Who’s Asking? Interviewer Coethnicity Effects in African Survey Data

Abstract

Face-to-face interviews constitute a social interaction between interviewer and respondent, and in the African context, social interactions are strongly shaped by ethnicity. Yet research using African survey data typically fails to account for the effect of shared ethnicity on survey responses. We find that respondents give systematically different answers to coethnic and noncoethnic interviewers across surveys in 14 African countries, but with significant variation in the degree of bias across question types and types of noncoethnic dyads, with the largest effects occurring where both the respondent and interviewer are members of ethnic groups with a history of political competition and conflict, and where the respondent or interviewer shares an ethnicity with the head of state. Our findings have practical implications for consumers of African survey data and underscore the context dependence of the social interaction that constitutes the survey experience.

Keywords

African politics survey design political psychology race ethnicity and politics

A recent survey in South Africa asked respondents whether they thought the African National Congress (ANC) government had succeeded in uniting the country following the collapse of apartheid. Whereas 69% of Black respondents felt that the government had done well in uniting South Africa, just 45% of Whites agreed. However, for White respondents interviewed by Black interviewers, this share jumped to 65%. In the same survey, respondents were asked whether they thought life was better today than under apartheid. When interviewed by a Black interviewer, 45% of Whites agreed, but when interviewed by a White interviewer, this number dropped to 17%. Something about the nature of the interaction between the interviewer and the interviewee—combined, no doubt, with the sensitive political implications of the questions—dramatically altered survey responses.

Similar patterns emerge in other African surveys. In Uganda, when survey respondents were asked to choose between identifying themselves as Ugandan or as a member of their ethnic group—a loaded question in a country where emphasizing one’s tribal identity is generally frowned upon (Matsiko, 2014)—the answers people gave were systematically different when the person asking the question was a fellow group member. Whereas 17% said they felt more strongly attached to their ethnicity when being interviewed by a noncoethnic, 23% admitted to holding this view when the interviewer was a coethnic. Kenyan survey respondents were similarly sensitive to who was conducting the interview. When asked how much they trusted Kenyans from other ethnic groups—a question that triggers strong social norms against openly admitting discriminatory sentiments—35% of those interviewed by a noncoethnic said they trusted noncoethnics, but when the interviewer was a coethnic, this dropped to 26%.¹

These discrepancies, which accord with findings from research on race-of-interviewer effects in the United States (e.g., Berinsky, 2004; Campbell, 1981; Conover, 1984; Converse & Schuman, 1974; Cotter, Cohen, & Coulter, 1982; Holbrook, Green, & Krosnick, 2003; Weeks & Moore, 1981), underscore the extent to which the opinions ventured on surveys in Africa may vary with the ethnic match between the respondent and the person administering the survey. Few studies using African survey data, however, take account of such interviewer coethnicity effects.² Given the large and growing body of research in Africa that relies on data drawn from attitudinal surveys, it is critical that we investigate the extent of this phenomenon and understand its implications for interpreting the findings of African public opinion research.

The ethnic match between an interviewer and respondent affects the answers provided in African survey data because the social norms about the attitudes one should not publicly express, as well as the desire to present oneself in a positive light, are likely to be stronger in interactions with ethnic outsiders than with members of one’s own community. Not all survey questions are equally sensitive to coethnicity response bias, however. Questions that mention ethnicity explicitly, such as whether a respondent believes her group is unduly favored with respect to economic circumstances or political influence, may prime respondents to the norms that govern coethnic versus noncoethnic interactions more than questions that are ostensibly more objective in nature, such as whether the respondent has access to clean water.

Coethnicity response bias is also likely to be stronger when the respondent–interviewer dyad contains members of politically relevant (as opposed to merely ethnographically distinct) groups and when either the respondent or the interviewer shares an ethnic link with the head of state—expectations that we test directly below. The motivation for the first test lies in the expectation that the degree of bias we observe will be related to the political or social salience of the ethnic divide between the interviewer and the respondent. For example, a Kikuyu respondent may give different answers when interviewed by a Luo (a group with which the Kikuyu have historically had contentious interactions) than by an Embu (a group that is ethnographically distinct from the Kikuyu but culturally and politically linked). In the first case, the difference in ethnic group memberships between the interviewer and respondent is likely to be highly relevant; in the latter case, much less so.

The motivation for the second test originates in the close association between heads of state in Africa and their ethnic groups. Because of this association, any evaluation of a head of state (or his government) is also implicitly an evaluation of his group. Hence, a respondent from the leader’s group who is asked about the leader’s performance will be evaluating the performance of her own ethnic community (and may be expected to provide a rosier opinion than she might if evaluating a member of another group). Similarly, when the interviewer shares the ethnicity of the head of state, the respondent is put in the position of evaluating the performance of the interviewer’s group. Given a desire not to offend the interviewer, we might expect respondents under such circumstances to evaluate the head of state more positively than they might if they were being interviewed by a fellow group member. Although these considerations should matter primarily for questions involving direct evaluations of the president, they may also shape responses to questions about support for the democratic political system, trust in political parties, and satisfaction with the government’s handling of the economy, among other politically related topics.

We test these expectations using data from the Afrobarometer in a cross-section of 14 African countries. Our investigation of coethnic interviewer effects (which we define as the difference in response patterns across interviews conducted by coethnics and noncoethnics) is made possible by the collection of original data on the ethnic identities of more than 1,100 Afrobarometer interviewers.³ We proceed in two stages. First, as the Afrobarometer does not randomly assign survey interviewers to respondents, we model the interviewer assignment process. Then, controlling for the factors that determine interviewer assignment, we estimate the effect of being interviewed by a noncoethnic on responses to a range of survey questions, including questions explicitly concerning ethnicity, measures of political attitudes and behavior, questions about socioeconomic status, and interviewer perceptions of the respondent’s cooperation and demeanor during the interview.

We find systematic effects of respondent–interviewer noncoethnicity on a variety of survey questions. Respondents who are interviewed by noncoethnics give answers that are almost always more socially desirable or biased so as to present themselves or their ethnic group more favorably. For example, respondents interviewed by noncoethnics are more likely to say that they prioritize their national identity over their ethnic identity, more likely to say that their ethnic group is economically and politically advantaged, less likely to say they think their ethnic group is treated unfairly, and more likely to say they prefer democracy, approve of the president’s performance, trust the ruling party, and voted in the last election. Respondents are also perceived to be more hostile and suspicious when interviewed by a noncoethnic. These findings point to systematic effects of interviewer coethnicity across a range of question topics.

The sizes of the effects we uncover are relatively modest, however—largely because our pooled analysis masks substantial heterogeneity in the characteristics of noncoethnic dyads. We therefore explore that heterogeneity in line with the expectations outlined earlier. First, we investigate how the political salience of the particular cleavage reproduced in the survey dyad affects our estimates. To explore the upper limit of this effect, we begin with one of the most notorious and salient political cleavages in Africa: between Black and White South Africans. Our analysis demonstrates that noncoethnic interviewer effects are massive in this context, as much as 10 times higher than the average effects we estimate for the full sample. Though obviously an extreme case, South Africa establishes just how important noncoethnic interviewer effects can be under certain circumstances. We then turn back to the pooled sample, where we show that the political relevance of the dyad between interviewer and respondent groups also matters, with politically relevant noncoethnic dyads generally producing larger effects than dyads containing groups that are not politically relevant.

Next, we explore how coethnicity with the leader shapes noncoethnic interviewer effects. We first show that White and Black South African respondents react to noncoethnic interviewers in sharply divergent ways, particularly when evaluating the performance of the president—a Black South African—and the party he leads, the ANC. While Whites tend to improve their evaluations of the incumbent when interviewed by a Black, Blacks do the opposite when interviewed by a White. We then return to the pooled sample to explore how the respondent’s or interviewer’s ethnic tie with the head of state affects the size of the bias generated by being interviewed by a noncoethnic. Although heterogeneity in the salience of the ethnic tie with the leader somewhat muddies the results, we find interesting and significant effects for some question items.

Taken together, our findings reinforce the view that survey data collection constitutes a social interaction (Berinsky, 2004): The same social norms that govern everyday conversations, such as the concern to present an admirable public impression, affect the responses generated in surveys. Our findings underline the importance of being attuned to the ways in which the ethnic or racial group memberships of interviewers and survey respondents help to define the nature of that social interaction in the African context. On one hand, our results highlight the need for users of African survey data to control for interviewer noncoethnicity so as to reduce the potential bias caused by these effects and thus to get their measures of public opinion “right.” On the other hand, they also suggest that there may be no such thing as an objective opinion or attitude (or even a report of one’s own socioeconomic circumstances) apart from the social context in which it is expressed, and that in an African setting—or, as the heterogeneity in the dyad- and question-level results make clear, in some social interactions revolving around some issues—social context is strongly shaped by ethnicity. Seen in this light, ethnicity becomes not something to control for but a critical causal input into how Africans perceive and answer questions regarding their views about the world.

Who’s Asking: When and Why Ethnicity Matters in African Surveys

Political scientists have relied on scientific surveys since the 1940s to gather observations about the political world, and these have proven to be a powerful research tool (Brady, 2000). Yet surveys also present a number of challenges, chief among them the difficulty of knowing whether respondents have provided truthful and accurate information. A vast literature on “response bias” attempts to identify the conditions under which self-reported attitudes and opinions may diverge from privately held views. In the U.S. context, a key emphasis in this literature is on race-of-interviewer effects. In one of its earliest demonstrations, Black respondents in Memphis were found to express more patriotic feelings when the interviewer was White than when the interviewer was Black (Hyman, Cobb, Feldman, Hart, & Stember, 1954). Similarly, White survey respondents expressed more support for interracial marriage and the racial integration of schools when interviewed by Black interviewers than by White interviewers (Hatchett & Schuman, 1975). More recently, Davis (1997) found that responses in over 60% of the attitudinal questions in the 1984 National Black Election Study correlated significantly with an interviewer’s race.

Notwithstanding the salience of social identity in other (particularly developing country) settings, there has, until recently, been surprisingly little research into ethnicity-of-interviewer effects in a comparative context. Four recent articles may signal a change in this trend. Relying on a randomized survey experiment in Egypt, Blaydes and Gillum (2013) find that female Muslim interviewers wearing Islamic headscarves elicit greater expressions of personal piety from survey respondents than the same interviewers dressed in secular garb. Dionne (2014) finds that interviewer coethnicity affects respondents’ willingness to answer sensitive questions related to sexual behavior in Malawi. In Burundi, Samii (2013) uses different nonresponse rates across coethnic and noncoethnic interviewer–respondent dyads as a measure of ethnic prejudice. And in a study of the impact of local diversity on interethnic trust in Kenya, Kasara (2013) uses respondent–interviewer coethnicity to rule out the possibility that social desirability bias underlies her result. While the attention these articles pay to ethnicity-of-interviewer effects in developing country settings is welcome, they all focus on single question topics (religiosity, sexual behavior, interethnic trust) or a particular form of noncooperation (refusal to answer) and draw on samples from within single countries. It is therefore not clear whether the patterns they identify generalize to the full breadth of topics likely to be affected by noncoethnic interviewers or whether the reported findings hold up in a broader set of contexts.

We address this gap in understanding through a systematic study of ethnicity-of-interviewer effects across Sub-Saharan Africa. Scholars of African politics have long recognized the social salience of ethnicity. Some trace this back to the continent’s experience with the slave trade, where intervillage competition and trade with European powers provided individuals an incentive to sell prisoners of war (or even their own kin) into slavery (Piot, 1996). This competition weakened intervillage ties and increased social fragmentation, with direct implications for ethnic diversity in Africa (Nunn, 2008). Others point instead to disruptive colonial policies, which simultaneously consolidated diverse tribes into single ethnic groups and partitioned others across national borders (Asiwaju, 1985; Englebert, Tarango, & Carter, 2002). Furthermore, Europeans governed and administered their colonial territories by reifying ethnic boundaries, setting the stage for an enduring legacy of ethnic–based politics long after independence (Bates, 1983; Laitin, 1986). Regardless of the origins of ethnic divisions in Africa, research has shown that ethnicity affects instability (Jackman, 1978), taxation (Kasara, 2007; Lieberman, 2003), access to education and health care (Franck & Rainer, 2010), voting (Adida, 2015 Carlson, 2015; Conroy-Krutz, 2013; Eifert, Miguel, & Posner, 2010; Ferree, 2006, 2011; Long & Hoffman, 2013), and patterns of interaction in behavioral games (Burns, 2012; Habyarimana, Humphreys, Posner, & Weinstein, 2009; Jeon, Johnson, & Robinson, 2013; Robinson, 2016). We draw from this rich literature to argue that ethnicity is also likely to matter during face-to-face survey interactions.⁴

Assignment of Interviewers to Respondents

Before estimating the effects of interviewer-respondent coethnicity, we investigate the process by which respondents are assigned to coethnic or noncoethnic interviewers in Afrobarometer surveys. If the assignment were random, then the estimation of ethnicity-of-interviewer effects would be straightforward: We could simply compare the responses of individuals interviewed by a coethnic interviewer with those of individuals interviewed by someone of a different ethnicity and conduct a difference-in-means test. However, the data suggest that interviewer–respondent assignment was far from random. In South Africa, for example, fully 82% of Xhosas and 65% of Zulus were interviewed by coethnics, whereas under random assignment these percentages should both be below 5% (see Table A1 of the Online Appendix for similar statistics on other South African ethnic groups).

Such departures from random assignment (which are found not just in South Africa, but in every country we study) stem from the practical complexities of fielding surveys in multiethnic settings. Afrobarometer interviewers deploy in teams, with each team assigned to specific geographic regions and interviewers assigned to teams so as to maximize the number of interviews that can be conducted in respondents’ home languages. This implies that respondents from large or regionally dominant groups, or from groups with unique or difficult languages, will experience a larger share of coethnic interviews than will respondents from smaller groups or who are living in more diverse (e.g., urban) communities. Afrobarometer country directors also strive to minimize the number of interview dyads containing groups with historically contentious relationships, so these are likely to be underrepresented in the sample.

The fact that interviewer assignment is not random compels us to identify and control for the factors that underlie the assignment process so that we can separate out the treatment effect of being interviewed by a coethnic interviewer from the selection effect of being assigned to a coethnic interviewer. We estimate the determinants of being assigned a noncoethnic interviewer in the pooled sample of 14 Afrobarometer countries for which we were able to collect information about the ethnic identity of survey interviewers: Benin, Burkina Faso, Ghana, Kenya, Malawi, Mali, Mozambique, Namibia, Nigeria, Senegal, South Africa, Uganda, Zambia, and Zimbabwe. To relate survey responses to the ethnic match between interviewer and respondent, we combine information from the Afrobarometer on the ethnicity of the respondent with original data that we collected on the race and ethnicity of the interviewers.⁵

Our model of treatment assignment includes both individual characteristics of the respondent (gender, age, education) and several factors that our qualitative understanding of the assignment process suggests may be relevant, such as the size (and the square of the size) of the respondent’s ethnic group, the size (and the square of the size) of the interviewer’s ethnic group, whether the respondent is a member of a regional minority, the respondent’s urban/rural location, and the administrative region in which the survey is conducted. This last factor may be important insofar as regions vary in their ethnic demography and are assigned interviewer teams with different ethnic compositions, hence altering the likelihood of being assigned a noncoethnic interviewer. We also include a dummy for the Afrobarometer survey round, because both interviewer teams and, at least potentially, the country organizers’ commitment to matching respondents and interviewers from the same groups, may vary across rounds. We estimate multiple versions of the model, varying the combinations of regional, respondent ethnic group, and interviewer ethnic group fixed effects that we include. The results are reported in Table 1.

Table 1.

Assignment to a Noncoethnic Interviewer in 14 African Countries.

	(1)	(2)	(3)	(4)
Male	0.00 (0.00)	−0.00 (0.00)	−0.01 (0.00)	−0.00 (0.00)
Age	−0.00 (0.00)**	0.00 (0.00)	0.00 (0.00)	0.00 (0.00)
Education	0.00 (0.00)	0.00 (0.00)***	0.00 (0.00)**	0.00 (0.00)**
Round 4	−0.06 (0.01)***	−0.06 (0.00)***	−0.05 (0.00)***	−0.05 (0.00)***
Urban	0.01 (0.01)	0.01 (0.01)	0.01 (0.01)***	0.01 (0.00)***
Regional ethnic minority	0.40 (0.00)***	0.25 (0.01)***	0.25 (0.01)***	0.25 (0.01)***
Respondent ethnic group size		−0.71 (0.07)***
Respondent ethnic group size²		0.07 (0.14)
Interviewer ethnic group size		−1.78 (0.08)***	−1.86 (0.08)***
Interviewer ethnic group size²		1.47 (0.16)***	1.59 (0.17)***
Country FEs	Yes	No	No	No
Region FEs	No	Yes	Yes	Yes
Respondent ethnic group FEs	No	No	Yes	Yes
Interviewer ethnic group FEs	No	No	No	Yes
Observations	32,911	32,900	32,900	32,900
R ²	.21	.38	.46	.52

Robust standard errors are in parentheses. FE = fixed effects.

p < .1. **p < .05. ***p < .01.

We find strong effects of respondent ethnic group size, interviewer ethnic group size, interviewer ethnic group size squared (suggesting that interviewers from both the smallest and largest groups are more likely to interview noncoethnics), and regional minority status. We also find evidence that interviewers and respondents were matched more systematically along ethnic lines in Round 4 than in Round 3. Individual level variables (except education) largely wash out once we control for group-level characteristics. The fact that the addition of each set of fixed effects pushes up the R² value for the model confirms the importance of unobserved regional and group factors in the assignment process—and of the need to include them as controls in our estimates of the effects of being interviewed by a noncoethnic (as we do below).

Ethnicity-of-Interviewer Effects in 14 African Countries

We can now estimate the effects of being interviewed by a noncoethnic interviewer, controlling for the correlates of treatment assignment identified above.⁶ We note that our empirical strategy, which involves comparing answers given by respondents who were interviewed by coethnics and noncoethnics, does not permit us to adjudicate whether the effects we find stem from the bias caused by being interviewed by a fellow group member or the bias caused by being interviewed by an ethnic outsider (or from both, in combination). We may have theoretical reasons to suspect one channel rather than the other for a particular question—and our strong hunch, reflected in our interpretation of most of our results below, is that most of the bias is due to the effect of being interviewed by a noncoethnic—but our empirical approach provides us with no leverage to disentangle these two effects. Even so, a finding of a significant difference in response patterns across coethnic and noncoethnic interview dyads tells us something important about the ways that ethnicity shapes the social interaction that constitutes the survey.

We also note that, in coding coethnic and noncoethnic dyads, we assume that respondents can correctly discern whether or not the interviewer is a member of their own group. Afrobarometer interviewers introduce themselves by name when they approach potential respondents at the start of the interview process, so this information, combined with the interviewer’s phenotypical characteristics, accent, and other visual cues, should put respondents in a good position to make an inference about shared ethnicity. However, we cannot rule out the possibility that these inferences may sometimes be incorrect. Such misidentifications would bias against finding an effect of being interviewed by a noncoethnic, so any results we report are despite this built-in downward bias in our estimates. Moreover, downward bias from misidentification is unlikely to be very large. As noted above, we asked the country directors who provided us with information about each interviewer’s ethnicity for their judgment about whether each interviewer was likely to be recognized as a member of his or her ethnic group, and our results are qualitatively unchanged if we restrict the sample to dyads containing interviewers who were judged by the country directors to be readily identifiable (see Online Appendix Figure A1).⁷

We focus on three types of questions that deal with topics for which social norms may generate varying degrees of pressure for certain types of responses in the African context we study. First, we consider seven questions that deal with explicitly ethnic topics such as the extent to which respondents privilege their national versus their ethnic self-identities, the degree to which they trust noncoethnics, and their perceptions of discrimination faced by members of their ethnic group. Given the negative stigma attached in most African countries to “tribalistic” attitudes (Juma, 2012), we might expect respondents interviewed by noncoethnics to be less willing to provide answers that suggest a preference for (or bias against) members of their own (or another) ethnic community. On the other hand, given the close connection between feelings of self-worth and perceptions of the status of one’s ethnic group (Bilig, 1976; Horowitz, 1985), we might expect respondents interviewed by noncoethnics to overstate their group’s economic well-being, political influence, and positive treatment by the government. For these reasons, questions dealing with explicitly ethnic issues are prime candidates for noncoethnic response bias, and where we expect to observe the largest effects of interviewer noncoethnicity.

Second, we examine 11 questions about political attitudes and behavior, including support for the head of state and ruling party, preferences for democracy, political engagement, and political knowledge. Insofar as there are strong social norms in the countries we study about supporting democracy (Leininger, 2014), voting in elections, and being actively engaged and interested in local political and community affairs, respondents might be less likely to admit to not doing these things when interviewed by someone from outside their own ethnic group. For similar reasons, respondents interviewed by noncoethnics might be less likely to admit that they disapprove of the president’s performance, distrust the ruling party, think that the government’s handling of the economy has been poor, or accepted (or even merely were offered) food or gifts in exchange for their vote. With respect to political knowledge, research suggests that noncoethnic interviewers may be less able to elicit effort from respondents, including the cognitive effort required for factual recall (Krosnick & Alwin, 1987; Weinreb, 2006). This tendency would generate a negative association between being interviewed by a noncoethnic and measures of political knowledge.

Third, we consider a set of questions about one’s own socioeconomic conditions and experience with poverty. Such questions might appear to involve little more than the reporting of basic facts, but, as Davis and Silver (2003) find in the U.S. context, because such self-reports have implications for one’s social status, they may be sensitive to who is asking. For example, we might expect respondents interviewed by noncoethnics to report higher levels of well-being than they do to coethnics. We also include a question asking whether the respondent knows someone who has died of AIDS. Given the stigma attached to this disease (Rankin, Brennan, Schell, Laviwa, & Rankin, 2005), we might expect to find hesitancy among survey respondents to admit to knowing a friend or relative who died of AIDS, and particular hesitancy when the interviewer is from a different racial or ethnic community.

Finally, we also examine four survey items in which interviewers are asked to rate the demeanor of their respondents. These items provide a test of whether—because of either the behavior of the respondent or the perceptions of the interviewer—respondents interviewed by noncoethnics are judged to be more hostile, uncooperative, impatient or suspicious. While not measures of social desirability bias, such responses provide insight into the ways in which the social interaction between the interviewer and respondent is affected by their shared or differing ethnic backgrounds. Question wording and response options for all questions are available in Table A2 of the Online Appendix.

The estimated effects of being interviewed by a noncoethnic interviewer are presented graphically for all 28 outcomes, broken down by question type in Figures 1 to 4.⁸ Because we present results for many outcome variables with different scaling and variability, we standardize all outcomes by country so that effect sizes are in country-specific standard deviation units.⁹ Thus, a coefficient of 0.1 indicates that being interviewed by a noncoethnic rather than a coethnic is associated with a 10th of a standard deviation change in a particular outcome.

Figure 1.

Impacts of a noncoethnic interviewer on responses to ethnic questions.

Figure 2.

Impacts of a noncoethnic interviewer on responses to political questions.

Figure 3.

Impacts of a noncoethnic interviewer on responses to economic questions.

Figure 4.

Impacts of a noncoethnic interviewer on interviewer’s reports of respondent’s demeanor.

Figure 1 focuses on explicitly ethnic questions. Consistent with a tendency for respondents to feel the need to assert their group’s self-worth and relative standing when confronted with an outsider, we find that respondents interviewed by noncoethnics are more positive about their group’s comparative economic conditions, political influence, and treatment by the government. We also find that respondents interviewed by noncoethnics are more likely to say they privilege a national over ethnic affiliation. This latter result, although just shy of statistical significance at conventional levels, is consistent with the expectation that being interviewed by a noncoethnic will generate more socially desirable responses.

We turn to political attitudes and behavior in Figure 2. We find that respondents interviewed by noncoethnics are statistically significantly more likely to express a preference for democracy, approval of the president’s performance, and trust in the ruling party, and to say they voted in the last national election—all more socially desirable responses. In addition, they are less likely to report knowing their Member of Parliament’s name, consistent with the expectation that respondents exert less effort in recalling information when asked by a noncoethnic interviewer.

With respect to their own socioeconomic conditions (Figure 3), respondents interviewed by noncoethnics are less likely to admit having gone without enough food, cash income, or clean water, to have feared crime in their own home, or to admit to knowing someone who died of AIDS. With the exception of fearing crime in their own home, none of these effects is statistically significant at conventional levels. But in every case, being interviewed by a noncoethnic generates survey responses that, in the local context, would be viewed as more socially desirable. The only result that runs counter to this pattern is with respect to respondents’ reports of their own present living conditions, which suggest that being interviewed by a noncoethnic is associated with more negative assessments.

Finally, Figure 4 gives us a sense of the extent to which the interview procedure itself is affected by the match between the ethnic group memberships of the interviewer and the respondent. We find evidence that it is. Specifically, we find that interviewers interacting with noncoethnics are more likely to report that the person they were interviewing was hostile or suspicious.

Overall, we find support for our expectation that respondents give systematically different answers to noncoethnic interviewers than to coethnic interviewers. In almost all cases, the direction of this effect is consistent with stronger social desirability bias in ethnically mismatched interviews. We anticipated that these effects would be largest for questions that deal explicitly with ethnicity, but this is not entirely borne out in our findings. While the largest effect is indeed on an ethnic question—whether the respondent believes her group is treated unfairly—not all ethnic questions are equally affected, and we find systematic effects for the other three question types, as well. This suggests that the ethnic match between interviewer and respondent is consequential for a wide range of question topics.

Alternative Explanations

Thus far, we have been interpreting differential response patterns across respondents interviewed by coethnics and noncoethnics as being due to social desirability bias—that is, due to a heightening, in the presence of a noncoethnic, of the natural desire to present oneself in a positive light and to be seen not to violate societal norms about what one should believe and how one should behave. But it is at least possible that these differences could stem instead from higher refusal rates in situations where an interviewer approaches a noncoethnic respondent or from lower response rates (especially for the sorts of sensitive questions we are studying) in noncoethnic dyads.¹⁰ Both could lead respondents with the most extreme beliefs and attitudes to select out, thus biasing average responses in noncoethnic dyads toward more socially desirable positions. This is still an ethnicity-driven bias, but the interpretation is somewhat different.

We do not, however, believe that this is the source of our findings. The rate of “don’t know” and “refused to answer” responses are no different in coethnic and noncoethnic dyads.¹¹ Interview refusal rates also do not appear to be significantly related to whether or not the interviewer approached a coethnic or noncoethnic potential respondent (see the Online Appendix for details).

Another potential concern is that regions with a history of intergroup conflict might be less likely to have noncoethnic interviewers assigned to them (as noted, Afrobarometer country directors strive to minimize the number of interview dyads containing groups with historically contentious relationships). This would lead noncoethnic dyads to be disproportionately located in places where peace and good intergroup relations prevail, which would obviously bias answers to several of our questions of interest. However, because our preferred specification includes region fixed effects, we can rule out this possibility as long as we believe that everyone in a conflict-exposed region is likely to have been affected by that exposure in the same way, which we think is reasonable.

A final concern is that the process used to assign interviewers to respondents may lead to an overrepresentation of highly educated, multilingual interviewers in noncoethnic dyads. To the extent that respondents react differently to more cosmopolitan, multilingual interviewers, what looks like a noncoethnic dyad effect may in fact be an artifact of the type of interviewer that is more likely to be present in such interactions. Again, we do not believe that this alternative explanation accounts for our findings. First, the vast majority of the 1,115 interviewers in our sample had at least some postgraduate education, and fully 84% of them were from an urban area. So the level of “cosmopolitanness” is likely to be universally high across coethnic and noncoethnic dyads. Indeed, 81% of both coethnic and noncoethnic interviews were conducted by interviewers with at least some postgraduate education. And while we do find a statistically significant difference in the share of coethnic and noncoethnic interviews that were conducted by interviewers from urban areas, it is the coethnic, not noncoethnic, interviews that were (slightly) more likely to have been conducted by an urbanite (80% vs. 77%, t = 6.81, p < .0001, two-tailed).

Second, as shown in Figure A2 of the Online Appendix, most of our results are unchanged if we re-estimate our main model using interviewer fixed effects. This highly conservative specification looks only at variation in answer patterns across coethnic and noncoethnic respondents interviewed by the same interviewer. It therefore controls for all possible interviewer characteristics that might provide alternative explanations for our findings. The fact that our findings are robust to this specification increases our confidence that the effects we estimate are driven by changes in the way respondents answer survey questions when those questions are asked by a member of a different ethnic group.

Variation in Effects Across Noncoethnic Dyads

Given the frequency with which Afrobarometer data are used to make generalizations about attitudes and behavior in Africa writ large, estimating the general effects of having a noncoethnic interviewer across multiple countries is a worthwhile undertaking. However, pooling the analysis in this way masks significant heterogeneity in the sizes of the effects across different types of noncoethnic dyads.¹² We explore two potential sources of this dyadic variation, both rooted in the nature of the political relationship between the ethnic groups to which interviewers and respondents belong. The first is whether the interviewer–respondent dyad is between members of groups with a history of political competition; the second is whether the interviewer or respondent is a coethnic of the president.

To explore how the political salience of the dyad shapes the effect of being interviewed by a noncoethnic, we begin with an illustration based on the most politically significant and historically laden social cleavage on the continent: race in South Africa. The salience of race in South Africa is in large part a product of the comprehensive set of policies initiated by the National Party in 1948 known as Apartheid. Building on prior policies of segregation, the state categorized all South Africans into one of four racial groups—White, Black, Coloured, and Asian—and used race to allocate essential goods and services like education, employment, public amenities, housing, and legal justice (Thompson, 1990). Even after the end of Apartheid and the rise to power of the ANC in democratic elections in 1994, the deep imprint of centuries of racially prejudicial policies persists, and large economic inequalities endure between racial groups (Leibbrandt, Finn, & Woolard, 2012; Seekings & Nattrass, 2002). Furthermore, voting continues to separate along racial lines. Although the ANC includes members of all groups and boasts a multiracial leadership, Blacks comprise all but a tiny portion of its support base while the primary opposition party, the Democratic Alliance (DA), attracts mostly non-Black voters (Ferree, 2011). Finally, social relations across racial lines may have loosened in 20 years of ANC rule, but many South Africans continue to find that racism plays a significant role in their society (Mangcu, 2003; Masombuka, 2014). We would therefore expect the effect of being interviewed by a noncoracial in South Africa to be particularly large.

Figure 5 presents results from South Africa for the same set of questions considered in the pooled, cross-country analysis presented in Figures 1 to 4.¹³ The black circles show the effect for Black respondents of being interviewed by a White interviewer (relative to being interviewed by a Black interviewer), and the white circles show the effect for White respondents of being interviewed by a Black interviewer (relative to being interviewed by a White interviewer).¹⁴ Effect sizes are as much as 10 times those reported in Figures 1 to 4—note the different scales—with the largest effects occurring for explicitly ethnic and political questions. For example, White respondents interviewed by Blacks are more than a half standard deviation more likely than those interviewed by Whites to say they identify as South African rather than as a member of an ethnic group. They are a full standard deviation more positive about their racial group’s comparative economic conditions and two thirds of a standard deviation more positive about the political influence their racial group wields. They also report substantially more political participation and interest in political affairs. Black respondents also demonstrate significant sensitivity to interviewer race, although often in a different direction than Whites—an asymmetry we will explore more thoroughly in the next section.

Figure 5.

Impacts of a noncoracial interviewer on survey responses in South Africa.

Although fewer outcomes in the socioeconomic category of questions were significantly affected by interviewer race, the findings among Blacks surprised us: In contrast to the results in the cross-country analysis, Black respondents interviewed by Whites were more likely to admit that they had gone without enough food or a cash income or had feared crime in their own home. We speculate that this result may be South Africa-specific, rooted in the desire by Black respondents to remind Whites that socioeconomic disparities persist even after the end of Apartheid. Similarly surprising is the fact that White respondents were more likely to admit to a Black than to a White interviewer that they knew someone who had died of AIDS. It may be that White respondents were trying to demonstrate awareness of, and sympathy for, what they view as an “African disease.”¹⁵

Finally, we find that interviewers evaluate coracial respondents differently than they do noncoracial respondents. Blacks interviewed by Whites were seen as more hostile, uncooperative, impatient, and suspicious than those interviewed by fellow Blacks, while Whites interviewed by Blacks were seen as more patient than Whites interviewed by fellow Whites. Being interviewed by a coracial thus affected not just the content of responses but also the tone of the interview in general.

In sum, we find potent race-of-interviewer effects for this highly politically relevant dyad. While the Black–White cleavage may represent an extreme in Africa, other groups with histories of deep political conflict like the Igbo and Hausa in Nigeria, the Baganda and Acholi in Uganda, or the Kikuyu and Luo in Kenya, are also likely to generate especially high levels of response bias. Therefore, while our analysis here almost certainly represents an upper bound, it nonetheless provides a sense of how much larger interviewer–respondent coethnicity effects can be when interviewers and respondents belong to groups with historically contentious relationships.

Figure 6 extends this analysis to the broader sample of Afrobarometer countries.¹⁶ Drawing on the coding of politically relevant groups provided in Posner (2004), we compare effect sizes in noncoethnic dyads where both the interviewer and the respondent are from politically relevant groups (Black circles) with effect sizes in noncoethnic dyads where either the interviewer or the respondent (or both) are from groups that are not politically relevant (White circles). In nearly every case, the effect is larger for the politically relevant dyads, with the strongest differences apparent for ethnic and political questions. Indeed, Figure 6 makes it clear that the findings reported earlier regarding the effect of a noncoethnic interviewer on a wide range of outcomes—from reports of one’s own group’s economic conditions and political influence to support for democracy, approval of the president’s performance, and trust in the ruling party—are all driven by the politically relevant dyads. Taken together, the results of these analyses confirm that the degree of historical conflict and competition between groups strongly shapes the impact of interviewer ethnicity on survey responses.

Figure 6.

Impacts of politically relevant and non-politically relevant noncoethnic interviewers on survey responses.

A different but potentially equally important source of variation across noncoethnic dyads lies in whether or not the respondent or interviewer is a coethnic of the leader of the country. As explained earlier, the close association in many African countries between the head of state and the ethnic group to which he belongs transforms evaluations of the leader (and answers to political questions more generally) into evaluations of the leader’s group. This can bias toward more positive assessments, either because the respondent seeks not to offend the interviewer who shares an ethnic tie with the leader or because, being a coethnic of the leader, the respondent seeks to bolster her own pride in her group by withholding criticism of the leader’s performance.

South Africa again provides a clear illustration. Given the strong association of the ruling party (ANC) with Blacks in the minds of most Whites (Ferree, 2011), Whites likely interpret questions about the ruling party, the president’s performance, the economic conditions of the country, and the treatment of their group by the government through a racial lens—as being an evaluation of Black rule in general. Even questions about the normative value of democratic government may reflect this dynamic, as democracy has been synonymous with ANC rule. Accordingly, many Whites may associate questions about participation in democracy as implicitly signaling their level of engagement with the new (Black) political dispensation. When interviewed by a Black interviewer, they may therefore respond by giving what they feel is the more positive answer to questions about government performance, support for democracy, and participation in the system. In contrast, Black respondents interviewed by Whites may feel the need to downplay evaluations of “their” party and underreport their engagement with democratic politics. We would therefore expect to see race-of-interviewer effects diverge for political questions depending on whether the respondent is Black or White.

This logic helps to explain the asymmetry we previously noted across Black and White respondents in Figure 5. In keeping with this interpretation, Whites interviewed by Blacks are much more positive about the performance of the president than Whites interviewed by Whites. In contrast, Blacks interviewed by Whites are much more negative about the president’s performance than Blacks interviewed by Blacks. Similar patterns hold for preference for democracy, trust in the ruling party, assessments of the country’s economic conditions, assessments of the government’s handling of the economy, and reported levels of political participation and interest in public affairs.

These results extend only modestly to our broader cross-country sample, in part because South Africa’s history makes it something of a unique case but also because the cross-country analysis pools results across dyads in which the bias may run in opposite directions. Figure 7a compares the size of the bias in our pooled sample between noncoethnic dyads in which the respondent is a coethnic of the head of state (black triangles) and is not (white triangles).¹⁷ While we do not see systematic differences for most questions, we do see statistically different patterns for questions asking about trust in the ruling party and the government’s handling of the economy. For these items, being a coethnic of the president pushes respondents toward systematically higher levels of reported trust in the ruling party and more positive assessments of the government’s performance.¹⁸

Figure 7.

Impacts of noncoethnic interviewers on responses to political questions when (a) Respondent is coethnic with head of state. (b) Interviewer is coethnic with head of state.

In Figure 7b, we explore how the effects of being interviewed by a noncoethnic change when the interviewer is a coethnic of the leader. Here, the clearest patterns are for questions about trust for opposition parties and (again) the government’s handling of the economy. Being interviewed by someone from a different ethnic group results in less reported trust for opposition parties when the interviewer shares an ethnicity with the leader.¹⁹ This is presumably because professing trust in opposition parties may be interpreted as professing a lack of support for the ruling party, which, through the interviewer’s ethnic tie with the leader, is likely to be associated with the interviewer. Being interviewed by a noncoethnic from the leader’s group is also associated with systematically higher evaluations of the government’s handling of the economy, almost certainly for similar reasons.

These results underscore that it matters not just whether interviewers and respondents are from different ethnic groups but whether those different groups have a history of political competition and conflict and whether either holds power in the country at the time of the interview. The practical implication for consumers of African survey data is that they need to be mindful of the potential for large response bias in some survey interactions, even if the average effect on response patterns in their broader sample is small. There is a deeper theoretical implication as well: If noncoethnic interviewer effects underscore the context dependence of the social interaction that constitutes the survey interview, then the results in Figures 5 to 7 remind us that the impact of being interviewed by a noncoethnic is itself dependent on other factors. Context operates at multiple levels and the answers respondents provide are potentially affected by all of them.

Implications for Extant Research Findings

Our analyses suggest that the ethnic match between an interviewer and respondent—a variable omitted in almost all prior research using African survey data—affects the answers respondents provide in surveys, especially when the interviewer and respondent are from politically relevant ethnic groups and/or when one of them is a coethnic of the head of state. The size of the bias we estimate is not always very large, but it is often statistically significant and it appears across a wide array of question types. A natural next question, then, is what these results imply for extant findings that draw on African survey data.

Because omitted variables create bias if and only if they are correlated with both treatment and outcome variables, the failure to control for the ethnic match between interviewers and respondents should only affect results under certain circumstances. For example, in an Afrobarometer-based study like Lieberman and McClendon’s (2013) analysis of how policy preferences vary along ethnic lines (in which both the main independent variable, ethnic group membership, and the main dependent variable, stated policy preferences, are correlated with whether or not the interviewer and respondent are from the same ethnic group) are likely to be biased by not including a control for interviewer–respondent coethnicity. However, studies like Nunn and Wantchekon’s (2011) analysis of the effect of exposure to the slave trade on contemporary levels of interpersonal trust (in which the outcome, trust, but not the main causal variable, exposure to the slave trade, is associated with whether the respondent was interviewed by a coethnic) are not likely to be biased by failing to control for the ethnic match between the interviewer and respondent. Indeed, when we replicate these two studies, we find that the inclusion of the interviewer–respondent coethnicity control does nothing to change Nunn and Wantchekon’s findings but does change—in fact, strengthens—the findings reported in Lieberman and McClendon (see Online Appendix Tables A9 and A10 for details of these replications). The lesson for researchers is that they should include the control for interviewer–respondent coethnicity if their independent and dependent variables of interest are likely to be correlated with whether or not the interviewer and respondent are from the same ethnic group. The results we report in Table 1 and Figures 1 to 4 provide a good starting point for identifying what these variables might be—although, as we have emphasized, these pooled results may mask significant associations that are present in particular countries or across particular interviewer–respondent dyads.

Conclusion

Ethnic group attachments are thought to affect the way people interact with one another in Africa. Combining data from the Afrobarometer with original data we have collected on the ethnic identity of Afrobarometer interviewers, we have shown that these effects extend to the interaction between interviewers and respondents during the course of administering public opinion surveys. In a sample of 14 African countries, we confirm that being interviewed by a noncoethnic is associated with statistically significant differences in the way people respond to survey questions. We find even stronger effects when we focus on politically salient noncoethnic dyads—both when we define political salience in terms of the history of competition and conflict between the interviewer and respondent’s ethnic communities and when we define it in terms of the ethnic group connection between the respondent or interviewer and the head of state. These findings accord strongly with (and, at least in the case of our results for the Black–White dyad in South Africa, are of the same magnitude as those found in) research on race-of-interviewer effects in the United States and can thus be viewed as an out-of-sample validation of that literature.

Our findings have important implications for survey work in Africa, where public opinion research is abundant and growing, and where dominant social cleavages tend to revolve around ethnicity. The findings are particularly significant given that one of the most common solutions to social desirability bias generated by interviewer effects in developed countries—the use of self-administered surveys (Krysan, 1998)—is often not feasible in Africa, where low rates of literacy necessitate the use of face-to-face interviews.

As social interactions, surveys are sensitive to the context in which they are administered. Our findings underscore the importance of ethnicity—and, in particular, the dyadic relationship between the ethnicity of the interviewer and the respondent—as one of the characteristics that defines that context in the African setting. Although our tests are necessarily limited to the way people respond to survey questions, our findings are strongly suggestive of the impact of shared (or divergent) ethnicity on beliefs, perceptions, and understandings of the world more broadly.

Like most treatments of survey bias in the literature, our discussion has proceeded from the starting point that respondents possess a “true” opinion and that the effect of being interviewed by a noncoethnic is to push responses away from that “truth.” However, it may be that there is no such “true” response apart from the context of the interview itself. For example, perhaps one does not report feeling more supportive of the president or more positive about one’s group’s economic conditions in response to a noncoethnic interviewer, but instead actually feels more positively about these things in a multiethnic context. In this case, the effect of a noncoethnic interviewer is not due to social desirability bias causing a respondent to report something other than his true opinion but instead due to a shift in opinions themselves that is caused by the social interaction with a noncoethnic. To the extent that this interpretation has merit, the implication is not simply that researchers should control for interviewer–respondent coethnicity but that they should think more fully about how ethnicity shapes the values, preferences, and perceptions of individuals in contexts where it is salient.

Footnotes

Author’s Note

Authors’ names are listed alphabetically to reflect their equal contributions.

Declaration of Conflicting Interests

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

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Claire L. Adida is assistant professor of political science at University of California, San Diego, and researches ethnic politics, African politics, and immigrant exclusion.

Karen E. Ferree is associate professor of political science at University of California, San Diego. Her research examines elections in emerging democracies, racial and ethnic politics in Africa, particularly South Africa, and survey design.

Daniel N. Posner is James S. Coleman Professor of International Development in the Department of Political Science at the University of California, Los Angeles. His research focuses on ethnic politics, research design, distributive politics and the political economy of development in Africa.

Amanda Lea Robinson is assistant professor of Political Science at Ohio State University. Her research focuses on the micro-foundations of national and ethnic identities, the political and social implications of ethnic geography, and the role of group identities in facilitating cooperation and trust, with a regional focus on Sub-Saharan Africa.

References

Adida

C. L.

(2015). Do African voters favor coethnics? Evidence from a survey experiment in Benin. Journal of Experimental Political Science, 2(1), 1-11.

Adida

C. L.

Ferree

K. E.

Posner

D. N.

Robinson

A. L.

(2015). Who’s asking? Interviewer coethnicity effects in African survey data (Unpublished manuscript). Afrobarometer Working Paper No. 158. Retrieved from http://afrobarometer.org/publications/wp158-whos-asking-interviewer-coethnicity-effects-african-survey-data

Asiwaju

A. I.

(1985). Partitioned Africans: Ethnic relations across Africa’s international boundaries, 1884-1984. New York, NY: St. Martin’s Press.

Asunka

(2015, November 19-22). Beyond ethnicity: Exploring other drivers of response bias in African survey data. Paper presented at the African Studies Association Conference, San Diego, CA.

Bates

R. H.

(1983). Modernization, ethnic competition, and the rationality of politics in contemporary Africa. In Rothchild

Olorunsola

V. A.

(Eds.), State versus ethnic claims: Africa Policy dilemma (pp. 152-171). Boulder, CO: Westview Press.

Berinsky

(2004). Can we talk? Self-presentation and the survey response. Political Psychology, 25(4), 643-659.

Bilig

(1976). Social psychology and intergroup relations. London, England: Academic Press.

Blaydes

Gillum

(2013). Religiosity-of-interviewer effects: Assessing the impact of veiled enumerators on survey response in Egypt. Politics & Religion, 6(3), 459-485.

Brady

(2000). Contributions of survey research to political science. Political Science and Politics, 33(1), 47-57.

10.

Burns

(2012). Racial stereotypes, stigma and trust in post-apartheid South Africa. Economic Modeling, 23(5), 805-821.

11.

Campbell

B. A.

(1981). Race-of-interviewer effects among Southern adolescents. Public Opinion Quarterly, 45(2), 231-244.

12.

Carlson

(2015). Ethnic voting and accountability in Africa: A choice experiment in Uganda. World Politics, 67(2), 353-385.

13.

Conover

(1984). The influence of group identifications on political perception and evaluation. Journal of Politics, 46(3), 760-785.

14.

Conroy-Krutz

(2013). Information and ethnic politics in Africa. British Journal of Political Science, 43(2), 345-373.

15.

Converse

J. M.

Schuman

(1974). Conversations at random: Survey research as interviewers see it. New York, NY: John Wiley.

16.

Cotter

P. R.

Cohen

Coulter

P. B.

(1982). Race-of-interviewer effects in telephone interviews. Public Opinion Quarterly, 46(2), 278-284.

17.

Davis

D. W.

(1997). Measurement error and race of interviewer effects among African Americans. Public Opinion Quarterly, 61(1), 183-207.

18.

Davis

D. W.

Silver

B. D.

(2003). Stereotype threat and race of interviewer effects in a survey on political knowledge. American Journal of Political Science, 47(1), 33-45.

19.

Dionne

K. Y.

(2014). The politics of local research production: Surveying in a context of ethnic competition. Politics, Groups, and Identities, 2(3), 459-480.

20.

Eifert

Miguel

Posner

D. N.

(2010). Political competition and ethnic identification in Africa. American Journal of Political Science, 54(2), 494-510.

21.

Englebert

Tarango

Carter

(2002). Dismemberment and suffocation. Comparative Political Studies, 35(10), 1093-1118.

22.

Ferree

K. E.

(2006). Explaining South Africa’s racial census. Journal of Politics, 68(4), 803-815.

23.

Ferree

K. E.

(2011). Framing the race in South Africa: The political origins of racial census elections. New York, NY: Cambridge University Press.

24.

Franck

Rainer

(2010). Does the leader’s ethnicity matter? Ethnic favoritism, education and health in sub-Saharan Africa. American Political Science Review, 106(2), 294-325.

25.

Habyarimana

Humphreys

Posner

D. N.

Weinstein

(2009). Coethnicity: Diversity and the dilemmas of collective action. New York, NY: Russell Sage.

26.

Hatchett

Schuman

(1975). White respondents and race-of-interviewer effects. Public Opinion Quarterly, 39(4), 523-528.

27.

Holbrook

Green

Krosnick

(2003). Telephone versus face-to-face interviewing of national probability samples with long questionnaires. Public Opinion Quarterly, 67(1), 79-125.

28.

Horowitz

D. L.

(1985). Ethnic groups in conflict. Berkeley: University of California Press.

29.

Hyman

Cobb

Feldman

Hart

Stember

(1954). Interviewing in social research. Chicago, IL: The University of Chicago Press.

30.

Jackman

R. W.

(1978). The predictability of coups d’Etat: A model with African data. American Political Science Review, 72(4), 1262-1275.

31.

Jeon

Johnson

Robinson

(2013, October 17-18). Nationalism and social sanctioning across ethnic lines: Experimental evidence from the Kenya-Tanzania border. Paper presented at the Midwest Group in African Political Economy Workshop, Bloomington, IN.

32.

Juma

(2012, November 27). How tribalism stunts African democracy. BBC News. Retrieved from http://www.bbc.com/news/world-africa-20465752

33.

Kasara

(2007). Tax me if you can: Ethnic geography, democracy, and the taxation of agriculture in Africa. American Political Science Review, 101(1), 159-172.

34.

Kasara

(2013). Separate and suspicious: Local social and political context and ethnic tolerance in Kenya. Journal of Politics, 75(4), 921-936.

35.

Krosnick

J. A.

Alwin

D. F.

(1987). An evaluation of a cognitive theory of response-order effects in survey measurement. Public Opinion Quarterly, 51(2), 201-219.

36.

Krysan

(1998). Privacy and the expression of white racial attitudes: A comparison across three contexts. Public Opinion Quarterly, 62(4), 506-544.

37.

Laitin

D. D.

(1986). Hegemony and culture: Politics and religious change among the Yoruba. Chicago, IL: The University of Chicago Press.

38.

Lee

Pérez

E. O.

(2014). The persistent connection between language-of-interview and Latino political opinion. Political Behavior, 36(2), 401-425.

39.

Leibbrandt

Finn

Woolard

(2012). Describing and decomposing post-apartheid income inequality in South Africa. Development Southern Africa, 29(1), 19-34.

40.

Leininger

(2014). A strong norm for democratic governance in Africa. Stockholm, Sweden: International Institute for Democracy and Electoral Assistance.

41.

Lieberman

E. S.

(2003). Race and regionalism in the politics of taxation in Brazil and South Africa. New York, NY: Cambridge University Press.

42.

Lieberman

E. S.

(2009). Boundaries of contagion: How ethnic politics have shaped government responses to AIDS. Princeton, NJ: Princeton University Press.

43.

Lieberman

E. S.

McClendon

(2013). The ethnicity–policy preference link in sub-Saharan Africa. Comparative Political Studies, 46(5), 574-602.

44.

Long

J. D.

Hoffman

(2013). Party attributes, performance, and voting in Africa. Comparative Politics, 45(1), 127-146.

45.

Mangcu

(2003). The state of race relations in post-apartheid South Africa. In Daniel

Habib

Southall

(Eds.), State of the nation: South Africa, 2003-2004 (pp. 105-117). Cape Town, South Africa: Human Sciences Research Council Press.

46.

Masombuka

(2014, February 24). Racism in SA alive and kicking. Times Live. Retrieved from http://www.timeslive.co.za/news/2014/02/24/racism-in-sa-alive-and-kicking

47.

Matsiko

(2014, September 1). The monster of tribalism. The Independent. Retrieved from http://www.independent.co.ug/cover-story/9283-the-monster-of-tribalism

48.

Nunn

(2008). The long-term effects of Africa’s slave trade. Quarterly Journal of Economics, 123(1), 139-176.

49.

Nunn

Wantchekon

(2011). The slave trade and the origins of mistrust in Africa. American Economic Review, 101(7), 3221-3252.

50.

Pérez

E. O.

(2009). Lost in translation? Item validity in bilingual political surveys. Journal of Politics, 71(4), 1530-1548.

51.

Piot

(1996). Of slaves and the gift: Kabre sale of kin during the era of the slave trade. Journal of African History, 37(1), 31-49.

52.

Posner

D. N.

(2004). Measuring ethnic fractionalization in Africa. American Journal of Political Science, 48(4), 849-863.

53.

Rankin

W. W.

Brennan

Schell

Laviwa

Rankin

S. H.

(2005). The stigma of being HIV-positive in Africa. PLoS Medicine, 2(8), e247

54.

Robinson

A. L.

(2016). Nationalism and ethnic-based trust: Evidence from an African border region. Comparative Political Studies. Advance online publication. doi:10.1177/0010414016628269

55.

Samii

(2013). Perils or promise of ethnic integration: Evidence from a hard case in Burundi. American Political Science Review, 107(3), 558-573.

56.

Seekings

Nattrass

(2002). Class, distribution and redistribution in post-apartheid South Africa. Transformation, 50(1), 1-30.

57.

Thompson

(1990). A history of South Africa. New Haven, CY: Yale University Press.

58.

Weeks

Moore

R. P.

(1981). Ethnicity-of-interviewer effects on ethnic respondents. Public Opinion Quarterly, 45(2), 245-249.

59.

Weinreb

A. A.

(2006). The limitations of stranger-interviews in rural Kenya. American Sociological Review, 71(6), 1014-1039.