INTRODUCTION
Ethnicity is a pervasive theme in African politics, associated with a wide range of political outcomes such as access to education and healthcare, taxation, and political instability.Footnote 1 The notion that African voters vote ethnically is not only a common storyline in journalistic accounts of the region; it emerges as an empirical pattern in a wide range of African elections. In Kenya’s contested 2007 election, President Kibaki won 46% of the national vote against his rival’s—Raila Odinga—44%. But as an ethnic Kikuyu, Kibaki won 97% of the vote in the Kikuyu-dominated Central province; and Odinga, an ethnic Luo, won 82% of the vote in the Luo-dominated Nyanza province. In 2011, Uganda’s Yoweri Museveni—an ethnic Banyankole from the West—obtained 68% of the national vote but an average 82% of the vote in the country’s Western region.Footnote 2 And in Ghana’s 2008 election, where a mere 40,000 votes determined the outcome in the run-off, the victor—late President John Atta Mills—won 50% of the national vote but 73% of the vote in his ethnic home-constituency of Mfantsiman East.Footnote 3 In these examples, the association between ethnicity and voting is unambiguous.
But does ethnicity motivate the African voter? On one hand, scholars describe African elections as ethnic head counts (Horowitz, Reference Horowitz1985) or emphasize the ethnic factor in African electoral contests (Eifert et al., Reference Eifert2010; Ndegwa, Reference Ndegwa1997). On the other hand, scholars attempting to tease out ethnicity from typical confounds paint a more nuanced picture of the African voter’s decision, highlighting performance evaluation and policy preference instead (e.g., Bratton et al., Reference Bratton2011; Bratton and Kimenyi, Reference Bratton and Kimenyi2008; Kimenyi and Romero, Reference Kimenyi and Gutierrez Romero2008; Lindberg and Morrison, Reference Lindberg and Morrison2008; Posner and Simon, Reference Posner and Simon2002). But the fact of the matter is that these observational studies can only speak to the correlation between ethnicity and support; they cannot demonstrate a causal effect. Indeed, key confounds may underlie the relationship instead. In Africa, for example, ethnicity and region highly overlap (Bates, Reference Bates, Rothchild and Olunsorola1983), such that leaders who have been found to reward their coethnics may instead be rewarding certain regions. In some cases, ethnicity and religion are also almost perfectly correlated, making it impossible to attribute causal effects to one or the other.Footnote 4
If we want to say something about the causal effect of ethnicity, we need to be able to randomly manipulate ethnic identity, or at least perceptions of ethnicity. In an observational setting, this is extremely difficult, if not impossible, to do. Ethnicity is a sticky social category, one that is defined as descent-based and thus into which individuals are usually born (e.g., Chandra, Reference Chandra2006). As Sen and Wasow aptly describe the issue with regard to race, with ethnicity, everything is post-treatment (Sen and Wasow, Reference Sen and Wasow2014). As a result, most recent work attempting to identify the independent effect of ethnicity on voter preferences in Africa has turned to hypothetical candidates: in such a context, ethnicity can be randomly assigned and isolated from potential confounds. Relying on this approach, for example, Dunning and Harrison (Reference Dunning and Harrison2011) show that, in Mali, a cross-cutting cleavage called cousinage reduces the extent of ethnic voting.
Building on these efforts, this paper introduces a survey experiment that allows us to draw causal inferences about the effectiveness of coethnic cues in boosting support for a real-world political actor. Scholars have argued that in information-poor contexts, candidates, parties, and voters can rely on ethnic cues as signals (Chandra, Reference Chandra2004; Conroy-Krutz, Reference Conroy-Krutz2013; Ferree, Reference Ferree2006). This paper proposes a survey experiment to assess the extent to which real world political actors can benefit from playing the ethnic card.
This empirical strategy is put to a test in Benin, a small democratic country in West Africa whose President since 2006, Thomas Boni Yayi, is of mixed ethnic heritage. This December 2011 survey experiment manipulates the ethnic cues associated with President Yayi, ethnic cues that reflect his actual ethnic identities: Nago from his paternal line and Bariba from his maternal line. Results confirm that coethnic cueing can move African voters’ expressed preferences. Additionally, a manipulation check gauges the intended effects of ethnic cueing. Results of this second survey experiment, conducted in August 2012, confirm the effectiveness of the treatment, and shed some light as to whether the coethnic effects are due to priming or learning (Lenz, Reference Lenz2009).
The empirical strategy introduced in this paper makes two contributions. First, it builds on experimental work that attempts to isolate the effect of ethnicity per se on voter preferences in an African democracy (Dunning and Harrison, Reference Dunning and Harrison2011). Second, it investigates the power of coethnicity in moving respondents’ political attitudes toward a real-world political actor (Wantchekon, Reference Wantchekon2003). Furthermore, by relying on a single political actor with two ethnic identities, it tests a hard case for gauging the effectiveness of coethnic cues by asking whether the same political actor can draw support from two different ethnic groups based solely on subtle ethnic cues. If the experimental results show that the same, real-world political actor can boost support from coethnics A by cueing ethnicity A, and from coethnics B by cueing ethnicity B, it will have unveiled just how powerful coethnicity can be in shaping political attitudes.
PRESIDENT YAYI OF BENIN: A MIXED-ETHNICITY PRESIDENT
Benin is a small West African country with a population of approximately 10 million. It gained its independence from France in 1960, and subsequently experienced 30 years of military coups and authoritarian rule. But in the 1990s, following a wave of democratization in sub-Saharan Africa, Benin transitioned peacefully to multiparty democracy.Footnote 5 On April 6, 2006, the Beninois elected Thomas Boni Yayi and in March 2011, reelected him for a second and final term. In both elections, candidate Yayi enjoyed high levels of support, winning the 2006 election with approximately 75% of the vote in the run-off, and the 2011 election with a majority vote in the first-round (see Figures SM-1 and SM-2 in the Supplementary Material).Footnote 6
The main cleavage in Benin politics has historically been defined as a regional one: since independence from France in 1960, Southeast, South–Center, and North have jockeyed for political power. Indeed, due largely to historical factors, these three regions delineated themselves in the wake of independence: the Southeast (the ancient kingdom of Porto-Novo), led by Sourou Migan Apithy and home of the Yoruba/Nago ethnic group; the South-Center (the ancient kingdom of Danhomé), led by Justin Ahomadegbé and home of the plurality Fon ethnic group; and the North (the ancient kingdom of Bariba), led by Hubert Maga and home of the Bariba ethnic group (Cornevin, Reference Cornevin1981; Decalo, Reference Decalo1976; Loko, Reference Loko2007). During the first 12 years of independence, this triumvirate struggled for power in coups and counter-coups. And as the most populous ethnic groups in each of these regions, respectively, the Yoruba, the Fon, and the Bariba make up some of Benin’s most politically relevant ethnic groups.
Ethnic groups come with rules of descent. President Yayi is the son of a Nago (YorubaFootnote 7 ) father and a Bariba mother. Nago and Bariba are patrilineal groups.Footnote 8 President Yayi, therefore, is theoretically a Nago. Yet empirically, President Yayi has leveraged his mixed ethnic roots effectively. When he was first running for President, candidate Yayi played up his Bariba, and more generally, his Northern connections. For example, Yayi actively sought out support from famous Bariba players in the field, such as Charles Toko, the owner of the newspaper Le Matinal (Loko, Reference Loko2007). This strategy seems to have been effective. President Yayi has been described in the popular media as Nago (BBC News, 2006), from the North (Freedom House, 2011), and multiethnic (Africa News, 2006). And a proportional sample of Beninois respondents, when asked in an open-ended survey question to what ethnic group President Yayi belongs, does not converge on a single ethnic identity (see Figure SM-3 in the Supplementary Material).
RESEARCH DESIGN
In December 2011, a survey with an embedded experiment in Cotonou, the economic capital of Benin, was administered.Footnote 9 This survey oversampled President Yayi's coethnics (see SM1); it collected basic sociodemographic data, presented the respondent with a short prompt about President Yayi’s performance, and gauged respondent support for Yayi. The prompt was identical in all surveys, except for a short phrase qualifying President Yayi with one key background characteristic. This phrase was randomly assigned, following a block-randomization design (Horiuchi et al., Reference Horiuchi2007; Imai et al., Reference Imai2008). A natural blocking category in estimating the effects of ethnic cueing on respondent support in Benin is ethnicity itself. Members of ethnic group A, for example, are likely to react differently to cueing of Yayi’s A identity than are members of any other ethnic group. Furthermore, members of ethnic groups to which President Yayi does not even belong are also likely to react differently from one another. And yet, there are close to 60 different ethno-linguistic groups in Benin (Bio Bigou, Reference Bio Bigou2011). For economy of resources, the block-randomization design in this survey thus follows a regional North/South cleavage.Footnote 10
After a set of pretreatment questions collecting data on the respondent’s basic socio-demographic characteristics, enumerators read a short biographical paragraph about President Yayi to the respondent. This section appeared in three versions. The control made no reference to ethnicity. The Nago prompt began with a reference to President Yayi’s Nago origins. The Bariba prompt began instead with a reference to President Yayi’s Bariba origins.
{∅; Born of a Nago father; Born of a Bariba mother}, Yayi Boni became President of Benin on April 6, 2006 and was just re-elected this year for a second term. He has led a presidential campaign based on economic growth and suppressing corruption. During his first presidential term, he invested in the agricultural sector to promote exports, and tackled corruption. However, some critics claim that the country’s economic growth has been disappointing, and that Boni’s administration is, itself, corrupt.
Immediately following the reading of the prompt, respondents were asked the following: “If there were no term limits and the election were held today, would you vote for Yayi Boni for President?” This question measures support for President Yayi, allowing us to identify the effect of coethnicity with Yayi on support for Yayi.Footnote 11
Table SM-2 in the Supplementary Material presents balance tests across the three treatment types, and shows that—across a wide array of pre-treatment variables—the three types of surveys were balanced. There are two important exceptions, however. First, respondents receiving the Bariba treatment are on average better educated than those receiving the control condition. Second, one enumerator ended up with fewer Bariba surveys than others.Footnote 12 In additional tests, the analysis includes an estimation that controls for pre-treatment confounds—namely education—to address these sources of imbalance (see Table SM-4).
The random assignment of cues about Yayi’s ethnic identities offers an opportunity to estimate the causal effect of ethnic cues on support for a real-world leader. Indeed, since everything else about the prompt remains the same—including the identity of President Yayi—and ethnic cues are randomly assigned, this design isolates the effect of ethnic cues on support for a real-world politician. This is as close as we can get to randomly assigning ethnic identities without relying on hypothetical candidates.
RESULTS
This section shows that coethnic cueing “worked” by boosting support among President Yayi’s coethnics. Table 1 provides the average effects of ethnic cueing on respondent support for Yayi: coethnic effects act as expected, yielding an 18% percentage-point boost in support for Yayi. When we restrict the analysis to the coethnic sample only, further isolating the effect of coethnic cueing rather than that of coethnic respondents, we find that the effect persists and is statistically significant at the 90% confidence level. By contrast, coethnic cueing has no effect on Yayi’s non-coethnic respondents, a result we expect to be true on average.Footnote 13 ,Footnote 14
Table 1 Average Treatment Effects, Difference-of-Means
Notes: The variable Vote for Yayi is a dummy variable that takes the value “1” if the respondent claimed that she would vote for Yayi, and “0” otherwise. Tests of statistical significance are survey-weighted two-tailed t-tests. “Coethnic cue” is a dummy variable that takes the value “1” if the respondent received a coethnic cue and “0” otherwise. The coethnic sample includes only Yoruba and Bariba respondents. The non-coethnic sample includes only non-Yoruba and non-Bariba respondents. President Yayi is a Nago from his paternal line and Bariba from his maternal line. The Nago are a sub-group of the Yoruba and are indigenous to the Southeastern and Central parts of the country. The Bariba are indigenous to the Northern part of the country. *, †, and ‡ indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
A simple cueing exercise that draws attention to coethnicity among the President’s coethnics provides an additional bump in support: although statistical significance drops to the 90% confidence level, due to a two-thirds reduction in sample size, the effect is stable: support for Yayi goes up by approximately 16 percentage-points when respondents are cued to a shared ethnic identity.Footnote 15
WHAT DID ETHNIC CUEING DO?
How do we know that the experimental intervention had its intended effect? In this section, I present results from manipulation checks of the ethnic prompts, and consider the extent to which we can determine whether the effects identified above are due to priming or learning.
Manipulation Checks
To verify that the manipulations worked, a follow-up survey in August 2012 intended to gauge manipulation checks was administered. The survey questionnaire was identical to the 2011 questionnaire in its collection of basic socio-demographic data. It then presented the respondent the same short prompt about President Yayi, with the same three randomly-assigned conditions (No ethnic reference, “Born of a Bariba mother”, “Born of a Nago father”).Footnote 16
This second survey experiment fulfills two objectives. First, it offers a replication of the main analysis on the causal effect of coethnic cues; second, it offers a manipulation check. Replication of the main analysis is provided in Tables SM-8 and SM-9. Both tables indicate a positive effect of coethnic cueing on support for Yayi, though conventional levels of statistical significance are not achieved in estimating the effect of the coethnic cue. This is not surprising: the sampling method for this second survey did not over-sample Yayi’s coethnics. With lower power, the effect is less precisely estimated. Still, its direction is as expected.
Second, this survey offers a manipulation check. After the prompt, one set of surveys immediately asked the respondent about the ethnic identity of President Yayi’s mother and father (the short questionnaire), and another set of surveys asked for that information a few questions down (the long questionnaire). The goal was to measure the extent to which respondents registered the information provided by the treatment immediately after the prompt, and a few questions after the prompt, so as to determine: (1) the presence of any effect at all, and (2) its persistence over the course of a short questionnaire.
Table SM-10 in the Supplementary Material presents the results of manipulation checks for the Bariba and Nago cues relative to the control condition where there is no mention of ethnicity. The manipulation worked: under the control condition, only 46% of respondents were able to say that Yayi’s mother is a Bariba and 66% of respondents were able to say that Yayi’s father is a Nago. This difference probably reflects the fact that these ethnic groups are patrilineal. But when respondents are cued to Yayi’s maternal ethnic roots, they get the message: 89% of them are able to identify Yayi’s mother as a Bariba. And when respondents are cued to Yayi’s paternal ethnic roots, the effect is also there: 92% of respondents are able to identify Yayi’s father as a Nago. The Bariba effect, a 42 percentage-point increase, is large and statistically significant at the 99% confidence level; the Nago effect, a 25 percentage-point increase, is also large and statistically significant at the 99% confidence level. In sum, not only did the manipulation increase the proportion of respondents who are able to identify Yayi’s parental ethnic membership, it made Yayi’s maternal ethnic origins just as salient as his paternal ethnic origins.
Table SM-10 in the Supplementary Material also illustrates the difference in manipulation effects when the manipulation check comes immediately after the prompt, and when it comes a few questions down.Footnote 17 The results indicate that the effect persists over the course of a slightly longer questionnaire: there is no statistically significant difference in the effect, whether the respondent is quizzed immediately after the prompt or a few questions later.
Priming or Learning?
An important literature on cueing experiments urges scholars to distinguish between priming and learning effects (Lenz, Reference Lenz2009). This question applies here as well: when we cue respondents to President Yayi’s Bariba or Nago origins, are we providing them new information, or are we reminding them – in other words, making salient—a piece of information they already knew? Although the methods proposed to distinguish between the two mechanisms are not applicable hereFootnote 18 , the manipulation check from August 2012 allows us to gain some ground on this question.
We can rule out learning for those respondents who, before the cue, already knew President Yayi’s ethnic origins. Indeed, a respondent who is able to identify the ethnic membership of President Yayi’s father under the Control condition is not a respondent who will learn new information from the Nago cue. When we break down the manipulation check by ethnic group, we find precisely this scenario for Bariba respondents on Yayi’s maternal and paternal ethnic heritage, and for Yoruba respondents on Yayi’s paternal ethnic heritage (see Table SM-11 in the Supplementary Material). For Yayi’s Bariba coethnics, the manipulation check showed no significant effect: 100% of Bariba respondents know Yayi’s father is a Nago without any cueing, and 77% know Yayi’s mother is a Bariba without any cueing. Cueing effects are not statistically significant. Similarly, 88% percent of Yoruba respondents know Yayi’s Nago heritage without any cueing, and the cueing effect is trivial.Footnote 19
In sum, among the Bariba, no learning occurred. Similarly, among the Yoruba exposed to the Nago cue, no learning occurred. Our manipulation check thus allows us to rule out a learning effect for three out of the four coethnic cueing scenarios. By contrast, among Yayi’s non-coethnics, ethnic cueing had a significant effect on knowledge of Yayi’s ethnic heritage: we thus cannot rule out a learning effect among Yayi’s non-coethnic respondents.Footnote 20
CONCLUSION
This paper measures the effectiveness of coethnic cues in boosting support for a real-world political actor in a democratic African context. It builds upon the recent literature aimed at using experiments and hypothetical vignettes to identify ethnic effects. In this paper, the experimental approach allows us to isolate the causal effect of coethnic cues; and the reliance on a real-world political actor allows us to speak directly to the extent to which politicians in Africa can play the ethnic card.
The results of this survey experiment in Benin, where President Yayi is of mixed ethnic heritage, provide evidence for coethnic favoritism: a subtle ethnic reference boosts coethnic support for President Yayi by approximately 16 percentage points. Results from a follow-up survey show that the ethnic cues have their intended effects of improving respondents’ knowledge or awareness of President Yayi’s ethnic origins; on top of that, by breaking down the manipulation checks by each ethnic group, we are able to specify, in some cases, whether the ethnic effect is due to learning or priming. In sum, this paper’s findings highlight the power of ethnic cues in moving individuals’ political attitudes toward a real-world politician.
More questions remain. First, although the identification strategy proposed here can successfully isolate the independent effect of coethnic cueing, it does not address why coethnic cues work. Do coethnic cues improve candidate credibility as a provider of patronage or development goods (Ichino and Nathan, Reference Ichino and Nathan2013; Posner, Reference Posner2005)? Do they improve voters’ self-esteem and sense of belonging (Horowitz, Reference Horowitz1985)? Do they provide information about candidate quality (Conroy-Krutz, Reference Conroy-Krutz2013)? Future work can, relying on the same identification strategy, hone in on which mechanisms are doing the work.
Future work should further leverage opportunities offered by survey experiments and the lessons learned from this study, to more precisely isolate the conditions that facilitate ethnic voting and those that attenuate it. For example, the empirical strategy introduced in this paper can be used to explore heterogenous treatment effects: are certain types of respondents more susceptible to ethnic cueing than others? On one hand, scholars have argued that education yields cosmopolitanism, implying that better-educated people are less likely to respond to ethnic cues (Hainmueller and Hiscox, Reference Hainmueller and Hiscox2007); yet others suggest instead that modernization brings about greater attachment to ethnic identity (Eifert et al., Reference Eifert2010). The question is an empirical one and merits further investigation.
In the meantime, the analysis herein has confirmed that, in one of the region’s most successful and stable democracies today, ethnic politics is alive and well; and that, as scholars in search of causal inference, we can leverage mixed identity leaders and experimental methods to test the power of ethnic cues.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/XPS.2014.26.
