Rejecting Ethnic Pandering in Urban Africa: A Survey Experiment on Voter Preferences in Nairobi,Kenya

Introduction

Ethnic mobilization is a routine feature of electoral competition in Africa’s emerging multiparty democracies. A standard assumption in the ethnic politics literature is that candidates bolster their support among coethnic voters by appealing to shared ethnic interests (Horowitz 1985; Posner 2005). Ethnic pandering—that is, promising to cater to the interests of one’s group, above all with respect to the allocation of patronage resources including jobs, scholarships, funds for roads, and other discretionary goods provided by the government—serves as useful strategy for augmenting coethnic support (Nathan 2019; Wantchekon 2003). Yet, ethnic pandering entails risk, particularly in Africa’s diverse urban areas. Promises to favor one group will alienate voters from other communities. Targeted promises may also run afoul with coethnic voters, the intended target for such appeals. Africa’s cities have become considerably more integrated in recent decades: inter-marriage is common (Bandyopadhyay and Green 2021; Crespin-Boucaud 2020), a growing share of mixed-ethnicity citizens trace their lineage to multiple communities (Dulani et al. 2021), and social networks frequently extend across ethnic divisions. Moreover, parochial attachments are weaker in urban areas (Kramon et al. 2021; Robinson 2014). In line with contact theory (Allport 1954), we propose that social integration in urban areas may temper ethnic preferences, leading voters—particularly longer-term residents and those whose social networks are more diverse—to disfavor narrow appeals to ethnic interests that promise to favor one group at the expense of others.

This proposition cuts against the grain of contemporary research. Recent scholarship shows that parties continue to rely on appeals to ethnic interests in urban Africa (Klaus and Paller 2017; Nathan 2019) and that ethnicity remains highly salient to urban voters. In Kenya, the focus of this study, ethnic bloc voting rates in Nairobi, the capital city, are on par with rural settings. However, we caution against drawing inferences about ethnic political preferences solely from observed voting behavior. As noted in prior literature, African voters may find themselves compelled to support coethnic leaders in order to avoid being left out of patronage allocations (Bratton and Kimenyi 2008; Posner 2005). Even those who disapprove of ethnic politics may choose to vote along ethnic lines given the limited “choice set” (Ferree 2020) available in contemporary elections. Thus, learning about voters’ preferences toward ethnic mobilization may require the use of experimental strategies that present voters with different options than those available in actual elections.

To test the proposition that urban voters disfavor certain types of ethnic pandering, we draw on data from a survey experiment conducted in Nairobi, Kenya in 2015. The experiment, implemented as part of a larger household survey, asked respondents to listen to a speech by a hypothetical parliamentary candidate planning to run for office in the next election. The experiment randomly assigned respondents to either an ethnically targeted message or an inclusive message and varied whether the candidate was identified as a coethnic or a non-coethnic. Standard account suggests that voters should disapprove of “mistargeted” ethnic appeals from non-coethnic leaders but should respond positively to pandering by coethnics. Our account suggests that voters should disfavor pandering regardless of whether the candidate is a coethnic.

Results show that support for the hypothetical candidate is lower in the ethnic pandering condition relative to the inclusive message, and that the penalty does not vary by coethnicity with the candidate. Additional tests show also that pandering reduces assessments of the candidate’s suitability for public office. We find that the penalty for ethnic pandering by coethnic leaders is greater among longer-term residents in Nairobi, suggesting that urban voters may come to disfavor ethnic pandering over time after migrating from rural areas. We draw on data from a second replication study, conducted in 2016, to confirm these results after making several modifications to the research design, including the use of a secret ballot to address concerns about social desirability bias. In the replication, the penalization for the coethnic pandering is greater among those who report more frequent interaction with out-group members, lending support to the notion that contact theory may partly explain the observed effects. Finally, to increase confidence in our theoretical account, we examine an alternative explanation, namely, that voters may reject ethnic pandering for strategic reasons, understanding that coethnic leaders will struggle to win election in Nairobi’s diverse constituencies if they couch their electoral bids in narrow ethnic terms. Our data suggests that strategic considerations are not the primary mechanism at work.

This study makes three contributions. First, it adds to the ethnic politics literature by showing that even in settings where ethnic voting is the norm, voters may disapprove of divisive forms of ethnic political mobilization. Second, this study contributes to the growing literature on campaign strategies in Africa, particularly research that employs experimental approaches to study voter preferences (Dunning and Harrison 2010; Fujiwara and Wantchekon 2013; Gutiérrez-Romero and Lebas 2020; Horowitz and Klaus 2020; Kim 2021; Rosenzweig 2021; Wantchekon 2003). While our evidence suggests that explicit promises to favor in-groups are disfavored—at least by some urban residents—we emphatically do not claim that all types of ethnic appeals are disdained. Candidates in Africa’s emerging democracies rely on a wide range of strategies to appeal to ethnic interests, often emphasizing threats to group interests rather than explicit promises of group favoritism (Chege 2008; Horowitz 2021; Posner 2005). Third, it contributes to the literature on instrumental theories of ethnic voting that imply that voters in ethnically diverse societies seek to maximize the benefits obtained by their own group, even if such gains come at the expense of others by demonstrating that there is greater support for inclusive politics among urban voters than often assumed. In doing so, we add to a growing literature that examines continuity and change in the politics of Africa’s rapidly growing cities (e.g., Berge et al. 2020; Nathan 2019; Paller 2019).

Ethnic Mobilization and Social Integration in Urban Africa

The existing literature offers contrasting expectations regarding the effects of urbanization on ethnic political dynamics. Early modernization theory (e.g., Lerner 1958) expected that as societies developed, national and class-based forms of identification and political organization would replace communal identities and political structures. These processes were expected to be most pronounced in urban areas, where access to education and formal-sector employment is higher. Later modernization theory, by contrast, argued the urbanization would intensify ethnic politics because cities, which are often more diverse than rural areas, bring individuals from different groups into sharper competition for access to jobs, education opportunities, markets, and other state-controlled resources (e.g., Bates 1974; 1983). Scholarship in this tradition claimed that urbanization would increase the reliance on ethnic networks and reinforce the salience of ethnic identities. In Nairobi, for example, studies show that competition for access to land, housing, and local markets can be intense (Gugler 2002; Huntington 1977). Conflicts over government-controlled land in the city’s informal settlements and control of local transportation sectors—particularly the matatu (mini-bus) routes—have at time escalated into inter-group violence, exacerbating patterns of ethnic segregation in the city (Amis 1984). Thus, Elfversson and Hoglund (2019, 355) observe that during the communal clashes following Kenya’s 2007 elections, “‘ethnic zones’ were established in the informal settlements, and when looting and arson became widespread people were forced to flee to their respective ethnic enclaves, which they regarded as safe since tribally-based gangs took over control.”

Current research in Africa has yet to settle the debate between the competing visions of urban politics offered in modernization theory. Robinson (2014) shows that ethnic identities are weaker relative to national identities in urban Africa, and Kramon et al. (2021) demonstrate that migration to urban areas in Kenya leads to a decrease in the perceived importance of ethnicity. ¹ Several contributions suggest that urban voters are less inclined to vote along ethnic lines (Conroy-Krutz 2009), hold more fluid electoral preferences (Harding and Michelitch 2019; Wahman and Boone 2018), and organize around class rather than ethnic interests (Resnick 2014). Others, however, show that forms of communal mobilization remain widespread. Klaus and Paller (2017) find that appeals to ethnic interests are commonplace in Accra, Ghana, particularly in areas where groups seen as “first-comers” are locally dominant. Nathan (2019) finds that ethnic voting in Accra remains on par with rural areas in localities that are dominated by a single ethnic group, while bloc voting declines in more heterogenous areas.

While recent literature has advanced understanding of ethnic politics in urban Africa, existing accounts have yet to consider how social integration—the process of forming inter-group connections and solidarities that bridge communal divides—might affect ethnic political preferences. In emphasizing the potential effects of social integration, we draw on a growing body of quantitative scholarship that documents significant changes in inter-ethnic relations in Africa over recent decades, especially in urban areas. First, inter-marriage is increasing. Based on data from Demographic and Health Surveys (DHS), Crespin-Boucaud (2020) shows that one in five marriages in a large sample of African countries now crosses ethnic divisions, documenting a steady increase in inter-marriage rates over recent decades in most countries. Bandyopadhyay and Green (2021) show that inter-marriage rates are nearly twice as high in urban versus rural areas (29% vs. 15%). As a result, there are a growing number of mixed-ethnic citizens—individuals who descend from diverse ethnic lineages—in Africa and especially in major cities (Dulani et al. 2021). Finally, some evidence indicates that social networks in urban areas are substantially more diverse than in the countryside. In an early study of social networks in a medium-sized Ugandan town, Jacobson (1970) finds that individuals in formal-sector jobs tended to have ethnically diverse friendship networks, while networks among “non-elite” residents—those with less education, wealth, or formal-sector employment—were more commonly organized along ethnic lines. However, more recent data from Nairobi, described below, shows inter-ethnic ties are now common in urban areas even among non-elite citizens.

We expect that inter-group ties may affect political preferences by reducing parochial attachments and increasing concern for the welfare of other groups. Contact theory (Allport 1954), which suggests that meaningful inter-group contact can attenuate negative biases toward out-groups, provides a foundation for these expectations. While Alport’s original formulation proposed that a set of “optimal conditions” were required for contact’s moderating effects, subsequent research finds that it can reduce inter-group prejudice in a wide array of contexts (Pettigrew and Tropp 2006; see also Paluck, Green, and Green 2019). Experimental studies indicate the relevance of contact theory to ethnic relations in Africa. Dionne (2015) shows, based on research in Malawi, that social networks supersede ethnic identities in explaining participant behavior in standard cooperation games. Scacco and Warren (2018) find that ethnic discrimination is attenuated by randomly assigning participants in a skills-training course to mixed-religious cohorts in Nigeria. The greater extent of meaningful inter-group contact in cities implies that such effects should predominate among urbanites.

In settings where inter-ethnic ties are commonplace, disdain for divisive forms of ethnic politics may become a generalized norm. Based on ethnographic research in a diverse area on the outskirts of Nairobi, Landau (2015, 59) documents the emergence of “a discursive field” that “works to discipline residents by stigmatizing…‘dangerous’ people and divisive practices.” He observes that the area’s residents “reject the validity of ethnic violence and politics while promoting a kind of universal inclusiveness.” The author attributes the development of inclusive norms to the large number of mixed-ethnic citizens in the area, suggesting the importance of inter-group ties in moderating ethnic preferences. ² The author argues, moreover, that Kenya’s history of inter-group conflict produces a desire to be seen as a “cosmopolitan” person—someone who embraces inter-group harmony and rejects overt manifestations of tribal politics. While these arguments rest on observations from a single settlement area, other research suggests the prevalence of inclusive norms more broadly in Nairobi. Berge et al. (2020) find little evidence of ethnic bias among participants in a set of behavioral games conducted in Nairobi. The authors observe ethnic bias only among recent migrants to the city, suggesting that gradual assimilation into “local norms” in Nairobi may account for the observed difference with Nairobi’s more egalitarian long-term residents.

In this study, we examine how urban voters respond to communal mobilization strategies employed by office-seeking elites. Parties and candidates routinely appeal to shared ethnic interests to shore up support and increase turn-out among coethnic voters in Africa’s emerging multiparty democracies (Ake 1976; Cheeseman and Larmer 2015; Posner 2005). A standard presumption in the literature is that messages that reinforce expectations of targeted patronage benefits serve as an effective strategy (e.g., Posner 2005; Wantchekon 2003). Our approach suggests, however, that urban voters may be averse to communal appeals that are seen as divisive. Thus, candidates who promise to direct benefits toward members of one group at the expense of others may be viewed as divisive and retrograde, relative to those who offer more inclusive distributive strategies. These expectations differ from standard accounts of group pandering, which typically predict that voters will react negatively to “mistargeted” panders—candidates’ promises to cater to the interests of other groups—but will generally react favorably to appeals to their own group’s interests (Hersh and Schaffner 2013). We propose that urban voters may disfavor ethnic pandering regardless of whether the candidate is a coethnic. These expectations cohere with recent research showing that divisive electoral strategies in multiethnic context are often less effective than assumed (e.g., Gutiérrez-Romero and LeBas 2020; Horowitz and Klaus 2020; Rosenzweig 2021).

We adopt an experimental approach to study reactions to ethnic pandering for three reasons. First, due to the association between ethnicity and partisanship, voters may fail to punish real-world leaders who engage in pandering even if they disfavor the practice in principle. Because partisan motivated reasoning (Taber and Lodge 2006) leads voters to rationalize bad behavior by preferred candidates, voters may fail to perceive transgressive conduct by politicians as such or may overlook behaviors they disapprove of when committed by co-partisans (Walter and Redlawsk 2019). Second, it is difficult to isolate reactions to campaign messages with observational approaches because of the bundled nature of campaign communication, particularly in developing countries, where campaign rallies—rather than paid media advertising—are central to campaigning (Paget 2019). Finally, observational studies of voter behavior may fail to detect changing ethnic preferences in urban Africa. Party systems offer voters a limited range of options on election day, and ethnic voting may remain the norm because voters have no option but to support the party or candidate associated with their group. Failure to do so means risking that one’s group may suffer neglect at the hands of leaders who represent the interests of other groups (Bratton and Kimenyi 2008; Posner 2005). Research strategies that abstract from such limited electoral contexts are needed to examine preferences.

We do not expect that all types of communal appeals will be disfavored by urban voters. Existing literature suggests that elites generally refrain from overt favoritism promises that may run counter to basic norms of fairness or that may alienate out-group voters (Posner 2005). Instead, candidates often emphasize threats to communal interests that portray opponents as ethnic chauvinists. Thus, in Kenya, Cheeseman, Kanyinga, and Lynch (2020, 7) observe that “leaders typically achieve effective ethnic politicization by manipulating the worst fears of their supporters in order to increase the perceive risks of allowing a leader from another community to emerge victorious.” Appeals that emphasize marginalization or neglect—actual or feared—may fail to trigger a backlash since these messages impugn an opponent’s divisive behavior rather than promising to provide disproportionate benefits to the candidate’s own group. For this reason, we focus in this research on overt appeals that might plausibly trigger a negative reaction—those that unambiguously promise to favor in-group members over others.

Finally, the approach outlined here implies one source of potential heterogeneity related to the recency of urban migration. Many inhabitants in Africa’s rapidly growing cities are recent transplants from the countryside. We expect that the rejection of ethnic pandering may be stronger among longer-term residents for two reasons. First, as shown below, the formation of inter-group ties increases in the years following arrival in the city. Second, if a distinct set of egalitarian norms exists in urban areas, as Berge et al. (2020) propose, the adoption of such norms ought to be contingent on the length of time one has lived in the city.

Context: Ethnic Politics in Nairobi, Kenya

It is often remarked that politics in Kenya revolves around ethnicity. Thus, a report by the Kenyan Human Rights Commission (2018, 16–17) notes that “political elites have traditionally used ethnic identity to pursue partisan ends to the extent that it is near impossible to discuss Kenyan politics without reference to the notion of ethnic identity.” Several studies document patterns of ethnic favoritism in Kenya since independence (e.g., Burgess et al. 2015; Kramon and Posner 2016; and Jablonski 2014), and a substantial literature provides evidence of ethnic voting since the return to multiparty competition in the early 1990s (e.g., Gutiérrez-Romero 2013; Ferree, Gibson, and Long 2014; Oyugi 1997). The perception that political leaders routinely favor their own communities over others leads many voters to prefer coethnic candidates when faced with a choice between candidates from distinct ethnic backgrounds—both in national races for the presidency and in local-level contests for governors, members of parliament, and councilors. Njogu (2001, 398–99) observes that in Nairobi, “because elections are seen as a platform through which individuals and constituents can be developed, socially and economically disadvantaged groups tend to congregate around individuals best suited to facilitate patronage resources…Powerful individuals drawn from one’s own ethnic group are more often than not viewed as ‘the safest’ patrons.” Various studies demonstrate that access to land, employment, and economic opportunities more broadly are thought to be mediated by ties to communal patrons and their ethnic networks (e.g., Amis 1984; Huntington 1977; Katumanga 2005; Leonard 1991; Nellis 1974; Ross 1975).

Voting patterns indicate the continued salience of ethnicity to Nairobi’s urban voters. Figure 1 shows bloc voting rates—defined as the share of voters in each ethnic group that register an intention to vote for the most-preferred candidate by members of the group overall—in Kenya’s 2007 presidential election for Nairobi relative to the rest of the country. Estimates are based on six national public opinion polls conducted in the three months prior to the election with a total sample size of 19,424 respondents, and include all groups with at least 50 respondents in Nairobi. Additional details are provided in the Supplemental Information (SI). Figure 1 shows that bloc voting rates in Nairobi are on par with the rest of Kenya, which is predominantly rural. Among the Kikuyu, for example, 88% of respondents in Nairobi registered an intention to vote for the group’s most-preferred candidate, Mwai Kibaki, relative to 93% in other areas. Among the Luo, support for the main opposition leader, Raila Odinga, was nearly universal both in Nairobi and elsewhere. While disparities are observed for some groups, the differentials in most cases are small in substantive terms, and for some groups—the Luhya, Kisii, and Somali—bloc voting rates were higher in Nairobi than elsewhere. While comparable data for lower-level elections is not available, these data suggest the enduring importance of ethnicity to voters in Nairobi.

OPEN IN VIEWER

Figure 1.

Bloc Voting Rates by Ethnic Group—2007 Presidential Election. Figure shows the percent of each ethnic group reporting an intention to vote for the most-preferred presidential candidate among members of the group in Kenya’s 2007 election (Mwai Kibaki for the Kikuyu and Meru; Raila Odinga for the Luo, Luhya, Kisii, Kalenjin, and Somali; and Kalonzo Musyoka for the Kamba). Data are from six national surveys conducted between October and December 2007 with a combined sample of 19,424. Additional details are provided in the SI.

Less is known about the ethnic mobilization strategies used by aspirants seeking to represent the city’s voters. Existing scholarship on national-level contests suggests that appeals to ethnic interests have been commonplace in Kenya since the return to multiparty competition. Typically, ethnic appeals are phrased as threats to communal interests, along with promises to represent group interests more effectively. Throup and Hornsby (1998, 349) provide an example from the 1992 campaigns: “In Molo … FORD-Asili’s Njenga Mungai campaigned on a strongly ethnic ticket, arguing that the Kikuyu needed to recapture the Presidency and the government for their tribe. He claimed, for example, that if the GEMA people [a reference to the Kikuyu, Embu, and Meru] did not vote for Matiba, the Presidency would never return to Kikuyuland.” Promises to serve communal interests are often made using implicit language or geographically targeted appeals that are understood as catering to distinct ethnic areas. Nonetheless, explicit references to ethnic interests are not uncommon. The following quote from Ali Mohamud Mohamed (MP for Moyale Constituency), speaking at a 2007 campaign rally in Garissa, provides an example: “The election card that you have in your pockets will determine your future…If we are the Somali community living in Kenya, our interest is in Kibaki…The other men want to get the seat of power in order for them to succeed in their personal interests. Something you have in hand is better than one which is far away from you. Us, we are among the government—we will build up; we benefit from it now and we shall also benefit in the future.” ³

Local elections in Nairobi, however, are distinct from those most other parts of country because ethnic diversity is greater. While disaggregated data on ethnicity from the Kenyan census is not publicly available, survey data provides a useful proxy. Data collected from multiple national opinion polls conducted between 2005 and 2008 show that the size of the largest ethnic group in Nairobi’s eight parliamentary constituencies used until redistricting in 2010 was 36.0%, relative to 78.2% for the other 202 constituencies outside Nairobi (see Horowitz 2019 for details on constituency-level ethnicity estimates). ⁴ Nairobi’s diversity has important implications for local elections. It means that candidates typically come from different ethnic groups. In Kenya’s 2007 elections, for example, the two leading parliamentary candidates were from different ethnic groups in 71% of all races in Nairobi’s urban constituencies, relative to only 8% of races in other parts of the country. ⁵ As a result, aspirants in urban contests must simultaneously balance the competing objectives of mobilizing coethnic voters with courting out-group supporters.

Data on inter-marriage, mixed ethnicity, and social networks suggests that contact theory could be relevant in this context. Data from a 2012 national survey conducted by one of the authors shows that in Nairobi 21.9% of marriages crossethnic lines, relative to 9.2% elsewhere (difference = 12.7 percentage points, p < .01). ⁶ Relatedly, a larger share of Nairobi’s residents come from multiethnic backgrounds (defined as having parents from different ethnic groups): 17.0% in Nairobi vs. 8.5% elsewhere (difference = 8.5, p < .05). Social networks are also more diverse in Nairobi. The 2012 survey asked respondents to report up to four individuals with whom they “discuss politics and elections.” It then recorded multiple attributes—including ethnic identity—for each named person. Respondents in Nairobi were substantially more likely to report at least one non-coethnic in their social networks, relative to respondents in other parts of the country: 29.2% vs. 7.2% (difference = 22.0, p < .01). The 2012 survey data also demonstrate that social network diversity is higher for longer-term residents, suggesting that migrants to the city form cross-ethnic ties over time after moving to the capital. Figure 2 plots the association between length of time in Nairobi, measured in years and segmented into quintiles, and the probability of reporting one or more non-coethnic in one’s social network. Because the sample from Nairobi is small (n = 106), the estimates are imprecise. Nonetheless, the data suggest a positive relationship, particularly in the first years after moving to Nairobi. Only 13.6% among in the first quintile (a year or less in Nairobi) report one or more non-coethnics in their social networks, while 33.3% of those in the second through fifth quintiles do (difference = 19.7, p < .10). Finally, though we are not aware of any research that compares norms in urban and rural Kenya, existing studies using both ethnographic and quantitative methods cited above suggest the presence of egalitarian norms that stigmatize divisive forms of ethnic politics in Nairobi (Berge et al. 2020; Landau 2015).

OPEN IN VIEWER

Figure 2.

Association between Length of Time in Nairobi and Social Network Diversity. Figure shows the share of respondents who report one or more non-coethnic in their social network, across quintiles of the number of years they have lived in Nairobi. Network diversity is based on self-reports of up to four individuals with whom respondents “discuss politics and elections.” Data are from the first round of a two-round panel study conducted prior to Kenya’s 2012 election. The figure is based on respondents in Nairobi Province (n = 106).

Research Design

We conducted a vignette experiment in which candidate appeals—an ethnically targeted messages or an inclusive appeal—were varied randomly (treatment scripts are shown below). Respondents were assigned to listen to one of two pre-recorded campaign speeches by a hypothetical candidate who was “planning to run for parliament in a nearby constituency in 2017” using headphones. The experiment was part of a larger household survey carried out in Nairobi in August 2015. It used a 2 × 2 between-subjects design in which appeal type (targeted vs. inclusive) was crossed with candidate ethnicity (coethnic vs. non-coethnic).

In the targeted appeal, the candidate promises to ensure that members of his own group receive various benefits—opportunities for employment, access to health and education services, funds for development, and security. We embedded the treatments in a longer speech based on Dunning and Harrison (2010), modified to fit our context. The speech was designed to prime respondents to think about distributive politics by discussing inequalities in access to public services and private goods (electricity, education, and government jobs). The specific promises made by the candidate cover a variety of areas that might be subject to ethnic favoritism in Nairobi, according to local residents consulted during the design stage. While the language used here to signal ethnic targeting (“I will fight to make sure that our tribesmen have opportunities…”) may be atypical of the language used to signal representational intentions on the campaign trail, we found that more subtle appeals were often not interpreted as ethnic messages. We therefore chose language that clearly conveyed the candidate’s intent in order to examine reactions to unambiguous promises of ethnic favoritism. In a replication study fielded in 2016 (described below), we found that most respondents believed the appeal to be similar to those heard during actual elections. ⁷ The two versions of the speech were recorded in Swahili by the same actor (and all respondents heard the recordings in Swahili). We chose an actor who had lived in Nairobi since birth and did not speak Swahili with a discernible ethnic accent. ⁸

TREATMENT SCRIPTS

Intro (read by enumerators):

Now I will ask you to listen to a speech made by a candidate named John [surname] who is planning to run for parliament in a nearby constituency in 2017. After you listen to the recording, I will ask you for your views about the speech.

For this part of the survey, you will need to wear headphones. Please put these on.

In this speech, [surname] will discuss the policies and programs he would support if elected.

Audio recording:

Hello. I am here today to ask for your vote in my candidacy for the Parliament.

As a resident of Nairobi for over 25 years, I am familiar with the challenges we face. Too many young people cannot find jobs. Crime and insecurity are major problems. And the rising costs of food and fuel make it hard for ordinary families to get by.

Although we strive for equality, Nairobi is still highly unequal. Why is it that some people have electricity while others do not? Why is it that some of our children are going to school regularly while others are not? Why are some people able to find good jobs in the government while others are not?

Targeted appeal: If you elect me as your representative, I will fight to make sure that our tribesmen have opportunities for employment, have access to health and education services, receive funds for development, and live in safe areas. I will work with the other members of the Parliament to ensure we receive the resources we deserve.

Inclusive appeal: If you elect me as your representative, I will fight to make sure that people from all ethnic communities have opportunities for employment, have access to health and education services, receive funds for development, and live in safe areas. I will work with the other members of the Parliament to ensure that people from every ethnic group – Kikuyu, Luo, Kamba, Kisii, Kalenjin, Luhya, all groups – receive the resources they deserve.

These are attainable goals and I have the determination to achieve them. After 25 years as a citizen of Nairobi, 15 years in business and 10 years in education, I am ready to put my experience to good use in fighting for your rights.

So please, my dear brothers and sisters, allow me to humbly ask for your support in the 2017 election. If you vote for me, I will listen to your concerns and I will address them because they are my concerns too. Thank you for your attention and your support.

To signal the candidate’s ethnicity, we altered the candidate’s surname in the introductory script read by enumerators. We used a list of 10 common surnames (see Supplementary Information B Table A3) for each of the six ethnic groups included in the sample (Kikuyu, Luo, Kamba, Kisii, Luhya, and Kalenjin). ⁹ We removed names that might prime respondents to think of specific politicians. Respondents in the coethnic condition were randomly assigned one of the 10 coethnic surnames, and those in the non-coethnic condition were randomly assigned one of the names from the five out-group lists. Enumerators repeated the candidate’s surname twice in the introductory script to emphasize the candidate’s identity. The first name for all candidates was John, a common Christian name in Kenya. ¹⁰

As a manipulation check, we asked respondent to provide their best guess of the candidate’s ethnicity after completing the outcome questions. The strategy proved less effective than expected: only about two-thirds (65.3%) of respondents correctly identified the candidate as a coethnic or non-coethnic, while 29.7% guessed incorrectly, and 5.0% refused to answer (see Supplementary Information C Table A4). We address concerns about the potential effects of this variation on the results below.

The primary outcome measure asked respondents how likely they would be to support the candidate on a five-point scale ranging from “very unlikely” to “very likely.” The sample included 757 respondents from Nairobi County’s 18 parliamentary constituencies. Enumerators used a random-walk procedure to select households and randomly selected respondents within households from among those at home at the time of the visit. Supplementary Information D Table A5 compares our sample to that obtained by the 2011 Afrobarometer Round 5 survey and finds a close match on key demographic variables. Treatment assignment is well balanced across demographic variables (Supplementary Information D Table A6).

Results

Main Results

Results are presented in Figure 3, which shows mean values on support for the candidate by appeal type and candidate ethnicity (regression results are reported in Supplementary Information E Table A7). Relative to the inclusive appeal, respondents penalize candidates for making targeted ethnic appeals regardless of whether the candidate is identified as a coethnic or not. For coethnic leaders, the mean value is 3.29 when the candidate makes a targeted ethnic appeal relative to 4.08 when the candidate offers an inclusive message (difference = 0.79, p < .001). This difference is substantively meaningful given that the scale used to measure support for the candidate. The results are similar for non-coethnic candidates: the penalty for targeted appeals is slightly larger (.91) and significant (p < .001). While the penalty for targeted appeals is larger for non-coethnic candidates (.91 vs. .79), the difference is not statistically significant, consistent with the argument that voters disfavor ethnic pandering not only by leaders from other ethnic groups but also by those from their own.

OPEN IN VIEWER

Figure 3.

Candidate Evaluations by Appeal Type and Candidate Ethnicity. Figure shows mean values of support for the candidate by message type and candidate ethnic match with the respondent (with 95% confidence intervals). Support is measured using a five-point scale ranging from very unlikely to very likely. Figure is based on the OLS estimates in model 1 in Supplementary Information E Table A7. *** p < .01, ** p < .05, * p < .1.

We also examine a set of secondary outcomes that probe respondents’ assessments of multiple candidate characteristics. Following the primary outcome question, respondents were asked to assess the candidate on four dimensions: (1) how well the candidate would represent “people like you” (represent), (2) whether he would be good for democracy (good for democracy), (3) whether he would be good for peace (good for peace), and (4) whether would fulfill the promises made in the speech if elected (fulfill promises). Each item is measured on a four-point scale ranging from “not at all” to “a lot.”

As shown in Figure 4, the targeted appeal reduced assessments on all four dimensions, relative to the inclusive appeal (full results are presented in Supplementary Information E Table A8). The negative effects are especially large on “good for peace” and “good for democracy,” suggesting that pandering especially activates concerns about the divisive effects of appealing to ethnic identities. Coethnicity does not condition these evaluations.

OPEN IN VIEWER

Figure 4.

Effects of Targeted Appeal on Candidate Evaluations. Figure shows marginal effects of the targeted appeal relative to the inclusive appeals on four candidate assessments, each of which is measured on a four-point scale (not at all, a little, somewhat, and a lot). OLS estimates are presented in Supplementary Information E Table A8.

Conditional Effects

The theoretical framework proposes that urban voters may be more inclined to penalize ethnic pandering than rural voters. While our survey, which draws on an urban-only sample, does not allow for a direct test of this hypothesis, we leverage the fact that many of Nairobi’s residents are recent transplants from rural areas to examine whether the effects of the targeted message are conditional on how long individuals have lived in Nairobi. Only 16.8% of respondents in the sample were born in Nairobi Province (now County). About 10% had lived in Nairobi for a year or less at the time of the survey, and more than a third (36.3%) had moved to the capital in the last five years. We estimate interactive models using OLS in which an indicator for treatment assignment is interacted with a continuous measure of how long respondents have lived in Nairobi, measured in years. ¹¹ We estimate separate models for coethnic and non-coethnic candidates. These tests do not merit a causal interpretation since the moderator (length of time in Nairobi) is confounded by several other factors—particularly age, education, employment status, and wealth—that could affect how voters respond to overt ethnic appeals. We therefore treat the results as suggestive evidence consistent with the proposed theoretical framework.

The results are shown graphically in Figure 5, which superimposes the linear interaction on binned estimates of the effect across terciles of the moderator, as suggested in Hainmueller, Mummolo, and Xu (2019). Regression results are shown in Supplementary Information E Table A10. For non-coethnic candidates, length of time in Nairobi has no bearing: penalization of the targeted appeal (relative to the inclusive message) does not depend on how long respondents have lived in Nairobi. This result is unsurprising since we would expect all voters to punish mistargeted appeals that promise benefits to other groups. However, the penalty for ethnic pandering does vary by length of time in Nairobi when the candidate is identified as a coethnic. Respondents who have been in Nairobi for 4 years or less (the first tercile) do not penalize coethnic pandering, while those in the second (5–15 years) or third (16–60 years) terciles do. These results are robust to the inclusion of controls for age, education, wealth (measured with an asset index), and employment status, as well as the interaction between treatment assignment and each of these variables (Supplementary Information E Table A10). Nevertheless, because length of time in Nairobi is not randomly assigned, these results should be taken as suggestive.

OPEN IN VIEWER

Figure 5.

Conditional Effects by Years in Nairobi. Figures show the marginal effect of the targeted appeal relative to the inclusive message, when the candidate is identified as a coethnic and a non-coethnic. The linear interaction is superimposed on estimates of the effects across terciles of the moderator—indicated by low (L), medium (M), and high (H)—as recommended in Hainmueller at al. (2019).

A Replication Study

We draw on data from a second survey experiment conducted in 2016 as part of a larger household survey to replicate the main findings. The replication not only allows us to confirm the results with a different sample but also to address several of the limitations of the initial research design. Because we sought to investigate reactions to appeals made by coethnic candidates, all candidates in the replication were identified as coethnics. We used a secret ballot to measure the key outcome of interest—support for the candidate—in order to reduce the potential for social desirability bias (Carlson 2016). To simplify the balloting, respondents were asked to indicate whether they would vote for the candidate or not, rather than how likely they would be to vote for him, as in the initial study. The survey included a control condition that makes no mention of ethnicity in order to test the effects of the targeted appeal against an ethnically neutral message rather than the inclusive appeal in the initial survey. We also added an implicit ethnic appeal condition to test whether more subtle ethnic pandering is also penalized. We modified the language used in the explicit ethnic appeal so that the candidate promises to deliver benefits for all [R’s ethnic group], rather than “our tribesmen” as in the initial study. We note again that because the goal of the survey is to examine reactions to overt ethnic appeals, we opted for language that unambiguously conveyed the candidate’s intent. We also included a post-treatment question that allows us to examine the external validity of the script. We employed actors from each target ethnic group to record the treatments, rather than using a single actor, as in the original survey. Finally, we added a question on out-group contact that allows for a more direct test of whether the penalty for ethnic pandering varies across respondents who have more/less frequent contact with members of other ethnic groups. Additional details on the sample, treatment scripts, descriptive statistics, and balance across treatment conditions are provided in SI F.

Table 1 presents the results. Model 1 shows that relative to the control condition the targeted appeal reduced the share of respondents saying they would vote for the candidate by about 6 percentage points, from 71.6% to 65.5%. ¹² Model 2 confirms that the effects vary according to how long respondents have lived in Nairobi, as in the initial study. We again estimate the effects across terciles of the moderator (see Supplementary Information F Figure A2). No penalty is observed among respondents in the lowest tercile (0–4 years). The negative effect is significant among those in the highest tercile (17–67 years), though not in the middle tercile (5–6 years). Model 3 interacts the treatment indicator with respondents’ self-reported frequency of inter-ethnic contact. ¹³ While self-reports on this question may be inflated by social desirability bias, they provide a useful first cut for testing the relevance of contact theory. ¹⁴ The results in Model 3 show that the penalty is greater among those who report more frequent out-group contact, as we would expect if inter-group connections increase concerns about the divisive effects of ethnic political mobilization.

OPEN IN VIEWER

Table 1.

Replication Study Results (DV = candidate support).

	(1)	(2)	(3)
Targeted appeal	−0.060*	0.011	0.089
	(0.031)	(0.045)	(0.085)
Targeted appeal × years in Nairobi		−0.006**
		(0.003)
Years in Nairobi		−0.003*
		(0.002)
Targeted appeal × out-group contact			−0.062*
			(0.034)
Out-group contact			−0.068***
			(0.025)
Constant	0.716***	0.756***	0.888***
	(0.021)	(0.031)	(0.063)
Observations	898	898	896
R-squared	0.004	0.028	0.036

Robustness and Alternative Explanations

We briefly address two robustness concerns and an alternative explanation. First, as mentioned earlier, 28.5% of respondents in the initial study misidentified the candidate’s ethnicity. The results presented above include the full sample since dropping those who misidentify the candidate could introduce post-treatment bias (Montgomery, Nyhan, and Torres 2018). As an alternative to the main specification, we adopt an instrumental variables approach (Gerber and Green 2012) that uses treatment assignment as an instrument for the respondent’s identification of the candidate as a coethnic or non-coethnic. Results (Supplementary Information G Table A17) show that the two-stage least squares (2SLS) estimates are almost identical to the main results presented above.

A second concern is that the observed penalization of targeted appeals by coethnic candidates may stem from social desirability bias rather than a genuine distaste for ethnic pandering. The data suggest that the effects cannot be explained solely by self-censoring. Building on prior literature that shows that enumerator ethnicity systematically affects survey responses in Africa (Adida et al. 2016; Carlson 2016), we examine whether the results vary by interviewer identity. Our expectation is that respondents should be less willing to express support for a coethnic candidate who makes ethnically targeted appeals when interviewed by a non-coethnic enumerator. Results (Supplementary Information H Table A18), however, show that the negative effect of pandering holds regardless of enumerator identity, and the observed penalty is actually smaller when the respondent is interviewed by a non-coethnic. Moreover, as noted, the negative effects obtain in the replication study when respondents are able to provide responses using a secret ballot.

Finally, we discuss an alternative explanation for the penalization of targeted appeals by coethnic candidates. Urban voters in diverse constituencies may disfavor ethnic pandering because they understand that coethnic leaders need to cultivate out-group support if they are to be electorally viable. To address this concern, we would ideally test whether the penalty for ethnic pandering is also observed in homogenous constituencies where strategic concerns would have less relevance. Because our sample is limited to Nairobi’s diverse parliamentary constituencies, we cannot implement this test. Yet, the available data suggest that strategic considerations do not account for the observed penalty. We expect that if strategic considerations related to candidate viability are at work, voters from ethnic groups that make up a smaller share of the electorate at the constituency level should penalize ethnic pandering more than voters whose groups comprise a larger share of the constituency-level population. ¹⁵ However, results from both the initial study and the replication (Supplementary Information I Table A19) show that the penalty for ethnic pandering does not vary systematically with constituency-level measures of group size. We also find that the effects for respondents from the local plurality group at the constituency level punish ethnic pandering equally with those from smaller groups (Supplementary Information I Table A18). Together, these results suggest that strategic concerns regarding electoral viability likely do not explain our main findings.

Conclusion

Appealing to ethnic interests is thought to be an effective strategy for increasing political support among coethnics in Africa’s diverse societies. Yet, some evidence suggests that urban voters may hold a preference for inclusive politics and scorn electoral strategies that run the risk of heightening ethnic divisions (Berge et al. 2020; Kim 2021; Landau 2015). This study examines reactions to ethnic pandering in Nairobi using data from a survey experiment conducted in 2015 and a replication study from 2016. Results show that respondents penalize candidates for using ethnically targeted messages relative to those making inclusive appeals regardless of shared ethnic ties with the candidate. Penalization for pandering by coethnic candidates is greater among longer-term urban residents and those who interact more frequently with out-groups. We propose that these results reflect a trend toward greater social integration in urban Africa, driven by the rise of inter-ethnic marriage, a growing population of mixed-ethnic citizens, dense social ties that bridge ethnic divides, and the emergence of norms that stigmatize divisive forms of ethnic political mobilization. Although the study examines voters in one urban area, we expect the results to hold in other urban settings in Africa where social ties cut across group lines.

We emphasize that our results should not be taken to mean that ethnic mobilization is a thing of the past in Kenya’s urban areas. Ethnic bloc voting remains the order of the day in Nairobi. And there is good reason to expect that voters may respond favorably to communal appeals framed in terms of threats to group interests rather than overt promises to favor one group over others (Cheeseman, Kanyinga and Lynch 2020). Thus, the results do not call into question the continued salience of ethnicity to voters in Nairobi or suggest that they will punish all types of ethnic political mobilization. Rather, they demonstrate that even in settings where ethnicity remains salient, social integration, particularly among long-term urban residents, may promote a preference for inclusion.

The findings suggest several avenues for future research. First, the effects of violence on norms related to ethnic political contestation merit additional exploration. Related research from Kenya suggest that the inter-group violence that followed Kenya’s 2007 election may have heightened the rejection of divisive ethnic politics among some segments of the electorate. Deacon (2015), for example, shows that in the subsequent elections in 2013, many church leaders throughout the country made a concerted effort to remain politically neutral in order to avoid a repeat of the 2007–08 election violence. Given that election-related violence is commonplace in Africa’s emerging democracies (Tyalor, Pevehouse, and Straus 2017), the potential effects of violence in crystalizing inclusive norms and preferences deserves greater attention. Second, we call for renewed effort to study the changing social and political dynamics in urban Africa. Particularly important is to distinguish between preferences, norms, and behaviors—recognizing that urbanization may have distinct effects on some aspects of ethnic identities and politics but not others. Patterns of ethnic bloc voting may remain the norm in settings—like Nairobi—where voters are compelled to support the party or candidate associated with their group for fear of being left out in distributive allocations. Yet, the persistence of bloc voting should not be taken as evidence that preferences and norms remain static among Africa’s increasingly integrated and cosmopolitan urban residents.

Supplemental Material