Literature review

There was an increase in the number of students attending higher education institutions in Brazil in the last decades (^{Caseiro, 2016}; ^{McCowan & Bertolin, 2020}; ^{Schwartzman, 2014}). However, participation in tertiary education in Brazil continues to vary remarkably among different population groups (^{Artes & Ricoldi, 2015}; ^{Pedrosa et al., 2007}; McCowan & Bertolin, 2020). For instance, the percentage of Brazilians aged 18 to 24 attending universities and colleges varied from 0.83% in the first income quintile to 43.7% in the last. 20.4% of Whites/Asians attended higher education institutions, while for Blacks/Pardos/Indigenous this number was only 5.3% (Pedrosa et al., 2007). In spite of these inequalities, McCowan and Bertolin (2020) argued that there were some improvements for minorities in the accessibility to higher education institutions in Brazil, in part due to affirmative action policies recently implemented in public institutions of higher education, and also because of loan and grant policies for private universities and colleges, such as the Student Financing Fund (In Portuguese: Fundo de Financiamento Estudantil, FIES) and the University for All (In Portuguese: Programa Universidade para Todos, Prouni) (McCowan & Bertolin, 2020; ^{Francis & Tannuri-Pianto, 2012a}, 2012b; Pedrosa et al., 2007; ^{Telles & Paixão, 2013}).

Different authors empirically addressed these points. ^{Artes and Ricoldi (2015}) analyzed the recent evolution in the student’s profiles using Census data from 2000 and 2010. They observed that the proportion of Whites attending higher education institutions in Brazil was much larger than the proportion of Blacks and Pardos, but that differences had shrunk in the period. Similarly, ^{Caseiro (2016}) analyzed the influence of race inequality on chances of ending secondary education and beginning tertiary education in 2004 and 2014. The author observed that the propensity of Whites to attend higher education intuitions was larger than for non-Whites, but that differences between races decreased, and ^{McCowan and Bertolin (2020}) verified similar trends.

Likewise, ^{Ribeiro et al. (2015}) analyzed transitions in the educational system. They concluded that Blacks and Pardos had lower probabilities to make transitions in general, including from the secondary to the higher educational level, even after controlling for socioeconomic variables. In addition, results for Blacks were even worse than for Pardos. However, similarly to the previously mentioned authors, Ribeiro et al. (2015) also observed a general decrease in the effects of race in the probabilities of transition.

In a similar vein, ^{Marteleto (2012}) studied educational disadvantages in Brazil from 1982 to 2007. She observed that inequalities associated with the dichotomy White versus non-White persisted, but declined. Moreover, differences between Blacks and Pardos became non-significant at the end of the analyzed period. She proposed two explanations for these results, structural changes and shifts in racial classification, and found evidence for both. Concerning the first, socioeconomic gaps between Blacks and other groups in the population decreased. Regarding the second, there was a “darkening” with education, as more educated individuals showed a greater propensity to self-declare Black as skin color.

Moreover, the proportion of non-Whites tends to be smaller in more prestigious courses (^{Artes & Ricoldi, 2015}; ^{Ferreira, 2020}). Similarly, ^{Carvalhaes and Ribeiro (2019}) and ^{McCowan and Bertolin (2020}) observed that disadvantaged students were mostly directed to lower quality for-profit institutions and to less prestigious courses.

There are other factors besides race that also impact higher education institutions attendance, such as household income, parents’ educational attainment and household arrangements, all also correlated with race (^{Caseiro, 2016}; ^{Silva & Hasenbalg, 2002}). Silva and Hasenbalg (2002) observed that Whites, individuals living in higher income households, with household heads with higher levels of formal education, and in smaller households tended to progress more effectively in the Brazilian educational system. Household income also impacts on the propensities of transition, as students from richer family show greater propensities to make the transitions in the different levels of the educational system. Nonetheless, it was also observed a decrease in the effects of income in the probabilities of transition for the elementary and secondary levels (Caseiro, 2016; ^{Marteleto, 2012}; ^{Ribeiro et al., 2015}).

These mentioned authors described some of the factors associated with the differences in higher education institutions attendance. However, the focuses here are the consequences of changes in the students’ selection process, including affirmative action policies. Concerning these points, ^{Epple et al. (2008}) analyzed the effects of a ban of an affirmative action based on race in the US and concluded that there would be a substantial impact on the racial composition of colleges, decreasing the proportion of Blacks and of racial diversity. In a similar vein, ^{Long (2004}) analyzed changes in application behavior of minorities and non-minorities due to a ban of race-based affirmative action in Texas and in California. He observed that both groups significantly changed their behavior and the gap between the number of SAT score reports sent by non-minority and minority students widened, possibly due to a decrease in the expectation of minorities of being accepted (^{Dickson, 2006}). ^{Bertrand et al. (2010}) examined an affirmative action program directed to individuals from a lower-caste in India. They observed that the displaced upper-caste applicants were richer than the displacing lower-caste applicants. Nonetheless, this affirmative action excluded other more disadvantaged groups, as the lower-caste admitted were those from stronger socioeconomic backgrounds among the caste.

Similarly, many authors analyzed the associations of different affirmative action policies in Brazil with the profile of the students poll at public institutions of higher education. ^{Piotto and Nogueira (2016}) studied data for a bonus policy that was created in 2006 in Universidade de São Paulo (USP). The results of this affirmative action indicated that the proportion of students from public schools and from low-income families attending the university increased; however, this same tendency was not observed for individuals whose father had low educational attainment. This indicated that the policy helped individuals to overcome some of the difficulties imposed by low SES, but was ineffective to the father’s poor cultural background. ^{Francis and Tannuri-Pianto (2012a}, 2012b) examined the quota policy in Universidade de Brasília (UnB), implemented in 2004, and observed that the proportion of black and dark-skinned students increased, and that displaced applicants were from higher SES level. ^{Ferreira (2020}) observed that affirmative action policies were effective in increasing the proportion of non-whites in most universities in Brazil.

In particular for UFMG, ^{Carnevali and Amaral (2016}) analyzed whether the bonus policy in UFMG had changed the student body between 2011 and 2014 and verified relative stable numbers, a initially non-expected consequence of the policy. In a similar vein, ^{Aranha et al. (2012}) analyzed the effects of the implementation of the bonus policy and the effects of the Reuni policy. Before both policies, the proportion of students from public schools who applied to UFMG was decreasing. After this policy, the tendency was of stabilization, with a proportion close to 45%. Hence, as emphasized by ^{Peixoto and Braga (2012}), the implementation of the bonus policy did not alter the demand for tertiary education among minorities as expected. The number of applications from public schools fell between 2002 and 2007 and was approximately stable between 2007 and 2011. Concerning the approved in the university entrance exam, before the policy they were around 25% of the total, a number much smaller than those who applied to a tertiary education slot, close to 50%. After the implementation of the bonus policy, close to 47% of those selected in the university came from a secondary public school, a similar number than the proportion who applied.

These studies described race and socioeconomic trends for higher education institutions attendance. These trends have as their determinants a myriad of racial and socioeconomic factors that are directly affected by the specific policies implemented in the selection process. Next section details the main changes that occurred in UFMG’s selection process.

Main policies implemented in the selection process at UFMG between 2006 and 2015

There were major changes in UFMG’s selection process between 2006 and 2015, as briefly mentioned in the introduction. They are further detailed in this section and are summarized in Chart 1.

In order to decrease attendance inequalities between population groups and increase the proportion of Blacks/Pardos/Indigenous students and those from low-income households in public universities, many institutions in Brazil implemented affirmative action policies in the last decades. The Universidade Estadual do Rio de Janeiro (UERJ) was the first to implement such policies in 2002 and then similar policies spread to a large number of higher education institutions all over Brazil (^{Telles & Paixão, 2013}).

More specifically at UFMG, it was implemented a bonus policy between the years of 2009 and 2012 (^{Aranha, 2009}). This policy increased 10% the grades obtained in the entrance exam for individuals who had attended public schools in the last seven years of elementary and secondary education. In addition, an extra 5% bonus was given to those who considered themselves Blacks/Pardos/Indigenous. This policy with minor changes was valid until the entrance exam of 2012. In 2012 it was approved the national federal law of quotas that was implemented in federal higher education institutions (^{Telles & Paixão, 2013}). Beginning at the entrance exam of 2013, at least 12.5% of the students in each course had to have done all their secondary schooling in a public institution. This number increased to 25%, 37.5% and 50% respectively in the years of 2014, 2015 and 2016. Since 2016 this is the implemented affirmative action policy in federal higher education institutions in Brazil.

Besides these affirmative action policies, other policies may also have affected the proportion of minorities in public institutions of higher education in Brazil, such as the Reuni, which promoted a remarkable increase in the number of slots in public federal universities in Brazil between 2008 and 2012. In particular at UFMG, the increase was from 4.6 thousands to 6.6 thousands slots annually (^{Aranha et al., 2012}; ^{Lima & Machado, 2016}).

There was a third major change in the selection process of public institutions of higher education in Brazil that was the introduction of Enem/SiSU as the main procedure in the students’ selection process. In particular at UFMG, there were two main changes in the exams that were applied to applicant’s since 2008. Until 2010 there was an UFMG’s exam in two stages. From 2011 to 2013, four exams of the Enem (mathematics, language, science and humanities) were used as the first stage, while the second stage continued to be an UFMG’s own exam with the incorporation of the composition from ENEN. Since 2014 the five exams of Enem and the Unified Selection System (SiSU) began to be used in UFMG as the main toll for student’s selection (^{Lima & Machado, 2016}; ^{Nogueira et al., 2017}; ^{Silveira et al., 2015}). All these changes may also have affected the proportion of minorities attending UFMG.

The paper analyses data of UFMG’s applicants between 2006 and 2015. Thus, most changes had already occurred in the end of the analyzed period. Notice, however, that the periods in which the policies were implemented are quite short and different between them. The bonus policy was the implemented affirmative action policy for only four years. The quota policy is the current affirmative action policy, nonetheless, the database used in this paper is from 2006 to 2015. The quota policy was still being implemented in this last year. Reuni was a policy that increased the number of slots at UFMG between 2009 and 2010. Before this policy the number of slots was smaller and, since then, UFMG has a larger number of slots. The main current method of students’ selection is the ENEN/SiSU. In our database it was present in the last three years, while UFMG´s exam in two stages, and the Enem in the first stage and a UFMG’s exam in the second stage are respectively present in five and three years in the analyzed period.

These differences are partially taken into account while discussing the empirical results because data is for each year, but pooled for all the period. Nonetheless, policies’ effects may be different depending on the time since its implementation. I do not tackle this limitation in the paper. Moreover, these policies interact with each other. For instance, the effect of a bonus policy may vary due to the Reuni policy. I acknowledge this aspect, however, the empirical analysis did not include interactions, in part due to problems of colinearity.

Based on Chart 1, there are different possibilities of comparative analysis. The main characteristics of the selection process did not change between 2006 and 2008. Between 2008 and 2009, two main changes occurred in the selection process at UFMG: the implementation of the bonus policy and the increase in the number of slots. Between 2009 and 2010, the number of slots increased further. The Enem replaced the UFMG´s own exam as first stage of the selection process in 2011. Between 2012 and 2013, the affirmative action policy changed from bonus to quota policy. The quota policy increased in coverage between 2013 and 2016. The second stage of the selection process was suppressed between 2013 and 2014.

All these changes presented in box 1 enables different comparisons regarding the diversity of the student body: 1) Which were the effects of the implementation of the bonus policy? 2) Which were the impacts due to the use of the Enem as the first stage exam of the selection process? 3) Which were the effects of the increase in the number of slots? 4) Which were the effects of the change of the bonus policy for the quota one? 5) Which were the impacts of the increase in coverage of the quota policy? 6) Which were the effects of the suppression of the second stage of the selection process? All these points are empirically addressed.

Methodology

Empirical strategy and variables

The empirical strategy is divided in two type of analysis. The first is a descriptive analysis of the main trends for the proportion of students at UFMG of different groups of the student body. The second is based on logistic models.

First, the paper describes the trend for low-income students. Students from low-income households are those who lived in households with a total income lower than two Brazilian minimum wages (BMW). A dummy was created with this information: (1 - Low-income, 0 - Non-low-income). Notice, however, that the real value of the BMW increased between 2006 and 2015. Thus, everything constant, it is expected an increase in the proportion of low-income students simply due to this real value change. This fact is taken into account in some analysis, which includes the real value of the BMW as a continuous variable.

After this, the paper describes the proportion of Whites and Blacks/Pardos. The database has information about ethnic group/skin color in five categories: White, Pardo, Black, Asian and Indigenous. These last two groups are not numerous at UFMG. As this variable is self-declared, there is a sixth possibility of answer for those who did not want to declare, which is a quite numerous category. A categorical variable was created with this information: (0 - Whites, 1 - Blacks/Pardos, 2 - Did not declare).

Then, trends for types of school administration are shown. Secondary schools in Brazil are classified into private and public. There are many differences between private and public secondary schools. For instance, ^{Cavalcanti et al. (2010}) found that the performance of students from public schools was smaller than those of student from private institutions, even after controlling for a set of controls. Among the public secondary schools, there are three types of administration: state, federal and municipal. Academic performance of students from federal secondary schools tend to be similar to those of students from private schools, and much higher than those of students from state or municipal schools (^{Golgher, 2010}). These differences in academic performance directly impact in the possibilities of becoming a student at higher education institutions, such as UFMG. A categorical variable was created with this information: (0 - Municipal, 1 - Private, 2 - Federal, 3 - State).

The descriptive study is complemented by logistic models for two types of analyses. The first one uses data from all years pooled with the objective to observe general associations between the policies that were described in the previous section and the variables cited above. The dependent variables are dummies derived from the variables cited above.

The explanatory variables in the first group of models are the following: whether in the year there were UFMG’s exams in both stages (1 - Yes, 0 - No); whether the year was before, during or after the Reuni policy (1 - After, ½ - During, 0 - Before), whether there was a Bonus policy (1 - Yes, 0 - No), whether there was a Quota policy (1 - Yes, 0 - No), whether the course was a SiSU one (1 - Yes, 0 - No), whether the student was selected for the second semester of the year (1 - Yes, 0 - No), and a continuous variable for BMW real value.

The second type of analysis aims to observe temporal changes regarding individual features for specific type of households. The data is specific for the years 2006, 2011 and 2015, which were analyzed separately. The dependent variables is whether the student was: a low-income (1 - Yes, 0 - No), Black/Pardo (1 - Yes, 0 - No), from a private secondary school (1 - Yes, 0 - No), or from a state secondary school (1 - Yes, 0 - No). The students from federal or municipal secondary schools were much less numerous and are not analyzed.

Concerning the individual explanatory variables, the models included dummies for: sex (1 - Male, 0 - Female), civil status (1 - Single, 0 - Other), whether the secondary school was at evening (1 - Yes, 0 - No), whether the student had attended a preparation course to do the higher education institution exams, none as pré-vestibular (1 - Yes, 0 - No), whether the student had already an undergraduate degree (1 - Yes, 0 - No), whether the student had the following assets in the household (Domestic servant, television, fridge, computer, car: 1 - Yes, 0 - No). The models also included categorical variables: skin color (the same as above); whether the student previously knew how to read foreign languages (0 - None, 1 - Spanish, 2 - Another language, 3 - More than one foreign language); whether the student had worked while studying at the secondary level (0 - No, 1 - Up to 20 hours weekly, 2 - More than 20 hours weekly); number of individuals in the household (0 - One, 1 - Two or three, 2 - Four or five, 3 - Six or more), father and mother schooling attainment (0 - Less than elementary, 1 - Elementary, 2 - Secondary, 3 - Tertiary, 4 - The student did not know his/her father/mother schooling level); the type of secondary school that the student had attended (0 - Municipal, 1 - Private, 2 - Federal, 3 - State).

All the analysis were done with Stata 12.

Results

Descriptive statistics

This section presents the trends for the proportion of students by semester for students from low-income households, Whites, Blacks/Pardos, students from private, state or federal secondary schools. The objective is to give an overview of the trends and to relate them with the main policy changes described in Chart 1.

Graph 1 shows the trend for the proportion of students from low-income households. The tendency is clear. Until 2008 the values were close to 6%, increased remarkably between 2008 and 2012, and remained stable between 2012 and 2015. As shown in Chart 1, two policies were implemented in 2009, the Reuni policy and the bonus policy. Both may have had an impact on this trend. The implementation of the quota policy in 2013 did not seem to have changed the trend, as the bonus policy already existed.

Source: Authors’ elaboration.

GRAPH 1 PROPORTION OF LOW-INCOME STUDENTS BY SEMESTER AT UFMG 

Graph 2 compares the proportions of Whites and of Non-whites (Notice that both categories sum 100%. Those who did not declare were excluded here). Before the Reuni and Bonus policies, the proportions of Whites were close to 70%. After these policies, until the implementation of Enem/SiSU, values were approximately 50%, and after the implementation of this last policy, the number was close to 52%.

Source: Authors’ elaboration.

GRAPH 2 PROPORTION OF WHITES AND NON-WHITES BY SEMESTER AT UFMG 

Graph 3 compares the proportions of students from public and from private secondary schools. Before the Reuni and Bonus policies, the proportions of students from private schools were around 65% and, after these policies, the values oscillated around 55%. The drop in the proportion of students from private schools in 2009 was accompanied by an increase in the proportion of students from public schools in the same year, when the bonus e Reuni policies were implemented.

Source: Authors’ elaboration.

GRAPH 3 PROPORTION OF STUDENTS FROM PUBLIC AND PRIVATE SECONDARY SCHOOLS BY SEMESTER AT UFMG 

Graph 4 shows the trends for the proportion of Whites and the proportion of students from private secondary schools. Tendencies are very similar, with a remarkable decrease in 2009, and reasonable stable numbers before and after this year. The variation for Whites was even larger than for private school, from 70% to 50% around this year. However, notice that there was an apparently slight increase in the proportion of Whites between 2011 and 2015, possibly due to the change from the bonus policy to the quota policy, which was in its process of implementation, or due to the implementation of the Enem/SiSU as the main method of students’ selection.

Source: Authors’ elaboration.

GRAPH 4 PROPORTION OF WHITES AND STUDENTS FROM PRIVATE SCHOOLS BY SEMESTER AT UFMG 

Graph 5 shows the proportions of students from state and federal secondary schools. The number of students from municipal secondary schools is small and the results are not shown. For the proportion of students from federal schools, the tendency is not similar to the one for students from public schools, shown in Graph 3, or form state schools, shown in Graph 5. Trends for public schools and for state schools are quite close, in part because students from state schools made the majority of students from public schools since 2009. For students from federal secondary schools, there was a slight decrease between 2006 and 2013 and a remarkable increase in 2014. This was the year of the implementation of the SiSU with the quota system. Besides, it can be noticed that there is a tendency that the proportions for students from federal secondary schools are greater in the first semester of the year, suggesting the higher level of performance of these students when compared to others, as already mentioned.

Source: Authors’ elaboration.

GRAPH 5 PROPORTIONS OF STUDENTS FROM STATE AND FEDERAL SCHOOLS BY SEMESTER AT UFMG 

Conclusion

Brazil is a highly unequal country and tertiary education attendance varies remarkably among different population groups. Many institutions in Brazil implemented affirmative action policies in the last decades and the quotas policy was implemented in federal public institutions of higher education in 2012 in order to decrease attendance inequalities (^{Aranha, 2009}; ^{Telles & Paixão, 2013}). Besides, other policies may have affected the proportion of minorities in higher education institutions in Brazil, such as Reuni and the introduction of Enem/SiSU in the students’ selection process (^{Lima & Machado, 2016}; ^{Nogueira et al., 2017}; ^{Silveira et al., 2015}).

This paper verified the evolution of the diversity of the student body at UFMG between 2006 and 2015, addressing the effects of these main changes in this institution’s selection process. The proportion of students from low-income households increased remarkably between 2008 and 2013. The Reuni, bonus and quota policies seemed to be effective in promoting this variation, however, changing from bonus to quota may not have being decisive for this result. The proportions of Whites and of students from private secondary schools showed similar trends with a remarkable decrease in 2009. Bonus and quota polices seemed effective to explain these results. However, contrary to these trends, the implementation of Enem/SiSU in 2014 apparently promoted an increased in Whites attendance. The drop in the proportion of students from private secondary schools in 2009 was accompanied by an increase in the proportion of students from state secondary schools, but the dynamics for students from federal secondary school differed. For the proportion of students from federal schools, there was a remarkable increase in 2014, the year of the implementation of the Enem/SiSU, a procedure for students´ selection that may have increased inequalities in attendance at UFMG.

Moreover, the determinants of the propensity of belonging to low-income households, being Black/Pardo, or from private or state secondary schools were assessed. Not surprisingly, SES levels were among these determinants. However, some other conclusions could be drawn. Being a female increased the propensity of being from a low-income household and of studying in private secondary schools. These features points for specific policies focusing young male students from low-income households while or after secondary school. Non-low-income students from state secondary schools had a greater propensity of pursuing further preparation to do the selection process of UFMG, while those who studied in private schools or low-income students had a lower propensity. These results suggest that specific policies of expansion of free or low-cost pré-vestibular courses should be promoted. The results for foreign language knowledge showed a large gap between those who attended private secondary schools in comparison to other groups of students, gap that should be shrank in higher education institutions. Thus, the empirical findings of this paper pointed to specificities that could be employed while designing policies facing some of the problems of higher education institutions attendance by minorities in Brazil.