Poverty and vote buying:

Survey-based evidence from Africa

Peter S. Jensen* Mogens K. Justesen+

Electoral Studies 33 (March 2014), 220-232.

Acknowledgements: We are grateful for constructive comments from Toke Aidt and two

anonymous reviewers. An earlier version of the paper was presented (under the title ‘The economic

origins of vote buying in Africa’) at the 2012 meeting of the European Political Science Association

in Berlin. Comments received on that occasion are much appreciated.

*
Department of Business and Economics, University of Southern Denmark, Campusvej 55, DK-5230 Odense M,

Denmark; E-mail: psj@sam.sdu.dk; phone: +45 6550 4472.

+
Department of Business and Politics, Copenhagen Business School, Porcelaenshaven 1, DK-2000 Frederiksberg,

Denmark; E-mail: mkj.dbp@cbs.dk; phone: +45 3815 3564.

1
Abstract. Alongside the spread of democracy in the developing world, vote

buying has emerged as an integral part of election campaigns. Yet, we know

little about the causes of vote buying in young democracies. In this paper,

we analyse the sources of vote buying in sub-Saharan 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 poor voters are significantly more likely to be targets of vote

buying than wealthier voters. This effect increases when elections are highly

competitive. Thus, micro-level poverty seems to be an important source of

vote buying in Africa and has major implications for the way electoral

democracy operates.

Key words: Vote buying; elections; new democracies; poverty; Africa;

Afrobarometer

2
1. Introduction

Vote markets used to be widespread during the early stages of democracy in Western Europe, but

largely seemed to disappear with economic development (Aidt and Jensen, 2012). In recent

decades, new waves of democratization have occurred around the globe, bringing democracy well

beyond the borders of the Western world. While these transitions are usually celebrated, they have

also lead to renewed debate about the operation of democracy in developing countries. Indeed,

alongside transitions to democracy, vote buying—understood as the direct exchange of money or

gifts for votes—has made a powerful return to the scene of democratic politics. Reports of vote

buying come from most regions of the world, including Asia (Hicken, 2007), Africa (Bratton, 2008;

Vicente and Wantchekon, 2009), the Middle East (Blaydes, 2006), and Latin America (Brusco et

al., 2004; Stokes, 2005; Gonzales-Ocantos et al., 2012). Thus, vote buying seems to be an integral

feature of electoral politics in new democracies across the world. In this paper we contribute to this

literature by providing new evidence on how vote markets operate in elections with de jure secret

ballot, with a particular focus on the relationship between poverty and voters’ experience with being

offered pre-election rewards in return for votes.

Vote buying is a particular form of political clientelism, i.e. the direct exchange at the

individual level of rewards and material goods by political patrons in return for electoral support by

voters (Stokes, 2007a; Hicken, 2011; Linos, 2013; Robinson and Verdier, 2013). It is widely

accepted that clientelistic politics create economic inefficiencies, reduce the supply of public goods,

and bias public policy in favour of elites (Stokes, 2007a; Vicente and Wantchekon, 2009; Robinson

and Verdier, 2013). Vote buying also raises questions about the character of democracy. While

elections involving vote buying may be ‘free’—allowing voters to choose between multiple

candidates in elections with universal suffrage—they collide with standards of democratic

‘fairness’, because the interests of some voters are bought by parties before the election, and may

3
therefore be ignored by political representatives after the election (cf. Elklit and Svensson, 1997).

According to Stokes (2007b: 96), the undemocratic nature of vote buying arises precisely because ‘

… it keeps vote sellers from having their interests accurately interpreted and made known, and in

that it makes them less autonomous than are the recipients of politically motivated public

programs.’ This tends to weaken or even reverse the accountability link between voters and

politicians (Stokes, 2005, 2007b).

Despite its consequences, empirical evidence suggests that vote buying may contribute to

increase electoral support (Wantchekon, 2003; Brusco et al., 2004; Blaydes, 2006; Vicente and

Wantchekon, 2009). In new democracies in particular, parties often rely on existing patron-client

networks and pre-election transfers to mobilize support (Keefer, 2007). However, there are good

reasons to suspect that political parties do not distribute their vote buying efforts randomly across

the electorate. Theoretical priors suggest that vote buying parties systematically target specific

groups in the electorate based on their socio-economic characteristics. Poverty in particular has

been emphasized as an important source of vote buying that enables political parties to exploit the

material needs of deprived voter groups by trading rewards for votes (Stokes, 2005, 2007a, 2007b;

Blaydes, 2006).

In this paper, we contribute to the study of vote buying by empirically examining how poverty

affects the likelihood that voters are targeted by vote buyers in a cross-section of African

democracies. In doing so, we follow the tradition of the seminal contributions in political science by

Scott (1969) and Lemarchand and Legg (1972), who analysed how modernization in new post-

colonial states affects—and ultimately undermines—patron-client networks in general and vote

markets more specifically. More recently, these issues have also been addressed in formal models in

political economy (Stokes, 2005; Aidt and Jensen, 2012; Robinson and Verdier, 2013). In this

paper, we focus on sub-Saharan African since democratic politics has only recently become the

4
norm in this region (van de Walle, 2007). In fact, no less than 40 African countries held their first

competitive election in a generation during the early 1990s (Bratton and van de Walle, 1997).

However, while multiparty politics and competitive elections have gained prominence, many

African countries are yet to complete their democratic transitions, some have reverted back into

non-democracy, and still others are plagued by violence and coups, as witnessed most recently in

Mali. Moreover—as we document later—vote buying is widespread in many countries that have

continued along the path towards democracy.

We make two contributions to the literature. First, while the relationship between vote buying

and poverty has been studied in Nigeria (Bratton, 2008), Kenya (Kramon, 2009), and Malawi and

Mozambique (Birch, 2011), this paper appears to be the first to provide a comprehensive analysis of

how poverty affect individuals’ propensity of being targeted by vote buyers across African

democracies, and, therefore, how vote markets operate during election campaigns in Africa.

Second, to the best of our knowledge, this study constitutes the most extensive cross-country

examination of vote buying to date, and the first attempt to study vote buying across a larger set of

African countries. Indeed, in a recent review of the literature, Stokes (2007a: 618) emphasizes that

there is a general lack of quantitative cross-country analyses of clientelism and vote buying. We

start to fill this gap by analyzing micro-level data on vote buying from 18 countries in sub-Saharan

African. Specifically, we use data from the third round (2005-2006) of the Afrobarometer survey.

This provides a unique source of data on vote buying in Africa, since it is so far the only round of

the Afrobarometer that contains information on voters’ experience with vote buying during

elections.

The fact that our study draws on data for a broader set of African countries has at least two

advantages. First, while the existing literature contains analyses of clientelism and vote buying in a

number of African countries, such as Benin (Wantchekon, 2003; Koter, 2013), Ghana (Lindberg,

5
2003), Sao Tome and Principe (Vicente, 2013), Nigeria (Bratton, 2008), Kenya (Kramon, 2009),

Malawi and Mozambique (Birch, 2011) and Senegal (Koter, 2013), using data for a larger set of

countries allows for more general inferences. Second, since the data contain micro-level survey

information from 18 countries, we are able to take account of the fact that voters vary in their

individual characteristics within countries and that cross-country differences may affect average

levels of vote buying between countries. In this way, we can both take into account the contextual

factors that Birch (2011: 106-107) emphasizes as important and investigate the role of inherently

interesting country characteristics such as electoral competitiveness. We do so by both conducting a

set of multilevel regressions and by controlling for country fixed effects which capture the influence

of country-level variables.

The rest of the paper is organized as follows. The following section provides a brief overview

of voters’ experiences with being targets of vote buying in 18 African countries, as well as

qualitative accounts of vote buying in Africa. After that, we briefly outline the theory relating

poverty to vote buying. Next, we describe the econometric model and data we use in the empirical

part. The next section presents empirical results from multilevel regressions, and the final section

concludes on the main findings.

2. Vote buying in Africa

The importance of vote markets in Africa is shown in Figure 1, which plots the percentage of voters

who report being targeted by vote buyers during the most recent presidential or parliamentary

elections for the 18 countries in our sample.

[Figure 1 about here]

6
Figure 1 shows that vote buying is pervasive in Kenya, Uganda, Benin, Madagascar, Nigeria, and

Mali; less widespread in a small group of countries including Senegal, Tanzania and South Africa;

and almost absent in Botswana and Lesotho.

Qualitative accounts of elections in 1990s and 2000s corroborate that vote buying is a common

electoral phenomenon in Kenya. Foeken and Dietz (2000) note that the support groups Youth of

KANU ’921 and Toroitich Till 20002 were highly active players during election campaigns and

undertook vote buying on a large scale, spending an estimated total of US$60 million on this

purpose alone. They also observe that the Kenyan Electoral Commission acknowledged that the

1997 election felt short of being ‘fair’. For instance, in 13 percent of polling stations, secret voting

was not guaranteed, and vote buying was quite common on the day of the election. According to

Mwangi (2008: 272), KANU’s election victories in 1992 and 1997 were largely due to electoral

corruption. Vote buying was also widespread during elections in 2002 and 2007, where voters were

simply being offered money upfront (Kramon, 2009). In line with this, Bryan and Baer (2005)

report that parties spend around 40 percent of their total campaign expenditure on vote buying. The

Kenyan example also suggests one obvious reason why vote buying may work in an environment

with secret ballot, which at least de jure has been introduced in all countries in the Afrobarometer—

it is simply violated, as it also was historically in, e.g., France and Germany (Aidt and Jensen,

2012).

Figure 1 also shows that around 7 percent of voters report experience with vote buying in

Senegal. The presence of vote buying in Senegal is supported by qualitative accounts. Schaffer

(1998) argues that even though the secret ballot was reintroduced in the 1990s, vote buying was still

possible due to gerrymandering that made Senegalese constituencies smaller—making it easier for

political party operatives to monitor deals made on vote markets. More recently, Osei (2011, 261)

1
KANU denotes the Kenya African National Union
2
Toroitich is the middle name of former Kenyan president Moi.

7
noted that the ruling party (PDS) is still actively buying votes during national elections. Similarly,

Bryan and Baer (2005) report that politicians in Senegal spend almost 20 percent of campaign

expenditures on vote buying—about half of what is used in Kenya.

Finally, Figure 1 shows that vote buying seems uncommon in Botswana as less than 2 percent

of voters report being targeted by vote buyers. In fact, Botswana is often considered the oldest,

established democracy in sub-Saharan Africa (Robinson and Parssons 2006) with ‘… little

repression and no exceptional allegations of fraud’ (Alvarez et al. 1997: 10). These descriptions fit

well with figure 1.

These examples are obviously only illustrations of the presence of—and differences between—

vote markets in Africa. However, they do support our data as being good reflections of the

magnitude of vote buying across countries in Africa.

3. Poverty and vote buying

The previous section shows that the magnitude of vote buying varies significantly between

countries, suggesting important differences in clientelistic electoral practices even within

democracies located in the same geographical region. At a very basic level, this raises the question

of how vote markets can operate in the presence of the secret ballot—a mechanism that was

designed precisely to allow voters to conceal their electoral choice from patrons and political parties

(Cox and Kousser 1981; Heckelman 1998; Aidt and Jensen 2012). While we do not deal with this

issue in detail, we emphasize two points. First, although the secret ballot allows voters to renege on

commitments to vote for a particular candidate, it does not necessarily eliminate vote buying.

Indeed, the literature has uncovered a number of strategies and mechanisms that enable parties to

uphold a certain level of monitoring of how—and if—people vote. These include party-issued

ballots, positioning of party agents inside polling stations, exploiting voters’ social network, paying

8
opposition voters to abstain – or supporters to turn out, and outright intimidation (Cox and Kousser

1981; Heckelman 1998; Stokes 2005; Medina and Stokes 2007; Nichter 2008). Second, data from

round four of the Afrobarometer show that around 25 percent African voters believe it is ‘somewhat

likely’ or ‘very likely’ that their vote choice can be monitored. 3 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’ vote buying campaigns target

particular voter groups. One of the key determinants of vote buying identified in the literature is

poverty. There are a number of reasons why vote buying thrives in the context of widespread

poverty. Firstly, poor voters generally lack access to resources—e.g. food, clean water, and medical

care—that politicians can promise to deliver during election campaigns. As emphasized by Scott

(1969), this paves the way for clientelistic relationships between voters and politicians and the use

of pre-election rewards as a means to mobilize electoral support among the poor. Secondly, on the

assumption that the marginal utility of income is higher for poor groups (Dixit and Londregan,

1996; Stokes, 2005), the utility derived from selling one’s vote is higher for poor people. Selling

one’s vote for material rewards may therefore be a rational course of action for people living in

material deprivation (Blaydes, 2006).4 Thirdly, from the perspective of political parties, votes at the

bottom of the income distribution are cheaper to buy (Dixit and Londregan, 1996; Blaydes, 2006;

Stokes, 2007b: 119). Parties can therefore buy more votes among the poor by offering even

relatively modest amounts to each voter. For instance, Bratton (2008) reports that during Nigeria’s

2007 elections the most common amount of money offered to voters was US$4. These economic

mechanisms are likely to make poor voters the prime targets of vote buying by clientelistic parties

who want to maximize their (re)election chances. We therefore expect that political parties are more

3
These data are not available in the third round.
4
This requires that voters are paid enough to be indifferent between selling their vote and receiving the benefit from
having their preferred party in power (see, e.g., Aidt and Jensen, 2012). For parties, this may imply that vote buying
trumps offering solid redistributive programs to poor voters.

9
likely to target and buy the votes of poor people. This is the hypothesis we test in the remainder of

the paper.

4. Econometric model

Our econometric approach uses a multilevel probability model for vote buying (Rabe-Hesketh and

Skrondal 2008). In the multilevel model with i, …, N individuals nested within j, …, J countries, the

individual-level equation is:

Pr⁡(𝑦𝑖𝑗 = 𝟏|𝛼j , 𝑿ij, 𝒁ij ) = 𝐆(𝛼j + 𝛽1𝑿ij ⁡ + 𝛽2 𝒁ij ) (1)

In equation (1), yij is a dummy variable equal to 1 if the individual has been approached by a vote

buyer and zero otherwise. Accordingly, Pr⁡(𝑦𝑖𝑗 = 𝟏|𝛼j , 𝑿ij, 𝒁ij ) is the probability that a voter reports

being targeted by a vote buyer during the most recent election. Model (1) is generalized via the

function G(.) and easily applicable to different probability models using appropriate link functions.

Here we use logit and linear link functions:

1
𝐆(𝛼j + 𝛽1 𝑿ij ⁡ + 𝛽2 𝒁ij ) = −(𝛼j +𝛽1 𝑿ij ⁡+𝛽2 𝒁ij ) (1a)
1+e

𝐆(𝛼j + 𝛽1 𝑿ij ⁡ + 𝛽2 𝒁ij ) = 𝛼j + 𝛽1 𝑿ij ⁡ + 𝛽2 𝒁ij (1b)

The X variable represents individual-level measures of poverty. The variables in Z are individual-

level controls. The variables are described below.

In (1), αj is a group-specific constant that varies across countries, and can be treated as random

or fixed. In the random effects model, αj is modelled as a function of country-level variables, as in

equation (2).

10
 𝛼j = 𝜇 + γ𝑪𝐣 + 𝜂j (2)

αj comprises a common intercept, μ, a non-random part explained the country-level variables, C,

and a country-specific random effect, ηj. C includes country-level variables such as GDP and

urbanization. Substituting (2) into (1) we obtain the random effects multilevel model (3).

Pr⁡(𝑦𝑖𝑗 = 𝟏|𝛼j , 𝑿ij, 𝒁ij ) = 𝐆(𝜇 + γ𝑪𝐣 + 𝜂j + 𝛽1 𝑿ij ⁡ + 𝛽2 𝒁ij ) (3)

The fixed effects model, which removes bias due to omitted variables at the country-level, is used

as an alternative. This model treats the αj’s in (1) as parameters to be estimated. While fixed effects

may be approximated by observable variables, it is arguably hard to control for all such country-

specific effects. For example, Acemoglu et al. (2008) argue that whether countries embarked on a

path towards electoral democracy and economic development was affected by critical junctures in

history. Thus, the country-level differences in vote buying shown in Figure 1 may have deep

historical roots that country fixed effects enable us to control for.

5. Data

Our data come from the Afrobarometer survey, which is a standard source of information on voter

attitudes and experiences with democracy in Africa (Bratton et al., 2005; Bratton, 2008; Justesen,

2011; Justesen and Bjørnskov, 2012).5 Summary statistics for the data are available in the online

appendix. We use the third round of the survey—collected in 2005 and 2006.6

5
An online appendix and replication data are available on https://sites.google.com/site/mkjustesen/
6
Details on the Afrobarometer methodology are available in Bratton et al. (2005) and on http://afrobarometer.org/

11
 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., 2005.). The sample size is

1200, except in three highly fractionalized countries—Nigeria, South Africa, and Uganda—where it

is 2400. The questionnaire consists of a standardized set of questions, making the data comparable

across countries. The interviews were conducted face-to-face, in the local language, and by people

outside the local community. This ensures that interviews are conducted with anonymity, and that

respondents can answer questions without fear of social repercussions from the local community,

even on potentially controversial issues like vote buying. Finally, while individual respondents are

randomly selected within countries, the countries in which the surveys are conducted are not. The

Afrobarometer countries are selected because they are minimally democratic and not part of armed

conflicts (Bratton et al. 2005).7 In the third round, the only country that may fall short of being an

electoral democracy is Zimbabwe. However, as shown below, our results are entirely robust to

excluding Zimbabwe. Otherwise, the Afrobarometer countries are similar to the sub-Saharan

average on a range of socio-economic indicators (Justesen, 2011: 8).

5.1. Measuring vote buying

We construct our vote buying measure from the question (Q57F): “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?” The question refers to the most recent national presidential or

parliamentary election in all 18 countries. While the question is part of a larger battery of survey

instruments concerning people’s experiences with paying bribes to government officials in return

for public services, it is distinct in asking respondent about their experience with being offered

7
According to Freedom House (http://www.freedomhouse.org), all countries except Zimbabwe are classified as ‘Free’
or ‘Partly Free’ on political rights and civil liberties. According to the regime classification of Cheibub et al. (2010) –
requiring that government power has alternated following elections – only nine countries were democracies in 2005.
These are Benin, Cape Verde, Ghana, Kenya, Madagascar, Malawi, Mali, Nigeria and Senegal.

12
rewards (rather than offering bribes themselves) by political candidates or party activist (rather than

to government officials) in return for their vote (rather than in return for a public service). The

question uncovers the property of electoral politics we are interested in: The use and frequency of

vote buying by political parties. However, it is important to note that the question concerns intended

vote buying by parties. That is, the question does not allow us to measure whether voters actually

accepted electoral bribes. But it does allow us to measure which voters political parties tend to

target during vote buying campaigns—and that it precisely what we want to examine.

Respondents can answer using the five categories ‘Never’, ‘No experience with this in the past

year’, ‘Once or twice’, ‘A few times’, or ‘Often’. We merge the two first categories and code them

as zero. The remaining categories are given the codes 1, 2, and 3, respectively, giving us an ordinal

measure of vote buying. This allows us to measure individuals’ experience with being approached

by vote buyers only in a relatively crude manner. In particular, we are not able to measure how

much money people are offered in return for their votes or what the monetary value of gifts offered

might be. With this caveat, we are nonetheless able to establish a measure of the prevalence of vote

buying both within and across African democracies. Figure 2a shows that 18 percent of Africans

report having been offered gifts or money in return for their vote, albeit with considerable

differences between countries (cf. Figure 1).

[Figures 2a and 2b about here]

Nevertheless, most respondents report no experience with being targeted by vote buyers. Therefore

we also construct a dichotomous variable—shown in Figure 2b—which simply distinguishes people

13
who have been offered rewards in return for their votes (1) from people who have no experience in

this regard (0). These measures are used as dependent variables in our estimation of equation (1).

5.2. Measuring poverty

To measure individual-level differences in poverty, we use survey questions from the

Afrobarometer to construct what the literature often refers to as the Index of Lived Poverty (Bratton

et al., 2005; Bratton, 2008; Mattes, 2008; Justesen, 2011; Justesen and Bjørnskov, 2012). The index

assesses individual-level poverty by asking respondents about their experiences with lacking access

to basic household necessities like food and clean water (Bratton, 2008: 31-32; Poku and Mdee,

2011: 54). Specifically, the survey contains five questions asking respondents how often during the

past year, they or anyone in their family have gone without: a) enough food to eat, b) enough clean

water for home use, c) medicines or medical treatment, d) enough fuel to cook food, e) a cash

income.8 Respondents’ answers are coded on a five-point scale from “never” to “always”.

Following the standard in the literature, we combine these five items into an aggregate index of

experienced poverty, and normalize it to vary between 0 and 1, with high values indicating that

people live in a state of severe poverty and are regularly deprived of access to basic material

necessities (Bratton et al. 2005; Bratton 2008; Mattes 2008; Justesen and Bjørnskov 2012). Low

values reflect that people live in good material conditions and do not suffer from deprivation due to

lack of basic household necessities. At the country-level, we include GDP per capita to capture the

impact of wealth and poverty on vote buying at the macro-level (Aidt and Jensen, 2012). This is

important because the vote market may diminish or break down once the share of poor voters at the

macro level reaches a low level. Data are from the Penn World Tables 7.0 (Heston et al., 2011).
8
We do not use question Q8F (concerning lack of school expenses for children), because it is not clear how people
without children should be classified in the index. However, our results are robust to including this item in the poverty
index.

14
5.3. Control variables

To alleviate problems of confounding caused by omitted variables, we include a number of

individual- and country-level control variables that may be correlated with both 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 exposed to vote buying attempts may be a

function simply of being in contact with political officials or party activists (Stokes, 2005, 2007b).9

We therefore include four variables (Q32A-Q32D), asking respondents how often during the past

year they have contacted: a) a local government councillor, b) a member of parliament, c) a

government ministry official, or d) a political party official, to solve a problem or express personal

opinions. This should ensure that we capture possible effects on vote buying arising from

respondents’ propensity to contact people affiliated with political parties and candidates.

Respondents can answer the four questions on a four-point scale, which we transform into dummy

variables, coded as 1 for people who have been in contact with political officials in one of the four

domains. However, since these questions were not asked in Zimbabwe, we also run regressions

excluding the four variables. This allows us to check that our results are not sensitive to including

Zimbabwe—the least democratic of the Afrobarometer countries—in the regressions.

Following debates about political machines’ propensity to offer rewards to opponents or

supporters (Stokes, 2005; Nichter, 2008), we also control for whether voters identify with a political

party, using a dummy variable coded as one for people who “feel close to a particular political

party”, and zero otherwise (Q85). However, the possibility of reverse causality is a major concern in

the relationship between vote buying and party identification, since it is plausible that voters say

9
However, Bratton et al. (2005: 151-153) find that people in Africa are more likely to address their grievances to
traditional/religious leaders in the community rather than to political officials.

15
they identify with a particular party because the party distributes rewards to voters. We therefore

control for party identification only in some models.

We also include five socio-economic controls that may correlate with vote buying and poverty.

First, differences in voters’ urban-rural residence may influence the potential for vote buying

(Lehoucq, 2007; Hicken, 2007). We therefore include a dummy variable indicating whether

respondents reside in urban (1) or rural (0) areas. Moreover, controlling for urban residence is

important since rural areas tend to be characterized by higher levels of poverty than urban areas

(Poku and Mdee, 2011: 54). Second, education may undermine the ability of parties to mobilize

support by buying votes because well-educated voters may be more likely to break traditional

clientelistic bonds (Sisson 1972). To measure education, we include three dummy variables

measuring whether respondents have a primary, secondary, or tertiary education, where respondents

with no formal schooling are the reference group. Third, a dummy variable distinguishes people

earning wages through formal employment (1) from people who do not (0). This is an important

control since formal employment is more widespread in urban areas and may also cause selection

into poverty. Fourth, we control for the respondents’ gender to capture bias in the tendency of vote-

buyers to target men. For similar reasons, we control for the age of respondents. Finally, ethnic

identities are often claimed to form an important basis for voting decisions in African elections

(Chandra, 2007; Kitschelt and 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 group affiliations because of beliefs that this will increase rewards for voters and re-election

chances for politicians. Vote buyers may therefore be more likely to target voters who identify with

an ethnic group, just as voters identifying with an ethnic group may be more likely to participate in

elections, making them more exposed to vote buying attempts. We therefore control for the salience

16
of ethnic identities, measured as respondents’ tendency to identify themselves with their nation (e.g.

Ghana or Kenya) or their ethnic group (Q82). High values indicate strong ethnic identity.10

At the country-level, we include a measure of urbanization (the proportion of the population

living in urban areas) and secondary school enrolment rates. Data are from the World Development

Indicators. However, since school enrolment correlates with GDP per capita at 0.94 (p<0.001) and

with urbanization at 0.73 (p<0.001), we include urbanization in most models and replace it by

school enrolment in some models to check that the results are not sensitive to this choice. We also

account for ethnicity by including a measure of nation-wide ethnic fractionalization (Alesina et al.

2003), which politicians may exploit to make ethnic identity a salient political issue (Lieberman and

Singh, 2012) or to privilege members of their own ethnic group using vote buying and other

clientelistic policies (van de Walle, 2007). Therefore, country-level ethnic fragmentation may be an

important pre-condition for vote markets to operate along ethnic lines. Some models also account

for the degree of inequality using data from Solt (2009) as a proxy for economic fractionalization

among voters that parties can exploit.

Finally, electoral competitiveness may affect nation-wide levels of vote buying (Kitschelt and

Wilkinson, 2007; van de Walle, 2007; Ferree and Long, 2011). If elections are close—and therefore

highly competitive—candidates may resort to vote buying strategies to maximize the chances of

winning. The relationship between competitiveness and vote buying can also appear in more subtle

ways. For instance, in the context of Indian elections, Aidt et al. (2011) argue that candidates with a

criminal record are more likely to appear on the ballot in constituencies where the election is highly

contested, arguably because such candidates may deter opposition voters from voting—a form of

10
It is also possible that belonging to – rather than identifying with – a specific ethnic group matters. To test this, we
have also controlled for ethnic group affiliation by including a full set of dummy variables for respondents’ ethnic/tribal
groups (Q79). This should capture the impact of confounding caused by (hardwired) ethnic group affiliations rather than
(socially constructed) ethnic identities. This approach also captures possible effects of being from the same (or a
different) ethnic group as the president/party in power (Bratton et al., 2012). Results (available upon request) show that
including the ethnic group indicators do not affect the impact of poverty on vote buying.

17
‘deflationary’ vote buying (Cox and Kousser, 1981; Heckelman, 1998). To account for electoral

competitiveness, we follow Eifert et al. (2010) and Ferree and Long (2011) and calculate the vote

margin at the most recent election, measured as the difference between the vote share of the

candidate receiving most votes and the candidate receiving second-most votes. When this difference

is small, elections tend to be close, and we expect higher levels of nation-wide vote buying. The

magnitude of electoral districts—i.e. the number of political candidates elected per electoral

district—may also affect the electoral strategies of political parties. For instance, Persson et al.

(2003) argue that small electoral districts—e.g. single-member constituencies—raise barriers to

entry into office for political candidates. By raising barriers to entry, small electoral districts may

also create incentives for political candidates to distribute pre-election rewards to voters in return

for their votes. In a related study, Jacob and Spierings (2010) argue that politicians are more likely

to use clientelist strategies in small electoral districts because voters can better identity the

politicians who deliver the benefits. We therefore expect that vote buying is more common when

electoral districts are smaller.11 Some models consequently control for mean district magnitude,

using data from the Database of Political Institutions (Beck et al., 2001).

6. Results

Table 1 presents results from 12 regressions with vote buying as the dependent variable. Models 1-

10 show results from random effects multilevel regressions and models 11-12 use fixed effects

estimators. Given the character of the dependent variables, we supplement binary logistic

regressions and linear regression with ordered logistic regressions.

A possible concern with using electoral competitiveness to explain vote buying is that parties’ attempts to buy votes
11

may also make elections closer. The potential for reverse causality is much less when we use electoral district
magnitude.

18
 [Table 1 about here]

6.1. Main results

Models 1-3 analyze data for all countries—including Zimbabwe. In these models, poverty has a

positive and highly significant effect on people’s propensity to experience vote buying. These

results suggest that differences in poverty are significantly related to vote buying at the micro-level,

with poor groups being much more likely targets of vote buying than wealthier groups. However, at

the country-level, GDP per capita does not display a particularly robust relationship with vote

buying. Although there are some indications that poorer countries have higher average-levels of

vote buying, the coefficients are insignificant in most models. While this result may seem puzzling,

part of the explanation is arguably the limited variation in GDP per capita in the group of African

countries we examine, with only a couple of countries (Botswana and South Africa) being

comparatively richer than the rest. This means that, by construction, a correlate of our sample is a

relatively low level of economic development. Moreover, across the 18 countries in the sample,

poverty is still so widespread that the share of poor voters at the macro level is large enough for

vote markets to operate, in spite of differences in average levels of GDP per capita.

In models 4-12 we include variables measuring ethnic identity and respondents’ contact with

representatives affiliated with political parties and candidates. In effect, data from Zimbabwe are

excluded. In models 4-6, individual-level poverty continues to have a positive and highly significant

impact on vote buying, even though the magnitude of the coefficient decreases a little. While only

indicative, this increases our confidence that omitted variables are not a major source of bias in the

estimates. In models 7-8, we control for party identification. This has little effect on the results.

19
Models 7-8 also replace GDP per capita and urbanization with secondary school enrolment rates.12

While the impact of cross-country differences in schooling is positive, it is significant only in the

ordered logistic regressions (model 8) and has no effect on the impact or significance of the

individual-level indicator of poverty.

Models 9-10 control for country-level inequality, which has little impact on the results. Finally,

in models 11 and 12 we run country fixed effects models using linear and binary (conditional)

logistic regressions. While the fixed effects transformation precludes the estimation of country-level

variables by construction, the advantage is that it allows us to test the robustness of individual-level

regressors to the impact of omitted variables at the country-level. However, it is clear that including

country fixed effects has little effect on the main result, namely that individual-level poverty has a

positive and highly significant effect on voters’ exposure to vote buying efforts by political

operatives. Overall, these results suggest that poverty—understood as lack of regular access to basic

household necessities—has a robust effect on vote buying in Africa. Indeed, poverty appears to one

of the most important sources of differences in exposure to vote buying. This result corroborates the

evidence from single-country studies (Bratton 2008; Kramon 2009; Birch 2011), and suggests that

the relationship applies more generally.13

The substantial effect of poverty on vote buying is not only statistically significant, it is also

quite large. We illustrate this in two ways. In the logistic regressions in models 5-12, the log odds

are fairly stable and vary between 0.67 and 0.72. Based on model 5, this corresponds to an

estimated effect on the odds ratio of e0.67=1.95. Substantively, this means that a change from the

lowest to the highest value on the poverty index doubles the odds of being exposed to vote buying

during election campaigns. All other things being equal, a poor voter is therefore twice as likely to

12
Results from linear regressions give similar results.
13
In other models (not shown) we have looked at the interaction of poverty and urban-rural residence. While poverty
has a highly significant effect in both urban and rural areas, the effect is slightly stronger in urban areas.

20
be targeted by vote buyers as a voter who is materially well-off. To get a more firm sense of the

effect of micro-level poverty on vote buying, Figure 3 plots predicted probabilities of being targeted

by vote buying political operatives during elections (based on model 5).

[Figure 3 about here]

Figure 3 shows predicted probabilities of vote buying for different values of micro-level poverty

and at fixed values of the remaining variables, with the random effects set to zero. Since model 5 is

a non-linear probability model, the predicted probabilities depend on the values at which the

explanatory variables are evaluated. Specifically, we plot probabilities for a male person, who is 37

years old, employed, lives in an urban area, has completed secondary school, and who has values of

ethnic salience and contact with political officials corresponding to the sample average. Figure 3

shows three lines. The bold line shows predicted probabilities for an individual living in a median

country.14 The punctuated line shows the equivalent plot with the electoral vote margin is set to

zero. All else equal, this shows the effect of poverty on vote buying in countries where elections are

narrow and extremely contested—in the limit approximating a difference of zero between the vote

shares of the winning candidate and the runner-up. Finally, the dotted line shows the relationship

between poverty and vote buying when the winning party wins by a wide vote margin, as is

typically the case in, e.g., Botswana.

14
We use the median because the mean of GDP/cap. ($2230) is almost twice as large as its median value ($1172)––the
value of Kenya in 2005. This is mainly due to comparatively high levels of GDP in Botswana ($9306) and South Africa
($6961). This makes the mean a relatively poor representative of the ‘typical’ value of GDP/cap. in the sample. In fact,
no country comes close to the mean of GDP/cap., while several countries (Benin, Ghana, Lesotho, Uganda, and
Zambia) have levels of GDP/cap. close to the median.

21
 With values of the country-level variables fixed at their median, the bold line shows that the

probability of being offered rewards for votes is 14 percent for people who are well-off—those with

the lowest values on the poverty index. This probability increases to 24 percent as poverty increases

to its highest value—a difference of 10 percentage points. This illustrates that the likelihood of

being offered money in return for votes varies markedly with poverty. Moreover, the

competitiveness of elections aggravates this difference. The correlation between vote buying and

the competitiveness of elections is consistent with arguments in the literature emphasizing that

clientelism is an important instrument for parties competing in democratic elections (Kitschelt and

Wilkinson, 2007; van de Walle, 2007). Indeed, it is not surprising that vote buying is an integral

part of closely contested elections in Africa, given that political power in most African countries is

highly concentrated, mostly with strong presidents. Presidential elections in particular are zero-sum

games where ‘the winner takes all’. When elections become highly contested, there are clear

incentives for politicians to use party operatives and clientelistic networks to launch vote buying

campaigns. So while highly competitive elections may make politicians and governments more

responsive to voter demands (Hobolt and Klemmensen, 2008), they also have a dark side, creating

strong incentives for parties and candidates to use pre-election vote buying—and not just promises

of post-election redistributive transfers—as a means to mobilize voter support, at least in weakly

institutionalized and newly established democracies.

The punctuated and dotted lines in Figure 3 illustrate the impact of variations in electoral

competitiveness. When elections are closely contested—the punctuated line—the probability of

being offered money for votes increases to 23 percent even for voters living in good material

conditions, while the corresponding probability for poor voters increases to a 37 percent—

amounting to a 14 percentage point difference between the most well-off and poorest voter groups.

In contrast, when electoral competition is low—the dotted line—voters are much less likely to be

22
offered rewards in return for their votes—7 and 13 percent, respectively, for well-off and poor

voters. Therefore competitive elections not only affect average levels of vote buying across

countries, they also amplify the general tendency of political operatives to target and mobilize poor

voters by trading money and material rewards for votes. Consequently, vote markets seem to thrive

when elections are close, making poor and deprived voter groups even more likely targets of vote

buying campaigns by political parties.

6.2. Results for control variables

The results for the control variables largely correspond to theoretical priors. Among the individual-

level controls, being in contact with political party operatives increases the chance that people

report experiences with vote buying. While there is little to suggest that government bureaucrats are

involved in vote buying, encounters with local councillors and political party officials seem to

matter. Interestingly, direct contacts with Members of Parliament have little effect. This is

consistent with Stokes’ (2005, 2007b) observation that the organizational structure of clientelistic

parties is ‘bottom-heavy’ and relies on a large network of local party operatives. Voters’ party

identification also has a positive and highly significant correlation with vote buying in models 7-8,

suggesting that voters who identify with a political party are more likely to experience vote buying

attempts. The positive correlation between party identification and vote buying could imply that

political parties target party loyalists or use material rewards to increase turnout among supporters

who identify with the party but are disinclined to vote (Nichter, 2008).15 However, since the

15
An alternative explanation is presented by Hicken (2011: 298) who summarizes ongoing work by Dunning and
Stokes (2010). This explanation stresses that party leaders target swing districts and use vote brokers to target voters
within these districts. However, vote brokers prefer to target party loyalists within district because it is cheaper than
targeting (and buying the voters of) voters supporting the opposing party. Our results are consistent with the notion that
vote brokers target (loyal) voters who identify with a political party.

23
correlation might also reflect reverse causality – i.e. that voters who get bought develop a stronger

sense of party identification – we do not go into more detail about the issue here.16

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 surprisingly – 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 Gonzales-Ocantos et al. (2012: 212-214) argues that the positive correlation may arise

because better educated people are more likely to participate in politics, making them more

disposed to experience vote buying attempts by political parties.17 The positive correlation could

also reflect an education-induced social desirability bias in reported vote buying attempts. For

instance, the literature on electoral turnout shows that better educated people are more likely to

overreport voting at elections in order to conform to norms of ‘good’ behaviour in democratic

societies (Karp and Brockington, 2005). While this cannot be ruled out, we have replicated the

results in Table 1 in regressions that exclude respondents with 1) tertiary education, 2) secondary

and tertiary education, and 3) primary, secondary, and tertiary education. The latter reduces the

sample to people without formal schooling. This does not change the results for poverty. Detailed

results are available in the appendix.

16
In additional tests, we looked at whether partisanship matters more among the poor relative to wealthier people. The
results show that partisanship matters for both groups. We have also tested whether party identification matters more
when elections are close, by running regressions which include an interaction of (individual-level) party identification
and (country-level) electoral competitiveness. These results suggest that party identification does, in fact, matter more
when elections are close. Results are available upon request.
17
However, education may also have an indirect negative effect on vote buying if education contributes to reduce
poverty.

24
 The findings for the other socio-economic controls show that employment increases the

likelihood of experiencing vote buying, and that men and young people are significantly more likely

to experience vote-buying. Finally, ethnic identity is positively related to vote buying, although the

significance of the coefficient varies somewhat. This suggests that strong ethnic identities may

increase the likelihood of being a target of vote buying campaigns.

Other than electoral competitiveness, ethnic fractionalization is the strongest country-level

predictor of vote buying in Africa. That is, political parties operating in ethnically fractionalized

societies are much more likely to use vote buying as an instrument to secure re-election than parties

in more homogenous societies. The details of the electoral system also seem to matter (models 7-8).

Consistent with the ideas of Persson et al. (2003), vote buying is most widespread in systems with

small electoral district magnitudes, where barriers to entry are high because the winning candidate

must gather support from a plurality in the constituency. Finally, urbanization, schooling, and

economic inequality (models 9-10) are mostly insignificant.

7. Conclusions

In his seminal study of machine politics in new post-colonial states, Scott (1969: 1150) stressed that

“perhaps the most fundamental quality shared by the mass clientele of machines is poverty”. More

than 35 years later, efforts to generate economic development as well as recent introductions of

electoral democracy in many African countries have not done much to alter this pattern. Indeed, the

findings of this paper suggest that the strong links between poverty and a particular form of

machine politics—vote buying—seem to persist. The key findings of this paper show that the

micro-economic conditions under which people live have large effects on the extent to which

African voters are targets of vote buying—an effect that seems to increase when elections are

25
closely contested. In the context of African democracies, it is perhaps not surprising that poverty is

the predominant socio-economic source of vote buying. In spite of recent economic progress and

the spread of competitive elections, poverty is still widespread (Poku and Mdee 2011). Poor voters

therefore constitute sizeable groups of the electorate in practically all of Africa’s democracies. This

may affect the way democracy operates in general (Mattes, 2008), the political issues voters are

concerned with (Justesen, 2011), and the prospects of developing and sustaining markets for buying

and selling votes during elections more specifically (Aidt and Jensen, 2012). And as the results of

this paper suggest, poverty also makes vote buying a more common political strategy during

elections. This also suggests that we should not dismiss the role of economic development at the

macro level, but that economic development may well make vote markets less vibrant if it

contributes to eradicate the material poverty that is instrumental for the operation of vote buying

political parties.

26
References

Acemoglu, D., Johnson, S., Robinson, J., Yared., P., Income and democracy. American Economic

Review 98 (3), 808-842.

Aidt, T.S., Jensen, P.S., 2012. From open to secret ballot: Vote buying and modernization.

Cambridge Working Papers in Economics 1102. University of Cambridge.

Aidt, T.S., Golden, M.A., Tiwari, D., 2011. Incumbents and criminals in the Indian legislature.

Cambridge Working Papers in Economics 1157. University of Cambridge.

Alesina, A., Devleesschauwer, A., Easterly, W., Kurlat, S., Wacziarg, R., 2003. Fractionalization.

Journal of Economic Growth 8, 155-194.

Alvarez, M., Cheibub, J.A., Limongi, F., Przeworski, A., 1996. Classifying political regimes.

Studies in Comparative International Development 31 (2), 3-26.

Beck, T., Clarke, G., Groff, A., Keefer, P., Walsh, P., 2001. New tools in comparative political

economy: The database of political institutions. The World Bank Economic Review 15 (1):

165-176.

Birch, S., 2011. Electoral Malpractice. Oxford University Press, Oxford.

Blaydes, L., 2006. Who votes in authoritarian elections and why? Determinants of voter turnout in

contemporary Egypt. Paper presented at the Annual Meeting of the American Political Science

Association, 2006, Philadelphia, PA.

Bratton, M., van de Walle, N., 1997. Democratic Experiments in Africa: Regimes Transitions in

Comparative Perspective. Cambridge University Press, Cambridge.

27
Bratton, M., Mattes, R., Gyimah-Boadi, E., 2005. Public Opinion, Democracy, and Market Reform

in Africa. Cambridge University Press, Cambridge.

Bratton, M., 2008. Vote buying and violence in Nigerian election campaigns. Electoral Studies 27

(4), 621–632

Bratton, M., Bhavnani, R., Chen, T-H., 2012. Voting intentions in Africa: Ethnic, economic or

partisan. Commonwealth and Comparative Politics 50 (1), 27-52.

Brusco, V., Nazareno, M., Stokes, S.C., 2004. Vote buying in Argentina. Latin American Research

Review 39 (2), 66-88.

Bryan, S., Baer, D., 2005. Money in Politics: A study of Party Financing Practices in 22 Countries.

National Democratic Institute for International Affairs. http://www.ndi.org/

Chandra, K., 2007. Counting heads: a theory of voter and elite behavior in patronage democracies.

In: Kitschelt, H., Wilkinson, W.I. (Eds.), Patrons, Clients and Policies: Patterns of Democratic

Accountability and Political Competition. Cambridge University Press, Cambridge, pp. 84-109.

Cheibub, J.A., Ghandi, J, Vreeland, J.R, 2010. Democracy and dictatorship revisited. Public Choice

143(1-2), 67-101.

Coalion for Accountable Political Finance (CAPF), 2008. The Money Factor in Poll Race: A

Monitoring Report of the 2007 General Election. www.capf.or.ke

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.

Dixit, A., Londregan, J. 1996. The determinants of success of special interests in redistributive

politics. Journal of Politics 58 (4), 1132-1155.

28
Dunning T., Stokes S. 2010. How does the internal structure of political parties shape their

distributive strategies? Paper presented at workshop on Political Parties in the Developing

World, Princeton University, April 30–May 1.

Eifert, B., Miguel, E., Posner, D.N., 2010. Political competition and ethnic identification in Africa.

American Journal of Political Science 54 (2), 494-510.

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 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., (Eds.), Election Observation and

Democratization in Africa. Macmillan Press, Basingstoke, pp. 122-149.

Gonzalez-Ocantos, E., Kiewiet de Jonge, C., Meléndez, C., Osorio, J., Nickerson, D.W., 2012. Vote

Buying and Social Desirability Bias: Experimental Evidence from Nicaragua. American

Journal of Political Science 56 (1), 202–217.

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.

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 buying? In: Schaffer, F.C., (Ed.),

Elections for Sale: The Causes and Consequences of Vote Buying. Lynne Rienner Publishers,

Boulder, pp. 47-60.

29
Hicken, A.D. 2011. Clientelism. Annual Review of Political Science 14, 289-310.

Hobolt, S.B., Klemmensen, R., 2008. Government responsiveness and political competition in

comparative perspective. Comparative Political Studies 41 (3), 309-337.

Jacobs, K, Spierings, N., 2010. District magnitude and voter turnout: A multilevel analysis of self-

reported voting in the 32 Dominican districts. Electoral Studies 29 (4), 704-718.

Justesen, M.K. 2011. Too poor to care? The salience of AIDS in Africa. Afrobarometer Working

Paper No. 133. www.afrobarometer.org

Justesen, M.K., Bjørnskov, C., 2012. Exploiting the poor: Bureaucratic corruption and poverty in

Africa. Afrobarometer Working Paper No. 139. www.afrobarometer.org

Karp, J.A., Brockington, D., 2005. Social desirability and response validity: A comparative analysis

of overreporting voter turnout in five countries. Journal of Politics 67 (3), 825-840.

Keefer, P., 2007. Clientelism, credibility, and the policy choices of young democracies. American

Journal of Political Science 51 (4), 804-821.

Kitschelt, H., Wilkinson, S.I., 2007. Citizen-politician linkages: an introduction. In: Kitschelt, H.,

Wilkinson, W.I. (Eds.), Patrons, Clients and Policies: Patterns of Democratic Accountability

and Political Competition. Cambridge University Press, Cambridge, pp. 1-49.

Koter, D., 2013. King Makers: Local Leaders and Ethnic Politics in Africa. World Politics 65 (2),

187-232.

Kramon, E., 2009. Vote-buying and political behavior: estimating and explaining vote-buying’s

effect on turnout in Kenya. Afrobarometer Working Paper no. 114. www.afrobarometer.org

30
Lehoucq, F., 2007. When does a market for votes emerge? In: Schaffer, F.C., (Ed.), Elections for

Sale: The Causes and Consequences of Vote Buying. Lynne Rienner Publishers, Boulder, pp.

33-46.

Lemarchand, R., Legg, K., 1972. Political clientelism and development: A preliminary analysis.

Comparative Politics 4 (2), 149-178.

Lieberman, E., Singh, P., 2012. The institutional origins of ethnic violence. Comparative Politics 45

(1), 1-24.

Lindberg, S.I., 2003. ‘It's our time to "chop”’: Do elections in Africa feed neo-patrimonialism rather

than counter-act it? Democratization 10 (2), 121-140.

Linos, E. (2013). Do conditional cash transfer programs shift votes? Evidence from the Honduran

PRAF. Forthcoming in Electoral Studies.

Mattes, R., 2008. The material and political bases of lived poverty: Insights from the

Afrobarometer. CCSR Working Paper No. 216. University of Cape Town.

Medina, L.F., Stokes, S.C., 2007. Monopoly and monitoring: An approach to political clientelism.

In: Kitschelt, H., Wilkinson, W.I. (Eds.), Patrons, Clients and Policies: Patterns of Democratic

Accountability and Political Competition. Cambridge University Press, Cambridge, pp. 68-83.

Mwangi, O.G., 2008. Political corruption, party financing and democracy in Kenya. Journal of

Modern African Studies 46 (2), 267–285.

Nichter, S., 2008. Vote buying or turnout buying? Machine politics and the secret ballot. American

Political Science Review 102 (1), 19-31.

31
Osei, A., 2012. Party-voter linkage in Africa: Ghana and Senegal in Comparative Perspective.

Springer VS, Wiesbaden.

Persson, T., Tabellini, G., Trebbi, F., 2003. Electoral rules and corruption. Journal of the European

Economic Association 1 (4), 958-989.

Poku, N.K., Mdee, A., 2011. Politics in Africa: A New Introduction. Zed Books, London.

Rabe-Hesketh, S., Skrondal, A., 2008. Multilevel and Longitudal Modeling Using Stata, 2nd edition.

Stata Press, Texas.

Robinson, J., Parsons, N.Q., 2006. State formation and governance in Botswana. Journal of African

Economies 15, 100-140.

Robinson, J., Verdier, T., 2013. The political economy of clientelism. Scandinavian Journal of

Economics 115(2), 260-291.

Scott, J., 1969. Corruption, machine politics, and political change. American Political Science

Review 63 (4), 1142-1158.

Sisson, R., 1972. The Congress Party in Rajasthan: Political integration and institution-building in

an Indian state. University of California Press, Berkeley.

Solt, F., 2009. Standardizing the World Income Inequality Database. Social Science Quarterly 9,

231-242.

Stokes, S.C., 2005. Perverse accountability: A formal model of machine politics with evidence from

Argentina. American Political Science Review 99 (3), 315-325.

Stokes, S.C., 2007a. Political clientelism. pp. 604-627. In: Boix, C., Stokes, S.C., (Eds.), The

Oxford Handbook of Comparative Politics. Oxford University Press, Oxford.

32
Stokes, S.C., 2007b. Is vote buying undemocratic? In: Schaffer, F.C., (Ed.), Elections for Sale: The

Causes and Consequences of Vote Buying. Lynne Rienner Publishers, Boulder, pp. 81-99.

Van de Walle, N., 2007. Meet the new boss, same as the old boss? The evolution of political

clientelism in Africa. In: Kitschelt, H., Wilkinson, W.I. (Eds.), Patrons, Clients and Policies:

Patterns of Democratic Accountability and Political Competition. Cambridge University Press,

Cambridge, pp. 50-67.

Vicente, P.C., 2013. Is vote-buying effective? Evidence from a field experiment in West Africa.

Forthcoming in Economic Journal.

Vicente, P.C., Wantchekon, L., 2009. Clientelism and Vote Buying: Lessons from Field

Experiments in African Elections. Oxford Review of Economic Policy 25 (2), 292-305

Wantchekon, L., 2003. Clientelism and Voting Behavior: Evidence from a Field Experiment in

Benin. World Politics 55 (3), 399-422.

33
 Figure 1. Vote buying in Africa
Afrobarometer countries
42.5
40

35.4
33.4 32.6
30

28.6
26.8
23.5
20

16.7

12.1
10.3 9.4
10

8.5
6.8 6.3
5.4 5.1
1.8 1.2
0

a a in ar li ia ia e a e i ia al ia a e a o
eny and en sc ma ger mb bw han iqu alaw ib eg an fric erd an oth
k ug b ga
a ni za ba g mb m nam sen anz th a e v tsw les
m a t u ap bo
ad zi oz so c
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?
Source: Afrobarometer, round 3

34
 Figure 2a. Vote buying in Africa: Four-point scale
80
60
40
20
0

) ) ) )
(0 (1 (2 (3
er ic
e es n
v fte
N
e tw tim O
or w
e fe
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?
Source: Afrobarometer, round 3

Figure 2b. Vote buying in Africa: Dichotomous variable

)
(1
(0
d

ed
re

r
fe

fe
of

of
ds

ds
ar
ar
w

ew
re

R
o
N

Vote buying: Dichotomous

35
 Figure 3. Poverty and vote-buying: Predicted probabilities
0.4

0.35

0.3
Narrow vote margin
0.25 Median country
Pr(Y=1|X)

Wide vote margin

0.15

0.1

0.05

0
0 0.25 0.5 0.75 1
Well-off Poverty Poor

36
Table 1. 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 Logit Ologit ML linear Logit Ologit Logit Ologit Logit Ologit Linear FE Logit
RE RE RE RE RE RE RE RE RE RE 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*
(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)
37
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)
Cut 2 5.21*** 6.89*** 6.98*** 5.29***
(31.80) (28.98) (23.49) (17.82)
Cut 3 5.99*** 7.68*** 7.77*** 6.08***
(41.98) (33.54) (26.91) (21.37)

Random effects
Level-2 std. dev. 0.143 0.547 0.327 0.139 0.471 0.445 0.406 0.392 0.449 0.396 0.253
Level-1 std. dev. 0.768 √(π2/3) √(π2/3) 0.779 √(π2/3) √(π2/3) √(π2/3) √(π2/3) √(π2/3) √(π2/3) 0.781
ρ (intra-class correlation) 0.034 0.142 0.056 0.031 0.109 0.098 0.083 0.078 0.100 0.0795 0.095

Log (pseudo)likelihood -28125.2 -9954.5 -14722.2 -24139.4 -8924.5 -13359.7 -8523.5 -12752.8 -8923.7 -13358.5 - -8831.9
Observations 24,316 24,316 24,316 21,464 21,464 21,464 20,615 20,615 21,464 21,464 21,464 21,464
Countries 18 18 18 17 17 17 17 17 17 17 17 17
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 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 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 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 √(π2/3) (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. *** p<0.001, ** p<0.01, * p<0.05.

38