Who Should Benefit from Group-Based Quotas? Experimental Evidence from Brazil

Abstract

Group-based quotas and affirmative action policies are common in many parts of the world. A large literature has focused on public debates over the perceived fairness of these policies. Yet in highly diverse countries, debates often centre on which groups should benefit and how to establish who counts as a member of those groups. We know comparatively little about these kinds of debates. In this article, we present new experimental evidence on public preferences over targeted race quotas in Brazil, a country where racial identity is famously ambiguous and where substantial controversy over the use of race quotas in university admissions remains. Placing respondents from an online convenience sample in the role of deciding whom to admit as part of a university race quota, the results indicate (1) that while economic class dominates other considerations in quota targeting, racial factors still matter; (2) the public places a large premium on indicators of merit; and (3) that these preferences are largely homogeneous across racial groups and resistant to priming.

Introduction

A remarkable number of countries have now adopted policies – such as affirmative action in the US, or group-based quotas in Latin America – aimed at redressing historical grievances among marginalised groups in the realms of education, business, and politics. Despite their pervasiveness, deciding who should benefit from quotas ¹ is not obvious and, as a result, often contentious. Two key questions drive debates over the targeting of quotas: (1) in most societies, and especially racially and ethnically diverse countries, numerous groups can have legitimate claims of marginalisation. Given budget constraints, how should quota seats be allocated across these groups? (2) Even if (1) is agreed upon, how should the relevant institutions determine who counts as a member of the relevant group? These questions are crucial to the design of quota systems but they also have broader resonance with redistributive policies that involve making choices about the targeting of beneficiaries (Hughes, 2011; Krook et al., 2009).

A broader literature on public opinion concerning redistribution and social policy has focused on whether social benefits should be targeted at all (Besley, 1990; Van Oorschot, 2002). Work that has focused on how social benefits should be targeted has found that who benefits matters for public support, often in ways that are racialised and rooted in people's beliefs about who the ‘deserving’ and ‘undeserving’ poor are (DeSante, 2013; Van Oorschot, 2002). Many of these patterns have also been found in public debates about affirmative action in the United States (Cahn, 2013). Yet much of this work suffers from a number of limitations.

First, very little of this research deals with public opinion on quotas directly. While in theory the targeting of benefits to some groups means that other groups may receive less, the ‘fixed pie’ nature of reserving seats in universities, public office, etc., present more obvious trade-offs than other forms of social spending. Second, studies of targeting typically test how broad, aggregate social categories influence public support for the reception of benefits – for instance, the race of the beneficiary, or their profession. However, it is hard to say which trait or characteristic of the categories in question is driving the observed effects. Aggregated as a ‘bundle of sticks’, the effects of race are particularly hard to identify given that many characteristics correlate within racial groups, and are difficult to disentangle (Sen and Wasow, 2016). Finally, although much of this work focuses on who the public believes should benefit, the question of how to identify members of a given group remains relatively unexplored. This is a particularly fraught question in post-colonial states, where group membership – especially race – is often more fluid than in Global North countries (Telles and Paschel, 2014).

In this article, we provide new, experimental evidence on public preferences over the targeting of group-based quotas. Empirically, we focus on the use of race-based quotas in Brazilian universities. Brazil is a compelling case study for our purposes. An early adopter of race-based quotas, Brazil is also a country where – relative to the United States and Western Europe – the boundaries between racial groups are often blurry, and where there can be reasonable disagreements about what racial ‘group’ someone belongs to (Telles and Paschel, 2014). This quality has led to substantial public controversy over who should benefit from the quotas. Controversies have involved accusations that some students have ‘cheated’ the system by misrepresenting their race, and even to the recent introduction of ‘verification committees’ tasked with validating the race of beneficiaries of race-based quotas (Silva et al., 2023). These patterns, however, are not unique to Brazil. They are echoed in debates in India about who belongs in the Scheduled Castes (Jenkins, 2003) or controversies in the United States about how politicians represent their race or ancestry.

To explore public preferences over the targeting of race quotas we employ two survey experiments on an online, convenience sample of 1,000 respondents. The first experiment aims to characterise which group attributes or dimensions of marginalisation citizens would choose to prioritise in race quotas. We focus on four major dimensions of marginalisation that have been discussed extensively in Brazil: gender, economic class, regional disparities and racial appearance, or phenotype (Silva et al., 2023). We use a discrete forced choice conjoint design (Hainmueller et al., 2014) that puts participants in the role of deciding who to select – from among pairs of randomly generated students – for reserved seats at a Brazilian university. The ‘forced’ choice nature of the design allows us to characterise which attributes matter most to respondents and how the public evaluates trade-offs when forced to choose between competing dimensions of marginalisation.

The second experiment tests the extent to which these preferences are malleable. A common criticism of public scholarship on affirmative action in the US is that polling can change dramatically depending on how questions are framed (Kopicki, 2014). More generally, prior research in social psychology suggests priming people to think about poverty – and the link between race and poverty – can shape attitudes towards broader redistributive policies (Krosch and Amodio, 2014; Rodeheffer et al., 2012; Valentino et al., 2002). Drawing on this body of work, we randomly exposed participants to images of (general) economic scarcity or racialised economic scarcity prior to completing the first experiment. We test both whether the primes move respondents to behave differently in the conjoint tasks, as well as whether they shape broader beliefs about redistribution and affirmative action policies. To our knowledge, we are among the first to use an experimental prime to test for differences in behaviour during conjoint tasks.

The study makes the following contributions. First, economic class appears to dominate other dimensions of marginalisation, producing the widest ‘spread’ in preferences among respondents. Even so, this does not necessarily mean that ‘class trumps race’ in preferences over quotas: a student's racial appearance can still move public opinion on who should benefit from quotas, holding class constant. This finding can speak to debates about the extent to which racial identity or class should be prioritised in varied redistributive policies (Kahlenberg, 1996) – particularly whether race should receive consideration independent of class. Second, we find that even while tasked to think in terms of redressing historical exclusion, respondents still place a large premium on indicators of ‘merit’ and reward students with stronger academic backgrounds. These results suggest that quotas designed with explicit academic performance ‘floors’ in place – or who frame quotas as benefiting particularly strong students – may overcome some of the public’s intransigence to these policies.

Lastly, and perhaps surprisingly, we find that these preferences are largely homogeneous across respondents and resistant to priming. Although some interesting differences emerge between White and non-White respondents, and between politically conservative and liberal respondents, they do not suggest these groups have diametrically different views on who should benefit from the quotas. Along these lines, we find null effects on the priming experiment, which indicates such preferences are not easy to shift. Jointly, this last set of results suggests that while there may be some room to move the opinions of important subgroups depending on how quotas are designed and framed to the public, there is a reasonable amount of consensus on how they should be implemented.

Quotas in the Public Sphere

Group-based quotas and affirmative action aim to increase the representation of historically marginalised or excluded communities in important institutions, such as the university, legislative bodies, and firms. These policies are not uncommon: Jenkins and Moses (2014) estimate that about a quarter of the world's countries have some form of affirmative action policy in place. Moreover, the Varieties of Democracy Project estimates that approximately 50 per cent of the world's countries in 2022 had at least a nominal gender quota for the lower house in national legislatures (Coppedge et al., 2011). Despite the fact that these policies are fairly widespread, however, they are often polarising. Debates in the public sphere typically turn on disagreements about whether quotas, affirmative action, or other redistributive policies are ‘fair’ or effective (Mosley and Capaldi, 1996).

There's a large body of research on why views differ on these questions. A central insight from the literature is that how people view those who stand to benefit from a policy – the targeted beneficiary – can have a strong impact on public preferences. At stake is whether the targeted beneficiary is seen, by the public, as deserving of the benefit or not (Gilens, 2009; Petersen, 2012; Van Oorschot, 2006). As one might expect, people's evaluations of deservingness are informed by a variety of psychological dispositions, biases, and other considerations (Crosby et al., 2006). There is significant debate, for example, on the extent to which opposition to affirmative action in the United States is primarily driven by racial sentiment as opposed to other considerations (DeSante, 2013; Kuklinski et al., 1997). Who benefits is thus very important for the political viability of social policies.

While these insights are valuable, we argue that there is much they leave unanswered about public preferences over the targeting of quotas and other social benefits. Part of the problem is that much of this work explores whether groups are seen as deserving or not in isolation; for example, the public might think veterans are deserving of benefits but the unemployed are undeserving. Yet its possible that the public might think that multiple social groups have legitimate claims to historical marginalisation (Davenport, 2020). The question then becomes how the public believes seats should be apportioned among these groups, given that granting one group more representation necessarily reduces access to others.

This question is particularly salient in diverse, multicultural societies where multiple and sometimes overlapping groups are marginalised or underrepresented in public life (Davenport, 2020). Indeed, the recent June 2023 United States Supreme Court decisions (Students for Fair Admissions, Inc. v. President & Fellows of Harvard College and SFFA v. University of North Carolina) on the use of affirmative action in university admissions hinged on the argument that the prioritisation of some historically excluded students (e.g. Hispanic and Black students) came at the expense of other, historically marginalised students (e.g. Asian-American students) (Montague, 2023). In our country of study, Brazil, multiple groups (e.g. women, sexual minorities, Afro-Brazilians, indigenous communities) are often seen as being historically marginalised. It is thus often the case that quotas must be apportioned among marginalised groups, yet we know little about public preferences over targeting in these situations.

A separate but much less studied problem is that, even if there was public consensus on which groups to prioritise in quotas, how to define who belongs to those groups (and who doesn't) is often unclear. Scholars of race have argued that group belonging can be fluid, ambiguous, and contextual in many parts of the Global South (Davenport, 2020). In Latin America, for instance, ‘blurred or ambiguous boundaries around racial categories’ prevail, which allow ‘people to move in and out of them’ with more ease (Telles and Paschel, 2014; Wade, 1993: 865). In India, there is debate not just about whether certain ‘Scheduled Castes’ should be targeted through quotas and other benefits but how to determine who is a member of those castes (Jenkins, 2003; Mishra, 2015). Indeed, new marginalised identities can even emerge over time as existing groups are recognised as having legitimate claims to marginalisation and exclusion (e.g. the LGBT community, indigenous peoples) (Proctor, 2020).

In sum, we know comparatively little about how the public thinks quotas should be targeted in situations where either group membership can be contested or where multiple, potentially competing identity categories are seen as having legitimate claims of marginalisation. These questions are important given that which criteria policymakers rely on to define access to group quotas has downstream consequences for applicants and how the state ‘sees race’ more generally (Silva et al., 2023). For instance, in a survey experiment, Bailey (2008) shows that people are more willing to identify as Black in the Brazilian context when primed to think about race-based quotas. In the realm of public opinion, how the public defines or understands membership in social groups is likely then to inform public debates about how quotas should be targeted.

A final note is that we believe many of these gaps exist in the literature as a result of an over-focus on Global North countries as empirical settings in the politics of group-based quotas and other forms of redistribution. Scholars of race have long argued that, due in part to differences in colonial experiences, group categories tend to be more stable and less ambiguous in the Global North, and especially the United States (Telles and Paschel, 2014; Wade, 1993). In Global South countries, such as in Brazil, the politics of group-based quotas are likely to be especially fraught and produce more varied coalitions of support and opposition. Figure 1 captures some of this dynamic at the aggregate level, where support for affirmative action maps neatly onto a left-right dimension in the United States but is much more heterogeneous in Brazil.

Figure 1.

Wording in Brazil Sample (LAPOP, 2012): ‘Universities Must Reserve Seats for Darker-Skinned Students, Even If They Have to Exclude Some Other Students. To What Extent Do You Agree or Disagree?’. Wording in US Sample (CCES, 2012): ‘Affirmative Action Programs Give Preference to Racial Minorities in Employment and College Admissions in Order to Correct for Past Discrimination. Do You Support or Oppose Affirmative Action?’.

Who Should Benefit? The Targeting of Quotas

Dimensions of Marginalisation

To summarise, our intervention in the literature is twofold: (1) to explore how people decide who to target for group-based quotas in settings where there are multiple, potentially overlapping groups to choose from; (2) to explore this question in settings where group boundaries are unstable and harder to define. We focus empirically on group-based quotas in higher education.

One approach to this question is to ask which group traits (or combinations of traits) the public thinks should be prioritised in the design of group-based quotas in education. As identity and intersectionality scholars have argued, marginalisation is constituted by the overlap or confluence of traits that shape a person's relationship to social status, or as Simien (2007) describes, a ‘simultaneity of oppression’. We focus on four dimensions of marginalisation that are critical in the Brazilian context but also important in other settings: gender, class, regional disparities, and racial appearance, or phenotype. ² We discuss each of these in turn.

Gender is a historically important trait in the design of group-based quotas. Typically, it deals with the allocation of seats to women (as opposed to men) as a means to redress gender-based marginalisation. There is reason to expect some level of public ambivalence on the extent to which gender should be prioritised over other considerations. On the one hand, there is a large literature on disparities between men and women across a variety of fields, including levels of employment, wages, and health outcomes (DeSouza et al., 2004). That said, women in many middle- and high-income countries (including Brazil) make up a significant proportion of tertiary students (OECD, 2021), which may undercut the belief that increasing women's presence in universities will be a palliative for these disparities. Moreover, there may be opposition in some societies to women taking on roles (such as entering university) that violate traditional patriarchal ideas of womanhood (Baldwin and DeSouza, 2001).

By contrast, there is reason to expect class will be a much less ambiguous determinant of who the public sees as deserving or undeserving of quota benefits. Indeed, part of the debate about the targeting of certain social benefits, at least in the US, is whether class should be the only factor that matters (Kahlenberg, 1996). The ambiguity that exists is typically about where to draw boundaries between those who should and should not benefit, for instance, in debates about the value of ‘means testing’ welfare (Van Oorschot, 2002). In Latin America, there is an especially large gap (relative to the general public) in the class backgrounds of politicians and other elites, and public resentment of this fact (Carnes and Lupu, 2015). To the extent that people see education as a pathway to class mobility, class background should drive a lot of the variation in who people believe should benefit from group-based quotas.

Where the literature is less clear – particularly in the Global South – is on the question of race, given that racial categorisation is often fluid in these contexts (Davenport, 2020). Rather than examine race as a category of racial identity, we follow a growing number of scholars in focusing on how the racial appearance, or phenotype, of potential quota beneficiaries shapes public preferences. We do this for two reasons. First, one of the central questions we are interested in is who is seen as belonging to a marginalised group in situations where group membership cannot be taken as given. Second, a growing body of work on ‘colourism’ in the United States and Latin America indicates physical appearance, such as how ‘dark’ or ‘light’ one’s skin is, may be as defining or more so of life outcomes than traditional sociopolitical constructions of racial categories (Dixon and Telles, 2017; Monk Jr, 2016; Telles, 2014). ³

In terms of expectations, our reading of the literature suggests potential beneficiaries at the ‘extremes’ of phenotypical appearance – that is, beneficiaries with physical traits strongly associated with either White or Black racial characteristics – are easy, in the public's eyes, to categorise as belonging to a marginalised or non-marginalised community (Telles and Paschel, 2014). The public should then see students who are easily categorised (in the Brazilian context) as White as less deserving, and non-White as more deserving, of a reserved seat in a quota system. The same is not true of beneficiaries with more ambiguous phenotypical characteristics, whom the public is likely to struggle to categorise (Telles and Paschel, 2014; Telles, 2014). The beneficiaries in these intermediate categories should thus be less easy to sort into more or less deserving of the quota system. The result is that phenotype should have non-monotonic effect on public preferences: strong effects at the extremes of appearance, weak or ambiguous effects among the intermediate categories on the continuum.

A final trait that we consider is the extent to which regional disparities factor into public preferences over quotas. Here, we are careful to conceptualise regional disparities independent of their correlates. For instance, race and class can geographically cluster in ways where regions become synonymous with poverty or certain racial groups (e.g. the South in the United States). Yet, development scholars have long noted the importance of regional disparities – even holding other correlates constant – across a host of economic and health outcomes (Laksono et al., 2020). People from more remote regions, or regions with less infrastructure, are likely to be locked out of economic and educational opportunities in ways that produce disparities in outcomes. ⁴ That said, these direct effects may be less obvious to citizens, and thus while people from less developed regions should be seen as more deserving of quota benefits, this effect is likely to be less powerful than its correlates.

While we estimate the effects of these dimensions of marginalisation in isolation, scholars of intersectionality have long warned against our ability to cleave off identities and measure their independent effects (Simien, 2007). As a result, we also consider how interactions among these dimensions affect ideas of deservingness in public quotas – for instance, the extent to which a woman of colour is seen differently from the sum of her parts (e.g. the effect of being a woman and a person of colour).

Priming Marginalisation

An important caveat in our discussion is that the extent to which people have these dimensions in mind – and especially the extent to which they understand how these traits produce marginalisation – is likely to influence their beliefs. ⁵ Given our interest in exploring how the public believes quotas should be targeted in contexts where group boundaries are fluid, one natural extension of the project is to consider whether people can be primed to value certain dimensions of marginalisation over others. Priming effects in our study would ultimately speak to the malleability of deservingness preferences: are these fairly static preferences, or can they be moved easily?

Towards this end, we draw on growing work in psychology that suggests economic scarcity shapes people's perceptions of racial identity. We build on the work of Rodeheffer et al. (2012) and Krosch and Amodio (2014), who show that participants who are primed to think about economic scarcity are more likely to identify racially ‘ambiguous’ faces with the out-group, effectively sharpening group boundaries. This change can have downstream consequences for policy – Krosch and Amodio (2014) find that scarcity can alter perceptions of race in ways that exacerbate discrimination. We anticipate that economic scarcity primes should both moderate how dimensions of marginalisation impact beliefs about deservingness, especially those bearing on class and phenotype, as well as general attitudes towards redistribution and affirmative action.

Race Quotas in Brazilian Universities

We focus on Brazil in this paper because of public controversies over the use of group-based quotas. Key for our purposes is the fact that, in Brazil, there are a wide range of racial categories in use and that race has a remarkably ‘elastic’ relationship with physical appearance: two people similar in appearance may nonetheless identify with different racial groups (Telles and Paschel, 2014). There are many reasons for this pattern, but chief among them are colonial legacies and a historically high level of social acceptance of interracial marriage, at least relative to the United States (Wade, 1997). Racial identity has also changed in the last few decades, as Black consciousness movements have advocated for greater public recognition (Covin, 1997; Dixon and Telles, 2017). While racial mixing is more prevalent, it has not prevented marginalisation and exclusion in Brazil. Physical appearance, particularly skin colour, plays a significant role in racial identification and shapes Brazilians' lived experiences, including their exposure to discrimination and marginalisation. Some argue that Brazilian society operates more as a ‘pigmentocracy’ than a traditional racial hierarchy (Telles, 2014).

We are interested in the way that Brazil's racial diversity has come into conflict with the use of race-based quotas in its public universities. National, standardised quotas come into effect in Brazil with the 2012 federal law (Law 12.711) (CÂMARA DOS DEPUTADOS, 2012). Brazil's quota system involves both a ‘social’ component for students who graduate from public high schools and students who are low income, and a separate racial component that reserves seats at public universities for black and brown (‘pardo’) students (Reis and Raimundo Santos, 2022). We focus on this latter, racial quota system, which, as Silva et al. (2023) notes, has generated more controversy than the other quotas. It is worth stressing here that, relative to affirmative action policies in the United States, Brazilian universities have much less discretion over the selection of new students. Admissions are primarily determined by standardised university entry tests, followed by the application of social and racial quotas, leaving less room for universities to evaluate student ‘merit’.

We focus specifically on debates over who should benefit from these quotas. In Brazil, the institutional answer to this question has changed over time: until 2016, quota benefits were apportioned based on racial self-identification, that is, students who identified as black (‘preto’) or brown (pardo). Over time, however, student groups mobilised for changes in quota criteria, either on the grounds that some students were ‘cheating’ the programme by misrepresenting their race, or on grounds that the relevant criteria should not be self-identification but rather a person's racialised physical appearance (Neves, 2022). Along these lines, some universities have developed ‘verification committees’ which task committee members (in some cases university faculty) with ‘verifying’ that a student should benefit from race quotas based on their phenotypical characteristics (Silva et al., 2023).

Race-based quotas are publicly divisive in Brazil, ⁶ and it is worth highlighting the exact nature of the controversy. Part of the controversy resembles debates in the US over affirmative action: namely, whether race (or gender, etc.) should be considered at all in admissions. The other part of the debate, however, deals with the problem of deciding who should benefit from the quota once it is in place. At stake here is how the Brazilian public and the state ‘see race’ (Silva et al., 2023), and how to decide which markers of physical appearance connote a marginalised identity. Our study aims to expand what we know about public preferences in these contexts.

Data and Methodology

The Survey

Our data are an online convenience sample of survey respondents in Brazil. Convenience samples are common in social science: they differ from probability samples in that not all members of the target population are equally likely to be sampled in the survey, which presents trade-offs for researchers that we discuss in more detail below.

Participants in our sample were selected by NetQuest in 2019. As with other online survey firms, NetQuest maintains an active panel of respondents from the countries in which it operates and offers respondents small financial incentives to complete the survey. Our only criteria for inclusion were that the respondent should be an adult (18 or over), a Brazilian citizen, and currently residing within the country. Given our interest in race and concerns that online, convenience samples are less diverse than the overall population, we also deliberately over-sampled non-white participants by placing higher weight on the likelihood that a non-white participant from the panel was chosen. While our final sample of participants still over-represents white Brazilians, the skew is less severe than it would have been in the absence of these measures.

Participants took the survey online, either on their computers or through a smart phone or tablet. To measure each respondent's skin-tone, we present respondents with the LAPOP colour palette shown in Figure A.5 and ask them to identify the colour that most closely matches their own skin-tone. ⁷ On the whole, 1,000 respondents were collected. More details on the survey, including demographics of our sample are available in the Appendix.

Representativeness of the Sample

The online convenience sample is an important limitation of our study: the Brazilians who are most likely to be sampled in our survey are likely to differ from the general population in important ways. To examine how our sample differs from the Brazilian population we display key demographic characteristics in our sample (Figures A.1 and A.3) and compare those to similar characteristics in the Latin American Public Opinion Project's (LAPOP) 2018 wave in Brazil, which is nationally representative (Figures A.2 and A.4). A caveat: while the wording and answer choices of our questions are similar to LAPOP's, they are not always identical.

Here, we highlight a few differences between the nationally representative sample in LAPOP and our sample. The results are largely unsurprising: relative to the nationally representative sample, our sample has more women, more people who identify as white (although notice the difference in available answer choices), and more people who are currently employed. Our respondents also appear to have higher educational attainment, are more ideologically moderate, and consume news at a lower frequency than the national average. These differences are largely in line with what we know about convenience samples: online respondents tend to be unrepresentative of the broader population in socioeconomic terms and tend to have higher educational attainment (Krupnikov et al., 2021).

These differences suggest we should be careful about over-attributing attitudes in our sample to the broader Brazilian population. That said, our primary goal in this study is the estimation of treatment effects in an experimental setting. What is at stake is whether the effects we estimate in our study would replicate in a nationally representative sample, not whether our sample perfectly mirrors national characteristics. There is past research that says this might be so: Coppock et al. (2018), for instance, successfully replicate 27 of 28 survey experimental results fielded on an Amazon MTurk sample on nationally representative samples. Along these lines, Clifford et al. (2015) show that even when convenience samples differ in important ways from the broader population, we can still draw broader inferences about our topic of study. The reliability of convenience samples is discussed in more depth in Krupnikov et al. (2021).

In sum, while our sample has limitations, there is evidence supporting the use of convenience samples in experimental settings.

The Experiments

To explore citizen's preferences over quotas, we rely on two experiments. The first experiment is a discrete choice conjoint design (Hainmueller et al., 2014). In this experiment, participants read the following prompt:

Many universities across the world have established programs that reserve a portion of seats in each incoming class for students who belong to historically marginalized groups. Imagine that you are on an admissions committee, and are being tasked with choosing students to fill these special seats for historically marginalized students. Please consider the characteristics of the two candidates for admission below and choose which of the two you think should be given entry. ⁸

Respondents are presented with a table that describes two hypothetical students that vary along the four dimensions of marginalisation described in the previous section. We operationalise the four dimensions as follows: (1) gender takes on the value of man or woman; (2) for class, we use the student's family income bracket, drawing the brackets from the LAPOP's income brackets for Brazil; (3) for phenotype, we use the student's skin-tone (Monk, 2016), and rely on the LAPOP ‘skin-tone palette’ that is standard in the literature; (4) for regional disparities, we use Brazil's standard five regions, where the South and Southeast are traditionally the more privileged parts of the country whereas the North and Northeast the more excluded. To these four we add the student's high school GPA as a relevant characteristic in university admissions (see Appendix Table A.1 for more details). Participants then select which hypothetical student to give entry. Participants repeat this process a total of five times.

In terms of analysis, we estimate both marginal means (MMs) and average marginal component effects (AMCEs), clustering standard errors by respondent. The MM estimate is a descriptive quantity, defined as the probability that a student with a given attribute level was chosen in the experiment, averaging over the student's other attributes (Leeper et al., 2020). In other words, the MM tells us how popular a given attribute level is with participants. The AMCE is a causal quantity, defined as how much more or less likely it is that a student with a given attribute level is chosen, relative to a student with a different, baseline characteristic. In other words, the AMCE lets us see whether a student's chances of being chosen increases if we change an attribute level of interest. ⁹

The second experiment builds on work by Krosch and Amodio (2014) and Rodeheffer et al. (2012) on the effects of primes on racial identity. We randomly assign participants to one of three priming conditions: a poverty prime, a racialised poverty prime, and a placebo condition. The primes are designed to mimic the ‘meme’ format that is increasingly popular in Brazil, with text and an image depicting either an impoverished landscape (poverty prime), an image of a Black man living in poverty (racialised poverty), or an innocuous ‘fun fact’ about the science of the brain (placebo). More details are available in the Appendix.

We estimate two types of priming effects. The first is in the conjoint experiment, where we test whether receiving one of the primes meaningfully alters who participants choose in the experiment. We follow Leeper et al. (2020) in estimating differences in MMs and conduct an F-test to evaluate the hypothesis that, overall, people were choosing differently depending on which kind of prime they received. The second priming effect is with regard to more general attitudes about redistribution and affirmative action. Here we look at three outcome variables bearing on: (1) support for redistribution (1–3, ordinal), support for affirmative action (1–5 ordinal), and support for making incomes more equal (1–10).

Results

Who Should Be Prioritised in Quotas?

The primary results of the study are depicted in Figure 2, which displays MM estimates across all respondents in the sample. The MM estimate is simply the probability that a student with a given attribute level was chosen in the experiment, averaging over the student's other attributes (Leeper et al., 2020). For instance, we see that students who are women are chosen about 53 per cent of the time, whereas students who are men are chosen about 47 per cent of the time. The difference between these two quantities is the AMCE: how much more likely a student is to be chosen if they are a woman as opposed to a man (in this case, women are about 6 per cent points more likely to be chosen). AMCEs are presented in Appendix Figure A.7 (regression table in Appendix Table A.6).

Figure 2.

Results from the Discrete Choice Conjoint Experiment. Point Estimates and 95% Confidence Intervals of Attribute Level Marginal Means. Regression Results in Appendix Table A.2.

The results suggest a number of interesting patterns. First, there is significant debate about the extent to which racial characteristics should be considered independently of class in these kinds of targeted benefits. Some opponents of race-based affirmative action, for instance, argue that only a student's class should be given consideration (Kahlenberg, 1996). Yet in our results we see that, holding class constant, respondents will still prioritise students from a more marginalised phenotype. That said, class does seem to overwhelm other attributes; it emerges as the most powerful predictor of who participants believe should be targeted. At the extremes, participants are more than twice as likely to choose the poorest student than the wealthiest student. These differences are substantially larger than other traditional markers of marginalisation in Brazil, including phenotype, gender, and home region.

To what extent do respondents prioritise class over race? Comparing the predicted probabilities of two students that are otherwise similar (one in the lowest income category, the other in the darkest skin-tone) students will select the lower-income student about nine percentage points more often (Figure 3). This difference is not huge, but it is not trivial either.

The effects of class also highlight how decoupling characteristics that are often strongly correlated in people's minds (e.g. wealth and home region) can help us make sense of their preferences: holding region constant, a student's wealth or colour still exert a powerful effect; yet holding other factors constant, a student's home region barely moves the dial. One reading of these results is that support for targeting marginalised regions is largely a function of citizen's desire to target the poor, and that other considerations – increasing representation of a region's distinct culture, history, etc. – have little weight in citizen's minds.

Second, participants place substantial weight on a student's high school GPA. In some respects, this is striking, given that respondents were tasked with choosing specifically in terms of redressing historical marginalisation, with no reference to prior student performance. A student with an A average, for instance, is 8 percentage points more likely to be chosen than a student with a B average, and this difference alone is larger than the difference in preferences for students from the (comparatively privileged) southeast and students from the (comparatively marginalised) north.

Finally, the results for candidate skin-tone merit discussion. Participants appear to have clear preferences at the extremes: students with the darkest skin tone are about 13 percentage points more likely to be targeted for affirmative action policy than students with the lightest skin tone. Yet participant preferences are much less clear for the intermediate skin-tone categories – many of these hover around 50 per cent (effectively the point of indifference) and there is no statistically significant difference between skin-tone categories that are in close proximity.

If respondents are seeking to prioritise Black students on the basis of skin-tone, these result in some ways differ from prior work, such as Telles (2014), which suggests skin-tones above ‘6’ are typically classified by survey participants as Black. In our results, only the students in the darkest skin-tone categories are prioritised. The lack of differentiation among the middle categories is also interesting and may reflect the centrality of these students in debates over racial quotas because their racial ‘status’ is ambiguous in the face of these policies (Telles and Paschel, 2014). Similar ambiguities exist over the categorisation of Brazilians as pardo (Nogueira, 2024).

One last possibility is that the effect of each attribute depends on the value of other attributes (i.e. the effect of region depends on class). We find no evidence, however, that the attributes meaningfully interact with one another (Appendix Table A.10). In short, when it comes to affirmative action policy respondents seem to prioritise economic class, though there is room for consideration over phenotype. They also, however, place significant weight on a student's grades, which may represent a belief that the student either ‘deserves’ to be given entry or has the necessary ‘aptitude’ for higher education. When it comes to race and identity, participants react to large differences in markers of race but have a harder time making sharp distinctions between students who are more similar in physical appearance.

Figure 3.

Predicted Probabilities of Being Selected in the Experiment for Two Hypothetical Students. Probabilities Generated from Model Reported in Appendix Table A.6.

How Stable are Affirmative Action Preferences Across Groups?

The results so far provide a view of how participants as a whole assess and respond to student attributes in forming preferences over affirmative action. But how do these effects vary across respondents? For instance, do white and non-white participants tend to have different preferences? And how malleable are these preferences – for example, with respect to elite cues or messaging in the media?

To answer these questions, we follow Leeper et al. (2020) and take the following steps. First, for the respondent attributes of interest, we conduct a nested model comparison test, comparing a null model (the simple features) and an interaction model (where every feature is interacted by the respondent characteristic). We only interpret results whose p-value in this test is less than the conventional .05. Second, for presentation purposes, we visualise differences in MM estimates. These can be interpreted as how much more (or less) likely a respondent with a given characteristic is to prefer a student with a given attribute than a respondent with a different characteristic. ¹⁰

Figure 4 presents the first set of these tests, comparing whether White and non-White respondents vary significantly in their behaviour in the experiment. We can see, for instance, that White respondents are 8 percentage points more likely to choose an A average student than a non-White respondent. One interpretation of this pattern is that, relative to non-White respondents, White respondents are more likely to place a premium on what they see as ‘merit’-based admission to the programme. In comparison, non-white respondents appear to have a broader conception of deservingness that may be more flexible to non-traditional considerations that could contextualise admitting a student with varying grades.

Figure 4.

Differences in Marginal Means Estimates, Comparing White Identifying Respondents and Non-White Respondents. Significant differences are highlighted in darker tone; non-significant differences are depicted in lighter tone. Regression Table in Appendix Table A.3.

We also compare respondents based on their ideological priors. At the time of the survey, whether or not someone supported Jair Bolsonaro was a major political cleavage in the country. Figure 5 examines respondents who either support or oppose ¹¹ President Jair Bolsonaro, a proxy for the respondent's political ideology given the highly polarising nature of the right-wing president. Here we see, for instance, that Bolsonaro supporters are more likely than opponents to select students who are in the wealthiest income category, and less likely than their counterparts to select students in the poorest category. Put another way, right-leaning Brazilians tend to ‘punish’ wealthy students disproportionately less than leftleaning Brazilians, while left-leaning Brazilians tend to disproportionately favour the poorest students.

Figure 5.

Source: Author. Differences in Marginal Means Estimates, Comparing Bolsonaro Supporters and Bolsonaro Opponents. Significant differences are highlighted in darker tone; non-significant differences are depicted in lighter tone. Sample Size = 9128. Regression Table in Appendix Table A.4.

How Malleable are Affirmative Action Preferences?

One final point ¹² that we consider is whether affirmative action preferences are malleable, both in terms of the conjoint experiment but also in general levels of support for affirmative action. Here, we leverage the second experiment that primes respondents to think about poverty and the relationship between race and poverty. We test whether respondents who were treated with these primes: (a) exhibit differences in preferences in the conjoint experiment; (b) exhibit differences in support for (1) affirmative action and (2) redistribution, a closely related yet more general set of policies that extend beyond education.

We find no evidence that respondents primed to think about poverty or race behave differently in the conjoint experiment than respondents in the control condition (Appendix Figure A.10, regression results in Table A.9). We further find no evidence that these primes can move respondent attitudes more broadly (Figure 6), either with respect to affirmative action policy in education (bottom panel) or in terms of attitudes towards redistribution (top panel). There also appear to be no major differences between White and non-White respondents.

Figure 6.

Results from Priming Experiment. Point Estimates and 95% Confidence Intervals. Regression Results in Appendix Table A.5.

Overall, the picture that emerges is that while there are some noticeable differences across respondents, we surprisingly find that affirmative action preferences are fairly homogeneous and largely stable or unresponsive to primes.

Conclusions

Group-based quotas are a common tool for redressing historical grievances among marginalised groups, but they face important challenges. First, multiple groups may have legitimate claims to historical marginalisation, and there is no clear, a priori way of determining how to apportion seats among those groups. Second, determining who belongs to a group that is a quota beneficiary is often not obvious. In these situations, who does the public think should benefit?

To be clear, our aim in this study is to characterise public opinion over the design of group-based quotas, not make normative claims about how these programmes ought to be designed. There may well be legitimate, normatively grounded reasons for policymakers to enact policies that diverge from public sentiment, for instance if such sentiment is influenced by prejudice, intolerance, or other biases. This is one of the fundamental tensions of democratic rule. Our study does not take a normative stance on the design of these programmes. However, we do believe there is value in democracies to knowing about how the public feels about these programmes and how they should be designed, in part because public opinion can shape public policy either directly or via who is elected (Kertzer and Zeitzoff, 2017).

Our results make a number of contributions. First, class emerges as a powerful predictor of who the public views as deserving of quota targeting. These patterns are largely similar across White and non-White respondents, and across respondents on the ideological Left and Right. The results show that the public wants to see the poor primarily benefit from these policies (Kahlenberg, 1996), which has implications for the design of policy. To the extent that class and race are highly correlated, programmes that target based on class are likely to be less controversial than their counterparts that emphasise race, and yet still benefit the same students. In countries where race and class are less correlated, however, there may be significant trade-offs in these designs.

Second, although participants were instructed to focus specifically on allocating seats based on deservingness in relation to marginalised experiences, the public clearly places a large premium on correlates of traditional measures of aptitude or merit. These findings are in line with prior work showing that differences in perceptions of merit are key points of debate over affirmative action (Son Hing et al., 2002). To the extent that conventional measures of ‘merit’ may reflect and reinforce inequities in the status quo, these findings raise debates about how much policymakers (and the public) are willing to promote equity for socially disadvantaged groups.

Here, it is worth connecting our findings to Brazil's institutional framework (the previously discussed, 2012 ‘Quota Law’). In some respects, our findings echo much of the existing law that reserves seats for low-income students and black and brown students: participants in our experiment similarly favoured low-income students and those with racialised appearances historically linked to marginalisation. That said, these are preferences that assume a quota system exists. A non-trivial portion of debates over quota systems, however, are about whether quotas should exist at all. Our findings thus raise a further question: why are certain quotas – such as those based on income – less polarising than those based on race, as observed in Brazil (Silva et al., 2023)?

Moreover, our findings highlight a challenge within the current institutional framework of Brazil and the challenge with crafting quota policies more broadly: the difficulty respondents have in categorising applicants based on physical appearance. The original framework of the Quota Law provides no clear criteria for determining who qualifies as black or brown, which creates troublesome ambiguity. This raises a broader question, also posed by Silva et al. (2023): why are such ambiguities less pronounced in other contexts, such as the United States?

Third, the few heterogeneous effects we do identify (comparing White and non-White respondents, left-leaning and right-leaning, and those who were experimentally primed to think about racial poverty) also raise some interesting questions. One possibility is that the ability of non-White Brazilians to shape public discourse around affirmative action and the use of racial quotas might move public opinion more than we might expect. For instance, Brazil's Black consciousness movement (‘movimento Negro’) has had some success at highlighting ongoing disparities faced by Afro-Brazilians in employment, education, and experiences with racial discrimination. This heightened awareness of inequities may shape in-group perspectives on necessary policy measures. At the same time, however, we do find what appears to be a marked consensus about who these quotas should target given that they exist. However, it is plausible that other psychological processes are at work that need further consideration. One approach would be to focus more directly on questions of merit and see how deeply held beliefs about what a ‘meritocratic’ society looks like modify preferences over group-based quotas (García-Sánchez et al., 2020).

The project also points towards new avenues for further research. Our study intentionally omitted categorising students into different racial or minority groups, instead giving participants the characteristics of the students and letting them sort the students into who they see as deserving and non-deserving. A strength of this approach is that it mimics the uncertainty about a student's group membership that exists in many countries, including Brazil. An interesting future line of inquiry would be to make group membership explicit and see how the public handles trade-offs among groups. How does the public believe seats should be allocated, for instance, among Indigenous and Black Brazilian students, given the relationship of each to marginalisation? And what internal calculus are respondents making when they prefer one over the other?

Supplemental Material

sj-docx-1-pla-10.1177_1866802X251317216 - Supplemental material for Who Should Benefit from Group-Based Quotas? Experimental Evidence from Brazil

Supplemental material, sj-docx-1-pla-10.1177_1866802X251317216 for Who Should Benefit from Group-Based Quotas? Experimental Evidence from Brazil by Anissa Nia Joseph, Nura Ahmad Sediqe and Juan Fernando Tellez in Journal of Politics in Latin America