An Inquiry into the Use of Illegal Electoral Practices and Effects of Political Violence and Vote-buying

Abstract

This article investigates whether vote-buying and the instigation of violence in the disputed 2007 Kenyan elections were strategically motivated and whether those affected by vote-buying or post-electoral violence changed their views toward ethnopolitics and the use of violence. To answer these questions, a panel survey conducted before and after the elections is combined with external indicators of electoral violence. We find that vote-buying was used to mobilize parties’ own strongholds. Political parties also targeted vote-buying and threats toward specific ethnic groups and areas, potentially to weaken the support of their political rivals. In addition, we find that the victims of post-electoral violence are more likely to identify in ethnic terms and support the use of violence. The non-victims of post-electoral violence, but who experienced vote-buying after our first survey are also more likely to support the use of violence.

Keywords

electoral violence vote-buying election fraud ethnic identity Kenya

Despite the progress in the introduction of political competition, a number of recent elections in Africa have been illegitimate and violent. This has called into question whether democracy is in retreat in the continent and how competitive electoral processes can be made to work in countries deeply divided by ethnic allegiances (Lynch and Crawford 2011). The debate is particularly pertinent as ethnic allegiances and conflicts have tended to strengthen at election times (Eifert, Miguel, and Posner 2010), with political parties relying on illegal electioneering strategies, which the generally weak electoral commissions and judiciary system have done little to prevent (Bratton 2008). Thousands of people have died as a result of electoral violence, putting into jeopardy the democratic and modest economic progress that some African countries have recently enjoyed. To assess this issue, it is crucial to analyze when political parties are more likely to resort to illegal practices and their consequences on the future of democracy in the continent.

In this article, we undertake a micro-level analysis of Kenya, in order to shed light on a macro-level phenomenon: electoral violence and vote-buying. The objectives of this article are threefold. First, to analyze whether political parties employed illegal electioneering practices strategically before and after the 2007 general elections. Second, to investigate whether vote-buying as well as the post-electoral violence affected people’s ethnic identity, their support for democracy, and the acceptability for the use of violence. Third, to link the micro-level findings to the macro-level debate on how the electoral process in Africa could be improved.

To explore these research questions, we conducted two surveys, one just two weeks before the 2007 elections and a second one in the summer of 2008 reinterviewing the previous respondents. We also use external data sources to measure the death toll and number of injured people. These indicators were obtained from the Commission in Kenya in charge of investigating the 2008 post-electoral conflict (Commission of Inquiry on Post Election Violence [CIPEV]) and were also estimated independently by monitoring Kenyan media outlets (newspapers, radio, and TV stations).

Recent literature has described the extent to which Kenyans were affected by the post-electoral violence in terms of personal injury, economic loss, and displacement (Anderson and Lochery 2008; CIPEV 2008; Human Rights Watch 2008; Yamano, Tanaka, and Gitau 2010). Previous studies have also identified that those living in areas where gangs operated, with alleged links to political parties, had a higher risk of experiencing post-electoral violence (CIPEV 2008; Dercon and Gutiérrez-Romero 2012). This article aims to contribute to this literature by analyzing whether vote-buying, the operation of political gangs, and the instigation of violence by political actors were strategic. The article also aims to go beyond previous studies by assessing to what extent having experienced vote-buying and separately violence at the personal or the district level has affected the stance of people on key issues such as the role of ethnopolitics. Since one cannot observe, for instance, what would have happened to the opinions of those affected by violence if the violence had not happened, the impact of post-electoral violence is assessed using a difference-in-difference estimator. This method allows the buildup of a counterfactual, where the change of opinions of those affected by violence is compared to those of individuals not affected by violence. We use the same method to assess the impact of those that experienced vote-buying.

The article finds that both the Party of National Unity (PNU) and the Orange Democratic Movement (ODM), the two leading political parties contending in the 2007 Kenyan presidential elections, used vote-buying practices extensively, targeting the supporters of their closest rivals and their own strongholds. The PNU, which backed the reelection bid of the incumbent president, was more frequently mentioned to have threatened people in areas with higher electoral competition. This was done potentially to weaken the support of the PNU rivals.

The electoral ordeal reduced the desire of the general population for holding elections. Among those affected by the post-electoral violence more now identify in ethnic terms and are more likely to accept the use of violence in support of a just cause. The non-victims of post-electoral violence but who experienced vote-buying after our first survey are also more likely to support the use of violence.

Previous studies have found that economic indicators, such as a fall in gross domestic product (GDP), are more important determinants of civil conflict than ethnic diversity (Collier 2007). Kenya actually experienced its highest rates of economic growth ahead of the disputed election. Thus, the Kenyan case provides an example where weak electoral institutions and parties seeking political profit at the expense of instigating ethnic divisions can overturn the democratic progress and lead to conflict, even in a phase of economic prosperity.

The article continues as follows: the next section provides an overview of the institutional failures that led to the post-electoral violence. The third section describes the data sources used, the instances in which political parties instigated violence and vote-buying. The fourth section presents evidence on whether parties behaved strategically and the effects of being a victim of post-electoral violence and vote-buying. The last section presents the conclusions.

The Buildup toward Post-electoral Violence

Kenya transferred power peacefully from the long-ruling Kenya African National Union (KANU) and Daniel arap Moi to Mwai Kibaki and the National Rainbow Coalition (NARC) in the elections of 2002. The optimism from this step toward democratic consolidation was short-lived. In the subsequent general elections of December 27, 2007, the political campaign appeared to be competitive, peaceful, and Kenyans turned up to vote in large numbers (Gibson and Long 2009). Opinion polls ahead of the elections revealed that Kibaki’s performance evaluations were high among the general population (Bratton and Kimenyi 2008). However, the three main presidential candidates were overwhelmingly supported by people from their own ethnic groups: the incumbent president Mwai Kibaki by the Kikuyus, Raila Odinga by the Luos, and Kalonzo Muzyoka by the Kambas. Other ethnic groups also divided their support (Table 1).

Table 1.

Distribution of Electoral Irregularities.

	Who would you vote for president?		Received offer for vote	Threatened to vote in certain way	Politicians instigated violence in community		Heard of gangs operating in community		Victim of violence (physical or economical)
	Kibaki	Odinga	Before election	Before election	Before election	After election	Before election	After election	After election
Kikuyu	90.2	6.1	26.1	8.9	33.2	34.2	30.0	35.4	21.8
Luo	3.7	94.9	25.0	10.8	26.4	36.5	25.6	36.5	24.4
Luhya	22.8	69.5	33.7	14.7	26.8	32.5	49.2	60.2	26.7
Kamba	18.3	1.0	23.2	5.4	31.3	14.8	13.0	18.3	11.3
Meru	88.9	5.6	31.3	8.1	23.2	3.2	1.1	3.2	3.2
Kissi	26.6	64.9	33.0	8.3	27.8	29.2	21.9	30.2	18.8
Kalenjin	14.3	80.6	16.5	7.8	28.2	41.3	25.0	43.5	30.4
Mijikenda	24.3	61.4	13.3	4.0	38.7	46.5	2.8	7.0	16.9
Somali	57.1	34.3	34.2	0.0	23.7	21.6	5.4	2.7	13.5
Others	34.5	56.6	30.3	10.1	37.0	19.8	14.4	21.6	13.5
National Average	40.1	46.7	26.9	9.0	29.9	29.1	23.8	31.6	19.6

Source: Author’s surveys.

Political parties conducted their campaigns in a confrontational manner, exploiting ethnic divisions (CIPEV 2008, 347-48). Several communities reported gangs intimidating people (Dercon and Gutiérrez-Romero 2012). Despite the heated political campaigning, only forty-one casualties as a result of the pre-electoral violence were reported.

Unprecedented levels of violence erupted shortly after the Election Day, once the electoral results were announced. Amid allegations of rigging, the incumbent president Kibaki (backed by the PNU) was declared the winner of the election with 51.3 percent of the votes compared to 48.7 percent for Odinga (backed by the ODM). A coalition government was agreed on February 28, 2008, after weeks of bloodshed. Kibaki was to remain the elected president, while Odinga would be the prime minister.

Several institutions failed and political choices made long before the elections contributing to the political crisis. Kibaki’s administration received criticisms that little efforts were made to tackle ethnopolitics, solve land disputes, and disarm political gangs which emerged with the introduction of multiparty politics in Kenya (CIPEV 2008). Interviews conducted by Human Rights Watch (2013, 50) in Kenya suggest that gang members, usually political party activists, see elections as an opportunity to sell their security services to politicians. The quid pro quo is either monetary payment or promised jobs and posts within the party. As a farmer in Central region, where the gang Mungiki operates, describes, “these gangs get paid by politicians. They move from door to door asking people to support one candidate. We do not argue with them. We just agree with what they say because arguing with them can cost you a life or that of your family members” (Human Rights Watch 2013, 46).

Illegal Electoral Practices: Panel Data and External Sources

To assess the scope of electoral irregularities and the effects of post-electoral violence, we conducted two nationally representative surveys, one two weeks before the 2007 general elections and a second one in August 2008 revisiting the previous respondents. The pre-electoral survey is based on a nationally and regionally representative sample of 1,207 individuals drawn from 77 of the 210 constituencies in forty-two of the seventy districts. The survey has a margin of sampling error of ±3 percent at a 95 percent confidence level. The sample captures the rural/urban split and its ethnic distribution is consistent with the Kenyan census. The post-electoral survey reinterviewed 54.3 percent of previous respondents. In order to make meaningful comparisons between the two surveys, the respondents from the pre-electoral survey that could not be reinterviewed were replaced by new respondents with the same overall characteristics (Table A1 in the Online Appendix).¹

At the time of the pre-electoral survey, most respondents claimed they were registered voters and that they planned to vote in the presidential election (93 percent). The responses to the question, “If elections were held tomorrow, which party’s candidate would you vote for as national President?” showed that 40.1 percent of respondents intended to vote for Kibaki and 46.7 percent for Odinga.² After the elections, the majority of respondents (61 percent) believed Odinga had won the election legitimately, while only 25 percent thought it was Kibaki.

Electoral Irregularities

To the question, “How free and fair do you expect the next elections of December 2007 to be?” only a small group expected that the elections would not be free and fair (11.4 percent). Nonetheless, Table 1 shows that vote-buying was widely spread before the elections, with 27 percent of respondents answering positively to “During the present campaign for the 2007 elections, did a candidate or someone from a political party offer you something, like food or a gift, in return for your vote?”

Direct threats from political parties were less common, with 9 percent of respondents answering positively to “During the present campaign for the December 2007 elections, has anyone threatened negative consequences to you in order to get you to vote a certain way?”

Close to 30 percent of respondents claimed, “Political representatives had been openly advocating violence in our community,” both before and after the elections. Nearly 24 percent of respondents before the elections answered positively to “Have you heard about violent groups such as gangs or youths connected with politics being active in your neighborhood.” This figure increased to 31.6 percent after the elections.

To assess whether the survey respondents were affected by the outbreaks of post-electoral violence the following was asked, “Were you personally affected in the outbreaks of violence after 2007 in any of the following ways?” Approximately 20 percent of respondents reported a specific personal impact after the elections, in terms of personal injury, being displaced from home, destruction of property, loss of jobs, or earnings. The Luos and the Kikuyus were among the most affected by the post-electoral violence. But, other ethnic groups also suffered from violence such as the Luhya, the Kissi, the Kalenjin, and the Mijikenda (Table 1).

Since it is difficult to gauge whether there is any systematic underreporting of violence in our surveys, we triangulated with other two independent sources described subsequently. We found that in all districts where violence occurred according to these independent sources, our respondents also reported violence or being injured in those areas.

Incidence of Violence according to CIPEV and the Media

The independent CIPEV estimated that 1,133 people died as a direct result of the post-electoral violence. These figures were obtained from reports of health institutions, police records, and witnesses who testified to the Commission (CIPEV 2008).

To independently verify the death toll and the incidences of violence, we monitored twelve of the major media outlets in Kenya on a twenty-four-hour and daily basis from December 1, 2007, until March 31, 2008.³ According to the media monitored, 1,128 people died as a result of the post-electoral violence. These estimates are very close to those obtained by CIPEV also at provincial (Table 2) and district levels.

Table 2.

Post-electoral Injuries and Death Toll.

	Injured	Death toll
	CIPEV	CIPEV	Media
Rift Valley	2,193	744	754
Nyanza	747	134	172
Central	0	5	5
Western	146	98	117
Coast	133	27	36
Nairobi	342	125	44
Total	3,561	1,133	1,128

Source: CIPEV(2008) and Kenyan media houses monitored by the Steadman Group.

Although some of the violence erupted spontaneously as a result of the belief that the election had been rigged, there were also premeditated attacks, involving politicians who enlisted criminal gangs even before the actual election (CIPEV 2008). According to the International Criminal Court (ICC), the violence in the Rift Valley was planned to attack supporters of President Kibaki after the election. In retaliation, organized gangs like Mungiki and the police were given authorization to use excessive force to attack ODM supporters.

Political Party’s Strategic Behavior and the Effects of Violence and Vote-buying

According to the literature, political parties wishing to maximize their profit may choose to employ illegal electoral strategies to advance their interests, taking into account their budget constraints and the strength of electoral institutions (Collier and Vicente 2012; Synder 2000). However, there is no theoretical consensus on the extent to which political parties relying on ethnic affiliations, like in Kenya, will focus their courting efforts on their supporters. For instance, Cox and McCubbins (1986) develop a theoretical model predicting that parties with a strong attachment to a particular group, an ethnic one for instance, will devote resources exclusively toward their supporters. According to this model, trying to lure swing voters is a risky strategy since parties cannot be sure whether their offers will be effective. Nitcher (2008) also argues that parties seeking to increase their turnout will seek to mobilize their own supporters, in what he defines as “turnout-buying.” In contrast, a theoretical model developed by Stokes (2005) suggests it is not an optimal strategy to devote resources exclusively to groups whose votes are already guaranteed. Instead, Stokes’s model predicts that political parties will target vote-buying toward “moderate voters” who are relatively indifferent between the contending candidates, thus having a higher chance of being influenced.

In Kenya, it is possible that political parties devoted resources toward swing voters given that in order to win the presidential election, 25 percent of the vote needs to be secured in five out of the eight provinces. Otherwise, the election would have to go into a second round between the two front runners. Nonetheless, parties might have had incentives to also devote resources to their own strongholds given how closely contested the presidential election was. Thus, using the definition of Nitcher (2008) parties might have combined “vote-buying” strategies toward swing voters and also “turnout-buying” in their own strongholds.

Collier and Vicente (2012) develop a more comprehensive model distinguishing between vote-buying, intimidation tactics, and rigging. The model assumes that swing voters can be effectively persuaded not to vote if presented with violence. Their model predicts that the chances of political parties relying on violence increases in areas where they have less electoral support, where the targets of intimidation will be swing voters. The instigation of violence is then used as a strategy of “damage control,” where political parties expect to deter people from voting for their rivals and since the core supporters might not be persuaded not to vote, the intimidation efforts are targeted toward the swing voters. This model also predicts that in areas where the incumbent party has electoral support, the party will prefer to use bribery or ballot fraud. Indeed, in Kenya, electoral observers reported the presidential turnout figures in Kibaki’s strongholds to be suspiciously high, and mysteriously higher than the observed turnout for the MP elections held on the same day. The same discrepancies in turnout were found in Odinga’s strongholds (Rice 2008). Based on the existing theories, the following four hypotheses will be tested:

Hypothesis 1: The ODM and the PNU devoted resources to “turnout-buying” in their own strongholds as a strategy to mobilize their own supporters.

Hypothesis 2: The ODM and the PNU devoted resources to “vote-buying” targeting people perceived to be “swing or moderate” voters as a strategy to weaken the support of rivals.

Hypothesis 3: Political actors were more likely to instigate violence in areas where their rival had an advantage, targeting particularly the “swing or moderate” voters.

Hypothesis 4: The strongholds of the PNU and ODM parties were more likely to have wider differences in turnout between the presidential and MP elections than other areas, suggesting ballot fraud.

Strategic Behavior of Political Parties

In this section, we estimate a series of probit regressions, as shown in equation (1) to estimate the factors correlated with the probability of political parties using illegal electoral practices. All results are shown as marginal effects.

\begin{aligned} P r (I = 1) \end{aligned} = ϕ (μ X_{i c} + δ_{1} {C l o s e n e s s o f t h e e l e c t i o n}_{i c} + δ_{2} {O D M h a s a d v a n t a g e o v e r P N U}_{i c} \times {C l o s e n e s s o f t h e e l e c t i o n}_{i c} + δ_{3} {P a r t i e s^{'} E l e c t o r a l S t r o n g h o l d s}_{i c}),

where I = 1 represents whether the respondent suffered an electoral malpractice, φ is the cumulative distribution function of the standard normal distribution. The covariates used, X, are the respondent’s ethnic origin, wealth,⁴ province of residence where the respondent i was living in constituency c in 2007. We use an ex ante measure of how close the respondents expected the presidential election to be using our pre-electoral survey. For this purpose, we asked, “Which party do you think will win the national presidential elections in December 2007?” and calculated the difference between the percentage of the two parties that got the most responses per constituency. We interact the expected closeness of the election variable with a dummy variable on whether the constituency expected the ODM to win more votes than the PNU in the presidential election. We also use the ethnic composition of the constituency to denote whether the constituency was an electoral stronghold of a main party, given that the parties contending are new and there are no previous electoral results.⁵

We focus exclusively on estimating the probability of respondents being victims of vote-buying and separately the probability of suffering intimidation by the two main political parties (the ODM and the PNU) as they accounted for the majority of illegal practices reported.⁶ To disentangle further the behavior of parties, we estimate separately the probability of respondents reporting being offered something in exchange for their vote in areas where these parties used vote-buying exclusively and in areas where parties mixed vote-buying and intimidation. The PNU and the ODM directly threatened relatively few people (sixty-two in our sample). In all constituencies where the PNU used threats it also used vote-buying and in some cases on the same individuals (44 percent of the people who received the threats). The same is true for the ODM where it threatened people (in twenty-two out of the twenty-four constituencies).⁷

Columns (1) and (3) in Table 3 show that respondents living in ODM and PNU strongholds had the same probability of receiving an offer for their vote in constituencies where the PNU used vote-buying exclusively and in constituencies where the ODM used vote-buying exclusively. Whenever the ODM combined both vote-buying and intimidation (column 4), respondents living in ODM strongholds had a higher probability of receiving an offer for their votes than those living in PNU strongholds. Thus, we find evidence to support Hypothesis 1 on “turnout-buying.”

Table 3.

Probit Marginal Effects of Respondents Being Approached by Main Political Parties.

	Received offer for vote from:				Received threats before election from:
	PNU in constituency where PNU tried vote-buying only (1)	PNU in constituency where PNU tried vote-buying and threaten (2)	ODM in constituency where ODM tried vote-buying only (3)	ODM in constituency where ODM tried vote-buying and threaten (4)	PNU in constituency where PNU tried vote-buying and threaten (5)	ODM in constituency where ODM tried vote-buying and threaten (6)
Difference in expected win presidential election constituency level^a	0.212*** (0.050)	−4.998* (2.874)	0.116 (0.075)	−0.091** (0.034)	−3.041** (1.342)	−0.148** (0.059)
More expected ODM to win over PNU × Expected win difference^a	−0.082 (0.053)	5.070* (2.673)	−0.016 (0.096)	0.438*** (0.095)	2.446** (1.234)	0.236** (0.115)
Ethnicity (Kikuyu reference)
Luo	0.074 (0.102)	−0.047 (0.128)	−0.110** (0.036)	−0.002 (0.082)	0.140 (0.129)	−0.056*** (0.017)
Luhya	0.125 (0.086)	0.048 (0.100)	−0.013 (0.098)	−0.006 (0.102)	0.094 (0.125)	−0.016 (0.027)
Kamba	0.095 (0.135)	0.095 (0.122)	−0.103 (0.110)	0.112 (0.188)		0.036 (0.075)
Meru	−0.008 (0.061)		−0.035 (0.045)	−0.203*** (0.019)	0.057 (0.236)	−0.084*** (0.013)
Kissi	0.379** (0.118)	0.186 (0.298)	−0.043 (0.061)	0.212 (0.131)	−0.002 (0.061)	−0.003 (0.036)
Kalenjin	0.029 (0.074)	−0.040 (0.079)	−0.064 (0.067)	−0.122*** (0.022)	0.085 (0.095)	−0.048*** (0.011)
Mijikenda	0.011 (0.086)	0.161 (0.131)	−0.016 (0.074)		−0.034 (0.053)
Somali	0.036 (0.122)		−0.101*** (0.026)
Others	0.116 (0.109)	−0.120** (0.057)	−0.070 (0.054)	−0.090** (0.041)	0.032 (0.100)	−0.060*** (0.016)
Ethnic stronghold (PNU reference)
Stronghold ODM	0.016 (0.040)	−0.240 (0.170)	−0.029 (0.039)	0.095* (0.053)	−0.631*** (0.141)	−0.105* (0.062)
Stronghold ODM-K	−0.008 (0.063)		0.028 (0.210)	−0.432*** (0.018)		−0.317*** (0.034)
Other	0.016 (0.045)		−0.035 (0.033)	0.222*** (0.052)		0.023 (0.082)
Wealth index	0.016 (0.090)	0.181 (0.139)	−0.021 (0.130)	0.405** (0.126)	0.024 (0.100)	0.143** (0.069)
Province fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Pseudo R ²	.04	.07	.05	.13	.12	.15
Observations	816	303	460	432	307	432

Source: Author’s surveys.

Note: ODM = Orange Democratic Movement; ODM-K = Orange Democratic Movement-Kenya; PNU = Party of National Unity.

Robust standard errors in parentheses clustered at constituency level.

^aExpected win difference taken from pre-election survey question, “Which party do you think will win the national presidential elections in December 2007?”

Significance level: *p < .10, **p < .05, ***p < .001.

We also find evidence to support Hypothesis 2 on vote-buying. Whenever the PNU mixed vote-buying and intimidation (column 2), it was more likely to do so in areas where more people expected the ODM to win over the PNU, perhaps to convey to these areas that the party would offer patronage goods in the future in order to lure voters in its favor. Both parties targeted specific ethnic groups within each area. For instance, in areas where the PNU used vote-buying exclusively, column (1), it targeted the Kissis, which could have been perceived as swing voters, given their relatively lower support for the ODM compared to other ethnic groups. In areas where the ODM used vote-buying exclusively, column (3), it was more likely to target the Kikuyus, the core supporters of the PNU, than the Luos. Thus, although swing voters were targeted for vote-buying, the core supporters of the parties’ rivals were also targeted.

Column (5) shows the PNU was more frequently mentioned threatening people in areas with higher electoral competition⁸ and in areas where more people expected the ODM to win over the PNU. The few reported threats issued by the PNU (twenty-eight) were mostly concentrated in areas where the ODM was expected to win. Only three respondents (of Kikuyu/Meru ethnicity) reported the PNU threatening them in the PNU’s own strongholds.

Column (6) shows the ODM focused their threats in areas where more people expected the party to win over the PNU, but was more likely to threaten the Kikuyu, core supporters of the PNU, than the Luo, the Meru, the Kalenjin, and the other smaller ethnic groups. In this case, we find mixed evidence to support Hypothesis 3, parties intimidated core supporters of their rivals, not necessarily moderate ones.

Next, we explore the factors correlated with some politicians instigating people to be violent before and after the elections. In neither of these two cases did the survey ask which specific candidates or parties were instigating the violence. Table 4, column (1) shows that before the elections, respondents living in ODM strongholds were less likely to report politicians instigating violence in their communities than those living in the strongholds of the PNU. However, after the elections the ODM and PNU strongholds reported equally that political actors instigated people to use violence (column 2).

Table 4.

Probit Marginal Effects of Respondents Reporting Gangs and Politicians Instigating Violence in Their Community.

	Politicians instigated violence		Heard gangs connected to politics		Heard of Mungiki
	Before elections (1)	After elections (2)	Before elections (3)	After elections (4)	Before elections (5)	After elections (6)
Difference in expected win presidential election constituency level^a	−0.030 (0.102)	−0.158 (0.130)	−0.157** (0.074)	−0.329* (0.170)	−0.133** (0.066)	−0.223** (0.081)
More expected ODM to win over PNU × Expected win difference^a	0.060 (0.102)	0.169 (0.137)	0.070 (0.087)	0.191 (0.167)	0.069 (0.068)	0.195** (0.085)
Ethnicity (Kikuyu reference)
Luo	0.035 (0.077)	0.040 (0.074)	0.109 (0.074)	0.085 (0.085)	0.028 (0.033)	0.027 (0.049)
Luhya	0.016 (0.062)	−0.011 (0.058)	0.046 (0.050)	−0.002 (0.068)	−0.016 (0.014)	−0.031 (0.020)
Kamba	0.014 (0.089)	−0.148** (0.052)	0.114 (0.094)	0.063 (0.103)	0.030 (0.032)	−0.014 (0.034)
Meru	−0.103 (0.070)	−0.201*** (0.054)	−0.146** (0.048)	−0.079 (0.112)	−0.028 (0.024)	−0.014 (0.055)
Kissi	0.041 (0.069)	−0.008 (0.076)	0.065 (0.066)	−0.010 (0.071)	0.051 (0.045)	0.003 (0.036)
Kalenjin	0.003 (0.068)	0.025 (0.088)	0.095 (0.063)	0.049 (0.058)	−0.005 (0.030)	−0.030 (0.026)
Mijikenda	0.147 (0.103)	0.024 (0.089)	−0.071 (0.078)	−0.125** (0.059)
Somali	−0.153** (0.054)	−0.085 (0.074)	−0.162*** (0.022)	−0.040 (0.142)	−0.042*** (0.009)
Others	0.038 (0.051)	−0.159*** (0.044)	0.017 (0.055)	−0.098** (0.047)	−0.009 (0.016)	−0.031 (0.020)
Ethnic stronghold (PNU reference)
Stronghold ODM	−0.132* (0.074)	−0.026 (0.057)	−0.036 (0.045)	−0.007 (0.085)	−0.012 (0.029)	−0.034 (0.039)
Stronghold ODM-K	−0.081 (0.104)	0.035 (0.109)	−0.043 (0.074)	0.004 (0.126)	−0.041** (0.016)	−0.033 (0.038)
Other	−0.049 (0.069)	−0.050 (0.050)	−0.068 (0.043)	−0.023 (0.076)	−0.044*** (0.011)	−0.036 (0.034)
Wealth index	0.106 (0.101)	0.088 (0.100)	0.077 (0.087)	0.050 (0.104)	0.012 (0.038)	0.061 (0.049)
Province fixed-effects	Yes	Yes	Yes	Yes	Yes	Yes
Pseudo R ²	.028	.098	.217	.220	.254	.194
Observations	1,207	1,184	1,188	1,184	1,117	894

Source: Author’s surveys.

Note: ODM = Orange Democratic Movement; PNU = Party of National Unity; ODM-K = Orange Democratic Movement-Kenya.

Robust standard errors in parentheses clustered at constituency level.

^aExpected win difference taken from pre-election survey question, “Which party do you think will win the national presidential elections in December 2007?”

Significance level: *p < .10, **p < .05, ***p < .001.

In Table 4, we also explore the involvement of organized gangs, allegedly hired by political actors. Columns (3) and (4) show that both before and after the elections respondents living in ODM and PNU strongholds were equally likely to report gangs operating in their areas. Similar results are found for people who reported the Mungiki gang, with alleged links to the PNU (columns 5 and 6). Both before and after the elections, that gang was equally likely to be reported in ODM and PNU strongholds. However, after the elections, this gang was more likely to have been reported in areas where more people expected the ODM to win the presidency over the PNU.

The irregular electoral practices analyzed so far were possible as the police and army failed to prevent these crimes (Human Rights Watch 2013, 2). The PNU, backing the reelection bid of President Kibaki, could have enjoyed an advantage in the use of state-controlled force over other parties. The police in Kenya are widely perceived as the most corrupt institution and in some instances in collusion with politically connected gangs (Human Rights Watch 2013, 17). Thus, we next explore the role of the police in the post-electoral violence.

In Table 5, we analyze the factors correlated with the police killing more people during the post-electoral violence in our sampled districts. We obtained the number of people that were killed as a result of bullets which were attributed to the police force and as a direct result of the post-electoral violence from the CIPEV report. As potential explanatory variables for police violence, we assess the role of the closeness of the elections at district level. We also include proxies for grievances, as there is international evidence that suggests rebellions may be explained by high inequality or ethnic divisions in society (Collier and Hoeffler 2004). Specifically, we use an index of the availability of three basic public goods (schools, hospitals, and police stations) near the respondent’s home in the districts we sampled and the polarization index proposed by Montalvo and Reynal-Querol (2003). Since there are no publicly available measures of ethnic diversity at district level in Kenya, we measure ethnic polarization using our pre-electoral survey.

P o l a r i z a t i o n_{d} = 1 - \sum_{g = 1}^{N} {(\frac{0.5 - s_{g d}}{0.5})}^{2} s_{g d},

where s_gd is the share of group g (g = 1 … N) in district d.

Table 5.

People Killed by Police per 100,000 Inhabitants in Sampled Districts and Discrepancies in Official Electoral Results in Sampled Constituencies.

	Number of people killed by police (bullets) after elections at district level^b(1)	Difference in official turnout between Presidential and MP elections at constituency level^c(2)
Difference in expected win presidential election^a	−0.232 (0.648)	−4.293 (3.906)
More expected ODM to win over PNU × expected win difference^a	0.054 (1.320)	3.475 (4.737)
Ethnic stronghold (PNU reference)
Stronghold ODM	2.025* (1.124)	−2.966 (2.952)
Stronghold ODM-K	−0.000 (0.385)	−6.305** (2.943)
Other	0.670 (0.609)	−5.959* (3.459)
Ethnic polarization index	2.147* (1.269)
Index public goods (schools, hospitals, and police stations nearby)	1.690 (1.634)
Province fixed effects	No	No
Constant	−1.936 (1.473)	5.409** (2.287)
R ²	.27	.10
Number of observations	42	76

Source: Author’s surveys, casualties by bullets CIPEV (2008), population at district level Kenyan Census 2009, official electoral results 2007, Kenyan Electoral Commission.

Note: ODM = Orange Democratic Movement; ODM-K = Orange Democratic Movement-Kenya; PNU = Party of National Unity.

Robust standard errors in parentheses clustered at district level in column 1, at constituency level in column 2.

^aExpected win difference taken from pre-election survey question: “Which party do you think will win the national presidential elections in December 2007?”

^bCovariates aggregated at district level.

^cCovariates aggregated at constituency level.

Significance level: *p < .10, **p < .05, ***p < .001.

The OLS results suggest that the number of casualties at the hands of police were not correlated with the degree of electoral competition (column 1 in Table 5). However, the strongholds of the ODM party experienced two more deaths (per 100,000 inhabitants) by the police than the strongholds of the PNU. This suggests that the police could have been more heavy-handed in the ODM areas.

Next, we analyze the allegations of electoral fraud. In Table 5, column (2), we estimate an OLS regression using as the dependent variable the raw difference between the official turnout in the presidential election and the turnout in the MPs elections at constituency level in the sampled areas of our pre-electoral survey. The results suggest that the strongholds of the PNU and the ODM were as likely to have had differences in turnout in the presidential and MP elections, as Hypothesis 4 predicts. The strongholds of the ODM-K and other parties were less likely to have differences in turnouts than were the PNU strongholds.⁹

In sum, the purpose of this analysis was to examine the correlation between the incidence of illegal electoral practices and the characteristics of respondents and their areas of residence. Since we have a weak identification strategy, we do not claim a direct causality relationship, such as what would have happened to the degree of electoral competition had all these illegal electoral practices not occurred. Nonetheless, the descriptive analysis suggests parties behaved strategically targeting specific groups. Only a minority of respondents (5 percent) reported having received both an offer in exchange for their vote and a threat, in the majority of those cases a different party had threatened them and offered something for the vote. The targeting of voters might be possible due to how easy it is to identify Kenyans by their ethnic origin, which is a good proxy for their political inclinations. People’s ethnicity can easily be deduced by their mother tongue, surname, and in some cases by physical appearance or area of residence.

Changes in Perceptions Following the Disputed Elections

This subsection explores the impact of post-electoral violence and separately the impact of vote-buying. First, we test whether the post-electoral violence affected respondents’ perceptions about democracy, political competition, ethnic identity and the use of violence. We use the difference-in-difference estimator, β, to identify the impact of violence by comparing the change in perceptions before and after the elections of the victims of violence to those of non-victims as expressed in equation (3):

β = E \{Y_{11} - Y_{10} | T = 1\} - E \{Y_{01} - Y_{00} | T = 0\},

where Y_st denotes the outcome in treatment status s and period t. T = {0, 1} is the indicator of exposure to violence, a person in state s = 1 is a respondent treated with violence.

To assess the impact of violence on perceptions, we focus exclusively on the very same respondents that answered both the pre- and the post-electoral surveys in 2007 and 2008, respectively. Since we cannot ignore the fact that our post-electoral survey suffered from attrition, we estimate the difference-in-difference estimator using two different specifications: a Heckit model and a panel fixed-effects with kernel matching. Both methods deal with different potential biases.

The Heckit model helps us to detect and correct (if necessary) in case there is self-selection bias due to attrition. This method is estimated in two stages. In the first stage, we estimate the probability of being able to reinterview the same respondent in 2008. We estimate this probability of response using a probit model, as shown in equation (4):

P r o b (R = 1 | Z) = ϕ (g Z),

where R indicates if the respondent was reinterviewed, Z is a vector of explanatory variables including respondent’s age, gender, ethnicity, and constituency’s poverty level in 2006, whether urban, and whether the district where the respondent was living experienced casualties or injuries after the election according to the CIPEV report. We also control for the interviewer’s education level, as the skills of the interviewer could have influenced the likelihood of filling the necessary contact information to reinterview the respondent. φ is the cumulative distribution function of the standard normal distribution.

In the second stage, we run a Heckit regression, which estimates the factors influencing a respondent’s change of views, as shown in equation (5). This regression corrects for a potential bias in self-selection by incorporating a transformation of these predicted individual probabilities of response, known as the inverse Mills ratio λ, which is added as an explanatory variable:

E [{c h a n g e i n v i e w s}_{i d t} | R = 1] = α + β {T r e a t e d}_{i d} + ρ σ_{u} λ (g Z) + u_{i d t},

where Treated is a dummy variable indicating whether respondent i was a victim of post-electoral violence in district d and β refers to the difference-in-difference estimator. ρ is the correlation between unobserved determinants of reinterviewing the respondent and unobserved determinants of change in views u_it at time t, σ _u is the standard deviation of u, and λ is the inverse Mills ratio evaluated at γZ.

The results of the Heckit models are shown in Tables A2 and A3 in the Online Appendix. These regressions show that our estimations do not suffer from a sample selection bias due to nonresponse, as the inverse Mills ratios estimated are not statistically significant for the great majority of the models estimated. In the few cases (four) where the inverse Mills ratios were statistically significant, the difference-in-difference estimator coincides with the one obtained under the Kernel matching method, which we explain subsequently.

Once we established that the difference-in-difference estimator does not suffer from systematic biases due to attrition, we estimated the difference-in-difference using propensity score Kernel matching. This is our preferred estimator, as it helps to control for individual and regional fixed effects, as well as to ensure that the two types of respondents being compared, those treated by violence, and those used as a control group were as similar as possible before the violence occurred. We estimate this method in three stages.

First, we use a series of probit models to estimate the probability of a respondent reporting having experienced a specific kind of post-electoral violence (direct, indirect, or exposed to violence, which we explain further subsequently). These estimated probabilities, known as propensity scores, are estimated as shown in equation (6):

P r o b (T r e a t e d = 1 | X) = ϕ (κ X) .

In these probit models, we use covariates X that jointly influence the likelihood of experiencing violence and that might also affect the outcomes we analyze, which is the change in respondent’s perceptions. The covariates X used are the respondent’s ethnicity, education, age, and gender, an index of availability of three basic public goods (schools, hospitals, and police stations) near the respondents’ home aggregated at district level, whether residing in an urban area, the incidence of poverty before elections and change in incidence of poverty between 2006 and 1999 in the constituency where the respondent was living before the outbreaks of post-electoral violence. We also include in the estimation of the propensity score whether the respondent’s household did not own a TV and radio, whether the respondent claimed to be interested in the presidential elections and an interaction between these two variables. This information helps us to control for a potential spillover effect among people not affected by violence directly, but simply by learning about these events through the media.

Table A4 (in the Online Appendix) shows the results from the probit regressions, where the coefficients are reported as marginal effects. The balancing property is satisfied in all the estimations shown, which ensures that the distribution of characteristics X is the same for those exposed to violence and those who were not and who fall in the region of common support. We identify the region of common support as the overlap between the two distributions of the propensity score for the treatment and control groups.

In the second stage, we use Epanechnikov kernel to match each respondent treated by violence to respondents in the control group who were not treated with violence by weighting for the differences according to the propensity score of the control group, as shown in equation (7). The farther away the respondent in the control group’s propensity score to that of the treated, the less weight they get (Heckman, Ichimura, and Todd 1998):

w_{i j} = \frac{G [\frac{p_{j} - p_{i}}{a_{n}}]}{\sum_{k = 1}^{n_{0}} G (\frac{p_{k} - p_{i}}{a_{n}})},

where G(·) is the kernel function, a_n is a bandwidth parameter, p_i and p_j are the estimated propensity scores of the respondent i being treated by the violence, and the respondent j in the control group. In all the estimations presented, we use a bandwidth equal to 0.06. The t-tests of mean equality show that none of the covariates compared between treated and control areas have a statistically significant difference.

In the third and last stage, we estimate the difference-in-difference estimator using a panel fixed-effects regression as expressed in equation (8). That equation estimates the factors that might have affected a respondent’s change in views, including having experienced violence, where the kernel weights w_ij adjust for the comparison between the respondents in the control group and the treated respondents by the violence:

Y_{i d t} = {α_{0} + α_{1} D u m m y Y e a r 2 008 + β T r e a t e d_{i d} \times D u m m y Y e a r 2 008, w i j},

where Y_idt is the perception of interest for the individual i living at district d at time t (t = 0 before and t = 1 after treatment), the coefficient α₀ indicates the views held before the election by those not affected in the post-electoral violence. α₁ refers to the change in views after the elections by those not affected in the post-electoral violence. β refers to the difference-in-difference estimator.

We identify separately the impact of three different kinds of violence that the respondent could have been exposed to:

Suffered direct violence, identified as those who answered positively to “Because of electoral conflict were you personally affected in the outbreaks of violence after 2007 in any of the following ways? Personal injury, damage to your personal property, the destruction of your home or business, being forced to leave your home or land.”

Suffered indirect violence, identified as those who answered positively to “Because of electoral conflict were you personally affected in the outbreaks of violence after 2007 in any of the following ways? The destruction or closure of a business, loss of earnings from your business or loss of job.” We exclude from this definition those that also experienced destruction of their home, forced to leave their land, or personal injury.

Non-victims but exposed to violence, identified as those who were living in districts which experienced violence according to the CIPEV report but who were non-victims of direct or indirect violence using the responses to the questions mentioned earlier.

The change in opinions in each of these three groups is compared to those respondents who did not report to have suffered from any of these instances of violence and who were living in districts that were not exposed to violence. The areas that were not exposed to violence are identified as the districts that did not report people being injured or killed as a direct result of the post-electoral violence according to the CIPEV report.

In addition to testing the impact of violence, we analyze separately the impact of another widely used electoral malpractice: vote-buying. We use the same method, the difference-in-difference estimator via Heckit and kernel propensity score matching. We focus on respondents who claimed not to have received an offer for their vote in the pre-electoral survey in 2007, but that in the post-electoral survey claimed to have received an offer for their vote after our first interview. Also, to isolate the impact of vote-buying on political perceptions, we focus exclusively on those respondents who experienced vote-buying alone (that is without being affected by the violence directly, indirectly, or living in districts affected by post-electoral violence). The change in opinions among the respondents “treated” by vote-buying are compared to those who did not experience violence or vote-buying, and who were similarly living in districts that did not experience violence. Table A4 shows the number of respondents (in the treated and control group) that fall in the area of common support, across all groups analyzed.

In Table 6, we present the changes in the perceptions before and after the elections regarding democracy, the role of political competition and ethnic identity. In each column, we estimate the impact of the three kinds of treatments of violence, as well as the impact of vote-buying.

Table 6.

Change in Perceptions on Democracy, Identity, and Electoral Competition Using Kernel Matching.

	Not satisfied how democracy works in Kenya				Since elections sometimes produce bad results, we should adopt other methods for choosing this country’s leaders				Identify first and foremost as Ethnic				Parties should not be allowed to form on tribe or religion basis
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)	(16)
Constant	0.362*** (0.025)	0.349*** (0.029)	0.410*** (0.031)	0.264*** (0.040)	0.095*** (0.020)	0.121*** (0.024)	0.117*** (0.029)	0.081*** (0.024)	0.184*** (0.017)	0.164*** (0.015)	0.171*** (0.012)	0.234*** (0.024)	0.641*** (0.022)	0.610*** (0.027)	0.632*** (0.030)	0.681*** (0.041)
Dummy year 2008	0.193*** (0.057)	0.178** (0.086)	0.160 (0.119)	0.332** (0.131)	0.223*** (0.060)	0.188** (0.078)	0.177 (0.113)	0.124* (0.066)	−0.184*** (0.041)	−0.174*** (0.037)	−0.149*** (0.039)	−0.162** (0.074)	−0.036 (0.055)	0.065 (0.079)	−0.000 (0.115)	−0.045 (0.140)
Victim of violence × dummy year 2008	0.121 (0.100)				−0.106 (0.081)				0.267*** (0.070)				0.189** (0.089)
Had economic losses × dummy year 2008		0.176 (0.118)				−0.113 (0.096)				0.261*** (0.060)				0.047 (0.106)
Non-victim but living in area that suffered violence × dummy year 2008			0.011 (0.125)				−0.068 (0.116)				0.183*** (0.047)				0.154 (0.121)
Experienced vote-buying after first interview in 2007 × dummy year 2008				−0.189 (0.159)				0.111 (0.096)				0.006 (0.098)				0.060 (0.164)
Observations	597	557	981	216	604	562	992	218	604	562	992	218	604	562	992	218

Source: Author’s surveys.

Note: Robust standard errors in parentheses.

*p < .10, **p < .05, ***p < .001.

In columns (1) to (4) we explore the change in support for the following statement: “Not satisfied with how democracy works in Kenya.” The difference-in-difference coefficients are not statistically significant for any of the three groups affected or exposed to violence, or for those who experienced vote-buying, relative to their control group. In other words, the increase in the level of dissatisfaction about how democracy works cannot be attributed alone to experiencing violence or vote-buying.

Columns (5) to (8) explore the changes in the support for the following statement: “Since elections sometimes produce bad results, we should adopt other methods for choosing this country’s leaders.” We find no change among those affected or exposed to violence or vote-buying relative to their control group.

Columns (9) to (12) explore whether there were any changes in respondents’ self-reported identity. We obtained this information from the question: “We have spoken to many Kenyans in this country and they have all described themselves in different ways. Some people describe themselves in terms of their language, religion, race, and others describe themselves in economic terms, such as working class, middle class or a farmer. Besides being a citizen of Kenya, which specific group do you feel you belong to first and foremost?” After the elections, the percentage of people who identified in ethnic terms increased among victims of direct violence (by twenty-seven percentage points), among the victims of indirect violence (by twenty-six percentage points), and among the non-victims of violence but living in districts that experienced violence (by eighteen percentage points), relative to their control group. In other words, violence reinforced ethnic identification, while vote-buying alone did not affect it.

The support for the statement “Parties should not be allowed to form on a basis of tribe or religion” increased by nineteen percentage points among the victims of direct violence, relative to their control group (column 13). We found no effect among the other “treated” groups analyzed by violence or vote-buying (columns 14 to 16).

Columns (1) to (4) in Table 7 show no change in the support for the statement “In deciding which party you most like, do you consider the ethnic or regional origin of the party’s leader?” among those affected or exposed to violence nor among those who experienced vote-buying relative to their control group. Similarly, we find no change in support for the statement “People in your community have been standing clearly against violence originated by politicians” as a result of being a victim of violence or vote-buying (columns 5 to 8).

Table 7.

Change in Expectations about Violence Using Kernel Matching.

	In deciding which party likes most, considers the ethnic or regional origin of the party’s leader				People in village have been standing clearly against violence originated by politicians				If were victim of a violent crime, would find another way to deal with the matter instead of calling the police				It is sometimes necessary to use violence in support of a just cause
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)	(16)
Constant	0.524*** (0.025)	0.542*** (0.025)	0.506*** (0.025)	0.601*** (0.035)	0.629*** (0.026)	0.646*** (0.026)	0.679*** (0.030)	0.645*** (0.036)	0.118*** (0.017)	0.119*** (0.016)	0.121*** (0.010)	0.159*** (0.027)	0.110*** (0.018)	0.099*** (0.018)	0.130*** (0.011)	0.150*** (0.026)
Dummy year 2008	0.103* (0.057)	0.067 (0.060)	0.034 (0.090)	0.060 (0.114)	0.034 (0.068)	0.025 (0.078)	−0.024 (0.115)	−0.030 (0.121)	−0.019 (0.030)	−0.031 (0.033)	−0.022 (0.030)	0.003 (0.090)	0.020 (0.040)	0.046 (0.034)	−0.003 (0.032)	−0.087 (0.081)
Victim of violence × dummy year 2008	−0.056 (0.100)				−0.057 (0.102)				0.125* (0.067)				0.156** (0.072)
Had economic losses × dummy year 2008		−0.079 (0.099)				0.038 (0.106)				0.144** (0.065)				0.141* (0.073)
Non-victim but living in area that suffered violence × dummy year 2008			0.065 (0.100)				0.014 (0.121)				0.012 (0.039)				0.071* (0.043)
Experienced vote-buying after first interview in 2007 × dummy year 2008				−0.044 (0.140)				0.062 (0.145)				−0.065 (0.110)				0.181* (0.105)
Observations	604	562	992	218	604	562	992	218	604	562	992	218	604	562	992	218

Source: Author’s surveys.

Note: Robust standard errors in parentheses.

*p < .10, **p < .05, ***p < .001.

Columns (9) and (10) show that the support for the statement “If you were a victim of a violent crime, you would find another way to deal with the matter instead of calling the police” increased among the direct victims of violence (by thirteen percentage points) and among the indirect victims of violence, those who had economic losses, (by fourteen percentage points). The heavy-handed role of the police in the violence, mentioned earlier, could have contributed to this change of perception.

Finally, columns (13) to (16) show the change in support for the statement “In Kenya it is sometimes necessary to use violence in support of a just cause.” After the elections, the percentage of people with tolerance for the use of violence went up among the victims of direct violence (by sixteen percentage points), among the victims of indirect violence (by fourteen percentage points), and to lesser degree among the non-victims of violence but living in districts that experienced violence (by seven percentage points), relative to their control group. These findings are in line with international evidence that violence breeds violence (Balcells 2010). Moreover, we find the support for violence also increased among those who experienced vote-buying alone (by eighteen percentage points), relative to their control group. This suggests that vote-buying can also reinforce people’s perceptions that other illegal electoral practices, such as violence, can be justified.

In general, although this analysis is innovative in presenting separately the effects of violence and of vote-buying, we also acknowledge its potential limitations. The change in self-reported perceptions or beliefs might not necessarily translate into changes in actions or behavior.

Conclusion

The micro-level findings suggest that political parties in Kenya used illegal electoral practices strategically, despite the fact that the main parties contending had formed only recently and lacked political infrastructure such as provincial branch offices, a characteristic shared with other African countries (Kramon 2011; LeBas 2011). The results suggest that the ethnopolitical cleavages that shape the formation of political parties and the mobilization of voters in African democracies (Mozaffar, Scarritt, and Galaich 2003) are also correlated with the use of illegal electoral practices.

We found three important implications for the future of democracy. First, the reasons for the widespread vote-buying in Kenya (with 27 percent of respondents reporting vote-buying), as in other similar countries, could be rooted in a prisoner’s dilemma. All political parties would be better off financially if no one buys votes, making more resources available for the provision of public goods. However, as Kramon (2011) argues, given that no politician can commit to not buying votes, the dominant strategy is to vote-buy, despite the effectiveness of vote-buying being reduced by all parties doing the same. In Kenya, the observed behavior of ethnic-voting is driven by the expectations that candidates will deliver patronage goods to their coethnics (Gutiérrez-Romero 2013); hence, vote-buying is an important signal of the candidate’s commitment. Evidence supporting this argument has been found in Ghana, where voters expect vote-buying as a sign of their candidate’s interest (Nugent 2007). Randomized studies in Africa have shown promise in reducing vote-buying practices through voter education campaigns (Vicente and Wantchekon 2009), an approach that deserves further investigation.

Second, while some of the electoral violence occurred opportunistically, the article found strong correlations that suggest political actors chose to reinforce their chances of election with the instigation of violence. The distribution of electoral support shaped the use of violence, both before and after the elections, as other studies have found (Dunning 2011). The results shown are consistent with the predictions of Collier and Vicente (2012), who suggest that if the challenger party has a large support base, the incumbent party will intimidate the challenger’s supporters, as a repression tactic for them not to vote. Our evidence suggest that parties used predominantly indirect intimidation tactics by instigating people to be violent through sectarian campaigns and gangs and resorted to direct-threats less frequently.

Third, more electoral irregularities were observed in the strongholds of the two main political parties, supporting the hypothesis of Collier and Vicente (2012) that the incumbent party will seek to vote-buy and rig in its own strongholds. However, the Kenyan case also provides evidence that a strong challenger party will also resort to rigging if the competition is close, and the incumbent candidate is expected to rig.

The electoral ordeal reduced the desire of the general population for holding elections. Among those directly affected or exposed to violence more identify in ethnic terms. Moreover, those affected by the post-electoral violence at personal level seem to be trapped in a vicious circle. After the elections, they are more likely to prefer that parties are no longer allowed to organize on ethnic or religious lines, but more identify in ethnic terms and still use the ethnicity of the candidates to gauge which party to support. Another important finding was that the acceptability of the use of violence increased among both the victims of violence, in line with international evidence, and those affected by vote-buying.

Kenya avoided major scale violence in the general elections of 2013. Nevertheless, 477 lives were still lost and near 118,000 people were displaced right before the 2013 elections (Human Rights Watch 2013, 1). Political parties have continued using gangs and other practices of intimidation despite the constitutional reforms implemented after the 2007 disputed election and the pending ICC trials against key Kenyan politicians. Therefore, the country is still at risk of reexperiencing electoral violence.

Footnotes

Acknowledgment

Thanks are due to Laia Balcells, Michael Bratton, James Fenske, John Githongo, Christoph Oberlack, Adam Pepelasis, Justin Sandefur, two JCR anonymous referees, the iiG, CSAE and UAB seminar participants for their suggestions on earlier versions of this article.

Author’s Note

The views expressed are not necessarily those of Department for International Development.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The surveys and media data presented in this article were funded by the UK Department for International Development (DFID) as part of the Improving Institutions for Pro-poor Growth (iiG), a research consortium aimed at studying how to improve institutions in Africa and South-Asia. The author also acknowledges financial support from the Spanish Ministry of Science and Innovation (reference ECO2010-21668-C03-02 and ECO2013-46516-C4-1-R).

Notes

Supplemental Material

Online Appendix is available from the authors or on the Journal of Conflict Resolution website at .

References

Anderson

David

Lochery

Emma

. 2008. “Violence and Exodus in Kenya’s Rift Valley, 2008: Predictable and Preventable?” Journal of East African Studies 2:328–43.

Balcells

Laia

. 2010. “Rivalry and Revenge. Violence against Civilians in Conventional Civil Wars.” International Studies Quarterly 54:291–313.

Bratton

Michael

. 2008. “Vote Buying and Violence in Nigerian Election Campaigns.” Electoral Studies 27:621–32.

Bratton

Michael

Kimenyi

Mwangi

. 2008. “Voting in Kenya: Putting Ethnicity in Perspective.” Journal of East African Studies 2:272–89.

CIPEV (Commission of Inquiry on Post Election Violence). 2008. Report of the Commission of Inquiry into Post-election Violence. Nairobi, Kenya: CIPEV.

Collier

Paul

Hoeffler

Anke

. 2004. “Greed and Grievance in Civil War.” Oxford Economic Papers 56:563–95.

Collier

Paul

. 2007. “Economic Causes of Civil Conflict and their Implications for Policy.” In Leashing the Dogs of War Conflict: Management in a Divided World, edited by Crocker

Chester A.

Hampson

Aal

Pamela

, 197-218. Washington, DC: United States Institute of Peace.

Collier

Paul

Vicente

Pedro

. 2012. “Violence, Bribery and Fraud: The Political Economy of Elections in Sub-Saharan Africa.” Public Choice 153:117–47.

Cox

Gary

McCubbins

Mathew

. 1986. “Electoral Politics as a Redistributive Game.” Journal of Politics 48:370–89.

10.

Dercon

Stefan

Gutiérrez-Romero

Roxana

. 2012. “Triggers and Characteristics of the 2007 Kenyan Electoral Violence.” World Development 40:731–44.

11.

Dunning

Thad

. 2011. “Fighting and Voting: Violent Conflict and Electoral Politics.” Journal of Conflict Resolution 55:327–39.

12.

Eifert

Ben

Miguel

Edward

Posner

Daniel

. 2010. “Political Sources of Ethnic Identification in Africa.” American Journal of Political Science 54:494–510.

13.

Gibson

Clark

Long

James

. 2009. “The Presidential and Parliamentary Elections in Kenya, December 2007.” Electoral Studies 28:497–502.

14.

Gutiérrez-Romero

Roxana

. 2013. “To What Extent Did Ethnicity and Economic Issues Matter in the 2007 Disputed Kenyan Elections?” Development Policy Review 31:291–320.

15.

Heckman

James

Ichimura

Hidehiko

Todd

Petra

. 1998. “Matching as an Econometric Evaluation Estimator.” Review of Economic Studies 65:261–64.

16.

Human Rights Watch. 2008. “Ballots to Bullets: Organized Political Violence and Kenya’s Crisis of Governance.” Human Rights Watch. Vol. 20. Accessed December 20, 2013. http://www.hrw.org/sites/default/files/reports/kenya0308web.pdf.

17.

Human Rights Watch. 2013. “High Stakes. Political Violence and the 2013 Elections in Kenya.” US: Human Rights Watch. Accessed December 20, 2013. http://www.hrw.org/sites/default/files/reports/kenya0213webwcover.pdf.

18.

Kramon

Erick

. 2011. “Why Do Politicians Buy Votes When the Ballot Is Secret? Theory and Experimental Evidence from Kenya.” University of California Working Paper, Berkeley, CA.

19.

LeBas

Adrienne

. 2011. From Protest to Parties: Party Building and Democratization in Africa. Oxford, UK: Oxford University Press.

20.

Lynch

Gabrielle

Crawford

Gordon

. 2011. “Democratization in Africa 1990-2010: An Assessment.” Democratization 18:275–310.

21.

Montalvo

Jose

Reynal-Querol

Marta

. 2003. “Religious Polarization and Economic Development.” Economic Letters 80:201–10.

22.

Mozaffar

Shaheen

Scarritt

James R.

Galaich

Glen

. 2003. “Electoral Institutions, Ethnopolitical Cleavages, and Party Systems in Africa’s Emerging Democracies.” American Political Science Review 97:379–90.

23.

Nitcher

Simeon

. 2008. “Vote Buying or Turnout Buying? Machine Politics and the Secret Ballot.” American Political Science Review 102:19–31.

24.

Nugent

Paul

. 2007. “Banknotes and Symbolic Capital.” In Votes, Money and Violence: Political Parties and Elections in Sub-Sahara Africa, edited by Basedau

Erdmann

Natal

A. Mehler, 253-275. KwaZulu

: South Africa: University of KwaZulu Natal Press.

25.

Rice

Xan

. 2008. “EU Observers Condemn Kenyan Election and Call for Investigation.” The Guardian, January 2, 2008. Available at: http://www.theguardian.com/world/2008/jan/02/eu.kenya

26.

Stokes

Susan

. 2005. “Perverse Accountability: A Formal Model of Machine Politics with Evidence from Argentina.” American Political Science Review 99:315–25.

27.

Synder

Jack

. 2000. From Voting to Violence: Democratization and Nationalistic Conflict. New York: W.W. Norton.

28.

Vicente

Pedro

Wantchekon

Leonard

. 2009. “Clientelism and Vote Buying: Lessons from Field Experiments in African Elections.” Oxford Review of Economic Policy 25:292–305.

29.

Yamano

Takashi

Tanaka

Yuki

Gitau

Raphael

. 2010. “Haki Yetu (It’s Our Right): Determinants of Post-election Violence in Kenya.” GRIPS Discussion Paper 10-20, National Graduate Institute for Policy Studies, Tokyo.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.15 MB

0.12 MB

0.00 MB