In diverse societies, do citizens receive equal treatment from their political representatives? Racial and ethnic diversity has frequently been identified as a potential impediment to good governance (Alesina et al., Reference Alesina, Baqir and Easterly1999; Easterly and Levine, Reference Easterly and Levine1997; Lieberman and McClendon, Reference Lieberman and McClendon2013; Miguel, Reference Miguel2004; Montalvo and Reynal-Querol, Reference Montalvo and Reynal-Querol2005). One of the ways such diversity may impede good governance is by shaping the day-to-day behavior of politicians responsible for responding to constituent needs. Politicians may seek to favor members of their own group and to discriminate against members of out-groups (Broockman, Reference Broockman2013; Butler, Reference Butler2014; Faller et al., Reference Faller, Nathan and White2015; Franck and Rainer, Reference Franck and Rainer2012; Kramon and Posner, Reference Kramon and Posner2012), thereby undermining principles of political equality (Dahl, Reference Dahl2006; Verba, Reference Verba2003).
In this article, I describe an adaptation of Butler and Broockman’s (Reference Butler and Broockman2011) experimental study of political discrimination in the U.S. to the South African context. Butler and Broockman (Reference Butler and Broockman2011) find that, on average, U.S. state legislators discriminate against black constituent requests for help with voter registration. Once broken down by legislator race, however, the authors find that white legislators are more responsive to white than to black constituents, while minority legislators are more responsive to black than to white constituents. In other words, the authors find evidence that politicians exhibit same-race favoritism in constituency service.Footnote 1
There are reasons to believe that the same results would hold in South Africa. South Africa shares with the U.S. a history of black–white racial segregation and subjugation. Indeed, South Africa’s racist political system was not formally dismantled until the 1990s, and South Africa’s apartheid policies were, arguably, the most extreme forms of racial segregation in the world (Marx, Reference Marx1998; Seidman, Reference Seidman1999). If political systems leave a lasting impact on racial categorization and out-group prejudices (Jackman, Reference Jackman1994; Krysan, Reference Krysan2000; Lieberman and Singh, Reference Lieberman and Singh2012) and if politicians carry personal prejudices with them into office, we might expect to find racial bias in politician behavior in South Africa.
On the other hand, one might not expect the results to travel. Racial bias is a sensitive subject in post-apartheid South Africa, precisely because of the recent memory of the transition to full democracy (Vincent and Howell, Reference Vincent and Howell2014). Most parties seem eager to demonstrate egalitarian norms (Friedman, Reference Friedman2002); this eagerness seems to be particularly the case within the Democratic Alliance (DA)—an opposition party sometimes perceived to have ties to the apartheid regime (Southern, Reference Southern2011; Democratic Alliance, 2013)—but “non-racialism” is a formal commitment of the dominant African National Congress (ANC) as well (Anciano, Reference Anciano2014). Although politicians may have their own prejudices, they might try to check them for the purposes of political appeal. Ordinary citizens in South Africa exhibit much lower in-group racial bias than might be assumed, given the country’s history (Hofmeyer and Burns Reference Hofmeyr and Burns2012; Van Der Merwe and Burns, Reference Van Der Merwe and Burns2008; c.f. Gibson and Claasen Reference Gibson and Claassen2010). Parties and politicians might risk their own popularity by obviously discriminating among constituents on the basis of race.
Unlike the U.S., South Africa also has a largely party-rather than candidate-centered system. While there is variation in electoral rules at the local level, with some local politicians elected through candidate-centered contests and some through closed list, party-centered ones, in practice, all local politicians who want to advance within national structures have incentives to focus on hobnobbing with higher-ups and catering to broader party bases rather than on cultivating the personal vote among local constituents. In this context, one might expect to observe constituency case work receiving little attention across the board, rather than to observe politicians selectively targeting some constituents with help and favor over others (André et al., Reference André, Depauw, Shugart, Martin, Saalfeld and Strom2013; Barkan and Mattes, Reference Barkan and Mattes2014; Lundberg, Reference Lundberg2007).Footnote 2 Furthermore, in South Africa, race and partisanship are tightly correlated. On average, white South Africans are more likely than black South Africans to vote for the DA, an opposition party, whereas black South Africans are more likely than white South Africans to support the dominant ANC (Ferree, Reference Ferree2006; Garcia-Rivera, Reference Garcia-Rivero2006). Even if we were to observe discrimination in constituency service on the basis of race, it might be due to politicians’ using race as a proxy for party support, rather than to racial prejudice per se.
No research has yet been done on whether, in a post-apartheid era, racial bias persists among South African political elites. I thus conducted an experiment involving 1,229 black and white politicians and tested whether they were more responsive to constituents of the same race group who submitted queries about public goods provision than to constituents of other race groups who did the same. In some conditions, politicians also received separate signals about the constituents’ partisan preferences.
The study reveals two key results. First, on average, local South African politicians—both black and white, both plurality- and proportionally-elected, belonging to both ruling and opposition parties—were more responsive to public goods inquiries from same-race, than from other-race, constituents, even when partisanship cues were provided separately from racial cues. In other words, findings of same-race favoritism from the U.S. extend to other issue areas and to a developing democracy with a different institutional environment, albeit one that shares a similar history of racial subjugation. Second, rates of responsiveness are very low even for same-race constituents. This latter finding suggests that achieving political equality requires addressing absolute, and not just relative, responsiveness to constituent needs.
THE EXPERIMENT
In July 2011, emails were sent to the Xhosa, Zulu, Afrikaaner, and English local councilors in the municipalities of four South African provinces (N=1,936). In South Africa, language groups are nested within race groups, and these four categories represent the largest and most politically salient linguistic categories in South Africa that nest within white and black race groups (Cederman et al., Reference Cederman, Wimmer and Min2009; Posner, Reference Posner2004). Because there is evidence that South Africans have trouble with cross-linguistic, within-race identification (Harris and Findley, Reference Harris and Findley2014), I implemented treatments at the level of race. Each email appeared to the councilor to be from a constituent living in her district and raised a concern about public goods provision.Footnote 3
Before conducting the experiment, I had the opportunity to observe municipal councilors from both the dominant ANC and the opposition DA parties in 2009, 2010, and 2012. I observed councilors receiving multiple communications a day from known and unknown constituents via email, phone, in-person visits, and websites. Councilors widely acknowledged that their richer and well-educated constituents (of any race group) contact them via email and websites.
In theory, it is a primary responsibility of municipal councilors to communicate directly with constituents. This is particularly true for ward councilors, who are elected in single-member, candidate-centered, plurality electoral contests within wards. In practice, however, behavioral differences between ward and proportional representation (PR) councilors often wash out. South Africa as a whole has a strong party system, and all contests except for ward elections are based on proportional representation and closed lists. Local councilors of all types thus have incentives to focus their time on networking within party structures and on building a reputation beyond the confines of their districts if they want to advance their careers and earn a place on local, regional or national lists.Footnote 4 National structures of both the ANC and the opposition parties closely control which candidates appear on party lists. They also control which candidates run in which wards, such that even politicians interested in remaining only ward councilors have to gain the attention of higher-ups. The main constituent-relations criterion for intra-party selection and promotion in both the dominant ANCFootnote 5 and the main opposition party, the DA,Footnote 6 is often not constituency-service per se but instead the ability of councilors to promote the party.
The result is that many (though certainly not all) local councilors spend significant amounts of time in party caucuses and in events to promote the party brand. In the 2008 Afrobarometer, South Africans were asked how often they have contacted their local councilors in the past year. A little over one quarter (27.5%) answered, “at least once”—a larger percentage than for any other authority mentioned. Nevertheless, only 16% of respondents who reported having contacted their local councilor thought local councilors “always” try their best to listen to people like them. The experimental results in this study support these perceptions of low levels of responsiveness overall.
I first measured responsiveness dichotomously, according to whether each councilor had replied or not two months after the emails were sent. Any return contact counted as a reply, including if (1) the councilor replied directly to the constituent with the requested information, (2) the councilor replied to ask the constituent for more information about his query, (3) the councilor forwarded the email to a bureaucrat, cc-ing the constituent, and (4) a bureaucrat responded directly to the constituent, having received a forwarded email from the politician to which the constituent alias was not originally cc-ed.Footnote 7 Additionally, in order to gauge the effort politicians put into replying, I also coded “answered” as a one, if the politician supplied the requested information directly or provided the contact information for the bureaucrat through a carbon copy. “Answered” was coded zero if the politician did not reply or replied only to ask for more information. (See Butler, Reference Butler2014 for a similar approach.) The “answered” measure can be somewhat unfair to councilors who, in asking for more information, were actually trying to be helpful and precise. I thus privilege the first measure of responsiveness in my discussion of the results.
The main treatment was a same-race/other-race treatment. Black (White) councilors assigned to the other-race condition thus received an email from a white (black) male alias. See Figure 1. Gmail aliases contained names signaling race.
Same/Other Race Treatment Assignment, Given Each Councilor’s Ethnic Group.
From : [ntobeko.yonzi@gmail.com, modise.sipho@gmail.com, walker.ianandrew@gmail.com or kobus.duplessis38@gmail.com]Footnote 8
To: Politician’s Email Address
Subject: Question about [Roads/Water]
Councillor,
My name is [Treatment Name], and I am troubled by the condition of [the roads/the water] in this municipality. [While I am a supporter of (politician’s party),] I think this should be a priority for the government to fix. Can you tell me the name of an administrator I should speak to about this problem and also how I can contact that administrator or department?
Thank you,
[Treatment Name] Footnote 9
Municipality/Ward Footnote 10
I took two steps to ensure that councilors were receiving credible treatments. First, white councilors assigned to the other-race condition saw an alias from the predominant black linguist group in their geographic area. For councilors in the Western Cape and the Eastern Cape provinces, this meant a Xhosa alias; for councilors in Gauteng and KwaZulu Natal (KZN) provinces, this meant a Zulu alias. All black councilors in the other-race condition received an email from the white, Afrikaaner alias because Afrikaaners make up the larger portion of the white population in the districts in the sample. Second, I tried my best to exclude councilors in whose municipality either whites or blacks made up less than 1% of the population or where less than 1% of the households in the municipality has access to internet and a computer in the home.Footnote 11
As a second treatment, adapted from Butler’s and Broockman’s design, I varied whether the constituent indicated that he was the politician’s co-partisan or not. South Africa has a dominant party (the ANC) and several smaller parties. The most powerful of those parties is currently the DA, but the ANC and the DA together control less than 85% of the national vote share and their respective shares, relative to other parties, vary considerable across space. In this context, there is not a single, obvious “out”-party treatment. In order to reduce noise, the treatments thus consisted either of a clear co-partisan signal or of no partisan signal. Were politicians to discriminate against other-race constituents simply because they assume those constituents would never support their party, the clear co-partisan signal should diminish any difference in responsiveness between same-race and other-race constituents.
Each email raised a specific public goods concern about the quality of either roads or water in the councilor’s district and asked the councilor for specific information about how to have the problem addressed. Both roads and water are issues over which local councilors have some jurisdiction and are high-priority issues for average South Africans (South Africa Afrobarometer, 2011). Using two different public goods meant both that I could test for racial favoritism outside of the specific domain of voter registration used in previous studies and that the results are not limited to one specific public goods issue (Kramon and Posner, Reference Kramon and Posner2013).
Together, the same-race, the co-partisan and the roads/water treatments formed a 2 × 2 × 2 design. I randomly assigned councilors to the eight treatment conditions, blocking on whether the councilor is a member of the ANC or not, on whether the councilor is ward or PR, and on the racial fractionalization of the councilor’s district.Footnote 12
The councilors’ ethnic categories were coded based on their surnames and first names. Where photographs of the councilors were available on the websites, I used these photographs to double-check the race of the councilor. I also created a dummy variable for uncertainty of ethnic coding, which was coded 1 if a photograph was not available but either the councilors surname or first name was Afrikaans. The latter was important because many colored South Africans have Afrikaans names and cannot be distinguished from a white Afrikaaner without a photograph. The variable, “uncertain” was also coded as 1 if the councilors first name and surname signaled membership in two different ethnic groups. The results are robust if I drop the subjects coded as uncertain. The experiment treats the email addresses of the councilors, not the councilors themselves, although instances of councilors with other-race staff should undermine any average treatment effect of being assigned to the same-race condition.
The use of email is not as obvious a choice in the South African context as in the U.S., but it allowed me to obtain some control over socioeconomic class. Between-group inequality is high in South Africa (Baldwin and Huber, Reference Baldwin and Huber2010); a councilor might assume that a black South African constituent is poorer and less well-educated than a white South African constituent unless given other indications. Email usage is rapidly growing in South Africa among individuals of both race groups, but it is still segmented by income and education. For this reason, constituents of both race groups who use email to contact their councilors should be close in socioeconomic profile even while differing in terms of race. Using email also reduces the burden placed on politicians by the experiment. Other ethical considerations given to the design are described at the end of the article in an appendix.
Finally, even though South Africa has 11 official languages, I used English in order to ensure uniform wording across conditions and to ensure that any racial bias was not due to councilors’ more easily reading emails from same-race (and same language) constituents than emails from other-race constituents. The use of grammatically correct English, like the use of email, is also intended to signal that the constituents, regardless of race, are socioeconomically similar. If a councilor receiving an email from a same-race constituent not written in their shared language regarded that constituent as less of a co-ethnic, this should have attenuated any effect of “same-race” on responsiveness.
RESULTS
There were clear same-race effects on the rate of response in the experiment. Table 1 shows the difference in the rate of response depending on whether the constituent was of the same race group as the councilor or a member of another race group. Across all councilors, 25.2% of the same-race constituents received a response, while only 16.8% of the other-race constituents did: a difference of 8.4 percentage points. Table A2 in the online supplemental materials shows differences in answer rates. On average, 15.6% of same-race constituents received a complete answer to their question, while only 10.4% of other-race constituents did: a difference of 5.2 percentage points, also statistically significant (p = 0.004).
Do Same-Race Constituents Receive More Responses Than Other-Race Constituents? Yes
P-values are from one-sided t-tests with unequal variances.
Racial bias in responsiveness persisted no matter the ethnic category of the councilor. In other words, as in Butler and Broockman’s study, this study provides evidence of in-group bias across the board, not solely of discrimination against the historically subordinate group (i.e. black South Africans or Black Americans). The bias among black politicians appears slightly smaller than among white politicians in magnitude, but as a percentage decrease from the level of responsiveness to same-race constituents, the bias among black councilors (a decrease of 34.9% among Zulus and 43.8% among Xhosas) is about the same or slightly larger than that among white councilors (a decrease of 27.5% among the English and 27.4% among Afrikaaners). These estimates of racial bias remain significant using a 90% confidence interval, even after correcting for multiple hypothesis testing, and are robust to a number of additional councilor and district characteristics.Footnote 13 All categories of councilors also exhibit racial bias in the rates at which they provided complete answers to constituent questions, with the exception of white Afrikaaner councilors (Table A2). White Afrikaaner councilors were still less likely to respond to other-race constituents than to same-race ones, but, when they did respond to either category, they more often asked for additional information from same-race constituents, perhaps in an effort to be more helpful.Footnote 14
Bolstering my confidence that the results are not due to the other-race treatment being incredible, evidence of racial bias was not driven by councilors in districts with low percentages of other-race constituents. See Table A5a in the online supplemental materials.Footnote 15
In addition, consistent with previous studies in the U.S., the co-partisanship treatment did not attenuate racial bias in the full sample. The overall difference in the rate of response to same-race and other-race constituents did not diminish when both types of constituents expressed explicit support for the councilor’s party. In the full sample of black and white councilors, when constituents did not mention any partisan affiliation, the difference in response rates between same-race and other-race constituents among black and white councilors was 7.6% points. When the constituents indicated that they were supporters of the councilors’ parties, the estimated difference in response rates was 9.2% points. The difference between these two estimates of racial bias was not statistically significant. See Table 2.
Racial Bias Toward Co-partisans and by Party/Role
P-values are from one-sided t-tests allowing for unequal variance.
Racial bias persisted in both the dominant party and in the main opposition party, and across both PR and ward councilors (Table 2). While the magnitude of racial bias in the dominant party appears larger, a two-way ANOVA test did not allow me to reject the null that racial bias is the same size in both parties.Footnote 16
Although beyond the purview of experimental inference, the overall response rates are also worth examining. Among all councilors, the average response rate to the emails was only 20.9%, and the average “answer” rate was only 12.9%. These rates contrast with the higher overall response rates observed in similar experiments in the U.S. (56.5% in Butler and Broockman, Reference Butler and Broockman2011; 71% in Faller et al., Reference Faller, Nathan and White2015) and are more similar to but still lower than response rates in China (37.2% in Distelhorst and Hou, Reference Distelhorst and Hou2014). In this experiment, white English-speaking councilors, who are likely to have the best access to email and for whom English is their home language, responded at the highest rate. But even in the same-race treatment condition, these councilors responded to well less than half of these unburdensome queries. The results are in keeping with the perceptions of nationally representative samples of South Africans, less than 20% of whom think that local politicians pay attention to their concerns when they try to contact them through any means. Although the discourse on local government performance has often portrayed local councilors as being dismissive specifically of poor South Africans (Alexander, Reference Alexander2010), the results of this experiment suggest that South African local councilors are also dismissive of constituents who, because of the medium of communication, are likely to be of higher income and well-educated.
CONCLUSION
South African local politicians—both black and white, both ward and PR, belonging to both the dominant and main opposition parties—privileged constituents of their own race group. They did so when responding to constituents of similar-seeming socioeconomic standing and when both same-race and other-race constituents were co-partisans, indicating that the bias was indeed due to perceived race, and not to other closely-correlated constituent characteristics. While there are without a doubt individual South African politicians who make equal efforts to reach out to all constituents, these findings reveal an on average pattern similar to that found in studies of U.S. politicians.
At the outset there were some reasons to expect limited racial discrimination in constituency service in the South African context. Precisely because of the recency and extremity of apartheid, racial bias is a sensitive subject in South Africa, and democratically elected politicians might be expected to go out of their way not to exhibit it. Constituency service is also arguably less important to South African politicians than it is to politicians in more candidate-centered electoral systems or to those in systems with significant intra-party electoral competition. South African politicians have fewer strategic reasons to respond in detail to anyone, much less to favor some constituents over others.
Yet, consistent with the prediction that political systems like apartheid leave a lasting impact on in- and out-group prejudices, and consistent with the notion that many politicians carry personal prejudices into office, racial bias in constituency service was evident among all types of local politicians. It is difficult to rule out the possibility that politicians have no strategic incentive to engage in racial bias, though other studies (e.g. Broockman, Reference Broockman2013; Distelhorst and Hou, Reference Distelhorst and Hou2014) have found evidence of non-strategic bias among politicians in the U.S. and in China. Future experimental studies in South Africa could directly manipulate the strength of strategic incentives by, for instance, randomly assigning whether constituent requests are sent just before or after elections, or by randomly including in requests from other-race constituents the mention of political connections within the politician’s party.Footnote 17 Nevertheless, that racial bias in responsiveness persisted in this study across a wide range of councilor types and in the presence of partisan cues suggests personal, rather than strategic, prejudice in the absence of further investigation.
There is some evidence that simply revealing patterns of bias can reduce discrimination (Pope et al., Reference Pope, Price and Wolfers2013), so perhaps simply documenting quotidian biases can bring change. Yet, it should be underscored again that observational rates of responsiveness in this study were low across the board. The low response rates may simply suggest that South African local councilors need more reliable access to email. But the finding is also in keeping with patterns in nationally representative surveys: South Africans do not generally perceive their local representatives to be likely to listen to people like them. Future research should thus also focus on incentives and technological resources to increase rates of responsiveness in absolute terms. Improving the quality of democratic governance requires both a reduction in in-group bias and higher levels of responsiveness to everyone.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/XPS.2015.10
APPENDIX: ETHICAL CONSIDERATIONS
The experiment was approved through a full Institutional Review Board (IRB) review,Footnote 18 but several ethical considerations merit mention. This study joins others that have used mild forms of deception and waived informed consent in order to test for racial biases in government, academia, the housing market, and in the workplace (e.g. Bertrand and Mulainathan, Reference Bertrand and Mullainathan2004; Broockman, Reference Broockman2013; Butler and Broockman, Reference Butler and Broockman2011; Pager and Quivallian, Reference Pager and Quillian2005). The decision to waive informed consent and to use fictitious constituents is not undertaken lightly here or in those studies. Whether politicians engage in discrimination based on race or ethnicity is not simply theoretically interesting. It is normatively and practically important for achieving political equality. South African citizens deserve to have accurate information about whether their representatives engage in biased behavior. Diverse democracies more generally can benefit from understanding whether there are contexts under which ethnic favoritism is reduced or eliminated.
Yet, it is impossible to randomly assign race and other background constituent characteristics to elected officials or to prospective employers while adequately controlling for other factors (the content of the communication, partisanship, etc.) without engaging in some degree of deception—with fictitious email addresses (Butler and Broockman, Reference Butler and Broockman2011; Distelhorst and Hou, Reference Distelhorst and Hou2014), fictitious résumés (Bertrand and Mulainathan, Reference Bertrand and Mullainathan2004), trained actors (Pager and Quivallian, Reference Pager and Quillian2005) or confederates (Butler et al., Reference Butler, Karpowitz and Pope2012). Because racial discrimination is socially undesirable behavior, politicians are also less likely to engage in it if they are informed that they are being studied even if they regularly practice discrimination in their official capacities under non-research conditions. Using fictitious constituents allows studies like this one to draw valid inferences about socially important questions.
In addition, although elected politicians acting in their official capacities are not naive subjects and often fall outside of IRB review, steps were also taken here to minimize additional risks to the politicians themselves. I ensured that all responses remain confidential so as to protect politicians’ individual reputations. I chose a between-subjects design so that the results of the study could not be used to draw inferences about individuals but only to detect racial bias on average. I also tried to minimize burdens on the politicians’ time by using email rather than some other form of communication. To respond, politicians needed only ask for more detail or provide a bureaucrat’s contact information. The politician could simply ignore or delete the email.
Because the observed behavior here is part of politicians’ official duties, is not time-consuming and yet sheds light on important normative issues, there is little evidence that politicians in South Africa or elsewhere have thus far reacted negatively to these kinds of studies, and the hope is that they will continue to see their value. Since the experiment reveals evidence both of same-race bias on average and of low response rates overall, any behavioral adjustment on the part of politicians who learn about the experiment should, if anything, be to increase response rates in case they are being studied. Nevertheless, these types of designs should continue to be scrutinized (e.g., McClendon, Reference McClendon2012; Riach and Rich, Reference Riach and Rich2004), and researchers should use them with caution.
