Review of Economic Studies (2025) 92, 404–441 https://doi.org/10.

1093/restud/rdae027
© The Author(s) 2024. Published by Oxford University Press on behalf of The Review of Economic Studies Limited.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License
(https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any
medium, provided the original work is properly cited.
Advance access publication 12 March 2024

Transhumant Pastoralism,
Climate Change, and Conflict

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

in Africa
Eoin F. McGuirk
Tufts University, NBER and BREAD, Medford, USA
and
Nathan Nunn
University of British Columbia and CIFAR, Vancouver, Canada

First version received December 2021; Editorial decision July 2023; Accepted March 2024 (Eds.)

We consider the effects of climate change on seasonally migrant populations that herd
livestock—i.e. transhumant pastoralists—in Africa. Traditionally, transhumant pastoralists benefit from
a cooperative relationship with sedentary agriculturalists whereby arable land is used for crop farming in
the wet season and animal grazing in the dry season. Rainfall scarcity can disrupt this arrangement by
inducing pastoral groups to migrate to agricultural lands before the harvest, causing conflict to emerge.
We examine this hypothesis by combining ethnographic information on the traditional locations of tran-
shumant pastoralists and sedentary agriculturalists with high-resolution data on the location and timing
of rainfall and violent conflict events in Africa from 1989 to 2018. We find that reduced rainfall in the
territory of transhumant pastoralists leads to conflict in neighbouring areas. Consistent with the proposed
mechanism, the conflicts are concentrated in agricultural areas; they occur during the wet season and not
the dry season; and they are due to rainfall’s impact on plant biomass growth. Since pastoralists tend to be
Muslim and agriculturalists Christian, this mechanism accounts for a sizable proportion of the rapid rise
in religious conflict observed in recent decades. Regarding policy responses, we find that development
aid projects tend not to mitigate the effects that we document. By contrast, the effects are reduced when
transhumant pastoralists have greater power in national government, suggesting that more equal political
representation is conducive to peace.

Key words: Transhumant pastoralism, Sedentary agriculture, Seasonal migration, Conflict, Weather
JEL codes: N10, Q54, Z1

1. INTRODUCTION
Climate change is one of the most important challenges facing society. A fundamental concern
is that more frequent extreme weather events may lead to violent conflict and political instability
in fragile parts of the world. Many African countries are believed to be especially vulnerable to
this threat, due in part to low economic development, weak state capacity, and a high reliance

The editor in charge of this paper was Elias Papaioannou.

404
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 405

on crop agriculture. In this paper, we study a particularly important characteristic of African

economic and cultural life that is susceptible to the deleterious effects of climate change. It
is estimated that 22% of Africa’s population obtains the majority of its income from animal
husbandry and 43% of the continent’s land mass is used to support pastoral activities (FAO
2018). Many of Africa’s pastoral ethnic groups engage in the practice of transhumance, which is
the seasonal movement of grazing animals. Transhumance creates interdependent relationships
that are potentially sensitive to the increased frequency of droughts brought on by climate change
in Africa.

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

In typical years, neighbouring transhumant pastoral and sedentary agricultural groups coexist
in a symbiotic relationship that is characterized by seasonal migration. In the wet season, agri-
culturalists cultivate crops on more productive lands while transhumant pastoralists exploit more
marginal lands that produce sufficient plant biomass (or phytomass) for their livestock. After the
final harvest, transhumant pastoralists migrate along well-established corridors to arrive at the
agricultural farmlands for the dry season, where they benefit from the year-round availability of
phytomass while providing organic fertilizer in exchange.
In low precipitation years, there may be insufficient phytomass produced on the marginal
grazing lands to sustain pastoralists’ livestock. This shortage forces pastoralists to migrate to
agricultural farmlands before the dry season begins. If the animals arrive before the final har-
vest, tensions can arise due to damaged crops and competition for resources, such as water and
pasture. The issue is well known, with many documented examples (Moritz 2010; Kitchell et al.
2014; Brottem 2016).
Whether this mechanism results systematically in violent conflict is an empirical question.
On the one hand, neighbouring groups may avoid conflict if they believe droughts to be suf-
ficiently rare events. In this case, the symbiotic relationship is worth preserving. On the other
hand, groups may have updated their expectations about the frequency of droughts due to cli-
mate change. In this case, the symbiotic relationship is unsustainable, and frictions may emerge
in the form of conflict events.
A related question concerns the recent rise of extremist-religious violence in Africa. Given
that transhumant pastoral groups tend to be Muslim and sedentary agriculturalists tend to be
Christian, it is possible that this mechanism also affects violence involving self-styled religious
groups.
Our study examines these empirical questions. We measure the incidence of conflict using
geocoded conflict data from the Uppsala Conflict Data Program (UCDP) and the Armed Conflict
Location & Event Data Project (ACLED). We construct ethnicity-level measures of transhumant
pastoralism by combining information, taken from the Ethnographic Atlas, on the historical
importance of animal herding with information on historical mobility.
We begin with a descriptive account of the extent to which violence is more prevalent in land
outside of the territory of groups that are transhumant pastoral. We examine variation across
0.5-degree grid cells. For each cell, we identify a nearest neighbour, which is the ethnic group,
among all ethnic groups that border a cell’s own ethnic group, that is geographically closest to
the cell. We find that grid cells that have a transhumant pastoral nearest neighbour experience
more conflict. When we distinguish between types of conflict, we find that the effect is present
for conflicts that involve state actors, such as the police or military, as well as for smaller-scale
conflicts that only involve non-state actors. The relationship with civil conflicts is consistent with
accounts of agricultural landowners being aided by state forces against transhumant pastoral
ethnic groups, who are coded as non-state combatants.
We then turn to the central question of the analysis, which is whether reduced rainfall in the
territories of transhumant pastoralists leads to conflict in nearby agricultural lands. Examining
variation across grid cells and years, we estimate a specification that includes grid-cell fixed
406 REVIEW OF ECONOMIC STUDIES

effects and country-year fixed effects. The variable of interest is an interaction between the
measure of transhumant pastoralism of a grid cell’s nearest neighbour and the average amount
of rain in the nearest neighbour’s territory in a year. The coefficient estimate tells us whether the
incidence of conflict in a cell is influenced by rainfall in the nearest neighbouring ethnic territory
when the nearest neighbour is transhumant pastoral.
We find that, consistent with the hypothesis, less precipitation in a cell’s nearest neighbouring
ethnic group increases conflict in the cell, but only if the neighbour is transhumant pastoral. The
estimated effects are sizable and significant. We find that a one standard deviation decline in

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

rainfall experienced by the median transhumant pastoral ethnic group raises the risk of conflict in
a nearby grid cell by around 24%, or 0.8 percentage points (from a mean of 3.5% to 4.3%). The
same shock experienced by a non-transhumant pastoral group has a negligible and statistically
insignificant effect (around 2%, or 0.08 percentage points).
The specifications that we estimate also allow for direct effects of rainfall experienced in the
grid cell itself and for any intra-ethnic effects of rainfall occurring in the same ethnic territory
as the grid cell. They also allow for the possibility that these effects might differ if the cell’s
own ethnic group is transhumant pastoral. We find that these estimated effects are small and
statistically insignificant. Thus, while we estimate sizable adverse spillover effects from reduced
rainfall experienced by neighbouring transhumant pastoral groups, we find no evidence of effects
due to reduced rainfall in a cell or in the cell’s own ethnic group.
The estimates are consistent with the hypothesis that low rainfall induces transhumant
pastoralists to migrate early—that is, before the end of the growing season—to agricultural farm-
lands, resulting in damaged crops, competition for resources, and conflict. This interpretation
has a number of falsifiable predictions that we take to the data. First, we check that the estimated
effects are due to nearest neighbours being transhumant pastoral; namely, the combination of
being both mobile and pastoral. We show that there are no significant effects arising from nearest
neighbours who are either mobile but not pastoral or pastoral but not mobile. Second, we check
that conflicts arising due to adverse rainfall in neighbouring transhumant pastoral territories tend
to be located on agricultural land. Third, we obtain very similar estimates when we examine the
spillover effects due to phytomass growth rather than rainfall, suggesting that the effects are due
to the reduced availability of plant matter for animal grazing. Fourth, we check that there is no
spillover effect when we substitute precipitation with temperature. This is informative, since in
the tropical and subtropical climates of the African continent, rainfall is more important for plant
growth than temperature. Fifth, we examine the timing of the spillover effects within the year
and find that they are concentrated during the growing season, when crops are being cultivated,
but not during the dry season, when the land is left fallow. This is consistent with our hypothesis
that conflict occurs when pastoral groups are forced to use farmland before harvest.
We then turn to additional questions of interest, starting with whether our findings are able to
explain part of the rise in religious conflict involving jihadist groups in Africa in the past decades.
Since transhumant pastoral groups tend to be Muslim and sedentary agricultural groups are gen-
erally Christian, conflicts between the two groups may be viewed as—or evolve into—religious
violence. To investigate this, we separate conflict events into ones that involve jihadist actors
and ones that do not. We find that our mechanism affects the incidence of both jihadist and non-
jihadist conflict similarly. However, since jihadist conflicts were very rare prior to 2000, these
similar marginal effects imply a much larger rate of growth for jihadist conflicts in the past two
decades. Importantly, we also control for the religious composition of the nearest neighbour and
find that transhumant pastoralism is considerably more important than religion in predicting the
incidence of jihadist conflicts due to adverse rainfall in a cell’s nearest neighbour.
We next consider the important question of what can be done to mitigate the effects that we
find. We first examine the role of international aid projects, focusing specifically on projects
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 407

aimed at curbing the effects of environmental degradation, such as irrigation, forestry, conser-
vation, land improvement, and other agricultural projects. To test for the effects of such aid
projects, we allow our main estimated effect to vary by the cumulative presence of foreign aid
projects in a country and year starting in 1947. We find suggestive evidence that our documented
effects are independent of these aid projects.
We also consider the effects of state-protected conservation areas, which aim to prevent envi-
ronmental degradation and promote decarbonization. While these conservation projects may
attenuate the effects of climate change, some have argued that they can exacerbate transhumant

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

pastoral conflict by limiting the movements of herds and contributing to the scarcity of grazing
pastures. To test this, we allow our main estimated effect to vary by the share of land in a country
that is designated as conservation land at each point in time. We find that our estimated effects
are greater in magnitude when countries have more land that has protected conservation sta-
tus. This result suggests that conservation areas, while potentially beneficial in other ways, may
exacerbate conflict stemming from adverse rainfall shocks in transhumant pastoral territories.
The last factor we consider is political power. In the absence of political power-sharing,
minority groups may have strong incentives to fight (Mueller and Rohner 2018). Greater repre-
sentation of pastoral groups in national government may therefore mitigate conflict arising from
low rainfall in pastoral territories. We test this using the Ethnic Power Relations dataset. We
calculate, for each year and country, the power held by transhumant pastoral groups in national
politics, and we allow our estimated effects to vary depending on this measure. We find that the
estimated effects approach zero as transhumant pastoral groups gain a higher share of national
power. This suggests that when both sides have fair representation in government, a peaceful
resolution between pastoral groups and farmers is more likely.
Our analysis uncovers how relations between transhumant pastoral and sedentary agricul-
tural groups are undermined by episodes of low rainfall, which are becoming more frequent in
Africa due to climate change. The mechanisms that underlie the analysis are informed by the rich
ethnographic literature on the nature of transhumance and its implications for seasonal interac-
tions between sedentary farmers and herders in Africa (Lewis 1961; Jacobs 1965; Konczacki
1978; Dyson-Hudson and Dyson-Hudson 1980). Our findings add to this descriptive literature
and to more recent studies that document how adverse climate shocks have affected African
pastoral groups (Little et al. 2001; McPeak and Barrett 2001; Maystadt and Ecker 2004; Bol-
lig 2006) and in particular how they affect relations between pastoral and agricultural groups
(Benjaminsen et al. 2012).
Our focus on transhumant pastoralism is complementary to studies that focus on either one
of the two dimensions of this practice—either seasonal migration or pastoralism—and their
connection to conflict and economic development. Various studies have shown the importance
of seasonal migration for helping to alleviate poverty (Bryan et al. 2014; Morten 2019). Others
have examined the long-term consequences of animal husbandry on cultural traits associated
with gender (Becker 2019) and the importance placed on maintaining one’s honor (Grosjean
2014; Cao et al. 2021). A number of studies have examined the long-term consequences that a
noteworthy nomadic pastoral group, the Mongols, had on state development in China due to the
threat of invasion, which was, in part, due to climate shocks (Bai and Kung 2011; Ko et al. 2018).
Our findings contribute to a better understanding of the salience of cross-ethnicity divisions
and the conditions under which ethnic differences can lead to conflict. In particular, our findings
provide insight into the recent finding in Depetris-Chauvin and Özak (2020) that conflict tends
to occur near ethnic boundaries. Our findings suggest that one important mechanism underlying
the relationship could be the disruption of the traditional symbiotic relationship between pas-
toralists and sedentary farmers. Eberle et al. (2020) also show that conflict at the boundaries
between nomadic and non-nomadic groups is greater when temperatures are higher, consistent
408 REVIEW OF ECONOMIC STUDIES

with existing studies showing that heat can increase violence through a variety of mechanisms,
including psychological channels (Hsiang et al. 2013; Hsiang and Burke 2014; Baysan et al.
2019). Our analysis indicates that the direct “heat and hate” thermal stress effect on conflict doc-
umented in Eberle et al. (2020) is distinct from the inter-ethnic spillover effect of rainfall and
phytomass documented here, which is due to the disrupted seasonal migration of transhumant
groups.
We contribute directly to the literature on climate and conflict by providing new evidence that
documents a precise mechanism through which climate change affects inter-group violence (see

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

Miguel et al. 2004; Solow 2013; Hsiang and Burke 2014; Burke et al. 2015). We also contribute
to the wider literature on the determinants of conflict within Africa, including studies that explore
the importance of historical factors (e.g. Besley and Reynal-Querol 2014; Depetris-Chauvin
2015; Michalopoulos and Papaioannou 2016; Moscona et al. 2020); ethnic or social factors
(Montalvo and Reynal-Querol 2005; Esteban et al. 2012; Rohner et al. 2013; Arbatli et al.
2020); and economic factors, especially shocks to the opportunity cost of conflict, which can be
challenging to distinguish empirically from shocks that affect other drivers of conflict (Dube and
Vargas 2013; McGuirk and Burke 2020).
One important aspect of our study is the spillover nature of the effect we identify—rainfall in
one location (transhumant pastoral territories) affects conflict in another (sedentary agricultural
territories). Our approach can be interpreted as recovering the exact structure of one mechanism
behind the spatial spillovers observed in the existing climate-conflict literature (e.g. Guariso
and Rogall 2017; Harari and Ferrara 2018). While prior studies take a more empirical approach
towards characterizing the nature of spillovers, our analysis starts with a specific mechanism
that is motivated by the ethnographic literature. We then build our estimator to capture this
mechanism while accounting for other, more general forms of spillover. In this way, our strategy
is similar to other studies that also specify a particular spillover mechanism ex-ante that is then
brought to the data (e.g. König et al. 2017).
The paper is organized as follows. In Section 2, we provide a description of the traditional
symbiotic relationship between transhumant pastoralists and sedentary farmers in Africa. We
also discuss recent changes in climate on the continent and how this has affected the nature of
the herder–farmer relationship. In Section 3, we describe the data used in the main analysis.
In Section 4, we examine the cross-sectional relationship between transhumant pastoralism and
conflict in neighbouring areas. In Section 5, we estimate the effect of lower rainfall in transhu-
mant pastoral territory on conflict in neighbouring locations. In Section 6, we present a series of
auxiliary tests of causal mechanisms. In Section 7, we turn to the implications of our findings,
including an examination of extremist-religious conflict and factors that may mitigate the effects
that we estimate. Section 8 concludes.

2. BACKGROUND AND CONTEXT

2.1. Transhumant pastoralism
A defining feature of transhumant pastoralism is that it results in regular seasonal interac-
tions with sedentary agricultural groups. Neighbouring herders and farmers develop a symbiotic
relationship that allows both groups to share resources in an efficient and mutually beneficial
manner.
In most of Africa, seasons are determined primarily by precipitation rather than temperature,
a fact highlighted by the typical description of the seasons as either the wet (i.e. growing) or
dry (i.e. fallow) season. The time of year when seasons occur depends on where one is on
the continent, particularly whether one is north or south of the equator. The seasonal variation
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 409

a b

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

F IGURE 1
Rainfall and seasonal migration in Africa. (a) Rainfall and migration during the dry season and (b) Rainfall and
migration during the wet season

is shown in Figure 1, which reports rainfall across the continent in two months: January and
August. January, shown in Figure 1b, is a dry-season month for most of the continent that lies
north of the equator and a wet-season month for areas south of the equator. By contrast, in
August, which is shown in Figure 1b, the north experiences a wet season and the south a dry
season.
The figure also provides a stylized illustration of transhumant migrations that occur in West
Africa. Hypothetical sedentary agricultural groups are shown in blue and transhumant pastoral
groups in red. During the wet season, when crops are cultivated, pastoralists keep their livestock
on marginal grazing land that is not suitable for agriculture but supports the growth of wild
grasses that provide sustenance to animals. During the dry season, this growth no longer occurs.
As a result, herds are moved to the more fertile farmlands that are used for agriculture during the
wet season but are left fallow during the dry season. This movement is shown by the arrows in
Figure 1a. Animal herds are allowed to graze on the farmland during this period. This arrange-
ment benefits both the pastoralists, who enjoy the dry-season production of animal feed, and the
farmers, whose land is improved by the animals’ manure, a form of nitrogen-rich organic fertil-
izer. At the end of the dry season, herds are moved from the agricultural lands and return to the
more marginal grazing lands. This is shown by the arrows in Figure 1b.
Due to the seasonal movements of herds, both sedentary farmers and transhumant pastoral-
ists are able to exploit the land efficiently and cooperatively. Stenning (1959, p. 6), in his study
of the pastoral Fulani, describes the symbiotic relationship between them and their agricultural
neighbours, the Uda’en: “Pastoral life is pursued not in isolation, but in some degree of symbio-
sis with sedentary agricultural communities. . . there have existed, possibly for many centuries,
arrangements for pasturing cattle on land returning to fallow, and for guaranteeing cattle tracks
and the use of water supplies. Pastoral Fulani did not, and do not, merely graze at will, but
obtained rights to the facilities they required from the acknowledged owners of the land.”
410 REVIEW OF ECONOMIC STUDIES
4

Precipitation Anomaly, cm/month

2

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

-2

-4
1900 1920 1940 1960 1980 2000 2020
Year

F IGURE 2
Climate change and annual historical precipitation in the Sahel
Source: Authors creation using the Global Precipitation Climatology Centre’s Sahel Precipitation Index. Anomalies reported are in
average cm/month for June through October over 20–10◦ N, 20 W–10◦ E, annually from 1900–2017. doi: 10.6069/H5MW2F2Q.

As a consequence of these traditional relationships, extensive transhumance is common in the

parts of Africa with ecological zones that have these features, the largest region being the Sahel.
Except in a few cases, with very small samples, information about the exact routes remains
undocumented. From these studies, which are summarized in Appendix Table A1, one can see
many aspects that are relevant for our analysis. The distance between the origin and destination
varies considerably, ranging from tens of kilometres to hundreds of kilometres. Although the
routes are meandering, they often have a north–south orientation, although many routes follow
an east–west orientation, especially near the west-facing coastal areas. Routes commonly cross
ethnic boundaries and sometimes national boundaries.

2.2. Effects of climate change

For much of the African continent, particularly the Sahel region, the most salient consequence
of climate change has been rainfall that is persistently below the long-run average. Increasing
temperatures, particularly outside of Africa, tend to reduce precipitation on the continent. For
example, increased Atlantic sea-surface warming causes lower rainfall and more droughts in the
Western part of the continent (Shanahan et al. 2009), while warming in the Middle East, South
Asia, and particularly the Indian Ocean affects precipitation in Eastern Africa (Cook and Vizy
2013).
The recent effects of global warming on precipitation within the continent can be seen in
Figure 2, which shows annual wet-season rainfall from 1901 to 2017 for the Sahel, a region
that is particularly relevant for our analysis. It is clear that since the late 1960s, there has been
a reduction in annual precipitation (Nicholson et al. 2018). Between 1970 and 2017, annual
average rainfall was below the long-run (1900–2017) mean in 36 of the 47 years (Schneider
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 411

et al. 2020). Although there is some slight attenuation in recent years, it is clear that global
warming is associated with reduced rainfall (Herrmann and Mohr 2012; Biasutti 2018).
During this same time, the region has seen an increase in the frequency and severity of con-
flicts between sedentary agriculturalists and transhumant pastoralists. According to numerous
accounts, the new climate regime has led to more variation in the timing and location of transhu-
mance movements, causing migrations that are earlier in the season and deeper into agricultural
lands (Ayantunde et al. 2014). A plausible explanation for the concurrent trends is the reduc-
tion in living organic plant matter, known as phytomass, which provides sustenance for grazing

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

herds. Rainfall is the primary determinant of living organic plant matter on the continent (Hein
2006). While temperature is also a factor, its importance for plant growth is primarily due to
its indirect effect on rainfall (Biasutti 2018). This contrasts with the situation in more temperate
regions outside of Africa, such as North America or Europe, where temperature is more impor-
tant for plant growth than precipitation (Moles et al. 2014). While temperature is the primary
constraint for plant growth in temperate climates, rainfall is the primary constraint in tropical
climates. Our own calculations, which we describe in detail below, show that within our sam-
ple, annual variation in rainfall explains about six times more of the variation in phytomass than
temperature. Given the importance of rainfall for phytomass growth on the African continent,
our analysis focuses on the consequences of rainfall scarcity.

3. DATA
3.1. Description, sources, and validation
Our analysis examines the relationship between conflict, rainfall (or phytomass), and transhu-
mant pastoralism. Below, we describe the data and the measurement of each variable.1
Conflict. Our baseline set of conflict variables are from two sources of geocoded data: the
Uppsala Conflict Data Program (UCDP) (Sundberg and Melander 2013), which covers 1989–
2018, and the Armed Conflict Location & Event Data project (ACLED) (Raleigh et al. 2010),
which covers 1997–2019. We use both sources throughout our analysis since they each have
strengths and weaknesses. While the UCDP data have longer temporal coverage, the ACLED
data have broader coverage of smaller-scale conflicts.
To be included in the UCDP database, a conflict event must have at least one fatality and
the pair of actors involved in the event (i.e. the conflict dyad) must have produced at least 25
fatalities in at least one calendar year throughout the series. Additionally, at least one of the
actors involved in the event must be an “organized actor,” such as the state or a politically
organized rebel group or militia. The ACLED data have weaker criteria for inclusion. There is
no requirement for a certain number of fatalities in a calendar year or a conflict event. Thus, the
ACLED data are better equipped to capture small-scale, localized conflict events.
Using the reported locations of conflict events, we create measures of the presence of conflict
in 0.5-degree (approx. 55 km × 55 km) grid cells during a calendar year. Our primary measures
are indicator variables that equal one if each of the following types of conflict occurs: All con-
flicts; State conflicts, where the state is involved as a participant in the event; and Non-state
conflicts, where only non-state actors are involved.2

1. For more detail on data sources, see Appendix A.

2. When constructing the ACLED measures, we focus on “battles” and “violence against civilians,” which are
analogous to the two- and one-sided events that comprise the UCDP data.
412 REVIEW OF ECONOMIC STUDIES

Summary statistics for the conflict measures are reported in Appendix Table A3. The uncon-
ditional probability of ACLED conflict incidence is much higher than that of UCDP incidence.
As expected, the difference is largest for non-state conflicts: 8% for ACLED versus 2% for
UCDP. Thus, we place particular importance on the ACLED data in our analysis of non-state
conflicts.
Transhumant Pastoralism. To identify transhumant pastoral societies, we use information
from the Ethnographic Atlas (EA), a database of 1,265 ethnic groups assembled in Murdock
(1967). We construct a composite index that captures the two key aspects of transhumant

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

pastoralism.
The first is that the group moves seasonally; namely, that they are mobile. There is exten-
sive information in the EA on the mobility of ethnic groups traditionally. Variable v30 of the
database codes groups as falling within one of the following categories that describe the nature
of settlement: (1) Nomadic or fully migratory; (2) Seminomadic; (3) Semisedentary; (4) Com-
pact but impermanent settlements; (5) Neighbourhoods of dispersed family homes; (6) Separated
hamlets; (7) Compact and relatively permanent; and (8) Complex settlements.
Although transhumance is not measured explicitly, nearly all forms of movement today are
seasonal. Nomadic activity that is not seasonal is now rare. Thus, we take being traditionally
nomadic as a proxy for being seasonally mobile. We create two indicator variables that allow
for two definitions of transhumance: a “narrow” definition that includes only groups that are
“nomadic or fully migratory” or “seminomadic” and a “broad” definition that also includes
groups that are “semisedentary” or have “compact but impermanent settlements.” The variants
differ in whether groups that are semi-mobile are coded as being transhumant or not. We denote
the variable T ranshumant e , where e indexes ethnic groups in our sample.
The second key aspect of transhumant pastoralism is the herding of animals. To capture this
dimension, we build on a measure developed in Becker (2019), which combines information
on the fraction of subsistence that is from animal husbandry (measured on a 0–1 scale, from
variable v4 in the EA) with an indicator variable that equals one if the primary large animal
is suitable for herding (from variable v40). Animals that require herding include sheep, goats,
equine animals, camels, and bovine animals, but not pigs, for example. Becker’s measure is
constructed as the interaction between these two measures. It ranges from 0 to 1 and it proxies for
the fraction of an ethnic group’s subsistence that is from herded animals. We denote this variable
Pastoral e .
We construct a measure of “transhumant pastoralism” by interacting the two
components: T ranshumant e × Pastoral e . The resulting variable, which we denote
T ranshumant Pastoral e , measures the fraction of a group’s subsistence that is from transhu-
mant pastoralism.
Since our measure is based on traditional practices from ethnographic sources rather than
current practices from contemporary surveys, it is predetermined and unaffected by the episodes
of conflict that we explain empirically. Reassuringly, the measure is still predictive of con-
temporary pastoralism. This can be seen in Appendix Figure A1, which shows the positive
relationship between the contemporary ownership of animals, measured in DHS surveys, and
the transhumant pastoralism of the ethnic group of the respondent.
To assign the variable to spatial units, we match each society from the EA to the ethnic
territories mapped by Murdock (1959). Using a variety of sources, documented in Kincaide et al.
(2020), we match around 96% of the ethnic territories in the map to an ethnic group in the EA.
Figure 3 shows the spatial distribution of the transhumant pastoralism measure. The intensity
of transhumant pastoralism is consistent with expectations based on the location of land most
suitable for animal grazing rather than agriculture. This can be seen in Appendix Figure A2,
which shows the spatial distribution of land suitable for transhumant pastoralism and sedentary
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 413
a b

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

F IGURE 3
Cross-ethnicity spatial variation in transhumant pastoralism. (a) Narrow definition and (b) Broad definition

agriculture, taken from Beck and Sieber (2010), and the boundaries of ethnic groups with some
form of traditional mobility. It is clear that the ecological environment is a key determinant of
transhumant pastoralism.
Rainfall and Phytomass. Pastoral groups rely on rain to produce the phytomass needed to
sustain their livestock. Our rainfall variable measures average monthly precipitation during a
calendar year in a 0.5 degree cell. The data are from the Global Precipitation Climatology Centre
and are based on interpolated land-surface precipitation data from approximately 85,000 rain
gauges across the globe (Schneider et al. 2020). The variable, which covers the full duration of
our conflict series (1989–2019), is measured in centimetres per month.
We verify the importance of rainfall for plant growth using satellite data on dry matter veg-
etation (i.e. phytomass). The phytomass data are derived from satellite images provided by the
Copernicus Global Land Service and is available at the 1 km pixel level weekly from 1999 to
2019 (Copernicus n.d.). We aggregate the data to the 0.5 degree cell–year level and measure the
final variable in average kilograms of plant growth per hectare per day.
We estimate the determinants of phytomass growth at the cell–year level, modelling phy-
tomass as a function of average annual precipitation and temperature, while conditioning on cell
fixed effects and country-by-year fixed effects. The estimates, reported in Appendix Table A2,
confirm the importance of precipitation for vegetation growth. Consistent with the environmen-
tal science literature (e.g. Waha et al. 2013; D’Onofrio et al. 2019), we find that rainfall is a
significant determinant of phytomass growth and is a more important factor than temperature.
After accounting for the fixed effects, rainfall explains 3.6% of the residual variation while
temperature explains 0.6%.
Given that rainfall is the main driver of phytomass growth on the African continent, we
use this as our primary climate variable. We use rainfall rather than phytomass as our baseline
measure since it is available for a much longer time series. In sensitivity checks, we show that
the estimates are nearly identical when we use either phytomass or phytomass predicted by
rainfall.
414 REVIEW OF ECONOMIC STUDIES

3.2. Summary of the data

The descriptive statistics for the variables used in the analysis are reported in Appendix Table A3.
In separate panels, we report variables that vary at the cell–year, cell, ethnic-group-year, and
ethnic group levels. At the cell–year level, the incidence of any conflict is 3% when using the
UCDP data and 8% when using the ACLED data. The average precipitation is 5.65 centimetres
per month and the average temperature is 24.5 degrees Celsius. The average of the ethnicity-
level measure of transhumant pastoralism is 0.08 when the narrow measure is used and 0.09
when the broad measure is used.

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

In Appendix Table A4, we report summary statistics for groups that are transhumant pastoral
(column 1), groups that are not (column 2) and their difference (column 3). We find that tran-
shumant pastoralism is associated with less conflict, less precipitation, less phytomass, higher
temperatures, land that is less suitable for agriculture, and land that is more suitable for tran-
shumant pastoralism. It is also associated with lower population, fewer nighttime lights, less
national political power, a higher share of Muslim people, and a lower share of Christian peo-
ple today. Looking at historical ethnographic traits, we see that transhumant pastoral groups,
not surprisingly, practiced less agriculture and were more developed politically (as measured by
levels of political authority beyond the local community).
These comparisons make clear that transhumant pastoralism is not randomly allocated across
the continent. The practice is determined, in part, by ecological conditions and is associated with
various other factors. This highlights the importance of our auxiliary analyses which look for
evidence of our specific mechanism of interest, test for the importance of other ethnicity-level
traits, and examine the importance of contemporary political power.

4. CROSS-SECTIONAL PATTERNS
We begin our analysis by estimating the relationship between being near transhumant pastoral
groups and conflict across 0.5-degree grid cells. The sample comprises 9,691 cells nested in 780
ethnic territories across the African continent. These are shown in Figure 4 for a region in Mali
that is traditionally inhabited by the Masina, Dogon, Zenega, Songhai, and others. The map also
shows the location of UCDP conflicts from 1989 to 2018.
For each cell, we identify the neighbouring ethnic group that is most relevant for that cell.
As illustrated by Figure 4, cells within an ethnic territory can have different neighbours that are
relevant. For example, consider cells located within the Masina ethnic territory. The relevant
neighbouring ethnic group varies depending on where a given cell is located in the territory.
For the cells in the northwestern portion of the Masina territory, the relevant neighbour is the
Zenega; for those in the eastern portion, the relevant neighbour is Udalan; and for cells in the
southeastern portion, the relevant neighbour is the Dogon, Mossi, or Deforo. This generates rich
variation in nearest neighbour characteristics even when holding constant the characteristics of
one’s own ethnic group. We identify each cell’s “nearest neighbour” (or “neighbour” for short)
as the ethnic group that is geographically closest to a cell’s centroid among all ethnic groups that
are contiguous to the ethnic group in which the cell is located.3

3. For cells that lie within multiple ethnic territories, we determine the ethnic group of a cell by the location of
its centroid.
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 415

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

F IGURE 4
Structure of data and analysis. The figure shows 0.5-degree cells, along with the boundaries of the ethnic groups, their
names, and their measure of transhumant pastoralism (THP) using the narrow definition of transhumant

With this data structure, we then estimate the following equation:

N eighbour
yiet = γ1 T ranshumant Pastoral i + γ2 T ranshumant Pastoral eOwnGr oup
+ γ3 ln( popi ) + αt + ηiet , (1)

where i indexes 0.5-degree grid cells, e ethnic groups, and t years (1989–2018 or 1997–2019).
The dependent variable, yiet , is conflict incidence in cell i, which lies within the territory
N eighbour
of ethnicity e, and in year t. The variable T ranshumant Pastoral i is the measure
of transhumant pastoralism of the nearest neighbouring ethnic group to cell i. The variable
T ranshumant Pastoral eOwnGr oup measures the transhumant pastoralism of the ethnic group in
which the cell is located. Lastly, ln( popi ) is the natural log of the population of cell i, measured
in 1990. The parameter of interest is γ1 , which represents the effect of the nearest neighbour-
ing ethnic group’s transhumant pastoralism on conflict in a cell. Standard errors are adjusted for
two-way clustering at the level of a cell and a climate zone-year.
Estimates of equation (1) are reported in Table 1. Panel A reports estimates using the nar-
row definition of transhumance, while panel B reports estimates using the broad measure. Each
column reports estimates using a different measure of conflict as the dependent variable: total
conflicts, state-involved conflicts, and non-state conflicts, each measured using either the UCDP
(columns 1–3) or ACLED (columns 4–6) data.
In all specifications, we find that having a nearest neighbour that is transhumant pastoral
is associated with significantly more conflict. While this relationship is present for all conflict
measures, it is much smaller for non-state conflicts measured using the UCDP data. This is
 416

TABLE 1
Cross-sectional evidence of conflict spillover from nearest neighbouring THP territory: cell level
Indicator for the presence of conflict

UCDP ACLED

(1) (2) (3) (4) (5) (6)

I(Any) I(State) I(Non-state) I(Any) I(State) I(Non-state)
Panel A: Transhumant definition includes only groups that are migratory or nomadic (narrow definition)
Neighbour 0.0273*** 0.0253*** 0.0057** 0.0702*** 0.0534*** 0.0698***
Transhumant (0.0054) (0.0049) (0.0025) (0.0095) (0.0077) (0.0095)
Pastoral [γ1 ]
Transhumant Pastoral 0.0081 0.0063 0.0017 0.0208** 0.0134* 0.0200**
[γ2 ] (0.0057) (0.0046) (0.0028) (0.0097) (0.0077) (0.0097)
Panel B: Transhumant definition includes all groups without fully permanent settlements (broad definition)
Neighbour 0.0301*** 0.0288*** 0.0050** 0.0671*** 0.0534*** 0.0667***
Transhumant (0.0053) (0.0049) (0.0023) (0.0089) (0.0074) (0.0089)
Pastoral [γ1 ]
Transhumant Pastoral 0.0076 0.0058 0.0012 0.0188** 0.0126* 0.0181*
[γ2 ] (0.0054) (0.0044) (0.0026) (0.0093) (0.0074) (0.0093)
Dep. Var. Mean 0.035 0.025 0.016 0.085 0.055 0.085
Year FE and ln Yes Yes Yes Yes Yes Yes
Population
REVIEW OF ECONOMIC STUDIES

Climate-Zone-Year Clusters 420 420 420 336 336 336

Cell Clusters 7,690 7,690 7,690 7,690 7,690 7,690
Observations 230,700 230,700 230,700 184,560 184,560 184,560
Notes: All outcome variables measure conflict incidence at the level of a cell–year. “I(Any)” is an indicator variable that equals one if at least one violent conflict occurs in a cell and
year. “I(State)” is an indicator variable that equals one if at least one conflict event involving the state occurs in a cell and year; “I(Non-state)” is an indicator variable that equals one if
at least one conflict event not involving the state occurs in a cell and year. All specifications include a control for the natural log of the population of a grid-cell in 1990. Standard errors,
which are reported in parentheses, are adjusted for clustering at the level of a grid-cell and a climate zone-year. ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 417

not surprising given that the UCDP data have more restrictive inclusion criteria that lower its
coverage of smaller-scale conflicts not involving the state.

5. RAINFALL SCARCITY AND AGRO-PASTORAL CONFLICT

We now turn to our baseline equation which estimates whether adverse rainfall in transhumant
pastoral territories results in conflict in neighbouring lands.
Estimating Equation. We estimate a variant of equation (1) that traces the effect of rainfall in

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

a neighbouring transhumant pastoral territory on conflict in a cell. The equation is given by:
N eighbour N eighbour N eighbour
yiet = γ0s Rain it + γ1s Rain it × T ranshumant Pastoral i
OwnGr oup OwnGr oup
+ γ2s Rain et + γ3s Rain et × T ranshumant Pastoral eOwnGr oup
+ γ4s Rain itOwnCell + γ5s Rain itOwnCell × T ranshumant Pastoral eOwnGr oup

+ X iet  + αis + αc(i)t
s
+ ηiet
s
, (2)

where yiet is an indicator for the incidence of conflict in cell i, located in ethnic territory e,
N eighbour
and in year t; Rain it measures average precipitation in the nearest neighbouring eth-
N eighbour
nic group to cell i in year t; T ranshumant Pastoral i is the transhumant pastoralism
OwnGr oup
measure for the neighbouring ethnic group; Rain et measures precipitation in group e in
year t; T ranshumant Pastoral eOwnGr oup is the transhumant pastoralism measure for ethnicity

e; and Rain itOwnCell measures precipitation in cell i in year t. The vector X iet includes additional
covariates used in auxiliary robustness and sensitivity checks; αi denotes cell fixed effects, which
capture time-invariant differences across grid cells; and αc(i)t denotes country-year fixed effects,
which capture any variation across time that is common to all grid cells in a country. To account
for serial and spatial dependence, our standard errors are two-way clustered at both the cell and
climate zone-year levels.
The parameter γ1s represents the differential effect of rainfall in a neighbouring ethnic ter-
ritory on conflict in cell i when the neighbouring ethnicity is transhumant pastoral relative to
when it is not transhumant pastoral. A negative estimate of γ1s indicates that, consistent with our
hypothesis, dry weather in pastoral territories causes additional conflict in neighbouring cells.
It is important to note that this specification accounts flexibly for many factors that have been
previously studied in the conflict literature. The cell fixed effects αis capture all time-invariant
determinants of conflict, such as geography, national boundaries, historical factors, and ethnic
traits (e.g. Besley and Reynal-Querol 2014; Michalopoulos and Papaioannou 2016; Moscona
et al. 2020). The country-year fixed effects αc(i)ts
capture time-varying national-level factors such
as changes in country GDP, national political or legal institutions, country-level ethnic fraction-
alization and polarization, resource endowments, and international geo-political characteristics,
all of which have been prominent in the cross-country literature on conflict (e.g. Collier and
Hoeffler 1998, 2004; Fearon and Laitin 2003; Ross 2004; Esteban et al. 2012). The control for
rainfall in a cell, γ4s Rain itOwnCell , captures the direct effects of rainfall on the opportunity costs
or logistics of fighting (e.g. Miguel et al. 2004; Jia 2014; Burke et al. 2015; Harari and Ferrara
OwnGr oup
2018). The control for rainfall in the territory of a cell’s ethnic group, γ2s Rain et , captures
intra-ethnic spatial spillover effects, which are also potentially important determinants of con-
flict in a location (Harari and Ferrara 2018). Our specification also allows for differential effects
of the rainfall controls by the transhumant pastoralism of the cell or ethnic group.
Results. Estimates of equation (2) are reported in Table 2 for the narrow definition of tran-
shumant pastoralism and in Appendix Table A5 for the broad definition. Each column reports
estimates for one of our six conflict measures. The first set of coefficients, reported under the
418 REVIEW OF ECONOMIC STUDIES

heading “Nearest Neighbouring Ethnic Group,” are for the effect of variables that measure
rainfall experienced by the nearest neighbouring ethnic group, γ0s , and its interaction with the
neighbour’s transhumant pastoralism measure, γ1s .
We find that less rainfall in a cell’s nearest neighbouring ethnic group leads to more conflict
in the cell, but only if the neighbour is transhumant pastoral. While the estimated effects for non-
transhumant pastoral groups are never statistically different from zero, the differential effects for
transhumant pastoral neighbours are always negative and, in all columns but one, are statistically
significant. Consistent with prior findings, the estimates for non-state conflict using the high-

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

threshold UCDP data are much smaller in magnitude and imprecisely estimated.
To assess the magnitude of the estimates, in the second panel, we report the predicted effect
(expressed as a percentage of the dependent variable mean) of a one standard deviation reduction
in rainfall. According to the estimates, this adverse rainfall shock causes an increase in conflict
that is equal to 37.5% of the mean of total UCDP conflict (column 1); for the ACLED measure
of conflict, which has a higher mean, the equivalent figure is 13.6% (column 4).
If we take into account the deficiency of the UCDP non-state conflict measure, the evi-
dence suggests that rainfall in the territory of transhumant pastoral nearest neighbours affects
both state and non-state conflict. This implies that herder–farmer conflicts can involve state
agents such as police, conservation officers, or the military, or they can occur absent government
involvement.
The tables also report the coefficients for γ2s . . . γ5s , which are the estimated effects of rainfall
in the cell’s own ethnic group e and in cell i itself, and the differential effects of the rainfall
measures when the ethnic group is transhumant pastoral. These are reported under the headings
“Own Ethnic Group” and “Own Cell.” Each of the estimated coefficients is small in magnitude
and almost never statistically different from zero. Only one of 24 coefficients is significant,
and that is at the 10% level. Thus, while we find that less rainfall in the territory of the nearest
neighbouring transhumant pastoral groups leads to greater conflict, there is no evidence of effects
for own-cell or own-group rainfall.
Robustness and Sensitivity Checks. We now turn to the sensitivity of our estimates. We have
shown that the estimates using the narrow and broad definitions of transhumant pastoralism are
qualitatively identical. Thus, for the remainder of the paper, we use the narrower definition as
our baseline measure. In addition, we limit our focus to four baseline outcome variables. We
retain both measures of overall conflict, but use the UCDP measure of state conflict (because
of the longer time series) and the ACLED measure of non-state conflict (because of the better
coverage of smaller-scale conflicts due to the lower threshold for inclusion).
A potential concern is that transhumant pastoralists might also have other characteristics that
are important for mediating the relationship between adverse rainfall and nearby conflict. Given
this, we check the sensitivity of our findings to accounting for other potentially important charac-
teristics of neighbouring ethnic groups; namely, precolonial political centralization, the presence
of segmentary lineage organization, and a traditional belief in a religion with a moralizing high
god.4 We re-estimate a variant of equation (2) controlling for each additional characteristic of a
cell’s nearest neighbour interacted with the neighbour’s rainfall. The estimates, which we report
in Appendix Table A6, show that our findings remain robust to the inclusion of these additional
controls.

4. Precolonial political centralization, which is measured by the levels of jurisdictional hierarchies beyond the
local community, has been shown to be an important determinant of public goods provision and economic development
(Gennaioli and Rainer 2007; Michalopoulos and Papaioannou 2013), both of which are relevant for conflict. Segmentary
lineage organization has been shown to be associated with conflict (Moscona et al. 2020). The presence of a moralizing
high god is believed to be an important factor for cooperation, conflict, and long-term economic growth (Norenzayan
2013).
 TABLE 2
McGuirk & Nunn

Effect of rain shock in nearest neighbouring THP territory on conflict in a cell: narrow definition of transhumance
Indicator for the presence of conflict

UCDP ACLED

(1) (2) (3) (4) (5) (6)

I(Any) I(State) I(Non-state) I(Any) I(State) I(Non-state)
Nearest Neighbouring Ethnic Group
Rain [γ0s ] −0.0005 0.0001 −0.0005 −0.0007 0.0004 −0.0008
(0.0006) (0.0006) (0.0005) (0.0011) (0.0009) (0.0011)
Rain × Transhumant Pastoral −0.0110*** −0.0121*** −0.0012 −0.0096** −0.0092*** −0.0096**
[γ1s ] (0.0033) (0.0031) (0.0021) (0.0038) (0.0035) (0.0038)
Own Ethnic Group
Rain [γ2s ] 0.0001 0.0014 −0.0002 0.0007 0.0014 0.0005
(0.0010) (0.0009) (0.0007) (0.0013) (0.0010) (0.0013)
Rain × Transhumant Pastoral −0.0014 −0.0046 0.0017 −0.0011 −0.0079 0.0005
[γ3s ] (0.0047) (0.0048) (0.0038) (0.0065) (0.0062) (0.0065)
Own Cell
Rain [γ4s ] −0.0002 −0.0005 −0.0001 −0.0004 −0.0007 −0.0002
(0.0007) (0.0006) (0.0005) (0.0010) (0.0009) (0.0010)
Rain × Transhumant Pastoral 0.0041 0.0056* −0.0008 0.0046 0.0052 0.0032
[γ5s ] (0.0035) (0.0032) (0.0024) (0.0051) (0.0039) (0.0051)
(continued)
THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA
419

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

 420

TABLE 2
Continued.
Indicator for the presence of conflict

UCDP ACLED

(1) (2) (3) (4) (5) (6)

I(Any) I(State) I(Non-state) I(Any) I(State) I(Non-state)
Nearest Neighbouring Ethnic Group: Additional Calculations
Effect of 1 Std. Dev. Rain Shock as % of Dep. Var. Mean:
Rain −1.88 0.57 −3.51 −0.95 0.83 −1.13
p-value [0.40] [0.83] [0.36] [0.53] [0.67] [0.46]
Rain × Transhumant Pastoral −37.51 −57.26 −8.68 −13.60 −20.12 −13.64
p-value [0.00] [0.00] [0.58] [0.01] [0.01] [0.01]
Rain + Rain × Transhumant Pastoral −39.39 −56.68 −12.19 −14.55 −19.29 −14.76
p-value [0.00] [0.00] [0.43] [0.01] [0.01] [0.00]
Dep. Var. Mean 0.035 0.025 0.016 0.085 0.055 0.084
Cell FE Yes Yes Yes Yes Yes Yes
Country × Year FE Yes Yes Yes Yes Yes Yes
Climate-Zone-Year Clusters 420 420 420 322 322 322
Cell Clusters 7,722 7,722 7,722 7,722 7,722 7,722
Observations 231,660 231,660 231,660 177,606 177,606 177,606
REVIEW OF ECONOMIC STUDIES

Notes: The unit of observation is a 0.5-degree grid-cell and year. “I(Any)” is an indicator variable that equals one if at least one violent conflict occurs in a cell and year. “I(State)”
is an indicator variable that equals one if at least one conflict event involving the state occurs in a cell and year; “I(Non-state)” is an indicator variable that equals one if at least one
conflict event not involving the state occurs in a cell and year. Nearest Neighbouring Ethnic Group refers to the nearest neighbouring ethnic territory to cell i. Own Ethnic Group
refers to the ethnic territory that contains cell i. Standard errors, which are reported in parentheses, are adjusted for clustering at the level of a grid-cell and a climate zone-year.
∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 421

Another potential concern is that transhumant pastoral groups tend to live in locations where
rainfall is more scarce. Thus, our findings might be biased by the differential spillover effects for
nearest neighbours that experience less rainfall in general. We check for this by estimating our
baseline equation while controlling for the rainfall of the nearest neighbour (normalized to lie
between 0 and 1) interacted with the group’s average rainfall during the period of our analysis,
1989–2019. As shown in Appendix Table A7, the estimates of interest are nearly identical with
the inclusion of this control.
It is possible that our measure of rainfall is correlated with other time-varying macro-level

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

factors that differentially affect the presence of conflict adjacent to transhumant pastoral groups.
Rainfall could be capturing the effects of other factors that are also trending over time, such as
the availability of firearms, population density, better communication technologies, and so forth
(Acemoglu et al. 2020; Manacorda and Tesei 2020). To account for this, we include a control
for a linear time trend interacted with each cell’s nearest neighbour’s measure of transhumant
pastoralism. While this control captures factors trending linearly over time, other factors exhibit-
ing more irregular movements may also be important for conflict, such as commodity prices
(Berman et al. 2017; McGuirk and Burke 2020). To account for this, we also interact the mea-
sure of a cell’s nearest neighbour’s transhumant pastoralism with numerous aggregate price
indices that may affect conflict differently across space. These include price indices for energy
(coal, crude oil, and natural gas), metals and minerals (aluminium, copper, iron ore, lead, nickel,
steel, tin, and zinc), and precious metals (gold, platinum, and silver), as well as a price index for
agricultural products (oils and meals, grains, and other food such as bananas, meat, and sugar).5
The estimates with these additional covariates, reported in Appendix Table A8, are similar in
magnitude and remain highly significant.
Our final check examines the robustness of our conclusions to various methods of calculating
standard errors, including clustering by country; clustering by country and climate zone; and
allowing for spatial correlation within 1,000 km of a cell. As we report in Appendix Table A9,
the precision of our estimates is similar in each case.

6. TESTING FOR SPECIFIC MECHANISMS

Our findings are consistent with adverse rainfall shocks inducing transhumant pastoral groups
to migrate to nearby agricultural lands before the harvest, resulting in conflict with sedentary
farmers. This explanation yields a number of additional testable predictions that we now take
to the data. These are: (1) the effects are due to the combination of mobility and pastoralism
(i.e. transhumant pastoralism) rather than either mobility or pastoralism alone; (2) transhumant
pastoral rainfall should primarily affect conflict on agricultural lands; (3) since rainfall matters
because it affects plant growth, we should observe similar patterns if we use phytomass rather
than rainfall; (4) we should not observe the same patterns if we examine other climatic traits,
like temperature, that are less important for plant growth in Africa; and (5) transhumant pastoral
rainfall should primarily affect conflict during the wet season (when groups are competing for
resources) and not the dry season (when they are not).
Test 1: Importance of transhumant pastoralism rather than mobility or pastoralism alone.
Our mechanism of interest suggests that both aspects of transhumant pastoralism are necessary;
namely, that groups move seasonally and they engage in animal herding. If an ethnic group is
characterized by only one of the two—they move without animals or they have animals but do
not move—then we do not expect to observe the same effects.

5. The data are from the World Bank’s “Pink Sheet” commodity price index dataset (World Bank 2021). All
indices are based on real prices.
422 REVIEW OF ECONOMIC STUDIES
TABLE 3
Robustness to controlling for the components of transhumant pastoralism
Indicator for the presence of conflict

(1) (2) (3) (4)

UCDP UCDP ACLED ACLED
I(Any) I(State) I(Any) I(Non-state)
Nearest Neighbouring Ethnic Group
Rain −0.0014 0.0005 −0.0020 −0.0021
(0.0011) (0.0008) (0.0015) (0.0015)

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

Rain × Pastoral 0.0043 −0.0022 0.0069 0.0067
(0.0044) (0.0035) (0.0061) (0.0061)
Rain × Transhumant 0.0040 0.0020 0.0029 0.0031
(0.0025) (0.0018) (0.0039) (0.0039)
Rain × Transhumant Pastoral −0.0191*** −0.0128** −0.0186** −0.0187**
(0.0069) (0.0057) (0.0088) (0.0088)
Dep. Var. Mean 0.0351 0.0253 0.0845 0.0842
Cell FE Yes Yes Yes Yes
Country × Year FE Yes Yes Yes Yes
Climate-Zone-Year Clusters 420 420 322 322
Cell Clusters 7,722 7,722 7,722 7,722
Observations 231,660 231,660 177,606 177,606
Notes: The unit of observation is a 0.5-degree grid-cell and year. “I(Any)” is an indicator variable that equals one if at
least one violent conflict occurs in a cell and year. “I(State)” is an indicator variable that equals one if at least one conflict
event involving the state occurs in a cell and year; “I(Non-state)” is an indicator variable that equals one if at least one
conflict event not involving the state occurs in a cell and year. Nearest Neighbouring Ethnic Group refers to the nearest
neighbouring ethnic territory to cell i. This regression controls for the corresponding variables at the Own Ethnic Group
level and the Own Cell level. Standard errors, which are reported in parentheses, are adjusted for clustering at the level
of a grid-cell and a climate zone-year. ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.

To test for this, we estimate a version of equation (2) that also includes each component of the
transhumant pastoralism measure—the mobility indicator and the pastoralism index—interacted
with rainfall. By including each component interaction, we are accounting separately for the role
of mobility and for the role of pastoralism. This is particularly important given the recent findings
in Eberle et al. (2020), which show the importance of mobility for mediating the effects of
temperature on conflict. This also addresses potential concerns arising due to factors associated
with pastoralism, such as the presence of a “culture of honour” and revenge-taking (Nisbett
and Cohen 1996; Grosjean 2014; Cao et al. 2021), which may be more acute in the absence
of rainfall. These effects are captured by the inclusion of the pastoralism measure (along with
relevant interactions) in the equation directly.
The estimates from the equation including the component interactions are reported in Table 3.
We find that our estimates of interest are robust to these additional controls and that the coeffi-
cients for the controls themselves are small and insignificant. This suggests that it is the seasonal
movement of migrating herd animals that is important for our findings and not either mobility
or the presence of herd animals alone.
Test 2: Concentration of conflict on agricultural land. The second testable prediction that
arises from our interpretation is that conflict due to adverse rainfall shocks in the territory of
transhumant pastoral groups should be concentrated in land that is agricultural. Using infor-
mation from variable v5 of the Ethnographic Atlas, we split the sample between cells that are
located within the territory of ethnic groups whose traditional reliance on agriculture for sub-
sistence exceeded 35% and those whose reliance was between 0 and 35%. We then re-estimate
equation (2) separately for agricultural and non-agricultural cells.
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 423

The estimates, reported in Table 4, show that our main effects are driven primarily by conflict
in agricultural cells. While the estimated coefficient for the interaction of interest, γ1s , is large
in magnitude and statistically significant for agricultural cells, it is much smaller in magnitude,
varies in sign, and is never statistically different from zero in non-agricultural cells. Thus, con-
sistent with our interpretation, it is agricultural grid cells that are primarily responsible for the
effects reported in Table 2.6
Test 3: Similar effects for phytomass. Our hypothesis implies that a lack of rainfall in the terri-
tory of transhumant pastoral groups leads to conflict because it reduces the amount of vegetation

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

available for herd animals, which are moved to more fertile agricultural lands as a consequence.
If this is the case, we should find that adverse phytomass growth in the territory of neighbouring
transhumant pastoral groups should be associated with increased conflict in precisely the same
manner as adverse rainfall.
We test for this by re-estimating equation (2) using the measures of phytomass in place of
rainfall. The estimates, reported in Panel A of Table 5, are very similar for UCDP and even larger
in magnitude for ACLED. For example, looking at overall conflict, we find that a one standard
deviation decrease in phytomass in the territory of a neighbouring transhumant pastoral group
increases conflict by 37.95% of the mean incidence when the UCDP measure is used (column 1)
and by 32.09% when the ACLED measure is used (column 3). The equivalent effects of rainfall
are 37.5% and 13.6%.7
Unlike rainfall, one might be concerned that our satellite measure of phytomass growth is
itself endogenous to both conflict and the location of grazing animals. To address this concern,
we create a Phytomass Suitability Index, which is phytomass predicted by rainfall at the level
of a cell and a year. Aggregating this measure to the level of an ethnic group and the level of a
nearest neighbour for each year, we then estimate a version of equation (2) where the six rainfall
variables are replaced by six corresponding Phytomass Suitability Indices. Estimates using the
indices are reported in Appendix Tables A10 and A11, where Table A10 uses a phytomass
suitability index predicted by a linear function of rainfall and Table A11 uses a measure that is
predicted by a quadratic function (i.e. rainfall and rainfall squared). The estimates are similar in
both magnitude and precision to our baseline estimates.
Test 4: No effects for temperature. According to our interpretation, we should not find the
same effects for temperature as we do for rainfall or phytomass since it is not as important for
plant growth in Africa. While it is well documented that temperature is linked to conflict through
many potential channels, we do not expect temperature to matter for conflict through our specific
interaction of interest.
We test for this by re-estimating equation (2) using temperature in place of rainfall. The find-
ings, reported in Panel B of Table 5, reveal that the same patterns are not present in the data when
we use temperature. We estimate a fairly precise zero coefficient for the interaction between
the temperature of a cell’s nearest neighbour and the neighbour’s measure of transhumant

6. We find similar results when we split cells into three groups: the effects are greatest in absolute magnitude for
cells in the highest agriculture category, defined as 66–100% reliance.
7. The robustness of our findings to the use of the phytomass measure alleviates the concern that imprecision in
the gridded rainfall data might be important for our estimates. While the underlying rainfall data are based on a dense
set of weather gauges, the gridded measure does rely on interpolation. By contrast, the phytomass measure is based on
satellite images measured weekly at the 1 km pixel level.
 424

TABLE 4
Effect of rain in nearest neighbouring THP territory on conflict in agricultural and non-agricultural cells
Conflict in Agricultural Cells Conflict in Non-Agricultural Cells

(1) (2) (3) (4) (5) (6) (7) (8)

UCDP UCDP ACLED ACLED UCDP UCDP ACLED ACLED
I(Any) I(State) I(Any) I(Non-state) I(Any) I(State) I(Any) I(Non-state)
Nearest Neighbouring Ethnic Group
Rain [γ0s ] −0.0006 0.0002 −0.0002 −0.0004 0.0000 −0.0001 −0.0105*** −0.0103***
(0.0007) (0.0006) (0.0011) (0.0011) (0.0026) (0.0024) (0.0036) (0.0036)
Rain × Transhumant Pastoral [γ1s ] −0.0119** −0.0121*** −0.0172*** −0.0180*** −0.0053 −0.0062 0.0052 0.0056
(0.0047) (0.0039) (0.0056) (0.0057) (0.0056) (0.0051) (0.0064) (0.0064)
Dep. Var. Mean 0.039 0.028 0.097 0.096 0.025 0.019 0.055 0.055
Cell FE Yes Yes Yes Yes Yes Yes Yes Yes
Country × Year FE Yes Yes Yes Yes Yes Yes Yes Yes
Climate-Zone-Year Clusters 390 390 299 299 390 390 299 299
Cell Clusters 5,482 5,482 5,482 5,482 2,240 2,240 2,240 2,240
Observations 164,460 164,460 126,086 126,086 67,200 67,200 51,520 51,520
Notes: The unit of observation is a 0.5-degree grid-cell and year. “I(Any)” is an indicator variable that equals one if at least one violent conflict occurs in a cell and year. “I(State)” is an
REVIEW OF ECONOMIC STUDIES

indicator variable that equals one if at least one conflict event involving the state occurs in a cell and year; “I(Non-state)” is an indicator variable that equals one if at least one conflict
event not involving the state occurs in a cell and year. Nearest Neighbouring Ethnic Group refers to the nearest neighbouring ethnic territory to cell i. Own Ethnic Group and Own Cell
covariates are included in the regressions but not reported. Standard errors, which are reported in parentheses, are adjusted for clustering at the level of a grid-cell and a climate zone-year.
∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 425
TABLE 5
Estimates using phytomass and temperature rather than rainfall
Indicator for the presence of conflict

(1) (2) (3) (4)

UCDP UCDP ACLED ACLED
I(Any) I(State) I(Any) I(Non-state)
Panel A: Effect of Phytomass
Nearest Neighbouring Ethnic Group
Phytomass 0.0001 0.0001 0.0003 0.0004

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

(0.0005) (0.0004) (0.0006) (0.0006)
Phytomass × Transhumant Pastoral −0.0043** −0.0041** −0.0085*** −0.0086***
(0.0018) (0.0016) (0.0018) (0.0018)
Effect of 1 Std. Dev. Phytomass Shock as % of Dep. Var. Mean:
Phytomass × Transhumant Pastoral −37.95 −50.73 −32.09 −32.70
p-value [0.02] [0.01] [0.00] [0.00]
Dep. Var. Mean 0.037 0.027 0.087 0.087
Climate-Zone-Year Clusters 280 280 294 294
Cell Clusters 7,722 7,722 7,722 7,722
Observations 154,440 154,440 162,162 162,162
Panel B: Effect of Temperature
Nearest Neighbouring Ethnic Group
Temperature 0.0019 0.0028** 0.0027 0.0026
(0.0016) (0.0013) (0.0027) (0.0027)
Temperature × Transhumant Pastoral 0.0022 0.0047 0.0030 0.0029
(0.0037) (0.0035) (0.0045) (0.0045)
Effect of 1 Std. Dev. Temp. Shock as % of Dep. Var. Mean:
Temp. × Transhumant Pastoral 5.55 15.89 3.48 3.42
p-value [0.54] [0.18] [0.52] [0.52]
Dep. Var. Mean 0.032 0.024 0.068 0.068
Climate-Zone-Year Clusters 364 364 252 252
Cell Clusters 7,722 7,722 7,722 7,722
Observations 200,728 200,728 138,968 138,968
Cell FE Yes Yes Yes Yes
Country × Year FE Yes Yes Yes Yes
Notes: The unit of observation is a 0.5-degree grid-cell and year. “I(Any)” is an indicator variable that equals one if at
least one violent conflict occurs in a cell and year. “I(State)” is an indicator variable that equals one if at least one conflict
event involving the state occurs in a cell and year; “I(Non-state)” is an indicator variable that equals one if at least one
conflict event not involving the state occurs in a cell and year. Nearest Neighbouring Ethnic Group refers to the nearest
neighbouring ethnic territory to cell i. Own Ethnic Group and Own Cell covariates are controlled for but not reported.
Standard errors, which are reported in parentheses, are adjusted for clustering at the level of a grid-cell and a climate
zone-year. ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.

pastoralism. This is consistent with our observation that, unlike rainfall, temperature is not a
first-order determinant of phytomass growth.8
Overall, the estimates indicate that the established mechanisms linking temperature to con-
flict in the literature cannot account for the effects we find here.9 This is particularly important

8. Interestingly, we find evidence of a direct relationship between temperature and conflict, as in the existing
literature. Specifically, we estimate that, in general, higher temperatures experienced by the ethnic group of a cell result
in more conflict in that cell.
9. As reported in Appendix Table A12, if we include both rainfall and temperature, our estimated rainfall spillover
effects from transhumant pastoral neighbours remain large and statistically significant, while we observe no equivalent
spillover effect from temperature shocks.
426 REVIEW OF ECONOMIC STUDIES

given the recent evidence that higher temperatures at the border between nomadic and sedentary
populations increase conflict. These null effects provide added assurance that our mechanism is
distinct from the “heat and hate” effects documented in Eberle et al. (2020).
Test 5: Concentration of conflict during the wet season. The fourth test focuses on the timing
of conflict within a year. Our mechanism of interest implies that adverse rainfall in transhumant
pastoral territories only generates conflict in nearby farmland during the wet season. A lack of
rain during the wet season forces transhumant pastoral groups to migrate early to neighbouring
farmlands, when land is still being used for cultivation, which generates conflict. By contrast,

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

during the dry season, there is no tension since land is fallow and animal grazing benefits both
groups.
We verify this prediction using a number of tests. In the first, we estimate a variant of equation
(2) where the dependent variable is a measure of conflict that is specific to each of the two
seasons. Because the length of the seasons differ across locations, we transform the dependent
variable to be a monthly average, either: (1) the fraction of months that have at least one conflict
incident, or (2) the average number of conflict incidents per month.
To separate wet-season conflicts from dry-season conflicts, we use data from the
MIRCA2000 global dataset (Portmann et al. 2010), which provides information on the begin-
ning and end of the growing season as of the year 2000 at a 5 arc minute (9.3 km at the equator)
resolution. We use the starting and final months of the growing season for the “main crop” of
a cell, defined as the crop with the greatest harvested area in the cell. Our sample is therefore
restricted to cells that contain some harvested cropland and experience both seasons. Among
these cells, the average duration of the main crop’s wet season is 5.75 months.
To ensure that we capture all conflict events due to the joint use of resources, we define
wet-season conflict as conflict events that begin during either the main crop’s growing season
or within a month after it ends. This allows for conflict events that coincide with the harvesting
period, which may extend beyond the estimated final month of the main crop’s growing season
according to the MIRCA2000 data. Dry-season conflicts are events that begin at any point during
the rest of the year.10
The estimates are reported in Panel A of Table 6. We find that our baseline effects are pri-
marily due to conflict events that occur in the wet season. The estimated effects on wet-season
conflict are about twice the magnitude and much more precisely estimated than the effects on
dry-season conflict. This is particularly striking because, without understanding the nature of
conflict that arises from transhumant pastoralism, one might expect rainfall to have the largest
effect on conflict during the dry season, when fresh water is more scarce.
The second test that we implement also measures season-specific rainfall. Thus, we esti-
mate the relationship between rainfall in a season and conflict in that season, and we do this
separately for both wet and dry seasons. The estimates, reported in Panel B of Table 6, show a
similar pattern. Our baseline finding is driven by adverse rainfall in the wet season causing con-
flict in the wet season rather than adverse rainfall in the dry season causing conflict in the dry
season.
Test 6: Combinations of predictions. The last exercise that we undertake is to combine the
prediction about the timing of the effects (wet season rather than dry season) with the importance

10. When dating conflicts, we use the earliest date indicated when multiple dates or a time interval is reported.
Thus, we focus on the first incident within a conflict event—which is our object of interest—rather than other incidents
that are more likely to be a continuation of previous clashes.
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 427
TABLE 6
Effects of neighbour’s rainfall on conflict during the wet and dry seasons
Wet Season UCDP Conflict Dry Season UCDP Conflict

(1) (2) (3) (4)

Incidence No. Events Incidence No. Events
Year Equiv. Year Equiv. Year Equiv. Year Equiv.
Panel A. Annual Rainfall and Conflict by Seasons
Nearest Neighbouring Ethnic Group
Annual Rain 0.0008 0.0039 −0.0022 −0.0017

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

(0.0023) (0.0048) (0.0032) (0.0113)
Annual Rain × Transhumant Pastoral −0.0346*** −0.1294** −0.0152 −0.0669
(0.0128) (0.0613) (0.0116) (0.0444)
Effect of 1 Std. Dev. Rain Shock as % of Dep. Var. Mean:
Annual Rain × Transhumant Pastoral −46.32 −94.31 −18.60 −44.40
p-value [0.01] [0.04] [0.19] [0.13]
Panel B. Seasonal Rainfall and Conflict by Seasons
Nearest Neighbouring Ethnic Group
Seasonal Rain 0.0013 0.0037 −0.0015 0.0008
(0.0016) (0.0041) (0.0022) (0.0073)
Seasonal Rain × Transhumant Pastoral −0.0184* −0.0834* −0.0043 −0.0103
(0.0104) (0.0486) (0.0103) (0.0214)
Effect of 1 Std. Dev. Rain Shock as % of Dep. Var. Mean:
Seasonal Rain × Transhumant Pastoral −41.73 −102.80 −6.75 −8.69
p-value [0.08] [0.09] [0.68] [0.63]
Dep. Var. Mean 0.090 0.165 0.098 0.181
Cell FE Yes Yes Yes Yes
Country × Year FE Yes Yes Yes Yes
Climate-Zone-Year Clusters 420 420 420 420
Cell Clusters 4,632 4,632 4,632 4,632
Observations 138,960 138,960 138,960 138,960
Notes: The unit of observation is a 0.5-degree grid-cell and year. “Incidence” is per-month UCDP conflict incidence in
either the wet season or the dry season as defined in the main text. “Number” is per-month number of UCDP conflict
events. Nearest Neighbouring Ethnic Group refers to the nearest neighbouring ethnic territory to cell i. Own Ethnic
Group and Own Cell covariates are included in the regressions but not reported. Standard errors, which are reported in
parentheses, are adjusted for clustering at the level of a grid-cell and a climate zone-year. ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p <
0.01.

of phytomass and the location of conflict events (agricultural land). Appendix Table A13 repro-
duces the estimates from Table 6, but using phytomass growth rather than rainfall. As shown, the
same pattern emerges in the data. It is during the wet season that we see the effects of phytomass
growth on conflict.
We next incorporate the prediction about the location of conflicts by re-estimating the
specifications reported in Table 6 and Appendix Table A13, but for agricultural cells and
non-agricultural cells separately. The estimates, reported in Appendix Tables A14 and A15
respectively, show that the seasonal patterns we identify (for both rainfall and phytomass) are
strongly present in agricultural cells but much less so in non-agricultural cells.
These specific patterns—on the timing of the effects during the year, the location of the
effects across groups, and the centrality of plant growth—are precisely what one would expect
according to our hypothesis: reduced rainfall in transhumant pastoral territories induces herders
to move to agricultural lands prior to the harvest, generating competition for resources that
ultimately results in conflict.
428 REVIEW OF ECONOMIC STUDIES

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

F IGURE 5
Total Jihadist and non-Jihadist conflicts over time in Africa

7. LEARNING FROM THE ESTIMATES

The estimates reported to this point provide evidence consistent with first-hand accounts of the
effects of climate change on conflict between transhumant pastoral groups and farmers. In this
section, we examine how this finding relates to extremist-religious conflict and whether or not
it is moderated by government policies or by the distribution of political power across ethnic
groups.

7.1. Examining religious extremism

We begin with the question of whether our estimated relationship can help to explain the rise
in religious conflict in Africa in the past two decades. This trend is shown in Figure 5, which
reports the average conflict incidence across cells in our UCDP data between 1989 and 2018
for events that involve at least one actor that is labelled as being a jihadist group and for those
events that do not.11 Jihadist conflicts have increased significantly since 2000, while non-jihadist
conflicts have remained relatively stable.
One apparent explanation for this is a rise in religious grievances or tensions between Islamic
and Christian groups. However, our findings raise the possibility that this trend is instead (or also)
due to the increased frequency of adverse rainfall shocks in transhumant pastoral territories. In
our data, groups with a value of transhumant pastoralism that is non-zero are 56.5% Muslim
and 27.8% Christian, whereas groups with a value of transhumant pastoralism equal to zero
are 24.6% Muslim and 48.4% Christian (see Appendix Table A14). Since the conflicts that
we study often involve a largely Muslim group on one side and a largely Christian group on

11. We identify jihadist conflict events as those for which (i) the word “jihad” is present in either the actor’s
name or in the source headline or (ii) the word “Islam-” appears in the source headline and one of the actors is explicitly
jihadist. The list of groups is in Appendix A.
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 429
TABLE 7
Jihadist violence
Indicator for the presence of conflict

(1) (2) (3) (4)

I(Jihadist) I(Non-Jihadist) I(Jihadist) I(Non-Jihadist)
Nearest Neighbouring Ethnic Group
Rain −0.0000 −0.0006 0.0006 0.0002
(0.0003) (0.0006) (0.0005) (0.0020)
Rain × Transhumant Pastoral −0.0051** −0.0063** −0.0056** −0.0056*

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

(0.0022) (0.0026) (0.0025) (0.0030)
Rain × Share Muslim −0.0020 −0.0016
(0.0015) (0.0025)
Rain × Share Christian −0.0003 −0.0007
(0.0006) (0.0028)
Nearest Neighbouring Ethnic Group: Additional Calculations
Effect of 1 Std. Dev. Rain Shock as % of Dep. Var. Mean:
Rain × Transhumant Pastoral −82.03 −27.05 −82.42 −21.41
p-value [0.02] [0.01] [0.03] [0.06]
Dep. Var. Mean 0.007 0.028 0.008 0.032
Cell FE Yes Yes Yes Yes
Country × Year FE Yes Yes Yes Yes
Climate-Zone-Year Clusters 420 420 420 420
Cell Clusters 7,722 7,722 6,507 6,507
Observations 231,660 231,660 195,210 195,210
Notes: The unit of observation is a 0.5-degree grid-cell and year. “Jihadist” is an indicator variable that equals one if at
least one UCDP conflict event occurs in a cell–year involving a self-styled jihadist group, as defined in the main text.
“Non-Jihadist” is an indicator variable that equals one if at least one UCDP conflict event occurs in a cell–year that does
not involve a self-styled jihadist group. Nearest Neighbouring Ethnic Group refers to the nearest neighbouring ethnic
territory to cell i. Own Ethnic Group and Own Cell covariates are included in the regressions but not reported. In columns
3 and 4, the covariates also include own ethnic group rainfall interacted with the share Muslim and the share Christian,
as well as own cell rainfall interacted with the same two variables. Standard errors, which are reported in parentheses,
are adjusted for clustering at the level of a grid-cell and a climate zone-year. ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.

the other, they may take the appearance of—or soon develop into—religious conflict. Tensions
between farmers and herders have also been known to generate support for jihadist groups,
which facilitates recruitment (Benjaminsen and Ba 2019). Jihadist groups may therefore become
involved in conflicts between farmers and herders that arise due to reduced rainfall.
We test for this possibility by estimating our baseline specification—equation
(2)—separately for jihadist and non-jihadist conflicts. The estimates are reported in columns
1 and 2 of Table 7. We find statistically significant and quantitatively similar estimates for the
coefficient on our interaction term for both types of conflict. This suggests that our mechanism
applies equally to both jihadist and non-jihadist conflict. The predicted effects of a one standard
deviation rainfall shock in terms of the mean of the dependent variable, reported in the second
panel of the table, are about three times greater for jihadist conflicts (82%) than non-jihadist
conflicts (27%). This is because our measure of jihadist conflict has a lower mean incidence,
which can be seen in Figure 5, particularly prior to 2000.
In columns 3 and 4, we check whether our findings are simply because transhumant pas-
toral groups are more likely to be Islamic, which may be correlated with other factors—such as
low educational mobility, as in Alesina et al. (2023)—that interact with low rainfall in a man-
ner that results in conflict spillovers. To account for the importance of religion, we include the
proportion of each ethnic group that is Christian and Muslim (as of 2020) interacted with each
of our three rainfall measures (own cell, own ethnic group, and nearest neighbouring ethnic
430 REVIEW OF ECONOMIC STUDIES

group) as controls in equation (2).12 The estimated effects are nearly identical in magnitude and
significance after accounting for contemporary religion.
Our findings suggest that extremist-religious violence responds to adverse rainfall in almost
the same manner as other types of violence. This is consistent with atavistic grievances not being
the sole determinant of religious conflict.

7.2. Policy responses: development aid projects and protected conservation areas

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

Development Aid Projects. In recent decades, many development organizations have designed
interventions to combat the adverse effects of climate change. Examples include projects that
aim to enhance agricultural productivity, improve irrigation infrastructure, or expand protected
conservation areas. A potential solution to the effects that we document is to implement more
of these interventions. However, given the specifics of the mechanism that we uncover, it is not
clear whether these policies will help. The effects that we identify are due to adverse rainfall
causing pastoral groups to migrate to nearby farmlands before harvest. Improving the agri-
cultural productivity of farmland does not solve this underlying problem. Moreover, irrigation
projects potentially facilitate the conversion of marginal lands to farmland, thus reducing the
land available for grazing. Land privatization and the creation of protected conservation lands
that ban animal grazing likely have the same effect. In general, any policy that constrains the land
available to pastoralists in response to adverse rainfall can potentially increase the likelihood that
they come into conflict with farmers during the growing season.
Against this backdrop, we examine whether our documented effects are stronger or weaker
in the presence of such projects. To do this, we allow our effects of interest to differ depend-
ing on the stock of aid projects present in a country and year. We measure the presence of aid
projects in a country over time using the Aid Data repository, which reports detailed information
on all bilateral and multilateral foreign aid projects from 1947 to 2013.13 We measure the cumu-
lative number of project locations that have been implemented in each country prior to that year
(since 1947) and normalize this by the number of cells in a country.14 We denote this variable
For eign Aid ct .15
We then estimate the following equation, which allows our effect of interest to vary by the
prevalence of foreign aid projects in a country:
N eighbour N eighbour N eighbour
yiet = ψ0s Rain it + ψ1s Rain it × T ranshumant Pastoral i
N eighbour N eighbour
+ ψ2s Rain it × T ranshumant Pastoral i × For eign Aid ct−1
N eighbour N eighbour
+ ψ3s Rain it × For eign Aid ct−1 + ψ4s T ranshumant Pastoral i
OwnGr oup OwnGr oup
× For eign Aid ct−1 + ψ5s Rain et + ψ6s Rain et

12. The data are constructed using information from the World Religion Database, which reports information on
the populations of 18 religions for each language group in the world (Johnson and Grim 2021). The data are reported
with Ethnologue identifiers which we match to our Ethnographic Atlas. Since multiple Ethnologue groups often match
to one Ethnographic Atlas group, we create Ethnographic Atlas level measures by taking population-weighted averages
across all Ethnologue groups that match to a Ethnographic Atlas group.
13. See AidData (2017) and Tierney et al. (2011).
14. For example, if one umbrella program is implemented in 10 locations in a country with 20 cells, it is measured
with a value of 10/20 = 0.5.
15. Descriptive statistics for all country-year level variables employed in this section are presented in Appendix
Table A16.
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 431
TABLE 8
Heterogeneity by the presence of international aid projects
Indicator for the presence of conflict

(1) (2) (3) (4)

UCDP UCDP ACLED ACLED
I(Any) I(State) I(Any) I(Non-state)
Nearest Neighbouring Ethnic Group
Rain × Transhumant Pastoral −0.0129*** −0.0122*** −0.0038 −0.0036

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

(0.0038) (0.0035) (0.0039) (0.0039)
Rain × Transhumant Pastoral × −0.0059 −0.0068 −0.0113 −0.0117
Total Agriculture Aid (0.0064) (0.0055) (0.0074) (0.0075)
Rain × Transhumant Pastoral × 0.0004 0.0004 0.0005 0.0005
Total Non-Agriculture Aid (0.0004) (0.0004) (0.0005) (0.0005)
Dep. Var. Mean 0.032 0.024 0.068 0.068
Cell FE Yes Yes Yes Yes
Country × Year FE Yes Yes Yes Yes
Climate-Zone-Year Clusters 364 364 252 252
Cell Clusters 7,722 7,722 7,722 7,722
Observations 200,772 200,772 138,996 138,996
Notes: The unit of observation is a 0.5-degree grid-cell and year. “I(Any)” is an indicator variable that equals one if at
least one violent conflict occurs in a cell and year. “I(State)” is an indicator variable that equals one if at least one conflict
event involving the state occurs in a cell and year; “I(Non-state)” is an indicator variable that equals one if at least one
conflict event not involving the state occurs in a cell and year. Nearest Neighbouring Ethnic Group refers to the nearest
neighbouring ethnic territory to cell i. This regression controls for the corresponding variables at the Own Ethnic Group
level and the Own Cell level. Standard errors, which are reported in parentheses, are adjusted for clustering at the level
of a grid-cell and a climate zone-year. ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.

× T ranshumant Pastoral eOwnGr oup + ψ7s Rain itOwnCell + ψ8s Rain itOwnCell

× T ranshumant Pastoral eOwnGr oup + αis + αc(i)t

s
+ ξiet
s
, (3)

where For eign Aid ct is as described above and all indices and other variables are as defined in
equation (2). The estimates of interest are ψ1s , which is our main spillover effect when transhu-
mant pastoral groups are in a country with no previous foreign aid, and ψ2s , which shows how
our effect of interest differs depending on the amount of past foreign aid projects in a country.
The first analysis that we undertake divides foreign aid projects into two categories: those
that are agricultural and those that are not. We identify agricultural projects as those for which
the reported sector code is “Agriculture” and non-agricultural projects as all others. We allow
our estimated effects of interest to differ depending on the cumulative presence of both types of
projects in a country and year. The estimates, which are reported in Table 8, show no evidence
that agricultural aid reduces the effects of rainfall in transhumant pastoral territories on conflict
in nearby cells. While the point estimates are imprecise, their sign and magnitudes suggest that
agricultural aid may actually exacerbate the effects of interest.
To investigate whether the estimates mask heterogeneous effects, we create even finer cate-
gories of aid projects, distinguishing between irrigation projects, forestry projects, conservation
projects, land projects, other agricultural projects, and other non-agricultural projects.16 The

16. We measure these variables by searching for relevant keywords in the set of variables that contain the project
descriptions or sectors. The keywords are, respectively, “irrigat” for irrigation; “forest” for forestry; “conserv” for con-
servation; and “land”, “tenure” or “titling” for land. We define the residual projects as agricultural or non-agricultural as
in the first analysis.
432 REVIEW OF ECONOMIC STUDIES

estimates, which are reported in Appendix Table A17, do not indicate that any of these types of
aid alleviate the effects of adverse rainfall shocks in transhumant pastoral areas on conflict.
Finally, because For eign Aid ct varies over time as well as between countries, we
N eighbour
estimate a version of (3) that additionally controls for Rain it × T ranshumant
N eighbour N eighbour N eighbour
Pastoral i × αc and Rain it
s
× T ranshumant Pastoral i × αts ; that is,
our double interaction of interest interacted with country fixed effects and with year fixed effects.
By including these interactions, we only exploit variation in foreign aid that is over time and
within-country rather than across countries. In effect, this implies that our triple-interaction

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

effect of interest, ψ2s , is identified using a difference-in-differences style estimator rather than
a cross-country estimator. The results of this procedure are presented in Panels A and B of
Appendix Table A18. The results again suggest that, if anything, agricultural projects may
exacerbate the main effects that we document.17
Conservation Areas. The next analysis that we undertake looks specifically at the role of
protected conservation lands in a country at a point in time. While conservation is an important
tool for environmental protection, it can also be disruptive for pastoral groups. Lands that are
converted into conservation areas may contain transhumant pastoral corridors or grazing pas-
tures. Since conservation areas typically forbid the use of protected lands for grazing or impose
regulations or fees when use is allowed, their expansion may disrupt existing transhumant migra-
tion routes and cooperative arrangements with farmers (Bergius et al. 2020; Cavanagh et al.
2020).
We measure the presence of conservation lands in each country and year using data from
Protected Planet, a global database of protected areas and other conservation measures,18 and
compile panel data that measures the share of a country’s total area that is under protection each
year. We then estimate a variant of equation (3) that uses this measure rather than foreign aid.
The estimates, reported in Table 9, suggest that conservation lands may exacerbate the effects
of adverse rainfall experienced by transhumant pastoral groups. To illustrate this, in the second
panel of the table, we report the predicted effect (relative to the mean of the dependent vari-
able) of a one standard deviation change in rainfall for different values of the conservation land
variable. We find that, in countries with a large share of protected conservation land (i.e. at the
90th percentile of the sample), lower rainfall in a neighbouring transhumant pastoral territory
significantly increases conflict by 26–87%, depending on the outcome variable. In countries with
minimal conservation (i.e. at the 10th percentile), the effects are only marginally significant and
range from 1 to 36%.
To explore these effects further, we disaggregate the country-level conservation measure into
two subnational measures: one for ethnicity e (that lies in country c) and one for its complement,
i.e. the rest of country c. This is motivated by observations that grazing bans in certain parts
of a country can displace conflict into neighbouring locations (e.g. Avuwadah 2021). We allow
for such spillover effects by estimating heterogeneous effects by both measures. The estimates,
reported in Appendix Table A19, reveal clear evidence that the exacerbating effect of conserva-
tion is driven entirely by the presence of conservation areas located elsewhere; namely, within
the country but outside of a cell’s ethnic territory. By contrast, the presence of conservation areas
in a cell’s own ethnic territory appears to reduce the effects of adverse rainfall in pastoral ter-
ritories on conflict. Thus, while conservation areas appear to reduce violent events locally, they

N eighbour N eighbour
17. We do not present coefficients for Rain it × T ranshumant Pastoral i since they are
relevant only for the omitted country and year.
18. See UNEP-WCMC and IUCN (2021). The database was accessed via the URL protectedplanet.net on 16
May 2021.
 TABLE 9
Heterogeneity by the share of conservation lands in a country
Indicator for presence of conflict
McGuirk & Nunn

(1) (2) (3) (4)

UCDP UCDP ACLED ACLED
I(Any) I(State) I(Any) I(Non-state)
Nearest Neighbouring Ethnic Group
Rain × Transhumant Pastoral −0.0082 −0.0073 −0.0005 −0.0003
(0.0050) (0.0047) (0.0055) (0.0055)
Rain × Transhumant Pastoral × −0.0248 −0.0390 −0.0626** −0.0638**
Share Protected Area in Country (0.0259) (0.0258) (0.0260) (0.0262)
Nearest Neighbouring Ethnic Group: Additional Calculations
Effect of 1 Std. Dev. Rain Shock as % of Dep. Var. Mean:
Rain × Transhumant Pastoral when Protected Area at 10th pctile −28.6 −35.6 −1.1 −0.9
p-value [0.09] [0.10] [0.88] [0.91]
Rain × Transhumant Pastoral when Protected Area at 90th pctile −52.2 −87.0 −25.9 −26.2
p-value [0.00] [0.00] [0.00] [0.00]
Dep. Var. Mean 0.035 0.025 0.085 0.084
Cell FE Yes Yes Yes Yes
Country × Year FE Yes Yes Yes Yes
Climate-Zone-Year Clusters 420 420 322 322
Cell Clusters 7,718 7,718 7,718 7,718
Observations 231,540 231,540 177,514 177,514
Notes: The unit of observation is a 0.5-degree grid-cell and year. “I(Any)” is an indicator variable that equals one if at least one violent conflict occurs in a cell and year. “I(State)” is an
indicator variable that equals one if at least one conflict event involving the state occurs in a cell and year; “I(Non-state)” is an indicator variable that equals one if at least one conflict
event not involving the state occurs in a cell and year. Nearest Neighbouring Ethnic Group refers to the nearest neighbouring ethnic territory to cell i. Relevant covariates at the Own
Ethnic Group and Own Cell levels are controlled for but not reported. Standard errors, which are reported in parentheses, are adjusted for clustering at the level of a grid-cell and a climate
zone-year. ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA
433

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

434 REVIEW OF ECONOMIC STUDIES

come at the cost of increasing violence elsewhere. The net effect, as documented in Table 9, is
an aggregate increase in the effect of interest.19
Overall, the results of this exercise are consistent with conservation areas leading to more
constraints faced by herders, resulting in a larger effect of adverse rainfall in pastoral territories
on nearby conflict. The effects appear to be due to a spillover mechanism, whereby conservation
areas deflect conflict towards other parts of a country.

7.3. Rainfall scarcity, pastoral representation in government, and conflict

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

These estimates suggest that conflict induced by adverse rainfall in transhumant pastoral terri-
tories may be exacerbated by government policies such as the expansion of conservation areas.
This raises the broader question of whether national political economy forces play an important
role in either moderating or amplifying the main relationship that we document. Here, we test
whether the same spillover effects are present when pastoral groups have more political power.
The motivation for the test comes from the fact that pastoral groups are less likely to be
afforded grazing rights when they are excluded from national politics. In this scenario, state
forces will serve to protect the property rights of landowning farmers via restrictions or outright
bans on grazing, which have recently been implemented in a number of countries (Avuwadah
2021). Another less obvious example is land titling programs, which weaken the legitimacy of
customary use rights that are important to pastoral groups (Boone 2019).
Numerous studies have documented cases of policy bias against pastoral groups. Often, this
stance is explicit, with transhumant pastoralism being viewed as inefficient and outdated. For
example, the president of Tanzania, Jakaya Kikwete, in his 2005 inaugural speech to Parliament,
argued: “Our people must change from being nomadic cattle herders to being modern livestock
keepers.” A year later, during a press conference, he asserted: “We are producing little milk,
export very little beef, and our livestock keepers roam throughout the country with their animals
in search for grazing grounds. We have to do away with archaic ways of livestock farming.”
(Mattee and Shem 2006, p. 4).
We measure the extent to which political power in a country is held by transhumant pastoral
groups using information from the Ethnic Power Relations (EPR) Dataset, which documents the
nature of political power held by ethnic groups (Cederman et al. 2010). We use this information
to construct a measure of the total amount of political power held by an ethnic group e in coun-
try c in year t, which we denote by Power ect . The categories, and their numerical values, are
given by: (0) Fully excluded from politics (self-exclusion or discrimination); (1) Powerless; (2)
Junior partner in government; (3) Senior partner in government; (4) Dominant power; and (5)
Monopoly power.
Our interest is in the share of total political power in a country that is held by transhumant
pastoral groups. We measure the amount of political power in country c in year t by aggregating
the power held by all ethnic groups e: e Power ect . We measure the amount of power held by
transhumant pastoral groups by: e T ranshumant Pastoral e × Power ect . The share of power
held by transhumant pastoral groups in a country and year is then:

T ranshumant Pastoral e × Power ect
THP
Power ct = e  .
e Power ect

19. Since variation in conservation land is likely endogenous to many relevant factors, in Panels C and D of
N eighbour N eighbour
Appendix Table A18 we again include controls for Rain it × T ranshumant Pastoral i × αcs and
N eighbour N eighbour
Rain it × T ranshumant Pastoral i × αts , finding similar effects to those reported above.
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 435

The distribution of the measure across countries and years is shown in Appendix Figure A3.
It is clear that the amount of political power held by pastoral groups is limited. The median
THP
value of Power ct is 0.09, and a third of the observations have a measure that is equal to
zero, indicating transhumant pastoral groups do not hold any political power. The highest value
of the measure is 0.61, which is for Mauritania from 1989 to 2017, when the Delim, Trarza,
Regeibat, Zenega, Tajakant, and Berabish pastoral groups were represented as junior partners in
government.
Using the transhumant political power measure, we estimate a variant of equation (2) that

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

allows our effect of interest to differ depending on the extent to which transhumant pastoral
groups hold political power in that country in year t − 1, Power ct−1
THP
. We use a lagged measure,
which helps to address the potential for reverse causality—that is, conflict in year t affecting a
change in power in year t. The estimating equation is:

N eighbour N eighbour N eighbour

yiet = φ0s Rain it + φ1s Rain it × T ranshumant Pastoral i
N eighbour N eighbour
+ φ2s Rain it × T ranshumant Pastoral i × Power c(i)t−1
THP

N eighbour N eighbour
+ φ3s Rain it × Power c(i)t−1
THP
+ φ4s T ranshumant Pastoral i
OwnGr oup OwnGr oup
× Power c(i)t−1
THP
+ φ5s Rain et + φ6s Rain et
× T ranshumant Pastoral eOwnGr oup + φ7s Rain itOwnCell + φ8s Rain itOwnCell
× T ranshumant Pastoral eOwnGr oup + αis + αc(i)t
s
+ ξiet
s
, (4)

where all indices and variables are as in equation (2). The estimates of interest are φ1s , which
is our main spillover effect when transhumant pastoral groups have no political power, and φ2s ,
which tells us how much the estimated spillover effect changes as transhumant pastoral groups
gain more political power.
Estimates of equation (4) are reported in Table 10. We find that the estimated coefficient for
the interaction between a nearest neighbour’s rainfall and that neighbour’s measure of transhu-
mant pastoralism, φˆ1s , is negative and statistically significant for all four measures. This is the
estimated effect for a country where the share of power held by transhumant pastoral groups
is zero. The estimated coefficient for the triple interaction, φˆ2s , is positive and generally signifi-
cant, indicating that the effect of rainfall in the territory of a neighbouring transhumant pastoral
group on conflict is closer to zero when transhumant pastoral groups have more national political
power.
To assess the importance of the estimated heterogeneity, in the bottom panel of each
N eighbour
table, we calculate the predicted effect and statistical significance of Rain it ×
N eighbour THP
T ranshumant Pastoral i at different values of Power c(i)t−1 . The first predicted effect
THP
that we report is for a value of Power c(i)t−1 that is equal to the 10th percentile of its distribution,
which is zero. Below this, we report the same statistic calculated at the 90th percentile (0.303).
We find that for country-years in which no transhumant pastoral groups share political power,
the estimated spillover effect is large. For example, a one standard deviation decrease in rain-
fall is associated with an increase of conflict of 58% for all conflicts using the UCDP measure
and 82% for all conflicts using the ACLED measure. When a country is at the 90th percentile
 436

TABLE 10
Heterogeneity by share of political power held by transhumant pastoral groups
Indicator for the presence of conflict

(1) (2) (3) (4)

UCDP UCDP ACLED ACLED
I(Any) I(State) I(Any) I(Non-state)
Nearest Neighbouring Ethnic Group
Rain × Transhumant Pastoral −0.0158** −0.0151*** −0.0513*** −0.0513***
(0.0062) (0.0054) (0.0091) (0.0091)
Rain × Transhumant Pastoral × 0.0458** 0.0367* 0.1834*** 0.1824***
THP Power Share (0.0231) (0.0211) (0.0392) (0.0393)
Nearest Neighbouring Ethnic Group: Additional Calculations
Effect of 1 Std. Dev. Rain Shock as % of Dep. Var. Mean:
Rain × Transhumant Pastoral when THP Power at 10th pctile −58.1 −74.0 −81.7 −82.1
p-value [0.01] [0.01] [0.00] [0.00]
Rain × Transhumant Pastoral when THP Power at 90th pctile −7.2 −19.4 6.8 6.3
p-value [0.64] [0.32] [0.52] [0.56]
Dep. Var. Mean 0.033 0.024 0.075 0.075
Cell FE Yes Yes Yes Yes
Country × Year FE Yes Yes Yes Yes
Climate-Zone-Years 406 406 308 308
Cells 7,018 7,018 7,015 7,015
REVIEW OF ECONOMIC STUDIES

Observations 195,975 195,975 149,290 149,290

Notes: The unit of observation is a 0.5-degree grid-cell and year. “I(Any)” is an indicator variable that equals one if at least one violent conflict occurs in a cell and year. “I(State)” is an
indicator variable that equals one if at least one conflict event involving the state occurs in a cell and year; “I(Non-state)” is an indicator variable that equals one if at least one conflict
event not involving the state occurs in a cell and year. Nearest Neighbouring Ethnic Group refers to the nearest neighbouring ethnic territory to cell i. This regression controls for the
corresponding variables at the Own Ethnic Group level and the Own Cell level. Standard errors, which are reported in parentheses, are adjusted for clustering at the level of a grid-cell
and a climate zone-year. ∗ p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 437

of transhumant pastoral political power, these effects are not statistically different from zero. In
addition, they are very small: -7% for UCDP and 7% for ACLED.20
While the estimates reported here are merely correlational, they are consistent with political
power playing an important role in determining whether episodes of low rainfall in pastoral
areas lead to conflict. They align with prior evidence showing that, in the absence of political
power-sharing, minority groups have stronger incentives to fight (Mueller and Rohner 2018).

8. CONCLUSIONS

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

We have studied the question of whether climate change is responsible for disrupting longstand-
ing relationships between transhumant pastoralists and neighbouring sedentary agriculturalists
in Africa. Traditionally, transhumant pastoralists benefit from a cooperative relationship with
sedentary agriculturalists whereby arable land is used for farming in the wet season and grazing
in the dry season. Our findings confirm anecdotal accounts that decreased rainfall in transhumant
pastoral territories is forcing herders to migrate to neighbouring agricultural territories before the
harvest, resulting in competition for resources and the emergence of conflict.
The core of our analysis documented a relationship between adverse rainfall in the territo-
ries of transhumant pastoralists and conflict in the territory of neighbouring ethnic groups. To
test for the mechanism of interest—disruption to the seasonal migrations of transhumant pas-
toralists—we confirmed the effects through a series of falsification exercises. We found that the
conflicts induced by rainfall scarcity are concentrated in nearby agricultural lands and tend to
occur during the wet season, which is when land is still used for cultivation, and not during the
dry season, when land is left fallow and available for grazing. We also found that the effect of
rainfall operates through its influence on phytomass growth, which grazing animals require for
sustenance.
Our estimates also shed light on a specific form of conflict that has become more pervasive
in Africa in recent decades, namely religious violence. Transhumant pastoral groups tend to be
Islamic, while sedentary agriculturalists tend to be Christian. Our estimates indicate that a large
proportion of extremist-religious violence involving jihadist groups is due to the mechanism we
document rather than primordial grievances alone. Our counterfactual exercise implies that if
rainfall were one standard deviation higher during our study period, jihadist conflict would be
lower by 31%.
Our analysis also generates important policy implications. We examined whether policies
that are commonly used to combat the effects of environmental degradation can alleviate the
destructive effects that we identify in this article. We found no evidence that implementing agri-
cultural development aid projects or expanding protected conservation areas contributes to the
reduction of conflict that occurs due to lower rainfall in transhumant pastoral locations. The
findings suggest that such projects do not address the root cause of the conflict and may even be
counterproductive.
By contrast, we did find evidence that political economy factors are important. The esti-
mated effects are closer to zero when pastoral ethnic groups have a greater share of national
political power. Since transhumant pastoral groups are typically under-represented in national

20. Again, in Panel E of Appendix Table A18, we report estimates from a specifica-
N eighbour N eighbour
tion that also includes controls for Rain it × T ranshumant Pastoral i × αcs and
N eighbour N eighbour
Rain it × T ranshumant Pastoral i × αts . The country fixed effects interacted with our double
T H P to produce our estimates of
interaction of interest ensures that we use only within-country variation in Power c(i)t−1
interest. As reported, we again find positive and significant estimates.
438 REVIEW OF ECONOMIC STUDIES

politics, this suggests that a more equitable distribution of political power could have signif-
icant dividends in the form of peace. Indeed, if taken literally, our estimates imply that more
equitable politics could fully eliminate the effects of adverse rainfall on the conflict that we
document.
Finally, our findings highlight the importance of understanding the ethnic and cultural
context when studying conflict and climate change. In particular, they illustrate the value of
understanding pastoral populations and their way of life, which remains understudied and under-
appreciated in development economics despite comprising perhaps more than a fifth of Africa’s

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

population.
Acknowledgments. We are grateful to Felipe Valencia Caicedo, Aerie Changala, Edward Glaeser, Eliana La Fer-
rara, Julian Marenz, Ted Miguel, Nonso Obikili, Alex Orenstein, Elias Papaioannou, Imran Rasul, Dominic Rohner,
François Seyler, Maria Micaela Sviatschi, and participants at various conferences and seminars for helpful comments
and feedback. We also thank Mohammad Ahmad, Laura Kincaide, Talha Naeem, Leo Saenger, and Satish Wasti for their
excellent RA work.

Supplementary Data
Supplementary data are available at Review of Economic Studies online.

Data Availability Statement

The data and code underlying this research are available on Zenodo at https://dx.doi.org/10.5281/zenodo.10543632.

REFERENCES
ACEMOGLU, D., FERGUSSON, L. and JOHNSON, S. (2020), “Population and Conflict”, Review of Economic Studies,
87, 1565–1604.
AidData (2017), “AidDataCore ResearchRelease Level1 v3.1 Research Releases Dataset” http://aiddata.org/research-
datasets (accessed 9 April 2021).
ALESINA, A., HOHMANN, S., MICHALOPOULOS, S., et al. (2023), “Religion and Educational Mobility in Africa”,
Nature, 618, 134–143.
ARBATLI, C. E., ASHRAF, Q. H., GALOR, O., et al. (2020), “Diversity and Conflict”, Econometrica, 88, 727–797.
AVUWADAH, B. Y. (2021), “Do Conflict Management Policies Work? Evidence from a Ban on Open Grazing Laws in
Nigeria” (Working paper, University of Florida).
AYANTUNDE, A. A., ASSE, R., SAID, M. Y., et al. (2014), “Transhumant Pastoralism, Sustainable Management
of Natural Resources and Endemic Ruminant Livestock in the Sub-Humid Zone of West Africa”, Environment,
Development and Sustainability, 16, 1097–1117.
BAI, Y. and KUNG, J. K.-S. (2011), “Climate Shocks and Sino-Nomadic Conflict”, Review of Economics and Statistics,
93, 970–981.
BAYSAN, C., BURKE, M., GONZÁLEZ, F., et al. (2019), “Non-Economic Factors in Violence: Evidence from
Organized Crime, Suicides and Climate in Mexico”, Journal of Economic Behavior & Organization, 168, 434–452.
BECK, J. and SIEBER, A. (2010), “Is the Spatial Distribution of Mankind’s Most Basic Economic Traits Determined
by Climate and Soil Alone?”, PLoS ONE, 5, e10416. https://doi.org/10.1371/journal.pone.0010416.
BECKER, A. (2019), “On the Economic Origins of Restrictions on Women’s Sexuality” (Working Paper, Harvard
University).
BENJAMINSEN, T. A., ALINON, K., BUHAUG, H., et al. (2012), “Does Climate Change Drive Land-Use Conflicts
in the Sahel?”, Journal of Peace Research, 49, 97–111.
BENJAMINSEN, T. A. and BA, B. (2019), “Why Do Pastoralists in Mali Join Jihadist Groups? A Political Ecology
Explanation”, Journal of Peasant Studies, 46, 1–20.
BERGIUS, M., BENJAMINSEN, T. A., MANGANGA, F., et al. (2020), “Green Economy, Degredation Narratives, and
Land-Use Conflicts in Tanzania”, World Development, 129, 104850.
BERMAN, N., COUTTENIER, M., ROHNER, D., et al. (2017), “This Mine Is Mine! How Minerals Fuel Conflicts in
Africa”, American Economic Review, 107, 1564–1610.
BESLEY, T. and REYNAL-QUEROL, M. (2014), “The Legacy of Historical Conflict: Evidence from Africa”, American
Political Science Review, 108, 319–336.
BIASUTTI, M. (2018), “Rainfall Trends in the African Sahel: Characteristics, Processes, and Causes”, WIREs Climate
Change, 10, e591.
BOLLIG, M. (2006), Risk Management in a Hazardous Environment: A Comparative Study of Two Pastoral Societies
(New York: Springer).
BOONE, C. (2019), “Legal Empowerment of the Poor Through Property Rights Reform: Tensions and Trade-Offs of
Land Registration and Titling in Sub-Saharan Africa”, Journal of Development Studies, 55, 384–400.
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 439

BROTTEM, L. V. (2016), “Environmental Change and Farmer-Herder Conflict in Agro-Pastoral West Africa”, Human
Ecology, 44, 547–563.
BRYAN, G., CHOWDHURY, S. and MOBARAK, A. M. (2014), “Underinvestment in a Profitable Technology: The
Case of Seasonal Migration in Bangladesh”, Econometrica, 82, 1671–1748.
BURKE, M., HSIANG, S. M. and MIGUEL, E. (2015), “Climate and Conflict”, Annual Review of Economics, 7,
577–617.
CAO, Y., ENKE, B., FALK, A., et al. (2021), “Herding, Warfare, and a Culture of Honor: Global Evidence” (Working
paper, Harvard University).
CAVANAGH, C. J., WELDEMICHEL, T. and BENJAMINSEN, T. A. (2020), “Gentrifying the African Landscape:
The Performance and Powers of For-Profit Conservation on Southern Kenya’s Conservancy Frontier”, Annals of the
American Association of Geographers, 110, 1594–1612.

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

CEDERMAN, L.-E., WIMMER, A. and MIN, B. (2010), “Why Do Ethnic Groups Rebel? New Data and Analysis”,
World Politics, 62, 87–119.
COLLIER, P. and HOEFFLER, A. (1998), “On Economic Causes of Civil War”, Oxford Economic Papers, 50, 563–573.
COLLIER, P. and HOEFFLER, A. (2004), “Greed and Grievance in Civil War”, Oxford Economic Papers, 56, 563–595.
COOK, K. H. and VIZY, E. K. (2013), “Projected Changes in East African Rainy Seasons”, Journal of Climate, 26,
5931–5948.
Copernicus. (n.d.), “Copernicus Global Land Service: Dry Matter Productivity, Global, 1 km, 1999–2019”.
DEPETRIS-CHAUVIN, E. (2015), “State History and Contemporary Conflict: Evidence from Sub-Saharan Africa”
(Working paper, Pontifical Catholic University of Chile).
DEPETRIS-CHAUVIN, E. and ÖZAK, Ö. (2020), “Borderline Disorder: (De Facto) Historical Ethnic Borders and
Contemporary Conflict in Africa” (Working paper, Pontificia Universidad Católica de Chile).
D’ONOFRIO, D., SWEENEY, L., VON HARDENBERG, J., et al. (2019), “Grass and Tree Cover Responses to Intra-
Seasonal Rainfall Variability Vary Along a Rainfall Gradient in African Tropical Grassy Biomes”, Nature, 9, 2334.
DUBE, O. and VARGAS, J. F. (2013), “Commodity Price Shocks and Civil Conflict: Evidence from Colombia”, Review
of Economic Studies, 80, 1384–1421.
DYSON-HUDSON, R. and DYSON-HUDSON, N. (1980), “Nomadic Pastoralism”, Annual Review of Anthropology, 9,
15–61.
EBERLE, U., ROHNER, D. and THOENIG, M. (2020), “Heat and Hate: Climate Security and Farmer-Herder Conflicts
in Africa” (CEPR Working Paper).
ESTEBAN, J., MAYORAL, L. and RAY, D. (2012), “Ethnicity and Conflict: An Empirical Study”, American Economic
Review, 102, 1310–1342.
FAO (2018), “Pastoralism in Africa’s Drylands: Reducing Risks, Addressing Vulnerability, and Enhancing Resilience”
Rome: Food and Agriculture Organization of the United Nations.
FEARON, J. D. and LAITIN, D. D. (2003), “Ethnicity, Insurgency, and Civil War”, American Political Science Review,
97, 75–90.
GENNAIOLI, N. and RAINER, I. (2007), “The Modern Impact of Precolonial Centralization in Africa”, Journal of
Economic Growth, 12, 185–234.
GROSJEAN, P. (2014), “A History of Violence: The Culture of Honor as a Determinant of Homicide in the U.S. South”,
Journal of the European Economic Association, 12, 1285–1316.
GUARISO, A. and ROGALL, T. (2017), “Rainfall Inequality, Political Power, and Ethnic Conflict in Africa” (Working
paper, University of British Columbia).
HARARI, M. and LA FERRARA, E. (2018), “Conflict, Climate, and Cells: A Disaggregated Analysis”, Review of
Economics and Statistics, 100, 594–608.
HEIN, L. (2006), “The Impacts of Grazing and Rainfall Variability on the Dynamics of a Sahelian Rangeland”, Journal
of Arid Environments, 64, 488–504.
HERRMANN, S. M. and MOHR, K. I. (2012), “A Continental-Scale Classification of Rainfall Seasonality Regimes in
Africa Based on Gridded Precipitation and Land Surface Temperature Products”, Journal of Applied Meteorology
and Climatology, 50, 2504–2513.
HSIANG, S. M. and BURKE, M. (2014), “Climate, Conflict, and Social Stability: What Does the Evidence Say?”,
Climate Change, 123, 39–55.
HSIANG, S. M., BURKE, M. and MIGUEL, E. (2013), “Quantifying the Influence of Climate on Human Conflict”,
Science, 341, 12353–67.
JACOBS, A. H. (1965), “African Pastoralists: Some General Remarks”, Anthropological Quarterly, 38, 144–154.
JIA, R. (2014), “Weather Shocks, Sweet Potatoes and Peasant Revolts in Historical China”, Economic Journal, 124,
92–118.
JOHNSON, T. M. and GRIM, B. J. (2021), “World Religion Database”.
KINCAIDE, L., MCGUIRK, E. and NUNN, N. (2020), “A Comprehensive Concordance between Murdock’s Map of
Ethnic Groups and the Ethnographic Atlas” (Mimeo, Harvard University).
KITCHELL, E., TURNER, M. D. and MCPEAK, J. G. (2014), “Mapping of Pastoral Corridors: Practices and Politics
in Eastern Senegal”, Pastoralism, 4, 17. https://doi.org/10.1186/s13570-014-0017-2.
KO, C. Y., KOYAMA, M. and SNG, T.-H. (2018), “Unified China and Divided Europe”, International Economic Review,
59, 285–327.
KONCZACKI, Z. A. (1978), The Economics of Pastoralism: A Case Study of Sub-Saharan Africa (London: Frank Cass
& Company Ltd.).
440 REVIEW OF ECONOMIC STUDIES

KÖNIG, M. D., ROHNER, D., THOENIG, M., et al. (2017), “Networks in Conflict: Theory and Evidence from the
Great War of Africa”, Econometrica, 85, 1093–1132.
LEWIS, I. M. (1961), A Pastoral Democracy: A Study of Pastoralism and Politics (Oxford: Oxford University Press).
LITTLE, P. D., SMITH, K., CELLARIUS, B. A., et al. (2001), “Avoiding Disaster: Diversification and Risk Management
Among East African Herders”, Economic Development and Cultural Change, 32, 401–433.
MANACORDA, M. and TESEI, A. (2020), “Liberation Technology: Mobile Phones and Political Mobilization”,
Econometrica, 88, 533–567.
MATTEE, A. Z. and SHEM, M. (2006), “Ambivalence and Contradiction: A Review of the Policy Environment in
Tanzania in Relation to Pastoralism” (Drylands Issue Paper No. 140. International Institute for Environment and
Development (IIED)).
MAYSTADT, J.-F. and ECKER, O. (2004), “Extreme Weather and Civil War: Does Drought Fuel Conflict in Somalia

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025

Through Livestock Price Shocks?”, American Journal of Agricultural Economics, 96, 1157–1182.
MCGUIRK, E. and BURKE, M. (2020), “The Economic Origins of Conflict in Africa”, Journal of Political Economy,
128, 3940–3997.
MCPEAK, J. G. and BARRETT, C. (2001), “Differential Risk Exposure and Stochastic Poverty Traps Among East
African Pastoralists”, American Journal of Agricultural Economics, 83, 674–679.
MICHALOPOULOS, S. and PAPAIOANNOU, E. (2013), “Precolonial Ethnic Institutions and Contemporary African
Development”, Econometrica, 81, 113–152.
MICHALOPOULOS, S. and PAPAIOANNOU, E. (2016), “The Long-Run Effects of the Scramble in Africa”, American
Economic Review, 106, 1802–1848.
MIGUEL, E., SATYANATH, S. and SERGENTI, E. (2004), “Economic Shocks and Civil Conflict: An Instrumental
Variables Approach”, Journal of Political Economy, 112, 725–753.
MOLES, A. T., PERKINS, S. E., LAFFAN, S. W., et al. (2014), “Temperature or Precipitation?”, Journal of Vegetation
Science, 25, 1167–1180.
MONTALVO, J. G. and REYNAL-QUEROL, M. (2005), “Ethnic Polarization, Potential Conflict, and Civil Wars”,
American Economic Review, 95, 796–816.
MORITZ, M. (2010), “Understanding Herder-Farmer Conflicts in West Africa: Outline of a Processual Approach”,
Human Organization, 69, 138–148.
MORTEN, M. (2019), “Temporary Migration and Endogenous Risk Sharing in Village India”, Journal of Political
Economy, 127, 1–46.
MOSCONA, J., NUNN, N. and ROBINSON, J. A. (2020), “Segmentary Lineage Organization and Conflict in Sub-
Saharan Africa”, Econometrica, 88, 1999–2036.
MUELLER, H. and ROHNER, D. (2018), “Can Power-Sharing Foster Peace? Evidence from Northern Ireland”,
Economic Policy, 33, 447–484.
MURDOCK, G. P. (1959), Africa: Its Peoples and Their Cultural History (New York: McGraw-Hill Book Company).
MURDOCK, G. P. (1967), Ethnographic Atlas (Pittsburgh: University of Pittsburgh Press).
NICHOLSON, S. E., FINK, A. H. and FUNK, C. (2018), “Assessing Recory and Change in West Africa’s Rainfall
Regime from a 161-Year Record”, International Journal of Climatology, 38, 3770–3786.
NISBETT, R. E. and COHEN, D. (1996), Culture of Honor: The Psychology of Violence in the South (Boulder: Westview
Press).
NORENZAYAN, A. (2013), Big Gods: How Religion Transformed Cooperation and Conflict (Princeton: Princeton
University Press).
PORTMANN, F. T., SIEBERT, S. and DÖLL, P. (2010), “MIRCA2000—Global Monthly Irrigated and Rainfed Crop
Areas Around the Year 2000: A New High-Resolution Data Set for Agricultural and Hydrological Modeling”, Global
Biogeochemical Cycles, 24, GB1011.
RALEIGH, C., LINKE, A., HEGRE, H., et al. (2010), “Introducing ACLED-Armed Conflict Location and Event Data”,
Journal of Peace Research, 47, 651–660.
ROHNER, D., THOENIG, M. and ZILIBOTTI, F. (2013), “War Signals: A Theory of Trade, Trust, and Conflict”, Review
of Economic Studies, 80, 1114–1147.
ROSS, M. L. (2004), “What Do We Know about Natural Resources and Civil War?”, Journal of Peace Research, 41,
337–356.
SCHNEIDER, U., BECKER, A., FINGER, P., et al. (2020), “GPCC Full Data Monthly Product Version 2020 at 0.5◦ :
Monthly Land-Surface Precipitation from Rain-Gauges built on GTS-based and Historical Data” (Dataset).
SHANAHAN, T. M., OVERPECK, J. T., ANCHUKAITIS, K. J., et al. (2009), “Atlantic Forcing of Persistent Drought
in West Africa”, Science, 324, 377–380.
SOLOW, A. R. (2013), “A Call for Peace on Climate and Conflict”, Nature, 497, 179–180.
STENNING, D. J. (1959), Savannah Nomads: A Study of the Wod̄aab̄e Pastoral Fulani of Western Bornu Province,
Northern Region, Nigeria (Oxford: Oxford University Press).
SUNDBERG, R. and MELANDER, E. (2013), “Introducing the UCDP Georeferenced Event Dataset”, Journal of Peace
Research, 50, 523–532.
TIERNEY, M. J., NIELSON, D. L., HAWKINS, D. G., et al. (2011), “More Dollars Than Sense: Refining Our
Knowledge of Development Finance Using AidData”, World Development, 39, 1891–1906.
UNEP-WCMC and IUCN (2021), “Protected Planet: The World Database on Protected Areas (WDPA) and The
World Database on Other Effective Area-based Conservation Measures (WD-OECM) [Online]” Available at:
www.protectedplanet.net (accessed 14 May 2021).
 McGuirk & Nunn THP, CLIMATE CHANGE, AND CONFLICT IN AFRICA 441

WAHA, K., MÜLLER, C. and ROLINSKI, S. (2013), “Separate and Combined Effects of Temperature and Precipitation
Change on Maize Yields in Sub-Saharan Africa for Mid- to Late-21st Century”, Global and Planetary Change, 106,
1–12.
World Bank (2021), “World Bank Commodity Price Data (The Pink Sheet),” https://www.worldbank.org/en/
research/commodity-markets. (accessed 18 March 2021).

Downloaded from https://academic.oup.com/restud/article/92/1/404/7627150 by guest on 21 January 2025