Using Brazil’s Racial Continuum to Examine the Short-Term Effects of Affirmative Action in Higher Education

A. Student Sample (PSEU)

The authors conducted a survey of UnB students who were admitted through the vestibular system and matriculated between 2003 and 2005, a period including two admissions cycles before quotas (2-2003 and 1-2004) and three after quotas (2-2004, 1-2005, and 2-2005). We refer to this survey by its Portuguese acronym PSEU. Interviews, done online and face-to-face with an interviewer, covered a variety of topics: family background, preuniversity education, university admissions, university education, employment, expectations, and self-identified race. Photos of those students who participated in face-to-face interviews and admissions/academic records of all students were also obtained by the authors.

Data collection was as follows. UnB provided the authors a database with the names and contact information of students in the population of interest. Interviews were completed between May 2007 and August 2008 beginning with the oldest cohort (2-2003) and ending with the youngest (2-2005). An initial invitation to participate in the face-to-face interview was done by email and a secondary invitation done by phone. We attempted to contact all students in the population of interest by email, but due to resource constraints, we were only able to attempt to contact a random subset of students by phone. Face-to-face interviews took place in the UnB Department of Economics. With consent, a photo was taken of the respondent’s student identification card, which had a standard photo taken upon matriculation. An invitation to participate in the online interview was sent by email to those who had not yet responded to our initial invitation. To minimize the salience of race and affirmative action in the recruitment process, the title of the survey was general (“University Education Survey”), and neither race nor affirmative action was raised in any contacts with potential respondents. 2,286 students in the population of interest completed the PSEU, which amounts to a participation rate of almost 40 percent. 799 face-to-face interviews were conducted, and 748 of these had viable photos. Some photos were taken but were too blurry to be useful, and others were not taken because the camera was temporarily out of batteries. About 9 percent of the email addresses provided by the university were invalid (misspelled or deactivated), and about 25 percent of the phone numbers were incorrect.

Appendix table 1, which the reader can find on the corresponding author’s website, investigates the representativeness of the PSEU. Given that we have comprehensive admissions and academic records, we are able to compare the sample and population with respect to a number of dimensions (gender, place of family residence, quota participation, semester of matriculation, subject of study, number of application attempts, vestibular score, college GPA, and QSC completion). The table suggests that there are only a small number of significant differences between the PSEU and population, and that such differences are modest. For example, the difference in GPA, albeit significant, is about 13 percent of one standard deviation of GPA. At 48.3 percent, the sample is more female than the population. But this appears to be a common feature of many surveys, for example, the General Social Survey. It turns out that the difference in social science is entirely explained by the overrepresentation of economics students in the sample, which undoubtedly arises from the fact that the interviews took place in the economics department. In any case, it may be helpful to mitigate potential bias through weighting. To do so, we run a probit regression of sample participation on the set of characteristics in Appendix table 1 and construct sample weights equivalent to the inverse of the predicted probability of participation.

B. Applicant Sample (QSC)

UnB provided the authors with a survey of applicants who took the vestibular exam between 2004 and 2005, a period including one admissions cycle before quotas (1- 2004) and three after quotas (2-2004, 1-2005, and 2-2005). We refer to this survey by its Portuguese acronym QSC. Applicants submitted the 18-question QSC upon registration for the vestibular, that is, prior to taking the exam. It asked about selfidentified race, socioeconomic status, and vestibular preparation. 1-2004 was the first admissions cycle that included questions on racial identity. However, note that the QSC was not the formal document with which applicants identified themselves as negro to apply under racial quotas. 24,252 applicants in the population of interest completed the QSC, which amounts to a participation rate of about 50 percent. An important caveat is that response rates were falling during the period from roughly 84 percent in 2-2003 to 36 percent in 2-2005. Appendix table 1 suggests that there are many significant differences between the QSC and population. Given these substantial biases, it is vital to use sample weights.

C. Empirical Strategy

In what follows, we estimate the effect of racial quotas on preuniversity effort, college academic performance, and black identity. To do so, we employ a differencein- difference model interpreting the policy as a treatment on individuals with brown or dark skin tone. The following equation is estimated:

where Y is an outcome of interest, I is an indicator for application/matriculation postquotas, R is an indicator for self-reported race/skin tone (pardo, preto, Asian, Indigenous), T is an indicator for semester of application/matriculation, and X is a vector of controls for socioeconomic status, gender, and subject of study. Thus, the effect of racial quotas on pardos is τ_pardo, and the effect of racial quotas on pretos is τ_preto.

Making use of photos taken in the face-to-face interviews, we asked a panel of Brazilian reviewers to rate the skin tone of the subject in each photo from one (light) to seven (dark). Scores were standardized by reviewer, standardized scores were averaged by photo, and average standardized scores were sorted into quintiles. In the tables and text, “lightest quintile” indicates the lowest 20 percent of average standardized scores, “second quintile” indicates the next 20 percent, and so on. Appendix table A1 provides the distribution of self-reported race by skin tone quintile. The following equation is estimated:

where Y is an outcome of interest, I is an indicator for application/matriculation postquotas, Q is an indicator for skin tone quintile, T is an indicator for semester of application/matriculation, and X is a vector of controls for socioeconomic status, gender, and subject of study. Thus, the effect of racial quotas on skin tone quintile q is τ_q.

A. Effects on the Student Profile

The principal objective of the policy was to raise the proportion of black and socioeconomically disadvantaged students at UnB. Based on the PSEU, table 4 compares the racial composition of students before and after quotas. The percentage of students who self-identified as negro, or black, increased from 15.3 percent prequotas to 20.6 percent postquotas. While the percentage of students who self-reported as preto (dark-skinned) increased from 5.6 percent to 9.0 percent, the percentage who self-reported as pardo (brown-skinned), branco (light-skinned), Asian, and indigenous remained roughly constant. Although table 4 suggests that racial quotas raised the proportion of negros, using self-reported race might be problematic. Figure 1 depicts histograms of average standardized skin tone based on ratings of student photos by a panel of Brazilian reviewers. Light skin tone is toward the left and dark skin tone toward the right. From pre to postquotas, the distribution of skin tone shifts to the right, which illustrates that students matriculating in the postquota period had darker skin tone than students matriculating in the prequota period.

However, the question remains whether racial quotas also raised the proportion of socioeconomically disadvantaged students. Using the QSC, table 5 compares applicants who were admitted but would not have been if the quota system had not existed (displacing) with those who were not admitted but would have been if the quota system had not existed (displaced). To identify the two groups, it was assumed that the counterfactual removal of the quota system would not have affected who applied or performance on the vestibular, and that the number of applicants admitted, by semester and department, would have remained identical.⁵ About 95 percent of the displacing identified as negro compared to 16 percent of the displaced.⁶ 71 percent and 27 percent of the displacing were pardo and preto, respectively, while about 31 percent and 2 percent of the displaced were. Furthermore, we find that displacing applicants were, by many measures, from families with significantly lower socioeconomic status. For instance, 34.5 percent of the displacing and 57.7 percent of the displaced had a mother with a college degree. 59.5 percent of displacing and 34.0 percent of displaced applicants had family residence in Distrito Federal outside of Brasilia, which is meaningful because the average household income of families living in Distrito Federal outside of Brasilia was considerably lower than that of families living in Brasilia (PDAD 2004). Differences in self-reported family income and public secondary school attendance paint the same picture.^7,8

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Distribution of Skin Tone

Source: PSEU.

Note: Histograms of average standardized skin tone are based on ratings of student photos. Light skin tone is toward the left and dark skin tone toward the right. Sample weights are used.

B. Effects on Preuniversity Effort

Examining both applicants and students, we estimate the effect of racial quotas on effort in university admissions. A race-based affirmative action program may theoretically have two kinds of effects on effort (Fryer and Loury 2005a; Fryer, Loury, and Yuret 2008; Holzer and Neumark 2000). On one hand, affirmative action may increase the marginal return to actions that raise the likelihood of college admission. Changes in admissions standards might relocate some individuals who otherwise would have had little chance of selection to the margin of selection, thereby inspiring effort. On the other hand, affirmative action may decrease the marginal return to actions that raise the likelihood of admission. Changes in admissions standards might relocate some individuals who otherwise would have been at the margin of selection to an intramarginal position, thus reducing effort. Hence, although racial quotas have an unambiguous effect on the conditional chances of admission, they have an ambiguous effect on returns to effort. Applying this logic to Brazil, it is important to remember that the margin of selection is actually multi-dimensional, as there exists a hierarchy of departments with vastly different minimum scores for selection. To the extent that the objective is to gain admission to the most selective department possible, few applicants are intramarginal.

To begin, we consider two measures of effort: the number of times that an applicant took the UnB vestibular and whether an applicant/student took a cursinho, a 6–12-month course offered by a private company to prepare for the UnB vestibular. While both variables are actions informative about returns to effort, they are imperfect, and it is not obvious what ideal measures one could obtain in practice. table 6 displays the regressions. As Columns A and B indicate, for each of the measures of effort, the coefficient on pardo applicant in the postquota period is positive and significant, whereas the coefficient on preto applicant in the postquota period is not significant. This may suggest that racial quotas increased the number of times that a self-reported pardo applicant took the vestibular by about 0.24 and increased the likelihood that a pardo applicant took a cursinho by about 5.8 percentage points. Alternatively, this may suggest that racial quotas tended to attract pardo applicants who were more likely to take the vestibular multiple times and more likely to take a cursinho. Comparing the profile of UnB applicants pre- and postquotas may help to investigate the issue of selection into the applicant pool. Appendix table 2, which the reader can find on the author’s website, indicates that postquotas the applicant pool became more pardo, slightly less female, and slightly more socioeconomically advantaged. For this reason, we control for these factors in the regressions. Note that these factors do not appear to explain the results, that is, the pool is less female and more socioeconomically advantaged, but female and public secondary school are positively associated with cursinho. Nevertheless, it is still possible that we are not adequately accounting for changes in the applicant pool.

We also estimate the effect of quotas on the likelihood that students took a cursinho to prepare for the UnB vestibular. In Column C, neither the coefficient on pardo student in the postquota period nor that on preto student in the postquota period is significant, so even if there may be an effect on pardo applicants, there may not be an effect on pardo students. However, in Column D, the coefficient on the lightest quintile in the postquota period is positive and significant, which may mean that students with the lightest skin tone were more likely to take a cursinho following the introduction of racial quotas.

Given that the margin of selection is multidimensional, it remains to establish whether the adoption of quotas in admissions inspired applicants/students with relatively dark skin tone to apply to more selective departments. Evidence on choice of department would yield insight regarding returns to effort. Based on complete admissions data from 2-2004, UnB departments were assigned values according to their rank by minimum score for selection; the most selective departments were assigned the highest values and the least selective the lowest values. Using this measure, the regressions in Columns E and F demonstrate that for applicants and students, preto in the postquota period is positive and significantly related to selectivity of department, while preto is negative and significantly related. Thus, although self-reported preto applicants/students tended to apply to less selective departments than brancos, the racial gap in selectivity decreased with the implementation of racial quotas. This may suggest that pretos are better characterized as marginal than intramarginal, indicating that it is unlikely that returns to effort had diminished with the introduction of quotas. Column G, which makes use of photo ratings, confirms that selectivity of department varied significantly with skin tone but does not confirm that the policy narrowed the gap in selectivity, as the coefficient on darkest quintile in the postquota period is positive but insignificant. Taken together, the findings in the table imply that racial quotas had not reduced the preuniversity effort of applicants or students. Some evidence, albeit tenuous, may signify that the policy raised effort.

C. Effects on Racial Gaps in College Academic Performance

Focusing on UnB students, we examine the effect of quotas on racial gaps in college GPA. Racial disparities in college academic performance reflect a myriad of historical, social, and behavioral factors. The adoption of racial quotas might impact such disparities by modifying behaviors and/or the composition of students enrolled. Regardless of the underlying mechanisms, changes in racial gaps in GPA measure the extent to which the policy ameliorates or exacerbates college “mismatch.” table 7 displays the regressions. Grades range from zero (bad) to five (good); the mean and standard deviation of GPA are 3.82 and 0.65, respectively. The first three columns include students in all departments, and the last six columns include students in only the 50 percent most selective departments.

Columns A and B imply that the policy had no effect on racial gaps in GPA and also that there were few racial gaps in the first place. Only the coefficient on indigenous is significant. Restricting the sample to students in the 50 percent most selective departments uncovers additional findings. Column D reveals significant disparities, especially between brancos and pretos. The coefficients on pardo in the postquota period and preto in the postquota period are insignificant, so the policy did not appear to widen or narrow racial disparities among students in selective departments. Controlling for vestibular score in Column E, racial gaps in GPA slightly decrease in magnitude but remain significant indicating that they are not merely an artifact of differential performance on the vestibular. Nevertheless, Columns F and G do not yield evidence for the existence of racial disparities; the puzzling coefficient on third quintile in the postquota period may perhaps be explained by changes in the composition of students and/or low sample size.

It is also fruitful to compare the academic performance of displacing and displaced students.⁹ As Column C shows, the GPA of displaced students is lower than that of intramarginal students but not significantly so, while the GPA of displacing students is significantly lower than that of displaced and intramarginal students. Restricting the sample to students in the 50 percent most selective departments (Column H) widens the gap in GPA between displacing students and everyone else. That difference is roughly the same as the difference between brancos and pretos in Column D. However, controlling for vestibular score in Column I, the coefficient on displacing students falls to the extent that it is no longer significantly different from the coefficient on displaced students.

In summary, there may have been modest racial disparities in GPA among students in selective departments, but the introduction of racial quotas did not seem to impact these in any way. Displacing students had somewhat lower GPAs than displaced and intramarginal students, an outcome partially attributable to differential performance on the vestibular. Thus, the evidence is inconsistent with large “mismatch” effects with respect to college GPA (see Rothstein and Yoon, 2008). Factors independent of the policy, including racial differences arising prior to college attendance, appear to be driving any racial disparities in academic performance.

D. Effects on Racial Identity

At UnB, quotas are for negros. Given the competitiveness of admissions, there is substantial incentive to self-identify as black. Moreover, the university provides quota students with an array of programs that reinforce and foster investments in black identity, including a space on campus to study, meet, and have activities. Therefore, the policy of racial quotas may have caused some students to consider themselves black thus placing them on a new life path, while it may have caused others to self-identify as black solely for the purpose of admissions. One of the main empirical challenges is to differentiate between actual change in black identity and racial misrepresentation. Actual change is enduring, substantial, and genuine, whereas misrepresentation is temporary, superficial, and opportunistic. Although we rely on both the QSC and PSEU to measure black identity, the PSEU better minimizes misrepresentation. PSEU respondents had little incentive to misrepresent themselves because they had already matriculated, understood that their responses were absolutely confidential, and were unaware of the research objectives.

We begin the analysis by using the QSC to estimate the effect of racial quotas on black identity. The first two columns of table 8 involve the sample of applicants. The coefficients on both pardo and preto applicants in the postquota period are positive and significant. This may suggest that racial quotas increased the likelihood that a pardo applicant self-identified as negro by about 25 percentage points and increased the likelihood that a preto applicant self-identified as negro by about 3 percentage points. Alternatively, this may suggest that racial quotas attracted pardo applicants who tended to consider themselves black. Appendix table 2, which the reader can find on the author’s website, may help to shed light on the issue of selection into the applicant pool. The appendix table shows that postquotas the applicant pool became more negro and more pardo, which certainly does not rule out the possibility that the rise in black identity was attributable to change in the applicant pool. More evidence is required to distinguish between the hypotheses.

The last six columns of table 8 involve the sample of students who completed both the QSC and the PSEU. They responded to the same race questions at two points in time, first as an applicant and later as a student. In Columns C and D, the dependent variable is negro on the QSC. Here, the results mirror the results obtained with applicants. It appears that racial quotas increased the likelihood that pardo students self-identified as negro on the QSC by about 27 percentage points. In Columns E to H, the dependent variable is negro on the QSC and non-negro on the PSEU. If change in the applicant pool explained the increase in black identity on the QSC, we would expect consistency between the QSC and PSEU. However, the evidence indicates otherwise. The coefficient on pardo in the postquota period is positive and significant, while the coefficient on fourth quintile in the postquota period is positive, though insignificant. This may suggest that quotas raised, by about 15 percentage points, the likelihood that pardos self-identified as negro on the QSC but not on the PSEU. Hence, there was systematic inconsistency in black identity associated with the adoption of quotas. If results from students who responded to both surveys may provide insight more generally, they imply that part of the rise in black identity among pardos on the QSC was not a manifestation of change in the applicant pool but of racial misrepresentation.

Using students who completed both surveys, table 9 further investigates the role of quotas in the evolution of responses between the QSC and PSEU. As the top panel shows, self-identified brancos and pretos exhibited consistency across surveys before and after the implementation of racial quotas. Most brancos reported that they were non-negro on both surveys, and most pretos reported that they were negro. Echoing the findings in table 8, table 9 illustrates that following the adoption of quotas pardos were significantly more likely to report that they were negro on the QSC but non-negro on the PSEU. 10.8 percent did so prequotas, whereas 31.2 percent did so postquotas. Moreover, switching responses in the other direction became less common. 12.8 percent of pardos prequotas and 5.8 percent postquotas said that they were non-negro on the QSC but negro on the PSEU. Intriguingly, although the overall rate of consistency between surveys decreased for pardos, the percentage who reported that they were negro on both surveys increased from 11.1 percent to 24.1 percent.

As the bottom panel shows, for students with skin tone in the lightest three quintiles according to photo ratings, there were no significant changes in the pattern of responses between the QSC and PSEU pre- and postquotas. In contrast, students in the fourth quintile demonstrated significant changes analogous to those demonstrated by pardos in the top panel. This makes sense given that Appendix table A1 indicates 70 percent of students in the fourth quintile self-identified as pardo and 23 percent as branco. While the rate of consistency between surveys remained constant for students in the darkest quintile, an insignificantly greater share of respondents shifted toward reporting negro on the QSC but non-negro on the PSEU, and a significantly greater share of respondents shifted toward reporting negro on both surveys. Indeed, the percentage of students in the darkest quintile who reported they were negro on both surveys increased from 45.2 percent prequotas to 72.2 percent postquotas. To summarize, table 9 reveals evidence that points to a move toward racial misrepresentation among some pardos, especially those in the fourth quintile, but also points to a move toward consistent black identity among other pardos, especially those in the darkest quintile. In what follows, we explore genuine change in identity using the entire sample of students, not just the subsample of students who completed both surveys.

That the policy of racial quotas induced racial misrepresentation among some individuals by no means precludes the possibility that it also inspired genuine change in black identity among others. Using the PSEU, which minimizes misrepresentation, table 10 compares black identity by race/skin tone and photo ratings pre- and postquotas. Both before and after quotas, almost all pretos and almost no brancos selfidentified as negro. However, the percentage of pardos who self-identified as negro increased significantly with the implementation of racial quotas from approximately 18.8 percent to 24.4 percent. Photo ratings are also informative. Notably, there was no significant change in black identity for students with skin tone in the fourth quintile, the group most likely to switch from negro to non-negro between the QSC and PSEU. About 20.5 percent self-identified as negro prequotas and 19.8 percent postquotas. In contrast, black identity increased from 57.1 percent to 77.1 percent for students with skin tone in the darkest quintile.

table 11 explores the effect of racial quotas on black identity with the PSEU and difference-in-difference models. As the table shows, the coefficients on 2-2003 and 1-2004 are not significant in any of the specifications indicating that there was no trend in black identity prior to the implementation of quotas. In Columns A and B, the coefficient on pardo in the postquota period is positive and significant, which may suggest that racial quotas increased the likelihood that a pardo student selfidentified as negro by 6 percentage points. In Columns C and D, the coefficient on darkest quintile in the postquota period is positive and significant. This may suggest that racial quotas increased black identity by roughly 20 percentage points among students with skin tone in the darkest quintile. To elucidate the findings, it is important to understand that of students with skin tone in the darkest quintile, about 47 percent self-identified as pardo and 48 percent as preto (see Appendix table A1). Therefore, the table may provide evidence of change in black identity among darkskinned students. These results are consistent with the incentives created by the policy. Indeed, the incentive to apply under the quota system was substantial given the competitiveness of admissions, and programs for quota students reinforced and fostered investments in black identity. In this way, the policy may have placed some students on a new life path, the initial steps in the dynamic construction of black identity. These results are also consistent with the literature on the determinants of racial identity in economics, sociology, and other social sciences (Darity, Mason, and Stewart 2006; Golash-Boza and Darity 2008; Harris and Sim 2002; Kibria 1997; Lee and Bean 2004; Nagel 1994).