Download - Ferree, Gibson, And Long JEAS 2014
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
1/21
This article was downloaded by: [University of California, San Diego]On: 10 February 2014, At: 10:48Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Journal of Eastern African StudiesPublication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/rjea20
Voting behavior and electoral
irregularities in Kenya's 2013 ElectionKaren E. Ferree
a, Clark C. Gibson
a& James D. Long
b
aUniversity of California, San Diego, La Jolla, CA, USA
b
University of Washington, Seattle, WA, USAPublished online: 07 Feb 2014.
To cite this article:Karen E. Ferree, Clark C. Gibson & James D. Long , Journal of Eastern African
Studies (2014): Voting behavior and electoral irregularities in Kenya's 2013 Election, Journal of
Eastern African Studies, DOI: 10.1080/17531055.2013.871182
To link to this article: http://dx.doi.org/10.1080/17531055.2013.871182
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (theContent) contained in the publications on our platform. However, Taylor & Francis,our agents, and our licensors make no representations or warranties whatsoever as tothe accuracy, completeness, or suitability for any purpose of the Content. Any opinionsand views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Contentshould not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,proceedings, demands, costs, expenses, damages, and other liabilities whatsoever orhowsoever caused arising directly or indirectly in connection with, in relation to or arisingout of the use of the Content.
This article may be used for research, teaching, and private study purposes. Anysubstantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
http://www.tandfonline.com/page/terms-and-conditionshttp://www.tandfonline.com/page/terms-and-conditionshttp://www.tandfonline.com/page/terms-and-conditionshttp://dx.doi.org/10.1080/17531055.2013.871182http://www.tandfonline.com/loi/rjea20http://www.tandfonline.com/page/terms-and-conditionshttp://www.tandfonline.com/page/terms-and-conditionshttp://dx.doi.org/10.1080/17531055.2013.871182http://www.tandfonline.com/action/showCitFormats?doi=10.1080/17531055.2013.871182http://www.tandfonline.com/loi/rjea20 -
8/13/2019 Ferree, Gibson, And Long JEAS 2014
2/21
Voting behavior and electoral irregularities in Kenyas 2013 Election
Karen E. Ferreea, Clark C. Gibsona and James D. Longb*
aUniversity of California, San Diego, La Jolla, CA, USA; bUniversity of Washington, Seattle,WA, USA
(Received 5 July 2013; accepted 27 November 2013)
Data from a unique nationwide exit poll of 6258 voters are employed to explore twocentral themes of the 2013 Kenyan Election: (1) the correlates of individual votechoice; and (2) the credibility of the electoral process. The analysis reveals several
striking relationships between an individuals vote choice, personal attributes, andperceptions of the campaign and candidates. We find that the leading coalitions mostlykept their co-ethnics together, although ethnic alliances proved somewhat less certainthan in the past. We find that, for the most part, voters treated Uhuru Kenyatta notsitting Prime Minister Raila Odinga as the incumbent. The data show that campaignissues also influenced the vote: Odinga garnered more support on issues related toconstitutional implementation, corruption, and the International Criminal Court (ICC),while Kenyatta won on the economy, employment, and security. Exit poll data alsoreveal irregularities in the electoral process, including some evidence of inflated votetotals benefitting the Jubilee coalition and illegal administrative activities. The data,while not definitive, are highly suggestive of a deeply flawed electoral process andchallenge claims that Kenyatta won a majority in the first round.
Keywords:voting behavior; ethnicity; performance; campaigns; electoral irregularity;exit poll; Kenya
Introduction
Roughly 12.3 million Kenyan voters went to the polls on March 4, 2013 to elect a new
president for the fifth time since the reintroduction of multi-party elections in 1992. Raila
Odinga, the sitting prime minister and leader of the Orange Democratic Movement (ODM)
in the Coalition for Reforms and Democracy (CORD) Alliance, ran against Uhuru
Kenyatta, sitting deputy prime minister and son of Kenyas founding President Jomo
Kenyatta, and leader of The National Alliance (TNA) in the Jubilee Alliance. After thecertification of Kenyatta as the winner of the race with 50.07% of the vote, a tiny fraction
above the 50% + 1 votes required to avoid a second round, Odinga refused to concede,
citing irregularities in electoral administration and possible vote rigging. CORD appealed to
the Supreme Court but, following a relatively short hearing and period of deliberation, the
court upheld the victory and Kenyatta was sworn in as Kenyas new president on April 9.
We employ data from a nationwide exit poll of 6258 voters to explore two central
themes of the 2013 Kenyan Election: firstly, the correlates of individual vote choice; and
secondly, the credibility of the electoral process. Unlike household surveys, exit polls
interview only actual voters and do so directly after they cast their ballots, when errors of
*Email:[email protected]
Journal of Eastern African Studies, 2014
http://dx.doi.org/10.1080/17531055.2013.871182
2014 Taylor & Francis
http://-/?-http://-/?-http://-/?-mailto:[email protected]://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182mailto:[email protected]://-/?-http://-/?-http://-/?- -
8/13/2019 Ferree, Gibson, And Long JEAS 2014
3/21
recall and projection are minimized. In addition to providing a unique window into voting
behavior, exit polls represent important tools for verifying official vote counts. The goal
is to demonstrate both of these facets of the 2013 Kenyan election.
We first use data from the exit poll to examine voting behavior. Studies of voting
behavior in Africa generally include three sets of factors to explain the vote: ethnicity,
performance, and issues. Ethnic voting is by far the most well-known approach, and uses
the affinity between co-ethnic citizens and candidates to establish voter motivations.1 Co-
ethnicity proves an especially strong predictor in Kenyan elections.2 Given the violence
that occurred afterdisputed results in 2007, many analysts projected equally tense ethnic
relations in 2013.3 Scholars also find incumbent performance important in explanations
of voting behavior in Kenya as well as other African elections.4 More recently, studies
suggest that individual beliefs about issues also structure vote choice in Africa.5
Features of the 2013 race form an especially interesting context to explore the effects
of ethnicity, performance, and campaign issues. The contest featured two unusual
coalitions: the Jubilee Alliance of Uhuru Kenyatta and William Ruto brought together
bitter rivals from the 2007 election. Moreover, conflict between Kenyatta and Rutosethnic groups, the Kikuyu and Kalenjin, has marred many previous elections, including
2007. The CORD alliance also featured an unusual duo: Raila Odinga, a Luo, and
Kalonzo Musyoka, a Kamba, who had been competitors in 2007. Would voters support
these alliances even if it meant supporting candidates they had previously voted against?
The 2013 race also lacked a clear incumbent to whom voters might attribute government
performance, although Odinga and Kenyatta had both served in the sitting government.
Would voters hold one, both, or neither responsible for the governments performance?
Last, although the campaign involved fundamental and recurring issues in Kenyan
politics like land, security, and corruption, the ongoing prosecution of Kenyatta and Ruto
by the International Criminal Court (ICC) also featured prominently. Would this issuetrump others, or outweigh the influence of ethnicity and performance?
We also consider irregularities in the management of the election and the vote count.
We examine anomalies between the official election results and the exit poll and analyze
the quality of the electoral process as evaluated by voters. We consider evidence from the
exit poll in conjunction with qualitative and quantitative data from other sources,
including pre-election surveys and observer group reports, to evaluate the credibility of
the election and its outcome.
The analysis reveals several striking relationships between an individuals vote
choice, personal attributes, and perceptions of the campaign and candidates. We find that
the leading coalitions received strong support from their co-ethnics, although ethnic
alliances proved somewhat less certain than in the past. We also find that, for the most
part, voters treated Kenyatta not sitting Prime Minister Odinga as the incumbent. The
data show that campaign issues also influenced the vote: Odinga garnered more support
on issues related to constitutional implementation, corruption, and the ICC, while
Kenyatta won on security, services, and the economy. Somewhat surprisingly, the
incendiary ICC issue figured relatively low on voters lists of important issues and
appeared to play less central a role in mobilizing support for Kenyatta and Ruto than
many have suggested.6
Exit poll data also reveal irregularities in the electoral process, including evidence of
inflated vote totals benefitting the Jubilee the Jubilee coalition and illegal administrative
activities. The data, while not definitive, are highly suggestive of a deeply flawedelectoral process and challenge claims that Kenyatta won a majority of votes in the first
round.
2 K.E. Ferreeet al.
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
4/21
Ethnicity, performance evaluations, and issues in Kenyan elections
Like other studies of African voting behavior, research exploring the motives of the Kenyan
voter highlights themes of ethnicity, performance evaluations, and issues. Ethnicity plays
an important but complex role in Kenyan elections. Because no single ethnic group
comprises even a quarter of the country
s population, Kenyan parties generally attempt tobuild multi-ethnic coalitions to capture the presidency. Many parties fail in their cross-
ethnic appeals, and where viable coalitions form, they rarely survive across elections,
presenting voters with new configurations in each contest. Alliances can also result in
groups without a co-ethnic candidate running for president. While some races have
included Kikuyu and Luo presidential candidates, many groups have not fielded a
presidential or vice/deputy presidential candidate. Further, elites from groups without
presidential candidates frequently split their endorsements, some favoring one candidate
and others urging support for a rival.
While ethnicity can strongly predict voting behavior in Kenya, the way it maps to
individuals decisions varies from race to race and from voter to voter, depending on
the nature of alliances and whether a voter has a co-ethnic candidate on a ticket. For
voters with a co-ethnic in the presidential race, ethnicity correlates very strongly with
vote choice, especially when that candidate is at the top of the ticket and stands a
reasonable chance of winning.7 In the 2007 election, 94% of Kikuyus voted for the
Kikuyu candidate (Kibaki) and 98% of Luos for the Luo candidate (Odinga).8 Kibaki and
Odinga were at the top of their tickets and both had a good chance of winning what was
predicted to be a close election. Musyoka, a Kamba, also ran for president (but had little
chance of winning) and received 85% of his ethnic groups votes.
Voters without a co-ethnic in the race are less predictable9 but ethnicity may still
influence decisions. Voters may follow one or another co-ethnic opinion leader. When
Ruto joined Odinga and ODM in the 2007 election, 88% of Kalenjinvoters followed him,
despite former president Daniel arap Mois endorsement of Kibaki.10 Non-co-ethnics may
count heads to select the party or coalition that features the most co-ethnics in its
leadership. Or they may evaluate the past performance of parties to see which groups
have been favored in the past.11 In the 2007 Kenyan race, the Kisii, Mijikenda, Maasai,
Luhya, and Kalenjin may have followed one or all of these strategies since, without a co-
ethnic presidential candidate, each group divided its loyalties between Kibaki and
Odinga. In sum, ethnicity correlates with the behavior of some but not all voters,
depending on which groups have candidates in the race, their viability, and how elites
join and endorse parties.
Scholars offer several causal mechanisms to explain the correlation between ethnicity
and voting behavior. Ethnicity may act as an informational shortcut, helping voters
predict how candidates will direct flows of patronage,12 choose policy,13 or protect the
security of different groups.14 Voters may choose co-ethnics for purely expressive reasons
to derive satisfaction for voting for one of their own.15 Ethnicity may also reflect issues
like the distribution of wealth and resources in the country.16 Regardless of the precise
cause, the correlation between ethnicity and behavior remains robust in Kenya as in many
other African countries.
Evaluations of government performance also shape voting behavior in Africa.
Numerous studies now document a link between how African voters useperceptions of
incumbents performance in their vote choice,17 including in Kenya.18 Doing well inoffice whether through providing services or enjoying the fruits of a good economy
clearly matters to African voters.
Journal of Eastern African Studies 3
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
5/21
Issue positions have received less attention in the broader literature on African voting
behavior. Scholars suggest it is difficult for candidates to differentiate themselves on
valence goods or on ideological differences in Africa.19 Bucking this broader trend,
recent work demonstrates that issues do in fact help explain Kenyans votes. Candidates
have presented distinct position on issues to voters, who have in turn demonstrated
divergent opinions and vote choices based upon them.20 In the 2007 election, for
example, the topic of constitutional reform differentiated Odinga from Kibaki and
influenced some Kenyans votes.21
Most recent studies acknowledge that both ethnic and non-ethnic phenomena play
important roles in accounting for African voting behavior.22 Scholars also emphasize
complex interdependent relationships between ethnicity, performance evaluations, and
issues: issues can include ethnic overtones and divide along ethnic lines in many
elections.23 Performance evaluations often reflect ethnic considerations: voters can give
co-ethnic candidates better evaluations for the same record, be more forgiving of co-
ethnics with poor records, and align their performance evaluations to their possibly
ethnically determined vote rather than the other way around.24
The evidence suggests thatKenyans, like Africans more generally, consider ethnicity, government performance, and
issues when casting their ballots.
The 2013 Election
Key aspects of the 2013 Kenyan Election make it a compelling context in which to
explore the effects of ethnicity, performance, and campaign issues on voting behavior.
The contest featured two coalitions of unlikely bedfellows, the result of an extensive
courting process and furious last minute negotiations resembling a game of musical
chairs. Most observers assumed that candidates would team with traditional allies, so theannouncement of the Jubilee Alliance, joining Uhuru Kenyatta and William Ruto on the
same ticket, surprised many. Jubilee brought together the very groups the Kikuyu and
Kalenjin who weredeeply involved in the post-election violence that tore through the
Rift Valley in 2008:25 their rancor had also infected the 1992 and 1997 elections.26
Kenyatta and Ruto seemed to have little in common besides facing trial at the ICC (for
crimes against humanity for their roles in supporting the post-election conflict) and a
desire to prevent Odinga from becoming president, who had started the campaign period
as the favorite. Odingas CORD alliance also featured an unusual duo: Odinga, a Luo,
and Musyoka, a Kamba, competitors in the last election.
These unusual alliances forced voters into potentially uncomfortable choices. To
support the Kenyatta/Ruto ticket, Kalenjins would have to vote for a Kikuyu presidential
candidate accused of bankrolling violence against them in the last election, while
Kikuyus would need to overlook accusations of deputy presidential candidate Ruto for
similar crimes against them. The CORD alliance, somewhat less controversially,
nonetheless asked Kambas to support a presidential candidate whom they had voted
against in 2007. The 2013 election thus allows us to examine how voters traded a desire
to vote for a co-ethnic against prior attachments and antagonisms. Could ethnic elites
orchestrate ethnic voting despite history or are there limits to the types of alliances that
individuals will support?
The 2013 race also presented dilemmas to performance-oriented voters. The record of
the KibakiOdinga government was decidedly mixed. During the 200713 period, theKenyan economy grew more slowly than in the previous five years.27 Unemployment
remained high, although official statistics showed some improvements in services, school
4 K.E. Ferreeet al.
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
6/21
attendance, literacy, and child mortality during the Kibaki period,28 and development
around the country especially Nairobi was visible, as evidenced by several large road
projects.
The record on constitutional reform and implementation also remained uneven. The
KibakiOdinga government successfully passed constitutional reform via referendum in
August 2010, one of its primary mandates upon taking office. Reforms significantly
restructured executive power, creating new gubernatorial offices and bolstering the
independence of the court system. However, implementation of key features was delayed
until after the 2013 election, giving voters only a partial record. The coalitions fight against
corruption was unimpressive, doing little to foster greater transparency, accountability, or
ending the culture of impunity. In spite of tough talk, from Odinga in particular, few senior
government officials were successfully prosecuted; Odinga himself faced allegations about
the misappropriation of World Bank monies intended to create jobs for unemployed youth.
The governments record on security was even more disappointing. While conflict
abated in most parts of Kenya after the formation of the coalition government, security
problems persisted. Heading into the 2013 election, security forces appeared in disarray,incapable of containing violent incidents in northern and coastal Kenya, Nairobi, and Rift
Valley, with causes ranging from local score settling and livestock rustling, to
international terrorism (al-Shabaab) and separatist agitation (the Mombasa Republican
Council). Reports in September 2012 suggested that local elites, vying for newly
devolved government offices, were already engaged in election related intimidation.
Election primaries turned violent in a few locales, particularly in ODM strongholds in
Nyanza province. Worries about the competence of security forces grew when a militia
gunned down thirty-eight police officers in Northern Rift Valley just before the
Independent Elections and Boundaries Commission (IEBC) started voter registration.
Attributing responsibility for performance presented voters with an additionalchallenge. The sitting government included a coalition of Kibaki, Odinga, and their
parties. Voters might blame or praise President Kibaki, Prime Minister Odinga, both, or
neither for government action or inaction. Since Kibaki was not a candidate in 2013,
voters might have instead transferred their assessment of the governments performance
to Kenyatta, the new flag carrier for the remnants of Kibakis party and sitting deputy
prime minster. Or, they might have viewed Kenyatta as challenger and Prime Minister
Odinga as the incumbent. Voters could have seen the race as a choice between two
incumbents, one incumbent and a challenger, or two challengers. Given these complex-
ities, perhaps voters would eschew performance considerations altogether, falling back on
ethnic factors to decide their vote. Mixed records and muddied lines of attribution also
created opportunities for politicians to spin stories about the countrys state through and
independent of ethnic lenses.
Odinga and Kenyatta campaigned on different issues. Odinga emphasized fighting
corruption, land reform, and social issues. He expressed a desire to increase access to
healthcare, improve primary education, and establish free secondary education. He had
pushed hard for the new 2010 constitution and made other efforts to decentralize power,
and wanted to see key constitutional provisions implemented, especially regarding land
reform. He took swipes at the extensive land holdings of Kenyatta and his family and
derided his rivals promise to resettle the landless, at one point saying Can you let a
hyena guard your goats? His advocacy for the poor and previous dabbling in socialist
policies made him suspect among the business class.Kenyatta was better known on the campaign trail for championing economic growth,
making him the favorite of the business elite. While he had campaigned against the 2010
Journal of Eastern African Studies 5
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
7/21
constitution, his strong performances during televised debates helped to establish him as a
leader capable of taking over the reins of the presidency. Despite running against the
sitting prime minister, voters could conceivably view Kenyatta as an incumbent, given his
familys name, varied government posts, and ties to the ruling party.
In addition to the issues regarding the economy, land, security and corruption, the ICCs
case against Kenyatta and Ruto figured centrally in the 2013 campaign. Odingas camp
sought to paint the pair as war criminals; Kenyatta and Ruto denied all charges and
portrayed the ICC as outsiders interfering in Kenyan politics. Indeed, over the course of the
campaign the issue evolved from one about responsibility for post-election violence to one
about Kenyan sovereignty. Kenyatta and Ruto claimed their coalition exemplified the
reconciliation between two previously antagonistic groups. Some observers asserted that
far from hurting Kenyatta and Ruto the ICC issue helped their campaign and subsequent
voter turnout.
The 2013 election campaign period thus witnessed a fascinating combination of ethnic
factors, performance considerations, and appeals to issues. Rumor and innuendo swirled
around likely voting patterns and possible outcomes, suffused by fears of a repeat of 2007selectoral disaster.
Methods
To examine patterns of vote choice and possible irregularities in the vote count, we use
data from an exit poll we conducted on March 4, 2013, election day in Kenya. Exit polls
interview randomly selected voters after they have voted and are leaving the polling
station. Since exit polls sample only voters, they provide some advantages over pre-
election surveys that may include respondents who change their minds about issues and
candidates, or choose not to vote at all. Exit polls also decrease the likelihood of poorrecall, which threaten accuracy in post-election surveys. Exit polls are an increasingly
important tool to study vote counts and voting behavior in African elections.29
Our exit poll used a nationwide random sample of 6258 voters representing all of
Kenyas 47 counties, and included 404 (out of 31,977) polling stations. Using the IBEC s
final registry of voters, we distributed surveys proportionate to the size of each county,
and within each county selected polling stations proportionate to size. Enumerators
randomly selected every second voter leaving the polling station. Respondents could
choose to take the survey in English or Swahili. The survey included questions about
respondents demographics (including ethnicity), vote choice (president and governor),
and perceptions of government performance, candidates, and issues. It had a response rateof 88%. Enumerators conducted interviews using LG Android smartphones, equipped
with a survey application and real-time results transmission.
Correlates of the vote: ethnicity, performance, and issues
This section examines the correlates of vote choice by looking at the contribution of
ethnicity, performance, and campaign issues to explaining patterns of candidate support.
Descriptive statistics
Table 1 presents vote choice results cross-tabulated by ethnicity. Ethnicity correlatesstrongly with vote choice for groups with presidential and deputy presidential candidates:
83% of the Kikuyu and 74% of Kalenjin expressed support for Kenyatta/Ruto, and 94%
6 K.E. Ferreeet al.
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
8/21
of the Luo and 63% of the Kamba supported Odinga/Musyoka. Interestingly, these rates
fall below their levels in the 2007 election, when 98% of Luos supported Odinga, 94% of
Kikuyus supported Kibaki, 85% of Kamba supported Musyoka, and 88% of Kalenjin
supported Odinga (whom Ruto endorsed).30 These results suggest that while co-ethnics
still form a core base of electoral support in Kenya, this support eroded somewhat since
2007. Votes from groups without viable candidates are more difficult to predict. Although
Musalia Mudavadi ran as a third-party candidate, 53% of his co-ethnic Luhya supported
Odinga, perhaps indicating voters prefer a viable non-co-ethnic candidate over a non-
viable co-ethnic. The Kisii, Mijikenda, and Maasai divided their votes, but leaned
towards Odinga. The Meru, close cousins of Kikuyu, supported Kenyatta at a lower rate
than they had Kibaki in the 2007 election. The Somali were split. People who refused to
identify their ethnic affiliation, instead choosing Kenyan,were equally likely to support
Odinga and Kenyatta (33%). Thus, while ethnicity partially correlated with vote choice,
even the leading candidates did not garner all of their co-ethnic supporters, some groups
chose close ethnic relations along with electoral viability over non-viable co-ethnics, and
non-co-ethnics demonstrated varying levels of support for the front runners. Twelve
percent refused to answer (RTA) the vote choice question. These could be ethnic
defectors who feared revealing their vote choice, people who felt it important to keep vote
choice secret even from survey enumerators, respondents wishing to provide socially
desirable answers, or minorities in majority areas who feared reprisals. We return to this
topic in greater depth below.
Next, we turn to the salience of government performance and campaign issues to
voting behavior (Table 2); for ease of presentation we exclude votes for non-viable
candidates and refusals. Did either Kenyatta or Odinga benefit from positive perceptions
of government performance or suffer from negative ones? Did voters distinguish between
them on the issues?Panel I ofTable 2examines whether perceptions of central government performance
regarding the provision of services, the national economy, and security in Kenya correlate
Table 1. Cross-tabulation results by ethnic group.
Presidential vote
Kenyatta/Ruto Odinga/Musyoka Othersa Refusals Total
Total sample 41 41 7 12 100
Ethnic groupKikuyu 83 4 3 10 21Kalenjin 74 11 4 12 11Luo 1 94 1 4 11Kamba 12 63 6 18 8Luhya 6 53 22 18 14Kisii 15 72 4 9 6Mijikenda 13 72 7 8 6Maasai 29 60 2 9 2Meru 75 10 4 11 6Somali 41 48 8 2 3Kenyan 33 33 8 26 3
Notes: a
Since Kenyatta and Odinga garnered an overwhelming share of the nationwide vote, we collapse votesfor all additional candidates, including Musalia Mudavadi, Peter Kenneth, Martha Karua, Paul Muite,Mohammad Dida, and James Ole Kiyapi. Table 5presents the full exit poll results.Values are percentages.
Journal of Eastern African Studies 7
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
9/21
with vote choice. Voters who had excellent or good perceptions of government
performance on services were more likely to vote for Kenyatta (58%) over Odinga (42%).
Those numbers reverse for voters with negative perceptions of the governments
performance on services. Similarly, voters with positive perceptions of the nations
economy were 60% likely to vote for Kenyatta (versus 40% for Odinga). Those with
negative perceptions were more likely to support Odinga 54% to Kenyattas 46%.
Positive perceptions of security again correlate with support for Kenyatta (59%) and
negative perceptions with Odinga (54%). Thus, Kenyatta garnered support from those
who had positive perceptions of government performance; the opposite held for Odinga.
Voters seemed to treat Kenyatta as the incumbent in the race, crediting him for the
governments achievements or failures.
Panel II of Table 2 shows results regarding the role of campaign issues. We asked
respondents to name the most important issue in deciding how they voted for president,
using the options presented in the table. The results most often mention the economy
(27%), constitutional implementation (18%), employment (17%), security (13%), and
corruption (9%). Several of these issues correlate with citizensvotes. Kenyatta collected
the majority of votes for individuals who thought bread and butter and law and order
issues were the most important, winning 54, 56, and 58% of those naming employment,
the economy, or security respectively as their top concern. Odinga garnered most of the
votes from the reform-minded, winning 62% of those citing corruption as their most
important issue, 58% naming the ICC-The Hague, and 60% constitutional implementa-
tion. Few people reported land or tribalism as their top issue and neither of these topics
differentiated Odinga from Kenyatta supporters. The ICC result is interesting: only 3%named it as the most important issue, a finding at odds with those thinking the issue
would either a death knell or rallying cry for Kenyatta and Ruto.
Table 2. Cross-tabulation results by performance and campaign issues.
Presidential vote
Kenyatta/Ruto Odinga/Musyoka Total
I. PerformanceGovernment services
Excellent/good 58 42 49Just fair/poor 42 58 51
National economyExcellent/good 60 40 27Just fair/poor 46 54 73
Security in KenyaExcellent/good 59 41 29Just fair/poor 46 54 71
II. Campaign issuesMost important issue
Employment 54 46 17Economy 56 44 27Corruption 38 62 9ICC-The Hague 42 58 3Land 45 55 5Tribalism 46 54 6Constitutional implementation 40 60 18Security 58 42 13
Note: Values are percentages.
8 K.E. Ferreeet al.
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
10/21
Multivariate tests
This section examines correlations between ethnicity, performance, and campaign issues
and vote choice using multivariate regression models. We restrict the analysis to voters
who selected either Kenyatta or Odinga, again dropping voters who selected non-viable
candidates and RTAs for ease of presentation. We create the dependent variable equal to 1if they voted for Kenyatta, and 0 for Odinga, and we create ethnic dummies for the
Kikuyu and Luo, taking a value of 1 if true and 0 otherwise. We also create performance
variables for Services, National Economy, and Security, 1 for positive responses
(excellent and good) and 0 otherwise. Finally, we create dummy variables equaling
1 if a respondent gave any of the issue questions inTable 2, Panel II, as their answer. We
include controls for female and education (equal to 1 if the highest education level
achieved was at most primary school, and 0 otherwise) and age (equal to 1 if the voter
was a youth, classified as a voter aged 1835 years, 0 if older than 35 years). Positive
coefficients indicate a greater likelihood for a Kenyatta vote; negative coefficients
indicate a greater likelihood of an Odinga vote.Table 3presents results from four logit models of vote choice betweenKenyatta and
Odinga, clustering standard errors by polling station, and significance.31 Results are
converted from raw logit coefficients to predicted changes in the probability of voting for
Kenyatta, with other variables held at their means. Models 13 examine three different
performance variables; Model 4 combines them.32 We note the two ethnic variables for
co-ethnic candidate are strongly and substantively significant across all models. In Model
1, Kikuyus are about 60 percentage points more likely to vote for Kenyatta than non-
Kikuyus, whereas Luos are about 59 percentage points more likely to vote for Odinga
than non-Luos. Performance variables also tell an important story: Voters with positive
Table 3. Logit models of ethnicity and performance on vote choice with substantive effects shown.The dependent variable = 1 if the vote is for Kenyatta and 0 if the vote is for Odinga.
Model 1 Model 2 Model 3 Model 4
I. PerformanceServices 0.125*** 0.103***
(0.03) (0.03)Economy 0.092** 0.046
(0.03) (0.03)Security 0.090** 0.047
(0.03) (0.03)II. ControlsKikuyu 0.603*** 0.601*** 0.602*** 0.600***
(0.03) (0.03) (0.03) (0.03)Luo 0.589*** 0.591*** 0.591*** 0.588***
(0.02) (0.02) (0.02) (0.02)Female 0.047* 0.048* 0.046* 0.047*
(0.02) (0.02) (0.02) (0.02)Education 0.045 0.041 0.051+ 0.049+
(0.03) (0.03) (0.03) (0.03)Age (= youth) 0.021 0.022 0.019 0.020
(0.02) (0.02) (0.02) (0.02)N 5087 5087 5087 5087
Pseudo-R2
0.30 0.29 0.29 0.3
Note: Models are estimated with a constant; standard errors are clustered by polling station, +p< 0.1, *p< 0.05,**p < 0.01 and ***p < 0.001.
Journal of Eastern African Studies 9
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
11/21
perceptions of services, the economy, and security are about 13, 9, and 9 percentage
points more likely to vote for Kenyatta (Models 1, 2, and 3 respectively), than voters with
negative perceptions. In Model 4, services retains significance, but economy (p = 0.116)
and security (p= 0.129) fall just below. Third, women are about 5 percentage points more
likely than men to vote for Kenyatta, and education is not consistently significant
(although in Models 3 and 4, people with a primary education are about 5 percentage
points more likely to vote Odinga). Taken together, ethnicity strongly predicts vote
choice, as do many performance categories. In models not shown, we rerun all of these
models with additional ethnic controls and find that performance coefficients remain
similarly signed and significant, the Kalenjin variable is positive (reflecting greater
support for Kenyatta because of alliance with Ruto), and the Kamba is negative
(reflecting greater support for Odinga because of alliance with Musyoka).
Table 4 explores the effect of campaign issues on vote choice. Again, we estimate
logit models where the dependent variable equals 1 for a Kenyatta vote and 0 for an
Odinga vote excluding other responses. Each model includes controls for ethnicity,
female, education, and age. The ethnic controls in Models 18 work similarly to earliermodels; females are slightly more likely to vote for Kenyatta whereas education and age
no longer seem to matter once issue positions are included. Models 18 differ over the
response given to the most important issue question. Models 1 and 2 reveal that
respondents who thought employment or the economy was the most important issue
were about 6 and 5 percentage points more likely to vote for Kenyatta, respectively.
Models 3 and 4 show that corruption and ICC-The Hague worked in Odingas favor,
increasing the probability of voting for him by about 13 and 12 percentage points
respectively. Land and tribalism (Models 5 and 6) do not predict support for either
candidate. In Model 7, constitutional implementation increases the probability of an
Odinga vote by 9 percentage points, but security increases the probability of a Kenyattavote by 10 percentage points in Model 8. Similar to ethnicity and performance,
campaign issues also correlated with vote choice in substantively important ways.33
Like many observational studies before this one, we cannot definitively sort out
whether voters arrive at their evaluations of performance or their issue stances completely
independent of ethnic considerations. Luos may evaluate the incumbent governments
performance and form opinions about issues in ways colored by their ethnic identities and
experiences and the same could be true for Kikuyus and other groups. We therefore resist
making strong claims about the independent effects of ethnic, performance, and issue
variables.
We do note, however, that ethnicity and performance evaluations do not correlatestrongly in this particular election perhaps because there was no clear incumbent. In
Online Supplemental Table 1,34 we cross-tabulate the major ethnic groups fromTable 1
by each of the performance variables. While groups do have somewhat different
evaluations, and Kikuyus are generally more positive than Luos, there is substantial
within-group variation: 57% of Kikuyus and 54% of Kalenjins had positive views of
services, as did 34% of Luos and 52% of Kambas. No ethnic group rated the economy
highly: the Kikuyu (37%), Kalenjin (27%), Luo (19%), and Kamba (28%). Similar
patterns emerge for security provision. We also include a correlation matrix in Online
Supplemental Table 2 of the ethnic and performance variables from Table 3. Results
confirm that the ethnic dummies do not correlate strongly with performance variables.The highest correlation occurs between Luo and Services (r= 0.10). Similarly, ethnicity
and issue priorities also did not correlate strongly in this election (see Online Supplemental
10 K.E. Ferreeet al.
http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182 -
8/13/2019 Ferree, Gibson, And Long JEAS 2014
12/21
Table 4. Logit models of campaign issues on vote choice with substantive effects shown. The dependent variable = 1 if thfor Odinga.
Model 1 Model 2 Model 3 Model 4 Model 5 M
I. IssuesEmployment 0.056+
(0.03)Economy 0.053*
(0.02)Corruption 0.132***
(0.03)ICC-The Hague 0.121*
(0.06)Land 0.029
(0.04)Tribalism 0.
(0
Constitutional implementation
Security
II. ControlsKikuyu 0.604*** 0.602*** 0.603*** 0.604*** 0.604*** 0.
(0.03) (0.03) (0.03) (0.03) (0.03) (0.Luo 0.592*** 0.592*** 0.593*** 0.593*** 0.592*** 0.
(0.02) (0.02) (0.02) (0.02) (0.02) (0.Female 0.047* 0.048* 0.047* 0.047* 0.046* 0.
(0.02) (0.02) (0.02) (0.02) (0.02) (0.Education 0.042 0.041 0.044 0.043 0.042 0.
(0.03) (0.03) (0.03) (0.03) (0.03) (0.Age (= youth) 0.017 0.022 0.021 0.021 0.021 0.
(0.02) (0.02) (0.02) (0.02) (0.02) (0.N 5087 5087 5087 5087 5087 50Pseudo-R2 0.29 0.29 0.29 0.29 0.29 0.
Note: Models are estimated with a constant; standard errors are clustered by polling station, +p < 0.1, *p < 0.05, **p < 0.01 and ***p
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
13/21
Table 3for cross-tabulated results andOnline Supplemental Table 4for a correlation matrix).
While there were mild differences across ethnic groups (Kikuyu, Kalenjin, and Meru were
somewhat less likely than other groups to prioritize constitutional reform), the economy and
jobs ranked highly across all groups. All told, within-group differences in opinion about
performance evaluations and issues swamp across-group differences during this particular
election. While this does not rule out the possibility that ethnicity may shape performance
evaluations and issue stances in diverse and heterogeneous ways across respondents, it does
give us confidence that correlation between our independent variables is not a major concern
for our analysis.
Election results and evaluations of election quality
This section considers what exit poll data reveal about the conduct of the election and its
outcome. Exit polls can be used as a tool to verify official vote counts. Unlike actual
ballots, exit poll results are immune to ballot-stuffing, tally manipulation, or technological
failures in the voting system. Therefore, they can reflect the vote more accurately than
certified vote counts in corrupt elections. In an election free of manipulation, assuming
voter misrepresentation is randomly distributed with regard to vote choice, exit poll
results should match certified results within a margin of error.
Do the exit poll results confirm the IEBC counts? Exit polls estimates inTable 5showmean values for each candidate and 95% confidence intervals in parentheses. We
calculate confidence intervals based on two-stage sampling of polling stations propor-
tionate to size, with random selection of voters exiting polling stations, taking into
account design effects due to clustering. We also use survey weights to adjust for any
variation between planned and achieved samples per polling station. The inclusion or
exclusion of survey weights does not significantly alter mean values.35
Three findings deserve particular attention. First, the two leading candidates, Odinga
and Kenyatta, are statistically tied in the exit poll, with Odinga garnering 40.9% of the
vote to Kenyattas 40.6%. These results are consistent with the final Ipsos pre-election
tracking poll implemented ten days before the election, which also showed a statistical tiewith Kenyattas at 44.8% and Odingas at 44.4%. Second, a significant discrepancy of
almost 9.5% exists between Kenyattas totals in the exit poll and the IEBC result. Finally,
Table 5. Exit poll results and certified Independent Elections and Boundaries Commission (IEBC)results.
Exit poll IEBC
Odinga 40.90% (37.4, 44.4) 43.31%
Kenyatta 40.61% (37.1, 44.2) 50.07%Mudavadi 3.74% (2.6, 4.9) 3.93%Kenneth 1.67% (1.2, 2.1) 0.59%Ole Kiyapi 0.09% (0.0, 0.1) 0.33%Karua 0.69% (0.2, 1.2) 0.36%Muite 0.02% (0.0, 0.1) 0.10%Dida 0.43% (0.2, 0.7) 0.43%Refused to answer (RTA) 11.84% (10.0, 13.7) n.a.Rejected votes n.a. 0.88%Turnout 85.91%
n.a., not applicable.
12 K.E. Ferreeet al.
http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182http://dx.doi.org/10.1080/17531055.2013.871182 -
8/13/2019 Ferree, Gibson, And Long JEAS 2014
14/21
the percentage of RTA responses on the exit poll 11.8 exceeds the difference between
Odinga and Kenyattas exit poll vote shares and certified vote shares.
What explains the difference between the exit and Ipsos polls from the IEBC
count? Some election analysts argue that Kenyattas victory flowed from his
followers high rate of registration. But both polls were sampled from the final
registry of voters, and thus the possible differential in registration numbers for
Kenyatta and Odinga should already be accounted for in the polls results. Others
believe that higher turnout for Kenyatta voters explains the difference betw een the
polls and the final tally. The Kenyatta campaign may have succeeded in getting more
of their followers to polling stations. This could cause problems for pre-election
surveys as they interview registered voters who might not have actually voted, so
differences in turnout between Kenyatta and Odinga supporters could produce bias in
projected results. The same problem does not apply to the exit poll, however, as it
interviewed only actual voters. Moreover, the official IEBC results show consistently
high turnout in both Kenyatta and Odinga strongholds. Even weighting the exit poll
by IEBC turnout figures produces results that are statistically equivalent to Kenyattaand Odingas vote shares in Table 2.
36
Could last minute swings in the share of votes for the third through sixth place
finishers to Kenyatta have pushed him over the 50% threshold? It is extremely
unlikely: the sum of all the presidential candidates besides Odinga who lost votes
moving from the Ipsos poll to the IEBC tally only produces 2.8%. This implies that
about half of the 5.27 percentage point increase Kenyatta received when moving from
the Ipsos poll to the IEBC count would have had to have come from Odinga
supporters the most unlikely group to switch their votes. Moreover, because the exit
poll interviews people directly after they voted, it should capture all last minute
swings in support. In sum, registration, turnout, and shifts in candidate preference donot seem to explain the difference between the exit poll and the IEBC results.
Perhaps the 12% of respondents who RTA the vote choice question explains the
exit poll/certified results gap. For this to be true, Kenyatta would have to have won
80% or more of the RTAs. We explore the plausibility of this scenario by evaluating
political, social, and geographic correlates of RTAs and whether they more closely
match the profile of a typical Kenyatta supporter (partisan of TNA, Kikuyu or
Kalenjin, inhabitant of Rift Valley) or Odinga supporter (partisan of ODM, Luo or
Kamba, inhabitant of Nyanza or Eastern Province). Table 6 shows logit results for
RTAs with various political, demographic and geographic controls. For partisanship,
we use two measures: feeling close toa particular party and feeling that a particular
coalition is very or somewhat trustworthy. Both strongly predict vote choice and
have considerably less non-response than the vote choice question.37
The coefficients on the political controls are most consistent with political
neutrality among RTA voters. RTA voters were very unlikely to be either TNA or
ODM partisans; they were also very unlikely to trust either TNA or ODM. The ethnic
coefficients, on the other hand, hint at the possibility of a slight bias in favor of
Kenyatta. Luos were about three percentage points less likely than all other Kenyan
groups to refuse to answer the vote choice question. As Luos were only around 11%
of all survey respondents, it seems unlikely that this could radically skew the RTAs in
favor of Kenyatta. Also noteworthy, naming Kenyan as ones ethnic group was
strongly correlated with refusing that answer the vote choice question. Theseindividuals might simply be voters who prefer to keep their identities and vote
choices private. If so, it is difficult to think of a reason that this group would priori
Journal of Eastern African Studies 13
-
8/13/2019 Ferree, Gibson, And Long JEAS 2014
15/21
favor ODM or TNA, since they were equally likely to support Odinga and Kenyatta in
Table 1. Northeast was the only provincial dummy variable that was significantlydifferent from the omitted category (Nyanza province), suggesting again politically
Table 6. Analysis of RTAs by political and demo-graphic characteristics. The dependent variable = 1 ifthe response is RTA and 0 otherwise.
ODM partisan 0.079***(0.01)
TNA partisan
0.055***(0.01)
Trusts Jubilee 0.066***(0.01)
Trusts CORD 0.061***(0.01)
Kikuyu 0.002(0.01)
Kalenjin 0.018(0.02)
Luo 0.034*(0.01)
Kamba 0.028(0.03)
Kenyan 0.082*(0.03)
Education 0.008(0.01)
Female 0.012+
(0.01)Youth 0.002
(0.01)Middle-aged 0.022+
(0.01)
Muslim
0.039*(0.02)Nairobi 0.039
(0.04)Coast 0.017
(0.03)Northeast 0.069***
(0.02)Eastern 0.032
(0.04)Central 0.02
(0.04)Rift 0.038
(0.04)Western 0.043
(0.05)N 6258Pseudo-R2 0.11
Note: Models are estimated with a constant; standard errorsare clustered by polling station, +p < 0.1, *p < 0.05, **p