DTOs. In addition, the model reveals that changes in the traditional family structure
have a modest negative effect on violence. Finally, the results question the centrality
of corruption and local drug markets as key explanatory factors for understanding
the wave of criminal violence; these variables report the expected positive sign but
the magnitude of their effect is small.

299
 TABLE 6.1

STRUCTURAL DETERMINANTS OF VIOLENT COMPETITION

AMONG DTOs
(1) (2) (3) (4) (5)
Poverty -0.05*** -0.25*** -0.20*** -0.20*** -0.20***
(0.02) (0.02) (0.02) (0.02) (0.02)
State GDP log 1.96*** 1.73*** 1.01*** 1.04*** 1.05***
(0.02) (0.02) (0.02) (0.02) (0.02)
Drug production 0.09*** 0.18*** 0.19*** 0.18***
(0.01) (0.01) (0.01) (0.01)
Gulf 0.38*** 0.48*** 0.48*** 0.47***
(0.05) (0.04) (0.04) (0.05)
North 0.38*** 0.54*** 0.52*** 0.60***
(0.03) (0.03) (0.03) (0.03)
Pacific 1.00*** 0.66*** 0.67*** 0.67***
(0.04) (0.04) (0.04) (0.04)
9/11 1.23*** 1.15*** 1.07*** 1.06***
(0.04) (0.04) (0.05) (0.05)
Rifles (100 K) 0.13*** 0.14*** 0.13***
(0.00) (0.00) (0.00)
Unemployment 0.10*** 0.09*** 0.09***
(0.00) (0.01) (0.01)
Corruption -0.01*** -0.01***
(0.00) (0.00)
Cocaine price -0.001*** -0.001***
(0.00) (0.00)
Drug markets 0.001*** 0.001***
(0.00) (0.00)
Divorce -0.00***
(0.00)
Young motherhood -1.21***
(0.15)
Population log 0.15*** 0.03*** 0.05*** 0.04*** 0.06***
(0.01) (0.01) (0.01) (0.01) (0.01)
Constant -42.43*** -38.97*** -27.71*** -27.89*** -27.76***
(0.41) (0.41) (0.38) (0.40) (0.40)
Observations 9,868,208 9,868,208 9,868,208 9,868,208 9,833,864
Coefficients are in log of expected counts. Standard errors in parentheses.
The levels of significance are: * p< 0.1, ** p< 0.05, *** p< 0.01

300
 In contrast to the widely held belief that low economic development is associated
with higher levels of criminal behavior and internal conflict, the statistical analysis
indicates that violence between criminal organizations is more intense in wealthy
areas. This relationship is reflected in the negative coefficient of poverty and the
positive coefficient of State GDP. These estimates show that as poverty decreases at
the municipal level, violence between DTOs increases. It also increases as state GDP
increases. The magnitude of the effect of state GDP is remarkable. Moving from the
poorest state with a GDP of 2.03 million dollars per year to the average state with a
GDP of 15.2 million dollars increases events from 0.5 to 4.7 times the average number
of confrontations between DTOs. Moving from the average state to the wealthiest
state producing 165.6 million dollars per year is associated with 51.3 times more
events of violence between DTOs. These results contradict the expectation of H9
linking poverty with violence and are quite illustrative of the core differences between
economically motivated violence perpetrated by organized criminals and politically
motivated rebels discussed in Section 1.2.
There are several reasons why violence between criminal organizations might be
more intense in wealthy areas. In contrast to poor and underdeveloped areas, wealthy
areas tend to be more attractive to organized criminals because they offer much
greater opportunity for money laundering in the formal economic sector. Urban
areas also provide favorable conditions for discretely establishing safe houses and
storage facilities. Wealthy areas offer the opportunity of developing economies of
scale in which drug-trafficking organizations can conveniently expand their activities
to other illicit sectors such as extortion and kidnapping. Areas with high levels of
economic development also offer greater potential markets for local drug consumption.
In addition, developed areas make it easier for high level drug-traffickers to conceal
their wealth and operate with a low profile. All these factors make wealthier areas
more attractive to criminal organizations. As explained in Section 2.4.1, a high

301
concentration of criminal groups in a given area can easily degenerate to a war of all
against all.
The statistical analysis also provides support for H8 , which states that valuable
territories are likely to experience higher levels of violence among criminal organiza-
tions. Violence between DTOs is more intense in areas favorable for the production of
drugs, reception of shipments and entry spots to the U.S. market. The results indicate
that moving from a municipality that does not produce drugs to another one with
high levels of production increases events of violence between criminal organizations
from 6.6 to 12.4 times more. In addition, being located in reception municipalities
on the Gulf or the Pacific Ocean coasts generates respectively 10.2 and 11.9 times
more expected events of criminal competition. Northern municipalities contiguous
to the U.S.–Mexico border are expected to have 11.8 times more episodes of criminal
violence than non-border municipalities. The statistical analysis also indicates that
there were 7.14 times as many events of violence between DTOs after 9/11 than
before the terrorist attacks. The territorial centrality of drug violence supports the
implications of the theoretical model and is consistent with other findings addressing
the relevance of subnational geographic variation to understanding the dynamics of
domestic conflict (Buhaug and Ketil Rod, 2006; Buhaug, Gates and Lujala, 2009).
The structural model also finds support for hypotheses H6 and H7 , indicating the
relevance of military damage and recovery capabilities. The positive coefficient of
rifles indicates that the increase in the number of assault weapons produced in 2004
before the expiry of the ban to the maximum production of 24.5 million rifles reached
in 2009 is associated with an increase from 9.7 to 13.8 times as many events of violence
between DTOs. The increased availability of assault weapons improves the ability of
criminals to inflict military damage on their rivals, thus leading to higher levels of
violence between DTOs. The results also indicate that unemployment contributes to
the escalation of criminal violence. An increase from the average unemployment rate

302
of 2.9 percent to the highest level of unemployment of 9.2 percent is associated with
an increase in episodes of violence between criminal organizations from 3.6 to 11.4
times. Based on the theoretical argument advanced in this research, higher rates of
unemployment can be interpreted as a larger human reserve available for recruiting,
which can increase the recovery capability of criminal organizations.
In contrast to the prominence of corruption in extant explanations of criminal
behavior and drug violence, the statistical assessment reveals a limited explanatory
leverage of corruption. As expected from hypothesis H10.2 , the results indicate that
high levels of corruption are associated with lower levels of competition between
DTOs. This provides some support for arguments emphasizing the role of peaceful
configurations and state-sponsored protection rackets (Snyder and Duran-Martinez,
2009). However, the empirical leverage of corruption is so limited that it is hard to
sustain its centrality for understanding the escalation of violence. According to the
results, increasing the levels of police corruption from zero to 57 percent is associated
with a slight reduction from 7.7 to 5.8 times fewer events of violence. In other words,
there are only 1.9 times fewer confrontations between rival DTOs in states where no
one is asked for a bribe than in states where one out of two citizens are victims of
police corruption. Of course there are reservations about the quality of the corruption
data reported by Transparency International, but it is still striking that such a large
increase of corruption generates such a modest reduction of violence.
The structural model generates mixed support for hypothesis H11 associating
drug markets with criminal violence. In contrast to the expectation that high drug
prices are linked with intensified violence between criminal organizations, the results
indicate that higher prices of cocaine in the U.S. market are related to lower levels
of violence in Mexico, although the effect is modest. Increasing the price of a gram
of pure cocaine by nearly $100 from its minimum of $128 to its maximum of $222
is associated with a reduction of the expected levels of violence from 7.4 to 6.2

303
times more events of violence between DTOs. At first glance, this result might
generate enthusiasm about the reduction of violence as drugs become more expensive.
However, as illustrated by Figure 6.5, the price of a gram of pure cocaine has been
declining steadily since the early 1980s. According to the Office of National Drug
Control Policy (2004), the prices of all other drugs show a similar negative trend.
Due to data limitations, this research only considers the study of drug violence in
Mexico between 2000 and 2010; therefore it only includes part of the variation in
cocaine prices where the slope is slightly negative and in excludes the sharp decline
observed in the 1980s. If drug prices in the U.S. continue to fall, as is likely, it is
plausible to expect an increase of violence between DTOs in Mexico.

900
Price per pure gram of cocaine in the U.S.

800

700

600

500
(dollars)

400

300

200

100

0
1981

1983

1985

1987

1989

1991

1993

1995

1997

1999

2001

2003

2005

2007

2009

Years

Figure 6.5. Price per pure gram of cocaine in the U.S.

304
 Why would DTOs fight each other if drug prices are declining? The reduction
in prices means that revenues are shrinking, which might increase the motivation for
DTOs to fight their rivals in order to control their share of the market. As indicated
by the Office of National Drug Control Policy (2004), the price of drugs has been
declining in recent decades at the same time that the purity of almost all drugs has
been increasing. This means that DTOs have to sell drugs of better quality at lower
prices. This might indicate that the profitability of drug markets is shrinking, thus
generating stress in criminal organizations. Profit shrinkage might motivate some
DTOs to employ violence to eliminate their competitors in order to capture their
share of the market and to increase their own revenues. Profit shrinkage in the drug
market may also motivate the organisations to diversify their illegal activities. If
drug trafficking is not as profitable as it used to be, criminal organizations might try
complement their income with other violence-intensive activities such as racketeering
and kidnapping, thus generating more violence.
The results also provide some support for hypothesis H11 in terms of the value of
local drug consumption. As mentioned before, variable drug markets measures the
number of hospital discharges diagnosed with intoxication form narcotics as a proxy
for the levels of drug consumption at the municipal level. This measure assumes that
a higher frequency of observed drug intoxication might reflect an equally distributed
unobserved frequency of drug consumption. Results report a positive coefficient
for the variable variable drug markets, yet the magnitude of the effect is modest.
Moving from the mean of 1.2 hospital discharges to one standard deviation above
the mean of 17.9 drug intoxication cases is associated with a slight increase of 0.46
times more expected events of violence among DTOs. An increase of ten standard
deviations above the mean of drug intoxication would produce 4.3 times more events
of violence and it would require an extraordinary increase of fifty standard deviations
above the mean to produce 16.9 times more confrontations among DTOs. In this

305
sense, these results challenge a broadly held assumption that the wave of violence in
Mexico is primarily caused by the fierce competition among drug cartels fighting to
control emerging local consumption markets. It is true that there is more violence
among criminals in municipalities with higher consumption of drugs, but the effect
of increasing local consumption on violence is so small that it does not constitute a
central explanation of violence in Mexico.
Finally, the results throw doubt on the relationship stated in hypothesis H12
arguing that the erosion of traditional social structures is associated with higher
levels of violence between DTOs. Municipalities with a larger number of divorces
and higher proportion of young mothers are associated with lower levels of violence
between DTOs. Although the magnitude of the effect is very small, the relationship
is negative. Increasing the number of divorces from the mean of 25 divorces by one
standard deviation above the mean to 200 divorces has a negative, although modest
effect, reducing the expected number of violent events among rival criminal groups
from 6.98 to 6.97 times. A larger increase from the median number of divorces
(3,525 cases) to the seventy-fifth percentile (5,288 cases) would reduce the expected
number of confrontations between DTOs from 6.2 to 5.5 times. The effect of young
mothers between the age of 12 and 19 show a similar modest effect on violence in
the negative direction. Increasing the average proportion of young mothers from 14
percent by one standard deviation above the mean to 21 percent has a slight negative
effect on the expected number of confrontations between DTOs, from 8.19 to 8.1
times.8 These results indicate that changes in the traditional family structure are not

8
Part of the effect of these variables in the model could be caused by using contemporary measures
of divorce and young mothershood instead of lagged variables that could assess the effect of these
factors over time. For example, Donohue and Levitt (2001) use time series of abortion rates from
the 1970s to assess their impact on criminal behavior in the 1990s. Data lagged by twenty years
enables the effect of structural changes on the next generation to be assessed. However, data on
divorce and young motherhood in Mexico is not available before 2000. Even if there were theoretical
justifications for lagging those variables by at least five years, doing so would eliminate half of the
database and generate problems of data truncation. In addition, this five-year lag would only leave

306
conducive to higher levels of violence between criminal organizations. The evidence
thus contradicts the conservative discourse of government authorities indicating that
family and social disintegration is a key factor for explaining the escalation of violence
in Mexico (see Presidencia de la República, 2012).

6.2.4
Overall
Model
Fit

The structural model discussed in the previous section provides some valuable
findings about the determinants of the wave of violence in Mexico. However, these
structural factors simultaneously affect several areas and usually evolve slowly over
time, thus providing limited explanatory power to account for the rapid variation in
violence. As discussed in Chapter 4, levels of violence between criminal organizations
vary substantially over time and across space, and the use of daily event data at the
municipal level reveals the limited analytical leverage of structural explanations and
aggregated data. However, the most important limitation of the structural approach
is not empirical but theoretical. Structural explanations and statistical models over-
look the highly interactive and dynamic characteristics of conflict, thus generating
problems of omitted variable bias.
In order to assess the fit of the model, researchers often rely on the R2 parameter
as way to evaluate the amount of variation of the dependent variable explained by
the regression analysis. Unfortunately, the R2 corresponds to linear models estimated
through ordinary least squares (OLS) and it is not applicable to non-linear models
for categorical data using maximum likelihood estimation (MLE). Some MLE models
allow goodness of fit to be estimated using McFadden’s pseudo R2 and its adjusted

data between 2005 and 2010 for conducting the analysis and that might bias the estimates upwards
because this is the period showing the highest levels of violence.

307
version. However, these alternative estimates are not available for negative binomial
models for panel data with random effects like the model used in this research.9
Since it is not possible to assess the goodness of fit of the model through standard
post-estimation parameters, the alternative is to use visual methods. Figure 6.6 pro-
vides a visual representation of the fit between the observed values of the dependent
variable and the predicted levels of violence between DTOs estimated by means of
the structural model. A perfect fit would mean exact correspondence between the
observed and the predicted outcome. Since the dependent variable has a negative
binomial distribution, the visual representation of a perfect fit would be the data
points being heavily clustered on zero and then perfectly aligned along a 45-degree
diagonal for the non-zero observations. Of course it is not plausible to expect a sta-
tistical model to have perfect fit, but a good fit should display these characteristics.
Unfortunately, Figure 6.6 reveals that even the fully-specified structural model does
a poor job in predicting violence between criminal organizations. The number of
predicted episodes of violence generated by the model do not go above 1.5 events per
municipality–day. By neglecting the violent interaction between the state and DTOs,
the structural model provides limited explanatory power for understanding the levels
of violent competition between criminal organizations.

6.3
Interactive
Model
of
Drug
Violence

Given the limitations of the structural model to explain the substantial variation
of violent competition among DTOs, this research relies on an interactive approach
that includes the various actions and reactions between the state and criminal orga-

9
The Stata command used for estimating the model with negative binomial models for panel data
with random effects is -xtnbreg, re-. As indicated in the command syntax, the post-estimation
options do not allow generation of a pseudo R2 , Count R2 or Adjusted Count R2 ; see http://www.
stata.com/help.cgi?xtnbreg.

308
 Figure 6.6. Goodness of fit of the structural model

nizations inherent to the war on drugs. This approach has the empirical advantage
of incorporating fine-grained data to understand the micro-dynamics of conflict and,
perhaps more importantly, it allows the ontology of violence portrayed in theoretical
model to be aligned with the empirical strategy. As stated by Hall (2003, 387), in
order to overcome the limitations of structuralist ontologies which assume that large
structural variables have a strong, consistent, independent effect across space and
time, it is necessary to incorporate alternative approaches that acknowledge more
complex interactions and deal with endogenous processes.
By aligning the ontology of the theoretical model and the empirical strategy, this
research contributes to a recent trend in political science known as the “Empirical
Implications of Theoretical Models” approach (EITM). According to Aldrich, Alt and
Lupia (2007), the EITM approach joins the analytical leverage of formal modeling
and the methodological rigor of quantitative inference to improve the precision and
credibility of research results. This research follows the EITM approach by deriving

309
empirical implications from the formal model and designing an empirical strategy
that allows those hypotheses to be tested in the data.

6.3.1
Hypotheses
to
Evaluate

The formal model presented in Chapter 2 offers an integrative explanation for

the onset, escalation and distribution of organized criminal violence. The theoretical
model departs from a status quo in which corruption enables peaceful coexistence
between government authorities and criminal organizations. The model argues that
democratization then erodes these corrupt agreements by altering the system of po-
litical incentives and motivating authorities to enforce the law against criminals. The
model additionally argues that in contexts of political strain when the legitimacy of
the government is called into question, authorities reap political benefits by imple-
menting harsh security policies that portray them as strong leaders and boost their
approval levels. The following hypotheses already stated in Section 2.4 are derived
from these arguments:

(H1 ) Democratization is associated with higher levels of law enforcement.

(H2 ) Increased political strain is associated with higher levels of law enforce-
ment.

(H3 ) High levels of corruption are associated with lower levels of law enforce-
ment.

Hypothesis (H2 ) is further refined to derive distinct hypotheses for different types
of violent and non-violent law enforcement actions. The use of violent tactics is
usually more costly for government authorities than non-violent tactics, therefore
violent enforcement is used in exceptional circumstances. In contexts of political
strain, the authorities reap higher political benefits from using violent enforcement
tactics than from non-violent enforcement despite the higher cost of the former. This
nuanced argument can be stated in the following terms:

310
 (H2.1 ) Increased political strain is associated with higher levels of violent en-
forcement than non-violent enforcement tactics.

The theoretical explanation also uses a contest success model for territorial com-
petition to explain the escalation of violence. According to the model, increasing law
enforcement triggers an escalation of conflict between the state and criminal organi-
zations and, most importantly, it unleashes a wave of violence between rival criminal
groups. State action has a disruptive effect on the relative military balance of crim-
inal organizations by weakening the ability of a DTO to protect its territory, thus
motivating an invasion from a rival DTO. The model thus argues that law enforce-
ment leads to violence between criminal organizations. As discussed in Section 2.4,
this relationship can be stated as:

(H5 ) Increasing law enforcement is associated with higher levels of violence

among DTOs.

In order to explain the geographic distribution of violence, the model explicitly

incorporates the importance of territorial value as a key determinant of conflict.
According to the theoretical expectation, valuable territories tend to contain higher
levels of violence among criminal organizations. This relationship was previously
stated in hypotheses H8 .

6.3.2
Data

Law enforcement is the outcome variable considered by hypotheses H1 , H2 , H3 and

H2.1 . Law enforcement is operationalized based on the set of violent and non-violent
tactics used by the state to fight criminal organizations. As discussed in Chapter 4,
these variables are identified by Eventus ID on a daily basis at the municipal level.
Violent enforcement includes events in which state security forces attacked, wounded
or killed suspected members of criminal organizations. Non-violent enforcement tac-

311
tics include events of arrests, seizure of criminal assets, drug interdiction, and seizures
of weapons.
As mentioned by Coppedge (2012), democracy is one of the thickest concepts in
political science, comprising several dimensions. We can mean many different things
when we talk about “democracy.” Similarly, democratization can also be understood
as a prolonged process of political change involving several dimensions. Instead of us-
ing a thick conceptualization of democratization that could lead to employing dozens
of indicators, this research focuses on a thin perspective that employes two variables;
the effective number of political parties, and divided government. These two measures
are used to operationalize hypothesis H1 .
The variable effective number of parties (ENP) measures the effective number of
political parties competing in the presidential elections when Ermesto Zedillo, Vicente
Fox and Felipe Calderón, respectively, were elected. The ENP index is calculated
using the formula proposed by Laakso and Taagepera (1979) which takes into account
the share of votes obtained by each party in the presidential election. An ENP value
of 1 means that the party system is dominated by a single political actor. Larger
values of the ENP index mean that there are more relevant political actors in the party
system, which is often interpreted as a measure of competition. Using the ENP at the
presidential level is a valid indicator of political competition in Mexico. As discussed
in Chapter 5, for more than seventy years the Institutional Revolutionary Party
dominated the political system in an autocratic fashion that permitted no genuine
electoral competition (Greene, 2007; Magaloni, 2006). During this period, the PRI
used a variety of tactics to manipulate election results in order to secure victory. The
increase that occurred in the effective number of parties reflects substantial changes
in the political scene. Analyzing the effect of increasing political competition at
the executive level is also important to assess the impact of democratization on law
enforcement against organized crime because the Mexican constitution also endows

312
the executive with the formal title of commander in chief of the armed forces. In
consequence, the president has the prerogative of commanding the Army, the Navy
and Federal Police in the fight against organized crime, a prerogative not extended
to governors or majors.
Another important effect of democratization is the disruption of the hierarchical
chain of command through the different levels of government that allowed implemen-
tation of corrupt agreements between political authorities and criminal organizations.
During the period of one-party dominance, the PRI held the executive office continu-
ously from 1929 to 2000, and the majority of Congress until 1997. At the sub-national
level, the PRI controlled every state until 1992 when they lost the first governorship,
and held most municipalities of the country until the 1980s. This long period of
cohesive party dominance at all levels of government enabled the construction of
a straightforward chain of command that facilitated political operations, including
non-aggression equilibrium with criminal organizations. With the advent of democ-
ratization, starting at the sub-national level, the Mexican political scene began to
change, and the chain of command was gradually eroded. To reflect the decentraliza-
tion of political power, variable divided government measures the degree of partisan
division in the executive office across the three different levels of government (federal,
state and municipal). This variable takes the value 0 if the three levels of government
belong to the same party, 1 if either the governor or the mayor belong to the same
party as the president, and 2 if neither the governor nor the mayor belong to the
same party as the president. Divided government thus serves as a proxy for the erod-
ing effect of democratization on the chain of command and the increased difficulty of
establishing or maintaining peaceful configurations between criminals and politicians.
In order to measure the concept of political strain, this research uses the margin
of victory, the difference in vote share between the winner and the runner-up in pres-
idential elections. This variable serves as a proxy for political strain, as it reflects the

313
political difficulties of the 2006 election when the PAN candidate, Felipe Calderón,
won the election by a margin of less than 1 percent of the votes. This election repre-
sented the most profound political crisis in the short democratic history of Mexico.
Due to the narrow margin of victory, the opposition candidate from the left-wing
Party of the Democratic Revolution (PRD), Andrés Manuel López Obrador, loudly
raised allegations of election fraud. López Obrador called for a recount of all the
ballots under the slogan “voto por voto, casilla por casilla” (vote by vote; booth by
booth) to clear away any doubts over the election result. In a country with such
a long history of electoral fraud, vote buying, voter coercion, control of election of-
ficials, political clientelism and result rigging, these allegations of fraud resonated
with broad sectors of the population. After the election, thousands of protesters
took the streets in different parts of the country. Remarkably, over a million and
a half people attended a rally in downtown Mexico City in the largest post-election
protest. Elite and mass polarization swelled during a period several months. The
crisis reached its peak when the members of the opposition PRD took over Congress
to prevent president-elect Calderón from making his pledge to the Constitution and
formally taking office. Members of the Mexican secret service had to push through
the crowded tribune to make space to allow Calderón to formally take office. After the
episode, López Obrador kept referring to Calderón as “The Illegitimate President.”
This period of political crisis has been widely characterised by political analysts, pub-
lic opinion leaders, scholars and journalists as severely damaging the legitimacy of
President Calderón from the the beginning of his administration (e.g. Álvarez Béjar,
2007; Castañeda, 2012; Castañeda and Aguilar, 2010; Pacheco, 2006).
The remaining variables are the same as the variables specified in the structural
model.

314
6.3.3
Model
Specification

The concept of political violence conflates highly dynamic and endogenous pro-
cesses of conflict. As stated by Kalyvas, Shapiro and Masoud (2008), violence is used
by those challenging the existing order and by those fighting to preserve it. One
of the central challenges of analyzing the micro-dynamics of violence, therefore, is
the high degree of endogeneity and reciprocal causation of overlapping but distinct
processes of violence. In this particular research, the risk of endogeneity rests primar-
ily in hypothesis H3 , which states that law enforcement generates violence between
criminal organizations. However, the relationship can also operate in the opposite
direction, as the legitimate use of force against potential or actual threats is inher-
ent to the process of state formation (Hobbes, 1651; Scott, 2009; Tilly, 1985, 1992;
Weber, 1978).
This endogenous relationship complicates the design of a proper identification
strategy capable of disentangling the reciprocal causation between enforcement and
violence between criminals. Taking the naive approach of simply regressing the levels
of violence between DTOs on the different types of law enforcement would certainly
identify a strong positive relationship. However, the estimates would be highly biased
because they would be incorporating the feedback effects caused by this endogenous
relationship, thus generating misleading conclusions. Instead of using a naive ap-
proach, the identification strategy used in this study follows research by Acemoglu,
Johnson and Robinson (2001); Angrist and Pischke (2009); Miguel, Satyanath and
Sergenti (2004) and Levitt (1997), who employ an instrumental variables approach
to overcome problems of endogeneity.
Instrumental variables methods are quasi-experimental methods used to estimate
the impact of a treatment variable when an experimental design is not feasible. In-
strumental variables are used as an exogenous source of variation that enables part
of the effect of a variable on the outcome to be isolated regardless of the endoge-

315
nous relationship among them. The identification strategy of this research uses a
two-stage instrumental variable approach. The first stage assesses the effect of de-
mocratization and political strain on levels of law enforcement. The second stage
then evaluates the impact of predicted levels of enforcement caused by the exogenous
variation of democratization and political strain on the levels of violence between
criminal organizations.
As mentioned earlier, the variables of interest (enforcement and violent competi-
tion) are distributed according to Poisson functions with hyperdispersion; therefore
the statistical analysis uses a negative binomial model for panel data. The unit of
analysis is the municipality–day for all municipalities of the country on a daily basis
between January 1, 2000 and December 31, 2010, producing a total of 9.8 million
observations. The time series cross-sectional data structure enables the fine-grained
variation of violence to be analyzed across time and space. The model includes a ran-
dom effects specification with errors clustered by municipality. The random effects
justification is the same as that for the structural model, as discussed above.
Unfortunately, there are no standardized commands in Stata or R for computing
an instrumental variable regression with a negative binomial distribution for panel
data. I therefore computed the instrumental variable assessment as a two-stage neg-
ative binomial regression analysis. As discussed in detail by Angrist and Pischke
(2009), computing the two stages in separate regressions does not induce bias in the
estimates of the second stage. However, the variance–covariance matrix produces
misleading residuals. For this reason, errors for the second stage were calculated by
bootstrapping. This is a method for assessing the accuracy of the estimates based
on taking several random samples from the observed distribution. Instead of imple-
menting the bootstrapping iterations from the full sample of 9.8 million data points,
the bootstrapping procedure used a more stringent test based on smaller samples of
300,000 observations. In addition, instead of running hundreds of iterations to assess

316
the accuracy of the estimates, the bootstrapping procedure implemented in this re-
search considered a conservative approach of only fifty iterations. This bootstrapping
strategy conducting only a few iterations based on relatively small random samples
of the data constitutes a highly astringent test for the precision of the confidence
intervals estimated in the second stage.
The two-stage analysis uses violent competition among DTOs and law enforce-
ment as the endogenous variables. The instrumental variables are the effective num-
ber of parties, the degree of divided government, and the margin of victory. The
other variables are considered to be exogenous covariates. The first stage model is
expressed in terms of the following equation:

Eit = α + θ1 Pit + θ2 Dit + θ3 Mit + δi Xit + μit , (6.3)

where Eit represents the levels of law enforcement in municipality i on day t, vector
Pit contains information on the effective number of parties, Dit is the degree of
divided government and Mit represents the margin of victory. As stated in equation
(6.2) for the structural model, Xit a vector containing some control variables. Scalars
θi and δi represent the coefficients for the corresponding variables and μit contains
the error terms. Equation (6.3) is the mathematical expression of the first stage,
comprising hypotheses H1 and H2 .
The reduced form of the instrumental variable model is the mathematical ex-
pression of the relationship stated in hypothesis H3 , in which predicted levels of
law enforcement caused by democratization and political strain have an impact on
violence between DTOs. The second stage is formulated mathematically as:

Yit = α + β1 E it + δ  Xit + μit . (6.4)

317
where Yit represents violence between DTOs in municipality i on day t and E it
represents the predicted level of law enforcement from the first stage caused by de-
mocratization and political strain. The vector of controls Xit contains the control
variables already discussed in Equation (6.2) of the structural model. Consider that
that Xit = Xit + Xit , where Xit is the set of controls of the first stage and Xit are
all other exogenous covariates.
Following Angrist and Pischke (2009), we can think of instrumental variables as
initiating a causal chain where the instruments Pit , Dit , and Mit affect the variable
of interest Eit , which in turn affects the outcome Yit . In this way, the model isolates
part of the effect of law enforcement on violent competition among criminals caused
by the exogenous variation of democratization and political strain. This identification
strategy addresses the problem of endogeneity while being consistent with the data
generation process expected from the theory, thus allowing consistency between the
ontology of the theoretical construct and the methodology used for evaluating it.
The instrumental variable approach helps remove the endogenous relationship
between the various processes of violence and enables the effect of law enforcement
on violence between DTOs to be identified. Variation in the number of political
parties, the degree of government division, and the margin of victory are plausibly
exogenous to levels of violence between criminal groups. In consequence, the change in
incentives caused by the process of democratization and political strain is conceivably
independent from the wave of drug-related violence. The employment of this type of
political variables as exogenous instruments is congruent with other political variables
such as local elections and close municipal victories used by other authors (Dell, 2011;
Levitt, 1997) to assess the quasi-experimental effect of law enforcement on criminal
violence.
One of the requirements of instrumental variables is the exclusion restriction,
which requires the instrument to be assigned ‘as if’ random (Angrist and Pischke,

318
2009; Dunning, 2012). This can be accomplished when the instrumental variable (Zi )
is correlated with the instrumented variable (Si ) of interest but uncorrelated with
other determinants of the outcome variable (Yi ). For example, researchers employing
instrumental variables have used truly random or as-if random factors such as colonial
settler mortality rates to study contemporary economic growth (Acemoglu, Johnson
and Robinson, 2001), rainfall to study land invasion (Hidalgo et al., 2010) and twin
births to study the economic consequences of unwanted motherhood (Bronars and
Grogger, 1994). In these examples, there is a strong presumption of compliance
with exclusion restriction. However, in non-experimental research, it is hard to find
adequate instruments that comply perfectly with the exclusion restriction. For the
specific case of this research, an ideal instrument would have to be correlated with
law enforcement but uncorrelated with any other determinant of violence between
drug trafficking organizations.
A critical assessment of using democratization and political strain as instrumental
variables allows the possibility of some correlation of these factors with economic
development and corruption, which are already identified as determinants of criminal
violence in the structural model. For instances, the relationship between high levels
of economic development and democracy is well known in the literature (Przeworski
et al., 2000). In the case of Mexico, Magaloni (2006) argues that economic conditions
are related to increasing levels of political competition that could be reflected in the
number of political parties and divided government. In addition, the literature on
accountability has identified a relationship between high levels of democracy and low
levels of corruption (Schedler, Diamond and Plattner, 1999). In order to minimize
bias in the generation of first stage estimates due to potential violation of the exclusion
restriction, the model specification includes some control variables in Xit in Equation
(6.3). This vector of covariates is composed of measures of poverty at the municipal

319
level, state GDP, population, and corruption. The inclusion of these variables helps to
control part of the variation caused by possible violations of the exclusion restriction.
An purist approach to the use of instrumental variables as a quasi-experimental
strategy would emphasize strict observance of the exclusion restriction. However,
perfect quasi-experimental instruments are hard to find, and researchers often use
control variables in the first stage to address concerns about the violation of some
assumptions. In this way, the inclusion of controls in the first stage is consistent with
the central objective of the instrumental variable approach of addressing problems
of endogeneity by trying to isolate the effect of the instrumented variable on the
observed outcome.

6.3.4
Effect
of
Violent
Enforcement
on
Violence
Among
DTOs

First stage of violent law enforcement

This part of the discussion of the results will focus on analyzing the relationship
between violent law enforcement and violence between DTOs. The second part will
compare the effect of different law enforcement tactics, both violent and non-violent,
on criminal competition.
Tables 6.2 and 6.3 present the results of the first and second stage respectively
for the relationship between violent law enforcement and violent competition between
DTOs. Models 1–3 in Table 6.2 sequentially assess the effect of the effective number of
parties, margin of victory and divided government on violent law enforcement without
controls. Model 4 shows the full model specification of political variables including
controls. In general, the first stage provides strong support for hypotheses H1 and

H2 , which propose that democratization and political strain motivate politicians to

use force to fight crime. The magnitude of coefficients of the political variables across
all models in Table 6.2 are stable, as are their high levels of statistical significance. In
addition, it is important to notice that the instruments are remarkably strong even

320
after inclusion of the control variables. According to Angrist and Pischke (2009)
and Stock, Wright and Yogo (2002), the standard criteria to evaluate the strength
of instruments is when the F-statistic is above 10 at acceptable levels of significance.
As reported in Table 6.2, the F-statistics of Models 1–4 are several times larger than
the baseline threshold, indicating the presence of strong instruments.

TABLE 6.2

FIRST STAGE: BASE-LINE AND FULL MODEL OF VIOLENT

ENFORCEMENT
(1) (2) (3) (4)
ENP 5.699*** 4.552*** 4.565*** 4.304***
(0.107) (0.210) (0.210) (0.214)
Margin of victory -6.682*** -6.490*** -4.583***
(1.134) (1.135) (1.144)
Divided government 0.095*** 0.121***
(0.031) (0.032)
Poverty -0.408***
(0.038)
State GDP (log) 0.589***
(0.049)
Population (log) 0.191***
(0.018)
Corruption -0.032***
(0.002)
Constant -22.213*** -18.339*** -18.518*** -31.224***
(0.351) (0.706) (0.708) (1.103)
Observations 9,868,208 9,868,208 9,868,208 9,868,208
F - statistic 2,842*** 1,426*** 953*** 539***
Coefficients are in log of expected counts. Standard errors in parentheses.
The levels of significance are: * p< 0.1, ** p< 0.05, *** p< 0.01

321
 The discussion of the results shown in Table 6.2 will be focused on the full spec-
ification of the first stage reported in Model 4. As expected from the theory, the
statistical analysis indicates that democratization and political strain increase politi-
cal incentives to enforce the law, thus providing strong support for hypotheses H1 and
H2 . The results show that increasing political competition measured by the effective
number of parties is associated with the intensification of violent law enforcement
against criminals. Figure 6.7 presents the marginal effect of increasing the effective
number of political parties on the use of violent enforcement. According to the figure,
the increase in competition from 2.9 effective parties in the Zedillo administration to
3.3 effective parties when Calderón became president increased violent enforcement
events from 3.8 to 21.8 times more expected events of state violence. This result indi-
cates how sensitive the use of violent enforcement is to slight changes in the political
scene: increased levels of political competition have a profound effect on the system
of incentives that motivate politicians to use force against criminals.
The results also indicate that the degree of divided government has a significant
effect on enforcement. Figure 6.8 shows that changing from a unitary government to a
divided government in which the governor or mayor are not from the president’s party
increases the expected number of events of violent law enforcement against crime from
9.8 to 14.9 times. In addition, comparing the levels of law enforcement in unified
government to a context where neither the governor nor the mayor belong to the
same party as the president increases episodes of violent enforcement from 9.8 to 15.2
times more. This result indicates that democratization has a considerable disruptive
effect on the uni-partisan chain of command that could facilitate the establishment
and feasibility of state-sponsored protection rackets for criminal organizations.
These results help to elucidate the political conditions that facilitated a peaceful
equilibrium between the state and DTOs during the period of political hegemony
dominated by the PRI. The hierarchical chain of command characteristic of the

322
 30

25

Predicted violent enforcement

20

times the mean

15

10

0
2.9
3.0
3.0
3.0
3.1
3.1
3.2
3.2
3.2
3.3
3.3
Effective Number of Parties

Figure 6.7. Effect of political competition on violent enforcement

hegemonic party system enabled non-aggressive coexistence between the state and
criminal groups. In addition, the lack of effective elite circulation through electoral
means lent these agreements stability and assurance. The gradual process of democ-
ratization deeply affected the Mexican political arena and disrupted these peaceful
configurations. As opposition parties entered the political scene, implicit agreements
between corrupt officials and DTOs became more difficult to achieve or to sustain.
Political competition also created incentives for government authorities to provide
public goods such as public security.
Improving democratic conditions is necessary yet not sufficient for launching a cru-
sade against organized crime. Such bellicose behavior requires a trigger. Model 4 also
shows that winning the presidency by a narrow margin of votes motivates politicians
to use the state’s coercive apparatus to fight crime. This result provides support for
hypothesis H2 which states that periods of political strain are associated with intensi-

323
 18
16
14

Predicted violent enforcement

12

times the mean

10
8
6
4
2
0
Unified Divided Divided
governement government 1* government 2**
Divided government
* The governor OR the Mayor are from a different party that the President's
** The governor AND the Mayor are from a different party that the President's

Figure 6.8. Effect of divided government on violent enforcement

fication of enforcement. Figure 6.9 shows that changing from the comfortable margin
of victory of 22.8 percent when Zedillo was elected to the narrow margin of victory
of less than 1 percent of the votes when Calderón was elected is associated with a
change from 5.7 to 14.9 times more expected events of violent enforcement. This
finding is consistent with argument that President Calderón launched a full-fledged
campaign against crime as a way to increase his popularity after winning the 2006
election by a very narrow margin in the midst of accusations of fraud (Castañeda,
2012).

Second stage of violent law enforcement

In general, results of the first stage indicate that democratization and periods
of political strain are associated with higher levels of violent law enforcement. The

324
 18

16

Predicted violent enforcement

14

12

times the mean

10

0
5%
7%
9%
11%
13%
1%
3%

15%
17%
19%
21%
23%
Margin of victory

Figure 6.9. Effect of political strain on violent enforcement

next step is assessing the effect of violent crackdowns generated by the exogenous
variation of political variables on the levels of conflict between criminal organizations.
The answer is presented in Table 6.3 which shows the estimates for the second stage.
The results provide strong support for hypothesis H5 which claims that government
efforts to fight crime trigger waves of violence between rival DTOs. This relationship
might be related to the theoretical expectation about the disruptive effect of law
enforcement on the relative military balance among DTOs that unleashes turf wars
among rival criminal groups.

325
 TABLE 6.3

SECOND STAGE: BASE-LINE AND FULL MODEL OF VIOLENT

COMPETITION AMONG DOTs
(1) (2) (3) (4)
Violent enforcement 49.860*** 49.777*** 48.873*** 11.501***
(2.826) (2.112) (2.609) (1.814)
Retaliation 0.593*** 0.594*** 0.590*** 0.594***
(0.183) (0.133) (0.147) (0.174)
Drug production 0.197*** 0.197*** 0.195*** 0.167***
(0.059) (0.041) (0.063) (0.057)
Gulf 0.509*** 0.509*** 0.481*** 0.582***
(0.163) (0.189) (0.172) (0.166)
North 0.464** 0.465** 0.456** 0.469***
(0.226) (0.232) (0.233) (0.167)
Pacific 0.378*** 0.379*** 0.425*** 0.761***
(0.098) (0.098) (0.094) (0.133)
9/11 0.896*** 0.831*** 0.839*** 1.085***
(0.242) (0.261) (0.249) (0.178)
Poverty -0.245*** -0.245** -0.256** -0.197**
(0.094) (0.095) (0.101) (0.083)
State GDP (log) 0.407*** 0.409*** 0.435*** 0.704***
(0.034) (0.036) (0.034) (0.116)
Population (log) 0.041 0.041 0.042 -0.098*
(0.036) (0.042) (0.039) (0.052)
Rifles (100 K) 0.008 0.009 0.011 0.111***
(0.033) (0.017) (0.034) (0.024)
Unemployment 0.081*** 0.081*** 0.081*** 0.061**
(0.024) (0.014) (0.024) (0.024)
Corruption -0.008*** -0.008*** -0.008*** 0.008***
(0.003) (0.003) (0.003) (0.003)
Cocaine prince 0.005** 0.005** 0.005** -0.001
(0.002) (0.002) (0.002) (0.002)
Drug markets 0.001 0.001 0.001 0.002
(0.001) (0.002) (0.001) (0.002)
Divorce 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
Young motherhood 1.153 1.132* 1.011 -0.174
(0.922) (0.606) (0.992) (0.940)
Constant -15.608*** -15.584*** -16.065*** -19.830***
(0.977) (0.917) (1.042) (1.754)
Observations 9,868,208 9,868,208 9,868,208 9,868,208
Coefficients are in log of expected counts.
Bootstrapped standard errors in parentheses.
The levels of significance are: * p< 0.1, ** p< 0.05, *** p< 0.01

326
 Models 1–4 in Table 6.3 report the effect of violent enforcement and other covari-
ates on violent competition among DTOs. The predicted levels of enforcement are
generated from the first stage, in which the first three models include no controls
and the fourth one has the full specification. As Table 6.3 shows, the size of the co-
efficients of violent enforcement in Models 1–3 is substantially larger than in Model
4. This suggests that not including controls in the first stage might overestimate
the predicted levels of law enforcement due to possible violations of the exclusion
restriction, which is then corrected by including control variables. In consequence,
the enforcement coefficient in Model 4 represents a more plausible estimate of the
isolated variation of violent law enforcement generated in the fully specified model in
the first stage. The discussion of results of the second stage is focused on Model 4.
According to Model 4 in Table 6.3, the use of state violence to fight crime has
a strong increasing effect on the levels of violence between criminal organizations.
Figure 6.10 illustrates the relationship between the predicted degree of violent en-
forcement caused by democratization and political strain on violence between DTOs.
The graph shows that intensifying the predicted levels of violent enforcement from
minimum to maximum generates an increase from 5 to 62.7 times more expected
events of violence among DTOs. The results indicate that after controlling for en-
dogeneity, violence among criminal organizations surges upward as the Mexican gov-
ernment intensifies its efforts to fight crime. This provides empirical support for the
theoretical argument about the disturbing effect of state intervention. According to
the formal model, government enforcement against a criminal organization weakens
that criminal group and is likely to motivate invasion by a rival criminal group into
the territory of the weakened cartel. If the government launches a generalized and
sustained campaign against all criminal organizations operating within its territory,
state action is likely to have a profound disturbing effect on the criminal groups and
the balance between them, and unleash a massive wave of violence of all against all.

327
 80

Predicted competition among DTOs

70

60

times the mean

50

40

30

20

10

0
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0.22
Predicted violent enforcement

Figure 6.10. Effect of violent enforcement on violent competition among

DTOs

The coefficients of all other covariates in Model 4 show very similar results to
those reported in the structural model. This indicates that even after modelling an
interactive approach to conflict caused by the disruptive effect of law enforcement on
violence among criminals, the effects of structural factors are remarkably robust. In
general, the interactive model indicates that higher levels of economic development
are associated with more intense violence between DTOs. However, it is important to
notice that the coefficient of state GDP in Model 4 in Table 6.3 is significantly smaller
than the coefficient estimated using the structural model. Drug strategic territories
are also relevant for understanding criminal competition. The availability of rifles
and the vast human reserve caused by unemployment also indicate that violence is
explained by criminals’ potential for inflicting military damage and their capability

328
of recovering from it. To avoid redundancy, this section does not discuss the effects of
the structural factors in the interactive model in detail, and the reader is referred to
Section 6.2.3 where these factors are discussed in the context of the structural model.
Instead, the remainder of this section provides a more analytical discussion about
the effects of increasing law enforcement in valuable territories on the dynamics of
violent competition between DTOs. The equilibrium conditions of the theoretical
model indicate that law enforcement provides incentives for criminals to use violence
if the territory is valuable enough that the fight is worthwhile (see Section 2.3.5).
In consequence, it is plausible to expect that law enforcement will have a more dis-
turbing effect in valuable territories and generate more violence between criminal
organizations in areas where DTOs have vested interests. The statistical analysis
provides overall support for this claim. The results of Model 4 indicate that violence
between DTOs tends to concentrate around drug-producing areas, in territories fa-
vorable for the reception of drugs along the Gulf of Mexico and the Pacific coasts,
and in municipalities located along the U.S.–Mexico border (and hence favourable for
staging shipments to the U.S.) In addition, the panels in Figure 6.11 show that the
deployment of state violence in drug-valuable territories generates higher levels of vi-
olent competition between DTOs than in non-strategic areas. This indicates that the
interaction between structural factors and dynamic mechanisms of violence provide
valuable insights for understanding the escalation of violence and its geographical
concentration in particular areas.
Panel (a) in Figure 6.11 shows that violent law enforcement generates more vi-
olence between DTOs in municipalities with high levels of drug production than in
areas where marijuana and poppy crops are detected less frequently. The higher levels
of predicted law enforcement in non–drug-producing areas generates 58.4 times more
expected events of violence between DTOs, and deploying the same levels of enforce-
ment in municipalities with high production of illicit crops generates 105 times more

329
expected episodes of violence between criminals. For easier visualization, the graph
in Panel (a) does not include the margins of error, but the the predicted marginal
effects are statistically significant at each level of drug production.

(a) Drug production (b) Reception along the Gulf

120 140
Predicted competition among DTOs

Predicted competition among DTOs

100 120

100
80
times the mean

times the mean

80
60
60
40
40

20
20

0 0
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0.20
0.22

0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0.20
0.22
Predicted violent law enforcement Predicted violent law enforcement
Drug production None Low Middle High Enforcement not on Gulf Enforcement on Gulf

(c) Reception along the Pacific (d) International distribution

140 120
Predicted competition among DTOs

Predicted competition among DTOs

120 100

100
80
times the mean

times the mean

80
60
60
40
40

20 20

0 0
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0.20
0.22

0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0.20
0.22

Predicted violent law enforcement Predicted violent law enforcement

Enforcement not on Pacific Enforcement on Pacific Enforcement not on North Enforcement on North

Figure 6.11. Effect of violent law enforcement on different territories

330
 Panels (b) and (c) confirm that increasing the levels of law enforcement on territo-
ries favorable for the reception of drug shipments along the Gulf or Pacific coastlines
generate more conflict between criminals than enforcement in areas away from the
coasts. High levels of law enforcement in municipalities not located along the Gulf
experience 57.6 times more expected events of criminal violence while the same levels
of enforcement in Gulf coast municipalities generate 102.3 times more expected events
of violence between DTOs. Similarly, intense law enforcement in non-Pacific areas
generates 55.6 times more criminal competition; application of the same intensity of
violent law enforcement in municipalities located along the Pacific coast generates
115.1 times more expected events of violence between rival DTOs.
Panel (d) also provides evidence for the disruptive effect of increased state action
in municipalities favorable for the international distribution of illicit drugs across the
Mexico–U.S. border. Deploying the highest levels of state violence in municipalities
not located along the northern border generates 58.9 times more events of criminal
competition; employing the same high levels of law enforcement in the strip of mu-
nicipalities bordering the U.S. generates 95.3 times more violence between competing
criminal organizations.
In general, the empirical assessment using the interactive model provides strong
support for hypotheses H1 , H2 , H5 and H8 derived from the theoretical model. De-
mocratization and political strain are plausible sources of exogenous variation in
levels of violent and non-violent enforcement. The instrumental variable approach
shows that the efforts of Mexican authorities to fight crime triggered an unprece-
dented wave of territorial violence among rival criminal groups. Criminal violence
is particularly intense in municipalities favorable for the production of illegal drugs,
entry points suitable for the reception of shipments along the Pacific and the Gulf of
Mexico coasts, and territories along the Mexico–U.S. border.

331
6.3.5
Effect
of
Violent
and
Non-Violent
Enforcement
on
Vi olence
Among
DTOs

As discussed in section 2.4, the theoretical model provides a set of empirical im-
plications to be tested in the statistical analysis. In general, these hypotheses argue
that democratization is associated with higher levels of law enforcement (H1 ) and
periods of political strain motivate authorities to enforce the law against criminal
organizations (H2 ). According to the model, increased law enforcement is expected
to increase violence among rival criminal groups (H5 ). If the theoretical model is
right, we should see these these relationships operating with regularity across differ-
ent types of violent and non-violent enforcement tactics used by the state to fight
criminal organizations. The theoretical model also provides a more nuanced expec-
tation about the effects of political strain on violent and non-violent enforcement
due to the different costs and benefits associated with different tactics. According to
hypothesis H2.1 , increased political strain is associated with higher levels of violent
enforcement than non-violent enforcement tactics. Tables 6.4 and 6.5 present the first
and second stage results for the effects of violent and non-violent law enforcement on
the intensity of violent competition among criminal organizations.

First stage of violent and non-violent law enforcement

For comparative purposes, Model 1 in Table 6.4 reports the same results as Model
4 in the first stage of the interactive model (see Table 6.2). Models 2–5 in Table 6.4
report the effects of political variables on the employment of non-violent enforcement
tactics such as arrests of suspected members of criminal organizations, seizure of
criminal assets, interdiction of drugs, and weapons seizures.

332
 TABLE 6.4

FIRST STAGE: POLITICAL DETERMINANTS OF VIOLENT AND

NON-VIOLENT ENFORCEMENT
(1) (2) (3) (4) (5)
Enforcement Arrests Assets Drugs Guns
ENP 4.304*** 2.401*** 2.720*** 1.775*** 3.521***
(0.214) (0.036) (0.062) (0.033) (0.084)
Divided government 0.121*** 0.077*** 0.068*** 0.109*** 0.111***
(0.032) (0.009) (0.016) (0.009) (0.019)
Margin of victory -4.583*** 2.845*** 2.309*** 1.804*** 1.561***
(1.144) (0.136) (0.235) (0.131) (0.343)
Corruption -0.032*** -0.009*** -0.005*** -0.004*** -0.004***
(0.002) (0.001) (0.001) (0.001) (0.001)
Poverty -0.408*** -0.326*** -0.303*** -0.328*** -0.519***
(0.038) (0.018) (0.027) (0.014) (0.029)
State GDP (log) 0.589*** 0.299*** 0.149*** 0.200*** 0.503***
(0.049) (0.016) (0.029) (0.012) (0.032)
Population (log) 0.191*** -0.034*** 0.154*** 0.097*** 0.093***
(0.018) (0.007) (0.012) (0.006) (0.011)
Constant -31.224*** -15.416*** -16.234*** -13.762*** -26.095***
(1.103) (0.288) (0.538) (0.233) (0.619)
Observations 9,868,208 9,868,208 9,868,208 9,868,208 9,868,208
F-statistic 539*** 1,383*** 628*** 1,149*** 884***
Coefficients are in log of expected counts. Standard errors in parentheses.
The levels of significance are: * p< 0.1, ** p< 0.05, *** p< 0.01

Results from the first stage provide strong support for hypothesis H1 which states
that democratization is associated with higher levels of law enforcement. Increasing
the effective number of political parties generates a consistent increase in the vari-
ous types of violent and non-violent law enforcement efforts used for fighting crime.
Figure 6.12 compares the predicted effect of increasing the number of effective par-
ties on the levels of violent law enforcement, arrests, seizure of assets, drugs and

333
guns. For easier visualization, standard errors are omitted from the graph, but the
Table 6.4 shows that all these coefficients are highly statistically significant. The
figure shows that increasing political competition has the largest effect on seizures
of criminal assets and arrests of suspected members of criminal organizations. Vi-
olent law enforcement is the tactic third-most increased by the effective number of
parties. When compared to the other trends, the graph shows that state use of vi-
olence is highly sensitive to increasing political competition; this trend grows faster
than the other types of non-violent tactics. Increasing the number of parties also
increases weapons seizures and interdiction of illicit substances. Figure 6.12 provides
strong and consistent evidence that increasing the conditions of political competition
motivates government authorities to fight crime with a broad menu of violent and
non-violent enforcement tactics.
The first stage also shows that having a divided government has a consistent effect
of increasing law enforcement efforts across different types of tactics. Figure 6.13
shows that changing from a unified government to a divided government where either
the governor or mayor belong to a different party than the president’s is associated
with higher levels of law enforcement using all violent and non-violent tactics. As
before, the graph omits the standard errors to facilitate the visualization of trends.
It should be noted that violent law enforcement shows the largest increase when the
government shifts from a united to a divided configuration. This effect is less sharp
on non-violent tactics. In general, these results indicate that the entrance of new
political actors at different levels of government broke the unified chain of command
that facilitated the creation and maintenance of non-aggression agreements between
politicians and criminals.
The results of the first stage reveal a fascinating finding about the differential effect
of political strain on violent and non-violent law enforcement tactics. The statistical
analysis indicates that violent law enforcement is used in extraordinary circumstances

334
 35

30

Predicted state actions

25

20

15

10

0 2.9

3.0

3.0

3.0

3.1

3.1

3.2

3.2

3.2

3.3

3.3
Effective Number of Parties

Violent enforcement Arrests

Seizure of assets Seizure of drugs
Seizure of guns

Figure 6.12. Effect of political competition on violent and non-violent

enforcement

associated with periods of political crisis, whereas non-violent tactics do not follow
this pattern. Figure 6.14 shows the differing effect of the margin of victory in the
presidential election on law enforcement tactics. The figure shows that the narrower
the margin of electoral victory, the more incentives politicians have to employ violent
tactics to fight crime. In contrast, as the margin of victory increases, government
authorities rely more on non-violent tactics. This result provides empirical support
for hypothesis H2.1 and suggests that violent law enforcement is used as a last resort

to boost presidential approval rankings in periods of political strain. The distinct

effect of political strain on the decision to rely on lethal force to fight crime provides
support for the expectations suggested by Goldstein (1978) and Muller (1970) that
displaying aggressive policies generates political benefits for government authorities
as they signal strong leadership and resolve.

335
 25

20

Predicted state action

15

10

0
Unified Divided Divided
governement government 1* government 2**
Divided government
Violent enforcement Arrests
Seizure of assets Seizure of drugs
Seizure of guns
* The governor OR the Mayor are from a different party that the President's
** The governor AND the Mayor are from a different party that the President's

Figure 6.13. Effect of divided government on violent and non-violent

enforcement

Second stage of violent and non-violent law enforcement

Table 6.5 reports the estimates for the second stage, evaluating the effect of the
different law enforcement tactics on violent competition between DTOs. The results
provide strong support overall for hypothesis H5 , which claims that fighting crime
has a disruptive effect on criminals and triggers waves of violence among DTOs.
Models 1–5 in Table 6.5 show that after controlling for the endogeneity of overlapping
processes of violence, this positive effect on competition between criminal groups
holds consistently across violent and non-violent enforcement including arrests of
suspected criminals and seizures of assets, drugs and weapons.
Figure 6.15 presents the effects of violent and non-violent enforcement on violence
between DTOs. Since each enforcement tactic has its own range of predicted intensity,
the horizontal axis presents them in a standardized manner as percentiles. Results

336
 35

30

25

Predicted state action

20

15

10

0 1%

15%
3%
5%
7%
9%
11%
13%

17%
19%
21%
23%
Margin of victory

Violent enforcement Arrests

Seizure of assets Seizure of drugs
Seizure of guns

Figure 6.14. Effect of political strain on violent and non-violent

enforcement

show that all enforcement tactics exacerbate the levels of criminal competition, yet
there is a wide variation in the magnitude of their effects. Drug seizures have the most
disruptive effect on criminal violence. Rising drug interdiction to its maximum level
generates 210.9 times more expected episodes of conflict between DTOs. Weapons
seizures have the second most disruptive effect on criminal competition and generate
131.2 times more events of violence when deployed at maximum intensity. The seizure
of assets also instigates conflict between DTOs and triggers 101.4 times more events
of violence among DTOs. The graph shows that arresting criminals also ignites
conflict and generates 96.8 times more violent events between DTOs when applied at
its highest level. Finally, violent enforcement also has disrupting effects on criminal
competition but, when analyzed in comparative perspective, the magnitude of its
effect is the lowest, as it generates 62.7 events of violence among DTOs.

337
 TABLE 6.5

SECOND STAGE: VIOLENT COMPETITION AMONG DTOs

(1) (2) (3) (4) (5)
Violent enforcement 11.501***
(1.814)
Arrests 11.757***
(0.681)
Seizure of assets 15.524***
(0.833)
Seizure of drugs 24.080***
(1.112)
Seizure of guns 13.021***
(1.367)
Retaliation 0.594*** 0.594*** 0.581*** 0.585*** 0.600***
(0.174) (0.150) (0.152) (0.144) (0.157)
Drug production 0.167*** 0.185*** 0.177*** 0.177*** 0.176***
(0.057) (0.058) (0.034) (0.036) (0.056)
Gulf 0.582*** 0.584*** 0.566*** 0.571** 0.571***
(0.166) (0.193) (0.220) (0.226) (0.199)
North 0.469*** 0.497*** 0.354** 0.391*** 0.504**
(0.167) (0.173) (0.154) (0.141) (0.203)
Pacific 0.761*** 0.574*** 0.688*** 0.737*** 0.726***
(0.133) (0.103) (0.120) (0.116) (0.108)
9/11 1.085*** 1.327*** 1.168*** 1.223*** 1.088***
(0.178) (0.151) (0.124) (0.123) (0.141)
Poverty -0.197** 0.199* -0.031 0.100 -0.100
(0.083) (0.111) (0.099) (0.093) (0.108)
State GDP (log) 0.704*** 0.089** 0.467*** 0.357*** 0.588***
(0.116) (0.041) (0.024) (0.030) (0.021)
Population (log) -0.098* 0.068 -0.245*** -0.207*** -0.069
(0.052) (0.047) (0.051) (0.054) (0.050)
Rifles (100 K) 0.111*** 0.052 0.062*** 0.067*** 0.097***
(0.024) (0.035) (0.011) (0.011) (0.034)
Unemployment 0.061** 0.059** 0.065*** 0.064*** 0.064***
(0.024) (0.025) (0.018) (0.018) (0.026)
Corruption 0.008*** 0.011*** 0.001 0.001 -0.003
(0.003) (0.004) (0.003) (0.003) (0.004)
Cocaine price -0.001 0.002 0.002 0.001 -0.001
(0.002) (0.002) (0.002) (0.002) (0.002)
Drug markets 0.002 0.001 0.001 0.001 0.001
(0.002) (0.002) (0.002) (0.002) (0.002)
Divorce 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Teenage mothers -0.174 0.493 0.715 0.450 -0.259
(0.940) (1.202) (0.620) (0.586) (1.201)
Constant -19.830*** -10.618*** -14.197*** -12.800*** -17.654***
(1.754) (1.067) (0.432) (0.385) (0.959)
Observations 9,868,208 9,868,208 9,868,208 9,868,208 9,868,208
Coefficients are in log of expected counts. Bootstrapped standard errors in parentheses.
The levels of significance are: * p< 0.1, ** p< 0.05, *** p< 0.01
338
 250

Predicted competition among DTOs

200

times the mean

150

100

50

0
p10 p20 p30 p40 p50 p60 p70 p80 p90 p100
Predicted state actions (percentile)
Violent enforcement Arrests
Seizure of assets Seizure of drugs
Seizure of guns

Figure 6.15. Effect of violent and non-violent enforcement on violent

competition among DTOs

Since drug seizures have the strongest impact in generating violence between crim-
inal organizations, the discussion of mechanisms will focus on analyzing this type of
enforcement tactic. According to the theoretical model, territorial competition is the
main mechanism driving the escalation of violence. In addition, interviews conducted
during fieldwork suggest a variety of other mechanisms that could aid in understand-
ing the effect of drug interdiction on violence between DTOs. These mechanisms
are compensation, reprisal, spill-over, substitution and discipline. Some of these
mechanisms have already been discussed by Osorio (2012). Although these specific
mechanisms might operate in slightly different ways, the underlying common factor

339
in their operation is consistent with the general mechanism of territorial competition
discussed in the theoretical model.
Territorial competition. The main causal mechanism suggested by the theoretical
model is that law enforcement may trigger an invasion from a challenger DTO into
the territory of the target criminal group affected by law enforcement. Drug seizures
could generate at least three different signs of the target DTOs diminished capability
to defend its territory. First, a series of government crackdowns signals that the target
DTO is no longer protected under corrupt agreements with government authorities,
thus increasing its vulnerability to raids from rival cartels. Second, the fact that the
target DTO has experienced a crackdown by the state and lost some of its goods
indicates that the criminal group was not able to protect its assets. This can signal
weakness or incompetence on the part of the internal branch in charge of providing
security to the organization, which can motivate their rivals to launch an invasion.
Third, if the target DTO suffered a large drug seizure or a significant number of
smaller seizures, the loss of merchandise could decrease the income of the criminal
organization. The loss of income might undermine the capability of the target DTO
to adequately carry out its activities, including provision of its own security.
Compensation. Since drugs provide the most important source of income for
DTOs, drug seizures might motivate the targeted DTO to launch an offensive against
its rivals in order to seize their drugs. The objective is to compensate for their loss
by raiding their competitors. DTOs make considerable investments of money and
reputation with international partners in producing and destination countries, which
might try to compensate for the loss of the seized cargo by capturing their rival’s
drugs so they can fulfill their part of the deal. A lost cargo might not only imply
losing the income expected from the seized drugs, but can also affect other business
that the target DTO has with its partners or even generate violent reactions from

340
disappointed associates. In addition, raiding their competitors might generate violent
retaliation as rivals try to protect their cargo and storage facilities.
Reprisal. Drug interdiction enacted on a target DTO might be caused by a rival
informing authorities about the opportunity of seizing drugs. This mechanism can
be a type of selective law enforcement in which a criminal group might cooperate
with government authorities to inform them about the operations of rivals criminal
organizations. Cooperation of this type might be based on corrupt agreements but
also extracted from interrogations of arrested criminals trying to reduce their time in
prison. In cases where information provided by a rival DTO caused a drug cargo to
be interdicted, the targeted DTO might launch an offensive against the rival group
who turned them in.
Spill-over. Dell (2011) argues that enforcement generates a spill-over effect of
violence. Increased law enforcement in some areas motivates DTOs to use different
drug-trafficking routes to avoid these operations. Changing routes generates a diffu-
sion of violence among criminal groups to areas not previously affected by violence.
Substitution. Another mechanism potentially employed by DTOs is instead of
expanding their territorial area of operations, reacting to drug seizures by expanding
their activities to other types of illicit markets. Some DTOs might use their com-
parative advantage as specialists in violence to engage in predatory activities such
as racketeering and kidnapping. Revenues from extortion and ransom might thus
help compensate for the loss of income incurred by the seizure of cargo. In addition,
engaging in violence-intensive activities such as racketeering and kidnapping sends a
clear signal about the aggressiveness of the criminal group that might help to keep
their rivals at bay.
Discipline. Drug interdiction can also generate violence within a targeted criminal
organization. In some cases, leaders of criminal organizations might use violence to
sanction their lower ranks as punishment for losing a valuable cargo. This disciplinary

341
mechanism also serves the purpose of reducing principal-agent problems when leaders
declare that they will not tolerate “errors” from their lower ranks. Ensuring discipline
is crucial for leaders because lower ranks can easily claim that drugs were seized, when
they actually were not, and attempt to sell the cargo independently for their private
gain. By using disciplinary violence, the leader sends a signal to the organization
itself as well as to other groups about his ability and willingness to use violence.

6.3.6
Overall
Model
Fit

As discussed in Section 6.2.4, the structural model has limited explanatory power
to account for the substantial variation in violence between DTOs. The goodness
of fit assessment showed that structural variables often used for explaining criminal
behavior and violence are so large and slow-moving that they cannot account for the
rapid variation in criminal violence across time and space. In fact, Figure 6.6 showed
that the structural model does a poor job in predicting violence between DTOs. In
contrast, the interactive model of large-scale organized criminal violence presented in
this research has stronger explanatory power than the structural model.
Figure 6.16 presents the scatter plot between the number of observed and pre-
dicted events of violent competition between DTOs as estimated by the second stage
of the interactive model. This figure shows that the interactive model is better at
explaining the variation in violent competition between criminal organizations.
Since the full database contains more than 9.8 million observations, the scatter
plots of the observed and predicted events of violent competition generated by the
structural model (Figure 6.6) and the interactive model (Figure 6.16) are somewhat
limited for visualization purposes. The limitation of the scatter plots is due to the
difficulty of depicting the concentration of a large number of observations around
particular values. This is especially relevant for this research because the dependent
variable is distributed as a negative binomial with a high concentration of data points

342
 Figure 6.16. Goodness of fit of the interactive model

around zero (see Section 4.3.1). Figure 6.17 overcomes the limitations of the scatter
plots by presenting heat-maps for the concentration of observed and predicted out-
comes of the structural and interactive models. Darker areas in the panels indicate
a higher concentration of observations around those values. As presented in the left
panel of the figure, the structural model does a poor job of explaining the variation
in violent competition between criminals. The predicted values generated by this
model are heavily concentrated at the lower-left edge of the graph. In contrast, the
right panel shows that the interactive model is substantially better at explaining the
variation of violence between criminal groups. As expected, most of the predicted
and observed events tend to cluster around zero. In addition, the gray areas of di-
minishing intensity indicate a broader variation between the predicted and observed
levels of violent competition among criminal groups. In consequence, the empirical
assessment indicates that the interactive explanation of violent competition between

343
criminal organizations is more useful than an explanation based exclusively on struc-
tural factors.

Structural model Interactive model

85.573

85.573
64.18

64.18
Competition (predicted structural)

Competition (predicted dynamic)

42.787

21.394 42.787
21.394

.00071
1.5

0 9 18 27 36 0 9 18 27 36
Competition (observed) Competition (observed)

Figure 6.17. Comparative model fit between structural and dynamic models

6.4
Conclusions

This chapter provides the main empirical support for the predictions derived from
the theoretical model. In a context where government authorities coexist with pow-
erful criminal organizations, democratization erodes preexisting agreements between
politicians and criminals, and periods of political strain can further motivate author-
ities to fight organized crime in an effort to gain citizen support. These increased
efforts to enforce the law have a strong disruptive effect on the relative balance among

344
criminal organizations and trigger waves of violent competition among rival criminal
groups fighting to control strategic territories.
The research design advances an interactive model of violent competition that
overcomes the limitations of structural explanations of violence. However, the use of
an interactive model of conflict carries the risk of finding spurious correlations due to
the interaction between law enforcement and criminal violence, potentially generating
reciprocal causation. The identification strategy used in this research relies on the
use of instrumental variables, a quasi-experimental methodological strategy capable
of overcoming the endogenous relationship among overlapping processes of conflict.
The empirical analysis reveals that the entrance of new parties to the political
scene is associated with increased levels of enforcement across a broad menu of vio-
lent and non-violent law enforcement tactics. This may be indicative of how political
competition motivates government authorities to provide public goods, including pub-
lic security. The results also indicate that divided governments are associated with
increased efforts to fight crime. This finding suggests that the entrance of new po-
litical actors across different levels of government disrupted the long-standing chain
of command that enabled a peaceful, stable coexistence between the state and crim-
inal organizations. In addition, the results indicate that narrow margins of election
victory are related to the intensification of violent, but not of non-violent enforce-
ment. This finding is consistent with the notion that political strain may motivate
politicians to implement aggressive security policies under exceptional circumstances.
The results show that the effect of increased law enforcement caused by the ex-
ogenous variation of democratization and political strain is to increase levels of vi-
olence between criminal groups. This effect holds for both violent enforcement and
non-violent tactics such as arrests and seizures of assets, drugs and weapons. The
comparative analysis of the effect of different tactics reveals that drug interdiction has
the most disrupting effect on violent competition between DTOs. Witness accounts

345
collected from interviews conducted during fieldwork suggest a variety of mechanisms
involved in the relationship between drug seizures and criminal violence. These mech-
anisms are consistent with the main theoretical expectation of territorial competition
among criminal organizations.
The statistical analysis supports the argument about the centrality of territorial
values to explain violence between criminal groups. The results indicate that criminal
violence tends to concentrate in territories favorable for the production of illegal drug
crops, entry points along the Gulf and Pacific coasts, and international distribution
spots on the U.S.–Mexico border. The analysis also shows that the deployment of
law enforcement in these strategic areas dramatically exacerbates levels of violent
competition between criminal organizations.
The research design of the interactive model using instrumental variables is not
only able to overcome the problems of endogeneity between enforcement and criminal
violence, but is also consistent with the data generation process described in the
theoretical explanation, thus providing an application of the EITM approach capable
of aligning the ontology of the formal model with the methodology used for testing
its empirical implications.
The interactive approach of the empirical strategy helps provide a better under-
standing of the substantial variation in the micro-dynamics of violence in a way that
structural explanations are not capable of doing. Despite of its limited analytical
leverage, the structural model furnishes some findings that challenge the dominant
theories of violence. One of the most striking findings related to structural factors is
that violence between organized criminal groupss is strongly associated with higher
levels of economic development. This finding is consistent across a variety of model
specifications of the structural and the interactive models, and contradicts a widely
held expectation among conflict scholars and criminologists that poverty is one of
the most important determinants of political violence and criminal behavior. This

346
finding, in contrast, is consistent with the theory advanced in this research about
the importance of valuable territories for understanding violent competition among
DTOs. In any case, the positive relationship between economic development and vi-
olence suggests the need to further analyze the conceptual distinction between rebels
and organized criminals.

347
 CHAPTER
7

DYNAMICS
OF
VIOLENCE

IN
THE
MEXICAN
WAR
ON
DRUGS

7.1
Introduction

The most vigorous debate in the Mexican war on drugs revolves around the effi-
cacy of the punitive strategy implemented by government authorities to fight crim-
inal organizations. On one side of the debate, Poiré and Martı́nez (2011) support
the official discourse, arguing that law enforcement does not increase the levels of
drug-related violence. To support their claim, these authors provide evidence of a
single case in which the Mexican Army killed Ignacio “Nacho” Coronel Villarreal,
a prominent leader of the Sinaloa Cartel, in 2010. According to their account, the
general trend of drug-related homicides did not increase after the Army shot Coronel.
On the same side of the debate, Villalobos (2010, 2012) offered an eloquent defense
of the government strategy but with no systematic evidence to support his claim.
On the other side of the debate, Guerrero (2011b) questioned the methodological ap-
proach used by Poiré and Martı́nez to conclude, based on a single data point, that the
government strategy did not increase violence. Instead, Guerrero relies on 28 cases

348
in which government authorities captured or killed prominent drug lords to argue
that the punitive strategy of the government triggers spirals of increasing violence
between criminal groups. Other analysts have added their voices to the criticism
of the full-fledged military campaign launched by President Calderón against crimi-
nal organizations, citing its deleterious consequences (Castañeda and Aguilar, 2010;
Domı́nguez Ruvalcaba, 2010; Escalante Gonzalbo, 2009, 2011; Escalante Gonzalbo
et al., 2011; Guerrero, 2009a, 2010a,b, 2011a; Merino, 2011).
Despite these keen insights, the empirical support for either side of the debate
is problematic. The arguments focus primarily on the actions of the state towards
criminal organizations but fail to consider the reactions of DTOs against the state
and the actions rival criminal groups undertake towards each other. This unilateral
approach largely ignores the dynamic and interactive characteristics of conflict. As
indicated in Section 4.2 of Chapter 4, this unilateral approach is caused in part by
the use of aggregated data focused exclusively on homicides, and provides no infor-
mation about other interactions between the perpetrators and victims of violence.
Another empirical problem of this debate is the narrow focus on analyzing a single
law enforcement tactic (usually the targeting of drug leaders) without considering the
broader menu of actions implemented by the state to fight crime. Not considering
the full set of law enforcement efforts incurs a problem of omitted variable bias that
is likely to generate misleading results.
However, the limitations of this debate are not only empirical and methodolog-
ical. The tendency to study violence by paying narrow attention to a single actor
and focusing on a reduced menu of actions reflects a pervasive theoretical limitation
characteristic of research on conflict. Although the ontology of conflict is often de-
picted as the opposing actions or interests of two or more antagonistic actors, most
theoretical developments and empirical assessments in conflict research tend to focus
on only one actor without considering their opponents. In addition, most research

349
on conflict is focused on a narrow repertoire of contentious or repressive tactics, but
fails to consider broader tactical alternatives. For example, those studying social
movements, terrorism, or insurgencies tend to focus exclusively on protestors, ter-
rorists or rebels, respectively, without taking into account the interactions between
their subjects of study and the state. Similarly, those studying state repression and
counter-insurgency tend to focus on the actions of government authorities without
considering the interplay of their dynamic interaction with protesters or insurgents.
Due to the theoretical complexities and the methodological sophistication required to
study conflict from a dynamic, interactive perspective, there are few theoretical and
empirical efforts that adopt this interactive approach. Perhaps the area of research
that has advanced the farthest in analyzing the dynamic and interactive characteris-
tics of conflict is the literature on the repression–dissent nexus (see McAdam, Tarrow
and Tilly, 2001). In this emerging area of research, Davenport, Johnston and Mueller
(2005) proposed the “Maryland model” of coercion and mobilization, arguing that
the study of conflict requires taking into account the interactions between the state
and challengers as they evolve over time and employ a diverse variety of tactics.
This chapter adopts the main insights of the “Maryland model” to assess the
empirical implications derived from the theoretical model presented in Chapter 2,
enabling the empirical analysis described in this section to provide an interactive,
dynamic, multimodal explanation of violence in the Mexican war on drugs. The in-
teractive nature of the method consists in analyzing the actions and reactions between
the state and criminal organizations and between rival criminal groups. The dynamic
character consists in explicitly incorporating the temporal evolution of individual and
joint processes of violence into the empirical assessment. The multimodal character
is based on simultaneous analysis of a variety of violent and non-violent tactics used
by the state to fight criminal organizations.

350
 Consistent with the theoretical expectations, the results show that a sustained
campaign of violent law enforcement generates a substantial escalation of conflict
between rival criminal organizations. In addition, the use of force to fight crime
triggers a wave of retaliatory attacks perpetrated by criminal organizations against
government authorities. The dynamic analysis also shows that sustained campaigns
of violence between DTOs generate a violent response from government authorities,
but this effect is moderate. The results provide, rather, the surprising finding that
direct criminal attacks against the state do not seem to generate a coercive response
from government authorities to counter or neutralize criminal hostilities.
The material presented in this chapter to support these claims is divided into
four sections. The first part presents the main hypotheses to be tested and briefly
discusses the research design based on methodologies of time series analysis. The
second section identifies the specific characteristics of each time series of violence.
The third segment presents the results of the dynamic and interactive analysis of the
different processes of violence inherent to the Mexican war on drugs. Finally, the
fourth section summarizes the findings and discusses the results.

7.2
Hypotheses
and
Research
Design

The literature review in the theoretical chapter discussed one of the most robust
findings in the research on political repression, known as the “law of coercive re-
sponsiveness” (Davenport, 2007). It is broadly recognized in this body of literature
that government authorities generally respond with the use of repressive behavior
to counter or neutralize dissenters attempting to subvert the political or economic
status quo. In other words, the state regularly employs force to protect itself. In
addition, theories on state formation largely agree that the state holds the legitimate
right to use violence to suppress any kind of behavior that could threaten the life and

351
property of the population living in its territory (Hobbes, 1651; Olson, 2000; Tilly,
1985; Weber, 1978). According to this perspective, the state uses force to protect its
citizens. These relationships can be expressed in terms of the following hypotheses,
which are useful for analyzing the behavior of government authorities in the context
of the Mexican war on drugs:

(H12.1 ) Criminal retaliation against government authorities is associated with

higher levels of law enforcement.

(H12.2 ) Violent competition between rival criminal organizations is associated

with higher levels of law enforcement.

In addition, Section 2.4 discussed the direct empirical implications derived from
the formal model of drug violence. According to the theoretical explanation, law
enforcement is expected to generate violent reactions from criminal groups against
the state as well as violence between rival criminal organizations. These arguments
are stated in the following hypotheses:

(H4 ) Increased law enforcement is positively associated with higher levels of

criminal retaliation against the state.

(H5 ) Increased law enforcement is positively associated with higher levels of

violent competition among criminal organizations.

To test these hypotheses, the empirical strategy relies on time series analysis of
daily data aggregated at the national level. The processes analyzed in this chapter
include violent enforcement used by the state to fight DTOs, competition among
rival criminal organizations, retaliation attacks perpetrated by criminals against gov-
ernment authorities, as well as non-violent law enforcement tactics such as arrests
of suspected members of criminal organizations, seizures of assets, seizures of drugs
and seizures of guns. The characteristics of these variables are discussed in Chapter
4 and their descriptive statistics aggregated at the national level on a daily basis can
be found in Appendix A.7.

352
 The empirical analysis consists of three sections. The first relies on autoregressive
integrated moving average (ARIMA) models to identify and estimate the appropriate
autocorrelation process generating the outcome in each individual time series. These
ARIMA models serve to generate reliable predictions about the dynamics of each
time series. The second and third sections use vector autoregressive (VAR) models
for analyzing the dynamic and reciprocal interactions among the different time series.
In particular, the second section applies a baseline VAR model to the interactions
between violent enforcement, competition between rival DTOs and criminal retalia-
tion against the state. The third section analyzes a fully specified VAR model that
includes the interactions between the violent and non-violent law enforcement tactics
used by the state to fight crime, as well as the reaction by criminal groups against
the state and the dynamics of competition between rival criminal organizations.

7.3
Time
Series
of
Organized
Criminal
Violence

The model identification of each time series process is based on the procedure
proposed by Box and Jenkins (1976), consisting of three stages: (i) identification,
(ii) estimation and (iii) diagnosis. For brevity, this section mainly discusses the
three stages for the time series of violent competition between criminal organizations.
However, each subsection presents the general results for all the other time series of
criminal retaliation, and violent and non-violent law enforcement.

7.3.1
Time
Series
Identification

There are three components of any time series process: trend, cyclical charac-
teristics and random variation. The first two are deterministic and can be removed
or filtered in order to reach stationarity, a crucial concept in time series analysis.
According to Enders (2009), a time series is stationary if its mean, variance and

353
autocorrelations are time invariant. Even if we do not know the data generation
process that produced the outcome, a stationary time series allows the use of infor-
mation from previous events – such as the mean, variance and stochastic error – to
predict the outcome in the future or changes in the outcome that would be caused
by external shocks. This is possible because the mean, variance and autocorrelation
of a stationary time series are stable over time. What happened in the past can be
thus informative of what will happen in the future in stationary series. A time series
is said to have unit root if it has the characteristics of a non-stationary process.
The first step in determining whether a time series is stationary is to make a
basic visual assessment. Figure 7.1 presents the time series of daily data of violent
competition between DTOs aggregated at the national level. A first examination of
the graph suggests that the series might be non-stationary because of the positive
trend and increasing variance towards the end of the series. However, the compression
of eleven years of daily data into a single graph makes it hard to assess the stationarity
of the series visually.
A more rigorous strategy for identifying the stationarity of the time series is to
conduct an Augmented Dickey-Fuller (ADF) test. The null hypothesis of the ADF
test is that the series is non-stationary. In consequence, if the statistic produced by
the test is large (in absolute terms) and statistically significant, we can reject the null
hypothesis and say that the series is stationary. The result of the augmented Dickey-
Fuller test indicates that the time series of violent competition between criminal
organizations follows a stationary process, but is influenced by a positive trend. The
main estimate of the ADF reports a Z(t) statistic of -46.909 with a MacKinnon p-
value of 0.0000, thus rejecting the null hypothesis of a unit root. In addition, the
Z(t) statistic is well below the critical value of -3.960. The lag coefficient of the ADF
is -0.71 and statistically significant. In addition, the trend component is positive and
significant, thus indicating the presence of a deterministic trend in the process.

354
 200
Number of daily events of competiton among DTOs
0 50 100 150

01jan2000 01jan2002 01jan2004 01jan2006 01jan2008 01jan2010

Time

Figure 7.1. Daily time series of violent competition among DTOs at the
national level

The augmented Dickey-Fuller test indicates that the time series of violence be-
tween DTOs is already stationary. Therefore it is not necessary to apply the difference
transformation to the time series to make the process mean stationary nor to apply
a logarithm transformation to generate variance stationarity. Differencing and log-
transforming the data are the usual strategies for filtering deterministic components
in non-stationary data. However, it is not necessary to apply such procedures to
this time series since it has no unit root. Attempting to difference or log-transform
the data would generate a largely negative lag coefficient of -1.42, thus indicating
problems of over-differencing in the time series.
Further diagnostics indicate that the time series of violence between DTOs is
highly correlated over time. The Autocorrelation Function (AC) in panel (a) of
Figure 7.2 shows that the AC decays slowly, thus suggesting an auto-regressive (AR)
process of temporal autocorrelation. In addition, the slow decay of the spikes in the
AC reveal a longterm autocorrelation process This long temporal influence is likely to

355
be caused by the use of daily data. In addition, the Partial Autocorrelation Function
(PAC) in panel (b) of Figure 7.2 confirms the AR process. The large number of spikes
above the 95 percent confidence intervals in the PAC function indicates that the
autocorrelation in the time series goes back up to a month. In more intuitive terms,
these diagnostics show that the events of violence between criminal organizations
that occurred in the previous day strongly influence the number of events occurring
in the current day and this temporal influence persists for approximately a month.

(a) AC function (b) PAC function

0.60

0.60
Partial autocorrelations of Competition
Autocorrelations of Competition
0.40

0.40
0.20

0.20
0.00

0.00
-0.20

0 10 20 30 40 0 10 20 30 40
Lag Lag
Bartlett's formula for MA(q) 95% confidence bands 95% Confidence bands [se = 1/sqrt(n)]

Figure 7.2. Autocorrelation (AC) and Partial Autocorrelation (PAC) of

violent competition among DTOs

Table 7.1 presents the results of the augmented Dickey-Fuller tests for all the
processes of conflict inherent to the Mexican war on drugs, including violent compe-
tition between DTOs, violent enforcement, criminal retaliation, arrests of organized
criminals, seizures of assets, events of drug interdiction, and seizures of weapons. The
analysis of the ADF tests indicate that all the time series are stationary processes

356
with a high degree of autocorrelation affected by positive trends. The AC and PAC
graphs of all the different time series are not reported here, but they present similar
characteristics of slow decay and longterm autocorrelation effects to those shown in
Figure 7.2, thus suggesting the presence of auto–regressive processes.

TABLE 7.1

AUGMENTED DICKEY-FULLER TEST FOR DIFFERENT

PROCESSES OF VIOLENCE
Time series Z(t) MacKinnon Critical Lag coeffi- Trend co-
statistic p-value value cient efficient
Competition -46.909 0.0000 -3.96 -0.71*** 0.007***
Violent enforcement -48.777 0.0000 -3.96 -0.74*** 0.001***
Retaliation -50.724 0.0000 -3.96 -0.78*** 0.002***
Arrests -48.134 0.0000 -3.96 -0.73*** 0.004***
Assets -52.452 0.0000 -3.96 -0.81*** 0.002***
Drugs -44.895 0.0000 -3.96 -0.67*** 0.004***
Guns -50.823 0.0000 -3.96 -0.78*** 0.002***

7.3.2
Time
Series
Estimation
and
Diagnosis

After the identification of the time series as a stationary process, the second
stage consists of estimating the structure of the autocorrelation using Autoregressive
Integrated Moving Average (ARIMA) models. These models are useful for predicting
future values of a time series by a linear combination of its past values and a series
of error terms. In order to generate accurate predictions it is necessary to identify
the order of temporal autocorrelations in the data realization process. The order

357
indicates the number of lags to be included in the estimation of the time series. After
assessing different model specifications, it was found that the time series of violence
between DTOs is best described by an autoregressive (AR) model of the 37st order,
represented by the acronym AR(37). This means that events of violence that took
place 37 days in the past still have an effect on the current levels of conflict between
DTOs. The assessment of different model specifications indicate that there is no need
to include moving-average processes or seasonal components.
The key diagnostic of an ARIMA model an analysis of the residuals generated by
the time series process, which in this case is an AR(37) model. If the specification of
the model is correct, the model should generate no autocorrelation in the residuals,
a characteristic known as white noise. If autocorrelation is absent from the error
terms, the residuals should be randomly distributed around zero. This indicates that
the disturbances are stochastic and this white noise does not affect the prediction
of future values. In contrast, if the model does not include the correct specification
of autoregressive or moving average terms with their respective order of lags, the
model will not show white noise, and will not be able to forecast accurately based
on the observed data. One strategy for detecting white noise is to visually inspect
the correlograms of the residuals. Panels (a) and (b) in Figure 7.3 present the AC
and PAC functions of the residuals generated after estimating the ARIMA AR(37)
model. Both panels show that the model generates white noise in the residuals. With
the exception of lag 40, there are no spikes outside the confidence intervals of either
the AC or PAC correlograms. Even the spike at day 40 is so small and so far back
in time that it is not likely to affect the model.
An alternative way to assess the level of autocorrelation in the residuals is to
consider a Portmanteau test for white noise. The null hypothesis of this test is
that there is no serial correlation. If the test produces a large statistic with high
levels of statistical significance, the null hypothesis is rejected, and autocorrelation

358
 (a) AC of residuals (b) PAC of residuals

0.04
0.04

Partial autocorrelations of residuals

Autocorrelations of residuals

0.02
0.02

0.00
0.00

-0.02
-0.02 -0.04

-0.04
0 10 20 30 40 0 10 20 30 40
Lag Lag
Bartlett's formula for MA(q) 95% confidence bands 95% Confidence bands [se = 1/sqrt(n)]

Figure 7.3. Autocorrelation (AC) and Partial Autocorrelation (PAC)

functions of residuals from the ARIMA AR(37) model

is suspected. In the case of an AR(37) model for the time series of violence between
DTOs, the Portmanteau test yields a Q statistic of 22.6 with a p-value of 0.9879,
thus failing to reject the null hypothesis. In consequence, it can be concluded that
there is white noise in the residuals and that this model specification is reliable for
generating accurate predictions about violence between rival criminal groups.
Instead of presenting the AC and PAC graphs of residuals for all the other time
series, Table 7.2 reports the Q statistics and their corresponding p-values of the
portmanteau test for white noise for all the other processes of violence. Based on
the proposed model specifications, the test fails to reject the null hypothesis, thus
suggesting that there is no serial correlation in the error terms. In consequence,
these ARIMA models can be used for accurate forecasting for each process of violence
inherent to the Mexican war on drugs.

359
 TABLE 7.2

PORTMANTEAU TEST FOR WHITE NOISE IN DIFFERENT

PROCESSES OF VIOLENCE
Time series ARIMA model Order (Q) statistic p-value
Competition AR 37 22.6 0.988
Violent enforcement AR 31 33.92 0.74
Retaliation AR 28 49.9 0.135
Arrests AR 21 50.25 0.129
Assets AR 22 31.09 0.84
Drugs AR 28 32.14 0.999
Guns AR 28 24.37 0.81

7.3.3
Model
Specification
of
Time
Series
Processes

Based on the time series identification procedure proposed by Box and Jenk-
ins (1976), the following set of equations describes the time series processes for vi-
olence between criminal groups, violent law enforcement, criminal retaliation and
non-violent enforcement tactics such as arrests, seizure of assets, drug interdiction
and seizures of weapons. Notice that these equations are presented in terms of au-
toregressive (AR) processes represented by including a series of lags of the outcome
variable. The identification and estimation strategy discussed above does not suggest
the need to include moving average (MA) processes in the residuals nor seasonal com-
ponents. These ARIMA models will be used in the next section to analyze the conflict
interactions between the state and criminal organizations and violence between rival
criminal organizations.
Equation 7.1 represents the time series of violent competition among DTOs:

Ct = α1 + β1 Ct−1 + β2 Ct−2 + · · · + βk Ct−k + u1t (7.1)

360
where Ct represents the number of events of violent competition between criminal
groups at time t, scalar α1 is the intercept, Ct−i represents the lags of violence between
DTOs for each day i ∈ [1 − −k], parameter βi represents the autoregressive effect
of each lag in time t − i and u1t represents the current error terms. The ARIMA
model of violent competition between criminal groups uses k=37, thus including an
autoregressive process of lags for 37 days.
Equation 7.2 represents the time series of violent enforcement from the state
against DTOs:

Et = α2 + θ1 Et−1 + θ2 Et−2 + · · · + θk Et−k + u2t (7.2)

where Et represents the daily number of events of violent law enforcement, scalar
α2 is the intercept of this time series, Et−i represents the lags of violent enforcement
back to k days, coefficient θi accounts for the lagged effect of the data in time t−i and
u2t captures the error terms. The ARIMA model of violent enforcement conducted
by the state against criminal organizations uses an autoregressive process of 31 lags.
Equation 7.3 represents the time series of criminal retaliation against the state:

Rt = α3 + Ω1 Rt−1 + Ω2 Rt−2 + · · · + Ωk Rt−k + u3t (7.3)

where Rt represents the number of events of violence against the state perpetrated by
criminal organizations, α3 is the intercept, Rt−i represents the events of retaliation in
previous days, parameter Ωi is the coefficient of lagged observations in time t − i and
u3t represents the error terms. The ARIMA model of criminal retaliation against the
state uses an autoregressive order of 28 days.
Equation 7.4 represents the time series of arrests:

At = α4 + δ1 At−1 + δ2 At−2 + · · · + δk At−k + u4t (7.4)

361
where At represents the number of daily arrests, α4 is the intercept, At−i represents
the number of arrests in the past, parameter δi represents the autoregressive effect of
lags in t − i and u4t corresponds to the error terms. The ARIMA model of the time
series of arrests of suspected members of criminal organizations uses an autoregressive
process of the 21st order.
Equation 7.5 represents the time series of seizures of criminal assets:

St = α5 + σ1 St−1 + σ2 St−2 + · · · + σk St−k + u5t (7.5)

where St represents the number of seizures of criminal assets, α5 corresponds to the

intercept, St−i are the number of seizures conducted in previous days, coefficient σi
is the coefficient of the lag at time t − i and u5t represents the error. The ARIMA
model of the number of events where authorities confiscate criminal assets uses an
order of 22 lags in the autoregressive process.
Equation 7.6 represents the time series of seizures of criminal assets:

Dt = α6 + μ1 Dt−1 + μ2 Dt−2 + · · · + μk Dt−k + u6t (7.6)

where Dt corresponds to the daily number of drug seizures, scalar α6 represents

the intercept, Dt−i indicates the previous events of drug seizures back to k days, μi
represents the effect of lags in t − i and u6t is the error term of drug interdiction. The
ARIMA model of the time series of drug interdiction considers autoregressive order
of 28 days.
Lastly, equation 7.7 represents the time series of the number of daily gun seizures:

Gt = α7 + ρ1 Gt−1 + ρ2 Gt−2 + · · · + ρk Gt−k + u7t (7.7)

362
where Gt is the number of events in which the state confiscates weapons to criminal
organizations, α7 is the intercept, Rt−i represents the number of seizures of guns that
took place in the k previous days, coefficients ρi account for the the autoregressive
effect of lagged observations in time t − i and u7t corresponds to the error terms. The
ARIMA model of gun seizures considers an autoregressive process of 28 lags.

7.4
Dynamic
Analysis
of
Drug-Related
Violence

7.4.1
Base-Line
Vector
Autoregressive
Model
for
the
Dynamics

of
Drug
Violence

The ARIMA models identified in the previous section are useful for generating
accurate forecasts for each individual time series. In addition, these models can also
be used as building blocks for assessing the simultaneous and dynamic interactions
among the different processes of conflict as they evolve over time. These types of dy-
namic, endogenous relationships can be analyzed with vector autoregressive (VAR)
models. In their seminal work, Freeman, Williams and Lin (1989) introduced the
use of VAR models in political science as a way of developing an empirical strat-
egy that could better approximate the complexity and endogeneity of theoretical
explanations. VAR models are particularly useful for testing hypotheses in dynamic
analysis because they require fewer theoretical assumptions and empirical restrictions
than traditional structural equation models (SEM). Due to their flexibility and an-
alytical leverage, VAR models have been used to analyze the dynamic interactions
in conflict processes including anti-terrorist strategies (Enders and Sandler, 1993),
strategic behavior at the intersection of domestic and international conflicts (Moore,
1995), and interactions between the state and dissidents in a variety of domestic
conflicts (Shellman, Hatfield and Mills, 2010; Shellman, 2004).

363
 Instead of predicting the outcome of a time series from the effect of exogenous
variables, VAR analysis treats the interactions between time series as endogenous.
Following Freeman, Williams and Lin (1989, 845), VAR models regress each of the
variables in their system on the past lags of those variables and on the past lags of
all the remaining variables.
The basic intuition behind VAR models is illustrated in Figure 7.4 depicting the
temporal evolution and reciprocal interactions among the time series of violent com-
petition between DTOs, violent enforcement from the state, and criminal retaliation
against government authorities. The first set of components in the figure refer to the
time series of competition between criminal groups, the second set to the time series
of violent enforcement and the third set to criminal retaliation against the state. Solid
arrows depict the effect of previous events of each series on its own outcomes, and
dashed arrows show the reciprocal interactions between different time series. Solid
arrows thus actually represent a combined effect of both the occurrence of previous
events and variation caused by the interaction with other time series in the past.
The example in this figure includes only two lags. As the diagram implies, the use
of VAR analysis enables prediction of the outcome of violent competition between
DTOs at time t caused by previous events of violence between criminal groups at
times t − 1 and t − 2, while also incorporating the simultaneous effect on competition
caused by current and previous shocks of law enforcement and events of criminal
retaliation. The other time series are simultaneously estimated based on their own
temporal inertia and the effects of other processes of violence affecting them.
The vector autoregressive model shown in Figure 7.4 can be useful for testing
the central hypothesis related to the dynamic interaction between the state and
criminal organizations and between rival criminal groups. According to the law of
coercive responsiveness, the state is likely to engage in repressive behavior if directly
challenged by criminal organizations (H12.1 ). In addition, the state is likely to use

364
 Competition t Competition t-1 Competition t-2

Enforcement t Enforcement t-1 Enforcement t-2

Retaliation t Retaliation t-1 Retaliation t-2

Figure 7.4. General intuition of Vector Autoregressive (VAR) models

force to impose order if rival criminal groups are fighting each other (H12.2 ). Based
on the expectations of the theoretical model, state violence is also likely to trigger
violent reactions from criminal organizations against government authorities (H4 )
and between rival criminal groups (H5 ). These dynamic and interactive relationships
can be simultaneously tested in VAR model 1 comprising the following system of
equations:

Ct = α1 + β11 Ct−1 + β12 Ct−2 + · · · + β1k Ct−k

+ θ10 Et + θ11 Et−1 + θ12 Et−2 + · · · + θ1k Et−k (7.8)
+ Ω10 Rt + Ω11 Rt−1 + Ω12 Rt−2 + · · · + Ω1k Rt−k + u1t

Et = α2 + θ21 Et−1 + θ22 Et−2 + · · · + θ2k Et−k

+ β20 Ct + β21 Ct−1 + β22 Ct−2 + · · · + β2k Ct−k (7.9)
+ Ω20 Rt + Ω21 Rt−1 + Ω22 Rt−2 + · · · + Ω2k Rt−k + u2t

Rt = α3 + Ω31 Rt−1 + Ω32 Rt−2 + · · · + Ω3k Rt−k

+ θ30 Et + θ31 Et−1 + θ32 Et−2 + · · · + θ3k Et−k (7.10)
+ β30 Ct + β31 Ct−1 + β32 Ct−2 + · · · + β3k Ct−k + u3t

365
 Equation (7.8) in VAR model 1 models the process of violent competition be-
tween rival criminal groups, denoted by Ct , which is generated by previous events of
violence between DTOs (represented by the parameters in the first row, the current
and past levels of violent law enforcement (represented by the second row) and the
current and previous levels of criminal retaliation (represented in the third row of
the equation). The outcomes from the time series of violent competition between
DTOs thus incorporates both internal dynamics of violence between criminal groups
over time and the interactive effect of state violence against criminals and violent
retaliation from criminal groups against government authorities.
Equation (7.9) models the time series of violent law enforcement, Et , which is
affected by past events of state violence (parameters in the first row), current and
past events of violence between criminal organizations (second row of the equation)
and criminal attacks against the state (third row).
Equation (7.10) models the time series process of criminal retaliation, Rt , which

is influenced by previous criminal attacks against government authorities (in the first
row of the equation), current and past levels of violent law enforcement (second row)
and current and past events of violent competition among criminal organizations
(represented by the third row of the equation).
This vector autoregressive model for the dynamic interaction between the state
and DTOs and between rival criminal groups employs an autoregressive process of 37
lags. The order of serial correlation is based on both the estimation of the ARIMA
model in equation (7.1) capable of generating white noise and the analysis of the
lag-order selection statistic computed in Stata by command varsoc. In consequence,
equations (7.8), (7.9) and (7.10), making up the system of equations in VAR model
1, are estimated with lags of 37 days. Finally, this base–line model specification
includes no exogenous variables.

366
 The particular regression results of each of the 118 parameters in VAR model 1
are not of specific interest of this research. For this reason the table of results is
not reported. Table 7.3 shows that the model provides a good fit for the data, with
R2 ranging from 32.3 percent to 50.1 percent. However, it is important to note that
VAR analysis usually yields high R2 values due to the high degree of autocorrelation
between the lagged terms.

TABLE 7.3

EXPLANATORY POWER OF VAR MODEL 1

Equation Parameters RMSE R2 χ2 p-value
Competition 112 13.2516 0.5089 4125.827 0.0000
Violent enforcement 112 3.08389 0.3201 1873.904 0.0000
Retaliation 112 5.27884 0.3929 2576.207 0.0000

Rather than specific parameter estimates, the discussion of the regression results
is focused on analyzing the causal relationships relating to the dynamic interaction
among actors as stated in Hypotheses H12.1 , H12.2 , H4 and H5 . The dynamic and
reciprocal interactions inherent to a VAR analysis require a Granger causality test to
be performed to assess the effect of one variable (or sets of variables) on another and
vice versa. The Granger test of causality indicates whether the lags of one variable
affect the outcome of another variable.1 This test is implemented by performing a

1
Following Enders (2009, 318), the Granger causality test analyzes the temporal effects of pre-
vious values of a variable x(t) on the outcome of another variable, y(t), such that x(t) → y(t) . In
this sense, the term causality is restricted to the temporal relationship between the two variables,
which is referred as Granger causality. Therefore, the Granger causality test should not be directly
interpreted as a test of causal inference as understood in the experimental literature (Dunning,
2012).

367
χ2 test for the joint hypothesis that a potential causal variable does not cause the
other variable. Performing a Granger causality test requires first estimating the full
model to calculate the likelihood of this specification and then excluding the time
series of another variable to calculate the likelihood of this restricted model. The
null hypothesis of the Granger causality test is that the excluded time series has no
effect on the outcome of interest. If the likelihood of the restricted and full models
are the same, the test fails to reject the null hypothesis. This means that extracting
the removed time series has no effect in the full model and the excluded variable does
not contribute to the explanation of the outcome of interest. In contrast, if the test
reports a large coefficient with high statistical significance, then we can reject the
null hypothesis and state that the excluded time series Granger-causes the outcome.
Table 7.4 presents the result of the Granger causality test for VAR model 1. The
χ2 statistics of all tests are large at high levels of statistical significance, thus providing
evidence for the endogeneity inherent to the dynamics of violence in the Mexican war
on drugs. In particular, the Granger causality test indicates that previous efforts by
government authorities to enforce the law Granger-cause violent competition among
criminals. In addition, events of criminal retaliation against government authorities
Granger cause violence among DTOs. The third row in Table 7.4 indicates that a
restricted model excluding time series of both violent enforcement and criminal re-
taliation generates a different likelihood than the full model, thus suggesting that the
levels of violent law enforcement and criminal retaliation jointly cause competition
among rival drug cartels. The remaining results in the Granger test of causality con-
firm that these relationships are reciprocal. The statistics indicate that competition
among DTOs and criminal attacks against government authorities independently and
jointly Granger-cause violent law enforcement by the state. In addition, the results
indicate that violent confrontations between rival criminal organizations and govern-

368
ment efforts to enforce the law by violent means Granger-cause criminal retaliation
against government authorities.

TABLE 7.4

GRANGER CAUSALITY TEST FOR VAR MODEL 1

Equation Excluded χ2 test d.f. p-value
Violent enforcement 84.31 37 0.000
Competition Criminal retaliation 84.77 37 0.000
All 195.52 74 0.000
Competition 118.8 37 0.000
Violent enforcement Criminal retaliation 59.85 37 0.010
All 185.06 74 0.000
Competition 122.91 37 0.000
Criminal retaliation Violent enforcement 116.99 37 0.000
All 246.54 74 0.000

Although the Granger causality test provides valuable information about the re-
ciprocal relationships among the different time series, the test only indicates whether
the temporal variation of one time series has an effect on the outcome of another se-
ries. It does not specify the direction or the magnitude of this feedback relationship.
Fortunately, the analytical leverage of VAR models helps provide an understanding
of the dynamic characteristics and interactive behavior of the series estimated in the
system. The empirical strategy for identifying the direction and magnitude of the re-
ciprocal effect of the variables relies on the use of impulse response functions (IRFs).
These enable the effect of an external shock on the behavior of another variable to
be analyzed. The external shock is thus modeled as an impulse and the IRF function

369
describes the reaction of the time series of interest. The dynamic decomposition of
the effect provides information about the direction of the effect, its magnitude, and
its duration over time.
A simple IRF gives the effect over time caused by a one-time unit shock on the
response function, holding all else constant. For example, suppose that we want to
assess the effect of an external shock of law enforcement on violence among criminals.
In principle, the IRF could give us information about the behavior of violence be-
tween DTOs as a response to a one-time impulse of violent enforcement. However, as
indicated by parameters θ10 Et in equation (7.8), the shock of law enforcement is con-

temporaneously correlated with the outcome of violent competition between criminals

and it cannot be held constant; therefore a simple IRF cannot have a causal inter-
pretation for the kind of specification used in VAR model 1. To solve this problem,
Orthogonal Impulse Response Functions (OIRFs) implement a Cholesky decomposi-
tion to treat the impulse as an exogenous shock and enable a causal interpretation
(Hill, 2006; Sims, 1980; Sims and Zha, 1999).
Figure 7.5 shows the orthogonal impulse response functions (OIRFs) of the various
processes of violence inherent to the war on drugs. The solid line represents the effect
of the shock, the gray area around the impulse depicts the 95 percent confidence
intervals,2 steps in the horizontal axis indicate the number of days after day 0, when
the external shock is applied. The analysis of response functions in this figure extends
up to 40 days following the initial shock.
The panel in the upper left of Figure 7.5 shows the OIRF effect of a one-unit
increase of violent law enforcement at time 0 on the behavior of violent competition

2
It is important to note that the impulse response functions presented in this research have very
narrow margins of confidence. This is unusual in VAR analysis, as most IRF display very wide
margins of error, thus reducing the levels of certainty about the results. Wide margins of confidence
are usually caused by the use of aggregated data in large temporal units (e.g. years). The use
of fine-grained daily data contained in OCVED produces narrow margins of error, thus increasing
confidence about the results.

370
 2 2

Number of events
1 1
.5
0 0

-1 -1
0 10 20 30 40 0 10 20 30 40
Days Days
95% CI Enforcement --> Competition 95% CI Competition --> Enforcement

3 2
Number of events

2
1
1
.2
.1
0
-.1
0 -.2

-1 -1
0 10 20 30 40 0 10 20 30 40
Days Days
95% CI Enforcement --> Retaliation 95% CI Retaliation --> Enforcement

Figure 7.5. Orthogonal Impulse Response Functions (OIRF) of violence in

the Mexican war on drugs

between criminals. The OIRF shows that a single shock of violent enforcement has
a large disturbing effect on violence between DTOs that endures over time. This
effect is highly irregular, consisting of a series of spikes of violence between rival
criminal organizations at days 5, 11, 14, 15, 21, 23–28, 33 and 35. In the last ten
days of the analysis, the response function shows a sustained positive trend. This
OIRF panel reveals that the use of force by government authorities to fight crime has
a lasting effect on criminal competition that causes several confrontations between
rival criminal groups. Moreover, even after 40 days from the initial crackdown from
the state, violence between DTOs does not fade. This impulse response function thus
provides support for hypothesis H5 which claims that violent enforcement generates

371
violence between criminal organizations. As expected from the theoretical model, law
enforcement actions from the state disrupt the balance among criminal organizations
and are likely to trigger a series of violent actions and reactions among DTOs as they
push the relative military balance back and forth between them.
The second panel at the upper right of Figure 7.5 shows the effect of violent
competition between criminal groups on the levels of violent enforcement employed
by the state. The figure shows that a one-time event of violence between DTOs
generates an immediate repressive reaction from government authorities but has no
lasting consequences on the behavior of the state. The OIRF indicates that the reac-
tion from government authorities is proportional to the violence between criminals,
to the extent that one event of criminal competition generates approximately one
violent reaction by the state on day 0. Analyzing the behavior of the state over a
40-day period shows that the event of violence between criminals does not have a long
term impact on law enforcement beyond the immediate reaction by government au-
thorities. This evidence provides support for Hypothesis H12.2 which claims violence
between DTOs causes the state to employ coercive force to impose order. However,
it is surprising that violence between DTOs generates such a modest reaction from
government authorities.
The third panel at the lower left of Figure 7.5 presents the OIRF function for
the effect of a one-time event of violent law enforcement on criminal retaliation. The
figure shows that criminal groups react immediately to violent law enforcement by
perpetrating approximately two attacks against government authorities in response
to a single event of violent enforcement. However, after a few days there are no more
attacks against government authorities. This result provides support for the theo-
retical expectation expressed in Hypothesis H4 , which claims that law enforcement
generates violent contestation from criminal organizations against the state. One of
the most remarkable findings of this panel is the scale of the response by organized

372
criminals to attacks from government forces; that is, retaliating twice for each single
action committed by the state.
Finally, the lower right panel in Figure 7.5 reports the behavior of the state in
response to a one-time attack from organized criminals. The panel shows the sur-
prising finding that attacks perpetrated by criminal organizations against the state
do not seem to generate a violent reaction from government authorities. In contrast
to the robust expectation of the “law of coercive responsiveness” that the state will
invariably deploy a repressive response when threatened by a hostile actor, the re-
sults show that the Mexican government has not reacted with the use of force when
attacked by criminal organizations. The evidence provided in this panel does not
support hypothesis H12.1 .
OIRFs provide valuable insights into the dynamic interactions of drug-related vi-
olence. However, these functions only provide information about the behavior of a
time series in reaction to a single shock in one point in time. Unfortunately, this
single-shot approach does not reflect the sustained events of violence characteristic of
the Mexican war on drugs. To overcome this limitation, the empirical strategy relies
on Cumulative Orthogonal Impulse Response Functions (COIRF) to assess the ag-
gregated response of a variable of interest in reaction to a sequence of shocks. Instead
of focusing on the effect of a single and isolated event, COIRFs enable the cumulative
effect of sustained campaigns of violence to be modeled. COIRFs describe the re-
sponse over time of an outcome variable in reaction to the application of a sequence of
one-unit shocks applied over a period of time. For example, in an observation period
of 40 days, the COIRF incorporates the cumulative effect of a single shock occurring
repeatedly every day over the 40-day period. The analytical leverage of this tool is
thus helpful for understanding the reciprocal and dynamic interactions characteristic
of sustained campaigns of violence.

373
 Figure 7.6 presents the cumulative orthogonal impulse response functions of the
Mexican war on drugs. The upper left panel shows the aggregated effect on violence
between DTOs caused by the government continously using violent enforcement over
a period of 40 days. The panel shows a sustained positive effect, indicating that the
sustained use of force against criminal groups generates a massive escalation of vio-
lence between criminal organizations. According to the results, by the fortieth day in
the campaign of violent law enforcement, a single additional government crackdown
triggers some 12 events of violence between criminal organizations the same day. Even
when considering the wide span of the 95 percent confidence intervals caused by the
irregular variation in each individual crackdown discussed above, the fortieth consec-
utive event of law enforcement generates between 6 and 19 events of violent criminal
confrontations. Consistent with theoretical expectations, this result indicates that
launching a full-fledged campaign against criminal organizations is likely to have a
large disturbing effect on the relative military balance among criminal groups, thus
unleashing a massive and lasting wave of violence among rival criminal groups. The
COIRF function in this panel thus provides strong support for hypothesis H5 .
The upper right panel in Figure 7.6 presents the COIRF for the effect of violent
competition between DTOs on the levels of violent law enforcement deployed by the
state. The results indicate that sustained confrontations among rival criminal groups
increase the number of violent crackdowns conducted by government authorities.
After a sustained sequence of 40 days of confrontations between DTOs the state
employs violent tactics to fight crime an average of five times a day. These results
provide support for hypothesis H12.2 which claims that the state uses force to impose
order. However, it is surprising that after 40 days of violent confrontations between
criminal groups the state displays such moderate behavior.
The panel in the lower left corner of Figure 7.6 shows the effect of violent law
enforcement on criminal retaliation against the authorities. The results show that

374
 15
15

Number of events
10
10
6
5 4
2
0 0
-2 -2
0 10 20 30 40 0 10 20 30 40
Days Days
95% CI Enforcement --> Competition 95% CI Competition --> Enforcement

15 15
Number of events

10 10
8
6
5
4
2
1
0 0
-1
-2 -2
0 10 20 30 40 0 10 20 30 40
Days Days
95% CI Enforcement --> Retaliation 95% CI Retaliation --> Enforcement

Figure 7.6. Cumulative Orthogonal Impulse Response Functions (COIRF)

of violence in the Mexican war on drugs

government efforts to fight crime generate an aggressive response against the state
from organized criminals. According to the OCIRF, the first crackdown from the
state generates two attacks against government authorities perpetrated by criminal
organizations. The intensity of contestation keeps increasing as the government con-
tinues to carry out a punitive strategy over the subsequent days and weeks. After
the fortieth day of consecutive state violence, criminal organizations retaliate with
an average of seven attacks against government authorities per event of violent en-
forcement carried out by the state. This indicates that instead of deterring violent
challenges against the state, the punitive strategy to fight crime exacerbates the hos-

375
tilities against government authorities. The escalating effect of law enforcement on
criminal retaliation provides empirical support for hypothesis H4 .
Finally, the lower right panel in Figure 7.6 shows the COIRF for the effect of
criminal retaliation on levels of violent law enforcement. The COIRF reveals that
even after a series of 40 days of continuous attacks from criminal organizations, there
is no evidence of the state engaging in violent behavior to counter or neutralize
criminal attacks. This languid response from the state contradicts the theoretical
expectation from the “law of coercive responsiveness,” thus failing to provide support
for hypothesis H12.1 .

7.4.2
Full
Specification
of
the
Vector
Autoregressive
Model
for
the
Dynamic
Analysis
of
Drug
Violence

The previous section analyzed the micro-mechanisms of drug violence using VAR
model 1. Although this analysis provides valuable understanding of the dynamics
of conflict in the Mexican war on drugs, it only considers the interactions between
criminal competition, violent enforcement, and criminal retaliation but ignores other
non-violent enforcement tactics also used by the state against criminal organizations.
This section presents the analysis of VAR model 2 incorporating the complete menu of
violent and non-violent anti-criminal tactics such as arrests; seizurea of assets, drugs,
and weapons; and it includes the time series of violent competition between DTOs,
and criminal retaliation against government authorities. This model also includes a
vector of structural factors as control variables.
Equations (7.11)–(7.17) constitute the system of equations estimated in the fully
specified vector autoregressive model, referred here as VAR model 2, which contains
the following parameters:

• Ct represents the outcome of the time series of violent competition between
criminal organizations and all the other series contained in equation (7.11).

376
• Et represents the outcome of the time series of violent law enforcement and all
the other series contained in equation (7.12).
• Rt represents the outcome of the time series of retaliation against the state and
all the other series contained in equation (7.13).
• At represents the outcome of the time series of arrests and all the other series
contained in equation (7.14).
• St represents the outcome of the time series of seizures of criminal assets and
all the other series contained in equation (7.15).
• Dt represents the outcome of the time series of drug interdiction and all the
other series contained in equation (7.16).
• Gt represents the outcome of the time series of seizures of guns and all the
other series contained in equation (7.17).
• Ct−i contains a vector of violent competition between criminal organizations at
time t − i, and βei is a vector of its corresponding coefficients in the system of
equations of the fully specified VAR model 2.
• Et−i contains a vector of violent law enforcement and θei is a vector of its
corresponding coefficients.
• Rt−i contains a vector of criminal retaliation against the state and Ωei is a
vector of its corresponding coefficients.
• At−i contains a vector of arrests and δei is a vector of its corresponding coeffi-
cients.
• St−i contains a vector of seizure of criminal assets and σei is a vector of its
corresponding coefficients.
• Dt−i contains a vector of drug interdiction and μei is a vector of its correspond-
ing coefficients.
• Gt−i contains a vector of seizure of guns and ρei is a vector of its corresponding
coefficients.
• Xt contains a vector of control variables and Πe is a vector of its correspond-
ing coefficients. The control variables included in this vector are poverty, state
GDP, drug production, 9/11, rifles and unemployment. The descriptive statis-
tics of the controls can be found Appendix A.7.
• αe give the intercepts in each equation.
• uet are the error terms for each equation.

The full specification of VAR model 2 consists of the following system of equations:

377
Ct = α1 + β11 Ct−1 + β12 Ct−2 + · · · + β1k Ct−k
+ θ10 Et + θ11 Et−1 + θ12 Et−2 + · · · + θ1k Et−k
+ Ω10 Rt + Ω11 Rt−1 + Ω12 Rt−2 + · · · + Ω1k Rt−k
+ δ10 At + δ11 At−1 + δ12 At−2 + · · · + δ1k At−k
(7.11)
+ σ10 St + σ11 St−1 + σ12 St−2 + · · · + σ1k St−k
+ μ10 Dt + μ11 Dt−1 + μ12 Dt−2 + · · · + μ1k Dt−k
+ ρ10 Gt + ρ11 Gt−1 + ρ12 Gt−2 + · · · + ρ1k Gt−k
+ Π1 Xt + u1t

Et = α2 + θ21 Et−1 + θ22 Et−2 + · · · + θ2k Et−k

+ β20 Ct + β21 Ct−1 + β22 Ct−2 + · · · + β2k Ct−k
+ Ω20 Rt + Ω21 Rt−1 + Ω22 Rt−2 + · · · + Ω2k Rt−k
+ δ20 At + δ21 At−1 + δ22 At−2 + · · · + δ2k At−k
(7.12)
+ σ20 St + σ21 St−1 + σ22 St−2 + · · · + σ2k St−k
+ μ20 Dt + μ21 Dt−1 + μ22 Dt−2 + · · · + μ2k Dt−k
+ ρ20 Gt + ρ21 Gt−1 + ρ22 Gt−2 + · · · + ρ2k Gt−k
+ Π2 Xt + u2t

Rt = α3 + Ω31 Rt−1 + Ω32 Rt−2 + · · · + Ω3k Rt−k

+ β30 Ct + β31 Ct−1 + β32 Ct−2 + · · · + β3k Ct−k
+ θ30 Et + θ31 Et−1 + θ32 Et−2 + · · · + θ3k Et−k
+ δ30 At + δ31 At−1 + δ32 At−2 + · · · + δ3k At−k
(7.13)
+ σ30 St + σ31 St−1 + σ32 St−2 + · · · + σ3k St−k
+ μ30 Dt + μ31 Dt−1 + μ32 Dt−2 + · · · + μ3k Dt−k
+ ρ30 Gt + ρ31 Gt−1 + ρ32 Gt−2 + · · · + ρ3k Gt−k
+ Π3 Xt + u3t

At = α4 + δ41 At−1 + δ42 At−2 + · · · + δ4k At−k

+ β40 Ct + β41 Ct−1 + β42 Ct−2 + · · · + β4k Ct−k
+ θ40 Et + θ41 Et−1 + θ42 Et−2 + · · · + θ4k Et−k
+ Ω40 Rt + Ω41 Rt−1 + Ω42 Rt−2 + · · · + Ω4k Rt−k
(7.14)
+ σ40 St + σ41 St−1 + σ42 St−2 + · · · + σ4k St−k
+ μ40 Dt + μ41 Dt−1 + μ42 Dt−2 + · · · + μ4k Dt−k
+ ρ40 Gt + ρ41 Gt−1 + ρ42 Gt−2 + · · · + ρ4k Gt−k
+ Π4 Xt + u4t

378
 St = α5 + σ51 St−1 + σ52 St−2 + · · · + σ5k St−k
+ β50 Ct + β51 Ct−1 + β52 Ct−2 + · · · + β5k Ct−k
+ θ50 Et + θ51 Et−1 + θ52 Et−2 + · · · + θ5k Et−k
+ Ω50 Rt + Ω51 Rt−1 + Ω52 Rt−2 + · · · + Ω5k Rt−k
(7.15)
+ δ50 At + δ51 At−1 + δ52 At−2 + · · · + δ5k At−k
+ μ50 Dt + μ51 Dt−1 + μ52 Dt−2 + · · · + μ5k Dt−k
+ ρ50 Gt + ρ51 Gt−1 + ρ52 Gt−2 + · · · + ρ5k Gt−k
+ Π5 Xt + u5t

Dt = α6 + μ61 Dt−1 + μ62 Dt−2 + · · · + μ6k Dt−k

+ β60 Ct + β61 Ct−1 + β62 Ct−2 + · · · + β6k Ct−k
+ θ60 Et + θ61 Et−1 + θ62 Et−2 + · · · + θ6k Et−k
+ Ω60 Rt + Ω61 Rt−1 + Ω62 Rt−2 + · · · + Ω6k Rt−k
(7.16)
+ δ60 At + δ61 At−1 + δ62 At−2 + · · · + δ6k At−k
+ σ60 St + σ61 St−1 + σ62 St−2 + · · · + σ6k St−k
+ ρ60 Gt + ρ61 Gt−1 + ρ62 Gt−2 + · · · + ρ6k Gt−k
+ Π6 Xt + u6t

Gt = α7 + ρ71 Gt−1 + ρ72 Gt−2 + · · · + ρ7k Gt−k

+ β70 Ct + β71 Ct−1 + β72 Ct−2 + · · · + β7k Ct−k
+ θ70 Et + θ71 Et−1 + θ72 Et−2 + · · · + θ7k Et−k
+ Ω70 Rt + Ω71 Rt−1 + Ω72 Rt−2 + · · · + Ω7k Rt−k
(7.17)
+ δ70 At + δ71 At−1 + δ72 At−2 + · · · + δ7k At−k
+ σ70 St + σ71 St−1 + σ72 St−2 + · · · + σ7k St−k
+ μ70 Dt + μ71 Dt−1 + μ72 Dt−2 + · · · + μ7k Dt−k
+ Π70 Xt + u7t

VAR model 2 comprises a complex system of reciprocal and dynamic interactions

among the processes of violent competition between DTOs, violent law enforcement,
criminal retaliation, and non-violent law enforcement tactics also used by the state to
fight criminals; arrests, seizures of assets, drug interdiction and seizures of weapons.
The analysis accounts for the temporal inertia of each of these series as it is affected
by past and contemporary events of other time series. In contrast to the model
specification of VAR model 1, which incorporates an AR process of 37 lags, the

379
analysis of the autoregressive order processes in VAR model 2 indicates that the
most appropriate specification is an AR model with only 15 lags.
The particular values of the coefficients of each of the 112 parameters in VAR
model 2 are not of specific interest of this research. For this reason the table of
regression results is not reported. Table 7.5 reports the fit of the model with R2
estimates ranging between 29.8 percent to 54.3 percent. However, as mentioned
earlier, it is not surprising to find high R2 in vector autoregressive analysis due to
the high degree of serial correlation in this type of model.

TABLE 7.5

EXPLANATORY POWER FOR VAR MODEL 2

Equation Parameters RMSE R2 χ2 p-value
Competition 112 13.3493 0.4998 3999.927 0.0000
Violent enforcement 112 3.07918 0.3189 1874.552 0.0000
Criminal retaliation 112 5.25578 0.3957 2620.75 0.0000
Arrests 112 12.1608 0.3680 2330.546 0.0000
Seizure of assets 112 5.68984 0.2549 1369.44 0.0000
Seizure of drugs 112 16.3792 0.3228 1908.427 0.0000
Seizure of guns 112 5.51704 0.3387 2050.348 0.0000

Table 7.6 presents the Granger causality test for the full model. Large χ2 statistics
at high levels of significance provide evidence of reciprocal interactions and feedback
effects between the different time series. Events of competition between criminal
organizations are individually and jointly Granger-caused by violent enforcement,
criminal retaliation, arrests, seizures of assets, drug interdiction, and gun seizures.
Observations in the time series of violent enforcement are individually and jointly

380
Granger-caused by violence between DTOs, criminal retaliation against the state, and
non-violent tactics such as arrests, seizures of assets and drugs; however, gun seizures
do not seem to affect violent enforcement. The time series of attacks perpetrated by
DTOs against the state is individually and jointly Granger-caused by competition
between rival criminal groups, violent enforcement, and all non-violent state actions.
The Granger test indicates that arrests are caused by violence between DTOs, violent
law enforcement, criminal retaliation, drug interdiction, and gun seizures, but not by
seizures of criminal assets. Events involving seizures of criminal assets are Granger-
caused by competition between criminal organizations, violent enforcement, criminal
retaliation, and arrests of suspected members of criminal organizations; however,
drug interdiction and gun seizures do not seem to affect the confiscation of criminal
assets. The time series of drug interdiction is individually and jointly Granger-caused
by violence between DTOs, violent law enforcement, criminal retaliation, and gun
seizures. Finally, the causality test indicates that gun seizures are caused by conflict
between rival criminal groups, violent enforcement, criminal retaliation, arrests, and
seizures of assets and of drugs.

The analysis of the impulse response functions of the processes of violence inherent
to the Mexican war on drugs is divided into two parts. The first part analyzes the
dynamics of conflict between DTOs and criminal reactions against the state caused by
government efforts to fight crime. The second part analyzes how the state implements
its various violent and non-violent security actions in reaction to increasing levels of
conflict between DTOs or as a response to direct attacks perpetrated by criminals.
Figure 7.7 reports the cumulative orthogonal impulse response functions (COIRF)
for the effect of the full menu of violent and non-violent tactics on the dynamics of vio-
lent competition between criminal organizations. To facilitate the visualization of the
results, the figure does not present the confidence intervals. However, as mentioned

381
 TABLE 7.6

GRANGER CAUSALITY TEST FOR VAR MODEL 2

Excluded
Compe- Enfor- Retali-
Equation tition cement ation Arrests Assets Drugs Guns All
Competition 84*** 86*** 49* 83*** 54** 89*** 471***
Enforcement 114*** 61*** 80*** 73*** 74*** 60 439***
Retaliation 91*** 101*** 99*** 69*** 85*** 56** 527***
Arrests 76*** 92*** 53** 31 68*** 66*** 395***
Assets 66*** 94*** 62*** 57*** 45 46 389***
Drugs 80*** 50** 64*** 40 45 63*** 343***
Guns 62*** 104*** 75*** 61*** 61*** 68*** 452***
Numbers represent the χ2 statistic for the Granger causality test.
The levels of significance are: * p< 0.1, ** p< 0.05, *** p< 0.01

above, the use of daily data yields narrow margins of error. The response function
indicates that sustained campaigns of violent enforcement and seizures of criminal
assets have the most disrupting effect on criminals and cause the highest levels of vi-
olent confrontations between them. The results of Figure 7.7 provide strong support
for the theoretical expectation that state efforts to fight crime disrupt the relative
military balance among criminals and trigger waves of violence between rival criminal
groups. According to the figure, the final event of enforcement after a campaign of
40 days of violent law enforcement generates 7.1 confrontations between criminals on
that day. By the end of the campaign, 40 consecutive events of violent law enforce-
ment have generated a cumulative count of 153 events of criminal competition. The
results also show that the last seizure of assets over a 40-day period triggers 7.8 events
of violence between rival DTOs. By the end of this period, 40 seizures of assets have
generatee a cumulative total of 184 events of violence between criminal groups. The
COIRF function indicates that this positive effect is also present in the time series of
arrests, although the effect is less marked. The fortieth arrest in a sustained series of

382
detentions generates about 2.3 events of violence between criminal organizations. At
the end of a sequence of 40 arrests, this campaign generates about 38 events of vio-
lence between criminal groups. These results provide strong support for hypothesis
H5 , which claims that law enforcement triggers waves of violent competition between
DTOs.
8
6
Number of events
2 0
-2 4

0 10 20 30 40
Days

Violent enfocement --> Competition Arrests --> Competition

Seizure of assets --> Competition Seizure of drugs --> Competition
Seizure of guns --> Competition

Figure 7.7. Cumulative Orthogonal Impulse Response Functions (COIRF)

for the effect of violent and non-violent enforcement on competition among
DTOs

In contrast to the theoretical expectations, the time series of drug interdiction and
gun seizures shown in Figure 7.7 seem to have a slight deterrent effect on criminal

383
competition. The fortieth seizure of drugs reduces events of violence between criminal
organizations by a count of about 1.2. By the end of a sustained campaign of 40 days
of drug interdiction there would be 34.4 fewer events of criminal competition. The
fortieth confiscation of weapons barely reduces conflict among criminals by generating
0.7 fewer events of violence. By the end of a sustained campaign of gun seizures
there would be 34 fewer events of criminal confrontation. It is important to note
that the margins of error of the time series of drug and gun seizures (not reported
in the figure) make this effect indistinguishable from zero. In any case, the COIRF
functions modeling drug interdiction and seizures of guns do not support H5 .
Figure 7.8 presents the COIRF function for the dynamic effect of violent and
non-violent enforcement on criminal retaliation against government authorities. The
results provide strong support for the empirical implication derived from the formal
model that law enforcement is likely to trigger counterattacks from criminal groups
against the state. The impulse response function of violent enforcement indicates that
the fortieth consecutive crackdown on criminals generates about 5.2 violent reactions
perpetrated by DTOs against government authorities. By the end of a campaign
of 40 days of consecutive events of violent law enforcement, the punitive strategy
will generate 172.2 attacks from criminal organizations against the state. The figure
also reports a positive relation between seizures of assets and criminal retaliation.
The last confiscation of assets in a 40-day campaign of seizures generates about 2.2
counterattacks from DTOs against the state. By the end of the sequence of seizures
of criminal assets, the cumulative count of acts of retaliatory aggression from DTOs
will be 55.1. These results provide strong support for the theoretical expectation
stated in hypothesis H5 , which claims that DTOs are likely to retaliate against law

enforcement.
The results also indicate that arrests generate violent reactions from criminal or-
ganizations against government authorities, although the effect is less acute than the

384
 6 4
Number of events
0 2-2

0 10 20 30 40
Days

Violent enforcement --> Retaliation Arrests --> Retaliation

Seizure of assets --> Retaliation Seizure of drugs --> Retaliation
Seizure of guns --> Retaliation

Figure 7.8. Cumulative Orthogonal Impulse Response Functions (COIRF)

for the effect of violent and non-violent enforcement on criminal retaliation

effect of state violence or asset confiscation. According to the figure, the fortieth
arrest generates about 0.9 events of violence against the state, and by the end of
a campaign of 40 consecutive arrests there will be an aggregated number of 17.7
criminal attacks against government authorities. The results depicted in Figure 7.8
show that the effect of gun seizures is not distinguishable from zero. By the end of
a campaign of 40 days of gun confiscations there will be a 0.8 reduction in the accu-
mulated count of retaliation events. The results from arrest and gun seizure models
indicate that by reducing the human reserve and firepower of criminal organizations,
government authorities are capable of neutralizing attacks from DTOs. In contrast
to the expectation derived from hypothesis H5 , the COIRF function of drug seizures
indicates that a sustained campaign of drug interdiction is capable of inhibiting hos-

385
tilities against government authorities. The fortieth confiscation of drugs can reduce
the number of daily counterattacks by 2.5 events and by the end of a sequence of 40
drug seizures the cumulative number of retaliation attacks will be reduced by 63.9.
Figure 7.9 presents the reactions implemented by government authorities as a
response to increasing levels of violence among criminal organizations. The COIRF
functions provide strong support for hypothesis H12.2 , which claims that the state
employs the use of force to impose order within its territory. In general terms, the
impulse response analysis in Figure 7.9 shows that the Mexican government relies
mostly on non-violent responses to increasing competition between DTOs. In this
sense, the use of lethal force is employed as a last resort. All the security tactics
available to the state are increasing as a reaction to violence among DTOs, but
the impulse response function of violent enforcement is less acute than that of non-
violent strategies. After a period of 40 days of consecutive events of violence between
criminal organizations, the state reacts with three violent crackdowns per day. At the
end of the wave of criminal violence, the state would have implemented 92.7 violent
crackdowns. In contrast, the COIRF of arrests indicates that the fortieth event of
violence between DTOs generates 9.9 daily arrests, and at the end of the period of
criminal confrontations the state would have arrested 277.8 suspected members of
criminal organizations. The models also estimate that government authorities will
implement 5.2 daily seizures of assets as a response to the fortieth event of violence
between criminals, and that the cumulative count of asset confiscation events will
be 143.2. Finally, the results indicate a similar trend for seizures of guns. The last
event of criminal competition in a consecutive series of 40 days will correspond to
4.7 weapons seizures carried out by the state. By the end of this period, government
authorities would have seized weapons 122.4 times.
Finally, Figure 7.10 presents the results for the COIRF functions analyzing the
reaction of the state against acts of aggression perpetrated by criminal organizations.

386
 10
8
Number of events
4 2
0 6

0 10 20 30 40
Days

Competition --> Violent enforcement Competition --> Arrests

Competition --> Seizure of assets Competition --> Seizure of drugs
Competition --> Seizure of guns

Figure 7.9. Cumulative Orthogonal Impulse Response Functions (COIRF)

for the effect of competition among DTOs on violent and non-violent
enforcement

In general, the results do not provide support for hypothesis H12.2 , which formulates
the “law of coercive responsiveness” for this context. The response from the Mexican
government in terms of the number of arrests, events of violent enforcement, seizures
of criminal assets, and confiscation of weapons are barely distinguishable from zero
even at the end of a 40-day period of consecutive attacks from criminal groups.
Moreover, the impulse response function of drug seizures shows a negative trend.
This suggests that a sustained campaign of criminal hostilities is capable of inhibiting
government efforts to seize illicit drugs from criminal groups. After the fortieth
attack from organized criminals, the number of daily drug seizures decreases by 2.7
events and by the end of a sustained campaign of aggression against the state the
estimated number of drug interdiction events is 60.2 lower. In any case, the lack of

387
responsiveness from the Mexican state in the face of criminal hostilities is a robust
and surprising finding.

1
Number of events
-1 -2
-3 0

0 10 20 30 40
Days

Retaliation --> Violent enforcement Retaliation --> Arrests

Retaliation --> Seizure of assets Retaliation --> Seizure of drugs
Retaliation --> Seizure of guns

Figure 7.10. Cumulative Orthogonal Impulse Response Functions (COIRF)

for the effect of criminal retaliation on violent and non-violent enforcement

As indicated by Olson, Shirk and Selee (2010), Mexican government officials have
repeatedly stated that the increase of violence between DTOs lends credence to the
claim that the strategy against criminal organizations is working. According to this
argument, violence between criminals might increase in the short term but will de-
crease in the long run. In order to explore the possibility of an inverse (U-shaped)
relationship, in which violence initially increases but eventually goes down over time,

388
Figure 7.11 explores the predictions of the impulse response functions extended to a
period of six months. This figure is based on the estimates of VAR model 2 but only
reports the dynamic interactions between violent law enforcement, competition be-
tween DTOs and criminal retaliation, omitting the effect of non-violent enforcement.
The COIRF impulse response functions of sustained campaigns of violence among the
various actors over six months did not show any of the processes of violence declining
over time. Instead, violent competition among rival DTOs generated by the use of
force by government authorities stabilizes at a higher level of violence.

15 15

10 10

5
5
4
3
2
0 1
0
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Days Days

95% CI Violent enforcement --> Competition 95% CI Competition --> Violent enforcement

15 15

10 10

8
6 5
4
2
2 1
0
0 -1
0 30 60 90 120 150 180 0 30 60 90 120 150 180
Days Days

95% CI Violent enforcement --> Retaliation 95% CI Retaliation --> Violent enforcement

Figure 7.11. Cumulative Orthogonal Impulse Response Functions (COIRF)

of sustained campaigns of violence during six months

389
 The upper left panel in Figure 7.11 shows that a series of consecutive violent
crackdowns from government authorities over a period of 180 days causes a marked
increase in violent competition between criminal organizations in the first two months.
Then, the series stabilizes with a response of approximately eight daily events of vio-
lence between criminals per each additional event of violent enforcement. The upper
right panel shows that the violent response of the state against violent competition
between DTOs also increases in the first two months of hostilities between criminals.
State violence stabilizes after 60 days and responds with 3.5 events of violent law
enforcement for each additional event of criminal competition. The lower left panel
shows that criminal organizations deploy a sustained response against the state when
government authorities use violent enforcement. The time series of criminal retal-
iation stabilizes after the first month with a response of about 5.8 attacks against
the state per every additional crackdown conducted by government authorities. Fi-
nally, the lower right panel shows that even after six months of direct hostilities from
criminal organizations, the state does not seem to respond with the use of violence to
counter criminal attacks. In general, none of the impulse response functions analyzed
in Figure 7.11 show that violence declines within a period of six months of sustained
hostilities. In consequence, there is no empirical support for the argument that the
punitive strategy of the Mexican government is capable of reducing violence in the
long run.

7.5
Discussion
of
Results

The statistical analysis of this chapter relies on vector autoregressive (VAR) mod-
els. This type of time series analysis enables reliable predictions to be generated about
future behavior of a violent process by incorporating the effect of previous events from
the same time series as well as dynamic and reciprocal effects from other time series

390
affecting the outcome. In this context, the VAR analysis thus takes into account
what happened in the past to inform the behavior of violent trends in the future.
In addition, the statistical analysis relies on cumulative orthogonal impulse response
functions (COIRF) to analyze the behavior over time of a violent process as it is
affected by external shocks from other time series.

State Action Leads to Violence Among Criminals

The statistical analysis of this section provides strong support for the theoretical
expectation that increasing levels of law enforcement are associated with higher levels
of conflict between rival criminal organizations. According to the formal model,
state actions disrupt the relative military balance among criminals by weakening
the target DTO, which improves the relative military position of a challenger DTO.
This non-neutral effect of law enforcement is capable of motivating an invasion by
an indirectly-empowered challenger DTO onto the weakened group, thus triggering a
wave of violent competition among rival criminal organizations as they try to capture
and defend valuable territories.
The impulse response analysis reveals that a single event of violent law enforce-
ment has a lasting effect on violence among DTOs capable of generating a series of
confrontations among criminals that continues for more than 40 days. The dynamic
analysis also shows that the use of violent law enforcement, seizures of criminal assets,
and arrests of suspected members of criminal organizations have the most disrupting
effects on violent competition between DTOs. In particular, the results indicate that
after a sustained campaign of 40 days of consecutive violent crackdowns by govern-
ment authorities, one additional event of violent law enforcement generates about
7.1 events of violence between criminal organizations that same day. By the end of
the punitive campaign, 40 consecutive events of violent law enforcement will have
triggered a cumulative count of 153 events of violence between DTOs.

391
 Some of the statistical results fail to provide support for the hypothesis that non-
violent state action generates violence between criminal organizations. The impulse
response analysis did not show that drug interdiction and seizures of guns have any
impact on the levels of violence between criminal organizations. These findings are
contrary to the results of Figure 6.15 in Chapter 6, which indicate that increasing
levels of drug seizures and confiscation of weapons have a large positive effect on
violence among DTOs. However, it is important to note that the statistical analy-
ses used in these two chapters are substantially different and their estimates are not
directly comparable. The empirical strategy in Chapter 6 relies on an instrumental
variable (IV) approach for a time-series cross-sectional (TSCS) data structure us-
ing municipality–days as the unit of analysis. In contrast, the empirical strategy of
the present chapter uses a VAR model for the dynamic analysis of data daily data
aggregated at the national level. Future research will focus on desegregating the dy-
namic analysis of VAR at the state and municipal levels to assess the effect of violent
and non-violent enforcement on violence among criminals. Besides the differences in
terms of units of analysis between these two empirical approaches, the key difference
rests on the plausibility of a causal claim. The IV design constitutes a stronger quasi-
experimental identification strategy capable of overcoming problems of endogeneity
and providing more plausible claims of causal inference. In contrast, causality in
VAR analysis refers to the temporal effects of one variable over another without the
characteristics of a quasi-experimental design.

Criminal Hostilities Against the State

Another important finding of the impulse response analysis is evidence about the
aggressive behavior of organized criminals as they react against the state when the
government uses violent law enforcement. According to the results, the last crack-
down of a consecutive series of violent law enforcement over 40 days generates an
immediate reaction of about 5.2 attacks by DTOs against state authorities. By the

392
end of this punitive campaign, state security forces would have suffered a cumulative
count of 172.2 retaliations from DTOs. This aggressive behavior from criminal or-
ganizations is more intense when criminals react to violent law enforcement than to
non-violent enforcement tactics.
The difference in the magnitude of the response from criminal organizations sug-
gests that DTOs are somewhat more tolerant to non-violent than to violent security
tactics. However, when the government authorities use lethal force against them, or-
ganized criminals launch an overwhelming reaction against the state. What explains
such an aggressive reaction from criminal organizations?
Interviews conducted during fieldwork suggest several mechanisms to explain the
intensity of this behavior. One explanation argues that organized criminals have
greater firepower than most state security forces, especially at municipal and state
level. Criminals can easily smuggle military style weapons across the U.S. border
to supply their forces with assault rifles and grenade launchers. This argument is
consistent with the argument of Dube et al. (2013) about the increase of violence in
Mexico after the expiration of the Federal Ban on Assault Weapons in the U.S. The
argument about sophisticated weaponry also has resonance in a larger theoretical
explanation about the role of material resources for conducting collective dissent
(McCarthy and Zald, 1977). The ample financial resources derived from illicit markets
give DTOs the ability to collect sufficient military, human and material resources to
launch aggressive offensives against the state. A similar use of illicit contraband has
been identified as an important factor to explain the duration of civil wars (Fearon,
2004).
At the sub-national level, heavily armed criminals outgun poorly equipped police
forces who usually carry old shotguns and small caliber handguns: weakly armed
security forces become easy targets of criminal violence. This argument is supported
by news reports stating that municipal police are often outgunned by criminal orga-

393
nizations who usually ambush them (Redacción de AFP, 2012), or use death squads
to hunt and kill municipal police officers (Pedraza, 2013), often using extreme forms
of violence when killing them (Redacción del Universal, 2010). The weapons superi-
ority of criminal organizations was recognized by federal authorities, and motivated
President Calderón to launch a program called Subsidio para la Seguridad en los
Municipios (SUBESMUN) in 2008 (see Secretarı́a de Gobernación, 2013). As often
reported by the press, most SUBESMUN resources are used for buying weapons and
ammunition to augment the tactical equipment supplies of local municipal police
forces (Mexico Seguridad, 2013). Future research should examine whether improve-
ments in the punitive capacity of the state at sub-national level had an effect on
deterring retaliatory attacks by DTOs.
Other witnesses are skeptical about the argument of criminal military superiority.
Although these interview subjects recognize that DTOs have high caliber weapons,
they point out that Army and Navy military personnel, and Federal Police officers,
are much better equipped and trained than state and municipal police forces. There-
fore the argument of military superiority of criminal organizations does not apply
equally to security forces at different levels. These contrasting arguments suggest an
area of future research; namely, disaggregating and identifying reciprocal actions and
reactions between criminal organizations and security forces at different levels.
Another not mutually exclusive argument suggests that criminal organizations
can easily attack the security forces of the state not only because some of them
are militarily inferior but because they are easy to identify and ambush. There
are several news reports about security forces being ambushed by criminal groups
(Covarrubias, 2010; Notimex, 2010; Redacción de AFP, 2012). DTOs usually rely on
a large network of spotters spread across their territory to provide timely information
about the situation on the ground. In the criminal jargon these type of informants
are known as “halcones” (falcons). The halcones are not usually employed for drug

394
trafficking activities or for committing violence; their job is simply to stay alert and
immediately notify their superiors whenever they spot an Army convoy, police car
or any other kind of relevant “unusual” activity. Criminal organizations usually
make use of taxi drivers as spotters (Martı́nez, 2013). These taxi driver informants
are called “garrapatas” (ticks) and they contribute by spotting or following security
forces (Prados, 2013). According to witness accounts from police officers interviewed
in this research, using taxis as spotters is a convenient strategy for DTOs because taxi
drivers are a legal business and constitute a large network of observers roaming the
streets. If the taxi notices that police officers realize that they are being followed by
a garrapata, the taxi usually drops the chase and is replaced by a different taxi from
the same network a few blocks ahead.3 There are reports that DTOs also use corrupt
transit police officers as informants (La Policiaca, 2013). In addition, according to
Bishop Raúl Vera, a religious leader and well-known human rights activist, criminal
organizations also recruit children approximately aged twelve to work as halcones;
they give them a cell phone and pay them $50 dollars a week (Proceso, 2013).

Moderate Use of State Violence

Another interesting finding of the statistical analysis is that conflict between
DTOs increases the punitive response from government authorities, but the magni-
tude of the response varies by government tactics. Criminal competition substantially
increases non-violent responses from the state in the form of arrests and seizures of
drugs, weapons and assets. However, the violent response from the state is moder-
ate. This finding contributes to a broader literature on state repression analyzing
the different tactics used by government authorities to confront challengers. One
possible explanation of the moderate use of state violence as a response to conflict

3
In these cases, police officers usually turn off their radio communications and switch to cell
phone because it is likely that the official frequency is being tapped by criminals. They then call
for reinforcements to put an unit behind the garrapata that is following them. This leads to what
police officers call a “who-follows-whom” sequence.

395
between criminal groups may be rooted in the political characteristics of the state.
The “domestic democratic peace” argument advanced by Davenport (2009b) claims
that democratic political systems are less repressive than authoritarian regimes for
three main reasons: democracies have alternative ways of imposing control, leaders
suffer political sanctions for engaging in violent repression, and democracies have
stronger institutions to control coercive power. The analysis of different repressive
tactics also indicates that democracy is more effective in limiting the use of lethal
force than other non-violent restrictions (Davenport, 2004). This argument is consis-
tent with the discussion about the different costs of applying violent and non-violent
enforcement tactics in Section 2.4 of the theoretical chapter. Although there is dis-
cussion about the extent of the democratic consolidation of the Mexican political
system, there is a broad consensus among scholars and analysts that Mexico is a
democratic system, especially after the political alternation of the executive in 2000.
Democratic norms or institutions might thus help to explain the moderate use of vio-
lent law enforcement employed by the Mexican state to counter the wave of violence
among criminal organizations.
An alternative but not mutually exclusive explanation of the moderate use of
violence to impose order among DTOs could be made on the basis of rational choice
calculations about the costs and benefits of state actions (Lichbach, 1987, 1998; Ma-
son and Fett, 1996). According to this approach, state authorities moderate the use
of violent enforcement based on the strategic expectations about the deleterious ef-
fect of violent law enforcement. The results indicate that the use of violent tactics to
fight crime generates a large increase of violence between criminal organizations and
also motivates a substantial number of counterattacks against the state. Probably,
the pernicious consequences of violent law enforcement motivate government author-
ities to self-restrain the use of lethal force when trying to contain conflict between
DTOs. In any case, the empirical finding about the moderate use of state violence

396
in the Mexican war on drugs should be explored by further theoretical and empirical
analysis.

The Languid State

Finally, one of the most surprising findings of this chapter is the languid behavior
of the Mexican state when directly challenged by criminal organizations. The empir-
ical results of this chapter show that even after 40 continuous attacks perpetrated by
criminal organizations against the state, government authorities do not seem to react
with violent or non-violent tactics against criminals. The testimony of one high-level
police officer interviewed in this research is illustrative of this finding. When asked
about the apparent languid behavior of the state he was not surprised and said “We
do not fall for provocation. Of course, we respond if we are attacked. But my men
go out there to do their duty, not to look for revenge.”
This result is puzzling because the literature on state repression has invariably
found that perceived or actual threats to the state trigger repressive reactions from
government authorities. Studies on the dissent–repression nexus have found mixed
results for the effect of state coercion on dissent. The use of repression sometimes
deters challengers and other times exacerbates contentious behavior (Lichbach, 1987).
However, as claimed by Davenport, Johnston and Mueller (2005, viii), “moving in
the opposite direction, from protest to repression, only one relationship has ever been
identified: dissent increases the application of state coercion.” The consistency of this
finding is such that “in every statistical examination of the subject, dissent increases
repressive behavior” (Davenport, 2009b, 39).
Part of the explanation might lie in the self-restraining argument of the “domestic
democratic peace” discussed earlier. The statement offered by the high-level police
officer stressed the role of duty and institutions in guiding the behavior of security
forces. Another part of the explanation about the languid behavior of the state
might also rest on the non-political nature of criminal organizations. As discussed in

397
Section 1.2, organized criminals are primarily motivated by economic goals, and use
violence for preserving the power structures that enable them to extract rents from
illegal markets. It may be that government authorities do not react with the severity
and alacrity usually expected by researchers on state repression because criminal
organizations do not represent a direct threat to the political system. In any case,
this finding about the languid behavior of the Mexican state in reaction to direct
attacks by criminal groups calls into question the universal applicability of the “law
of coercive responsiveness” and suggests the need for further theoretical and empirical
investigation.

398
 CHAPTER
8

CONCLUSION

This research provides theoretical foundations and solid empirical evidence spec-
ifying the conditions of why, how, when, where, by whom, to what extent and for
how long violence constitutes a valuable form of behavior for organized criminals.
Located in a Hobbesian tradition of conflict, this research shows that violence in
Mexico emerged as the consequence of the collapse of order. In the absence of regula-
tion mechanisms, the disrupting effect of state actions unleashed a war of all-against-
all. As indicated by the central argument, democratization eroded the system of
incentives that had allowed a peaceful configuration between the state and crimi-
nal organizations for several decades. Under the new system of political incentives,
democratization gave authorities motivations to fight crime, which triggered an un-
precedented wave of violence between the state and organized criminals and between
rival criminal groups fighting to control drug-strategic territories.
A central contribution of this study is the integrated effort of conceptualizing,
measuring and estimating violence as a dynamic and interactive manifestation of
conflict operating within structural factors. In order to understand the escalation
and geographic concentration of violence, it is necessary to grasp both its external

399
factors and its internal processes. This research provides theoretical reasons and em-
pirical evidence showing that the largest share of violence is caused by confrontations
between rival criminal groups trying to seize control over territories that grant access
to enormous economic benefits. In consequence, violence tends to cluster around
valuable areas favorable for the reception, production and international distribution
of drugs. The opportunities to seize control over those territories are primarily gener-
ated by the state through the implementation of punitive law enforcement operations
that weaken target criminal organizations and indirectly empower their rivals.
This research addresses the urgent need to understand the wave of drug violence
in Mexico by analyzing three distinct, yet interrelated aspects of this conflict; the
onset, escalation and geographic distribution and concentration of organized criminal
violence.

8.1
Why
do
Politicians
Decide
to
Fight
Criminal Organizations

After
Having
Peacefully
Coexisted
With
Them?

Theoretical expectations
To explain the onset of the war on drugs, the theoretical explanation claims that
democratization disrupts the peaceful configurations that enable coexistence between
corrupt government authorities and criminal organizations in contexts of authoritar-
ian rule. In addition, democratization motivates politicians to enforce the law against
criminals. To sustain this claim, the formal model argues that in an authoritarian
regime government authorities receive more benefits from corrupt agreements than
from enforcing the law against criminals. In contrast, political actors in a democratic
system obtain more political benefits from enforcing the law than from making and
adhering to corrupt agreements with criminals. This argument is based on a set

400
of premises relating to the number of political actors and temporal expectations of
tenure in office.
Authoritarian regimes are characterized by a reduced number of political actors
and the prospect of long terms in power due to the lack of elite circulation. The small
number of political actors reduces the costs of collective action, thus facilitating bar-
gaining and reducing the costs of bribes. A limited number of actors also facilitates
collective action among corrupt politicians, reduces the chances of defection, and in-
creases the probability of sanctions for non-compliers. A cohesive hierarchical chain
of command facilitates the implementation of agreements and compliance across the
government structure. The lack of effective elite circulation through electoral means
favors credible expectations about future benefits for complying with the pact. Fi-
nally, the likelihood of a long term in office increases the credibility of future sanctions
for not complying with the agreement. The direct implication of this system of in-
centives is the lack of enforcement against crime.
In contrast, democratic regimes are characterized by a large number of political
actors with time-limited terms in office due to effective elite circulation. The increased
number of actors augments the costs of collective action and the costs of bribes. A
large number of actors also reduces the possibility of monitoring defection and the
probability of sanctioning non-compliers. The entrance of new actors at various
levels of government disrupts the chain of command, thus reducing the feasibility
of implementing corrupt agreements. In addition, effective elite circulation through
electoral means reduces the duration and stability of agreements. Elections also
introduce uncertainty about the next actor in office, thus hindering the expectations
of long-term agreements.
Democratization also takes the incentives for political survival out of the hands
of the elite and makes the prospects for their political careers dependent on the
support that politicians can win from the electorate. Democratization thus generates

401
personal incentives for politicians to provide public goods such as public security. In
addition, new political actors have incentives to enforce the law as a way to signal their
difference from old corrupt politicians. The theoretical explanation also argues that
government authorities obtain political benefits from implementing harsh security
policies in times when their legitimacy is threatened by political strain. These factors
all serve to erode the prospects of peaceful coexistence between political actors and
organized criminals and motivate government authorities to fight crime.

Empirical results
This research presents both qualitative and quantitative evidence in support of the
argument about the onset of the Mexican war on drugs. Evidence shows that democ-
ratization increased the number of relevant political actors at all levels of government
and favored the effective circulation of political elites. In consequence, democratiza-
tion subverted the political structures that enabled the existence of non-aggressive
agreements between authorities and criminals. In addition, democratization sub-
verted the preexisting system of political incentives and generated a new configuration
favoring law enforcement, thus motivating government authorities to fight crime.
The process tracing analysis elucidates the historical sequence that led to the
emergence and consolidation of a system of incentives that favored the peaceful co-
existence between criminals and corrupt government authorities during the political
hegemony of the PRI. This analysis shows how political order emerged after the Mex-
ican revolution with the formation of the PRI as a political agreement to regulate
access to power in a peaceful manner. This pact gave relative autonomy to local
leaders at the periphery, who benefited from the economic opportunities created by
Prohibition in the U.S. At the peak of its political hegemony, the PRI imposed order
on the criminal sector as well as in the political sphere. The political elite central-
ized an extensive network of criminals not only to obtain economic gains from illicit
sectors but also to monitor clandestine political activities that could threaten the

402
regime. To instill and impose discipline on criminals, the PRI relied on a centralised
hierarchical system of political incentives aligned with the overall government and
party structures. The effectiveness of this system depended on the concentration of
power in the political elite and on the hegemonic presence of the party in all govern-
ment branches across levels of government. In addition, the recurrent use of electoral
fraud and mass electoral mobilization through vote buying and clientelism prevented
the circulation of political elites outside the PRI, thus favoring the stability of the
pacts and incentives for long-term compliance.
The historical evaluation also sheds light on the erosion and subsequent collapse
of order. According to the analysis, the process of democratization slowly but sub-
stantially eroded the hegemony of the PRI and subverted the system of political
incentives that had enabled it to maintain order in the criminal sector. Thirty years
of gradual electoral reforms allowed the entrance of opposition parties at all levels
of government and the effective circulation of elites. The diversity of party labels at
the federal and local levels disrupted the chain of command that made those agree-
ments feasible. Sound rules of electoral competition dissolved the certainty of a PRI
victory and introduced uncertainty about establishing corrupt agreements with the
new political actors. By the time the presidential transition occurred in 2000, po-
litical diversity at sub-national level was entrenched. Most importantly, increasing
political competition infused by democratization generated personal incentives for
authorities to fight crime in an effort to gain citizen support. Finally, the historical
assessment analyzes the conditions under which president Calderón dispatched the
Army to fight drug organizations in an effort to boost his legitimacy after a contested
election tainted by accusations of fraud. This massive deployment of troops gener-
ated a profound disruptive effect on the already precarious equilibrium among drug
trafficking organizations in Mexico.

403
 The statistical analysis also provides strong support for the theoretical expecta-
tions about increased law enforcement caused by democratization. The quantitative
analysis indicates that increasing the effective number of political parties at the presi-
dential level increases the use of violent law enforcement against criminals. Increasing
the effective number of political parties also increases the number of arrests, seizures
of criminal assets, events of drug interdiction and seizures of weapons. In addition,
as expected from the theory, a change from a unified government at federal and sub-
national level generates a substantial increase in the levels of violent enforcement,
although its effect on non-violent tactics is less acute. Finally, the statistical analysis
provides a surprising finding about the effect of political strain on law enforcement.
The results indicate that a narrow margin of electoral victory increases the levels
of violent law enforcement but has the opposite effect on non-violent tactics. This
finding suggests that violent enforcement is used as a last resort in periods of politi-
cal strain. In general, the statistical analysis supports the theoretical argument that
democratization erodes peaceful agreements and motivates government authorities
to fight crime.

Implications
To some sectors of the population, the authoritarian past came to be remembered
in a more favorable light when the PRI hegemony managed to maintain order and
security. During the presidential campaign of 2012, billboards on the streets presented
slogans presumably attributed to the PRI saying “against the ineptitude of dealing
with the drug trafficking, the experience of bargaining with them (PRI)” (Robles
de la Rosa, 2012). Even president Calderón expressed his concern about the PRI
returning to power and falling into a corrupt relationship with organized crime as
an effort to pacify the country. As Calderón stated in an interview with the New
York Times, “there are many in the PRI who think the deals of the past would
work now” (The New York Times, 2011). Eventually, in July 2012, the PRI won

404
the election by a comfortable margin of 6.6 percent against the PRD, whereas the
PAN was relegated to a distant third position. The newly elected president, Enrique
Peña Nieto, repeatedly stated that his administrations has no intentions to negotiate
with drug trafficking organizations (Villamil, 2012). However, the question still is
relevant. Would the PRI, or any other party in the presidency, be able to negotiate
peace with criminal organizations in a democratic context? Probably not. The
protracted process of democratization eroded the authoritarian characteristics of the
PRI dominant era well beyond the presidential office and deeply affected the system
of incentives thought the entire political system.
There are two key ingredients that would be necessary for a non-aggression agree-
ment to be feasible: political homogeneity across government tiers and a long time
horizon for the stability of such pact. None of those factors exist in the current demo-
cratic political system in Mexico. The diversity of political actors is well entrenched
at the municipal level, as well as in elections for governors and state legislatures. The
plurality of political actors is also deeply rooted at the federal level in the Chamber of
Deputies and the Senate. With such a large number of political actors across govern-
ment tiers, it would be incredibly difficult to coordinate the necessary actors and align
their interests behind a consistent effort of government authorities to negotiate with
organized criminals. Even if such complex coordination is achieved, the recurrent
elections at the municipal, state and federal levels are likely to bring new political
actors to government positions, which would require renewed efforts of coordination
from the government side. In addition to the characteristics of political plurality
and elite circulation in a democratic regime, the current structure of the organized
criminal sector might not be conducive to negotiations with government authorities.
In contrast to the reduced number of large criminal organizations operating in the
country during the heydays of PRI hegemony, the Mexican war on drugs witnessed
the multiplication and expansion of criminal organizations. The plurality of criminal

405
organizations would impose enormous difficulties of coordination and alignment of
incentives for a peaceful agreement with government officials. In consequence, such
a pact is not likely to be feasible.

8.2
Why
Does
Drug-Related
Violence
Escalate so
Rapidly?

Theoretical expectations
The escalation of violence is based on the mechanics of a contest success model for
territorial competition. According to the theory, increased levels of law enforcement
trigger an escalation of conflict between authorities and criminals, and violence be-
tween rival criminal groups. The action of the state has a highly disruptive effect on
the relative military balance among criminal organizations. Law enforcement actions
weaken the capability of a criminal group to protect its territory, which can moti-
vate an invasion from a competing criminal group that observes a weaker rival. The
equilibrium conditions indicate that violence committed by criminal organizations
– against either the state or a rival group – primarily depends on the value of the
territory in dispute, although another determining factor is the severity of military
damage inflicted on them and their ability to recover from it. The core element of
the model is that criminals will engage in violence as long as the economic benefits
obtained from capturing or defending the territory are worth the fight. The model is
used for deriving a set of empirical implications that are tested in the data. The two
main hypotheses are that law enforcement triggers violent competition between crim-
inal organizations and that it also generates criminal retaliation against the state.
The basic derivatives of the model are extended to a generalized punitive campaign
where the state simultaneously battles several criminal groups. The implications re-
main consistent with respect to the disturbing effects of enforcement on competition

406
among criminals and violent retaliation against the state, but suggest that law en-
forcement will generate a substantially larger wave of violence between rival criminal
organizations than will attacks against the state. Together, the synergy of these
mechanisms has produced a Hobbesian war of all-against-all.

Empirical results
An empirical evaluation of the micro-mechanisms of violence suggested by the
theory requires dealing with the challenge of endogeneity. To minimize this risk, the
identification strategy relies on a quasi-experimental instrumental variables research
design. The statistical analysis provides strong confirmation of the deleterious effect
of law enforcement on violence between criminal groups. Increasing the levels of
violent law enforcement generates an exponential increase in the number of violent
confrontations among rival criminal organizations. In addition, the results reveal
that non-violent tactics also generate violence between criminal groups. Moreover,
the effect of non-violent tactics is stronger than the effect of the use of lethal force.
In particular, drug interdiction and seizures of weapons generate dramatic spikes of
violence. The statistical model also confirms the theoretical expectation for the effect
of military damage – measured by U.S. production of assault weapons – and recovery
capability – measured by unemployment – on the levels of violence among criminals.
The statistical analysis also relies on time-series techniques to assess the dynamic
and reciprocal interactions between the various processes of violence over time. Using
forecasting techniques of vector autoregressive (VAR) models and cumulative orthog-
onal impulse response functions (COIRF), this approach shows that law enforcement
actions have a highly disturbing and enduring effect. The results show that a sus-
tained campaign of violent law enforcement conducted over several days generates
a dramatic escalation of violence among organized criminals that accumulates over
time. In addition, sustained law enforcement efforts relying on the use of lethal force
generate a cumulative aggressive reaction from criminals against government authori-

407
ties. However, the effect of law enforcement on violent competition between criminals
is substantially larger than the effect on criminal retaliation against the state. The
results provide strong support for the theory in terms of the direction and magnitude
of the disturbing effect of law enforcement and generate new insights about its lasting
effects. The analysis of the temporal dynamics of conflict also provides novel find-
ings about the state’s response to violent criminal behavior. The results show that a
sustained wave of confrontations between criminals generates a larger reaction from
the state in terms of non-violent tactics than in the use of lethal force. In particular,
government authorities primarily react by arresting criminals and seizing drugs, but
state violence is rarely used. One of the most surprising findings of the dynamic anal-
ysis reveals the languid behavior of the state when directly challenged by organized
criminals. The results show that a sustained campaign of attacks against the state
generates no response from government authorities in terms of violent or non-violent
tactics, and actually manages to deter government efforts to seize drugs.
Finally, the statistical analysis reveals that most structural factors suggested by
alternative explanations offer limited analytical leverage for understanding the highly
dynamic characteristics of violence. This finding confirms the importance of concep-
tualizing and empirically evaluating violence from a micro-dynamic perspective. The
results show no support for the explanations usually advanced by government au-
thorities to justify the war on drugs. The argument about the erosion of the social
fabric – measured by the number of divorces and the percentage of young mothers
(ages 12–19) – shows no impact on the levels of violence among criminals. The ar-
gument about increased drug consumption in Mexico – measured by the number of
hospitalizations due to drug abuse – fails to find support across any model specifi-
cation. The prices of drugs in the U.S.market do not show any effect on the levels
of criminal violence. Finally, the statistical model reveals that violent competition
between criminal groups is more intense in states characterized by high economic

408
development than in poor states. The result is robust across different model spec-
ifications and different law enforcement tactics. This finding is consistent with the
theoretical expectations of the model about the value of strategic territories. Most
importantly, this finding challenges mainstream explanations of political violence and
criminal behavior emphasizing the role of low levels of economic development as a
key determinant of violence.

Implications
This research provides a sound theoretical explanation supported by rigorous em-
pirical evidence about the deleterious effects of implementing quasi-military strategies
to fight drugs. The unprecedented use of force to counter organized criminal groups
launched by president Calderón triggered a massive wave of drug related violence in
Mexico. The analysis of different law enforcement tactics conducted in this research
can help to address some policy recommendations for government authorities.
Results invariably show that violence delivered by the state has a disruptive ef-
fect on criminal organizations as it is capable of generating spirals of violence among
rival criminal groups. The basic intuition behind the rationalist approach of law
enforcement advanced by Becker (1968) indicates that intensifying the severity and
probability of sanctions against criminals increases the costs of engaging in criminal
activities, thus expecting to reduce such behavior. However, this rationalist approach
is wrong for explaining large scale organized criminal violence. Instead of imposing
unbearable costs – as high as the cost of being killed – the use of violent law enforce-
ment generates opportunities for rival criminal organizations to launch an invasion
on a weakened target organization in order to control a valuable territory, thus in-
stigating violence among criminal organizations. In consequence, the efforts to fight
crime through the use of state violence have the counter productive consequence of
exacerbating violence among criminals.

409
 The empirical assessment in Chapter 6 shows that drug interdiction triggers
episodes of violence among criminal organizations. However, as discussed in that
chapter, the prevalence of drug consumption in Mexico is remarkably low. The vital
problem in Mexico is not drug use, the problem is violence. Instead of focusing on
seizing drugs, which in turn generate more violence against the state and among
criminal organizations, the priority for government authorities should be on reducing
the levels of violence.
The statistical analysis also provides evidence about the importance of the capa-
bility of inflicting military damage of criminal organizations. In particular, results
indicate that the increased availability of assault weapons in the U.S. is associated
with higher levels of violence in Mexico. The right to keep and bear arms in the
U.S. is a highly political issue and efforts of gun regulation face fierce opposition
from citizens, interest groups and political actors. In this sense, efforts of Mexican
government authorities to lobby in favor of gun control would not only encounter the
indifference of large sectors of the U.S. public about the deaths generated by these
guns on the Mexican side, but might also face the discredit from gun supporters and
the political consequences for tying to intervene on such a sensitive issue. However,
there are a few things Mexican government authorities can do for trying to reduce
the flow of assault weapons into Mexico. One strategy might be requesting U.S. gov-
ernment authorities to increase the number of southbound security check points on
main border crossing areas. Another strategy could be to strengthen security checks
on the Mexican side of the border and customs in an effort to reduce the flow of illicit
weapons.
The responsible implementation of public policies in any sector should not ignore
the negative consequences of conducting such policies. Economists refer to these
effects as “negative externalities” and political scientists use the term “unintended
consequences.” Regardless the specific terms used, the fact is that the implementation

410
of a large scale punitive policy against organized crime generated an unprecedented
wave of violence in Mexico. Responsible government authorities must not neglect the
consequences of implementing security policies with these characteristics. Rendering
the social, economic and, most importantly, the human costs as acceptable casualties
in service for a political cause would certainly constitute a perverse calculation.
Governments face no easy policy choices in terms of confronting organized crime.
However, due to the deleterious consequences of the implementation of a large-scale
punitive campaign against criminal organizations in Mexico, it is time to consider
alternative security policies prioritizing harm reduction. A promising alternative is
the implementation of a “dynamic concentration” approach to fight crime proposed
by Kleiman (2009). The basic intuition behind this security policy indicates that if
government authorities are not capable of controlling all different criminal organi-
zations at the same time, it is preferable for authorities to concentrate their efforts
on a specific criminal organization. To do so, Kleiman proposes a strategy based on
thee central axis. First, it is crucial to define a specific criteria to target criminal
organizations. For the Mexican case, the criteria should be the levels of violence,
not the size of the organization nor the amount of drugs it transports. In doing so,
authorities should focus on the most violent criminal organizations. Second, author-
ities must effectively communicate the criteria and the organizations that meet such
characteristics. In that way, authorities would indicate that they are going after the
most violent organizations. Third, instead of fighting all criminal groups at the same
time, a more efficient use of limited capabilities is to concentrate law enforcement
efforts on the selected criminal group. In this way, the probability and severity of
state actions against the selected target become more credible. Once violence from
that specific organization is reduced, authorities can move on to the next one in the
list. As indicated by Kleiman (2011), when deterrent threats are sufficiently credible
and clearly communicated, they generate incentives for criminal groups to reduce the

411
overall levels of violence to the extent that the dynamic concentration approach does
not need to be carried out very often. Security analysts such as Guerrero (2010a)
consider that the implementation of Kleiman’s dynamic concentration approach is
an attractive option for the Mexican case, yet there are no clear signs that Mexican
government authorities are reorienting their security policies towards this approach.

8.3
Why
is
Violence
More
Concentrated
in
Some Areas
Than
in
Others?

Theoretical expectations
In order to explain the geographic distribution and concentration of organized
criminal violence, the model explicitly incorporates a measure of territorial value as
a key determinant of conflict. Departing from the assumption that criminals are
primarily motivated by economic gains, the model indicates that criminal organiza-
tions are willing to engage in violent confrontations to capture or defend strategic
territory that give them access to profitable illicit activities. In this way, the value of
a territory serves as a contextual factor comprising the highly dynamic and interac-
tive micro-mechanisms of violence between the state and criminals and among rival
criminal organizations. The equilibrium conditions of the model indicate that drug
trafficking organizations will engage in violent behavior as long as the economic ben-
efits of capturing or defending a specific territory are worth the cost. In consequence,
violence tends to concentrate in and around drug-valuable territories.

Empirical results
The statistical analysis also provides support for the theoretical expectation ar-
guing that valuable territories are likely to experience higher levels of violence among
criminal organizations. The results indicate that violence among criminal organi-
zations is more intense in areas favorable for the production of drugs, reception of

412
shipments along the Gulf of Mexico and the Pacific coast, and at entry spots to the
U.S. market located along the northern border of Mexico. In addition, the results
show that violence between criminal groups increased after the September 11 terrorist
attacks when the U.S. increased security along the border. The statistical analysis
provides another remarkable set of findings, namely that the disrupting effect of vi-
olent law enforcement on criminal competition is contingent on the strategic value
of the territory. The results indicate that increasing the use of violent enforcement
in areas of high production of illicit crops generates more violence between crim-
inals than in areas that do not produce drugs. In addition, the intensification of
enforcement along the Pacific and the Gulf coasts generates substantially more vio-
lent competition between organized criminal groups than in territories away from the
coast. Finally, increasing the use of lethal force along the U.S.–Mexico border gener-
ates more violence in areas favorable for international distribution of illicit drugs than
away from the border. In all these cases, as government efforts to fight crime intensify
in strategic areas favorable for drug-related activities, the levels of violence between
criminal organizations increase exponentially. These findings provide strong support
for the assumption that criminal organizations are primarily motivated by economic
incentives. The results also show that the opportunities for confrontations between
criminals generated by law enforcement have more disrupting effects in valuable ter-
ritories than in non-strategic areas. Finally, these findings reinforce the importance
of integrating the macro and micro-determinants of violence in order to understand
the dynamic mechanisms of conflict contained in large structural factors.

Implications
Due to its geographic location next to the U.S., Mexico is particularly affected by
the economic attractiveness of the world’s largest drug consumption market. As long
as there is a large demand of drugs in the U.S. and anti-drug regulations in the U.S.
consider drugs as illicit commodities, there will be drug traffickers willing to kill and

413
die for obtaining the enormous profits associated with supplying the U.S. demand
of drugs. The economic benefits generated by illegal markets constitute the driving
force behind the strategic value of some territories favorable for the reception, pro-
duction and international distribution in Mexico. In addition, it has been argued that
the strategic importance of Mexico as a drug-trafficking route increased as the U.S.
increased the aerial and maritime monitoring of drug transportation routes trough
the Caribbean. Unfortunately, in the short term, it is not realistic to expect that the
U.S. will conduct a radical change in its drug policies by focusing on attending the
consumption as a problem of public health rather than a national security concern,
or by decriminalizing drugs at the federal level. This situation, represents a challenge
for Mexican government authorities since the economic value of drug-valuable terri-
tories in Mexico is externally defined by U.S. drug policies and the characteristics of
international drug routes.
However, there are a few things that Mexico could do to reduce the risk of violence
associated with these strategic territories. As indicated in this research, increasing
law enforcement in drug valuable areas generates intense waves of violence among
criminal groups. The direct implication of this result is the recommendation of min-
imizing disruptive law enforcement actions in drug-valuable territories as a strategy
to reduce violence. In addition, Mexican government authorities would benefit from
engaging in a serious, evidence-based and open debate about adopting alternative
legal and regulatory drug regimes in the country. More importantly, Mexican gov-
ernment authorities should play a more active role the in ongoing efforts led by the
Organization of American States (OAS) on rethinking drug policies from a coordi-
nated and regional perspective. These efforts include exploring possible scenarios
for implementing alternative drug regulatory frameworks and even entertaining the
possibility of abandoning the use of punitive counter-narcotic policies in some coun-

414
tries heavily affected by the production and transportation of drugs (Organization of
American States, 2013).

8.4
Contributions

Theoretically, this research redefines the study of organized criminal violence

within a broader literature of conflict research by calling the attention of scholars
to the study of this pervasive and highly lethal form of violence and by providing
direct contributions to the field. Based on an explicit set of scope conditions and as-
sumptions, this research advances a formal model for the micro-dynamics of conflict
among multiple actors operating within structural contexts. The emphasis on theo-
retical desegregation allows to specify the conditions for the behavioral manifestation
of violence. According to the model, opportunities for violence among rivals are unin-
tentionally generated by the intervention of a third actor. This analytical framework
could help to understand the challenges of combatting organized criminal violence in
the developing world, especially in new democracies. In addition, the mechanisms of
conflict laid out in this model could contribute to analyzing the micro-dynamics of
violence in other types of conflicts.
Empirically, this research contributes to the analysis of conflict by presenting
“Organized Criminal Violence Event Data in Mexico 2000–2010” (OCVED), a novel
data set of event data comprising daily geo-referenced information on who did what
to whom, when and where in the Mexican war on drugs. With more than 9.9 million
observations covering all Mexican municipalities over a time span of eleven years,
this database contributes to the recent trend of using “big data” for conflict research.
The use of fine-grained data also provides unprecedented analytical leverage for test-
ing the micro-mechanisms of violence derived from the theoretical model, thus con-
tributing to the efforts of the EITM program bridging theoretical developments and

415
empirical inferences. As discussed in the statistical assessment, the use of micro-level
data reveals the highly dynamic and interactive character of violence and challenges
mainstream explanations that focus primarily on structural factors.
Methodologically, this research relies mainly on a rigorous quantitative assessment
of violence, but it also incorporates deep insights gathered through several months of
fieldwork in Mexico. The research makes a deliberate effort to address the empirical
challenges of causal inference generated by the endogenous relationships between dy-
namic and interactive processes of violence. The use of a quasi-experimental research
design helps increase confidence about the causal inferences generated by the statisti-
cal assessment. In addition, the use of impulse–response analysis helps to disentangle
the dynamic and lasting effects of different processes of violence.
Technologically, this research brings together cutting-edge advances in computer
science with recent methodological developments in conflict research to develop Even-
tus ID, a novel automated protocol for coding event data from text in Spanish. This
software helps researchers to reduce the financial and labor costs associated with man-
ual coding strategies, thus favoring the creation of new databases on a wide variety
of topics. In addition, the accuracy of event coding from sources in Spanish generates
an unprecedented possibility for coding timely, detailed information written in this
language.
Substantially, this research contributes to the task of observing and understand-
ing organized criminal violence in Mexico. It offers a robust explanation and solid
empirical evidence about the causes, mechanisms, magnitude, scope, and length of
the wave of criminal violence caused by the implementation of a full-fledged mili-
tary campaign to fight crime in Mexico. More broadly, the results of this research
challenge the international paradigm that encourages the implementation of punitive
strategies to fight drugs.

416
8.5
Looking
Ahead

This research has accomplished its goal by providing some answers to urgent
questions about the onset, escalation and concentration of organized criminal violence
in Mexico. The theoretical foundations and empirical results should be subjected to
close scrutiny by other researchers in order to advance our understanding of this
complex and pressing phenomenon. As this research seeks to provide some answers,
it also opens the door to further theoretical and empirical questions. The extension of
this project constitutes a research agenda in its own right that will be accomplished
through four tasks: bridging, disaggregating, expanding and reversing.

Bridging
The development of the study of the micro-dynamics of organized criminal vio-
lence requires building a bridge between carefully crafted theoretical developments
and fine-grained evidence. Some of the empirical findings that emerged out of this
research set a base for further theoretical development. The most important of these
findings is the seemingly languid behavior of the state when directly challenged by
criminal organizations. In contrast to the theoretical expectation of the “law of co-
ercive responsiveness,” we should expect the state to react with the use of repression
to counter or neutralize hostile actors. However, the results indicate that the state
barely reacts when directly attacked by criminal organizations.
Another area of future research requires building bridges between diverse method-
ological approaches to the study of organized criminal violence. With the exception
of the macro-historical analysis of the emergence, consolidation and collapse of order,
this research relies heavily on the quantitative analysis of violence. Further exten-
sions of this work will incorporate the qualitative analysis gathered during fieldwork
to disentangle fine-grained causal mechanisms operating behind the general trends
identified through quantitative analysis. However, the integration of qualitative in-

417
sights and quantitative regularities faces the challenge of acquiring systematic and
reliable qualitative information in the midst of an active conflict. Journalists can
testify to the lethality of drug-related violence against “those who know too many
details.” Therefore, research should be conducted maximizing security conditions.

Disaggregating
Further research on drug-related violence will also involve theoretical and empir-
ical disaggregation. One of the directions that this research will take is towards the
disaggregation of criminal organizations. So far this research has treated criminal
organizations as a homogeneous category. This decision was primarily based on the-
oretical grounds with the objective of generating a general encompassing explanation
about their motivations and behavior that goes well beyond narratives relating to the
current period. However, clearly not all criminal groups are the same. After identify-
ing a set of common characteristics of criminal organizations, the next step will be to
focus on analyzing variation between the different criminal groups in terms of their
levels of violence, the brutality of their tactics, the different histories of territorial
expansion and their different reactions towards law enforcement.
Another direction disaggregation will take is to look more carefully at the state.
Future research will analyze the behavior of the different federal and sub-national
security forces and agencies. Further research should analyze variations in their use
of violent and non-violent tactics, as well as the effect of those tactics on criminal
behavior. A breakdown of state actors also requires generating theoretical ideas and
empirical information about principal–agent relations between political authorities
and security agents across the different levels of government.

Expanding
One of the methodological contributions of this research is the development of
Eventus ID. This tool can be used to expand the present work in various directions.
The most immediate task will be to update the data base of organized criminal vio-

418
lence in Mexico. This is crucial for providing a public good to analyze the dynamics
of violence in Mexico in a timely manner. This task is particularly urgent because
at the time of this writing, the Mexican government had recently stopped generat-
ing and updating publicly available data on organized criminal violence. Another
task will be to launch an ambitious collaborative project for replicating the event
coding strategy implemented for Mexico in this research in other Latin American
countries. This task would provide a massive amount of fine-grained data to conduct
comparative analysis of organized criminal violence across Latin America. Such an
empirical platform would set the foundations for theoretical and empirical develop-
ments about the domestic and transnational behavior of criminal organizations and
different state responses. A third research direction would be to expand the use of
Eventus ID to code different types of event data to analyze dissent–repression dynam-
ics in Latin America. In addition, contingent on collaboration with other regional
exports, Eventus ID’s coding algorithms could be adapted for analyzing information
in other languages.

Reversing
This project is primarily focused on analyzing the causes of large-scale organized
criminal violence. However, future research developments will focus on “reversing
the terms of the equation” in order to analyze the consequences of drug-related vio-
lence. Studying the effects of drug=related violence is part of an emerging research
agenda on human security. In order to make progress, this research agenda should
rely on sound theories and rigorous empirical analysis to assess the effects of orga-
nized criminal violence on the prospects of economic development, the protection of
human rights, and the expectations of democratic governance. Analyzing the effect of
organized criminal violence on each of these sectors constitutes a research agenda in
its own right. Addressing these topics is of paramount importance for understanding
and meeting urgent global challenges.

419
"11&/%*9
"

420
A.1
Infolatina
Query

("crimen organizado" OR "delincuencia organizada" OR AFI OR amapola

OR arma* OR armas OR asesin* OR atac* OR ataque OR capo OR cartel
OR cocaina OR cuerp* OR crim* OR delincuen* OR decapit* OR dispar*
OR drog* OR ejecucion* OR ejecuta* OR ejercito OR enfrenta* OR
herid* OR mariguana OR marina OR mata OR mata* OR mato OR militar*
OR muert* OR narco* OR PFP OR pistol* OR polici* OR restos OR rifl*
OR secuestr* OR sicari* OR SSP OR tortur* OR zeta*) NOT (Qaeda OR
Oppenheimer OR "apoyo al campo" OR "Banco de Mexico" OR Laden OR
"bolsa Mexicana" OR "canasta basica" OR "Carlos Marin" OR Aristegui
OR "Gomez Leyva" OR invitado OR Dresser OR "desastre natural" OR
"desastres naturales" OR Bartolome OR Fondevilla OR forestal OR
"JAQUE MATE" OR Chabat OR "Fuentes Aguirre" OR Zuckermann OR Curzio
OR "liga mexicana" OR lector OR Meyer OR Loret OR "Luis Rubio" OR
Merino OR "medio ambiente" OR "Obra publica" OR "Obras publicas"
OR "Plaza Publica" OR "regreso a clases" OR Sarmiento OR "sintesis
de medios" OR "sintesis de prensa" OR "sintesis informativa" OR
"tala clandestina" OR taxi OR "templo mayor" OR tolvanera OR "toma
clandestina" OR accidente OR actor OR actriz OR acuerdo OR adelata
OR Afganistan OR aguacero OR AH1N1 OR alimento OR AMLO OR anuncia
OR aprueba OR atropella OR bacteria OR ballena OR beatificacion
OR belleza OR bono OR bronquitis OR caiman OR calor OR cancer OR
campeonato OR cantante OR carnaval OR ciclista OR cine OR clima OR
colera OR comentario OR comisionado OR cortocircuito OR cultura
OR debate OR delfin OR deporte OR desbordamiento OR dialogo OR
ebrio OR ecologia OR educacion OR Egipto OR empresa OR escritor
OR espectaculos OR estrellas OR Europa OR eutanasia OR evenenado
OR felicita OR festejo OR fiscal OR frio OR futbol OR Gaddafi OR
ganancia OR granizo OR gripa OR gripe OR inaugura OR indigna OR
infarto OR influenza OR infraccion OR infraccion OR Ingalterra
OR intoxicado OR inundacion OR invito OR Irak OR islam OR juego
OR jugar OR justific* OR laboral OR Latinoamerica OR Libia OR
llam* OR afirm* OR nieg* OR exig* OR reclam* OR critic* OR promet*
OR llueven OR lluvia OR Marruecos OR medicament* OR medicin*
OR metrobus OR Mitofsky OR modernizacion OR multa OR multa* OR
mundial OR msica OR nepotismo OR nuclear OR Obama OR olimpiada OR
opinion OR Osama OR Paulette OR pelicula OR peticion OR pirata OR
pirotecnicos OR PROFEPA OR promo* OR propon* OR rabia OR reclam*
OR reconoc* OR religi* OR reproch* OR ring OR rugido OR sarampion
OR sindica* OR sismo OR supervisa OR talamontes OR Tabaco OR teatro
OR temperatura OR terremoto OR tormenta OR torneo OR trascendio OR
tuberculosis OR urge OR urgen OR utilidad* OR vacacio* OR varicela
OR veneno OR videojuego OR viru* OR EZLN OR EPR OR Zapatist* OR
nota diplomatica OR fraude OR subcomandante marcos OR ambulantes OR
pescado OR tortuga OR IFE OR dia del trabajo OR desvio de recursos
OR campeonato OR dictador OR suicid* OR casas de bolsa OR Iraq
OR hipotec* OR Appo OR Appistas OR Boletin de prensa OR misa OR
(reforma & (energetica OR laboral OR politica OR tributaria)))

421
A.2
Number
of
Information
Sources
by
Year

Federal government 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Army
Federal Police
Navy
Attorney General
Local government
State Attorney Generals
National newspapers
Servicio Universal de Noticias
El Economista
El Financiero
Excélsior
Notimex - Nacional
Reforma
La Jornada
El Sol de México
Milenio Diario
Revista Proceso
La Crónica de Hoy
Local newspapers
Aguascalientes
Baja California
Baja California Sur
Campeche
Chiapas
Chihuahua
Coahuila
Colima
Distrito Federal
Durango
Estado de Mexico
Guanajuato
Guerrero
Hidalgo
Jalisco
Michoacán
Morelos
Nayarit
Nuevo León
Oaxaca
Puebla
Querétaro
Quintana Roo
San Luis Potosí
Sinaloa
Sonora
Tabasco
Tamaulipas
Tlaxcala
Veracruz
Yucatán
Zacatecas

Color code: number of information sources per year

0 1 2 3 4 5+

Figure A.1. Number of information sources by year

422
A.3
List
of
Information
Sources

This is the complete list of information sources used in this research. It contains
a total of 105 sources including:

• 4 from federal government agencies

• 32 from local government agencies

• 11 national newspapers

• 58 local newspapers

TABLE
A.1:

LIST
OF
INFORMATION
SOURCES

Source type Name of information source

Federal government agencies: 4

Army: Secretarı́a de la Defensa Nacional (SEDENA)

Federal Police: Secretarı́a de Seguridad Pública (SSP)

Navy: Secretarı́a de Marina Armada de México (SEMAR)

Attorney General: Procuradurı́a General de la República

(PGR)

Local government agencies: 32

State Attorney Generals for all states: Procuradurı́as Es-

tatales

National level newspapers and magazines: 11

Servicio Universal de Noticias

El Economista

El Financiero

Excélsior

Continued on next page

423
 TABLE A.1 – Continued from previous page

Source type Name of information source

Notimex - Nacional

Reforma

La Jornada

El Sol de México

Milenio Diario

Revista Proceso

La Crónica de Hoy

Local level newspapers: 58

Aguascalientes El Sol - Regional Newspapers

Baja California El Mexicano

El Sol - Regional Newspapers

La Voz de la Frontera

Baja California Sur El Sudcaliforniano

Campeche Diario de Yucatán - Campeche

Chiapas Notimex - Estados

Chihuahua Diario de Juárez

El Diario de Chihuahua

El Diario de Delicias

El Diario de Nuevo Casas Grandes

El Diario de Parral

El Sol - Regional Newspapers

Coahuila La Opinión

Milenio de Torreón

El Sol - Regional Newspapers

Continued on next page

424
 TABLE A.1 – Continued from previous page

Source type Name of information source

Colima Notimex - Estados

Distrito Federal El Sol - Regional Newspapers

Durango El Sol - Regional Newspapers

Estado de México Milenio Estado de México

Guanajuato Periódico A.M. Celaya

Periódico A.M. Guanajuato

Periódico A.M. Irapuato

Periódico A.M. La Piedad

Periódico A.M. León

Periódico A.M. San Francisco del Rincón

Milenio - León

El Sol - Regional Newspapers

Guerrero El Sol - Regional Newspapers

Hidalgo Milenio Pachuca

El Sol - Regional Newspapers

Jalisco Mural - Newspaper

Milenio Guadalajara

Michoacán El Sol - Regional Newspapers

Morelos Ecos de Morelos - La Unión de Morelos

El Sol - Regional Newspapers

Nayarit Notimex - Estados

Nuevo León Milenio Diario de Monterrey

El Norte - Newspaper

Oaxaca Notimex - Estados

Continued on next page

425
 TABLE A.1 – Continued from previous page

Source type Name of information source

Puebla Milenio Puebla

El Sol - Regional Newspapers

Querétaro Diario de Querétaro

El Occidental

El Sol - Regional Newspapers

Quintana Roo Notimex - Estados

San Luis Potosı́ El Sol de San Luis

El Sol - Regional Newspapers

Sinaloa El Sol - Regional Newspapers

Sonora Notimex - Estados

Tabasco Milenio Villahermosa

Tamaulipas Milenio Diario de Tampico

El Sol - Regional Newspapers

Tlaxcala El Sol - Regional Newspapers

Veracruz Milenio Xalapa

El Sol - Regional Newspapers

Yucatán Diario de Yucatán

Zacatecas El Sol - Regional Newspapers

426
 A.4
List
of
Mexican
States

TABLE A.2

LIST OF MEXICAN STATES

Code Name Short name Abbreviation Code Name Short name Abbreviation
1 Aguascalientes AGS Ags. 17 Morelos MOR Mor.
2 Baja California BC B.C. 18 Nayarit NAY Nay.
3 Baja California Sur BCS B.C.S 19 Nuevo León NL N.L.
4 Campeche CAMP Camp. 20 Oaxaca OAX Oax.
5 Coahuila COAH Coah. 21 Puebla PUE Pue.
6 Colima COL Col. 22 Querétaro QRO Qro.

427
7 Chiapas CHIS Chis. 23 Quintana Roo QROO Qroo.
8 Chihuahua CHIH Chih. 24 San Luis Potosı́ SLP S.L.P.
9 Distrito Federal (Mexico City) DF D.F. 25 Sinaloa SIN Sin.
10 Durango DUR Dur. 26 Sonora SON Son.
11 Guanajuato GTO Gto. 27 Tabasco TAB Tab.
12 Guerrero GRO Gro. 28 Tamaulipas TAM Tam.
13 Hidalgo HGO Hgo. 29 Tlaxcala TLAX Tlax.
14 Jalisco JAL Jal. 30 Veracruz VER Ver.
15 Estado de México MEX Edomex. 31 Yucatán YUC Yuc.
16 Michoacán MICH Mich. 32 Zacatecas ZAC Zac.
 A.5
Map
of
Mexican
States

_
1 Aguascalientes 17 Morelos
2 Baja California 18 Nayarit
3 Baja California Sur 19 Nuevo León
2 4 Campeche 20 Oaxaca
5 Coahuila 21 Puebla
26
6 Colima 22 Querétaro
_
8 7 Chiapas 23 Quintana Roo
_
8 Chihuahua 24 San Luis Potosí
9 Distrito Federal 25 Sinaloa
5
10 Durango 26 Sonora
11 Guanajuato 27 Tabasco
3
_ 19 12 Guerrero 28 Tamaulipas
_
25 13 Hidalgo 29 Tlaxcala
_ 10 14 Jalisco 30 Veracruz
_
_ _ 28 15 Estado de México 31 Yucatán
32 16 Michoacán 32 Zacatecas

428
_
24
1 _
18 _
_ _ 31
_
11
_ _22 13
14 _
_ 23
_ 29 _
_ 16 _ _
15 9 _ 30 4 _
6 _ _
17 21
_27
12 _
_ 20 _
7

_ State Capital
States 0 212.5 425 850 Kilometers Ü
Figure A.2. Map of Mexican states
A.6
Descriptive
Statistics
of
Data
at
Municipal Level
on
a
Daily
Basis

TABLE
A.3:

DESCRIPTIVE
STATISTICS
OF
DATA
AT
MUNICIPAL
LEVEL

ON
A
DAILY
BASIS

Variable Obs. Mean Std. Dev. Min. Max.

Arrests 9,868,208 0.006 0.121 0 60

Arrests (predicted) 9,868,208 0.092 0.059 0.01 0.40

Cocaine Price 9,868,208 163.903 29.141 128.32 231.97

Competition 9,868,208 0.005 0.114 0 36

Competition (predicted) struc. 9,833,864 0.033 0.057 0.0003 1.49

Corruption 9,868,208 22.677 12.321 0 56.90

Divided government 9,868,208 1.181 0.589 0 2

Divorce 9,868,208 25.437 174.684 0 7054

Drug markets 9,868,208 1.241 16.610 0 740

Drug Production 9,868,208 0.417 0.766 0 3

Effective number of Parties 9,868,208 3.080 0.198 2.91 3.32

GDP State log 9,868,208 19.101 0.757 17.09 21.49

Gulf 9,868,208 0.114 0.317 0 1

Margin of victory 9,868,208 0.053 0.048 0.01 0.23

North 9,868,208 0.055 0.227 0 1

Pacific 9,868,208 0.145 0.352 0 1

Population log 9,868,208 9.364 1.527 4.53 14.41

Poverty 9,868,208 0.000 0.996 -2.38 4.50

Retaliation 9,868,208 0.002 0.052 0 11

Rifles 9,868,208 15.911 2.602 12.85 22.49

429
 TABLE A.3: Continued

Variable Obs. Mean Std. Dev. Min. Max.

Seizure of assets 9,868,208 0.002 0.056 0 16

Seizure of assets (predicted) 9,868,208 0.040 0.031 0.00 0.25

Seizure of drugs 9,868,208 0.008 0.144 0 26

Seizure of drugs (predicted) 9,868,208 0.039 0.024 0.00 0.20

Seizure of guns 9,868,208 0.002 0.058 0 20

Seizure of guns (predicted) 9,868,208 0.017 0.022 0.00 0.27

Teenage mothers 9,833,864 0.141 0.068 0 1

Unemployment 9,868,208 2.911 1.560 0.2 9.7

Violent enforcement 9,868,208 0.001 0.033 0 9

Violent enforcement (predicted) 9,868,208 0.009 0.015 0.0 0.23

9/11 9,868,208 0.846 0.361 0 1

Notes:

“predicted” refers to eXβ where (Xβ) is the predicted log of expected counts

“struc.” refers to the predicted outcome of the structural Model 5 in Table 6.1

430
A.7
Descriptive
Statistics
of
Data
at
the
National Level
on
a
Daily
Basis

TABLE
A.4:

DESCRIPTIVE
STATISTICS
OF
DATA
AT
NATIONAL
LEVEL

ON
A
DAILY
BASIS

Variable Obs. Mean Std. Dev. Min. Max.

Violent enforcement 4018 1.52887 3.673342 0 58

Competition 4018 11.82106 18.59076 0 164

Retaliation 4018 3.74216 6.65784 0 63

Arrests 4018 15.73868 15.0746 0 127

Seizure of assets 4018 5.054007 6.491895 0 110

Seizure of drugs 4018 20.37382 19.60744 0 176

Seizure of guns 4018 4.251618 6.68392 0 63

Drug production 4018 0.4172612 0.0650845 0.3147394 0.52443

State GDP log 4018 19.1005 0.3274759 18.5609 19.51432

Poverty 4018 0.0001695 0.0011577 -0.0014269 0.0029951

Population log 4018 9.364474 0.0255989 9.336298 9.41745

Rifles 4018 1591124 260266.2 1284554 2248851

431
A.8
Hospitalizations
by
Drug
Intoxication

Total number of hospital discharge records where the diagnostic indicates intox-
ication of cannabis, cocaine, opium, hallucinogens, solvents and multiple or other
drugs (Secretarı́a de Salud, 2012a).

AGS BC BCS CAMP

500 1000 1500

500 1000 1500

500 1000 1500

500 1000 1500

0

0
COAH COL CHIS CHIH
500 1000 1500

500 1000 1500

500 1000 1500

500 1000 1500

0

0
DF DUR GTO GRO
500 1000 1500

500 1000 1500

500 1000 1500

500 1000 1500

0

0
HGO JAL MEX MICH
500 1000 1500

500 1000 1500

500 1000 1500

500 1000 1500

0

0
MOR NAY NL OAX
500 1000 1500

500 1000 1500

500 1000 1500

500 1000 1500

0

PUE QRO QROO SLP

500 1000 1500

500 1000 1500

500 1000 1500

500 1000 1500

0

SIN SON TAB TAM

500 1000 1500

500 1000 1500

500 1000 1500

500 1000 1500

0

TLAX VER YUC ZAC

500 1000 1500

500 1000 1500

500 1000 1500

500 1000 1500

0

2000 2002 2004 2006 2008 2010 2000 2002 2004 2006 2008 2010 2000 2002 2004 2006 2008 2010 2000 2002 2004 2006 2008 2010

Figure A.3. Hospitalizations by drug intoxication at state level

432
 BIBLIOGRAPHY

Acemoglu, A., S. Johnson and A. Robinson. 2001. “The Colonial Origins of Compar-
ative Developmnet: An Empirical Investigation.” The American Review of Eco-
nomics 91(5):1369–1401.

Adcock, Robert and David Collier. 2001. “Measurement Validity: A Shared Standard
for Qualitative and Quantitative Research.” The American Political Science Review
95(3):529–546.

Aguilar Camı́n, Héctor. 2004. “Una novela sin ficción.” Nexos . Last accessed on
26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=2102091

Aguilar Camı́n, Héctor. 2006. La tragedia de Colosio. Mexico: Alfaguara.

Aguilar Camı́n, Héctor. 2009. “Narco Historias Extraordinarias.” Nexos . Last ac-
cessed on 26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=73

Akçomak, I. Semih and Bas ter Weel. 2008. “The Impact of Social Capital on Crime:
Evidence from the Netherlands.”.

Aldrich, John H., James Alt and Arthur Lupia. 2007. The EITM Approach: Origins
and Interpretations. In The Oxford Handbook of Political Methodology, ed. Janet M.
Box-Steffensmeier, Henry E. Brady and David Collier. New York: Oxford Univer-
sity Press pp. 828–843.

Almeida, Paul. 2008. Waves of Protest: Social Struggle in El Salvador, 1925-2005.

Minneapolis: University of Minessota Press.

Álvarez Béjar, Alejandro. 2007. “Mexico after the Elections: The Crisis of Legiti-
macy & the Exhaustion of Predatory Neoliberalism.” Monthly Review 59(3). Last
accessed on 26/6/2013.
URL: http://monthlyreview.org/2007/07/01/mexico-after-the-elections-the-crisis-
of-legitimacy-the-exhaustion-of-predatory-neoliberalism

American Chamber. 2013. “Sondeo de Seguridad Empresarial en México.”.

433
Anderson, Annelise. 2005. Organized crime, mafia and governments. In The eco-
nomics of organized crime, ed. Gianluca Fiorentini and Sam Peltzman. Cambridge,
MA: Cambridge University Press pp. 33–60.
Andreas, Peter. 1998. “The Political Economy of Narco-Corruption in Mexico.”
Current History (April):160–165.
Andreas, Peter. 2009. Border Games. 2 ed. Ithaca: Cornell University Press.
Angrist, Joshua and Jorn-Steffen Pischke. 2009. Mostly Harmless Econometrics.
Princeton, New Jersey: Princeton University Press.
Article 19. 2013. Mexico: Serious crimes against free expression in 2012. Technical
report Article 19 Mexico: . Last accessed on 26/6/2013.
URL: http://www.article19.org/resources.php/resource/3655/en/mexico:-serious-
crimes-against-free-expression-in-2012
Astorga, Luis. 1995. Mitologı́a del narcotraficante en México. Mexico: Plaza y Janés.
Astorga, Luis. 1999. “Drug Trafficking in Mexico: A First General Assessment.” Last
accessed on 26/6/2013.
URL: http://www.unesco.org/most/astorga.htm
Astorga, Luis. 2003. Drogas Sin Frontera. Mexico City: Grijalbo.
Astorga, Luis. 2005. El siglo de las drogas: el narcotráfico, del porfiriato al nuevo
milenio. Mexico City: Plaza y Janés.
Astorga, Luis. 2010. “A Pax Mafiosa in Mexico?” Project Syndicate . Last accessed
on 26/6/2013.
URL: http://www.project-syndicate.org/commentary/a-pax-mafiosa-in-mexico
Azar, Edward E. 1970. “Analysis of international events.” Peace Research Reviews
4(November):1–106.
Azar, Edward E. 1975. Behavioral forecasts and policy making: An events data ap-
proach. In nternational Events and the Comparative Analysis of Foreign Policy, ed.
Jr. Charles Kegley, Gregary A. Raymond, Robert M. Rood and Richard Skinner.
University of South Carolina Press pp. 215–236.
Azar, Edward E. 1980. “The Conflict and Peace Data Bank (COPDAB).” Journal
of Conflict Resolution 24:143–152.
Azar, Edward E. and Joseph D. Ben-Dak. 1975. Theory and Practice of Events
Research: Studies in Inter-Nation Actions and Interactions. New York: Gordon
and Breach Science Publishers.
Bailey, John and Matthew M Taylor. 2009. “Evade, Corrupt, or Confront? Organized
Crime and the State in Brazil and Mexico.” Journal of Politics in Latin America
pp. 3–29.

434
Bailey, John and Roy Godson, eds. 2000. Organized crime and democratic governabil-
ity: Mexico and the US-Mexican borderlands. Pittsburgh: University of Pittsburgh
Press.

Banks, Arthur. 1971. Cross-Polity Time-Series Data. Cambridge, MA: MIT Press.

Bartra, Roger. 2013. The Mexican Transition: Politics, Culture, and Democracy in
the Twenty-First Century. Cardiff: University of Wales Press.

Bates, R., A. Greif and S. Singh. 2002. “Organizing Violence.” Journal of Conflict
Resolution 46(5):599–628.

Bates, Robert H. 2008. When Things Fell Apart: State Failure in Late Century
Africa. New-York: Cambridge University Press.

Baumgartner, Frank, Bryan Jones and Michael MacLeod. 1998. “Lessons from the
Trenches: Ensuring Quality, Reliability, and Usability in the Creation of a New
Data Source.” The Political Methodologist 8(2):1–10.

Becerra, Ricardo, Pedro Salazar and José Woldenberg. 1997. La reforma electoral de
1996. Una descripción general. Mexico: Fondo de Cultura Económica.

Becerra, Ricardo, Pedro Salazar and José Woldenberg. 2011. La mecánica del cambio
polı́tico en México. elecciones, partidos y reformas. Mexico: Cal y Arena.

Becker, Gary S. 1968. “Crime and Punishment. An Economic Approach.” Journal

of Political Economy 76(2):169–217.

Berman, Eli, Jacob N Shapiro and Joseph H Felter. 2011. “Can Hearts and Minds
Be Bought? The Economics of Counterinsurgency in Iraq.” National Bureau of
Economic Research 119(4):766–819.

Blancornelas, Jesús. 2002. El Cártel. Mexico: Plaza y Janés.

Blattman, Christopher and Edward Miguel. 2010. “Civil War.” Journal of Economic
Literature 48(1):3–57. Last accessed on 26/6/2013.
URL: http://pubs.aeaweb.org/doi/abs/10.1257/jel.48.1.3

Bogan, Jesse. 2009. “Billionaires List. Cocaine King.”. Last accessed on 26/6/2013.
URL: http://www.forbes.com/forbes/2009/0330/102-cocaine-king.html

Bollen, Kenneth A. 1989. Structural Equations with Latent Variables. New York:
Wiley.

Bond, Doug, Craig J. Jenkins, Charles L. Taylor and Kurt Schock. 1997. “Mapping
Mass Political Conflict and Civil Society: Issues and Prospects for the Automated
Development of Event Data.” Journal of Conflict Resolution 41(4):Mapping Mass
Political Conflict and Civil Society:.

435
Bond, Doug, Joe Bond, Churl Oh, Craig J. Jenkins and Charles L. Taylor. 2003.
“Integrated Data for Events Analysis (IDEA): An Event Typology for Automated
Events Data Development.” Journal of Peace Research 40(6):733–745.

Borjón Nieto, José J. 2008. La lucha contra el crimen organizado y las sobmras de la
militarización. In Primer Congreso Nacional y Segundo Internacional de Derecho
Administrativo. Boca del Rı́o, Veracruz: . Last accessed on 26/6/2013.
URL: http://biblio.juridicas.unam.mx/libros/7/3005/4.pdf

Box, George E.P. and Gwilym M. Jenkins. 1976. Time Series Analysis: Forecasting
and Control. San Francisco,: Holden-Day.

Brachet-Marquez, Viviane. 1992. “Explaining sociopolitical change in Latin America:

The case of Mexico.” Latin American Research Review 27(3):91–122.

British Broadcasting Corporation. 2006. “More than two months after the elections
in Mexico, the conservative Felipe Calderon has been declared president-elect. But
left-wing candidate Andres Manuel Lopez Obrador insists the election was rigged
and does not accept the official result.”. Last accessed on 26/6/2013.
URL: http://news.bbc.co.uk/2/hi/americas/5326452.stm

Brito, Dagobert L. and Michael D. Intriligator. 1992. “Narco-traffic and guerrilla

warfare: A new symbiosis.” Defence Economics 3:263–274.

Bronars, Stephen and Jeff Grogger. 1994. “The Economic Consequences of Unwanted
Motherhood: Using Twin Births as Natural Experiment.” he American Economic
Review 84(5):1141–1156.

Bruce, Victoria and Karin Hayes. 2010. Hostage nation. Colombia’s guerrilla army
and the failed war on drugs. New York: Knopf.

Bruhn, Kathleen. 1996. Taking on Goliath: The Emergence of a New Left Party
and the Struggle for Democracy in Mexico. Pennsylvania: Pennsylvania State
University Press.

Buhaug, Halvard and Jan Ketil Rod. 2006. “Local determinants of African civil wars,
19702001.” Political Geography 25(3):315–335.

Buhaug, Halvard, Scott Gates and Paivi Lujala. 2009. “Geography, Rebel Capability,
and the Duration of Civil Conflict.” Journal of Conflict Resolution 53(4):544–569.

Buonanno, Paolo, Daniel Montolio and Paolo Vanin. 2009. “Does Social Capital
Reduce Crime?” Journal of Law and Economics 52(February):145–170.

Bureau of Alchohol, Tobacco, Firearms and Explosives. 2012. “Annual Firearms

Manufacturers And Export Report.”. Last accessed on 26/6/2013.
URL: http://www.atf.gov/statistics/

436
Cáceres-Delpiano, Julio and Eugenio Giolito. 2008. “The Impact of Unilateral Divorce
on Crime.”.
Calderón, Gabriela, Alberto Dı́az-Cayeros, Beatriz Magaloni, Gustavo Robles and
Jorge Olarte. 2012. “The Temporal and Spatial Dynamics of Violence in Mexico.”
Last accessed on 26/6/2013.
URL: https://www.dropbox.com/s/ce0lhfs9unsuqa3/paper policies violence Aug30.pdf
Camarena, Salvador. 2010. “El asesinato de un candidato a gobernador no detiene
las elecciones estatales en México.”. Last accessed on 26/6/2013.
URL: http://goo.gl/ztIgY
Campbell, Howard. 2009. Drug War Zone. Frotnline Dispatches From The Streets of
El Paso and Juarez. Austin: University of Texas Press.
Cardie, Claire and John Wilkerson. 2008. “Guest Editor’s Introduction: Text An-
notation for Political Science Research.” Journal of Information Technology and
Politics 5(1):1–6.
Castañeda, Jorge. 2012. “Time for an Alternative to Mexico’s Drug War.” Cato
Institute. Economic Development Bulletin .
Castañeda, Jorge and Rubén Aguilar. 2010. El narco: La guerra fallida. Mexico City:
Punto de Lectura.
Castellanos, Laura. 2007. México Armado. 1943-1981. Mexico City: Era.
Cederman, Lars E., Simon Hug and Lutz F. Krebs. 2010. “Democratization and Civil
War: Empirical evidence.” Journal of Peace Research 47(4):377–394.
Cedillo, Adela. 2009. “Cronologı́a de la represión en México (2 mitad del siglo XX).”.
Last accessed on 03/17/2013.
URL: http://guerrasuciamexicana.blogspot.com/2009 01 01 archive.html
Cedillo, Juan Alberto. 2011. La Cosa Nostra en México (1938-1950). Mexico City:
Grijalbo.
Chao Romero, Roberto. 2010. The Chinese in Mexico, 1882-1940. Tucson: University
of Arizona Press.
CNN-Editor. 2012. “La violencia desplazó a más de 26,000 personas en México en
2011: estudio.”. Last accessed on 26/6/2013.
URL: http://mexico.cnn.com/nacional/2012/04/19/la-violencia-desplazo-a-mas-
de-26000-personas-en-mexico-en-2011
Collier, David and Henry E Brady. 2004. Rethinking Social Inquiry: Diverse Tools,
Shared Standards. Maryland: Rowman & Littlefield Publishers.
Collier, David and Steven Levitsky. 1997. “Democracy with Adjectives: Conceptual
Innovation in Comparative Research.” World Politics 49(3):430–451.

437
Collier, Paul. 2000. “Rebellion as a Quasi-Criminal Activity.” Journal of Conflict
Resolution 44(6):839–853.

Collier, Paul and Anke E. Hoeffler. 1998. “On the economic causes of civil war.”
Oxford Economic Papers 50:563–573.

Collier, Paul and Anke E. Hoeffler. 2004. “Greed and grievance in civil war.” Oxford
Economic Papers 56(4):563–595.

Comanor, Williams and Llad Phillips. 2002. “The Impact of Income and Family
Structure on Delincuency.” Journal of Applied Economics 5(2):209–232.

Congreso de la Unión. 1978. “Ley Federal de Amnistı́a.”. Last accessed on 26/6/2013.

URL: http://info4.juridicas.unam.mx/ijure/tcfed/17.htm

Consejo Nacional de Evaluación de la Polı́tica de Desarrollo Social. 2010.

“Metodologı́a para la medición multidimensional de la pobreza en México.”. Last
accessed on 2013/04/13.
URL: http://goo.gl/ySUy0

Consejo Nacional de Evaluación de la Polı́tica de Desarrollo Social. 2012. “Medición

de la Pobreza por Municipio.”. Last accessed on 9/4/2011.
URL: http://web.coneval.gob.mx

Coppedge, Michael. 2012. Democratization and Research Methods: Strategies for

Social Inquiry. Cambridge, MA: Cambridge University Press.

Coppedge, Michael and H. Wolfgang Reinicke. 1990. “Measuring Polyarchy.” Studies

in Comparative International Development 25(1):51–72.

Covarrubias, Adriana. 2010. “Matan a dos policı́as municipales en Acapulco.”. Last

accessed on 26/6/2013.
URL: http://www.eluniversal.com.mx/notas/722177.html

Craig, Richard. 1980. “Operation Condor: Mexico’s Antidrug Campaign Enters a

New Era.” Journal of Interamerican Studies and World Affairs 22(3):345–363.

Crook, Clive. 2009. “A criminally stupid war on drugs in the US.”.

Cruz, Francisco. 2008. El Cártel de Juárez. Mexico City: Planeta.

Cunningham, David. 2003. “The Patterning of Repression: FBI Counterintelligence

and the New Left.” Social Forces 82(1):209 – 240.
URL: http://www.jstor.org/stable/3598144

Dallin, Alexander and Gerorge W. Breslauer. 1970. Political Terror in Communist

Systems. Stanford, CA: Stanford University Press.

Daly, By Catherine, Kimberly Heinle and David A Shirk. 2012. Armed with Impunity.
Technical report Trans-Border Institute San Diego, CA.: .

438
Davenport, Christian. 1995. “Multi-dimensional Threat Perception and State Repres-
sion: An Inquiry into Why States Apply Negative Sanctions.” American Journal
of Political Science 39(3):683–713.

Davenport, Christian. 2004. “The Promise of Democratic Pacification : An Empirical

Assessment.” International Studies Quarterly 48(3):539–560.

Davenport, Christian. 2005. “Understanding covert repressive action: the case of

the U.S. government against the Republic of New Africa.” Journal of Conflict
Resolution 49(1):120–140.

Davenport, Christian. 2007. “State Repression and Political Order.” Annual Review
of Political Science 10(June):1–23.

Davenport, Christian. 2009a. Media Bias, Perspective, and State Repression: The
Black Panther Party. New York: Cambridge University Press.

Davenport, Christian. 2009b. State repression and the domestic democratic peace.
Cambridge University Press.

Davenport, Christian, Hank Johnston and Carol McClurg Mueller. 2005. Repression
and mobilization. Minesota University Press.

Davenport, Christian and Patrick Ball. 2002. “Views to a kill: exploring the im-
plications of source selection in the case of Guatemalan State Terror, 1977-1995.”
Journal of Conflict Resolution 46(3):427–450.

Davies, John L. 1998. “The Global Event-Data System (GEDS).” Last accessed on
26/6/2013.
URL: http://www-rohan.sdsu.edu/ alexseev/RussiaInAsia/GEDSCodebook800 3.pdf

Davis, David and Will H. Moore. 1995. “International Political Interactions (IPI).”.
Last accessed on 26/6/2013.
URL: http://goo.gl/kHt1z

Davis, Diane E. 2009. “Non-State Armed Actors, New Imagined Communities, and
Shifting Patterns of Sovereignty and Insecurity in the Modern World.” Contempo-
rary Security Policy 30(2):221–245.

Dell, Melissa. 2011. “Trafficking Networks and the Mexican Drug War.”.

Department of Defense. 1968. “DIA Intelligence Information Report, Status of Brig.

Gen. Jose Hernandez Toledo.”. Last accessed on 26/6/2013.
URL: http://www.gwu.edu/ nsarchiv/NSAEBB/NSAEBB10/docs/doc24.pdf

Diario Oficial de la Federación. 1998. “Acuerdo por el que se establecen los Lin-
eamientos a que se sujetará la guarda, custodia y plazo de conservación del Archivo
Contable Gubernamental.”. Last accessed on 26/6/2013.
URL: http://info4.juridicas.unam.mx/ijure/nrm/1/102/1.htm?s=is

439
Dixon, William J. 1981. “Measuring Interstate Affect.” American Journal of Political
Science 27(4):828–851.
Dominguez, Jorge and Chappell Lawson, eds. 2003. Mexico’s Pivotal Democratic
Election: Candidates, Voters, and the Presidential Campaign of 2000. Stanford:
Stanford University Press.
Domı́nguez Ruvalcaba, Héctor. 2010. “Ciudad Juárez: La vida breve.”. Last accessed
on 9/4/2011.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=73223
Donohue, John J and Steven D Levitt. 2001. “The Impact of Legalized Abortion on
Crime.” Quarterly Journal of Economics 116(2):379–420.
Doyle, Kate. 2003. “Operation Intercept. The perils of unilateralism.” The National
Security Archive Electronic Briefing . Last accessed on 26/6/2013.
URL: http://www.gwu.edu/ nsarchiv/NSAEBB/NSAEBB86/
Doyle, Kate. 2006a. “Human Rights and the Dirty War in Mexico.” The National
Security Archive Electronic Briefing . Last accessed on 26/6/2013.
URL: http://www.gwu.edu/ nsarchiv/NSAEBB/NSAEBB89/
Doyle, Kate. 2006b. “Official Report Released on Mexico’s ”Dirty War”. Govern-
ment Acknowledges Responsibility for Massacres, Torture, Disappearances and
Genocide.” The National Security Archive Electronic Briefing . Last accessed on
26/6/2013.
URL: http://www.gwu.edu/ nsarchiv/NSAEBB/NSAEBB209/index.htm
Drug Enforcement Administration. 1995. “DEA Congressional Testimony. Thomas
A. Constantine, Administrator Drug Enforcement Administration United States
Department of Justice.”. Last accessed on 26/6/2013.
URL: http://www.justice.gov/dea old/pubs/cngrtest/ct950808.htm
Drug Enforcement Administration. 2012. “Stats & Facts.”. Last accessed on
26/6/2013.
URL: http://www.justice.gov/dea old/statistics.html
Drug Enforcement Administration. 2013a. DEA History in Depth. 1985-1990. Tech-
nical report Washington, D.C.: . Last accessed on 26/6/2013.
URL: http://www.justice.gov/dea/about/history/1985-1990.pdf
Drug Enforcement Administration. 2013b. “Wall of Honor. Enrique S. Camarena.”.
Last accessed on 02/06/2013.
URL: http://www.justice.gov/dea/about/wall-honor/wall-of-honor bios.shtml
Duarte, Enrique. 2008. “Secuestro de Martı́ conmociona a México.”. Last accessed
on 26/6/2013.
URL: http://www.cnnexpansion.com/actualidad/2008/08/04/secuestro-de-marti-
conmociona-a-mexico

440
Dube, Arindrajit, Oeindrila Dube and Omar Garcia-Ponce. 2013. “Cross-Border
Spillover: U.S. Gun Laws and Violence in Mexico.”.

Dunning, Thad. 2012. Natural Experiments in the Social Sciences: A Design-Based

Approach. New York: Cambridge University Press.

Dunning, Thad and Leslie Wirpsa. 2004. “Oil and the Political Economy of Conflict
in Colombia and Beyond: A Linkages Approach.” Geopolitics 9(1):81–108.

Duran-Martinez, Angelica. 2012. “Criminals, Cops, and Politicians: Dynamics of

Drug Violence in Colombia and Mexico.”.

Duvall, Raymond D. and Michael Stohl. 1988. Governance by terror. In The Politics
of Terrorism, ed. Michael Stohl. New York: Marcel Dekker pp. 231–272.

Eisenstadt, Todd A. 2003. Courting Democracy in Mexico: Party Strategies and

Electoral Institutions. Cambridge: Cambridge University Press.

Ellingsen, Tanja and Petter Gleditsch. 1997. Democracy and armed conflict in the
third world. In Causes of conflict in third world countries, ed. Ketil Volden and
Dan Smith. Oslo: Nort-South Coalition and International Peace Research Institute
pp. 69–81.

Enders, W. and T. Sandler. 2006. Political Economy of Terrorism. Cambridge:

Cambridge University Press.

Enders, Walter. 2009. Applied Econometric Times Series. New York: Wiley.

Enders, Walter and Todd Sandler. 1993. “The Effectiveness of Antiterrorism Policies:
A Vector-Autoregression- Intervention Analysis.” The American Political Science
Review 87(4):829 – 844.

Escalante Gonzalbo, Fernando. 2009. “Homicidios 1990-2007.”. Last accessed on

26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=776

Escalante Gonzalbo, Fernando. 2011. “Homicidios 2008-2009 La muerte tiene per-

miso.”. Last accessed on 26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=1943189

Escalante Gonzalbo, Fernando. 2012. El crimen como realidad y representacion:

contribuicion para una historia del presente. Mexico: El Colegio de México.

Escalante Gonzalbo, Fernando, Eduardo Guerrero Gutiérrez, Hope Alejandro, Denise

Maerker, Ana Laura Magaloni, Héctor de Mauleón, Natalia Mendoza Rockwell,
Guillermo Valdés and Joaquı́n Villalobos. 2011. “Nuestra guerra: Una conver-
sación.”. Last accessed on 26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=2102417

441
Espinosa, Verónica. 2011. “Intromisión de Fox y dinero sucio en las campañas, im-
pidieron el triunfo de AMLO en 2006: Cárdenas.” Proceso . Last accessed on
26/6/2013.
URL: http://www.proceso.com.mx/?p=270796

Fajnzylber, Pablo, Daniel Lederman and Norman Loayza. 2000a. “Crime and
Victimization: An Economic Perspective.” Economı́a 1(1):219–278. Last accessed
on 26/6/2013.
URL: http://muse.jhu.edu/content/crossref/journals/economia/v001/1.1fajnzylber.html

Fajnzylber, Pablo, Daniel Lederman and Norman Loayza. 2000b. “Crime and Vic-
timization: An Economic Perspective.” Economı́a 1(1):219–278.

Fajnzylber, Pablo, Daniel Lederman and Norman Loayza. 2002a. “Inequality and
Violent Crime.” Journal of Law and Economics 45(1):1–40.

Fajnzylber, Pablo, Daniel Lederman and Norman Loayza. 2002b. “What causes
violent crime?” European Economic Review 46:1323–1357.

Fearon, James D. 2004. “Why Do Some Civil Wars Last So Much Longer than
Others?” Journal of Peace Research 41(3):275–301.

Fearon, James D. 2005. “Primary Commodity Exports and Civil War.” Journal of
Conflict Resolution 49(4):483–507.

Fearon, James D. and David D. Laitin. 2003. “Ethnicity, Insurgency, and Civil War.”
The American Political Science Review 97(1):75 – 90.

Felbab-Brown, Vanda. 2010. Shooting up. Counterinsurgency and the war on drugs.
Washington, D.C.: Brookings Institution.

Felsen, David. 2009. An Incomplete Transition: Mexican Politics and the Transition
to Democracy. Toronto: University of Toronto Press.

Francisco, Ronald A. 1996. “Coercion and Protest: An Empirical Test in Two Demo-
cratic States.” American Journal of Political Science 40(4):1179 – 1204. Last ac-
cessed on 26/6/2013.
URL: http://www.jstor.org/stable/2111747

Freeman, John R., John T. Williams and Tse-min Lin. 1989. “Vector Autoregression
and the Study of Politics.” American Journal of Political Science 33(4):842–877.

Gambetta, Diego. 1993. The Sicilian Mafia: The Business of Private Protection.
Harvard University Press.

Garay Salamanca, Luis Jorge and Eduardo Salcedo-Albarán. 2012. Narcotrafico,

corrupcion y estados : como las redes ilicitas han reconfigurado las instituciones
en Colombia, Guatemala y Mexico. Mexico City: Debate.

442
Garay Salamanca, Luis Jorge, Eduardo Salcedo-Albarán, Isaac de León Beltrán and
Bernardo Guerrero. 2008. La Captura y Reconfiguración Cooptada del Estado en
Colombia. Bogotá, Colombia: Fundación Método, Fundación Avina y Transparen-
cia por Colombia.

Gartner, Scott Sigmund and Patrick M Regan. 1996. “Threat and Repression : The
Non-Linear Relationship Between Government and Opposition Violence.” Journal
of Peace Research 33(3):273–287.

Geddes, Barbara. 2003. Paradigms and Sand Castles: Theory Building and Research
Design in Comparative Politics. Ann Arbor, MI: University of Michigan Press.

Gerner, Deborah, Philip A. Schrodt, Omur Yilmaz and Rajaa Abu-Jabr. 2002. The
Creation of CAMEO (Conflict and Mediation Event Observations): An Event Data
Framdwork for a Post Cold War World. In Prepared for delivery at the 2002 Annual
Meeting of the Americal Political Science Association. 01/10/2013. Last accessed
on 26/6/2013.
URL: http://web.ku.edu/ keds/papers.dir/Gerner.APSA.02.pdf

Glaeser, Eduard L., Bruce Sacerdote and José Scheinkman. 1996. “Crime and social
interactions.” The Quarterly Journal of Economics 111(2):507–548.

Glaeser, Edward L and Bruce Sacerdote. 1999. “Why Is There More Crime in Cities
?” Journal of Political Economy 107(S6):S225–S258.

Global Commission on Drug Policy. 2012. The War on Drugs and HIV/AIDS. How
the Criminalization of Drug Use Fuels the Global Pandemic. Technical report
Global Commission on Drug Policy Zurich: . Last accessed on 26/6/2013.
URL: http://globalcommissionondrugs.org/wp-content/themes/gcdp v1/pdf/GCDP HIV-
AIDS 2012 REFERENCE.pdf

Global Commission on Drug Policy. 2013. The Negative Impact Of The War On
Drugs On Public Health: The Hidden Hepatitis C Epidemic. Technical report
Global Commission on Drug Policy Zurich: . Last accessed on 26/6/2013.
URL: http://www.globalcommissionondrugs.org/hepatitis/gcdp hepatitis english.pdf

Goertz, Gary. 2005. Social Science Concepts: A User’s Guide. Princeton, New Jersey:
Princeton University Press.

Goldstein, Joshua S. 1992. “A Conflict-Cooperation Scale for WEIS Events Data.”

Journal of Conflict Resolution 36(2):369–385.

Goldstein, Robert Justin. 1978. Political Repression in Modern America: From 1870
to 1976. First illi ed. Chicago: University of Illinois Press.

González Ruiz, Samuel, Ernesto López Portillo and José Arturo Yáñez. 1994. Se-
guridad Pública en México. Mexico: Universidad Nacional Autónoma de México.

443
Gootenberg, Paul. 2011. “Cocaine s Blowback North : A Pre-History of Mexican
Drug Violence.” LASA Forum 42(2):7–10.

Graham, Hugh Davis and Ted Gurr. 1969. The History of Violence in America:
Historical and Comparative Perspectives. New York: Praeger.

Greene, Kenneth F. 2007. Why Dominant Parties Lose: Mexico’s Democratization

in Comparative Perspective. Cambridge University Press.

Greene, Kenneth F. 2010. “Ousting Autocrats: The Political Economy of Competi-

tive Authoritarism.”.

Grimmer, Justin and Brandon M. Stewart. 2013. “Text as Data: The Promise and
Pitfalls of Automatic Content Analysis Methods for Political Texts.” Political
Analysis Published. Last accessed on 26/6/2013.
URL: http://pan.oxfordjournals.org/content/early/2013/01/21/pan.mps028.abstract

Grossman, David. 1995. On Killing. The Psychological Cost of Learning to Kill in

War and Society. New York: Back Bay Books.

Grossman, Herschel I. 1999. “Kleptocracy and revolutions.” Oxford Economic Papers

51(267-283).

Guerrero, Eduardo. 2009a. “Las tres guerras.”. Last accessed on 26/6/2013.

URL: http://www.nexos.com.mx/?P=leerarticulo&Article=777

Guerrero, Eduardo. 2009b. “Narcotráfico, S.A.” Nexos . Last accessed on 26/6/2013.

URL: http://www.nexos.com.mx/?P=leerarticulo&Article=54

Guerrero, Eduardo. 2010a. “Cómo reducir la violencia en México.” Nexos . Last

accessed on 26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=1197808

Guerrero, Eduardo. 2010b. “Los hoyos negros de la estrategia contra el narco.”. Last
accessed on 26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=248547

Guerrero, Eduardo. 2011a. “2011: La dispersión de la violencia.”. Last accessed on

26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=2102543

Guerrero, Eduardo. 2011b. “La Raı́z de la Violencia.” Nexos . Last accessed on

26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=2099328

Guerrero, Eduardo. 2011c. Security, Drugs, and Violence in Mexico: A Survey.

Technical report 7th North American Forum Washington, D.C.: .

444
Guerrero Robledo, José Guadalupe. 2009. “Mis sexenios (13).” El Periódico de Saltillo
. Last accessed on 26/6/2013.
URL: http://www.elperiodicodesaltillo.com/2009 abril/abril09ms.html

Guillén Jiménez, Gonazlo. 2007. Los Confidentes de Pablo Escobar. Bogotá: News
Design.

Gurr, Ted. 1968. “Psychological Factors in Civil Violence.” World Politics 20(2):245–
278. Last accessed on 26/6/2013.
URL: http://www.journals.cambridge.org/abstract S0043887100014908

Gurr, Ted. 1989. Violence in America: The history of crime. Newbury Park: Sage.

Gurr, Ted Robert. 1970. Why Men Rebel. Princeton: Princeton University Press.

Hall, Peter. 2003. Aligning Ontology and Methodology in Comparative Politics. In

Comparative Historical Analysis in the Social Sciences, ed. J. Mahoney and D.
Rueschemeyer. New York: Cambridge University Press pp. 373–404.

Harff, Barbara. 2003. “No Lessons Learned from the Holocaust? Assessing Risks
of Genocide and Political Mass Murder since 1... more.” The American Political
Science Review 97(1):57 – 73. Last accessed on 26/6/2013.
URL: http://www.jstor.org/stable/3118221

Harvey, Anna. 2008. What Makes a Judgment ”Liberal”? Coding Bias in the United
States Supreme Court Judicial Database. In 3rd Annual Conference on Empirical
Legal Studies Papers. Last accessed on 26/6/2013.
URL: http://papers.ssrn.com/sol3/papers.cfm?abstract id=1120970

Hayden, Tom, ed. 2002. The Zapatista Reader. 5th ed. New York: Nation Books.

Hechter, Michael. 2001. Containing Nationalism. Oxford: Oxford University Press.

Hegre, Havard, Tanja Ellingsen, Scott Gates and Nils Petter Gleditsch. 2001. “Toward
a Democratic Civil Peace? Democracy, Political Change, and Civil War, 1816-
1992.” The American Political Science Review 95(1):33–48.

Held, David. 2006. Models of Democracy. 3rd ed. Stanford, CA: Stanford University
Press.

Herbst, Jeffrey. 2004. “African Militaries and Rebellion: The Political Economy of
Threat and Combat Effectiveness.” Journal of Peace Research 41(3):357–369.

Hermann, Charles. 1975. Comparing the Foreign Policy Events of Nations. In Inter-
national Events and the Comparative Analysis of Foreign Policy, ed. Charles Jr.
Kegley, Gregary A. Raymond, Robert M. Rood and Richard Skinner. Columbia,
South Carolina: University of South Carolina Press pp. 145–158.

445
Hermann, Charles, Margaret G. Hermann, Stephen A. Salmore, Maurice A. East,
Barbara G. Salmore and Linda P. Brady. 1977. Comparative Research on the
Events of Nations (CREON) Project: Foreign Policy Events Data Set. Ann Arbor,
MI: Inter-University Consortium for Political and Social Research.

Hermann, Charles, Maurice A. East, Margaret G. Hermann, Barbara G. Salmore and

Stephen A. Salmore. 1973. CREON: A Foreign Event Data Set. Beverly Hills, CA:
Sage.

Hernández, Anabel. 2011. Los Señores del Narco. Mexico: Grijalbo.

Herrera, Raúl. 2013. “Búfalo: 29 años después del gran decomiso de mariguana.”.
Last accessed on 26/6/2013.
URL: http://www.oem.com.mx/elsoldeparral/notas/n2848782.htm

Hibbs, D. A. 1973. Mass Political Violence. New York: John Wiley.

Hidalgo, Daniel F., Suresh Naidu, Simeon Nichter and Neal Richardson. 2010. “Eco-
nomic Determinants of Land Invasion.” The Review of Economics and Statistics
92(3):505–523.

Hill, Jonathan B. 2006. “Strong Orthogonal Decompositions and Non-Linear Im-

pulse Response Functions for Infinite-Variance Processes.” The Canadian Journal
of Statistics / La Revue Canadienne de Statistique 34(3):453–473.

Hirschi, Travis. 1969. Causes of Delinquency. Berkeley: University of California

Press.

Hirshleifer, Jack. 1991. “The paradox of power.” Economics & Politics 3:177–200.

Hobbes, Thomas. 1651. Leviathan. 1994 ed. Indianapolis: Hackett Publishing Co.

Hoover, Dean and David Kowalewski. 1992. “Dynamic Models of Dissent and Re-
pression.” Journal of Conflict Resolution 36(1):150–182.

Howell, Llewellyn D. 1983. “A comparative study of the WEIS and COPDAB data
sets.” International Studies Quarterly 27(2):149–159.

Human Rights Watch. 2011. Neither Rights Nor Security. Technical report Human
Rights Watch Mexico: . Last accessed on 26/6/2013.
URL: http://www.hrw.org/reports/2011/11/09/neither-rights-nor-security-0

Human Rights Watch. 2013. Mexico’s Disappeared. Technical report Human Rights
Watch Mexico: . Last accessed on 26/6/2013.
URL: http://www.hrw.org/reports/2013/02/20/mexicos-disappeared

Humphreys, Macartan. 2005. “Natural Resources, Conflict, and Conflict Resolution:

Uncovering the Mechanisms.” Journal of Conflict Resolution 49(4):508–537.

446
Huntington, Samuel P. 1968. Political order in changing societies. Yale University
Press.
Instituto Federal Electoral. 2012. “Elecciones en México.”. Last accessed on
26/6/2013.
URL: http://www.eleccionesenmexico.org.mx/

Instituto Federal Electoral. 2013. “Estadı́sticas y Resultados Electorales.”. Last

accessed on 26/6/2013.
URL: http://www.ife.org.mx/portal/site/ifev2/Estadisticas y Resultados Electorales/
Instituto Nacional de Estadı́stica y Geografı́a. 2008. “Clasificación del delito en
México.”. Last accessed on 26/6/2013.
URL: http://goo.gl/IbU9B
Instituto Nacional de Estadı́stica y Geografı́a. 2011a. “Banco de Información
Económica.”. Last accessed on 26/6/2013.
URL: http://dgcnesyp.inegi.org.mx/bdiesi/bdie.html

Instituto Nacional de Estadı́stica y Geografı́a. 2011b. “Marco Geoestadı́stico Na-

cional.” Last accessed on 26/6/2013.
URL: http://www.inegi.org.mx/geo/contenidos/geoestadistica/default.aspx
Instituto Nacional de Estadı́stica y Geografı́a. 2011c. “Sistema Estatal y Municipal
de Bases de Datos.”. Last accessed on 9/4/2011.
URL: http://sc.inegi.org.mx/sistemas/cobdem/
Instituto Nacional de Estadı́stica y Geografı́a. 2013a. Anuario estadı́stico de los
Estados Unidos Mexicanos. Technical report Instituto Nacional de Estadı́stica y
Geografı́a Mexico: . Last accessed on 26/6/2013.
URL: http://www3.inegi.org.mx/sistemas/productos/?undefined
Instituto Nacional de Estadı́stica y Geografı́a. 2013b. “Mortalidad. Conjunto de datos:
Defunciones por homicidios.”. Last accessed on 04/28/2013.
URL: http://goo.gl/Z6amr

Jenkins, Craig J. and Charles Perrow. 1977. “Insurgency of the Pwerless: Farm
Worker Movements (1946-1972).” American Sociological Review 42(2):249–268.
Jia, Hao, Stergios Skaperdas and Samarth Vaidya. 2012. “Contest Functions : Theo-
retical Foundations and Issues in Estimation.” International Journal of Industrial
Organization pp. 1–25.
Kalyvas, Stathis N. 2006. The Logic of Violence in Civil War. Cambridge, MA:
Cambridge University Press.

Kalyvas, Stathis N. 2012. “Micro-level studies of violence in Civil War: Redefining

and exending the control-collaboration model.” Terrorism and Political Violence
24(4):658–668.

447
Kalyvas, Stathis N., Ian Shapiro and Tarek Masoud. 2008. Order, Conflict and
Violence. Cambridge: Cambridge University Press.
Kenny, Paul and Mónica Serrano. 2012a. The Mexican State and Organized Crime:
An Unending Story. In Mexico’s Security Failure. Collapse into Criminal Violence,
ed. Paul Kenny, Ménica Serrano and Arturo Sotomayor. New York: Routledge
chapter 1, pp. 29–53.
Kenny, Paul and Mónica Serrano. 2012b. Transition to Dystopia. 1994-2008. In Mex-
ico’s Security Failure. Collapse into Criminal Violence. New York: Routledge
chapter 2, pp. 54–88.
Kilmer, Beau, Jonathan P. Caulkins, Brittany M. Bond and Peter H. Reuter. 2010.
“Reducing Drug Trafficking Revenues and Violence in Mexico. Would Legalizing
Marijuana in California Help?”.
Kilmer, Beau and Rosalie Liccardo Pacula. 2009. “Estimating the size of the global
drug market. A demand-side approach. Report 2.”.
King, Gary, Robert O. Keohane and Sidney Verba. 1994. Designing Social Inquiry.
Princeton University Press.
Kleiman, Mark A. R. 2009. When Brute Force Fails: How to Have Less Crime and
Less Punishment. Princeton, New Jersey: Princeton University Press.
Kleiman, Mark A. R. 2011. “Surgical Strikes in the Drug Wars. Smarter Policies for
Both Sides of the Border.” Foreign Affairs . Last accessed on 26/6/2013.
URL: http://www.foreignaffairs.com/articles/68131/mark-kleiman/surgical-
strikes-in-the-drug-wars?page=show
Knight, Alan. 1990a. The Mexican Revolution. Volume 1: Porfirians, Liberals, and
Peasants. Lincoln NE: University of Nebraska Press.
Knight, Alan. 1990b. The Mexican Revolution. Volume 2: Counter-revolution and
Reconstruction. Lincoln NE: University of Nebraska Press.
Knight, Alan. 2012. Narco-Violence and the State in Modern Mexico. In Violence,
Coercion, and State-Making in Twentieth-Century Mexico: The Other Half of the
Centaur, ed. Wil G Pansters. Stanford, CA: Stanford University Press pp. 115–134.
Koonings, Kees and Dirk Kruijt. 2004. Armed Actors: Organized Violence and State
Failure in Latin America. New York: Zed Books.
Kornhauser, Ruth R. 1978. Social Sources of Delinquency: An Apprisal of Analytic
Models. Chicago: Chicago University Press.
La Policiaca. 2013. “Dictan formal prisión a agente de tránsito.”. Last accessed on
26/6/2013.
URL: http://www.lapoliciaca.com/nota-roja/dictan-formal-prision-a-agente-de-
transito/

448
Laakso, Markku and Rein Taagepera. 1979. “The Effective Number of Parties: A
Measure with Application to West Europe.” Comparative Political Studies 12(1):3–
27.

Langston, Joy. 2001. “Why rules matter: Changes in candidate selection in Mexico’s
PRI.” Journal of Latin American Studies 33(3):485–511.

Langston, Joy. 2006. The birth and transofrmation of the Dedazo in Mexico. In Infor-
mal institutions and democracy. Lessons from Latin America, ed. Gratchen Henke
and Steven Levitsky. Maryland: The Johns Hopkins University Press pp. 143–.

Latin American Commission on Drugs and Democracy. 2009. “Drugs and Democracy:
Toward a Paradigm Shift.”. Last accessed on 26/6/2013.
URL: http://www.drogasedemocracia.org/Arquivos/declaracao ingles site.pdf

Laurell, Asa Cristina. 1992. México, una democracia restringida. Mexico: Casa del
Sol.

Lazer, David, Alex Pentland, Lada Adamic, Sinan Aral, Albert-László Barabási,
Devon Brewer, Nicholas Christakis, Noshir Contractor, James Fowler, Myron Gut-
mann, Tony Jebara, Gary King, Michael Macy, Deb Roy and Marshall Van Alstyne.
2009. “Computational Social Science.” Science 323(6 February):721–723.

Lederman, Daniel, Norman Loayza and Ana Maria Menendez. 2002. “Violent
Crime: Does Social Capital Matter?” Economic Development and Cultural Change
50(3):509–539.

Leetaru, Kalev and Philip A. Schrodt. 2013. “GDELT: Global Data on Events,
Location and Tone, 1979-2012.”.

Lessing, Benjamin. 2012. “The Logic of Violence in Criminal War: Cartel-State

Conflict in Mexico, Colombia, and Brazil.”.

Levitt, Steven D. 1996. “The Effect of Prison Population Size on Crime Rate: Ev-
idence from Prision Overcrowding Litigation.” Quarterly Journal of Economics
111:319–352.

Levitt, Steven D. 1997. “Using electoral cycles in police hiring to estimate the effect
of police on crime.” American Economic Review 87:270–290.

Levitt, Steven D. and Stephen Dubner. 2005. Why do drug-dealers still live with
their moms? In Freakonomics: A Rogue Economist Explores the Hidden Side of
Everything. New York: William Morrow pp. 85–114.

Lichbach, Mark I. 1984. “An economic theory of governability : Choosing policy and
optimizing performance.” Public Choice 44(2):307–337.

Lichbach, Mark I. 1987. “Deterrence or Escalation?: The Puzzle of Aggregate Studies

of Repression and Dissent.” Journal of Conflict Resolution 31(2):266–297.

449
Lichbach, Mark I. 1998. The Rebel’s Dilemma. University of Michigan Press. Last
accessed on 26/6/2013.
URL: http://books.google.com/books?id=CWJIl5A5NOMC&pgis=1

Lieberman, Evan S. 2005. “Nested Analysis as a Mixed-Method Strategy for Com-

parative Research.” American Political Science Review 99(3):435–452.

Linz, Juan and Alfred Stepan. 1996. Problems of Democratic Transition and Consoli-
dation: Southern Europe, South America, and Post-Communist Europe. Maryland:
The Johns Hopkins University Press.

Loaeza, Soledad. 1994. El Partido Acción Nacional: la larga marcha, 1939-1994.

Oposición leal y partido de protesta. Mexico: Fondo de Cultura Económica.

Loaeza, Soledad. 1999. “La crisis electoral del 6 de julio de 1988.” Revista Mexicana
de Sociologı́a 61(3):163–182.

López, Claudia. 2010. Y refundaron la patria? De como mafiosos y politicos recon-

figuraron el Estado colombiano. Barcelona: Random House Mondadori.

Lujambio, Alonso. 2000. El Poder Compartido. Ensayo Sobre la Democratización

Mexicana. Mexico City: Oceano.

Lupsha, Peter A. 1995. “Transnational Narco-Corruption and Narco Investment: A

Focus on Mexico.” Transnational Organized Crime 1(1):84–101.

Lustig, Nora. 1990. “Economic Crisis, Adjustment and Living Standards in Mexico,
1982-85.” World Development 18(10):1325–1342.

Luttwak, Eduard N. 1999. “Give War a Chance.” Foreign Affairs 78(4):36–44.

Madrazo, Alejandro and Ángela Guerrero. 2012. “Más Caro el Caldo que las
Albóndigas.” Nexos . Last accessed on 26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=2103069

Magaloni, Beatriz. 2006. Voting for Autocracy. New York: Cambridge University
Press.

Maltz, Michael D. 1976. “On Defining ’Organized Crime’: The Development of a

Definition and a Typology.” Crime and Delinquency 26(July):338–346.

Mandel, Robert. 2011. Dark Logic: Transnational Criminal Tactics and Global Se-
curity. Stanford, CA: Stanford University Press.

Mansfield, Edward D. and Jack Snyder. 2005. Electing to Fight: Why Emercing
Democracies Go to War. Cambridge, MA: MIT Press.

Martı́nez, Fabiola, Ciro Pérez, Gabriel León and Georgina Saldierna. 2007. “Fox sigue
cosechando crı́ticas por su intromisión electoral.”. Last accessed on 26/6/2013.
URL: http://goo.gl/8A208

450
Martı́nez, Sanjuana. 2013. “Taxistas o halcones?” Sinembargo . Last accessed on
26/6/2013.
URL: http://www.sinembargo.mx/opinion/16-04-2012/6272

Marvell, Thomas B and Carlisle E Moody. 1994. “Prision Population and Crime
Reduction.” Journal of Quantitative Methodology 10:109–139.

Mason, T. David and Dale A. Krane. 1989. “The Political Economy of Death Squads:
Toward a Theory of the Impact of State-Sanctioned Terror.” International Studies
Quarterly 33(2):175 – 198.

Mason, T. David and Patrick J. Fett. 1996. “How Civil Wars End: A Rational Choice
Approach.” Journal of Conflict Resolution 40(4):546–568.

Maurer, Noel. 2012. “Criminal Organizations and Enterprise.” Revista. Harvard

Review of Latin America . Last accessed on 26/6/2013.
URL: http://www.drclas.harvard.edu/publications/revistaonline/winter-
2012/criminal-organizations-and-enterprise

McAdam, Doug. 1982. Political Process and the Development of Black Insurgency,
1930-1970. Chicago: Chicago University Press.

McAdam, Doug, Sidney G. Tarrow and Charles Tilly. 2001. Dynamics of Contention.
Cambridge, MA: Cambridge University Press.

McCarthy, John D. and Mayer N. Zald. 1977. “Resource mobilization and social
movements: A partial theory.” The American Journal of Sociology 82(6):1212–
1241.

McClelland, Charles A. 1978. “World Event/Interaction Survey (WEIS) Project,

1966-1978.”.

McClelland, Charles A. and Gary D. Hoggard. 1969. Conflict patterns in the inter-
actions among nations. New York: Free Press.

McKenley, James C. 2006. “Electoral Crisis in Mexico as Top 2 Declare Victory.”.

Last accessed on 26/6/2013.
URL: http://goo.gl/nCHmQ

Medina Peña, Luis. 1994. Hacia el nuevo Estado. México, 1920-1993. Mexico City:
Fondo de Cultura Económica.

Merino, José. 2011. “Los operativos conjuntos y la tasa de homicidios: Una

medición.”. Last accessed on 26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=2099329

Merino, Mauricio. 2003. La Transición Votada. Crı́tica a la Interpretación del Cambio

Polı́tico en México. México D.F.: Fondo de Cultura Economica.

451
Merritt, Richard L., Robert G. Muncaster and Dina A. Zinne. 1993. International
event-data developments : DDIR phase II. Ann Arbor, MI: University of Michigan
Press.
Merton, Robert K. 1938. “Social Structure and Anomie.” American Sociological
Review (3):672–682.
Merton, Robert K. 1957. Social Theory and Social Structure. New York: The Free
Press.
Mexico Seguridad. 2013. “Renovará armamento Acámbaro, Guanajuato, con
recursos del Subsemun.”. Last accessed on 26/6/2013.
URL: http://mexicoseguridad.mx/renovara-armamento-acambaro-guanajuato-
con-recursos-del-subsemun/
Miguel, Edward, Shanker Satyanath and Ernest Sergenti. 2004. “Economic Shocks
and Civil Conflict: An Instrumental Variables Approach.” Journal of Political
Economy 112(4):725–753.
Miller, Walter. 1958. “Lower Class Culture as a Generating Mileu of Gang Delin-
quency.” Journal of Social Issues 14:5–19.
Mitofsky, Consulta. 2012. México: Confianza en instituciones. Technical report Con-
sulta Mitofsky Mexico: . Last accessed on 26/6/2013.
URL: http://goo.gl/zh0iL
Molinar Horcasitas, Juan and Jeffrey A. Weldon. 2003. Reforming Electoral Systems
in Mexico. In Mixed-Member Electoral Systems: The Best of Both Worlds?, ed.
Matthew Soberg Shugart and Martin P. Wattenberg. Oxford: Oxford Scholarship
Online pp. 209–230. Last accessed on 26/6/2013.
URL: http://www.oxfordscholarship.com/view/10.1093/019925768X.001.0001/acprof-
9780199257683-chapter-11
Monreal, Ricardo. 2008. “Narcoterrorismo.”. Last accessed on 26/6/2013.
URL: http://lacolumna.wordpress.com/2008/09/23/narcoterrorismo-ricardo-
monreal-avila/
Montemayor, Carlos. 2007. La guerrilla recurrente. Mexico: Debate.
Montemayor, Carlos. 2009. Guerra en el paraı́so. Mexico: Random House Mondadori.
Montemayor, Carlos. 2010. La Violencia de estado en Mexico. Mexico: Grijalbo.
Moore, Will H. 1995. “Action-Reaction or Rational Expectations? Reciprocity and
the Domestic-International Conflict Nexus during the ”Rhodesia Problem”.” Jour-
nal of Conflict Resolution 39(1):129–167.
Moore, Will H. 1998. “Repression and Dissent: Substitution, Context, and Timing.”
American Journal of Political Science 42(3):851 – 873. Last accessed on 26/6/2013.
URL: http://www.jstor.org/stable/2991732

452
Moore, Will H. 2000. “The Repression of Dissent: A Substitution Model of Govern-
ment Coercion.” Journal of Conflict Resolution 44(1):107–127.
Moreno, Alejandro. 2010. Confianza en instiuciones públicas. México en perspectiva
comparada. Technical report Centro de Estudios Sociales y de Opinión Pública
Mexico: .
Morris, Stephen D. 2012. “Corruption , Drug Tra cking, and Violence in Mexico.”
Brown Journal of World Affairs 18(2):29–43.
Morris, Stephen D. 2013. “The impact of Drug-Related Violence on Corruptio in
Mexico.” The Latinamericanist 57(1):43–64.
Muedano, Marcos. 2011. “El presidente que aprendió el negocio de las drogas.”. Last
accessed on 26/6/2013.
URL: http://www.eluniversal.com.mx/nacion/186543.html
Muller, John E. 1970. “Presidential Popularity from Truman to Johnson.” The Amer-
ican Political Science Review 64(1):18–34.
Nagin, Daniel S. 1998. Deterrence and Incapacitation. In The Handbook of Crime
and Punishment, ed. Michael Tonry. New York: Oxford University Press.
Nagin, Daniel S, Greg Pogarsky and David P Farrington. 1997. “Adolescent Mothers
and the Criminal Behavior of Their Children.” Law & Society Review 31(1):137–
162.
National Advisory Committee on Criminal Justice Standards and Goals. 1976. Or-
ganized Crime: Report of the Task Force on Organized Crime. Technical report
Government Printing Office Washington, D.C.: .
Notimex. 2006. “Interpone PRD denuncia contra Fox en la PGR.”. Last accessed on
26/6/2013.
URL: http://www.eluniversal.com.mx/notas/350653.html
Notimex. 2010. “Acribillan a tres policı́as municipales en Ciudad Juárez.”. Last
accessed on 26/6/2013.
URL: http://http://www.cronica.com.mx/notas/2009/438136.html
Notimex. 2012. “Desplaza guerra del narco a 160 mil personas en México: António
Guterres.”. Last accessed on 26/6/2013.
URL: http://www.animalpolitico.com/2012/04/desplaza-guerra-del-narco-a-160-
mil-personas-en-mexico-onu/#axzz2W1WHgBDW
Obert, Jonathan M. 2011. “The Illusion of a Violent Monopoly: Explaining Trajec-
tories of Security Institutionalization.”.
O’Donnell, Guillermo A. 1993. “On the State, Democratization, and Some Concep-
tual Problems: A Latin American View with Glances at some Post-communist
Countries.” World Development 21(8):1355–1369.

453
O’Donnell, Guillermo A. 1994. “Delegative Democracy.” Journal of Democracy
5(1):55–69.
O’Donnell, Guillermo A., Philippe C. Schmitter and Laurence Whitehead. 1986.
Transitions from authoritarian rule: Comparative perspectives. Vol. 3 JHU Press.
Office of National Drug Control Policy. 2004. Technical Report For the Price and
Purity of Illicit Drugs: 1981 Through the Second Quarter of 2003. Technical report
The Executive Office of the President, Office of National Drug Control Policy
Washington, D.C.: .
Olson, Eirc, David A. Shirk and Andrew Selee, eds. 2010. Shared Responsibility:
U.S.-Mexico Policy Options for Confronting Organized Crime. Washington, D.C.:
.
Olson, Mancur. 1965. The Logic of Collective Action. Cambridge: Harvard Economic
Studies.
Olson, Mancur. 2000. Power And Prosperity: Outgrowing Communist And Capitalist
Dictatorships. New York: Basic Books.
Oppenheimer, Andres. 1996. Bordering on Chaos: Mexico’s Roller-Coaster Journey
Toward Prosperity. Boston: Little, Brown & Company.
Organization of American States. 2013. “OAS General Assembly: Countries Discuss
Drug Policies and Other Matters.”. Last accessed on 7/7/2013.
URL: http://www.oas.org/en/media center/press release.asp?sCodigo=E-230/13
Osorio, Javier. 2012. Las causas estructurales de la violencia: Evaluación de algunas
hipótesis. In Las Bases Sociales del Crimen Organizado y la Violencia en México,
ed. José Antonio Aguilar Rivera. Mexico City: Centro de Investigación y Estudios
en Seguridad - Secretarı́a de Seguridad Pública pp. 73–132. Last accessed on
26/6/2013.
URL: http://www.cies.gob.mx/pdf/Bases Sociales Crimen Organizado Mexico.pdf
Osorno, Diego E. 2009. El Cartel de Sinaloa. México D.F.: Random House Mon-
dadori.
Osorno, Diego E. 2012. La Guerra de los Zetas. Mexico: Grijalbo.
Pacheco, Istra. 2006. “Mexico Candidate Claiming Election Fraud.”. Last accessed
on 26/6/2013.
URL: http://goo.gl/WBa6E
Parametrı́a. 2012. “Confianza en instituciones.”. Last accessed on 26/6/2013.
URL: http://www.parametria.com.mx/DetalleEstudio.php?E=4480
Paras, Pablo. 2007. Unweaving the Social Fabric: The Impact of Crime on Social Cap-
ital. In Reforming the Administration of Justice in Mexico, ed. Wayne A. Cornelius
and David A. Shirk. South Bend: University of Notre Dame Press pp. 323–348.

454
Partido Nacional Revolucionario. 1929. “Estatutos del Partido Nacional Revolu-
cionario.”. Last accessed on 26/6/2013.
URL: http://goo.gl/2B084

Pedraza, Iván. 2013. “Asesinan a balazos a policı́a municipal de Sonora.”. Last

accessed on 26/6/2013.
URL: http://www.eluniversal.com.mx/notas/917610.html

Peschard, Jaqueline. 1995. “México 1994: un nuevo marco electoral para la elección
presidencial.” Perfiles Latinoamericanos (5):105–129.

Petraeus, David. 2006. The Army Counterinsurgency Manual. US Army.

Piven, Frances Fox and Richard Cloward. 1977. Poor People’s Movements: Why
They Succeed, How They Fail. New York: Vintage Books.

Poiré, Alejandro and Marı́a Teresa Martı́nez. 2011. “La caı́da de los capos no multi-
plica la violencia El caso de Nacho Coronel.”. Last accessed on 26/6/2013.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=2099273

Poniatowska, Elena. 1971. La noche de Tlatelolco. Tesimonios de historia oral. Mex-

ico City: Era.

Poniatowska, Elena. 1980. Fuerte es el silencio. Mexico City: Era.

Posen, Barry R. 1993. “The security dilemma and ethnic conflict.” Survival: Global
Politics and Strategy 35(1):27–47. Last accessed on 26/6/2013.
URL: http://www.informaworld.com/10.1080/00396339308442672

Prados, Luis. 2013. “Código rojo en el norte de México.”. Last accessed on 26/6/2013.
URL: http://goo.gl/HwntQ

Presidencia de la República. 2005. “Ordena Presidente Vicente Fox puesta en marcha

del ‘Operativo México Seguro’ contra el crimen organizado.”. Last accessed on
26/6/2013.
URL: http://fox.presidencia.gob.mx/actividades/orden/?contenido=18872

Presidencia De La República. 2010. “El Gobierno Mexicano. Crónica Mensual No.

39 Febrero 2010.”. Last accessed on 9/4/2011.
URL: http://goo.gl/dAljf

Presidencia de la República. 2012. “Sexto Informe de Gobierno.”. Last accessed on

26/6/2013.
URL: http://goo.gl/tBKtW

Proceso. 2013. “Recluta narco a niños como halcones en Coahuila, denuncia obispo
de Saltillo.” Proceso . Last accessed on 26/6/2013.
URL: http://www.proceso.com.mx/?p=275827

455
Procuradurı́a General de la República. 2001a. “Boletı́n No. 046/01.”. Last accessed
on 26/6/2013.
URL: http://www.pgr.gob.mx/cmsocial/bol01/ene/b04601.html

Procuradurı́a General de la República. 2001b. “Boletı́n No. 102/01.”. Last accessed

on 26/6/2013.
URL: http://www.pgr.gob.mx/cmsocial/bol01/feb/b10201.html

Przeworski, Adam. 1991. Democracy and the market. Political and economic reforms
in Eastern Europe and Latin America. Cambridge, MA: Cambridge University
Press.

Przeworski, Adam, Michael E. Alvarez, José Antonio Cheibub and Fernando Limongi.
2000. Democracy and Development. New York: Cambridge University Press.

Putnam, Robert D. 1993. Making Democracy Work: Civic Traditions in Modern

Italy. Princeton: Princeton University Press.

Putnam, Robert D. 2000. Bowling Alone. New York: Simon and Shuster.

Quinn, Kevin, Burt L. Monroe, Michael Colaresi, Michael H. Crespin and

Dragomit R. Radev. 2010. “How to Analyze Political Attention with Minimal
Assumptions and Costs.” American Journal of Political Science 54(1):209–228.

Ravelo, Ricardo. 2007a. Crónicas de Sangre. Cinco Historias de los Zetas. Mexico
City: Debolsillo.

Ravelo, Ricardo. 2007b. Herencia Maldita. El Reto de Calderón y el Nuevo Mapa del
Narcotráfico. Mexico: Grijalbo.

Ravelo, Ricardo. 2009. Osiel. Vida y Tragedia de un Capo. México D.F.: Grijalbo
Mondadori.

Redacción de AFP. 2012. “Asesinan a cuatro policı́as municipales en México durante

emboscada.”.

Redacción de El Universal. 2009. “Habrá un nuevo Lucio Cabañas?”. Last accessed

on 26/6/2013.
URL: http://blogs.eluniversal.com.mx/wweblogs detalle.php?p fecha=2009-12-
02&p id blog=63&p id tema=9354

Redacción del Universal. 2010. “’Los Zetas’ torturan y matan a seis agentes munici-
pales en NL.”. Last accessed on 26/6/2013.
URL: http://www.eluniversal.com.mx/estados/76448.html

Regan, Patrick M. 2009. “Political Repression as a Public Good.”.

Reuter, Peter. 1989. Disorganized Crime. The Economics of the Invisible Hand.
Cambridge: MIT Press.

456
Reuter, Peter. 2008. “Can Production and Trafficking of Illicit Drugs be Reduced or
Merely Shifted ?”.

Reuter, Peter H. 2009. “Systemic Violence in Drug Markets.” Crime, Law and Social
Change 52(3):275–284.

Revueltas, José. 1978. México 68. Juventud y revolución. Mexico City: Era.

Rios, Viridiana. 2012a. “How Government Structure Encourages Criminal Violence.

The causes of Mexico’s Drug War.”. Last accessed on 26/6/2013.
URL: http://www.gov.harvard.edu/files/Rios PhDDissertation.pdf

Rios, Viridiana. 2012b. “Tendencias y explicaciones al asesinato de periodistas y al-

caldes en México: El crimen organizado y la violencia de alto perfil.” Last accessed
on 26/6/2013.
URL: http://www.gov.harvard.edu/files/Rios2012 PoliticalAssassinations.pdf

Rivera, Rubén and Jesús Alberto Valdez. 2007. “Lucio Cabañas Barrientos: Las
cuentas siguen pendientes.” Militante . Last accessed on 26/6/2013.
URL: http://www.militante.org/node/6

Robles de la Rosa, Leticia. 2012. “PRI denuncia campaña negra en parabuses.”. Last
accessed on 26/6/2013.
URL: http://www.excelsior.com.mx/2012/03/06/nacional/816001

Rodrı́guez Araujo, Octavio, ed. 2009. Mexico: un nuevo regimen politico? Mexico:
Siglo XXI.

Rohner, Ronald P. and Leonard Katz. 1970. “Testing for Validity and Reliability in
Cross-Cultural Research.” American Anthropoligist 72:1068–1073.

Ross, Michael. 2006. “A Closer Look At Oil, Diamonds, and Civil War.” Annual
Review of Political Science 9(1):265–300.

Rother, Larry. 1989. “Mexicans arrest top drug figure and 80 policemen.”. Last
accessed on 26/6/2013.
URL: http://www.nytimes.com/1989/04/11/world/mexicans-arrest-top-drug-
figure-and-80-policemen.html

Rubio, Luis and Susana Kaufman, eds. 2006. Mexico, democracia ineficaz. Mexico:
H. Camara de Diputados, LIX Legislatura : CIDAC : Miguel Ángel Porrua.

Rubio, Mauricio. 1997. “Perverse Social Capital: Some Evidence from Colombia.”
Journal of Economic Issues 31(3):805–816.

Rummel, Rudolph J. 1966. “A Foreign Conflict Code Sheet.” World Politics

18(2):283–296.

457
Rummel, Rudolph J. 1976a. Dimensionality of Nations Project: Dyadic Foreign
Conflict Variables, 1950-1965. Ann Arbor, MI: Inter-University Consortium for
Political and Social Research.
Rummel, Rudolph J. 1976b. The Dimensionality of Nations Project. In Quantita-
tive International Politics, ed. Francis W. Hoole and Dina A. Zinnes. New York:
Praeger pp. 149–154.
Rummel, Rudolph J. 1979. Understanding Conflict and War. Vol. 4. War, Power
and Peace. Beverly Hills, CA: Sage.
Salmi, Venla and Janne Kivivuori. 2006. “The Association between Social Capital
and Juvenile Crime. The Role of Individual and Structural Factors.” European
Journal of Criminology 3(2):123–148.
Sambanis, Nicolas. 2004. “What Is Civil War?: Conceptual and Empirical Complex-
ities of an Operational Definition.” Journal of Conflict Resolution 48(6):814–858.
Sampson, Robert J, John H Laub and Christopher Wimer. 2006. “Does Marriage
Reduce Crime? A Counterfactual Approach to Within-Individual Causal Effects.”
Criminology 44:465–508.
Schedler, Andreas. 2002. “The Menu of Manipulation.” Journal of Democracy
13(2):36–50.
Schedler, Andreas. 2006. Electoral Authoritarianism. The Dynamics of Unfree Com-
petition. London: Lynne Rienner Publishers.
Schedler, Andreas, Larry Diamond and Marc Plattner. 1999. The Self-Restraining
State. Boulder, CO: Lyne Rienner.
Schelling, Thomas. 1960. The strategy of conflict. Cambridge: Harvard University
Press.
Schelling, Thomas. 1967a. Economic Analysis and Organized Crime. Washington,
D.C.: Task Force Report: Organized Crime, Annotations and Consultant’s Papers.
Schelling, Thomas. 1967b. “Economics and Criminal Enterprise.” Public Interest
Spring:61–78.
Schelling, Thomas. 1971. “What is the Business with Organized Crime?” Journal of
Public Law 20(1):71–84.
Schrodt, Philip A. 2001. Automated Coding of International Event Data Using
Sparse Parsing Techniques. Chicago: Paper presented at the annual meeting of
the International Studies Association. Last accessed on 26/6/2013.
URL: http://polmeth.wustl.edu/media/Paper/schro01b.pdf
Schrodt, Philip A. 2006. “Twenty Years of the Kansas Event Data System Project.”
The Political Methodologist 14(1):2–6.

458
Schrodt, Philip A. 2009. “TABARI. Textual Analysis by Augmented Replacement
Instructions.”.

Schrodt, Philip A. and Deborah Gerner. 2012. Fundamentals of Machine Coding. In

Analyzing International Event Data: A Handbook of Computer-Based Techniques.
Last accessed on 26/6/2013.
URL: http://eventdata.psu.edu/papers.dir/AIED.2012.ch2.pdf

Schrodt, Philip A., Shannon G. Davis and Judith L. Weddle. 1994. “Political Science:
KEDS - A program for the machine coding of event data.” Social Science Computer
Review 12(4):561–587.

Schwedler, Jillian. 2005. “Cop Rock: Protest, Identity, and Dancing Riot Police in
Jordan.” Social Movement Studies 4(2):155–175. Last accessed on 26/6/2013.
URL: http://www.informaworld.com/10.1080/14742830500191410

Scott, James C. 2009. The Art of Not Being Governed: An Anarchist History of
Upland Southeast Asia. New Heaven, CT: Yale University Press.

Scott, Peter D. 2010. American War Machine. Deep politics, the CIA, Global Drug
Connection, and the Road to Afghanistan. Maryland: Rowman & Littlefield Pub-
lishers.

Secretarı́a de Gobernación. 2013. “Subsidio para la Seguiridad en los Municipios

(SUBSEMUN).”. Last accessed on 5/12/2013.
URL: http://www.secretariadoejecutivo.gob.mx/es/SecretariadoEjecutivo/Subsemun

Secretarı́a de la Defensa Nacional. 2011. “Solicitud de acceso a la información No.

0000700066311.”. Last accessed on 5/5/2011.
URL: http://buscador.ifai.org.mx/buscador/buscar.do

Secretarı́a de la Defensa Nacional. 2012. “Solicitud de acceso a la información No.

0000700074711.”. Last accessed on 26/6/2013.
URL: http://buscador.ifai.org.mx/buscador/buscar.do

Secretarı́a de Salud. 2012a. “Base de datos sobre egresos hospitalarios.”. Last accessed
on 12/09/2012.
URL: http://www.sinais.salud.gob.mx/basesdedatos/std egresoshospitalarios.html

Secretarı́a de Salud. 2012b. Encuesta Nacional de Adicciones 2011. Technical report

Secretaı́a de Salud, Centro Nacional para la Prevención y el Control de las Adic-
ciones, Comisión Nacional contra las Adicciones Mexico City: . Last accessed on
26/6/2013.
URL: http://encuestas.insp.mx/ena/ena2011.html#.UZHBSbXql8E

Secretarı́a de Salud. 2013. “Defunciones Generales.”. Last accessed on 26/6/2013.

URL: http://www.sinais.salud.gob.mx/basesdedatos/defunciones.html

459
Secretarı́a de Salud y Secretarı́a de Desarrollo Social. 2009. “Campaña Nacional de
Información para una Nueva Vida: 10 Recomendaciones para prevenir que tus
hijos utilicen drogas.”. Last accessed on 26/6/2013.
URL: http://www.conadic.salud.gob.mx/pdfs/nueva vida/nv10 recomendaciones.pdf

Serrano, Mónica. 2007. “México: narcotráfico y gobernabilidad.” Pensamiento

Iberoamericano (1):251–278. Last accessed on 26/6/2013.
URL: http://goo.gl/dPP7R

Service, Mexico Media. 2012. “Gobierno oculta resultados de encuesta 2011 sobre
adicciones.”. Last accessed on 26/6/2013.
URL: http://mmspress.com.mx/noticias/gobierno-oculta-resultados-de-encuesta-
2011-sobre-adicciones/

Shellman, S. M., C. Hatfield and M. J. Mills. 2010. “Disaggregating actors in intra-

national conflict.” Journal of Peace Research 47(1):83–90.

Shellman, Stephen M. 2004. “Time series intervals and statistical inference: The
effects of temporal aggregation on event data analysis.” Political Analysis 12(1):97–
104.

Shellman, Stephen M. 2008. “Coding Disaggregated Intrastate Conflict: Machine

Processing the Behavior of Substate Actors Over Time and Space.” Political Anal-
ysis 16(4):464–477.

Shellman, Stephen M. 2013. “Violent Intranational Political Conflict & Terrorism

(VIPCAT).”. Last accessed on 01/10/2013.
URL: http://www.wm.edu/offices/itpir/vipcat/index.php

Shirk, David A. 2005. Mexico’s new politics. The PAN and Democratic Change. New
York: Lynne Rienner Publishers.

Simon, Marc V. 2008. “Hawks, doves, and civil conflict dynamics : A strategic
action-reaction model.” International Interactions 19(3):213–239.

Sims, Christopher. 1980. “Macroeconomics and Reality.” Econometrica 48(1):1–48.

Sims, Christopher and Tao Zha. 1999. “Error Bands for Impulse Responses.” Econo-
metrica 67(5):1113–1155.

Sipes, R. G. 1976. “A Test For Coder Bias.” Cross-Cultural Research 11(3):149–168.

Sistema Nacional de Información en Salud. 2013. “Mortalidad. Información tabular.

Principales causas de mortalidad general.”. Last accessed on 04/28/2013.
URL: http://www.sinais.salud.gob.mx/mortalidad/index.html

Sistema Nacional de Seguridad Pública. 2011a. “Base de Datos de Fallecimientos

Ocurridos por Presunta Rivalidad Delincuencial.”. Last accessed on 9/4/2011.
URL: http://200.23.123.5/GraficasSimples.aspx

460
Sistema Nacional de Seguridad Pública. 2011b. “Metodologı́a de la Base de Datos de
Fallecimientos Ocurridos por Presunta Rivalidad Delincuencial.”. Last accessed on
9/4/2011.
URL: http://200.23.123.5/Documentos.aspx

Skaperdas, Stergios. 2002. “Warlord Competition.” Journal of Peace Research

39(4):435–446.

Skaperdas, Stergios and Constantinos Syropoulos. 1995. Gangs as primitive states.

In The economics of organized crime, ed. Gianluca Fiorentini and Sam Peltzman.
Cambridge, New York: Cambridge University Press.

Skocpol, Theda. 1979. States and Social Revolutions. Cambridge University Press.

Snyder, David and William R. Kelly. 1977. “Conflict Intensity, Media Sensitivity and
the Validity of Newspaper Data.” American Sociological Review 42(2):105–123.

Snyder, Jack. 2000. From Voting to Violence. New York: Norton.

Snyder, Richard. 2006. “Does Lootable Wealth Breed Disorder?: A Political Economy
of Extraction Framework.” Comparative Political Studies 39(8):943–968.

Snyder, Richard and Angelica Duran-Martinez. 2009. “Does illegality breed violence?
Drug trafficking and state-sponsored protection rackets.” Crime, Law, and Social
Change 52:253–273.

Snyder, Richard and Ravi Bhavnani. 2005. “Diamonds, Blood, and Taxes: A
Revenue-Centered Framework for Explaining Political Order.” Journal of Conflict
Resolution 49(4):563–597.

Stock, James H, Jonathan H Wright and Motohiro Yogo. 2002. “A Survey of Weak
Instruments and Weak Identification in Generalized Method of Moments.” Journal
of Business & Economic Statistics 20(4):518–529.

Stockwell, Robert P., Donal D. Bowen and John W. Martin. 1965. The Gramatical
Structures of English and Spanish. Chicago: The University of Chicago Press.

Subramanian, Devika and Richard J. Stoll. 2006a. Events, Patterns, and Analysis:
Forcasting International Conflict in the Twenty-First Century. In Programming for
Peace: Computer-Aided Methods for International Conflict Resolution and Preven-
tion, ed. Robert Trappl. Dordrecht: Springer pp. 145–160.

Subramanian, Devika and Richard J. Stoll. 2006b. “Events, Patterns, and Analysis:
Forecasting International Conflict in the Twenty-First Century.”. Last accessed on
26/6/2013.
URL: http://www.cs.rice.edu/ devika/projects/stoll.html

Tanter, Raymond. 1972. “International system and foreign policy appraoches: Im-
plications for conflict modelling and management.” World Politics 24:7–39.

461
Tarrow, Sidney G. 1998. Power in movement: social movements and contentious
politics. Cambridge: Cambridge University Press.
The New York Times. 2011. “Excerpts From an Interview With President Felipe
Calderón.”. Last accessed on 26/6/2013.
URL: http://www.nytimes.com/2011/10/16/world/excerpts-from-an-interview-
with-president-felipe-calderon.html? r=0
The Vienna Declaration. 2010. The Vienna Declaration: A Global Call to Action for
Science-based Drug Policy. Technical report XVIII International AIDS Conference.
Last accessed on 26/6/2013.
URL: http://www.viennadeclaration.com/the-declaration/
The World Bank. 2010. “Gross national income percapita 2010, Atlas method and
PPP.”. Last accessed on 04/08/2013.
URL: http://siteresources.worldbank.org/DATASTATISTICS/Resources/GNIPC.pdf
Thomas, Troy S., Stephen D. Kiser and William D. Casebeer. 2005. Warlords Rising:
Confronting Violent Non-State Actors. New York: Lexington Books.
Thomson, Janice E. 1996. Mercenaries, Pirates, and Sovereigns: State-Building and
Extraterritorial ... Princeton University Press.
Tilly, Charles. 1978. From mobilization to revolution. New York: Addison Wesley.
Tilly, Charles. 1985. War Making and State Making as Organized Crime. In Bringing
the State Back In, ed. Peter Evans, Dietrich Rueschemeyer and Theda Skocpol.
Cambridge: Cambridge University Press pp. 169–191.
Tilly, Charles. 1992. Coercion, Capital and European States, AD 990-1992. Malde,
MA: Blackwell Publishers.
Tilly, Charles, ed. 1975. The Formation of Nation States in Western Europe. Prince-
ton, New Jersey: Princeton University Press.
Tonry, Michael. 2009. Crime and Public Policy. In The Oxford Handbook of Crime
and Public Policy. New York: Oxford University Press pp. 3–24.
Toro, M C. 1999. “The internationalization of police: the DEA in Mexico.” Journal
of American Hstory 86(2):623–40.
Torres, Mauricio. 2013. “26,121 personas desaparecidas en el gobierno de Calderón,
reporta Segob.”. Last accessed on 26/6/2013.
URL: http://mexico.cnn.com/nacional/2013/02/26/26121-personas-
desaparecidas-en-el-gobierno-de-calderon-reporta-segob
Transparencia Mexicana. 2012. “Índice Nacional de Corrupción y Buen Gobierno.”.
Last accessed on 02/15/2013.
URL: http://www.tm.org.mx/indice-nacional-de-corrupcion-y-buen-gobierno-
incbg/

462
Trejo, Guillermo. 2009. “Religious Competition and Ethnic Mobilization in Latin
America: Why the Catholic Church Promotes Indigenous Movements in Mexico.”
American Political Science Review 103(03):323. Last accessed on 26/6/2013.
URL: http://www.journals.cambridge.org/abstract S0003055409990025

Tuckman, Jo. 2012. Mexico: Democracy Interrupted. New Heaven, CT: Yale Univer-
sity Pres.

Tulchinm, Joseph S., Hugo H. Frühling and Heather Golding. 2003. Crime and Vio-
lence in Latin America: Citizen Security, Democracy, and the State. Washington,
D.C.: Woodrow Wilson Center Press.

Tullock, G. 1980. Efficient rent seeking. In Toward a Theory of Rent Seeking Society,
ed. M. Buchanan, R. Tollison and G. Tullock. College Station, TX: Texas A&M
University Press.

Tversky, a and D Kahneman. 1974. “Judgment under Uncertainty: Heuristics and

Biases.” Science (New York, N.Y.) 185(4157):1124–31.

United Nations. 1961. “Single Convention on Narcotic Drugs of 1961 as amended by

the 1972 Protocol.”. Last accessed on 26/6/2013.
URL: http://www.unodc.org/unodc/en/treaties/single-convention.html

United Nations. 1971. “Convention on Psychotropic Substances, 1971.”. Last ac-

cessed on 9/1/2012.
URL: http://www.unodc.org/unodc/en/treaties/psychotropics.html

United Nations. 1988. “United Nations Convention against Illicit Traffic in Narcotic
Drugs and Psychotropic Substances, 1988.”. Last accessed on 9/1/2012.
URL: http://www.unodc.org/unodc/en/treaties/illicit-trafficking.html

United Nations. 2000. “United Nations Convention against Transnational Organized

Crime and the Protocols Thereto.”. Last accessed on 9/1/2012.
URL: http://www.unodc.org/unodc/en/treaties/CTOC/index.html

United Nations. 2002. “United Nations Convention against Corruption.”. Last ac-
cessed on 9/1/2012.
URL: http://www.unodc.org/unodc/en/treaties/CAC/index.html

United Nations Human Rights. 2011. “United Nations Working Group on Enforced
or Involuntary Disapperances concludes visit to Mexico.”. Last accessed on
4/11/2011.
URL: http://www.ohchr.org/en/NewsEvents/Pages/DisplayNews.aspx?NewsID=10907

United Nations Office on Drugs and Crime. 2008a. “The 1912 Hague International
Opium Convention.”.

463
United Nations Office on Drugs and Crime. 2008b. “The Shanghai Opium Commis-
sion.”. Last accessed on 25/5/2013.
URL: http://www.unodc.org/unodc/en/data-and-analysis/bulletin/bulletin 1959-
01-01 1 page006.html

United Nations Office on Drugs and Crime. 2013a. “Drug trafficking.”. Last accessed
on 04/07/2013.
URL: http://www.unodc.org/unodc/en/drug-trafficking/
United Nations Office on Drugs and Crime. 2013b. World Drug Report 2012. Tech-
nical report United Nations Office on Drugs and Crime Vienna: . Last accessed
on 26/6/2013.
URL: http://www.unodc.org/unodc/en/data-and-analysis/WDR-2012.html

UNODC. 2003. Terminology and information on drugs. 2nd ed. Viena: United
Nations Office on Drugs and Crime.
U.S. Department of Justice. 2008. National Drug Threat Assessment 2009. Technical
report Washington, D.C.: . Last accessed on 26/6/2013.
URL: http://www.justice.gov/archive/ndic/pubs31/31379/index.htm
Valentino, Benjamin. 2000. “Final solutions: The causes of mass killing and geno-
cide.” Security Studies 9(3):1–59.
Varese, Federico. 2013. Mafias on the Move: How Organized Crime Conquers New
Territories. Princeton, New Jersey: Princeton University Press.

Vargas Llosa, Mario. 1990. “Vargas Llosa: ‘México es la dictadura perfecta’.”. Last
accessed on 26/6/2013.
URL: http://elpais.com/diario/1990/09/01/cultura/652140001 850215.html

Vasquez, Juan M. 1985. “Mexico Accuses 2 Former Security Chiefs Over Amassed
Wealth; 427 Agents Fired.”. Last accessed on 26/6/2013.
URL: http://articles.latimes.com/1985-06-07/news/mn-16013 1 camarena-case

Villalobos, Joaquı́n. 2010. “Doce mitos de la guerra contra el narco.”. Last accessed
on 9/4/2011.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=72941
Villalobos, Joaquı́n. 2012. “Nuevos mitos de la guerra contra el narco.”. Last accessed
on 9/4/2011.
URL: http://www.nexos.com.mx/?P=leerarticulo&Article=2102505
Villamil, Jenaro. 2012. “Reitera Peña Nieto que no negociará ni pactará con el narco.”
Proceso . Last accessed on 26/6/2013.
URL: http://www.proceso.com.mx/?p=320310
Villareal, Angeles M. 2009. NAFTA and the Mexican Economy. Technical report
Congressional Research Service Washington, D.C.: .

464
Volgy, Thomas J. and Jon E. Quistgard. 1974. “Correlates of organizational rewards
in the United Nations: An analysis of environmntal and legislative variables.”
International Organization 28(2):179–205.
Volkov, Vadim. 2002. Violent Entrepreneurs. The Use of Force in the Making of
Russian Capitalism. Ithaca: Cornell University Press.
von Lampe, Kalus. 2013. “Definitions of Organized Crime.”. Last accessed on
04/06/2013.
URL: http://www.organized-crime.de/organizedcrimedefinitions.htm
Wagner, R. Harrison. 1993. The Causes of Peace. In Stopping the Killing: How Civil
Wars End, ed. Roy Licklider. New York: New York University Press pp. 235–268.
Walker, Victor E. 1969. Terror and Resistance: A Study of Political Violence. New
York: Oxford University Press.
Wallensteen, Peter. 1981. Patterns of Armed Conflict Since 1945: An Overview and
Some Implications. In Report for the Independent Commission on Disarmament
and Security Issues. Uppsala: Department of Peace and Conflict Studies.
Wasserman, Mark. 1993. Persistent Oligarchs. Elites and Politics in Chihuahua,
Mexico. 1910-1940. Durham: Duke University Press.
Weber, Max. 1978. Economy and Society. Los Angeles: University of California
Press.
Weinberg, Leonard, Ami Pedahzur and Sivan Hirsch-Hoefler. 2004. “The Challenges
of Conceptualizing Terrorism.” Terrorism and Political Violence 16(4):777–794.
Weinstein, Jeremy. 2007. Inside Rebellion: The Politics of Insurgent Violence. Cam-
bridge: Cambridge University Press.
Wickham-Crowley, Timothy P. 1992. Guerrillas and revolution in Latin America: a
comparative study of ... Princeton University Press. Last accessed on 26/6/2013.
URL: http://books.google.com/books?id=K08AHKJnFRUC&pgis=1
Womack, John. 1999. Rebellion in Chiapas. New York: The New Press.
World Health Organization. 2010a. “International Classification of Diseases (ICD).”.
Last accessed on 02/21/2013.
URL: http://www.who.int/classifications/icd/en/
World Health Organization. 2010b. “The ICD-10 Classification of Mental and
Behavioural Disorders: Clinical descriptions and diagnostic guidelines.”. Last
accessed on 26/6/2013.
URL: http://www.who.int/substance abuse/terminology/ICD10ClinicalDiagnosis.pdf
Youngers, Coletta A. and Eileen Rosin. 2005. Drugs and democracy in Latin America.
The impact of U.S. policy. Boulder, CO: Lynne Rienner Publishers.

465
Zárraga, Paola. 2006. “Mercado critica intromisión de Fox en las campañas.”. Last
accessed on 26/6/2013.
URL: http://www.eluniversal.com.mx/nacion/137536.html

Zedillo, Ernesto and Haynie Wheeler, eds. 2012. Rethinking the ”War on Drugs”
Through the US-Mexico Prism. New Heaven, CT: Yale Center for the Study of
Globalization. Last accessed on 26/6/2013.
URL: http://www.ycsg.yale.edu/center/forms/rethinking-war-on-drugs.pdf

466
This document was prepared & typeset with pdfLATEX, and formatted with
nddiss2ε classfile (v3.2013β[2013/01/31]) provided by Sameer Vijay and
updated by Megan Patnott.

467