Electoral Studies 33 (2014) 220–232

Contents lists available at ScienceDirect

Electoral Studies
journal homepage: www.elsevier.com/locate/electstud

Poverty and vote buying: Survey-based evidence from Africa

Peter Sandholt Jensen a,1, Mogens K. Justesen b, *
a
Department of Business and Economics, University of Southern Denmark, Campusvej 55, DK-5230 Odense M, Denmark
b
Department of Business and Politics, Copenhagen Business School, Porcelaenshaven 1, DK-2000 Frederiksberg, Denmark

a r t i c l e i n f o a b s t r a c t

Article history: Alongside the spread of democracy in the developing world, vote buying has emerged as
Received 27 March 2013 an integral part of election campaigns. Yet, we know little about the causes of vote buying
Received in revised form 4 July 2013 in young democracies. In this paper, we analyse the sources of vote buying in sub-Saharan
Accepted 10 July 2013
African. Using data from the Afrobarometer, we focus on the impact of poverty on vote
buying at the individual- and country-level. Results from multilevel regressions show that
Keywords:
poor voters are significantly more likely to be targets of vote buying than wealthier voters.
Vote buying
This effect increases when elections are highly competitive. Thus, micro-level poverty
Elections
New democracies
seems to be an important source of vote buying in Africa and has major implications for
Poverty the way electoral democracy operates.
Africa Ó 2013 Elsevier Ltd. All rights reserved.
Afrobarometer

1. Introduction Ocantos et al., 2012). Thus, vote buying seems to be an in-

tegral feature of electoral politics in new democracies
Vote markets used to be widespread during the early across the world. In this paper we contribute to this liter-
stages of democracy in Western Europe, but largely seemed ature by providing new evidence on how vote markets
to disappear with economic development (Aidt and Jensen, operate in elections with de jure secret ballot, with a
2012). In recent decades, new waves of democratization particular focus on the relationship between poverty and
have occurred around the globe, bringing democracy well voters’ experience with being offered pre-election rewards
beyond the borders of the Western world. While these in return for votes.
transitions are usually celebrated, they have also lead to Vote buying is a particular form of political clientelism,
renewed debate about the operation of democracy in i.e. the direct exchange at the individual level of rewards
developing countries. Indeed, alongside transitions to de- and material goods by political patrons in return for elec-
mocracy, vote buying – understood as the direct exchange toral support by voters (Stokes, 2007a; Hicken, 2011; Linos,
of money or gifts for votes – has made a powerful return to 2013; Robinson and Verdier, 2013). It is widely accepted
the scene of democratic politics. Reports of vote buying that clientelistic politics create economic inefficiencies,
come from most regions of the world, including Asia reduce the supply of public goods, and bias public policy in
(Hicken, 2007), Africa (Bratton, 2008; Vicente and favour of elites (Stokes, 2007a; Vicente and Wantchekon,
Wantchekon, 2009), the Middle East (Blaydes, 2006), and 2009; Robinson and Verdier, 2013). Vote buying also rai-
Latin America (Brusco et al., 2004; Stokes, 2005; Gonzales- ses questions about the character of democracy. While
elections involving vote buying may be ‘free’ – allowing
voters to choose between multiple candidates in elections
* Corresponding author. Tel.: þ45 3815 3564.
E-mail addresses: psj@sam.sdu.dk (P.S. Jensen), mkj.dbp@cbs.dk
with universal suffrage – they collide with standards of
(M.K. Justesen). democratic ‘fairness’, because the interests of some voters
1
Tel.: þ45 6550 4472. are bought by parties before the election, and may

0261-3794/$ – see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.electstud.2013.07.020
 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232 221

therefore be ignored by political representatives after the cross-country examination of vote buying to date, and the
election (cf. Elklit and Svensson, 1997). According to Stokes first attempt to study vote buying across a larger set of
(2007b: 96), the undemocratic nature of vote buying arises African countries. Indeed, in a recent review of the litera-
precisely because ‘ . it keeps vote sellers from having ture, Stokes (2007a: 618) emphasizes that there is a gen-
their interests accurately interpreted and made known, eral lack of quantitative cross-country analyses of
and in that it makes them less autonomous than are the clientelism and vote buying. We start to fill this gap by
recipients of politically motivated public programs’. This analyzing micro-level data on vote buying from 18 coun-
tends to weaken or even reverse the accountability link tries in sub-Saharan African. Specifically, we use data from
between voters and politicians (Stokes, 2005, 2007b). the third round (2005–2006) of the Afrobarometer survey.
Despite its consequences, empirical evidence suggests This provides a unique source of data on vote buying in
that vote buying may contribute to increase electoral Africa, since it is so far the only round of the Afrobarometer
support (Wantchekon, 2003; Brusco et al., 2004; Blaydes, that contains information on voters’ experience with vote
2006; Vicente and Wantchekon, 2009). In new de- buying during elections.
mocracies in particular, parties often rely on existing The fact that our study draws on data for a broader set
patron-client networks and pre-election transfers to of African countries has at least two advantages. First,
mobilize support (Keefer, 2007). However, there are good while the existing literature contains analyses of clientel-
reasons to suspect that political parties do not distribute ism and vote buying in a number of African countries, such
their vote buying efforts randomly across the electorate. as Benin (Wantchekon, 2003; Koter, 2013), Ghana
Theoretical priors suggest that vote buying parties sys- (Lindberg, 2003), Sao Tome and Principe (Vicente, 2013),
tematically target specific groups in the electorate based Nigeria (Bratton, 2008), Kenya (Kramon, 2009), Malawi
on their socio-economic characteristics. Poverty in and Mozambique (Birch, 2011) and Senegal (Koter, 2013),
particular has been emphasized as an important source of using data for a larger set of countries allows for more
vote buying that enables political parties to exploit the general inferences. Second, since the data contain micro-
material needs of deprived voter groups by trading re- level survey information from 18 countries, we are able
wards for votes (Stokes, 2005, 2007a, 2007b; Blaydes, to take account of the fact that voters vary in their indi-
2006). vidual characteristics within countries and that cross-
In this paper, we contribute to the study of vote buying country differences may affect average levels of vote
by empirically examining how poverty affects the likeli- buying between countries. In this way, we can both take
hood that voters are targeted by vote buyers in a cross- into account the contextual factors that Birch (2011: 106–
section of African democracies. In doing so, we follow the 107) emphasizes as important and investigate the role of
tradition of the seminal contributions in political science by inherently interesting country characteristics such as
Scott (1969) and Lemarchand and Legg (1972), who ana- electoral competitiveness. We do so by both conducting a
lysed how modernization in new post-colonial states af- set of multilevel regressions and by controlling for country
fects – and ultimately undermines – patron-client fixed effects which capture the influence of country-level
networks in general and vote markets more specifically. variables.
More recently, these issues have also been addressed in The rest of the paper is organized as follows. The
formal models in political economy (Stokes, 2005; Aidt and following section provides a brief overview of voters’ ex-
Jensen, 2012; Robinson and Verdier, 2013). In this paper, we periences with being targets of vote buying in 18 African
focus on sub-Saharan African since democratic politics has countries, as well as qualitative accounts of vote buying in
only recently become the norm in this region (Van de Africa. After that, we briefly outline the theory relating
Walle, 2007). In fact, no less than 40 African countries poverty to vote buying. Next, we describe the econometric
held their first competitive election in a generation during model and data we use in the empirical part. The next
the early 1990s (Bratton and van de Walle, 1997). However, section presents empirical results from multilevel re-
while multiparty politics and competitive elections have gressions, and the final section concludes on the main
gained prominence, many African countries are yet to findings.
complete their democratic transitions, some have reverted
back into non-democracy, and still others are plagued by 2. Vote buying in Africa
violence and coups, as witnessed most recently in Mali.
Moreover – as we document later – vote buying is wide- The importance of vote markets in Africa is shown in
spread in many countries that have continued along the Fig. 1, which plots the percentage of voters who report
path towards democracy. being targeted by vote buyers during the most recent
We make two contributions to the literature. First, presidential or parliamentary elections for the 18 countries
while the relationship between vote buying and poverty in our sample.
has been studied in Nigeria (Bratton, 2008), Kenya Fig. 1 shows that vote buying is pervasive in Kenya,
(Kramon, 2009), and Malawi and Mozambique (Birch, Uganda, Benin, Madagascar, Nigeria, and Mali; less wide-
2011), this paper appears to be the first to provide a spread in a small group of countries including Senegal,
comprehensive analysis of how poverty affects individuals’ Tanzania and South Africa; and almost absent in Botswana
propensity of being targeted by vote buyers across African and Lesotho.
democracies, and, therefore, how vote markets operate Qualitative accounts of elections in 1990s and 2000s
during election campaigns in Africa. Second, to the best of corroborate that vote buying is a common electoral phe-
our knowledge, this study constitutes the most extensive nomenon in Kenya. Foeken and Dietz (2000) note that the
222 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232

Afrobarometer countries
42.5

40
35.4

Vote buying (percent)

33.4 32.6

30
28.6
26.8
20 23.5

16.7

12.1
10.3 9.4
10

8.5
6.8 6.3
5.4 5.1
1.8 1.2
0

r
ni li

ba ia

h ia

v a

o
ug ya

le na
da

za ria

ts e
se ibia

so nz l
ag in

m ue
bi a
m g we

na awi

ta g a
ca

th
pe ric
bo erd
am n
zi mb

ut an
n

m
n

a
ge
an

oz ha
b

ne
ad be

so
as

al
m

ca af
ke

w
m
m

Question: And during the [20xx] election, how often (if ever) did a candidate or someone from
a political party offer you something, like food or a gift, in return for your vote?

Fig. 1. Vote buying in Africa.

support groups Youth of KANU ’922 and Toroitich Till20003 politicians in Senegal spend almost 20 percent of
were highly active players during election campaigns campaign expenditures on vote buying – about half of
and undertook vote buying on a large scale, spending an what is used in Kenya.
estimated total of US$60 million on this purpose alone. Finally, Fig. 1 shows that vote buying seems uncommon
They also observe that the Kenyan Electoral Commission in Botswana as less than 2 percent of voters report
acknowledged that the 1997 election felt short of being being targeted by vote buyers. In fact, Botswana is often
‘fair’. For instance, in 13 percent of polling stations, secret considered the oldest, established democracy in sub-
voting was not guaranteed, and vote buying was quite Saharan Africa (Robinson and Parsons, 2006) with ‘. lit-
common on the day of the election. According to Mwangi tle repression and no exceptional allegations of fraud’
(2008: 272), KANU’s election victories in 1992 and 1997 (Alvarez et al., 1996: 10). These descriptions fit well with
were largely due to electoral corruption. Vote buying was Fig. 1.
also widespread during elections in 2002 and 2007, where These examples are obviously only illustrations of the
voters were simply being offered money upfront (Kramon, presence of – and differences between – vote markets in
2009). In line with this, Bryan and Baer (2005) report that Africa. However, they do support our data as being good
parties spend around 40 percent of their total campaign reflections of the magnitude of vote buying across coun-
expenditure on vote buying. The Kenyan example also tries in Africa.
suggests one obvious reason why vote buying may work in
an environment with secret ballot, which at least de jure 3. Poverty and vote buying
has been introduced in all countries in the Afrobarometer –
it is simply violated, as it also was historically in, e.g., The previous section shows that the magnitude of vote
France and Germany (Aidt and Jensen, 2012). buying varies significantly between countries, suggesting
Fig. 1 also shows that around 7 percent of voters report important differences in clientelistic electoral practices
experience with vote buying in Senegal. The presence of even within democracies located in the same geographical
vote buying in Senegal is supported by qualitative ac- region. At a very basic level, this raises the question of how
counts. Schaffer (1998) argues that even though the secret vote markets can operate in the presence of the secret
ballot was reintroduced in the 1990s, vote buying was still ballot – a mechanism that was designed precisely to allow
possible due to gerrymandering that made Senegalese voters to conceal their electoral choice from patrons and
constituencies smaller – making it easier for political political parties (Cox and Kousser, 1981; Heckelman, 1998;
party operatives to monitor deals made on vote markets. Aidt and Jensen, 2012). While we do not deal with this
More recently, Osei (2012, 261) noted that the ruling party issue in detail, we emphasize two points. First, although
(PDS) is still actively buying votes during national elec- the secret ballot allows voters to renege on commitments
tions. Similarly, Bryan and Baer (2005) report that to vote for a particular candidate, it does not necessarily
eliminate vote buying. Indeed, the literature has uncov-
ered a number of strategies and mechanisms that enable
2
KANU denotes the Kenya African National Union. parties to uphold a certain level of monitoring of how –
3
Toroitich is the middle name of former Kenyan president Moi. and if – people vote. These include party-issued ballots,
 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232 223

positioning of party agents inside polling stations, otherwise. Accordingly, Prðyij ¼ 1 aj ; Xij ; Zij Þ is the proba-
exploiting voters’ social network, paying opposition voters bility that a voter reports being targeted by a vote buyer
to abstain – or supporters to turn out, and outright during the most recent election. Model (1) is generalized
intimidation (Cox and Kousser, 1981; Heckelman, 1998; via the function G(.) and easily applicable to different
Stokes, 2005; Medina and Stokes, 2007; Nichter, 2008). probability models using appropriate link functions. Here
Second, data from round four of the Afrobarometer show we use logit and linear link functions:
that around 25 percent African voters believe it is ‘some-
1
what likely’ or ‘very likely’ that their vote choice can be G aj þ b1 Xij þ b2 Zij ¼ (1b)
ðaj þb1 Xij þb2 Zij Þ
1þe
monitored.4 This indicates that political parties in Africa
actively try to circumvent ballot secrecy to monitor voters’
electoral choices.

This raises the question of whether – and why – parties’ G aj þ b1 Xij þ b2 Zij ¼ aj þ b1 Xij þ b2 Zij (1c)
vote buying campaigns target particular voter groups. One The X variable includes individual-level measures of
of the key determinants of vote buying identified in the poverty. The variables in Z are individual-level controls. The
literature is poverty. There are a number of reasons why variables are described below.
vote buying thrives in the context of widespread poverty. In (1a), aj is a group-specific constant that varies across
Firstly, poor voters generally lack access to resources – e.g. countries, and can be treated as random or fixed. In the
food, clean water, and medical care – that politicians can random effects model, aj is modelled as a function of
promise to deliver during election campaigns. As empha- country-level variables, as in Equation (2).
sized by Scott (1969), this paves the way for clientelistic
relationships between voters and politicians and the use of aj ¼ m þ gCj þ hj (2)
pre-election rewards as a means to mobilize electoral
support among the poor. Secondly, on the assumption that aj comprises a common intercept, m, a non-random
the marginal utility of income is higher for poor groups part explained by the country-level variables, C, and a
(Dixit and Londregan, 1996; Stokes, 2005), the utility country-specific random effect, hj. C includes country-
derived from selling one’s vote is higher for poor people. level variables such as GDP and urbanization. Substitut-
Selling one’s vote for material rewards may therefore be a ing (2) into (1a) we obtain the random effects multilevel
rational course of action for people living in material model (3).
deprivation (Blaydes, 2006).5 Thirdly, from the perspective


of political parties, votes at the bottom of the income dis- Pr yij ¼ 1 aj ; Xij; Zij ¼ G m þ gCj þ hj þ b1 Xij þ b2 Zij (3)
tribution are cheaper to buy (Dixit and Londregan, 1996; The fixed effects model, which removes bias due to
Blaydes, 2006; Stokes, 2007b: 119). Parties can therefore omitted variables at the country-level, is used as an alter-
buy more votes among the poor by offering even relatively native. This model treats the aj’s in (1) as parameters to be
modest amounts to each voter. For instance, Bratton (2008) estimated. While fixed effects may be approximated by
reports that during Nigeria’s 2007 elections the most observable variables, it is arguably hard to control for all
common amount of money offered to voters was US$4. such country-specific effects. For example, Acemoglu et al.
These economic mechanisms are likely to make poor voters (2008) argue that whether countries embarked on a path
the prime targets of vote buying by clientelistic parties who towards electoral democracy and economic development
want to maximize their (re)election chances. We therefore was affected by critical junctures in history. Thus, the
expect that political parties are more likely to target and country-level differences in vote buying shown in Fig. 1
buy the votes of poor people. This is the hypothesis we test may have deep historical roots that country fixed effects
in the remainder of the paper. enable us to control for.

4. Econometric model 5. Data

Our econometric approach uses a multilevel probability Our data come from the Afrobarometer survey, which is a
model for vote buying (Rabe-Hesketh and Skrondal, 2008). standard source of information on voter attitudes and ex-
In the multilevel model with i, ., N individuals nested periences with democracy in Africa (Bratton et al., 2005;
within j, ., J countries, the individual-level equation is: Bratton, 2008; Justesen, 2011; Justesen and Bjørnskov,


2012).6 Summary statistics for the data are available in
Pr yij ¼ 1 aj ; Xij ; Zij ¼ G aj þ b1 Xij þ b2 Zij (1a) the online appendix. We use the third round of the survey –
In Equation (1), yij is a dummy variable equal to 1 if the collected in 2005 and 2006.7
individual has been approached by a vote buyer and zero The data are gathered based on a stratified random
sampling procedure, generating a largely representative
sample of adult individuals in all countries (Bratton et al.,
4
These data are not available in the third round.
5
This requires that voters are paid enough to be indifferent between
6
selling their vote and receiving the benefit from having their preferred An Online Appendix and replication data are available on https://
party in power (see, e.g., Aidt and Jensen, 2012). For parties, this may sites.google.com/site/mkjustesen/.
7
imply that vote buying trumps offering solid redistributive programs to Details on the Afrobarometer methodology are available in Bratton
poor voters. et al. (2005) and on http://afrobarometer.org/.
224 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232

2005.). The sample size is 1200, except in three highly an ordinal measure of vote buying. This allows us to
fractionalized countries – Nigeria, South Africa, and measure individuals’ experience with being approached by
Uganda – where it is 2400. The questionnaire consists of a vote buyers only in a relatively crude manner. In particular,
standardized set of questions, making the data compa- we are not able to measure how much money people are
rable across countries. The interviews were conducted offered in return for their votes or what the monetary
face-to-face, in the local language, and by people outside value of gifts offered might be. With this caveat, we are
the local community. This ensures that interviews are nonetheless able to establish a measure of the prevalence
conducted with anonymity, and that respondents can of vote buying both within and across African democracies.
answer questions without fear of social repercussions Fig. 2a shows that 18 percent of Africans report having
from the local community, even on potentially contro- been offered gifts or money in return for their vote, albeit
versial issues like vote buying. Finally, while individual with considerable differences between countries (cf.
respondents are randomly selected within countries, the Fig. 1).
countries in which the surveys are conducted are not. The Nevertheless, most respondents report no experience
Afrobarometer countries are selected because they are with being targeted by vote buyers. Therefore we also
minimally democratic and not part of armed conflicts construct a dichotomous variable – shown in Fig. 2b –
(Bratton et al., 2005).8 In the third round, the only country which simply distinguishes people who have been offered
that may fall short of being an electoral democracy is rewards in return for their votes (1) from people who
Zimbabwe. However, as shown below, our results are have no experience in this regard (0). These measures are
entirely robust to excluding Zimbabwe. Otherwise, the used as dependent variables in our estimation of Equation
Afrobarometer countries are similar to the sub-Saharan (1).
average on a range of socio-economic indicators
(Justesen, 2011: 8).
5.2. Measuring poverty

5.1. Measuring vote buying

To measure individual-level differences in poverty, we
use survey questions from the Afrobarometer to construct
We construct our vote buying measure from the ques-
what the literature often refers to as the Index of Lived
tion (Q57F): “And during the [20xx] election, how often (if
Poverty (Bratton et al., 2005; Bratton, 2008; Mattes, 2008;
ever) did a candidate or someone from a political party
Justesen, 2011; Justesen and Bjørnskov, 2012). The index
offer you something, like food or a gift, in return for your
assesses individual-level poverty by asking respondents
vote?” The question refers to the most recent national
about their experiences with lacking access to basic
presidential or parliamentary election in all 18 countries.
household necessities like food and clean water (Bratton,
While the question is part of a larger battery of survey in-
2008: 31–32; Poku and Mdee, 2011: 54). Specifically, the
struments concerning people’s experiences with paying
survey contains five questions asking respondents how
bribes to government officials in return for public services,
often during the past year, they or anyone in their family
it is distinct in asking respondent about their experience
have gone without: a) enough food to eat, b) enough
with being offered rewards (rather than offering bribes
clean water for home use, c) medicines or medical
themselves) by political candidates or party activist (rather
treatment, d) enough fuel to cook food, e) a cash income.9
than to government officials) in return for their vote (rather
Respondents’ answers are coded on a five-point scale
than in return for a public service). The question uncovers
from “never” to “always”. Following the standard in the
the property of electoral politics we are interested in: The
literature, we combine these five items into an aggregate
use and frequency of vote buying by political parties.
index of experienced poverty, and normalize it to vary
However, it is important to note that the question concerns
between 0 and 1, with high values indicating that peo-
intended vote buying by parties. That is, the question does
ple live in a state of severe poverty and are regularly
not allow us to measure whether voters actually accepted
deprived of access to basic material necessities (Bratton
electoral bribes. But it does allow us to measure which
et al., 2005; Bratton, 2008; Mattes, 2008; Justesen and
voters political parties tend to target during vote buying
Bjørnskov, 2012). Low values reflect that people live in
campaigns – and that is precisely what we want to
good material conditions and do not suffer from depri-
examine.
vation due to lack of basic household necessities. At the
Respondents can answer using the five categories
country-level, we include GDP per capita to capture the
‘Never’, ‘No experience with this in the past year’, ‘Once or
impact of wealth and poverty on vote buying at the
twice’, ‘A few times’, or ‘Often’. We merge the two first
macro-level (Aidt and Jensen, 2012). This is important
categories and code them as zero. The remaining cate-
because the vote market may diminish or break down
gories are given the codes 1, 2, and 3, respectively, giving us
once the share of poor voters at the macro level reaches a
low level. Data are from the Penn World Tables 7.0
8 (Heston et al., 2011).
According to Freedom House (http://www.freedomhouse.org), all
countries except Zimbabwe are classified as ‘Free’ or ‘Partly Free’ on po-
litical rights and civil liberties. According to the regime classification of
9
Cheibub et al. (2010) – requiring that government power has alternated We do not use question Q8F (concerning lack of school expenses for
following elections – only nine countries were democracies in 2005. children), because it is not clear how people without children should be
These are Benin, Cape Verde, Ghana, Kenya, Madagascar, Malawi, Mali, classified in the index. However, our results are robust to including this
Nigeria and Senegal. item in the poverty index.
 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232 225

80
Vote buying (percent)
60
40
20
0

)
)

(1

(3
(2
(0

n
es
er

ic

fte
ev

tim
tw

O
N

or

w
fe
e
nc

A
O

Vote buying: Four-point scale

Question: And during the [20xx] election, how often (if ever) did a candidate or someone from
a political party offer you something, like food or a gift, in return for your vote?

b
80
Vote buying (percent)
60
40
20
0

)
(1
(0
d

d
re
re
fe

fe
of

of
s

ds
d

ar
ar
w

ew
re

R
o
N

Vote buying: Dichotomous

Question: And during the [20xx] election, how often (if ever) did a candidate or someone from
a political party offer you something, like food or a gift, in return for your vote?
('No rewards offered=category 'Never')
(Rewards offered=categories 'Once or twice', 'A few times', and 'Often')

Fig. 2. a). Vote buying in Africa: Four-point scale. b). Vote buying in Africa: dichotomous variable.

5.3. Control variables (Stokes, 2005, 2007b).10 We therefore include four vari-
ables (Q32A–Q32D), asking respondents how often during
To alleviate problems of confounding caused by omitted the past year they have contacted: a) a local government
variables, we include a number of individual- and country- councillor, b) a member of parliament, c) a government
level control variables that may be correlated with both ministry official, or d) a political party official, to solve a
vote buying and the processes causing selection into
poverty. At the individual-level, we include up to 13 control
variables. Most importantly, the chances that people are 10
However, Bratton et al. (2005: 151–153) find that people in Africa are
exposed to vote buying attempts may be a function simply more likely to address their grievances to traditional/religious leaders in
of being in contact with political officials or party activists the community rather than to political officials.
226 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232

problem or express personal opinions. This should ensure identify themselves with their nation (e.g. Ghana or Kenya)
that we capture possible effects on vote buying arising from or their ethnic group (Q82). High values indicate strong
respondents’ propensity to contact people affiliated with ethnic identity.11
political parties and candidates. Respondents can answer At the country-level, we include a measure of urbani-
the four questions on a four-point scale, which we trans- zation (the proportion of the population living in urban
form into dummy variables, coded as 1 for people who have areas) and secondary school enrolment rates. Data are from
been in contact with political officials in one of the four the World Development Indicators. However, since school
domains. However, since these questions were not asked in enrolment correlates with GDP per capita at 0.94
Zimbabwe, we also run regressions excluding the four (p < 0.001) and with urbanization at 0.73 (p < 0.001), we
variables. This allows us to check that our results are not include urbanization in most models and replace it by
sensitive to including Zimbabwe – the least democratic of school enrolment in some models to check that the results
the Afrobarometer countries – in the regressions. are not sensitive to this choice. We also account for
Following debates about political machines’ propensity ethnicity by including a measure of nation-wide ethnic
to offer rewards to opponents or supporters (Stokes, 2005; fractionalization (Alesina et al., 2003), which politicians
Nichter, 2008), we also control for whether voters identify may exploit to make ethnic identity a salient political issue
with a political party, using a dummy variable coded as one (Lieberman and Singh, 2012) or to privilege members of
for people who “feel close to a particular political party”, their own ethnic group using vote buying and other cli-
and zero otherwise (Q85). However, the possibility of entelistic policies (Van de Walle, 2007). Therefore, country-
reverse causality is a major concern in the relationship level ethnic fragmentation may be an important pre-
between vote buying and party identification, since it is condition for vote markets to operate along ethnic lines.
plausible that voters say they identify with a particular Some models also account for the degree of inequality
party because the party distributes rewards to voters. We using data from Solt (2009) as a proxy for economic frac-
therefore control for party identification only in some tionalization among voters that parties can exploit.
models. Finally, electoral competitiveness may affect nation-
We also include five socio-economic controls that may wide levels of vote buying (Kitschelt and Wilkinson,
correlate with vote buying and poverty. First, differences in 2007; Van de Walle, 2007; Ferree and Long, 2011). If elec-
voters’ urban-rural residence may influence the potential tions are close – and therefore highly competitive –
for vote buying (Lehoucq, 2007; Hicken, 2007). We there- candidates may resort to vote buying strategies to maxi-
fore include a dummy variable indicating whether re- mize the chances of winning. The relationship between
spondents reside in urban (1) or rural (0) areas. Moreover, competitiveness and vote buying can also appear in more
controlling for urban residence is important since rural subtle ways. For instance, in the context of Indian elections,
areas tend to be characterized by higher levels of poverty Aidt et al. (2011) argue that candidates with a criminal
than urban areas (Poku and Mdee, 2011: 54). Second, ed- record are more likely to appear on the ballot in constitu-
ucation may undermine the ability of parties to mobilize encies where the election is highly contested, arguably
support by buying votes because well-educated voters may because such candidates may deter opposition voters from
be more likely to break traditional clientelistic bonds voting – a form of ‘deflationary’ vote buying (Cox and
(Sisson, 1972). To measure education, we include three Kousser, 1981; Heckelman, 1998). To account for electoral
dummy variables measuring whether respondents have a competitiveness, we follow Eifert et al. (2010) and Ferree
primary, secondary, or tertiary education, where re- and Long (2011) and calculate the vote margin at the
spondents with no formal schooling are the reference most recent election, measured as the difference between
group. Third, a dummy variable distinguishes people the vote share of the candidate receiving most votes and
earning wages through formal employment (1) from peo- the candidate receiving second-most votes. When this
ple who do not (0). This is an important control since difference is small, elections tend to be close, and we expect
formal employment is more widespread in urban areas and higher levels of nation-wide vote buying. The magnitude of
may also cause selection into poverty. Fourth, we control electoral districts – i.e. the number of political candidates
for the respondents’ gender to capture bias in the tendency elected per electoral district – may also affect the electoral
of vote-buyers to target men. For similar reasons, we con- strategies of political parties. For instance, Persson et al.
trol for the age of respondents. Finally, ethnic identities are (2003) argue that small electoral districts – e.g. single-
often claimed to form an important basis for voting de- member constituencies – raise barriers to entry into office
cisions in African elections (Chandra, 2007; Kitschelt and for political candidates. By raising barriers to entry, small
Wilkinson, 2007; Van de Walle, 2007; Bratton et al.,
2012). For instance, Chandra (2007) suggests that voters
and politicians tend to build coalitions according to ethnic 11
It is also possible that belonging to – rather than identifying with – a
group affiliations because of beliefs that this will increase
specific ethnic group matters. To test this, we have also controlled for
rewards for voters and re-election chances for politicians. ethnic group affiliation by including a full set of dummy variables for
Vote buyers may therefore be more likely to target voters respondents’ ethnic/tribal groups (Q79). This should capture the impact
who identify with an ethnic group, just as voters identi- of confounding caused by (hardwired) ethnic group affiliations rather
fying with an ethnic group may be more likely to partici- than (socially constructed) ethnic identities. This approach also captures
possible effects of being from the same (or a different) ethnic group as the
pate in elections, making them more exposed to vote president/party in power (Bratton et al., 2012). Results (available upon
buying attempts. We therefore control for the salience of request) show that including the ethnic group indicators do not affect the
ethnic identities, measured as respondents’ tendency to impact of poverty on vote buying.
 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232 227

electoral districts may also create incentives for political variables are not a major source of bias in the estimates. In
candidates to distribute pre-election rewards to voters in models 7–8, we control for party identification. This has
return for their votes. In a related study, Jacobs and little effect on the results. Models 7–8 also replace GDP per
Spierings (2010) argue that politicians are more likely to capita and urbanization with secondary school enrolment
use clientelist strategies in small electoral districts because rates.13 While the impact of cross-country differences in
voters can better identity the politicians who deliver the schooling is positive, it is significant only in the ordered
benefits. We therefore expect that vote buying is more logistic regressions (model 8) and has no effect on the
common when electoral districts are smaller.12 Some impact or significance of the individual-level indicator of
models consequently control for mean district magnitude, poverty.
using data from the Database of Political Institutions (Beck Models 9 and 10 control for country-level inequality,
et al., 2001). which has little impact on the results. Finally, in models 11
and 12 we run country fixed effects models using linear and
6. Results binary (conditional) logistic regressions. While the fixed
effects transformation precludes the estimation of country-
Table 1 presents results from 12 regressions with vote level variables by construction, the advantage is that it al-
buying as the dependent variable. Models 1–10 show re- lows us to test the robustness of individual-level regressors
sults from random effects multilevel regressions and to the impact of omitted variables at the country-level.
models 11–12 use fixed effects estimators. Given the char- However, it is clear that including country fixed effects
acter of the dependent variables, we supplement binary has little effect on the main result, namely that individual-
logistic regressions and linear regression with ordered lo- level poverty has a positive and highly significant effect
gistic regressions. on voters’ exposure to vote buying efforts by political op-
eratives. Overall, these results suggest that poverty – un-
derstood as lack of regular access to basic household
6.1. Main results necessities – has a robust effect on vote buying in Africa.
Indeed, poverty appears to be one of the most important
Models 1–3 analyze data for all countries – including sources of differences in exposure to vote buying. This
Zimbabwe. In these models, poverty has a positive and result corroborates the evidence from single-country
highly significant effect on people’s propensity to experi- studies (Bratton, 2008; Kramon, 2009; Birch, 2011), and
ence vote buying. These results suggest that differences in suggests that the relationship applies more generally.14
poverty are significantly related to vote buying at the The substantial effect of poverty on vote buying is not
micro-level, with poor groups being much more likely only statistically significant, it is also quite large. We illus-
targets of vote buying than wealthier groups. However, at trate this in two ways. In the logistic regressions in models
the country-level, GDP per capita does not display a 5–12, the log odds are fairly stable and vary between 0.67
particularly robust relationship with vote buying. Although and 0.72. Based on model 5, this corresponds to an esti-
there are some indications that poorer countries have mated effect on the odds ratio of e0.67 ¼ 1.95. Substantively,
higher average-levels of vote buying, the coefficients are this means that a change from the lowest to the highest
insignificant in most models. While this result may seem value on the poverty index doubles the odds of being
puzzling, part of the explanation is arguably the limited exposed to vote buying during election campaigns. All
variation in GDP per capita in the group of African coun- other things being equal, a poor voter is therefore twice as
tries we examine, with only a couple of countries likely to be targeted by vote buyers as a voter who is
(Botswana and South Africa) being comparatively richer materially well-off. To get a more firm sense of the effect of
than the rest. This means that, by construction, a correlate micro-level poverty on vote buying, Fig. 3 plots predicted
of our sample is a relatively low level of economic devel- probabilities of being targeted by vote buying political
opment. Moreover, across the 18 countries in the sample, operatives during elections (based on model 5).
poverty is still so widespread that the share of poor voters Fig. 3 shows predicted probabilities of vote buying for
at the macro level is large enough for vote markets to different values of micro-level poverty and at fixed values
operate, in spite of differences in average levels of GDP per of the remaining variables, with the random effects set
capita. to zero. Since model 5 is a non-linear probability model, the
In models 4–12 we include variables measuring predicted probabilities depend on the values at which
ethnic identity and respondents’ contact with representa- the explanatory variables are evaluated. Specifically, we
tives affiliated with political parties and candidates. In ef- plot probabilities for a male person, who is 37 years old,
fect, data from Zimbabwe are excluded. In models 4–6, employed, lives in an urban area, has completed secondary
individual-level poverty continues to have a positive and school, and who has values of ethnic salience and contact
highly significant impact on vote buying, even though the with political officials corresponding to the sample average.
magnitude of the coefficient decreases a little. While only
indicative, this increases our confidence that omitted

13
Results from linear regressions give similar results.
12 14
A possible concern with using electoral competitiveness to explain In other models (not shown) we have looked at the interaction of
vote buying is that parties’ attempts to buy votes may also make elections poverty and urban-rural residence. While poverty has a highly significant
closer. The potential for reverse causality is much less when we use effect in both urban and rural areas, the effect is slightly stronger in urban
electoral district magnitude. areas.
Table 1

228
Poverty and vote Buying in Africa: estimates from multilevel regressions.

Model 1 2 3 4 5 6 7 8 9 10 11 12
Method ML linear RE Logit RE Ologit RE ML linear RE Logit RE Ologit RE Logit RE Ologit RE Logit RE Ologit RE Linear FE Logit FE
Individual-level variables
Poverty 0.27*** 0.86*** 0.91*** 0.23*** 0.67*** 0.72*** 0.70*** 0.71*** 0.67*** 0.68*** 0.23*** 0.67***
(4.98) (10.71) (5.28) (4.20) (7.84) (4.68) (8.00) (5.40) (7.84) (4.75) (4.22) (4.48)
Local councillor contact 0.11*** 0.35*** 0.31*** 0.33*** 0.32*** 0.35*** 0.36*** 0.11*** 0.35***
(4.18) (7.26) (7.43) (6.81) (6.47) (7.26) (9.46) (4.14) (9.28)
MP contact 0.03 0.06 0.08 0.05 0.08 0.06 0.07 0.03 0.07
(0.99) (0.98) (1.55) (0.80) (1.44) (0.98) (1.32) (1.00) (1.31)
Bureaucracy contact 0.00 0.05 0.02 0.06 0.06 0.05 0.03 0.00 0.05
(0.08) (0.68) (0.29) (0.93) (0.79) (0.68) (0.44) (0.08) (0.59)
Political party contact 0.18*** 0.65*** 0.63*** 0.61*** 0.58*** 0.65*** 0.63*** 0.18*** 0.65***
(4.18) (10.86) (6.99) (10.01) (6.09) (10.88) (6.71) (4.20) (6.70)
Party identification 0.24*** 0.21*

P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232

(5.65) (2.00)
Urban 0.00 0.00 0.01 0.01 0.01 0.00 0.03 0.03 0.01 0.02 0.01 0.01
(0.14) (0.08) (0.17) (0.28) (0.22) (0.05) (0.76) (0.33) (0.24) (0.29) (0.27) (0.11)
Primary education 0.05 0.17** 0.16** 0.04 0.13* 0.17*** 0.12* 0.14** 0.13* 0.12* 0.04 0.14*
(1.94) (3.02) (2.34) (1.75) (2.28) (3.34) (2.00) (2.99) (2.23) (2.03) (1.75) (2.53)
Secondary education 0.08*** 0.25*** 0.26*** 0.06** 0.14* 0.19*** 0.12 0.15** 0.14* 0.14** 0.06** 0.14**
(3.39) (4.08) (4.06) (2.86) (2.20) (4.44) (1.84) (2.58) (2.12) (2.66) (2.81) (2.74)
Tertiary education 0.13** 0.33*** 0.36*** 0.08 0.14 0.21* 0.13 0.16 0.14 0.16 0.08* 0.15
(2.98) (4.34) (3.43) (1.94) (1.76) (2.42) (1.60) (1.82) (1.69) (1.47) (1.96) (1.31)
Employment 0.04 0.19*** 0.19 0.03 0.14*** 0.12 0.13** 0.13 0.14*** 0.13 0.03 0.14
(1.09) (4.93) (1.88) (0.78) (3.51) (1.21) (3.11) (1.44) (3.52) (1.17) (0.74) (1.35)
Gender 0.06*** 0.16*** 0.18*** 0.04* 0.08* 0.10* 0.07 0.09 0.08* 0.10* 0.04* 0.08
(3.47) (4.34) (4.37) (2.42) (2.09) (2.12) (1.81) (1.64) (2.10) (2.15) (2.43) (1.73)
Age 0.00 0.00* 0.00 0.00** 0.01*** 0.01*** 0.01*** 0.01*** 0.01*** 0.01*** 0.00** 0.01***
(0.88) (2.21) (1.91) (2.66) (4.09) (3.73) (4.24) (3.57) (4.10) (4.25) (2.65) (3.68)
Ethnic identity 0.02 0.05** 0.04 0.07*** 0.05 0.05** 0.05 0.02 0.05*
(1.54) (3.01) (1.41) (3.87) (1.66) (2.99) (1.82) (1.52) (2.00)
Country-level variables
GDP/cap. 0.01 0.10 0.14*** 0.02 0.02 0.02
(0.30) (1.18) (17.25) (0.95) (0.26) (1.06)
Urbanization 0.01 0.00 0.00 0.01 0.01 0.00 0.01 0.01***
(1.18) (0.19) (0.92) (1.30) (0.40) (1.84) (1.32) (7.81)
Schooling – – – – – – 0.00 0.01***
(0.81) (5.34)
Ethnic fractionalization 0.89*** 4.52*** 4.76*** 1.04*** 5.54*** 6.71*** 5.87*** 6.25*** 5.78*** 4.97***
(5.92) (5.80) (33.41) (5.91) (6.88) (22.21) (8.41) (18.58) (7.76) (21.89)
Vote margin 0.50* 1.58 1.62*** 0.54* 2.05** 2.35*** 1.81** 2.07*** 2.65*** 1.74***
(2.02) (1.92) (18.43) (2.25) (2.76) (13.77) (3.14) (9.51) (3.75) (13.03)
Mean district magnitude 0.02* 0.02***
(2.35) (5.42)
Inequality 0.02 0.01**
(1.30) (2.87)
Constant 0.08 4.61*** 0.25 5.45*** 6.14*** 6.15*** 0.20***
(0.41) (5.96) (1.15) (7.35) (10.32) (6.82) (7.07)
Cut 1 4.54*** 6.20*** 6.29*** 4.60***
(28.03) (25.72) (21.79) (16.30)
 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232 229

Fig. 3 shows three lines. The bold line shows predicted

dependent variable. Low values on the poverty variable means that people are well-off; high values denote poverty. All results are generated in Stata 12. ML Linear RE denotes linear maximum likelihood models

command. In the binary and ordered logistic models, the coefficient are log odds, and the level-1 standard deviation follows the logistic distribution and is defined as O(p2/3) (Rabe-Hesketh and Skrondal 2008: 256-
257). Cut 1–3 are the cut points for the ordered logistic regressions. In models 1, 3, 4, 6, 8, 10, 11, 12 standard errors are robust and clustered by country. Absolute value of z-statistics are shown in parentheses. ***
Linear models (1, 4, 11) and ordered logistic models (3, 6, 8, 10) use the four-point scale of vote buying as dependent variable. Binary logistic models (2, 5, 7, 9, 12) use the dichotomous measure of vote buying as

with group random effects, estimated using the xtmixed command. Logit RE denotes random effects logistic regression, estimated using the xtmelogit command. Ologit RE denotes random effects ordered logistic
regression, estimated using the gllamm command. Linear FE are fixed effects regressions, implemented using the xtreg, fe. Logit FE is conditional (fixed effects) logistic regression, implemented using the clog
8831.9
Logit FE probabilities for an individual living in a median country.15

21,464
The punctuated line shows the equivalent plot with the
12

17
electoral vote margin set to zero. All else equal, this shows
the effect of poverty on vote buying in countries where
Linear FE

elections are narrow and extremely contested – in the

21,464
0.253
0.781
0.095
limit approximating a difference of zero between the vote
11

17
–
shares of the winning candidate and the runner-up.
13358.5
Finally, the dotted line shows the relationship between
Ologit RE

poverty and vote buying when the winning party wins by

O(p2/3)
(17.82)

(21.37)
5.29***

6.08***

0.0795

21,464
a wide vote margin, as is typically the case in, e.g.,
0.396
10

17
Botswana.
With values of the country-level variables fixed at their
median, the bold line shows that the probability of being
8923.7
Logit RE

O(p2/3)

21,464

offered rewards for votes is 14 percent for people who are

0.449

0.100

well-off – those with the lowest values on the poverty

17
9

index. This probability increases to 24 percent as poverty

increases to its highest value – a difference of 10 per-
12752.8
Ologit RE

O(p2/3)
(23.49)

(26.91)

centage points. This illustrates that the likelihood of being

6.98***

7.77***

20,615
0.392

0.078

offered money in return for votes varies markedly with

17
8

poverty. Moreover, the competitiveness of elections ag-

gravates this difference. The correlation between vote
8523.5
Logit RE

O(p2/3)

buying and the competitiveness of elections is consistent

20,615
0.406

0.083

with arguments in the literature emphasizing that cli-

17
7

entelism is an important instrument for parties competing

in democratic elections (Kitschelt and Wilkinson, 2007;
13359.7
Ologit RE

Van de Walle, 2007). Indeed, it is not surprising that

O(p2/3)
(28.98)

(33.54)
6.89***

7.68***

21,464
0.445

0.098

vote buying is an integral part of closely contested elec-

17

tions in Africa, given that political power in most African

6

countries is highly concentrated, mostly with strong

presidents. Presidential elections in particular are zero-
8924.5
Logit RE

O(p2/3)

21,464
0.471

0.109

sum games where ‘the winner takes all’. When elections

17

become highly contested, there are clear incentives for

5

politicians to use party operatives and clientelistic net-

ML linear RE

works to launch vote buying campaigns. So while highly

24139.4

competitive elections may make politicians and govern-

21,464

ments more responsive to voter demands (Hobolt and

0.139
0.779
0.031

17

Klemmensen, 2008), they also have a dark side, creating

4

strong incentives for parties and candidates to use pre-

election vote buying – and not just promises of post-
14722.2
Ologit RE

O(p2/3)

election redistributive transfers – as a means to mobilize

(31.80)

(41.98)
5.21***

5.99***

24,316
0.327

0.056

voter support, at least in weakly institutionalized and

18
3

newly established democracies.

The punctuated and dotted lines in Fig. 3 illustrate the
9954.5
Logit RE

O(p2/3)

impact of variations in electoral competitiveness. When

24,316
0.547

0.142

elections are closely contested – the punctuated line – the

18
2

probability of being offered money for votes increases to

23 percent even for voters living in good material condi-
ML linear RE

tions, while the corresponding probability for poor voters

28125.2

increases to a 37 percent – amounting to a 14 percentage

24,316
0.143
0.768
0.034

point difference between the most well-off and poorest

18
1

voter groups. In contrast, when electoral competition is

p<0.001, ** p<0.01, * p<0.05.
r (intra-class correlation)
Log (pseudo)likelihood

15
We use the median because the mean of GDP/cap. ($2230) is almost
Table 1 (continued )

twice as large as its median value ($1172) – the value of Kenya in 2005.
Level-2 std. dev.
Level-1 std. dev.
Random effects

This is mainly due to comparatively high levels of GDP in Botswana

Observations

($9306) and South Africa ($6961). This makes the mean a relatively poor
Countries

representative of the ‘typical’ value of GDP/cap. in the sample. In fact, no

Method
Model

country comes close to the mean of GDP/cap., while several countries

Cut 2

Cut 3

(Benin, Ghana, Lesotho, Uganda, and Zambia) have levels of GDP/cap.

close to the median.
230 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232

stronger sense of party identification – we do not go into

more detail about the issue here.17
Living in urban or rural areas is not related to vote
buying. That is, vote markets seem to operate in both urban
and rural areas. The education variables show that people
who have completed some level of schooling are more
likely to report experiences with vote buying than people
without formal schooling, suggesting – somewhat sur-
prisingly – that education does not erode vote buying.
However, we note that this finding is similar to results from
Nicaragua reported by Gonzales-Ocantos et al. (2012) and
Malawi reported by Birch (2011). Birch (2011: 103–105)
suggests that this reflects that vote buyers are more likely
to trust deals made with better educated people, while
Fig. 3. Poverty and vote buying: predicted probabilities. Gonzales-Ocantos et al. (2012: 212–214) argue that the
positive correlation may arise because better educated
people are more likely to participate in politics, making
low – the dotted line – voters are much less likely to be
them more disposed to experience vote buying attempts by
offered rewards in return for their votes – 7 and 13
political parties.18 The positive correlation could also reflect
percent, respectively, for well-off and poor voters. There-
an education-induced social desirability bias in reported
fore competitive elections not only affect average levels of
vote buying attempts. For instance, the literature on elec-
vote buying across countries, they also amplify the general
toral turnout shows that better educated people are more
tendency of political operatives to target and mobilize poor
likely to overreport voting at elections in order to conform
voters by trading money and material rewards for votes.
to norms of ‘good’ behaviour in democratic societies (Karp
Consequently, vote markets seem to thrive when elections
and Brockington, 2005). While this cannot be ruled out, we
are close, making poor and deprived voter groups even
have replicated the results in Table 1 in regressions that
more likely targets of vote buying campaigns by political
exclude respondents with 1) tertiary education, 2) sec-
parties.
ondary and tertiary education, and 3) primary, secondary,
and tertiary education. The latter reduces the sample to
6.2. Results for control variables
people without formal schooling. This does not change the
results for poverty. Detailed results are available in the
The results for the control variables largely correspond
Appendix.
to theoretical priors. Among the individual-level controls,
The findings for the other socio-economic controls
being in contact with political party operatives increases
show that employment increases the likelihood of expe-
the chance that people report experiences with vote
riencing vote buying, and that men and young people are
buying. While there is little to suggest that government
significantly more likely to experience vote buying. Finally,
bureaucrats are involved in vote buying, encounters with
ethnic identity is positively related to vote buying,
local councillors and political party officials seem to matter.
although the significance of the coefficient varies some-
Interestingly, direct contacts with Members of Parliament
what. This suggests that strong ethnic identities may in-
have little effect. This is consistent with Stokes’ (2005,
crease the likelihood of being a target of vote buying
2007b) observation that the organizational structure of
campaigns.
clientelistic parties is ‘bottom-heavy’ and relies on a large
Other than electoral competitiveness, ethnic fraction-
network of local party operatives. Voters’ party identifica-
alization is the strongest country-level predictor of vote
tion also has a positive and highly significant correlation
buying in Africa. That is, political parties operating in
with vote buying in models 7 and 8, suggesting that voters
ethnically fractionalized societies are much more likely to
who identify with a political party are more likely to
use vote buying as an instrument to secure re-election than
experience vote buying attempts. The positive correlation
parties in more homogeneous societies. The details of the
between party identification and vote buying could imply
electoral system also seem to matter (models 7 and 8).
that political parties target party loyalists or use material
Consistent with the ideas of Persson et al. (2003), vote
rewards to increase turnout among supporters who iden-
buying is most widespread in systems with small electoral
tify with the party but are disinclined to vote (Nichter,
2008).16 However, since the correlation might also reflect
reverse causality – i.e. that voters who get bought develop a
17
In additional tests, we looked at whether partisanship matters more
among the poor relative to wealthier people. The results show that
16
An alternative explanation is presented by Hicken (2011: 298) who partisanship matters for both groups. We have also tested whether party
summarizes ongoing work by Dunning and Stokes (2010). This explana- identification matters more when elections are close, by running re-
tion stresses that party leaders target swing districts and use vote brokers gressions which include an interaction of (individual-level) party iden-
to target voters within these districts. However, vote brokers prefer to tification and (country-level) electoral competitiveness. These results
target party loyalists within district because it is cheaper than targeting suggest that party identification does, in fact, matter more when elections
(and buying the voters of) voters supporting the opposing party. Our are close. Results are available upon request.
18
results are consistent with the notion that vote brokers target (loyal) However, education may also have an indirect negative effect on vote
voters who identify with a political party. buying if education contributes to reduce poverty.
 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232 231

district magnitudes, where barriers to entry are high Aidt, T.S., Jensen, P.S., 2012. From open to secret ballot: vote buying and
modernization. In: Cambridge Working Papers in Economics 1102.
because the winning candidate must gather support from a
University of Cambridge.
plurality in the constituency. Finally, urbanization, Aidt, T.S., Golden, M.A., Tiwari, D., 2011. Incumbents and criminals in the
schooling, and economic inequality (models 9 and 10) are Indian legislature. In: Cambridge Working Papers in Economics 1157.
mostly insignificant. University of Cambridge.
Alesina, A., Devleesschauwer, A., Easterly, W., Kurlat, S., Wacziarg, R.,
2003. Fractionalization. Journal of Economic Growth 8, 155–194.
7. Conclusions Alvarez, M., Cheibub, J.A., Limongi, F., Przeworski, A., 1996. Classifying
political regimes. Studies in Comparative International Development
31 (2), 3–26.
In his seminal study of machine politics in new post- Beck, T., Clarke, G., Groff, A., Keefer, P., Walsh, P., 2001. New tools in
colonial states, Scott (1969: 1150) stressed that “perhaps comparative political economy: the database of political institutions.
the most fundamental quality shared by the mass clientele The World Bank Economic Review 15 (1), 165–176.
Birch, S., 2011. Electoral Malpractice. Oxford University Press, Oxford.
of machines is poverty”. More than 35 years later, efforts to Blaydes, L., 2006. Who votes in authoritarian elections and why? De-
generate economic development as well as recent in- terminants of voter turnout in contemporary Egypt. In: Paper Pre-
troductions of electoral democracy in many African coun- sented at the Annual Meeting of the American Political Science
Association, 2006, Philadelphia, PA.
tries have not done much to alter this pattern. Indeed, the Bratton, M., van de Walle, N., 1997. Democratic Experiments in Africa:
findings of this paper suggest that the strong links between Regimes Transitions in Comparative Perspective. Cambridge Univer-
poverty and a particular form of machine politics – vote sity Press, Cambridge.
Bratton, M., Mattes, R., Gyimah-Boadi, E., 2005. Public Opinion, De-
buying – seem to persist. The key findings of this paper mocracy, and Market Reform in Africa. Cambridge University Press,
show that the micro-economic conditions under which Cambridge.
people live have large effects on the extent to which African Bratton, M., 2008. Vote buying and violence in Nigerian election cam-
paigns. Electoral Studies 27 (4), 621–632.
voters are targets of vote buying – an effect that seems to Bratton, M., Bhavnani, R., Chen, T.-H., 2012. Voting intentions in Africa:
increase when elections are closely contested. In the ethnic, economic or partisan. Commonwealth and Comparative Pol-
context of African democracies, it is perhaps not surprising itics 50 (1), 27–52.
Brusco, V., Nazareno, M., Stokes, S.C., 2004. Vote buying in Argentina.
that poverty is the predominant socio-economic source of
Latin American Research Review 39 (2), 66–88.
vote buying. In spite of recent economic progress and the Bryan, S., Baer, D., 2005. Money in Politics: a Study of Party Financing
spread of competitive elections, poverty is still widespread Practices in 22 Countries. National Democratic Institute for Interna-
(Poku and Mdee, 2011). Poor voters therefore constitute tional Affairs. http://www.ndi.org/.
Chandra, K., 2007. Counting heads: a theory of voter and elite behavior in
sizeable groups of the electorate in practically all of Africa’s patronage democracies. In: Kitschelt, H., Wilkinson, W.I. (Eds.), Pa-
democracies. This may affect the way democracy operates trons, Clients and Policies: Patterns of Democratic Accountability and
in general (Mattes, 2008), the political issues voters are Political Competition. Cambridge University Press, Cambridge,
pp. 84–109.
concerned with (Justesen, 2011), and the prospects of Cheibub, J.A., Ghandi, J., Vreeland, J.R., 2010. Democracy and dictatorship
developing and sustaining markets for buying and selling revisited. Public Choice 143 (1–2), 67–101.
votes during elections more specifically (Aidt and Jensen, Cox, G.W., Kousser, M.j, 1981. Turnout and rural corruption: New York as a
test case. American Journal of Political Science 25 (4), 646–663.
2012). And as the results of this paper suggest, poverty Dixit, A., Londregan, J., 1996. The determinants of success of special in-
also makes vote buying a more common political strategy terests in redistributive politics. Journal of Politics 58 (4), 1132–1155.
during elections. This also suggests that we should not Dunning, T., Stokes, S., 2010. How does the internal structure of political
parties shape their distributive strategies?. In: Paper Presented at
dismiss the role of economic development at the macro Workshop on Political Parties in the Developing World, Princeton
level, but that economic development may well make vote University, April 30–May 1.
markets less vibrant if it contributes to eradicate the ma- Eifert, B., Miguel, E., Posner, D.N., 2010. Political competition and ethnic
identification in Africa. American Journal of Political Science 54 (2),
terial poverty that is instrumental for the operation of vote
494–510.
buying political parties. Elklit, J., Svensson, P., 1997. What makes elections free and fair? Journal of
Democracy 8 (3), 32–45.
Ferree, K., Long, J.D., 2011. Violating the Secret Ballot: the Political Logic of
Acknowledgements Vote Monitoring in Ghana’s 2008 Elections (Working Paper).
Foeken, D., Dietz, T., 2000. Of ethnicity, manipulation and observation:
the 1992 and 1997 elections in Kenya. In: Abbink, J., Hesseling, G.
We are grateful for constructive comments from Toke
(Eds.), Election Observation and Democratization in Africa. Macmillan
Aidt and two anonymous reviewers. An earlier version of Press, Basingstoke, pp. 122–149.
the paper was presented (under the title ‘The economic Gonzalez-Ocantos, E., Kiewiet de Jonge, C., Meléndez, C., Osorio, J.,
Nickerson, D.W., 2012. Vote buying and social desirability bias:
origins of vote buying in Africa’) at the 2012 meeting of the
experimental evidence from Nicaragua. American Journal of Political
European Political Science Association in Berlin. Comments Science 56 (1), 202–217.
received on that occasion are much appreciated. Heston, A., Summers, R., Aten, B., 2011. Penn World Table Version 7.0.
Center for International Comparisons of Production, Income and
Prices. University of Pennsylvania.
Appendix A. Supplementary material Heckelman, J.C., 1998. Bribing voters without verification. Social Science
Journal 35 (3), 435–443.
Hicken, A.D., 2007. How do rules and institutions encourage vote
Supplementary material related to this article can be
buying? In: Schaffer, F.C. (Ed.), Elections for Sale: the Causes and
found at http://dx.doi.org/10.1016/j.electstud.2013.07.020. Consequences of Vote Buying. Lynne Rienner Publishers, Boulder,
pp. 47–60.
Hicken, A.D., 2011. Clientelism. Annual Review of Political Science 14,
References 289–310.
Hobolt, S.B., Klemmensen, R., 2008. Government responsiveness and
Acemoglu, D., Johnson, S., Robinson, J., Yared, P., 2008. Income and de- political competition in comparative perspective. Comparative Polit-
mocracy. American Economic Review 98 (3), 808–842. ical Studies 41 (3), 309–337.
232 P.S. Jensen, M.K. Justesen / Electoral Studies 33 (2014) 220–232

Jacobs, K., Spierings, N., 2010. District magnitude and voter turnout: a Nichter, S., 2008. Vote buying or turnout buying? Machine politics and the
multilevel analysis of self-reported voting in the 32 Dominican dis- secret ballot. American Political Science Review 102 (1), 19–31.
tricts. Electoral Studies 29 (4), 704–718. Osei, A., 2012. Party-voter Linkage in Africa: Ghana and Senegal in
Justesen, M.K., 2011. Too Poor to Care? The Salience of AIDS in Africa. Comparative Perspective. Springer VS, Wiesbaden.
Afrobarometer Working Paper No. 133 www.afrobarometer.org. Persson, T., Tabellini, G., Trebbi, F., 2003. Electoral rules and corruption.
Justesen, M.K., Bjørnskov, C., 2012. Exploiting the Poor: Bureaucratic Journal of the European Economic Association 1 (4), 958–989.
Corruption and Poverty in Africa. Afrobarometer Working Paper No. Poku, N.K., Mdee, A., 2011. Politics in Africa: a New Introduction. Zed
139 www.afrobarometer.org. Books, London.
Karp, J.A., Brockington, D., 2005. Social desirability and response validity: Rabe-Hesketh, S., Skrondal, A., 2008. Multilevel and Longitudinal
a comparative analysis of overreporting voter turnout in five coun- Modeling Using Stata, second ed. Stata Press, Texas.
tries. Journal of Politics 67 (3), 825–840. Robinson, J., Parsons, N.Q., 2006. State formation and governance in
Keefer, P., 2007. Clientelism, credibility, and the policy choices of young Botswana. Journal of African Economies 15, 100–140.
democracies. American Journal of Political Science 51 (4), 804–821. Robinson, J., Verdier, T., 2013. The political economy of clientelism.
Kitschelt, H., Wilkinson, S.I., 2007. Citizen-politician linkages: an intro- Scandinavian Journal of Economics 115 (2), 260–291.
duction. In: Kitschelt, H., Wilkinson, W.I. (Eds.), Patrons, Clients and Schaffer, F., 1998. Democracy in Translation: Understanding Politics in an
Policies: Patterns of Democratic Accountability and Political Compe- Unfamiliar Culture. Cornell University Press, Ithaca.
tition. Cambridge University Press, Cambridge, pp. 1–49. Scott, J., 1969. Corruption, machine politics, and political change. Amer-
Koter, D., 2013. King makers: local leaders and ethnic politics in Africa. ican Political Science Review 63 (4), 1142–1158.
World Politics 65 (2), 187–232. Sisson, R., 1972. The Congress Party in Rajasthan: Political Integration and
Kramon, E., 2009. Vote-buying and Political Behavior: Estimating and Institution-building in an Indian State. University of California Press,
Explaining Vote-buying’s Effect on Turnout in Kenya. Afrobarometer Berkeley.
Working Paper no. 114 www.afrobarometer.org. Solt, F., 2009. Standardizing the world income inequality database. Social
Lehoucq, F., 2007. When does a market for votes emerge? In: Schaffer, F.C. Science Quarterly 9, 231–242.
(Ed.), Elections for Sale: the Causes and Consequences of Vote Buying. Stokes, S.C., 2005. Perverse accountability: a formal model of machine
Lynne Rienner Publishers, Boulder, pp. 33–46. politics with evidence from Argentina. American Political Science
Lemarchand, R., Legg, K., 1972. Political clientelism and development: a Review 99 (3), 315–325.
preliminary analysis. Comparative Politics 4 (2), 149–178. Stokes, S.C., 2007a. Political clientelism. In: Boix, C., Stokes, S.C. (Eds.), The
Lieberman, E., Singh, P., 2012. The institutional origins of ethnic violence. Oxford Handbook of Comparative Politics. Oxford University Press,
Comparative Politics 45 (1), 1–24. Oxford, pp. 604–627.
Lindberg, S.I., 2003. It’s our time to "chop”: do elections in Africa feed Stokes, S.C., 2007b. Is vote buying undemocratic? In: Schaffer, F.C. (Ed.),
neo-patrimonialism rather than counter-act it? Democratization 10 Elections for Sale: the Causes and Consequences of Vote Buying.
(2), 121–140. Lynne Rienner Publishers, Boulder, pp. 81–99.
Linos, E., 2013. Do conditional cash transfer programs shift votes? Evi- Van de Walle, N., 2007. Meet the new boss, same as the old boss? The
dence from the Honduran PRAF. Electoral Studies (Forthcoming). evolution of political clientelism in Africa. In: Kitschelt, H.,
Mattes, R., 2008. The Material and Political Bases of Lived Poverty: In- Wilkinson, W.I. (Eds.), Patrons, Clients and Policies: Patterns of
sights from the Afrobarometer. CCSR Working Paper No. 216. Uni- Democratic Accountability and Political Competition. Cambridge
versity of Cape Town. University Press, Cambridge, pp. 50–67.
Medina, L.F., Stokes, S.C., 2007. Monopoly and monitoring: an approach to Vicente, P.C., 2013. Is vote-buying effective? Evidence from a
political clientelism. In: Kitschelt, H., Wilkinson, W.I. (Eds.), Patrons, field experiment in West Africa. Economic Journal (Forthcoming).
Clients and Policies: Patterns of Democratic Accountability and Po- Vicente, P.C., Wantchekon, L., 2009. Clientelism and vote buying: lessons
litical Competition. Cambridge University Press, Cambridge, pp. 68– from field experiments in African elections. Oxford Review of Eco-
83. nomic Policy 25 (2), 292–305.
Mwangi, O.G., 2008. Political corruption, party financing and democracy Wantchekon, L., 2003. Clientelism and voting behavior: evidence from a
in Kenya. Journal of Modern African Studies 46 (2), 267–285. field experiment in Benin. World Politics 55 (3), 399–422.