Interpersonal trust, when sufficiently generalized, reduces viscosity within vital democratic, economic, and social institutions. Yet citizens may discriminate trust—that is, trust some individuals more than others—on the basis of stereotypes associated with social and even political identities. Indeed, in many democracies co-partisans trust each other more than they trust rival partisans, and this partisan trust discrimination is exacerbated by fierce and salient competition over resources (Carlin and Love Reference Carlin and Love2018; Michelitch Reference Michelitch2015). But do rival partisans discriminate trust in polities where one party dominates electoral competition?
We address this question with two studiesFootnote 1 in South Africa, where the African National Congress (ANC) has won between 60% and 70% of the vote in the five elections from the end of Apartheid in 1994 to 2014, with margins of victory ranging from 40% to 57%. Consistent with our expectations, opposition party partisans of the Democratic Alliance (DA) exhibit strong trust discrimination characterized more by out-group derogation against the dominant ANC partisans than in-group favoritism toward fellow DA partisans. ANC partisans do not discriminate trust, i.e. they tend to trust ANC and DA partisans equally. This asymmetry is not caused by the racial identities nested in partisanship but by partisanship itself.
Competition, Identity and Trust
While prejudice, hostility, and discrimination are well documented across other types of groups, their existence across party lines or “partyism” (Sunstein Reference Sunstein2014) has only recently drawn scholarly attention. Two conclusions from this research are critical.
First, political competition exacerbates perceptions of in-group moral superiority and out-group threat and fear, which can, in turn, fuel in-group favoritism as well as out-group derogation (Brewer Reference Brewer1999). To wit, Carlin and Love (Reference Carlin and Love2018) find positive correlations between partisan trust gaps and perceived party polarization and more consistent evidence of rival-partisan derogation than co-partisan favoritism. Party polarization in the United States is considered a scope condition for out-partisan derogation in trust (Carlin and Love Reference Carlin and Love2013), economic decisions (McConnell et al. Reference McConnell, Margalit, Malhotra and Levendusky2018), or both (Iyengar and Westwood Reference Iyengar and Westwood2015). Michelitch (Reference Michelitch2015) shows that partisanship influences market-price bargaining net of ethnicity—but only at election time—in Ghana. Political competition, thus, creates amity toward co-partisans and enmity between rival partisans.
Second, partisanship acts as a superordinate identity into which corresponding social identities nest—especially where political cleavages reinforce social cleavages and when competition raises partisanship’s salience (Carlin and Love Reference Carlin and Love2018; Michelitch Reference Michelitch2015). Partisanship predicts more robust trust discrimination than the social identities undergirding competitive party systems (Carlin and Love Reference Carlin and Love2018; Iyengar and Westwood Reference Iyengar and Westwood2015; Martini and Torcal Reference Martini and Torcal2019; Westwood et al. Reference Westwood, Walgrave, Leonisio, Miller and Stribjis2018). Cross-cutting cleavages tend to lower intergroup trust discrimination (Bossuroy and Selway Reference Bossuroy and Selway2011), and partisanship nests in a national superordinate identity (Carlin and Love Reference Carlin and Love2018; Levendusky Reference Levendusky2018).
These two conclusions shape our study. The first sets up our theoretical critique. The second informs our research design and empirical analyses.
Theoretical Expectations
Based on the foregoing discussion, we levy one critique and posit one expectation regarding the influence of political competition on interpartisan trust. Our critique is that mutual distrust of rival partisans implies all partisans perceive roughly equal degrees of political competition. But beyond robust multiparty democracies, opposition and ruling-party identifiers may (correctly) perceive political competitiveness much differently. That is, some contexts may lack the main requirement for rival-partisan derogation—a shared sense of political competition.
This begs for additional theorizing. We argue large electoral margins lower the salience of political competition. Hence where elections are less competitive, ruling-party supporters feel less threatened by, afraid of, and morally superior to opposition partisans. For opposition partisans, the opposite is true. Indeed, in less-competitive contexts—the modal regime in Africa—opposition success depends on social networks to overcome coordination problems that underlie supporters’ beliefs about opposition viability (Weghorst Reference Weghorst2018). This should heighten the salience of competition and, in turn, the animus opposition partisans feel toward supporters of the ruling party. Electoral domination should, thus, make rival-partisan distrust asymmetrical—stronger among opposition partisans than among dominant partisans. Put in different terms, perpetual losers should be more politically distrustful of their partisan rivals than perennial winners are of them. We test this expectation in two studies in South Africa.
Study Area
South Africa’s European settlers constructed the Apartheid regime (1948–1994), which systematically restricted the political, social, and economic rights of the black African majority and non-white minorities. Against this backdrop, the democratic transition culminated in an election that pitted a mostly black ANC, led by Nelson Mandela, against the white National Party, Apartheid’s architects and curators. An era of undisturbed ANC dominance ensued, approximating a predominant party system, defined as “pluralism in which—even though no alternation in office actually occurs—alternation is not ruled out” (Sartori Reference Sartori1976, 200).
Race remains nested within partisanship in South Africa (see Harding and Michelitch Reference Harding and Michelitch2018), and racial identities fuel group threat, intolerance, negative affect (Gibson and Gouws Reference Gibson and Gouws2000), and distrust (Burns Reference Burns2006; Haile et al. Reference Haile, Sadrieh and Verbon2008). But race may masquerade as politics in South Africans’ psyches, and do so asymmetrically thanks to ANC predominance. Circa our studies, 94% of ANC identifiers were Black and 54% of identifiers with the DA, the largest opposition party and the ANC’s main political adversary since 2004, were White.Footnote 2
Experimental Design
We designed two online survey experiments conducted in Qualtrics.Footnote 3 Both treat on partisan and/or racial identities and gauge trust discrimination via trust games. Following Berg, Dickaut, and McCabe (Reference Berg, Dickaut and McCabe1995), our trust games are played by two randomly assigned players who never meet and are told their partner lives in South Africa. Player 1 receives a sum of money and learns she can share some, none, or all of it with Player 2, who received the same sum.Footnote 4 Player 1 is told that any sum shared will be tripled before Player 2 receives it, and that Player 2 will, then, have the same options—to return some, none, or all of it to Player 1. Trust is measured as the amount Player 1 sends to Player 2. Instead of hard cash incentives, for logistical reasons, players were endowed with raffle tickets for several cash prizes.Footnote 5 After playing the trust games participants answered a questionnaire.
Note that this measure matches a well-recognized definition of trust: “intention to accept vulnerability based upon positive expectations of another” (Rousseau Reference Rousseau, B., Ronald S. Burt and Camerer1998, 395). Thus, trust reflects the perceived trustworthiness of others (Hardin Reference Hardin2006). Players should translate limited information about partisanship and race into trustworthiness stereotypes (e.g. Carlin and Love Reference Carlin and Love2013, Reference Carlin and Love2018). With these measures, definitions, and assumptions, we proceed. Our analyses and results below follow the reporting standards of the APSA Organized Section on Experimental Research.
Study 1Footnote 6
In November 2011, 1,624 student subjects from two major South African universities (University of Capetown and Witwatersrand University) logged on to participate in our online study. In total, 1,206 subjects completed some or all of the instrument. Each played five trust games. The first “anonymous” game imparted no information about Player 2. Then subjects played games in which we randomized information about Player 2’s identity: identifying with the ANC, the DA, as Black, or as White. After game play, subjects were asked their party identification (for Player 1’s, 103 ANC, 235 DA) and racial identification (for Player 1’s 203 Black, 197 White, 105 Indian, Colored, or other). Only subjects who identified with the ANC or DA are included the analyses involving partisan games. For analyses involving racial trust discrimination, only subjects who identified White or Black are included (485, over 78% of subjects).
The online environment presented coordination challenges. Thus, approximately half of the subjects (551) were randomly assigned to the role of the trustor (Player 1, the subject of this study) and the other half (654) to the role of the trustee (Player 2). Player 2s fitting the partisan or racial description in the treatments were randomly matched with real ticket allocations from a Player 1 in the same condition. Final ticket allocations were determined by the joint decisions made by the matched players. Hence, subjects were not deceived when told that they were playing with actual partners (Carlin and Love Reference Carlin and Love2013, Reference Carlin and Love2018; Fehr et al. Reference Fehr, Fischbacher, Rosenbladt, Schupp and Wagner2003).
We acknowledge student samples raise questions of external validity. Though students may be less trusting, we find no such evidence: in Study 1’s no-information or “anonymous” trust game Player 1s sent, on average, 0.49 of their endowment. Johnson and Mislin’s (Reference Johnson and Mislin2011) meta-analysis of anonymous trust-game studies on student and non-student samples indicates the average proportion sent by Player 1s is 0.50 (s.d. = 0.12, range 0.22–0.89). But we recognize two potential biases associated with recently socialized South African university students. First, they may have weaker attachments than those who forged party identities before or during the transition from Apartheid. This would bias our tests against observing partisan trust discrimination. Similarly, they have limited experienced with ANC predominance. This would bias against finding asymmetric partisan trust discrimination. Hence, ours is a hard test for our expectations.
As shown in Figure 1,Footnote 7 ANC and DA partisans in the role of Player 1 trust co-partisans and anonymous players to roughly the same degree; DA partisans trust “anonymous” South Africans slightly more than ANC identifiers (.75 tickets, p=.01). However, we observe dramatic differences between ANC and DA partisans vis-à-vis trust in rival partisans. While ANC partisans trust rival partisans marginally less than co-partisans, DA identifiers trust rival partisans significantly less than co-partisans. We call the ticket difference between what subjects send to co- and to rival partisans the “partisan trust gap,” and we call the difference based on race, discussed below, the “racial trust gap.” While the partisan trust gap for both groups of partisans is distinguishable from zero, it is nearly twice as large for DA partisans (see Figure 2; difference, p=.03 in paired t-test).
Mean ticket allocations by treatment and partisanship, 95% c.i.
Trust gaps by type and partisanship, 95% c.i.
Considering the racial dynamics and histories of the two parties is the asymmetry in partisan trust gaps mirrored in racial trust gaps? That is, do ANC and DA identifiers employ partisanship and race differently in trust decisions after two decades of ANC electoral dominance? To compare partisanship’s influence on trust relative to race we examine the political and racial trust gaps by partisanship.
Figure 2 shows no significant differences in racial trust gaps between DA and ANC partisans. For ANC subjects, racial and partisan cues produce equal trust gaps. However, for DA subjects, the partisan cue produces substantially greater trust discrimination than the racial cue. We calculate a gap-of-gaps, by subtracting the racial trust gap from the partisan trust gap, to gauge how much more distrust partisanship triggers than race. For DA identifiers the partisan-racial trust gap is 1.18 tickets (p<0.001, paired t-test). For ANC identifiers, however, the partisan-racial trust gap is indistinguishable from zero. This provides evidence consistent with the notion that electoral “losers” in South Africa discriminate trust more along party lines than racial lines relative to electoral “winners.”
Since a substantial proportion of our ANC and DA partisans identify as Black, we can isolate any possible racial dimensions that partisan cues contain from their political components by restricting the analysis to Black subjects. When paired with DA partisans, Black ANC partisans and Black DA partisans sent an equal number of tickets; however, when paired with ANC partisans, Black DA partisans sent 1.3 tickets less than Black ANC partisans (p=.02), mirroring the behavior of White DA partisans. Further evidence of the value of the political component inherent in partisan cues is observed in the difference between Black DA partisans and Black non-identifiers. Black non-identifiers send similar number of tickets in both the ANC and DA games (3.7 and 3.5) while Black DA partisans send 1.1 fewer tickets (p = .04) in the ANC game than the Black non-identifiers.
Study 2Footnote 8
To test the potential interactive effects of partisanship and race treatments, and to bolster external validity, we conducted a follow-up study in July–August 2015 that differed in two key ways. First, Study 2 paired an anonymous game with 2×2 factorial design of racial and partisan treatments. In this “fully-crossed” design subjects play four randomly ordered games with both partisan and racial cues (ANC or DA and Black or White). While Study 1 allows us to assess the effect of partisanship or race (and their relative effect sizes), Study 2 allows us to test the effect of each treatment conditional on the other.Footnote 9 Specifically, we can test whether race has any effect, directly or interacted, with partisanship.
A second difference is that Study 2 was fielded with a sample of 534 adult South Africans ranging in age from 18 to 82 (mean = 33) recruited via seven Facebook advertisements following Samuels and Zucco (Reference Samuels and Zucco2014).Footnote 10 Subjects were again asked their partisan and racial identification following game play (for Player 1s, 64 ANC, 115 DA, 22 other, 90 no party; 93 Black, 136 White, 55 Indian, Colored, or other). Taken together, Study 1 and Study 2’s variation in samples (student vs. adult)Footnote 11 and study design (treatment-by-treatment vs. factorial) provides strong external and internal validity checks on our inferences.
Table 1 reports the effects of partisanship, race, and their interaction on trust allocations in Study 2.Footnote 12, Footnote 13 Results mirror Study 1: subjects sent, on average, 1 ticket more to co-partisans than to rival partisans and rival-partisan distrust is asymmetrical—largely driven by subjects identified with DA.Footnote 14 DA identifiers sent, on average, 1.34 more tickets to fellow DA identifiers than to ANC identifiers (i.e. partisan trust gap), yet ANC subjects showed no significant rival-partisan trust discrimination (either directly or in interaction with the racial cue). This effect size is nearly identical to the one observed in Study 1 (1.43 tickets). The difference in partisan treatment effect between DA and ANC identifiers is 1 ticket (p=.015). Results are unchanged when estimated with a mixed-effects model (see results in Figure SM3 and Tables SM2–4 in Supplementary Material; ANOVA Table SM5 in Supplementary Material). Furthermore, as with Study 1, DA identifiers sent a similar number of tickets to anonymous Player 2s (6.1) as to fellow DA identifiers (5.8), indicating strong out-group discrimination toward ANC identifiers and not in-group favoritism. Dovetailing with Study 1, then, partisan distrust is most pronounced amongst opposition partisans.
Effects of partisanship and race treatments on interpersonal trust, study 2
Study 2 finds no racial discrimination when racial and partisan cues are combined. Nor do racial cues spur partisan discrimination. Whereas Study 1 subjects received racial information in isolation from partisanship, Study 2 subjects received race and partisan information in combination. Study 1’s design led to moderate trust discrimination along racial lines. In Study 2’s factorial design race has neither direct nor conditional effects on trust allocations. Though these findings contrast with evidence of politicians discriminating among constituents on the basis of race (McClendon Reference McClendon2016), they bolster the interpretation that, at least among ordinary citizens, partisanship acts a heuristic for electoral competition not racial difference.
In sum, in Study 2 only partisanship affects trust decisions. This supports our contention from Study 1—in line with Michelitch (Reference Michelitch2015)—that race is a fairly weak “stand in” for partisanship in social interaction. When race and partisanship are both presented, race’s predictive power evaporates and partisanship’s power changes little. Finally, the lack of an interaction effect between race and partisanship indicates that co-racials who identify with the rival party are not treated differently than non-co-racials rival-party supporters.
Discussion
Partisanship discouraging trust and legitimizing animus between rival partisans is common in competitive political systems, where multiple parties vie for and, at least occasionally, win power. In such contexts, the feelings are mutual—partisans of all stripes distrust their partisan rivals. Our research shifts the lens to a less-competitive context—a predominant party system—to test whether this symmetry holds, or whether lopsided competition creates asymmetry in rival-partisan distrust. Since such uncompetitive regimes are increasingly common, and the modal regime in Africa (Weghorst Reference Weghorst2018), our case study’s findings may speak to a great number of polities.
Drawing on behavioral games with experimental treatments in South Africa, after more than two decades of post-Apartheid ANC dominance, partisan trust discrimination is not symmetrical. ANC identifiers are no more or less trusting of their DA counterparts, yet supporters of the main opposition, the DA, strongly distrust partisans of the predominant ANC. This could be troubling given the importance of losers’ consent in democracy (Riker Reference Riker and Finifter1983) and winners–losers gaps in democratic legitimacy in Africa (Moehler and Lindberg Reference Moehler and Lindberg2009). More optimistically, neither DA nor ANC identifiers display substantial trust discrimination based on racial identities, despite South African partisan politics often falling along racial cleavages. The relative importance of partisanship to race sheds new light on Gibson’s (Reference Gibson2006) conclusion that race’s role in social interactions is declining in South Africa. More generally, this study’s finding help define the scope conditions under which we should observe mutual distrust between rival partisans. These may include, but are not limited to, systems lacking competitive elections and parties that correspond neatly to social (racial, ethnic, etc.) cleavages.
A lingering question is whether South Africans ought to discriminate trust on racial or partisan stereotypes, heuristics, or cues. After all, if trust is misplaced in individuals of different political or racial identities, then discrimination is wholly rational. The answer, of course, depends on the reciprocating actions of Player 2. But here the predictions for discrimination diverge somewhat. Although Player 2s receive the same information about partisanship and race of Player 1, they receive an important bit of hard data—the amount of trust Player 1s have placed in them—which turns out to be consequential for Player 2’s reciprocity in the US (Carlin and Love Reference Carlin and Love2013). Similarly, our analysis of Player 2 behavior in South Africa finds only a modest amount of partisan reciprocity discrimination: Player 2s in Study 2 returned 4 percent more of their tickets (based on their initial endowment plus the tickets received from Player 1) to co-partisans than to rival partisans; the correlation between the number of tickets Player 2 receives from and returns to Player 1 is 0.36 (p<.000). As with trust (the Player 1s), partisan reciprocity discrimination is restricted to DA identifiers who show out-group derogation toward ANC identifiers (see Tables SM6 and SM7 in Supplementary Material for full results). Thus, ANC Player 1s should discriminate against DA player 2s (but they do not) and DA Player 1s should not discriminate against ANC Player 2s (though they do). The discriminatory actions of Player 1s in trust decisions cannot be rationally predicted by the realized actions of Player 2s. In all, considering partisanship in social interactions involving trust is socially inefficient.
While our findings are consistent across differing experimental designs and samples, we recognize our study’s limitations. One is that we lack nationally representative samples. That said, subjects in our student sample (Study 1) are disproportionately likely to influence future decision-making and norms of social interactions in South Africa. Secondly, we cannot assess if or how political dynamics in the three years between our two studies may have shaped the behavior we observed. Having recognized that, the fact that our core results remain unchanged between the differing samples, time periods, or experimental design bolsters our findings.
In this vein, we conclude by noting that the validity of our theory can be tested as South African party competition unfolds. Any combination of corruption (de Kadt and Lieberman Reference De Kadt and Lieberman2017), ANC supporters socialized around the founding election aging out of the electorate (de Kadt Reference De Kadt2017), and shifting loyalties between the ANC and tribal chiefs (de Kadt and Larreguy Reference De Kadt and Larreguy2018) could cause the ANC to lose further ground to the DA. Such heightened competition should stoke threat and fear among ANC partisans and embolden the DA vis-à-vis the ANC. In such circumstances, we would expect rival-partisan derogation to increase among ANC supporters. Distrust would become mutual. When and if scholars will have a chance to test these expectations is an open question.
Supplementary Material
To view supplementary material for this article, please visit https://doi.org/10.1017/XPS.2019.17.
