Post on 21-Jun-2020
transcript
1
Ethnic Cues and Ballot Design
Jeffrey Conroy-Krutz
Department of Political Science Michigan State University
conroyk6@msu.edu
Devra C. Moehler Annenberg School for Communication
University of Pennsylvania dmoehler@asc.upenn.edu
[Please do not cite or circulate without permission from the authors]
Draft: 3 January 2014
Abstract Researchers often assume that ethnic cues affect voting in developing countries by informing citizens of candidates’ identities, thereby affecting expectations of how candidates would behave if elected. However, ethnicity is only one of multiple heuristics available to voters, and it might not be the most salient on election day. We posit that ethnic cues might affect voting not only by providing information about candidate ethnicity, but also by priming identity. In order to test this theory, we conducted an experiment days prior to the 2011 Ugandan elections. We measured the effects of a brief stimulus immediately before a vote is cast: exposure to candidate photographs on ballots. We find that the inclusion of photographs increased ethnic voting, and our evidence suggests a priming, rather than an informational mechanism. We argue that subtle stimuli at the end of a campaign can have sizeable effects on ethnic-voting rates by altering identity salience. We would like to thank Diana Mutz and Robin Pemantle for their help in designing analytical strategies for this paper; Rosario Aguilar Pariente, our co-researcher in aspects of this project relating to data collection and party-based voting; and André Blais, Joseph Cappella, Monica Schneider, Karleen Jones West, Susanna Wing, and attendees at talks at Michigan State University, the 2012 Midwest Political Science Association Annual Meeting, 2012 African Studies Association Annual Meeting, and the 2013 Pre-APSA Workshop on Electoral Integrity for helpful comments. We would also like to thank John Kavuma, James Odongo, Jin Woo Kim, and Douglas Allen for their assistance with the research, and the University of Pennsylvania Annenberg School for financial support. All remaining errors and omissions are our own.
2
Ethnic Cues and Ballot Design
A growing body of literature suggests that voters in many contexts use ethnicity as an
indicator for how candidates would behave in office. Instrumental voters often presume that co-
ethnics will deliver benefits to in-group members, and cast their ballots accordingly (Chandra
2004; Birnir 2007; Ferree 2011). Information that can enable ethnic identification, such as
candidate name, dialect or accent, dress, and facial phenotype are presumed to be readily
accessible to voters. When other types of political information are scarce, as is often the case in
the developing world, voters might therefore rely heavily upon such ethnic cues (Chandra 2004).
In short, scholars argue that ethnic information is cheap and provides utility to the voter, while
other kinds of information are more costly.
However, information about candidate ethnicity might not be readily available and
inexorably salient for voters. Even in low-information environments, there are multiple potential
heuristics for evaluating candidates, some of which might be more observable or relevant to
voters than ethnicity.1 We hypothesize that some observed variation in ethnic-voting rates can
be explained by the presence or absence of ethnic cues proximate to the time of the electoral
decision. Such signals, even when subtle, might have sizeable effects on behaviors.
Importantly, while these cues may provide information about candidate ethnicity, they might also
prime ethnic considerations. Both mechanisms should theoretically increase the probability of
ethnic voting.
Scholars tend to interpret cue effects in different ways based on their region of study.
Literature on the developing world usually assumes that cue effects are informational, while
those studying developed democracies usually focus on a priming mechanism. Few studies
1 Alternative criteria include candidate party affiliation, place of origin, age, gender, status, education, and delivery of largesse.
3
make an empirical distinction between these two causal pathways, and many likely conflate the
two (Lenz 2009).2 Our research is designed to evaluate the effects of brief stimuli on voting and
to probe the mechanisms responsible for cue effects.
We evaluate the effects of one potential source of ethnic cues: candidate photographs on
election ballots. Ballots often include candidate images to facilitate voting by illiterate and
otherwise under-informed populations, especially in the developing world. Since the ballot is the
last external stimulus on a voter’s decision making, any cues that it provides might have
especially significant impacts on vote choices.3 If images signal ethnic identities, then the
inclusion of candidate photographs on ballots could increase citizens’ likelihood of casting a vote
in favor of co-ethnics, through informational and/or priming mechanisms.
To identify the effects of cues on ballots, we conducted an experiment in Uganda, a
country where ethnicity is politically salient. In early February 2011, just days prior to national
and local elections, we randomly assigned subjects to mark different types of mock ballots that
included or excluded party names, party symbols, and candidate photographs. In this article we
focus on how candidate images affect ethnic voting.
Our findings indicate that ballot design significantly impacts ethnic-voting rates. The
inclusion of photographs resulted in a 61.6% increase in the likelihood that subjects would mark
a ballot in favor of a co-ethnic. Evidence suggests that increases in ethnic voting are not due to
an informational mechanism. Subjects looking at candidates’ photographs were no better at
identifying candidates’ ethnicities than those who did not see photographs. Rather, we do find
2 While Lenz (2009) reevaluates existing research and concludes that many effects that have been characterized as priming likely occurred through learning, we consider that effects that have largely been characterized as informational (i.e., learning about candidates’ ethnicity through cues) could be due to priming. 3 In a study on such last-minute cutes, Berger, et al. (2008) find that the type of building in which a polling station is located can affect local vote outcomes. For example, support for initiatives to finance schools is significantly higher at polling stations located in schools, which the authors attribute to a priming effect.
4
suggestive evidence in support of a priming mechanism. Subjects who received ballots that
contained candidate photographs were subsequently more likely to stress ethnic aspects of their
identity over national ones, indicating that the treatment effects increased the salience of ethnic
considerations. We conclude that subtle stimuli at the end of an election campaign can have
sizeable effects on ethnic-voting rates, perhaps by altering identity salience, even when
knowledge is unaffected. We argue that scholars of the developing world should be more
attentive to information-processes biases when theorizing about the causes of ethnic voting.
The paper proceeds as follows. The first section provides an overview of existing
literature on cues and ethnic voting. In the second, we present hypotheses on the relationship
between the inclusion of candidate images on ballots and vote outcomes. The third section
discusses the Uganda case and provides an overview of our survey and experimental
methodologies, as well as the measures we use to test the hypotheses. Section four covers
analyses and presents our findings, while the fifth concludes with a discussion of potential
theoretical and normative implications.
Cues and Ethnic Voting
Many societies experience high rates of ethnic voting, which is associated with a range of
pathologies, including limitations on democratic accountability and exacerbated inter-group
tensions. To explain the especially high prevalence of such behavior in the developing world,
informational theories (Chandra 2004; Posner 2005; Birnir 2007; Ferree 2011; Conroy-Krutz
2013) start with the premise that such societies tend to be marked by costly and scarce political
information, owing to factors such as lack of access to detailed information about political
candidates and high electoral volatility (Keefer 2007; Birnir 2007). Like their counterparts in
5
developed contexts (Popkin 1994), voters in such societies find cues helpful in reducing the costs
of information-gathering and –processing.
Ethnicity is one such cue, and one that can often be ascertained relatively cheaply, from
observations of a candidate’s name, face, skin color, dress, dialect or accent (Chandra 2004).
Voters can then draw upon stereotypes and real or imagined political histories to determine
whether candidates, given their identities, are likely to implement favorable policies (Bates 1974;
Chandra 2004; Posner 2005). This results in apparently high rates of support for co-ethnic
candidates.4
However, there are at least two significant limitations of such informational theories, at
least in terms of how they are applied to studies of elections in the developing world. First, even
though ethnic information might be relatively cheap, it is neither costless nor perfect. In some
situations, name will be a highly accurate indicator of ethnicity (Isaacs 1975; Fershtman and
Gneezy 2001); in others, it might be more ambiguous, and voters interested in determining
candidates’ ethnic identities will have to turn to other potential indicators, such as dress, or facial
structure or markings (Habyarimana, et al. 2009: 54-5). In sum, citizens’ abilities to code
others’ ethnic identities correctly are highly variable, and often surprisingly limited, even in
societies in which ethnicity is politically and socially relevant (Harris and Findley forthcoming).
There are likely instances in which citizens enter the voting booth without previously having
made accurate determinations of all candidates’ ethnic affiliations, even if ethnicity is politically
salient, given the cost of information gathering. In fact, the average subject in our experiment
who received our control mock ballot (i.e., including candidates’ names as the only piece of
information) could only correctly identify the ethnicity of 52% (7.8 of 15) of candidates included
4 Voters might also draw conclusions about candidates’ electoral viability from their ethnic identity, which might discourage members of less-populous groups from voting for co-ethnics (Chandra 2004; Conroy-Krutz 2013) or associated parties (Ichino & Nathan 2013).
6
in the study.5 Given that the study was conducted in an area with only two ethnic groups, and
just days (and sometimes hours) before the actual election, this finding suggests that ethnic
information is not, in fact, perfect on election day.
Second, even in situations in which ethnic information is relatively cheap, other types of
cheap cues might also be prevalent. Such cues include the candidate’s partisan affiliation, place
of origin, gender, status, education, or past delivery of largesse to the voter or his or her
community.6 Voters may find these alternate indicators more accessible or reliable predictors of
future behavior, and thus non-ethnic considerations might trump ethnic ones (Casey 2013;
Conroy-Krutz 2013).
Theories of ethnic voting should account for variation in both the availability and relative
salience of ethnic information. This variation is important, in that it could explain, at least in
part, differences in rates of ethnic voting, across countries, over time, and between individuals.7
To date, theories that attempt to explain such variation have tended to focus on how institutional
(Rokkan 1970; Posner 2005; Huber 2012) and demographic (Chandra 2004; Miguel 2004;
Posner 2004a; Dunning and Harrison 2010; Ichino and Nathan 2013) factors impact the strategic
considerations of elites and/or voters. Others have examined how the availability of information
that would facilitate more informed, policy-based evaluations affects support for co-ethnics
(Banerjee, et al. 2011; Pande 2011; Casey 2013; Conroy-Krutz 2013). We propose that the
availability of certain ethnic cues could also affect ethnic-voting rates. Notably, even ephemeral
5 This low figure is not driven by subjects’ inabilities to identify low-viability candidates. Those assigned to the control performed about the same on the ethnic-identification task—53%, or 2.1 out of 4—when considering only candidates from the two largest parties (the National Resistance Movement and the Forum for Democratic Change). 6 Even in Africa, where ethnicity is held to be a particularly dominant consideration, studies have demonstrated that individuals use other information, some derived from cues, in their electoral decision making (Posner and Simon 2002; Lindberg and Morrison 2008; Bratton, et al. 2011; Conroy-Krutz 2013; Hoffman and Long 2013). 7 For a discussion of this variation in contemporary Africa, for example, see Cheeseman and Ford (2007).
7
cues may generate sizeable shifts in vote choice, whereas these other factors found to affect
ethnic-voting rates are relatively stable.
While the literature on ethnic voting in the developing world tends to focus on how these
cues help voters learn about candidates’ identities (i.e., an informational mechanism), we argue
that cues can affect ethnic-voting rates through a different pathway: by priming individuals to
consider ethnicity over other judgment criteria, or stimulating previously dormant attitudes.
Increases in either—knowledge or salience —could increase rates of ethnic voting.
Although a significant literature on developed democracies highlights the priming
potential of cues in electoral decision-making processes (Chaiken 1980; Petty & Cacioppo 1986;
Krosnick & Kinder 1990; Mondak 1993; Rahn 1993; Kam 2005; Lau, et al. 2008), scholarship
on the developing world has tended to privilege the informational mechanism, largely
overlooking the possibility of cues’ priming effects.8 This oversight could stem from the fact
that learning and priming effects may be observationally equivalent. In fact, as Lenz (2009)
suggests, much of the research identifying priming effects in contexts such as the United States
might have mischaracterized mechanisms that actually constitute learning. This demonstrates
the need for research, in both developed and developing settings, that focuses on elucidating the
mechanisms behind ethnic cue effects.9 In the next section, we discuss how one factor that often
provides such cues—election ballots—could impact the availability and salience of ethnic
information, and thus significantly impact overall rates of ethnic voting.
8 Adida (n.d.) on Benin is an exception. She finds that a brief ethnic cue increases ethnic-based candidate preferences and provides suggestive evidence that the change is due to priming, but only for the subset of her subject population with near perfect knowledge of candidate ethnicity. 9 While other experimental research has determined that brief stimuli can affect ethnic voting (Dunning and Harrison 2011; Conroy-Krutz 2013; Hoffman and Long 2013), our research design allows us to the test the mechanisms through which ethnic cues might affect decision-making in an actual election.
8
Last-Minute Cues: Ballot Design and Ethnic Voting
One source of ethnic cues—and one that has been overlooked in the literature on
voting—is ballots themselves. There has been significant variation in ballot design around the
world and over time. More limited designs include only the candidates’ names, or, in partisan
competition, candidates’ names and affiliated parties. In other instances, information such as
candidates’ mailing addresses or occupations appears. And visual features, such as candidate
photographs, are often prominent, especially in the developing world (Reynolds and Steenbergen
2006). Advocates of the inclusion of images maintain that such features facilitate voting by
citizens who have little education, access to political information, or previous experience with
voting (Smith, et al. 2009).
While there is a literature that focuses on how ballot design affects the accurate recording
of voters’ choices,10 little research has been conducted on the possible effects of including
various pieces of information about competitors, such as candidate images, on ballots (Reynolds
and Steenbergen 2006). Such elements could have (seemingly) unintended consequences on
vote outcomes (Katz, et al. 2010). We contend that one of these consequences could be an
increase in ethnic voting.
Candidates’ images on ballots can affect ethnic voting through at least two pathways:
providing information on candidate identity or priming photo-related considerations. First, facial
phenotype and clothing often presumably convey information about ethnic identity (Chandra
2004).11 One of the most notorious recent examples comes from Rwanda, where Tutsi were held
10 For examples from the US, see Wand, et al. 2001; Herron and Sekhon 2003; Ansolabehere and Stewart 2005; Herron and Wand 2007; and Herrnson, et al. 2012. 11 Candidate images can also affect voters’ evaluations in other ways, and there is a broad literature on how candidate attractiveness or general image during campaigns (Antonakis and Dalgas 2009; Bailenson, et al. 2008; Boudreau 2009; Lawson, et al. 2010; Lenz and Lawson 2011; Sigelman, et al. 1987; Todorov, et al. 2005; Tsfati, et al. 2010) and on ballots (Banducci, et al. 2008; Buckley, et al. 2007; Dumitrescu 2010; Johns & Shephard 2011) affects electoral outcomes.
9
to have slender noses, lighter skin tones, and more angular faces, in contrast to the flatter noses
and darker skin of Hutu. If citizens prefer voting for co-ethnics, but lack knowledge of
candidates’ identities before entering the voting booth, then ethnic cues conveyed through
candidates’ photographs on ballots could facilitate ethnic voting. In other words, photographs
could enable otherwise uninformed individuals to make use of ethnic heuristics.
Second, candidates’ images might prime voters to weight ethnic considerations more
heavily in their electoral decision making.12 There are a number of manners through which
photographs might increase the salience of ethnic considerations. The ethnic heterogeneity of
the candidate pool could be more apparent to voters when photographs are present, thus
underlining the importance of inter-ethnic competition in a diverse environment. An alternate
expectation is that voters find it psychologically more difficult to vote against a co-ethnic, and
for a non co-ethnic, when a recognized co-ethnic is, in essence, staring them in the face. Finally,
photos might call attention to personal characteristics of the candidate, one of the more important
being ethnicity, over alternate characteristics such as party affiliation.
If either mechanism—informational or priming—is at work, we should therefore expect
the following:
H1: The inclusion of candidate photographs on ballots will increase rates of support for
co-ethnic candidates.
Empirical support for H1 does not in itself identify the mechanism or mechanisms leading
to increases in co-ethnic voting, however. If the informational mechanism is operative, we
should expect that the inclusion of candidates’ photographs on ballots will improve voters’
abilities to identify candidates’ ethnic identities correctly. Candidate faces on ballots will
12 In our study of party-based voting, evidence indicates that the presence of partisan identifiers on ballots encourages voters to weight the importance of party more significantly in their electoral decision-making process ([working paper by authors]).
10
improve voters’ assessments, provided that a) name alone does not indicate ethnicity perfectly to
all voters, b) facial features and dress are, in fact, indicators of ethnic identity in the particular
context, and c) voters do not obtain or retain sufficient information about candidates’ ethnic
identities elsewhere before they view the ballot. In sum, we hypothesize that:
H2: The inclusion of candidate photographs on ballots will increase voters’ abilities to
code candidates’ ethnic identities correctly.
Finally, if candidate images on ballots increase ethnic voting through a priming
mechanism, then we should expect that photographs will result in voters’ attaching greater
importance to their ethnic identity after receiving such a ballot. One way of measuring the
salience of ethnic identity is by asking individuals to assess the relative importance of their
ethnic identity, vis-à-vis their national identity.13 In doing so, we expect that:
H3: The inclusion of candidate photographs on ballots will increase voters’ attachment to
their ethnicity identity, vis-à-vis their national identity.
In sum, we expect that the inclusion of candidate photographs on ballots will increase
ethnic voting, and that this increase will occur via an informational and/or a priming mechanism.
Case Selection and the Experiment
Country Selection
Uganda was an ideal country in which to test these hypotheses, for a number of reasons.
First, it has a history of including various elements on its ballots in order to facilitate correct and
autonomous voting by all individuals.14 Candidate photographs, for example, first appeared on
13 Eifert, et al. (2010) use the same type of question to assess the effects of political competition (i.e., proximity to elections) on the salience of ethnic identity in Africa. 14 In many countries, voters are allowed to bring another individual into the voting booth if they claim they cannot read or cast a ballot on their own. This raises fears of diminished privacy and manipulation.
11
ballots for the 1994 Constituency Assembly elections, and party symbols were added starting in
2006. It would not seem particularly odd to Ugandans to mark ballots that included (or
excluded) various visual elements. Furthermore, the implications of including certain ballot
features are especially relevant in Uganda given variation in practice over time.
Second, ethnicity has been central to Ugandan politics since independence. Regardless of
the metric used (Posner 2004b: 856), the country ranks as one of the most ethnically diverse in
Africa and, indeed, the world. The largest group—the Baganda, who are the dominant group in
the politically and economically important Central Region of the country, around Kampala—
comprise only 16.9% of the population, according to the 2002 Census. Other major groups
include President Yoweri Museveni’s Banyankole (9.5%), Basoga (8.4%), Bakiga (6.9%), Iteso
(6.4%), and Langi (6.1%).15 Ugandan parties have, since independence, been structured along
regional, ethnic, and religious schisms (Kasfir 1976). Two of the earliest, the Uganda People’s
Congress (UPC) and the Democratic Party (DP), eventually became associated with Northerners
(i.e., Acholi and Langi) and Catholic Baganda, respectively. Ethnic schisms were exacerbated
by events such as the abolition of traditional kingdoms in 1966, Idi Amin’s violent purges
against Acholi and Langi soldiers in the 1970s, and President Milton Obote’s brutal counter-
insurgency campaign against the primarily Baganda population in the Luweero Triangle in the
1980s.
The so-called Movement system, which banned election-related activities by parties until
it was abolished in favor of multipartyism in 2005, was ostensibly an attempt to de-ethnicize
Ugandan politics (Museveni 1997), and in many respects Museveni’s National Resistance
Movement (NRM) was successful in rebuilding the Ugandan state after decades of instability
15 No other group claimed more than 5.0% of the population; nearly half (45.8%) of Ugandans are affiliated with one of these smaller groups.
12
(Rubongoya 2007; Gilley 2009). However, while the NRM maintains a multiethnic base,
anchored by strong electoral support from Western groups like the Banyankole and Central
groups like the Baganda, opposition parties have had more limited ethnic bases of support, and
these tendencies only increased following the 2006 elections (Cheeseman and Ford 2007). At
least prior to the 2011 elections, which involved a potential realignment of support amongst
some Northern groups toward the NRM (Conroy-Krutz and Logan 2013), the main opposition
Forum for Democratic Change (FDC) performed well among Acholi, Langi, Iteso and other
Northern groups, while the UPC and DP remain largely Langi and Baganda parties, respectively.
It is, however, important to note that Ugandans have exhibited a strong willingness to
cross ethnic lines during elections (Tripp 2011: 54-6). After all, the two leading candidates in
the last three presidential elections—Museveni and the FDC’s Kizza Besigye—are both
Banyankole. Together, these two have won between 94 and 97% of the vote since 2001, despite
the fact that their group comprises less than 10% of the population.
This reality—salience of ethnicity in political competition, but demonstrated willingness
to cross ethnic lines in voting—made Uganda an ideal site for our study of the effects of ballot
design on ethnic voting. Selection of a case at either extreme—one in which ethnicity is
politically unimportant and one in which inter-ethnic schisms are so deep as to prevent any
cross-ethnic voting—would likely lead to Type II errors in our attempts to draw generalizable
lessons about ballot design and ethnic voting. If ethnicity is not an important consideration at all
in voting, then elements of ballot design that contain ethnic cues (e.g., names, candidate faces)
should have no effect on likelihood of supporting a co-ethnic. On the other hand, if ethnicity is
an overwhelming consideration, then no element of ballot design is likely to be associated with
lower rates of ethnic voting. Since most African electoral competitions fall somewhere between
13
these two extremes—ethnicity being an important, but not completely overriding
consideration 16 —conducting the experiment in an area that falls close to either would be
inadvisable.
Experimental Design
In order to test the effect of ballot design on outcomes such as ethnic voting, we
conducted a survey experiment in Uganda in the days prior to the country’s 2011 general
elections. Subjects filled out sample ballots to indicate their preferred candidate in each of four
upcoming races: president, Member of Parliament, district chairperson, and district women’s
Member of Parliament.17 The ballots that subjects filled out varied in design, however. By
randomly varying ballot type, we could be confident that any observed differences in outcomes
across treatment groups could be attributed to the difference in ballot design, rather than to other
factors.
Subjects were randomly assigned to one of five treatment conditions. Ballots across all
treatment conditions included the candidates’ names, and Treatment 1 (the control) provided no
additional information. Treatment 2 also included party name. Treatment 3 augmented this, by
including party symbols as well. 18 In Treatment 4, candidate photographs were the only
elements other than candidate names included. Treatment 5 included all elements: candidate
names, party names, party symbols, and candidate photographs; this condition most closely
approximates the current design of actual Ugandan ballots. In short, Treatments 2, 3, and 5
16 See, for example, Posner and Simon (2002); Cheeseman and Ford (2007); Bratton, et al. (2011); Hoffman and Long (2013). 17 Uganda is divided into 111 districts, the populations of which each elect a district chairperson (akin to a governor). In addition, the electorate of each district elects one woman to Parliament. 18 Political parties select their own symbols, although all are subject to final approval by the Electoral Commission (EC). Independent candidates (running for offices other than president) choose a symbol from a preset list established by the EC. In treatments in which we include party symbols (i.e., Treatments 3 and 5), we also included the symbols assigned to each independent candidate (e.g., a soccer ball, a radio, a clock, a saucepan, etc.).
14
included information about candidates’ partisan affiliations, and Treatments 4 and 5 included
candidate photographs. These treatment conditions are summarized in Table 1. In all four races,
respondents were assigned to the same treatment (e.g., Treatment 5 for president, Member of
Parliament, district chair, and district women’s representative, etc.).
[Table 1 goes around here]
The information provided across the treatments always accurately portrayed the actual
candidates competing in the upcoming election. Given that experiments on information and
ethnicity often include hypothetical candidates (Dunning and Harrison 2011; Gibson and Long
2012; Conroy-Krutz 2013), our design offers significant advantages in terms of external validity.
Ballot designs for a parliamentary election are available in Appendix A.19
Site Selection
The experiment was conducted in one parliamentary constituency in Uganda. We
selected Soroti County, which is located in the Teso Sub-Region, near the eastern tip of Lake
Kyoga, for two reasons. First, we wanted to work for practical reasons in a district in which one
of us was already conducting fieldwork as part of a broader data-collecting effort on the 2010-11
election campaign. Second, since we were interested in how ballot design affects ethnic voting,
we needed to select an electoral area in which an ethnically diverse candidate pool was
competing for the votes of an ethnically diverse population.
19 Our sample ballots did, however, differ in key ways from actual ballots. First, they lacked the official seal of the Electoral Commission, and also were very clearly marked as “Sample Ballots.” Subjects were reminded of this multiple times by interviewers: “Please keep in mind that this is only a sample ballot, and there is language on the ballot that indicates that. If you want to cast an actual ballot, you will have to go to your local polling station to do so.” Finally, at the close of questions, respondents were told “We also want to remind you that the ballots you marked were not real ballots for the election. You must vote again at your polling station on voting days if you want your choice to count for the Ugandan election.” These strategies were employed because we did not want any of our subjects to eschew casting a real ballot because they thought they had done so through our experiment.
15
Soroti fits this second criterion well; it is an ethnically divided constituency, in that about
two-thirds of the population (69.1%, according to the 2002 Census) is Iteso, while most of the
remainder (29.1%) is Kumam. In addition, we identified two inter-ethnic races in Soroti County
in 2011: Member of Parliament and district chairperson, both of which included both Kumam
and Iteso candidates.20
Further, in recent years, the Iteso-Kumam division in the Teso Sub-Region has been
politically salient. In the last presidential election before we conducted our experiment (2006),
Besigye of the FDC won at every polling station in Soroti County, but he did best in the more
heavily Kumam areas, and there is a significant, positive correlation between Besigye share at a
polling station and Kumam share (r=.371, p=.00, N=101 stations).21
In the parliamentary election of that year, the leading Kumam candidate in the race—
Peter Omolo, of the FDC—ran strongest in Kumam areas, and there is a positive, significant
relationship between Kumam share and Omolo’s share in that election (r=.609, p=.00).
However, Omolo’s overall share also suggests that some significant portion of Iteso voters was
willing to support a Kumam candidate. After all, Kumams represent less than one third of the
population of the constituency, yet Kumam candidates managed to win 72% of the vote in
2006.22 Finally, in order to measure more precisely rates of ethnic voting in Soroti County’s
recent past, we utilize King’s (1997) ecological inference (EI) method to generate point estimates
of the share of Kumams and Iteso voting for each candidate at each polling station in Soroti in
20 In the MP race, Kumam candidates included Peter Omolo of the FDC (the incumbent), Vincent I. Enomu (NRM), and Simon Peter Ebitu (independent), while Iteso candidates included Engirot Lawrence Okae (UPC), Raphael Okoropot (DP), Jimmy Oriokot (People’s Progressive Party, PPP), and independents Samuel Anyolo, John Lule, and William Obit. In the district chairperson race, George Michael Egunyu (NRM) was Kumam, while Daniel Ediau (FDC), Napoleon Martina Oliba (UPC), and independents Leonard Otekat Ekapu, George William Okwaput, and Jorem Obicho Opian were Iteso. All candidates for district women’s MP were Iteso, while there were no Kumam or Iteso candidates for president. 21 Ethnic measures are only available at the parish level; Soroti parishes in 2006 had between two and six polling stations apiece. 22 Omolo won 62%, while the other Kumam candidate—Ateker Ejalu, of the NRM—won 10%.
16
2006.23 Not surprisingly, candidates tended to perform better with members of their own ethnic
groups.24 The estimated mean share of Kumams voting for one of the two Kumam candidates
was .88 (SD=.006), while the candidates’ estimated mean share among Iteso was only .65
(SD=.002).25 The estimated mean shares for the two Iteso candidates26 were .33 (SD=.002)
among co-ethnics and .09 (SD=.005) among non co-ethnics. In short, these histories suggest
that, while ethnicity appears to have political salience in Soroti County, schisms between
Kumam and Iteso residents are also not so stark as to prevent cross-ethnic voting.
Respondent Selection
Within Soroti, respondents were selected through a multi-stage design. The first stage
involved the selection of forty-five Enumeration Areas (EAs), with each EA’s probability of
selection being directly proportional to its population size as of the 2002 census. The selected
EAs cover all seven of the sub-counties and nineteen of the twenty-six parishes in Soroti County
(Figure 1). Within selected EAs, interviewers selected households via a random-walk pattern,
while subjects were recruited from selected households using a kish grid. Subjects had to be at
least eighteen years of age, a Ugandan citizen, and be able to participate in an interview
conducted in one of the three survey languages (English, Iteso, and Kumam). We imposed no
requirement that respondents had to be registered to vote in the upcoming election. The vast
majority (93.5%) of those selected consented and successfully completed the survey. All the
surveys were conducted between 10 February and 17 February 2011.
23 Since there are not reliable estimates of turnout at each polling station, the analysis assumes that there were not significant differences across ethnic groups in turnout rates in Soroti. 24 The exception was Ateker, whose estimated mean share was higher among Iteso (.12, SD=.001) than amongst fellow Kumams (.05, SD=.003). This was likely due to the fact that Kumam support is particularly strong for the FDC, and Ateker was a NRM candidate. 25 These means are weighted by the number of votes cast per polling station. 26 Samuel Anyolo, an independent, and Engirot Lawrence Okae, of the UPC.
17
[Figure 1 goes around here]
The external validity of our study benefits from the fact that research was conducted just
days before the actual elections.27 In a real-world setting, any stimulus that individuals receive
from ballots themselves will come after a campaign period. Studies on ballot design conducted
earlier in, or even prior to, actual campaigns, are likely to find larger effects, but since we are
interested in the effects of the inclusion (or exclusion) of certain ballot elements on vote choice,
we sought to conduct the experiment as close to the election as possible.
Early in the interview, respondents were asked to fill out mock ballots, in private, for
presidential, Member of Parliament, district chairperson, and district women’s Member of
Parliament races. Respondents then placed their ballots in envelopes, which were serialized to
allow for later matching with completed questionnaires.
Finally, we should of course be careful about making generalizations about any findings
from one electoral constituency to the rest of Uganda. According to the 2002 Census, Soroti
County is somewhat poorer than the rest of Uganda, although it is only slightly less literate
(61.2% versus 68.2% nationally). Our sample is representative of Soroti County at the
household level, but our analyses of the ballot conditions are not representative of individuals
within households. 28 Despite this, we are not aware of any specific issues that would suggest
that our experimental results should not be generalizable (Kam, et al. 2007).
27 Presidential and parliamentary elections were held on 18 February, while district elections were held on 23 February. 28 Stemming from the fact that questionnaire numbers were used both to select respondents according to kish grid rules and to conduct the randomization, an unanticipated interaction occurred, in which not all positions on the kish grid had an equal probability of being assigned to each ballot treatment. Therefore, in all of the treatment comparisons reported in the analysis section, we compare respondents only from those positions on the kish grid (which were determined by the number of individuals in a household and age rankings within that household) that had an equal probability of being assigned to the paired conditions. This maintains the experimental design, but does cost us considerable statistical power, in that it reduces the number of observations per comparison by about 50%. It also limits external validity, in that the analysis population is not representative of the population at the within-household level. For example, respondents in single-person households are more prevalent in our analysis of
18
Results
Again, we are primarily interested in this paper in the effects of ballot design—
specifically, the inclusion of candidate photographs—on ethnic voting. Therefore, our primary
dependent variable is support for co-ethnics. We employed a number of local research assistants
to code the ethnicity of all parliamentary and district candidates in Soroti County. In cases of
inter-coder discrepancy, family members of the candidates were interviewed to ascertain ethnic
identities. To code support for co-ethnic candidates we matched the self-reported “tribal
identity” of the respondents with the ethnicity of candidates for whom they voted.29 The
resulting variable ranges from 0 to 2, indicating the number of times that an individual voted for
a co-ethnic.30 English-language wordings for all relevant questions are reported in Appendix B,
while descriptive statistics for all dependent variables are presented in Appendix C.
In order to test H1, we run an ordered logit model, in which we regress this dependent
variable on a dummy variable indicating whether or not the subject received ballots including
candidate photographs, as well as a dummy indicating whether or not the ballot included a party
identifier. The results are statistically significant (p=.06), and in the direction hypothesized
(Table 2). 31 The results are also substantively large; subjects who received ballots with
the combined party cues (27%) than they are in the general (representative) sample (16%). The gender distribution is statistically indistinguishable in our analysis sample and the general sample, but respondents are about 4 years older on average, and the age difference is statistically significant. 29 63.2% of our respondents were Iteso, while 34.8% were Kumam. 1.8% of the sample either does not have an ethnicity identified, or has an ethnicity identified that was neither Kumam nor Iteso. These individuals are excluded from all analyses. 30 We exclude the presidential race because no subjects had co-ethnics in the race. We also exclude the district women’s Member of Parliament race from this analysis, since all candidates here were Iteso. 31 The difference in ethnic voting does not seem to be the result of different voting rates. An ordered logit model in which a variable measuring the number of ballots marked for the two races studied here (0-2) is regressed on a picture dummy and a control indicating inclusion of a party identifier finds no significant effect of pictures on the dependent variable (see Appendix D, column A).
19
photographs were 61.6% more likely to support a co-ethnic candidate than those whose ballots
did not include them, thus supporting H1.32
[Table 2 goes around here]
Are these changes in voting behavior the result of an informational mechanism, a priming
mechanism, or possibly both? Evaluating causal mechanisms is a difficult task with respect to
causal inference (Green, et al. 2010).33 We can only provide suggestive evidence of mediation,
which we do by examining whether our experimental treatments are associated with the
hypothesized mediating variables.
We first evaluate the hypothesis of an informational mechanism. H2 states that subjects
who receive ballots with photographs will code candidates’ ethnic identities more accurately than
those who do not. Our results do not support this expectation about an informational
mechanism. After “casting” their ballots, subjects were shown blank ballots, of the type they had
just been asked to complete, and were asked a series of questions about each of the candidates.34
In other words, subjects in the photograph conditions were looking at photographs both when
they marked their ballots and again when asked about the ethnicity of each candidate, while
those in non-photograph conditions never saw candidate photographs at any stage in the process.
Subjects were then asked to identify each candidate’s ethnicity. A correct answer was coded as
1, while an incorrect answer or a “don’t know” response was coded as 0. These scores were then 32 We find no significant evidence that individuals are more likely to vote for perceived co-ethnics when photographs are present (see Appendix D, column B). We construct a dependent variable similar to that used for H1, except here the consideration is not whether the subject cast a ballot for an actual co-ethnic, but rather for one he or she perceived to be a co-ethnic. We were forced to exclude the many respondents who answered “don’t know” to the relevant questions about candidate ethnicity, which might account for the puzzling findings. 33 We must assume sequential ignorability to identify the model. The first assumption, that the treatment is independent of all potential values of the outcome and mediating variables, is satisfied due to random assignment. The second assumption, that the observed mediator is independent of all potential post-treatment and pre-treatment confounding variables, is a stronger one. This assumption implies that knowledge (or ethnic salience) is unrelated to other outcomes of exposure to photographs as well as to pre-treatment subject characteristics (Imai, et al. 2010). 34 Here, interviewers were instructed to point to the relevant candidate as each question was asked; protocols stipulated that the interviewers provide no assistance whatsoever during the earlier phase, in which subjects filled out and cast their ballots.
20
summed for all fifteen MP and district chair candidates. Scores could therefore range from 0 to
15, with an individual’s score indicating the number of candidates he or she correctly identified.
OLS regression analysis yields null results, indicating that the photographs did not help
respondents correctly infer candidate ethnic identity.35 There are three possible explanations for
this finding. First, given that the experiment was conducted at the end of a campaign, subjects
might have already known most candidates’ ethnic identities before the treatment was
administered. In other words, the null finding here might be due to a ceiling effect. However,
the fact that the mean number of candidates correctly coded by respondents in our control
category was 7.8, out of a possible 15, suggests there was room for improvement here.
Second, subjects might have devoted minimal attention to questions about minor or non-
preferred candidates, and only used photographs to gather ethnic information about those
candidates they deemed viable, or for whom they considered voting. While we cannot identify all
candidates a subject considered, if subjects did use the photographs to learn about candidates’
ethnic identities, they should at least be better able than those in non-photograph treatments to
identify the identity of the candidates for whom they voted. However, the mean number of
correctly identified preferred candidates (i.e., those MP and district chair candidates for which
the subject “voted”) did not differ significantly between the control (1.3) and treatments
involving pictures (1.4 without party identifiers and 1.3 with party identifiers). Ordered logit
regression analysis conducted to measure the effect of including photographs on ballots on
likelihood of knowing a favored candidate’s ethnicity also yields insignificant results (Appendix
35 A similar analysis including district women’s candidates (i.e., a pool of twenty-three candidates, rather than just the fifteen MP and district chair competitors) also yields null results (b=-.19, p=.76).
21
D, column C). Finally, photographs do not seem to have helped subjects perform better on the
ethnic-identification task if we limit our focus to major-party (read, NRM and FDC) candidates.36
The last possibility, and the one that seems most likely, is that the photographs actually
did not provide clear signals about candidate ethnicity for the uninformed. In fact, our subjects
do not seem to have had any expectation that the photographs were increasing their coding
capabilities, as we find no evidence that subjects in photograph treatments had significantly
higher confidence in their knowledge of candidate ethnicities (see Appendix D, column E).In
short, it does not appear that an informational mechanism can explain the photograph-induced
observed increase in support for co-ethnics.
Even if the photographs did not increase ethnic voting through an informational
mechanism, there is potential that they affected subjects’ decisions by making ethnic identity
more salient. In order to test this priming hypothesis (H3), we measured the strength of subjects’
ethnic attachments after exposure to the different ballots. If the increase in ethnic voting that we
observe under photograph conditions is attributable to a priming mechanism, then we should
expect that individuals who filled out ballots with photographs will demonstrate stronger ethnic
sentiments than those who did not. To measure these sentiments, subjects were asked to weigh
the value of their ethnic versus national identities. Individuals had the option of identifying as
“only Ugandan” (coded as 0), “more Ugandan than [ethnic group]” (1), “equally Ugandan and
[ethnic group]” (2), “more [ethnic group] than Ugandan” (3), and “only [ethnic group]” (4).
Here, higher values on the generated variable indicate increasing embrace of ethnic identity, and
decreasing embrace of Ugandan identity.
36 Here, the mean number of correctly identified candidates (out of four) was 2.1 in the control, and 2.3 and 2.2 in photograph treatments that excluded and included, respectively, party identifiers. An ordered logit regression yields insignificant effects (Appendix D, column D)
22
The ordered logit results demonstrate that subjects in the photograph treatment groups were
significantly more likely to weight their tribal identity as more important than their Ugandan
identify (p=.04), suggesting that ballot design here might indeed have primed subjects to
consider their ethnic identity when “voting.”37 These results suggest that the increase in ethnic
voting associated with the inclusion of photographs on ballots is more likely attributable to a
priming mechanism, rather than an informational one.
We want to caution readers against generalizing too broadly from these results. Pictures
might not always lead to ethnic priming, and, under some conditions, they may provide ethnic
information, even though we did not find evidence of an information effect in our experiment.
Nonetheless, we want to stress that the information mechanism is likely to work under far more
restrictive conditions than is generally thought. Ethnic cues can affect voting through the
information mechanism only where the cues convey ethnic information, ethnicity is relevant, and
there is no ceiling effect on knowledge. In short, voters must care about ethnicity but not yet
know the ethnicity of candidates. While the information mechanism is possible under a limited
set of circumstances, we expect that the priming mechanism is more likely to explain ethnic
voting than has been previously thought, and should therefore receive greater scholarly attention.
Conclusion
Ethnic voting is potentially problematic for democratic accountability because it reduces
the likelihood that poor-performing and corrupt politicians will be voted out of office. Further, it
can generate instability and violence by exacerbating inter-group competition and animosity.
Conventional wisdom holds that ethnicity is a key determinant of voting behavior in much of the
37 We also measured perceptions of ethnic inequality or discrimination, but neither was significantly related to treatment assignment in regression analysis. See Appendix D for discussion of analyses and results (columns F, G, and H).
23
developing world and that many countries’ political woes are due, at least partially, to the
politicization of ethnicity. In order to understand the link between elections and governance, we
need to better understand the processes generating ethnic voting.
Current literature on voting in developing countries tends to contrast low-information
ethnic voting with high-information performance-based voting. Scholars posit that instrumental
voters use ethnicity as a proxy for candidate performance when they lack credible information
about candidate preferences and actions. These scholars assume that candidate ethnicity is both
known by and relevant to otherwise poorly informed voters, and they conclude that ethnic voting
is widespread in developing countries because of weak information infrastructures. They also
conclude that individuals will eschew ethnic voting in favor of performance-based judgment
criteria when they are privy to such information.
We extend upon the informational theory of ethnic voting by theorizing about how subtle
stimuli proximate to voting decisions can affect ethnic voting. Knowledge of candidate ethnicity
might not be universal, and other heuristics might be more salient than ethnicity, even in low-
information environments. We posit that ethnic cues can have a large effect on ethnic-voting
rates if they provide information about candidate ethnicity and/or prime ethnic considerations.
While racial priming has been a staple of research on voting in the United States, it has not been
a topic of consideration in most research on ethnic voting in developing countries, largely
because scholars have tended to assume that ethnicity is automatically the most salient criteria,
and thus the default condition when policy information is lacking. In contrast, scholars of the
United States have tended to assume priming effects even when information effects are equally
consistent with the evidence (Lenz 2009). Few studies are designed to adjudicate between the
two mechanisms.
24
Here, we test whether subtle ethnic cues--candidate photographs on election ballots--
affect ethnic voting and examine whether results suggest an informational or priming
mechanism. Evidence from our ballot experiment in Uganda supports our hypothesis that ethnic
cues can significantly alter rates of ethnic voting. Our subjects who received mock ballots with
candidate photographs were 61.6 percent more likely to support a co-ethnic candidate than those
whose ballots did not include them. We find no evidence to suggest that these increases in ethnic
voting can be attributed to learning, but we do find evidence in support of a priming mechanism.
Notably, we provide more convincing evidence of priming than most research on the topic, even
though we examined cue effects in a context where existing scholarship would lead us to expect
an informational mechanism.
Our evidence also has important implications for policy makers. The current
instrumental view of ethnic voting has generated a strong policy recommendation in favor of
educating citizens about candidate performance; if voters employ ethnic heuristics because they
lack knowledge about how candidates will behave once in office, then the solution is to educate
citizens about past performance and policy preferences of candidates. Our research suggests that
this approach is neither sufficient nor efficient. Even if citizens are aware of candidate
performance, they might still vote ethnically if primed to do so. A successful civic education
program must both inform citizens about candidate performance and policies and ensure that this
information is at the top of the mind when voting decisions are made (often weeks if not months
after the educational intervention). Given that election campaigns are often infused with ethnic
primes, 38 citizens’ knowledge of performance criteria may still be overwhelmed by ethnic
considerations by the time they go to vote.
38 For example, Eifert, et al. (2010) found that the salience of ethnicity increased among Africans as elections drew near.
25
Furthermore, our research suggests that simple priming campaigns may be more efficient
means to the same end. For example, civically minded posters placed near a voting booth can
remind citizens about alternative judgment criteria, thereby yielding equal or greater declines in
ethnic voting, at a fraction of the cost of more-intensive campaigns. Or, more directly related to
our research, leaving candidate photos off of ballots would both reduce printing costs and
potentially decrease ethnic voting, while including other elements on ballots—such as partisan
identifiers39 or national flags—could prime alternate considerations.40
In sum, this article offers an extension to the informational theory of ethnic voting by
illuminating the varying effects of ethnic cues. The experimental results show that ethnic cues
generated significantly more ethnic voting even though knowledge of candidate ethnicity (or
candidate performance, for that matter) was not affected. Instead, the evidence suggests that
ethnic cues increased ethnic voting by priming ethnic considerations. We argue that the role of
information in driving ethnic voting has been overstated, and scholars and policy makers should
be more attentive to information-processing biases when theorizing about the causes of ethnic
voting.
39 In another paper on the same experiment, we find that including partisan identifiers on ballots increased the likelihood that individuals voted for major parties and cast straight-ticket votes ([working paper by authors]). 40 Of course, the inclusion of photographs on ballots could have effects not tested here, including increasing turnout and improvements in voters’ abilities to mark ballots correctly and autonomously. However, as noted earlier, we find no evidence that subjects were more likely to mark ballots when photographs were present.
26
Table 1: Treatment Conditions by Ballot Elements
No Party Names Party Names
No Party Symbols
Party Symbols
No Party Symbols
Party Symbols
No photographs 1 2 3 Photographs 4 5
Notes: Empty cells represent potential treatment conditions that were not included in the experiment. Shaded cells represent the treatment conditions that are included in this paper.
Table 2: Effects of Photographs on Ethnic Voting and Mechanism Tests H1
Ethnic voting H2
Informational mechanism
H3 Priming mechanism
Photographs .48 (.25)
p=.06 .04 (.40)
p=.92 .50 (.25)
p=.04
Party Cues – .51 (.26)
p=.05 .47 (.41)
p=.25 – .34 (.25)
p=.17 Constant 7.82 (.33) Cut Points – .79 (.22) – 1.70 (.25) – .98 (.23) – 1.34 (.23) 1.23 (.23) 2.33 (.27) N 217 259 245 Model ologit ols ologit Notes: Cell entries represent coefficient estimates followed by standard errors in parentheses and p-values below. P-values are for two-tailed tests. Outcome variables are votes for co-ethnic candidates (H1), correct coding of candidate ethnicity (H2), and importance of ethnic vis-à-vis national identity (H3).
27
Figure 1: Maps of Research Sites
28
Appendix A: Experimental Ballots for MP Race and Official Ballot for Presidential Race
Samuel Anyolo
Simon Peter Ebitu
Vincent I. Enomu
John Lule
Engirot Lawrence Okae
Raphael Okoropot
Peter Omolo
William Opit
Jimmy Oriokot
Samuel Anyolo
Simon Peter Ebitu
Vincent I. Enomu
John Lule
Engirot Lawrence Okae
Raphael Okoropot
Peter Omolo
William Opit
Jimmy Oriokot
SAMPLE BALLOT 1 SAMPLE BALLOT 3
29
Samuel Anyolo
Independent
Simon Peter Ebitu
Independent
Vincent I. Enomu
National Resistance Movement (NRM)
John Lule
Independent
Engirot Lawrence Okae
Uganda People’s Congress (UPC)
Raphael Okoropot
Democratic Party (DP)
Peter Omolo
Forum for Democratic Change (FDC)
William Opit
Independent
Jimmy Oriokot
People’s Progressive Party (PPP)
Samuel Anyolo
Independent
Simon Peter Ebitu
Independent
Vincent I. Enomu
National Resistance Movement (NRM)
John Lule
Independent
Engirot Lawrence Okae
Uganda People’s Congress (UPC)
Raphael Okoropot
Democratic Party (DP)
Peter Omolo
Forum for Democratic Change (FDC)
William Opit
Independent
Jimmy Oriokot
People’s Progressive Party (PPP)
SAMPLE BALLOT 4 SAMPLE BALLOT 5
30
Appendix B: English-Language Question Wordings
Assessment of Candidate Ethnicity What would you say is the tribe of this candidate? [Options not read] How certain are you about this tribal identity? Are you very sure, somewhat sure, somewhat unsure, or not sure at all? Respondent Ethnic Identity What is your tribal identity? [Options not read] Let’s suppose that you had to choose between being a Ugandan and being a [respondent’s tribal group]. Which of the following statements best expresses your feelings? --I feel only Ugandan --I feel more Ugandan than [respondent’s ethnic group] --I feel equally Ugandan and [respondent’s ethnic group] --I feel more [respondent’s ethnic group] than Ugandan --I feel only [respondent’s ethnic group] Perceptions of Ethnic Status and Discrimination Now, think about the condition of [respondent’s ethnic group]. Do they have less, the same as, or more political influence than other tribal groups here in Soroti County? [If less or more]: Are they a lot [less/more] or only somewhat [less/more]? How often are [respondent’s ethnic group] treated unfairly by the government? All the time, only sometimes, or never?
31
Appendix C: Descriptive Statistics for Dependent Variables Variable Mean Std. Dev. Min. Max.
Votes for actual co-ethnics 1.05 .75 0 2
Votes for perceived co-ethnics 1.02 .73 0 2
Candidates’ ethnicities correctly coded 7.95 3.21 0 15
Supported candidates’ ethnicities correctly coded 1.31 .73 0 2
Major party candidates’ ethnicities correctly coded 2.27 1.14 0 4
Confidence in coding of candidate ethnicities 28.34 10.54 0 45
Number of marked ballots 1.75 .64 0 2
Ethnic over national identity 2.02 1.01 0 4
Perceptions of ethnic advantage/disadvantage 2.20 1.19 0 4
Perceptions of ethnic-based differences in advantages .93 .76 0 2
Perceptions of ethnic discrimination 1.28 .96 0 3
32
Appendix D: Additional Tests A: Number of marked ballots Dependent variable measures the number of ballots (for the MP and District Chair races) on which the subject registered a preference. Null results here do not appear to be due to a ceiling effect, as the mean number of marked ballots in the control group is 1.7 (out of possible 2). B: Vote for perceived co-ethnic Subjects were asked to identify the ethnic identity of each candidate. Only MP and District Chair candidates are included in this analysis. A dependent variable was constructed measuring the number of times that the subject marked a ballot indicating a preference for a candidate that he/she later coded as a co-ethnic. Null results here do not appear to be due to a ceiling effect, as the mean number of supported perceived co-ethnics in the control group is 1.1 (out of possible 2). C: Knowledge of supported candidates Dependent variable measures the number of candidates (for the MP and District Chair races) for whom the subject “voted” later correctly identified by the subject in regards to ethnic identity. Null results here not likely due to ceiling effect, as the mean number of correctly coded supported candidates in the control group is 1.3 (out of possible 2). D: Knowledge of major party candidates Dependent variable measures the number of candidates (for the MP and District Chair races) from major parties (i.e., the NRM and FDC) correctly identified by the subject in regards to ethnic identity. Null results here not likely due to ceiling effect, as the mean number of correctly coded major party candidates in the control group is 2.1 (out of possible 4). E: Confidence in ethnic coding After each determination of candidate ethnicity, the subject was asked to state how confident he or she was about the assessment: “very sure” (coded as 3), “somewhat sure” (2), “somewhat unsure” (1), or “not at all sure” (0). Responses of “don’t know” were also coded as 0. These scores were then summed, generating an overall confidence measure ranging from 0 (not at all sure about any of the fifteen candidates) to 45. Null results here do not appear to be due to ceiling effects, as the mean confidence level of subjects in the control group is 28.3. F: Perceptions of ethnic advantage Subjects indicated whether they believed that their ethnic group had more, less, or the same amount of political influence as other groups in Uganda. Responses were coded on a five-point scale, with 0 representing “much less,” 1 “somewhat less,” 2 “the same,” 3 “somewhat more,” and 4 “much more” perceived influence. Null results not likely due to ceiling effects, as the mean response of subjects in the control is 2.1. G: Perceptions of ethnic-based differences in advantage Variable constructed to indicate the level of difference the subject sees between his or her group’s influence, versus that of other ethnic groups. Responses indicating no perceived
33
difference were coded as 0, while those indicating “somewhat more/less” were coded as 1, and those indicating “much more/less” as 2. Null results here not likely due to ceiling effects, as mean response in the control group was .9. H: Perceptions of ethnic discrimination Subjects were asked how often they felt that their ethnic group was treated “unfairly” by the government. Responses range from “never” (0) and “only sometimes” (1) to “often” (2) and “all the time” (3). Null results do not seem to be due to ceiling effects, as mean response in control group was 1.2.
34
Notes: Cell entries represent coefficient estimates followed by standard errors in parentheses and p-values below. P-values are for two-tailed tests.
A Number of
marked ballots
B Vote for
perceived co-ethnic candidates
C Knowledge
of supported candidates
D Knowledge
of major party
candidates
E Confidence
in ethnic coding
F Perceptions
of ethnic advantage
G Perceptions of ethnic-
based differences
in advantage
H Perceptions of
ethnic discrimination
Photographs .14 (.35)
p=.70 .01 (.28)
p=.97 -.00 (.25)
p=.99 .19 (.22)
p=.38 1.01 (1.35)
p=.45 .12 (.24)
p=.61 .08 (.25)
p=.75 -.03 (.25)
p=.90
Party Cues .61 (.38)
p=.11 – .19 (.28)
p=.50 .09 (.26)
p=.72 .26 (.23)
p=.26 1.41 (1.37)
p=.30 – .12 (.24)
p=.61 .33 (.26)
p=.20 – .09 (.25)
p=.71
Constant 27.39 (1.10)
p=.00
Cut Points -1.96 (.29) – 1.17 (.25) -1.56 (.25) -2.46 (.28) –2.27 (.27) – .80 (.21) – 1.41 (.23) -1.45 (.27) .74 (.24) .26 (.22) -.79 (.20) – .58 (.20) – 1.20 (.22) .48 (.22) .44 (.19) .57 (.20) 1.83 (.26) 2.03 (.23) 1.59 (.23) N 259 179 220 259 259 227 227 220 Model ologit ologit ologit ologit ols ologit ologit ologit
35
Works Cited Adida, Claire L. 2012. “Do African Voters Favor Co-Ethnics? A New Identification Strategy.”
http://ssrn.com/abstract=2054853 Ansolabehere, Stephen, and Charles Stewart III. 2005. “Residual Votes Attributable to
Technology.” Journal of Politics 67(2): 365-89. Antonakis, John, and Olaf Dalgas. 2009. “Predicting Elections: Child’s Play!” Science
323(5918): 1183. Bailenson, Jeremy N., Shanto Iyengar, Nick Yee, and Nathan A. Collins. 2008. “Facial
Similarity Between Voters and Candidates Causes Influence.” Public Opinion Quarterly 72(5): 935-61.
Banducci, Susan A., Jeffrey A. Karp, Michael Thrasher, and Colin Rallings. 2008. “Ballot
Photographs as Cues in Low-Information Elections.” Political Psychology 29(6): 903-17. Banerjee, Abhijit V., Selvan Kumar, Rohini Pande, and Felix Su. 2011. “Do Informed Voters
Make Better Choices? Experimental Evidence From Urban India.” Working Paper. Bates, Robert H. 1974. “Ethnic Competition and Modernization in Contemporary Africa.”
Comparative Political Studies 6(4): 457-84. Berger, Jonah, Marc Meredith, and S. Christian Wheeler 2008. “Contextual Priming: Where
People Vote Affects How They Vote.” PNAS: Proceedings of the National Academy of Sciences of the United States of America 105(26): 8846-9.
Birnir, Jóhanna Kristín. 2007. Ethnicity and Electoral Politics. New York: Cambridge
University Press. Boudreau, Cheryl. 2009. “Closing the Gap: When Do Cues Eliminate Differences Between
Sophisticated and Unsophisticated Citizens?” Journal of Politics 71(3): 964-76. Bratton, Michael, Ravi Bhavnani, and Tse-Hsin Chen. 2011. “Voting Intentions in Africa:
Ethnic, Economic, or Partisan?” Afrobarometer Working Paper, No. 127. Buckley, Fiona, Neil Collins, and Theresa Reidy. 2007. “Ballot Paper Photographs and Low-
Information Elections in Ireland.” Politics 27(3): 174-81. Casey, Katherine. 2013. “Crossing Party Lines: The Effects of Information on Redistributive
Politics.” Working Paper. Chaiken, Shelly. 1980. “Heuristic Versus Systematic Information Processing and the Use of
Source Versus Message Cues in Persuasion.” Journal of Personality and Social Psychology 39(5): 752-66.
36
Chandra, Kanchan. 2004. Why Ethnic Parties Succeed: Patronage and Ethnic Headcounts in
India. New York: Cambridge University Press. Cheeseman, Nicholas, and Robert Ford. 2007. “Ethnicity as a Political Cleavage.”
Afrobarometer Working Paper, No. 83. Conroy-Krutz, Jeffrey. 2013. “Information and Ethnic Politics in Africa.” British Journal of
Political Science 43(2): 345-73. Conroy-Krutz, Jeffrey, and Carolyn Logan. 2013. “Museveni and the 2011 Ugandan Election:
Did the Money Matter?” Journal of Modern African Studies 50(4): 625-55. Dumitrescu, Delia. 2010. “Know Me, Love Me, Fear Me: The Anatomy of Candidate Poster
Designs in the 2007 French Legislative Elections.” Political Communication 27(1): 20-43. Dunning, Thad, and Lauren Harrison. 2010. “Cross-Cutting Cleavages and Ethnic Voting: An
Experimental Study of Cousinage in Mali.” American Political Science Review 104(1): 21-39.
Eifert, Ben, Edward Miguel, and Daniel N. Posner. 2010. “Political Competition and Ethnic
Identification in Africa.” American Journal of Political Science 54(2): 494-510. Ferree, Karen E. 2011. Framing the Race in South Africa: The Political Origins of Racial
Census Elections. Cambridge: Cambridge University Press. Fershtman, Chaim, and Uri Gneezy. 2001. “Discrimination in a Segmented Society: An
Experimental Approach.” Quarterly Journal of Economics 116(1): 351-77. Gilley, Bruce. 2009. The Right to Rule: How States Win and Lose Legitimacy. New York:
Columbia University Press. Green, Donald P., Shang E. Ha, and John G. Bullock. 2010. “Enough Already about ‘Black
Box’ Experiments: Studying Mediation Is More Difficult than Most Scholars Suppose.” The Annals of the American Academy of Political and Social Science 628(1): 200-8.
Habyarimana, James, Macartan Humphreys, Daniel N. Posner, and Jeremy M. Weinstein. 2009.
Coethnicity: Diversity and the Dilemmas of Collective Action. New York: Russell Sage Foundation.
Harris, Adam S., and Michael G. Findley. Forthcoming. “Is Ethnicity Identifiable? Lessons
From an Experiment in South Africa.” Journal of Conflict Resolution. Herrnson, Paul S., Michael J. Hanmer, and Richard G. Niemi. 2012. “The Impact of Ballot
Type on Voter Errors.” American Journal of Political Science 56(3): 716-30.
37
Herron, Michael C., and Jasjeet S. Sekhon. 2003. “Overvoting and Representation: An Examination of Overvoted Presidential Ballots in Broward and Miami-Dade Counties.” Electoral Studies 22(1): 21-47.
Herron, Michael C., and Jonathan Wand. 2007. “Assessing Partisan Bias in Voting Technology:
The Case of the 2004 New Hampshire Recount.” Electoral Studies 26(2): 247-61. Hoffman, Barak D., and James D. Long. 2013. “Parties, Ethnicity, and Voting in African
Elections.” Comparative Politics 45(2): 127-46. Huber, John D. 2012. “Measuring Ethnic Voting: Do Proportional Electoral Laws Politicize
Ethnicity?” American Journal of Political Science 56(4): 986-1001. Ichino, Nahomi, and Noah L. Nathan. 2013. “Crossing the Line: Local Ethnic Geography and
Voting in Ghana.” American Political Science Review 107(2): 344-61. Imai, Kosuke, Luke Keele, and Teppei Yamamoto. 2010. “Identification, Inference, and
Sensitivity Analysis for Causal Mediation Effects” Statistical Science 25(1): 51-71. Isaacs, Harold. 1975. “Idols of the Tribe.” In Ethnicity: Theory and Experience, eds. Nathan
Glazer and Daniel P. Moynihan. Cambridge: Harvard University Press. Johns, Robert, and Mark Shephard. 2011. “Facing the Voters: The Potential Impact of Ballot
Paper Photographs in British Elections.” Political Studies 59(3): 636-58. Kam, Cindy. 2005. “Who Toes the Party Line? Cues, Values, and Individual Differences.”
Political Behavior 27(2): 163-82. Kam, Cindy, Jennifer Wilking, and Elizbaeth Zechmeister. 2007. “Beyond the ‘Narrow Data
Base’: Another Convenience Sample for Experimental Research.” Political Behavior 29(4): 415-40.
Kasfir, Nelson. 1976. The Shrinking Political Arena: Participation and Ethnicity in African
Politics, with a Case Study of Uganda. Berkeley: University of California Press. Katz, Gabriel, R. Michael Alvarez, Ernesto Calvo, Marcelo Escolar, and Julia Pomares. 2011.
“Assessing the Impact of Alternative Voting Technologies on Multi-Party Elections: Design Features, Heuristic Processing and Voter Choice.” Political Behavior 33(2): 247-70.
Keefer, Philip. 2007. “Clientelism, Credibility, and the Policy Choices of Young Democracies.”
American Journal of Political Science 51(4): 804-21. King, Gary. 1997. A Solution to the Ecological Inference Problem: Reconstructing Individual
Behavior from Aggregate Data. Princeton: Princeton University Press.
38
Krosnick, Jon A., and Donald R. Kinder. 1990. “Altering the Foundations of Support for the President Through Priming.” American Political Science Review 84(2): 497-512.
Lawson, Chappell, Gabriel S. Lenz, Andy Baker, and Michael Myers. 2010. “Looking Like a
Winner: Candidate Appearance and Electoral Success in New Democracies.” World Politics 62(4): 561-93.
Lau, Richard R., David J. Andersen, and David P. Redlawsk. 2008. “An Exploration of Correct
Voting in Recent U.S. Presidential Elections.” American Journal of Political Science 52(2): 395-411.
Lenz, Gabriel S. 2009. “Learning and Opinion Change, Not Priming: Reconsidering the
Priming Hypothesis.” American Journal of Political Science 53(4): 821-37. Lenz, Gabriel S., and Chappell Lawson. 2011. “Looking the Part: Television Leads Less
Informed Citizens to Vote Based on Candidates’ Appearance.” American Journal of Political Science 55(3): 574-89.
Lindberg, Staffan I., and Minion K. C. Morrison. 2008. “Are African Voters Really Ethnic or
Clientelistic? Survey Evidence From Ghana.” Political Science Quarterly 123(1): 95-122. Miguel, Edward. 2004. “Tribe or Nation? Nation Building and Public Goods in Kenya versus
Tanzania” World Politics 56(3): 327-62. Mondak, Jeffery J. 1993. “Public Opinion and Heuristic Processing of Source Cues.” Political
Behavior 15(2): 167-92. Museveni, Yoweri K. 1997. Sowing the Mustard Seed: The Struggle for Freedom and
Democracy in Uganda. London: Macmillan. Pande, Rohini. 2011. “Can Informed Voters Enforce Better Governance? Experiments in Low-
Income Democracies.” Annual Review of Economics 3(1): 215-37. Petty, Richard E., and John T. Cacioppo. 1986. Communication and Persuasion: Central and
Peripheral Routes to Attitude Change. New York: Springer-Verlag. Popkin, Samuel L. 1994. The Reasoning Voter: Communication and Persuasion in Presidential
Campaigns. Chicago: University of Chicago Press. Posner, Daniel N. 2004a. “The Political Salience of Cultural Difference: Why Chewas and
Tumbukas are Allies in Zambia and Adversaries in Malawi.” American Political Science Review 98(4): 529-45.
Posner, Daniel N. 2004b. “Measuring Ethnic Fractionalization in Africa.” American Journal of
Political Science 48(4): 849-63.
39
Posner, Daniel N. 2005. Institutions and Ethnic Politics in Africa. New York: Cambridge University Press.
Posner, Daniel N., and David J. Simon. 2002. “Economic Conditions and Incumbent Support in
Africa’s New Democracies: Evidence From Zambia.” Comparative Political Studies 35(2): 313-36.
Rahn, Wendy M. 1993. “The Role of Partisan Stereotypes in Information Processing About
Political Candidates.” American Journal of Political Science 37(2): 472-96. Reynolds, Andrew, and Marco Steenbergen. 2006. “How the World Votes: The Political
Consequences of Ballot Design, Innovation and Manipulation.” Electoral Studies 25(3): 570-98.
Rokkan, Stein 1970. Citizens, Elections, Parties: Approaches to the Comparative Study of the
Process of Development. Oslo: Universitetsforlaget. Rubongoya, Joshua B. 2007. Regime Hegemony in Museveni’s Uganda: Pax Musevenica. New
York: Palgrave Macmillan. Sigelman, Lee, Carol K. Sigelman, and Christopher Fowler. 1987. “A Bird of a Different
Feather? An Experimental Investigation of Physical Attractiveness and the Electability of Female Candidates.” Social Psychology Quarterly 50(1): 32-43.
Smith, Benjamin, Sharon Laskowski, and Svetlana Lowry. 2009. “Implications of Graphics on
Usability and Accessibility for the Voter.” In E-Voting and Identity, eds. P. Ryan and B. Schoenmakers. Berlin and Heidelberg: Springer, 54-74.
Todorov, Alexander, Anesu N. Mandisodza, Amir Goren, and Crystal C. Hall. 2005. “Inferences
of Competence from Faces Predict Election Outcomes.” Science 308(5278): 1623-6. Tripp, Aili Mari. 2011. Museveni’s Uganda: Paradoxes of Power in a Hybrid Regime.
Boulder: Lynne Rienner Publishers. Tsfati, Yariv, Dana Markowitz Elfassi, and Israel Waismel-Manor. 2010. “Exploring the
Association Between Israeli Legislators’ Physical Attractiveness and Their Television News Coverage.” The International Journal of Press/Politics 15(2): 175-92.
Wand, Jonathan N., Kenneth W. Shotts, Jasjeet S. Sekhon, Walter R. Mebane, Michael C.
Herron, and Henry E. Brady. 2001. “The Butterfly Did it: The Aberrant Vote for Buchanan in Palm Beach County, Florida.” American Political Science Review 95(4): 793-810.