Figure 7.4 presents graphical results of regression analyses that include a triple interaction between treatment, coethnicity, and respondent income. Table G.1 presents the full regression results. In each model, the sum of the coefficients on the Electoral Clientelism Treatment variable and the Electoral Clientelism Treatment×Coethnic interaction termrepresents the treatment effect among the candidate's poorest coethnics. The triple interaction, Electoral Clientelism Treatment × Coethnic × Income, tells us what happens to this treatment effect among coethnics as participant wealth increases. In each model, the pattern of results is the same. The treatment effect is big and positive among the candidate's poor coethnics, and the treatment effect decreases with income. Among non-coethnics, on the other hand, the electoral clientelism treatment effect is negative among voters at all income levels.
Table G.2 examines how voter income, clientelism, and coethnicity interact to shape prospective expectations. Since the triple interactions are difficult to interpret substantively, the main result is presented graphically in Figure 7.6. Consistent with the patterns presented above, the results on prospective expectations are driven primarily by the poorer coethnics of the candidate. The treatment effect among coethnics decreases with participant income and eventually becomes about equal to zero at higher income levels.
All models are OLS. The sample includes only those who heard either about a coethnic or about a non-coethnic. Participants who heard about a non-coethnic are the omitted reference category. The dependent variable in column 1 is participant degree of agreement with the statement: “I would vote for the candidate.” The dependent variable in column 2 is participant degree of agreement with the statement: “All candidates should be like the one in the recording.” The dependent variable in column 3 is participant degree of agreement with the statement: “I would like the candidate to run in the next election.” The dependent variable in column 4 is the mean of all three support measures.
Column 1 dependent variable is degree of agreement with the statement: “The candidate will help people like you who are living in poverty.” Column 2 dependent variable is degree of agreement with the statement: “If the candidate wins, he will help people in an emergency.” Column 3 dependent variable is degree of agreement with the statement: “If the candidate wins, I will receive resources such as cash or food.” Column 4 dependent variable is the mean response to each of these three responses.
Chapter 4 relies in part on Afrobarometer survey data collected in Kenya. The Afrobarometer round 5 survey was conducted in Kenya in November 2011 and includes responses of 2,399 individuals. Each Afrobarometer survey is designed to be nationally representative. Survey questions probe attitudes and experiences related to political, economic, and social life.
In this appendix, I describe in further detail the survey questions and measures used in Chapter 4. Along with the respondent's report of whether they voted in the 2007 election as well as their age, gender, and residence in a rural location—each which fall straightforwardly from survey responses—I create the following variables.
Offered bribe or gift in 2007. To generate this measure, I used the survey question that asked: “And during the last national election in 2007, how often, if ever did a candidate or someone from a political party offer you something, like food or a gift or money, in return for your vote?” I created a dichotomous measure that takes a value of 1 if the respondent reports that this happened to them once or more, and 0 otherwise: 32 percent of the sample report receiving a bribe or gift in 2007.
Ballot not secret. To generate the measure of perceptions of ballot secrecy, I used the survey question that asked: “How likely do you think it is that powerful people can find out how you voted, even though there is supposed to be a secret ballot in this country?” Responses could range from Not at all likely (0), Not very likely (1), Somewhat likely (2), and Very likely (3)—82 percent of the sample reports Not at all likely.
Freedom to vote without pressure. To generate the measure of how free or autonomous respondents feel with respect to their voting decisions, I used the question that asked: “In this country, how free are you to choose who to vote for without feeling pressured?” Responses range from Not at all free (1), Not very free (2), Somewhat free (3), and Completely free (4)—78 percent responded that they are completely free.
Feels close to a political party. To determine whether the respondent has a partisan identity, I used the survey question that asked: “Do you feel close to any particular political party?” Responses are either yes or no—57 percent report that they feel close to a particular party.
Why do politicians hand out money during elections in Kenya? According to a resident of Nairobi, “they insist on this money we are giving you is just a start of the many projects that I will do once you elect me in office … they are trying to show that by givingmoney they will bring better amenities to the people” (Nairobi-02, 2015). “It's a show of development and of concerns to the people,” said a respondent from Kisii (Kisii-02, 2015). A respondent from Kakamega in Western Province explained that “by giving peoplemoney, politicians show that they will in future be able to take good care of their people, provide them with resources, create employment, and improve their living standards” (Kakamega-04, 2015).
This chapter addresses a simple but important question: Are these Kenyans right? I examine this question by investigating the relationship between electoral clientelism and the provision of local public goods in the period following the elections. I address two questions: First, is electoral clientelism associated with lower overall levels of public goods investment in an electoral district? That is, do politicians expend more or less effort and resources providing local public goods in contexts where electoral clientelism is prevalent? Second, do communities within electoral districts that experience more electoral clientelism receive fewer public goods in the period following the campaign? The goal is thus to examine how electoral clientelism relates to the overall provision and distribution of public goods.
In answering these questions, this chapter serves a dual purpose. First, the chapter tests empirical implications of the informational theory advanced in this book. Indeed, the theory produces potentially counter-intuitive implications with respect to the relation between electoral clientelism and public goods provision. If electoral handouts convey information to voters about the future flow of clientelist benefits, then we may observe more local public goods provision by MPs where electoral clientelism is concentrated. These predictions contrast with the notion that clientelism undermines public goods provision and that voters who can be mobilized with handouts during a campaign will be less likely to receive benefits in the future (Hanusch and Keefer, 2013; Kitschelt and Wilkinson, 2007; Stokes, 2007a).
This chapter serves two purposes. First, I introduce the Kenyan case. Second, I address a number of descriptive, but difficult to answer, questions about electoral clientelism in Kenya. Indeed the anecdotal evidence suggests that electoral clientelism is widespread in the country. However, important descriptive questions about electoral clientelism remain unanswered. How much electoral clientelism is there really? How much does it cost? Does electoral clientelism impact how people vote? Answers to these descriptive questions provide a crucial starting point for an analysis of electoral clientelism's causes and consequences.
I analyze original nationally representative survey data about electoral clientelism during Kenya's 2007 elections. To address the challenge of response bias, I implement a survey list experiment. As I detail below, survey list experiments provide an indirect method for eliciting less biased responses to sensitive survey questions. I complement the main survey findings with descriptive data from two sources: an in-depth survey I conducted in Kenya's Nairobi and Rift Valley Provinces; and data from an unusually rich campaign finance survey conducted with candidates for parliament in 2007.
The quantitative results confirm that electoral clientelism is widespread and expensive. About 20 percent of those I surveyed received a cash handout during Kenya's 2007 campaign. These handouts range in value from 1 to 20 US dollars. For political candidates, electoral clientelism is costly: the average candidate for parliament spent about 3 million Kenya shillings on electoral clientelism (about 48,000 US dollars), a substantial amount in a country where the GDP per capita is about 900 US dollars per year. This investment is by far the biggest campaign expenditure item, tripling the amount that candidates spend on campaign rallies, travel, and campaign materials. Results from the survey list experiment show, importantly, that electoral clientelism influenced the vote choice of about 20 percent of Kenyans in 2007.
Kenya is located in East Africa. With its eastern border along the Indian Ocean coastline, the country's neighbors include Tanzania to the south, Uganda to the west, and Sudan and Somalia to the north. As of 2014–2015, Kenya's GDP per capita was about 1,350 US dollars and the population was about 46 million.
Note that this is a rough and simplified version of the discussion that covers the main points in the recording. The discussion in the recording is in Swahili and follows this general framework, but is recorded in a vernacularway that would be familiar to listeners. As such, this is not an exact translation but rather a general outline of the discussion.
Person 1: Did you hear about the campaign rally in [insert location of survey] the other day?
Person 2: No, who was here?
Person 1: Oh, it was very interesting. A politician from [insert location] was here to try to gain support for his campaign to be a member of parliament.
Person 2: I see. What happened at the event?
Person 1: First some people from the local community spoke in support of his candidacy. There was also music and a lot of dancing.
Person 2: Ah, sounds like fun.
Person 1: Yes, and then he gave a speech to the crowd. He said that education is a big priority for him. He also said that he would work hard to bring roads, electricity, and clean water to our area. Everyone cheered when he complained that politicians have ignored us for too long and that we need jobs to make life better.
Person 2: I see. Did he mention corruption? Politicians always talk about corruption, but I wonder if they ever do anything.
Person 1: Yes, he did mention corruption. He promised he would do his best to make sure that the wananchi [the people] benefit from the country's resources, and that thieve can no longer be in government.
Person 2: Ah, I see. And does he have any qualifications? Person 1: He says that he does. He studied business at university, so he says that he can help make the economy grow.
Person 1: And after the speech, people from the campaign went out into the crowd handing out cash to people who were there.
Person 2: They handed out cash? I wonder how much they gave to people?
Person 1: My friend who was there got 500 Kenya shillings.
Person 2: Wow, sounds like a big event. How many people were there?
Person 1: About 1,000 people were there.
Person 2: That is a good turnout. I wonder if he will win the election.
Person 1: We will have to wait and see!
Prior to Kenya's March 2013 elections, the country's Human Rights Commission released footage of a secretly videotaped campaign event. The video features Ferdinand Waititu—a now-former member of parliament from Nairobi's Embakasi Constituency—speaking to voters at a campaign rally in the city's Donholm neighborhood. Waititu delivers a passionate speech in which he promises that he will bring jobs, help women and youth, and improve food security. After the address, he personally distributes money to members of the audience.
Many observers would call this “vote buying”: an attempt to directly exchange money for votes. This book shows that this is often amischaracterization. A central theme is that gift giving is often not a strategy to buy votes, but instead a mechanism through which politicians convey information to voters. According to the informational theory I advance, politicians such as Waititu distribute handouts to make their promises to deliver development resources and particularistic benefits to the communities that they representmore credible and, as a result, to connect with and persuade poor voters. Thus, what observers often interpret as “vote buying” is actually not a transaction at all.
In developing and testing this argument, this book contributes to the literature on electoral clientelism—the allocation of private and material benefits to voters during elections (Gans Morse et al., 2014). Electoral clientelism has been widespread in a range of contexts, including the late Roman Republic (Lintott, 1990; Yakobson, 1995) and elections in eighteenth- and nineteenth-century Britain and the United States (Bensel, 2004; O'Leary, 1962). In contemporary Latin America, electoral clientelism is prevalent in Argentina (Auyero, 2001; Brusco et al., 2004), Mexico (Magaloni, 2006), and Nicaragua (Gonzalez-Ocantos et al., 2012). In the Middle East, the strategy is important to elections in Egypt (Blaydes, 2010), Jordan (Lust-Okar, 2006), and Lebanon (Corstange, 2010). In Asia, electoral clientelism is pervasive in such countries as the Philippines (Khemani, 2012) and Taiwan (Wang and Kurzman, 2007). Electoral clientelism is common to election campaigns in much of Africa.1 In Ghana, for example, “campaigning is often about walking around various neighborhoods, talking to people about what they do and what their life is, while one of ‘the boys’ … continues to feed the MP with small notes for handouts from a small envelope” (Lindberg, 2003, p. 129).
Voting is based on the secret ballot. So there is no way the aspirant will know that you did not vote for him. Even if he knew that I did not vote for him, he will still not be able to punish me. It is very difficult.
No, [accepting money from one candidate and voting for another] is not immoral, after all no force is used upon an aspirant in order to give out money. Politicians give out money willingly.
Survey data highlight the prevalence of electoral clientelism to political campaigns in Kenya. In all regions of the country, politicians invest substantial amounts of money in allocating handouts to voters. What the survey data cannot tell us, however, is what electoral clientelism looks like in action. This chapter therefore examines the “mechanics” of electoral clientelism, addressing questions about how electoral clientelism is experienced in the context of political campaigns in Kenya. For example: where and how does electoral clientelism happen? Do politicians’ agents seek to monitor the voting behavior of those who receive handouts? Do voters think that they are being monitored? Do voters feel a moral responsibility to vote for those who give them handouts?
In addressing these questions, this chapter provides descriptive evidence that supports the informational theory. As I outline in Table 3.1, the theory generates a set of expectations about what electoral clientelism should look like in practice—where it happens, how voters get targeted, and what the relationships are between those who distribute handouts and those who receive them. To test these implications, I provide qualitative evidence from semi-structured interviews with voters and quantitative evidence from original surveys.
The evidence is consistent with the informational theory. It is also at odds with alternative explanations for the electoral effectiveness of electoral clientelism. For example, the first quote in the epigraph casts doubt on one plausible explanation: that parties directly enforce vote-buying transactions by violating ballot secrecy, or inferring how people vote, and credibly threatening to punish or deny future benefits to vote sellers that do not follow through with their end of the bargain. This Kenyan, a 24-year old woman who received 200 shillings at a political rally before the 2013 elections, expresses doubts that ballot secrecy might be violated and that parties can credibly punish voters.
This book argues for a shift in the way scholars often think about electoral clientelism during elections. Instead of focusing on electoral handouts as purely an element of direct exchange between candidate and voter, I argue that more attention should be paid to the signals handouts convey. Just as individuals often give one another gifts to signal goodwill, generosity, and awareness of social responsibility, I argue that politicians in Africa and elsewhere often distribute electoral handouts to signal to voters that they will represent their interests and distribute resources to them in the future. In short, electoral handouts are not always transactional: they can also be informational.
This chapter tests a central assumption of this theory: that electoral clientelism sends a credible signal to voters. Two challenges present obstacles to the empirical evaluation of this assumption. First is the problem of response bias. If we ask individuals what they think about candidates who distribute electoral handouts, they may give socially desirable answers which conform to norms that suggest that distributing handouts is immoral. The second challenge is presented by voters’ cognitive constraints. Individuals in all environments have difficulty identifying the determinants of their beliefs and behaviors, most of which result from complex cognitive processes about which they are often only partly aware (Nisbett and Wilson, 1977). Thus, even if individuals are trying to be honest, they may have difficulty to accurately explain the sources of their beliefs about political candidates. To overcome these challenges, I analyze data collected from a set of field experiments. These experiments, which were conducted with 1,000 randomly selected participants in two regions of Kenya, present participants with radio recordings describing hypothetical political candidates.
The recordings manipulate information about the candidate's engagement in electoral clientelism—in some recordings, the candidate distributes electoral handouts and in some he does not. As this information is subtle and randomly assigned to participants—ensuring that the only difference between the recordings and the populations in each experimental group is the short reference to electoral handouts—I am able to determine how electoral handouts influence electoral support and to identify the informational content of their signals. The results provide evidence that corroborates the informational theory.
E.0.1 Experiment 1
This section provides evidence about the sample analyzed in Experiment 1 (presented in Chapter 5). To draw the sample, I first identified sampling sites by mapping the sampling locations of the nationally representative Demographic and Health Surveys (DHS). I then selected sites within Nairobi and the surrounding area to maximize diversity in the types of locations sampled for the experiment. At each sampling site, participants were sampled using a random walk, a common approach to sampling in areas where the entire sample space cannot be enumerated.
Experiments are useful because, with randomization and a sufficiently large sample, treatment and control groups should “look the same.” It is this similarity that permits us to infer the causal influence of the treatment—in this case, the impact of the information about electoral handouts on electoral support and perceptions of political candidates. Table E.1 presents descriptive information and evidence of this covariate balance in treatment and control groups in Experiment 1. In the section above, I provide details about how each of these variableswas constructed. About 30 percent of those in both the treatment and control group report receiving an electoral handout ahead of the 2007 election. This is comparable to estimates from the nationally representative Afrobarometer survey. The table further shows that treatment and control groups are comparable on a number of covariates. There is slight imbalance is with respect to education, age, partisanship. More participants in the control group have some secondary education, more treatment group participants are aged 31–35, and the treatment group has more ODM supporters. I therefore augment the main analyses with regression analyses that include control variables.
E.0.2 Experiment 2
Table E.2 presents descriptive information about the sample for Experiment 2 (analyzed in Chapter 3). It also provides evidence of covariate balance in treatment and control groups. Participants in treatment and control groups are comparable with respect to their past experience with electoral clientelism, perceptions of ballot secrecy, partisan attachments, gender, age, income, and education. They are also comparable on a number of attitudinal measures, including attitudes about reciprocity, government redistribution, and Kenyans from other regions and ethnic groups. The previous section describes how the variables were constructed.
Politicians in Kenya are cunning; they have always been able to convince their tribesmen that if leadership is from their community they are in a better position to access government jobs, resources, infrastructure, and so on.
Scholarly and journalistic accounts of African politics often emphasize the important role of ethnic identity. The notion that ethnicity is important on the continent underpins dominant theories of voting behavior (Bratton and Kimenyi, 2008; Posner, 2005), elite coalition building (Arriola, 2009), the production and distribution of public goods (Franck and Rainer, 2012; Habyarimana et al., 2009), and the structure of economic markets (Fafchamps, 2002). Further, the salience of ethnicity in politics is often linked to a number of poor development outcomes, including slower economic growth (e.g., Easterly and Levine, 1997; Montalvo and Reynal-Querol, 2005) and the under-provision of public goods (e.g., Habyarimana et al., 2009; Miguel and Gugerty, 2005).
This chapter examines the connection between electoral clientelism and ethnic politics, testing the informational theory's implications for the local ethnic demographic conditions in which electoral clientelism is likely to bemost effective. In the Kenya context, where ethnicity is highly salient, the informational theory implies that electoral clientelism should have the most impact on voting decisions in conditions where political candidates cannot be distinguished by their ethnic identity. This hypothesis follows from the fact that, for both historical and psychological reasons, ethnic labels tend to send a signal to voters about how likely candidates will be to support their interests and channel resources to them (Chandra, 2007; Conroy- Krutz, 2012; Ferree, 2006; Posner, 2005). In races with a diverse candidate pool, electoral clientelism will convey information, but it will not be the only heuristic upon which voters can rely, diminishing its overall impact and importance. By contrast, in areas where candidates are all members of the same ethnic group—a large proportion of parliamentary and local government races in Kenya—voters cannot rely on ethnic labels to differentiate candidates. In these contexts, the information conveyed by electoral clientelism should be most influential.
The evidence in this chapter is drawn from two sources. First, I return to the list experimental and survey data about Kenya's 2007 elections introduced in Chapter 2. With those data, I show that electoral clientelism was most effective in ethnically homogenous districts where parliamentary candidates are generally members of the same ethnic group.
Personally, if the politician is a first timer I would not believe [his campaign promises] at all. But given the situation that the politician is someone who is coming for a second or third time, their background will greatly determine my outcome.
A central claim of this book is that electoral handouts convey information to voters. The previous chapters test different assumptions and implications of this argument, addressing questions about how electoral clientelism manifests in practice, how voters interpret the distribution of handouts, and which types of voters are most likely to be responsive to them. This chapter turns to two additional and important questions in the study of electoral clientelism. First, which types of political candidates are most likely to distribute electoral handouts? And, second, which types of candidates reap the greatest electoral benefits from electoral clientelism?
This chapter shows that the informational theory can help to answer these critical questions. Specifically, the theory implies that political candidates that are relatively unknown to voters or who lack other tools and time to establish a personal reputation for the delivery of resources are those who will invest most heavily in electoral handouts. They are also the types of candidates that will benefit most electorally from their use. This is because, as the epigraph to this chapter highlights, voters are likely to have the least developed beliefs about these types of candidates, which means that the information conveyed by electoral handouts will be most important in shaping voters’ views about them. In contrast, political candidates with well-established reputations due to their visibility or their time in office, should be less likely to rely on electoral handouts in the cultivation of electoral support. The electoral handout signal will be weak relative to all of the other information voters have about these types of established candidates.
I test these expectations by combining survey data about electoral handouts during Kenya's 2007 elections (detailed below) with official constituency level election results and original data collected about incumbents and challengers competing in the parliamentary elections. I show how these data allow me to overcome challenges that have hindered past attempts to use survey data to identify the influence of electoral handouts on voter behavior.
Electoral clientelism is a common, and often colorful, feature of elections in many low-income democracies around the world. The goal of this book has been to advance understanding of the role of electoral clientelism during elections, and to better understand the link between electoral clientelism and democratic accountability and responsiveness.
In the course of conducting fieldwork for this project, I found that the literature's dominant models of electoral clientelism were not sufficient to explain patterns of electoral clientelism in Kenya. I found that politicians use electoral handouts not to directly buy votes but instead to convey to voters their commitment to delivering resources to them and to serving their interests in the future. I further found that distributing money on the campaign trail is critical to being taken seriously as a candidate. Those who do not hand out cash are widely perceived as too weak to win and thus not worthy of electoral support. These observations form the basis of the informational theory.
While one goal of this book has been to develop a theoretical framework more consistent with patterns of electoral clientelism in Kenya, an additional goal has been to subject the theory to a set of diverse empirical tests that take seriously the inferential and measurement challenges involved in the study of clientelism. Most empirical research on clientelism is either purely qualitative or based upon survey data.While both can yield rich insights—the former was central to the development of the informational theory—they are both vulnerable to the critique that their inferences may be biased due to measurement problems, difficulties isolating clientelism's causal impact, or both. To address these empirical challenges, this book has adopted a pluralistic approach. I present evidence from indepth and semi-structured interviews with voters in seven of Kenya's eight provinces, a list experiment conducted with a nationally representative sample of 2,000 Kenyans, and a field experiment designed to identify electoral clientelism's causal impact and to assess the mechanisms through which clientelism influences voter behavior. While each of these approaches has its strengths and weaknesses, taken together the results paint a picture that is consistent with the informational theory.
According to the constitution, it is illegal. The constitution interprets [it] as corruption, as if you were buying the vote. But [he] does not buy your vote because he is a kind of a person who is ready to listen to your problems. If the problem touches his heart, he's ready to get in his pocket and help you according to your need, but not to buy your vote.
This chapter advances a theory to explain the prevalence and effectiveness of electoral clientelism. I test the implications of this theory, as well as some of the assumptions that underpin it, in the empirical chapters that make up the remainder of this book. My theoretical framework departs from the dominant approach in the political economy literature on electoral clientelism. While existing theoretical models differ in important ways, including in some of their empirical predictions and normative implications, they are unified in that they apply a transactional framework. Each suggests that electoral handouts are used to purchase desired political behaviors from voters. The objects of purchase—the vote, turnout, abstention—are the subjects of debate, as are the mechanisms of contract enforcement.
Yet, as the quote above illustrates, voters in some contexts do not conceptualize electoral handouts in a transactional way. Indeed, this Kenyan explicitly rejects the notion that the candidate from his constituency distributes money in an attempt to buy votes. This suggests that a transactional framework may be too narrow to understand fully the causes and consequences of electoral clientelism.
Building upon observations from fieldwork and ethnographic literature on elections in Africa, I advance an alternative framework that conceptualizes electoral handouts as gifts that contain social information. I argue that politicians distribute electoral handouts to convey information to voters who are uncertain about which candidates will best serve their interests. More specifically, electoral clientelism is effective because it conveys information that helps politicians to establish credibility with respect to the future provision of resources to the poor. For a number of reasons that I outline in detail below, credibility in this respect is important for political candidates in much of Africa.
F.0.1 Prospective Expectations: Results on Individual Survey Items
Figure 5.3 presents results of experimental analyses in which the dependent variable is a mean index capturing participants’ prospective expectations of the candidates presented in the field experiment. This section presents the results on each of the individual items that make up the index.
F.0.2 Norms of Reciprocity Tests
This section presents the measures and results used to address the social norms of reciprocity alternative explanation. I first describe the measures and then present the results.
I measure positive reciprocity, which captures the degree to which someone is willing to incur costs to pay back those who have helped them, and negative reciprocity, which captures the degree to which someone is willing to incur costs to seek revenge against someone who has wronged them. To capture positive reciprocity, I measure the degree of agreement (on a 5-point scale) with the statements: (1) “If someone does me a favor, I am prepared to return it,” and (2) “I will go out of my way to help somebody who has been kind to me before.” To measure negative reciprocity, I assess participants’ degree of agreement with the following statements: (1) “If somebody insults me, I will insult him or her back,” and (2) “If somebody puts me in a difficult position I will do the same to him or her.” I average the two responses in each category to get measures of positive and negative reciprocity, and average those two scores to produce a measure of average reciprocity. I present results using all three measures— average, positive, and negative—in part because positive and negative reciprocity are weakly correlated (r = −0.1) and because they may capture different traits (Dohmen et al., 2009). These measures are admittedly less precise than behavioral measures, such as those used in Finan and Schechter (2012). This is in part because they are prone to response bias and in part because the questions are open to interpretation and therefore provide a less uniform decision-making environment. However, they are widely used survey-based measures of positive and negative reciprocity.
The results are presented below. Columns 1, 3, and 5 present the interaction between treatment and each of the individual reciprocity measures in models without control variables.
Email your librarian or administrator to recommend adding this to your organisation's collection.