Brazil

The two percentage-point gaps for Brazil are shown next to the unweighted mean across settings. Both are defined as female minus male. A negative choice gap means that among students who have already cleared the same academic bar, women are less likely than men to rank a STEM program first.

The table summarizes features of the admission system that bear on the construction and interpretation of the analysis. Untranscribed values are marked as pending.

All public tertiary institutions in Brazil can opt to participate in SISU, a centralized digital platform that matches students to degree programs. Centralized matching for public universities was introduced in 2010, and by 2016, the year of our data, 57% of the 4.8 million university entrance exam takers applied to a degree program through the platform. Students are assigned to degree programs by an iterative deferred acceptance algorithm that takes into account performance in the national university entrance exam (ENEM), student preferences, and affirmative action status.

The iteration is what distinguishes Brazil from a standard one-shot deferred acceptance. The application window runs for a few days in January, during which provisional admission cutoffs are updated daily and students may revise their choices. A student may list at most two programs, chosen from roughly 6,310 college-major combinations offered in a typical year, the tightest list cap among the ten systems the paper studies. To implement affirmative action, each degree program sets aside quotas reserved for students from public high schools, so targeted students face lower admission cutoffs than non-targeted students.

Because ENEM scores are released before the window opens, applicants rank programs already knowing exactly where they stand, and the daily cutoff updates let them condition their short list on how the competition is unfolding. The paper notes that strategy-proofness weakens when lists are restricted and that the evidence on how well iterative designs sustain truthful reporting is mixed; it also observes that across the ten settings the list caps do not order the choice gaps, with Brazil's two-cell list, the tightest cap observed, sitting in the middle of the range.

The ENEM is sat in the closing months of the Brazilian academic year, and individual scores are released before the SISU window opens, so applicants submit their ranked lists with their scores in hand. The window itself runs for a few days in January; during those days provisional cutoffs are updated daily and students may revise their two choices.

The assignment is published in the third week of January, and students then have until the end of the month to accept or reject the offer. Private universities, which do not participate in the platform, run decentralized admissions right after the assignment by SISU.

In the year we observe, 132 of Brazil's 2,448 tertiary institutions admitted students through the centralized platform. Participation is opt-in for public institutions, and the country's 2,152 private institutions do not participate; they use decentralized admissions that begin right after the SISU assignment. Among the participating universities there is no tuition fee, and financial aid is available.

One boundary of the platform matters especially for a study of STEM: the São Paulo state universities, which host some of the country's strongest STEM programs, run their own admissions. The Brazilian sample therefore covers the federal system rather than the full public sector. In most of the other settings the paper studies, the centralized platform is the route into the most selective universities and their STEM programs; Brazil is one of two settings, with Uganda, where the paper flags a departure from that pattern.

The performance measure is the student's ENEM score, the national university entrance exam on which SISU itself allocates seats, with program-specific weighting of the exam. The data record both overall and subject-level scores. The measure is constructed identically for women and men, and no admission adjustment, weighting, or bonus enters it; the affirmative-action quotas that lower cutoffs for public high school students operate on admission, not on the score used here to rank students.

Performance groups are formed within setting and cohort, over the full distribution of score holders, 2,711,049 students in the 2016 cycle. The groups fill cleanly at the intended thresholds, and the manuscript records no setting-specific caveat for Brazil's performance measure.

The sample covers the 2016 admission cycle, drawn from the SISU platform and ENEM records. The full distribution against which performance groups are defined contains 2,711,049 score holders. The top decile contains 271,049 students, of whom 133,899 are women and 137,150 are men, making women 49.4% of the high-achieving group, close to parity, so the pipeline gap is small in this setting. Women are 57.1% of all applicants, against 49.4% of the top decile. Brazil is also the largest setting in the study, with 2.71 million yearly applicants against 18.1 thousand in Uganda at the other extreme.

Because the São Paulo state universities run their own admissions and private institutions sit outside the platform, the analysis sample covers applications within the federal system rather than the Brazilian public sector as a whole. Within that system we observe all applications submitted to the centralized platform together with the applicants' test scores.

Additional record-level exclusions and field mappings are documented in the replication code.

No admission rule in Brazil's centralized system differs by sex: the paper's Table A.III comparison of the ten systems documents no sex-conditional rule for this setting, and the ENEM-based performance measure is constructed identically for women and men. The only documented admission adjustment is the affirmative-action quota system, under which each degree program reserves seats for students from public high schools; that adjustment conditions on school background, not sex, and it does not enter the performance measure used to form the groups analyzed in the paper.

Our data for Brazil are two datasets, drawn from the SISU platform and ENEM records for the 2016 cycle, in which we observe all applications submitted to the centralized admission system and students' test scores. The records include students' overall and subject-level ENEM scores, their ranked preferences for specific college-major combinations, demographic characteristics, namely gender and age, and self-reported household income.

Access conditions and the agency contact route have not yet been supplied for publication.