Does it Matter Which Top Institution You Choose? A Case Study of Brazilian Graduate Admissions

Full text

Does It Matter Which Top Institution You Choose?
A Case Study of Brazilian Graduate Admissions∗
Fernanda Estevan†Kelly Santos‡
January 2, 2023
Abstract
Does attending a particular graduate school matter for academic outcomes once we consider stu-
dents’ selection into graduate programs? This paper sheds light on this previously unexploited ques-
tion by investigating the impact of attending a selective master’s institution in Economics on Ph.D.
placement. Using data from the ANPEC exam, widely used for admission to Brazilian master’s pro-
grams in Economics, we can control for relevant variables, such as applicants’ undergraduate college
and ANPEC scores. To address the potential selection biases, we compare students who applied and
were accepted to the top four master’s programs, as in Dale and Krueger (2002). When we account
for students’ observable and unobservable characteristics, we show that seemingly large differences
between programs mostly vanish. Therefore, we cannot rule out that top master’s programs perform
similarly in Ph.D. placements abroad.
Keywords: selective universities, graduate school, Ph.D. enrollments, higher education, returns to
education.
JEL Classification Codes: I23, I26.
∗We gratefully acknowledge funding from the Sao Paulo Research Foundation (FAPESP - grant # 2015/21640-3 and 2019/25033-5) and
CNPq. Kelly received financial support from Coordenac¸ ˜
ao de Aperfeic¸oamento de Pessoal de N´
ıvel Superior - Brasil (CAPES - Finance Code
001). We are indebted to the ANPEC exam organizers, in particular, L´
ızia de Figueirˆ
edo, Marco Fl´
avio da Cunha Resende, and Rodrigo Zadra
Armond, for providing the data and outstanding assistance during the project. We also thank FGV-EESP, FGV-EPGE, IPE-USP, and PUC-Rio
master’s coordinators who provided data on Ph.D. placement for their respective programs. This research has benefited from discussions with
Braz Camargo, Bruno Ferman, Gustavo Gonzaga, Ricardo Madeira, Enlinson Mattos, Luis Meloni, Naercio Menezes Filho, Renata Narita,
Paula Pereda, Vladimir Ponczek, and Vitor Possebom. All remaining errors are ours.
†Sao Paulo School of Economics - FGV, Rua Itapeva, 474, Sao Paulo, SP, Brazil; phone: + 55 11 3799-3359; email:
fernanda.estevan@fgv.br.
‡Sao Paulo School of Economics - FGV, Rua Itapeva, 474, Sao Paulo, SP, Brazil; email: kelly.santos@fgv.br.

1 Introduction
There is fierce competition among top universities to attract high-achieving students in many educational
settings. As education markets become more nationally integrated, colleges seek to differentiate them-
selves from their competitors to influence students’ choices (Hoxby,1997). In doing so, institutions may
build a reputation based on their students’ future outcomes and attract individuals inclined to fulfill those
expectations. Therefore, an important question is whether the payoffs of a given educational choice are
due to the school or individuals’ preexisting observable and unobservable characteristics.
At the college level, there is quite some evidence that the positive impact of selective versus non-
selective schools on students’ labor market outcomes is severely reduced or even eliminated once we
account for students’ self-selection into institutions (e.g., Dale and Krueger,2002;Mountjoy and Hick-
man,2021). However, we know much less about the relative contribution of graduate schools, a level
at which most relevant players are selective universities, and their graduates’ academic rather than labor
market outcomes may be better metrics of a school’s performance.
To the best of our knowledge, we are the first paper to investigate the impact of attending a particular
top master’s program on students’ subsequent Ph.D. enrollment abroad. Master’s programs are the most
common pathway for Brazilian students to Ph.D. programs abroad, as in other Latin American and
developing countries. A priori, there is no reason to believe that the results obtained in the undergraduate
literature on labor market outcomes extend to the graduate-level setting. First, selective Ph.D. programs
arguably attach more weight to applicants’ academic credentials than potential employers. Second, the
number of available slots is much smaller than in the labor market, so the master’s program institution
may be an important tie-breaker in Ph.D. admissions.
We use data from the National Association of Postgraduate Programs in Economics (ANPEC) exam,
widely used for admission by Brazilian master’s programs. By linking the ANPEC dataset to Ph.D. en-
rollment data obtained from various sources, we show that 57.5% of students enrolled in a top four mas-
ter’s program1pursued a Ph.D. in Brazil or abroad after completing their master’s degree. Furthermore,
between 20.0% and 40.0% enroll in a Ph.D. program abroad. Moreover, these four schools account for
76.2% of Ph.D. placements abroad, confirming that they are the relevant setting for investigating this
research question.
Apart from their reputation, master’s programs may impact Ph.D. enrollment through several chan-
nels. For instance, their professors may incentivize their students to apply for Ph.D. programs abroad
1The top four master’s programs are FGV-EESP, FGV-EPGE, IPE-USP or PUC-Rio. We explain how we distinguish this
group in Section 2.
1

directly or indirectly through role models. Some programs may also have better technology for assisting
their graduates in obtaining admission to Ph.D. programs by helping them apply to programs that better
match their profiles and writing more convincing reference letters. While we cannot disentangle these
different mechanisms in our analysis, our results provide the combined (relative) impact of all those
potential influences on Ph.D. enrollment abroad for our top four programs.
We benefit from an unusually favorable setting to separate the impact of graduate schools from
students’ sorting into institutions. Postgraduate admission to Economics programs in Brazil mainly
relies on ANPEC scores rankings.2Thus, we observe in our data nearly the entire set of applicants’
information relevant for admission in contrast with most settings that adopt more decentralized and
holistic admission procedures.
Importantly, we observe applicants’ undergraduate degree institutions and a measure of pre-graduate
school performance in Economics, the ANPEC exam scores, allowing us to overcome one of the main
challenges in the measurement of value-added in the higher education literature (Cunha and Miller,
2014). As expected, both variables are crucial controls in our empirical specifications.
Moreover, applicants must list up to six (unordered) choices of master’s programs upon registering
for the ANPEC exam. We can also infer master’s program admissions based on the ANPEC rankings
and actual enrollment in these programs. Thus, we can limit our analysis to applicants who applied
and were likely admitted by the top four programs, allowing us to minimize concerns related to unob-
servables as suggested by Dale and Krueger (2002).3Finally, our top four programs clearly distinguish
themselves from the remaining master’s programs in attracting the best-ranked ANPEC applicants and
placing students in Ph.D. programs abroad. Still, we can exploit a significant degree of variation within
and among these top four programs regarding their incoming students’ profiles and Ph.D. placement
over the years.
Since we compare the top four master’s programs, our sample includes all ANPEC applicants en-
rolled in one of these programs. For this sample, we start by showing that FGV-EPGE and PUC-Rio
appear significantly more successful than FGV-EESP in terms of Ph.D. placement abroad, especially
in top-ranked programs. When we include controls for undergraduate college fixed effects, the effects
correspond to increases of 42.9% and 70.2% in the probability of attending a Ph.D. program abroad for
2Annually, ANPEC releases an ‘Applicants Handbook’ (see http://www.anpec.org.br/novosite/br/exame for the current
version). This handbook lists all master’s and Ph.D. programs adopting the ANPEC exam for admission, exam weights by
program, and any additional requirements. For example, some programs may require letters of recommendation and under-
graduate transcripts but still place a relatively larger weight on the ANPEC exam, especially for top-ranked applicants. We
provide more detail in Section 2.
3Unfortunately, a regression discontinuity design is unfeasible in our setting. Indeed, we cannot identify clear-cut cutoffs
among these four programs because their admission lists overlap considerably and also due to data limitations (see Section 3).
2

FGV-EPGE and PUC-Rio students relative to FGV-EESP’s, respectively. Furthermore, when we restrict
our outcome variable to the top eight Ph.D. programs abroad, that relative advantage corresponds to an
even larger increase of 119.0% and 159.5%, respectively.4
Importantly, these differences nearly vanish once we account for students’ observable and unob-
servable characteristics. More precisely, when we control for students’ ANPEC scores, the impacts
of FGV-EPGE and PUC-Rio relative to FGV-EESP on the likelihood of pursuing a Ph.D. abroad are
close to zero and become statistically insignificant. Moreover, all coefficients become statistically in-
significant when restricting our analysis to students who selected and received offers from the four top
institutions. For the top eight Ph.D. programs abroad, the coefficients of FGV-EPGE and PUC-Rio re-
main positive, and the coefficient for IPE-USP is negative. We perform a minimum detectable effects
analysis and show that we have enough power to rule out effects for PUC-Rio and IPE-USP on the top
eight Ph.D. programs. Still, we may imprecisely estimate the impact of FGV-EPGE on the top eight
Ph.D. placements due to our relatively small sample. Taken together, our results suggest that controlling
for scores and unobservables substantially reduces the causal effects of particular schools.
Our study relates to the literature that aims to identify the impact of selective colleges on students’
labor market outcomes.5The seminal contribution is Dale and Krueger (2002), which propose restricting
the analysis to students with the same application and admission profiles to deal with the selection of
students into colleges based on unobservables. Using data from the College and Beyond data set and the
National Longitudinal Survey of the High School Class of 1972, they show no differences between the
earnings of students who attended more or less selective colleges once they take selection into account,
except for minority students.
Similarly, Cunha and Miller (2014) show large earnings differences across colleges in Texas that
significantly reduce once they control for selection using a similar strategy. More recently, Mountjoy
and Hickman (2021) expanded Dale and Krueger (2002)’s methodology to allow for heterogeneous
treatment effects and confirm that the value-added of postsecondary institutions is small relative to the
effects of students’ sorting. Interestingly, they show a short-term selectivity premium that fades away a
few years into the job market. This last finding helps reconcile the results from Broecke (2012) that finds
4The top eight Ph.D. programs abroad are MIT, Harvard, Stanford, Chicago, Princeton, Yale, Berkeley, and Northwestern.
We explain in Section 2how we select this group of programs.
5A related literature investigates the impact of college attributes, the so-called school quality, on students’ future earnings
(e.g., Fox,1993;Loury and Garman,1995;Daniel et al.,1997;Brewer et al.,1999;Behrman et al.,1996;Altonji and Dunn,
1996;Black and Smith,2004). These papers suggest that college quality positively affects students’ labor market performance.
Still, such effects tend to be overestimated in studies not considering the self-selection of students into colleges. Grove and
Hussey (2014) reach a similar conclusion investigating returns to MBA programs using survey data of registrants for the GMAT
exam.
3

a positive wage premium to attending selective universities in the UK around four years after individuals
graduate, even accounting for selection on observables and unobservables.6
We also contribute to the tiny literature on the determinants of Ph.D. enrollment. A few studies link
enrollment in graduate programs with college quality and labor market conditions. Eide et al. (1998) find
that the quality of college significantly increases the probability of attending graduate school at a major
research institution. Bedard and Herman (2008) conclude that labor market conditions do not affect
graduate school enrollment while Johnson (2013) shows that enrollments respond to unemployment
only for women.
Our findings show that differences in Ph.D. placements abroad among top master’s programs are
mostly due to the (self-) selection of students in observable and unobservable characteristics. Once we
account for their ANPEC scores and compare students with the same choices/options, most master’s
programs are similar in their placement profiles. While our results do not speak directly to the determi-
nants of Ph.D. enrollment, they suggest that individuals’ characteristics play a major role in explaining
graduate school continuation. Our results also have policy implications, as programs’ evaluations typi-
cally rely on students’ outcomes. Finally, while our analysis does not allow us to distinguish whether top
institutions are relevant in terms of value-added or as signaling/network devices, our findings suggest
that the differences within these top are, at most, much smaller than anecdotal evidence would suggest.
We organize this paper as follows. Section 2presents the ANPEC exam characteristics and contex-
tualizes the top four master’s programs. Then, section 3describes the data, and Section 4reports the
empirical strategy. Finally, section 5presents the main results and some robustness checks, and Section
6concludes.
2 Background
In Brazil, undergraduate degrees in Economics last for four or five years, depending on the university
and stream (typically daytime and evening). While some Ph.D. programs in Brazil admit a few students
straight after their undergraduate studies, most students start their postgraduate studies with a master’s
degree. Moreover, Brazilian master’s programs, especially at top institutions, are sought-after by stu-
dents wishing to pursue Ph.D. programs abroad.
The ANPEC exam is the centralized postgraduate admission exam in Economics organized by the
6A recent literature (e.g., Hoekstra,2009;Hastings et al.,2013;Zimmerman,2014) finds large impacts of selective univer-
sities for academically marginal students, employing regression discontinuity designs. Since we focus on top-ranked students,
those findings do not speak directly to our results.
4

Brazilian National Association of Postgraduate Programs in Economics (ANPEC, Associac¸ ˜
ao Nacional
dos Centros de P´
os-Graduac¸ ˜
ao em Economia in Portuguese). Brazilian master’s and (some) Ph.D.
programs adopt the ANPEC exam for admission. Some schools also require letters of recommendation
and undergraduate transcripts, mainly used for applicants outside the top of the ANPEC ranking.7
The ANPEC exam takes place annually and does not require any specific academic degree, although
only college graduates can enroll in master or Ph.D. programs in Brazil. Applicants register for the
ANPEC exam by simply filling out a form with personal data, paying a fee, and applying to (up to six)
master’s program choices.8
The ANPEC exam evaluates students in six subjects: Microeconomics, Macroeconomics, Mathe-
matics, Statistics, Brazilian Economy, and English. The exam is the same for all applicants, consisting
of true or false and open-ended questions. ANPEC calculates each subject’s final score considering that
an item answered incorrectly cancels the score obtained in a correct item in the true or false questions
and then standardizes each subject-specific score at the year level. Based on the exam scores, ANPEC
provides two general rankings, with and without the Brazilian Economy score, and then specific rank-
ings based on each postgraduate institution’s weighting criteria. The ANPEC general rankings equally
weight all tests except English (and Brazilian Economy, if applicable). Per its pre-established rules, each
program uses the general or specific ranking to fill vacancies. In most years of our sample, top programs
use the ANPEC ranking without the Brazilian Economy score, i.e., the ranking that uses the arithmetic
average of the Microeconomics, Macroeconomics, Mathematics, and Statistics tests.9Therefore, we
use the ANPEC ranking without the Brazilian Economy score throughout our paper (henceforth, AN-
PEC ranking). Accordingly, our ANPEC score equally weights the Microeconomics, Macroeconomics,
Mathematics, and Statistics standardized scores.
Each institution sends admission offers to the best-ranked applicants and may invite them to campus
visits. Importantly, institutions compete for better-ranked students and may send offers even to students
who did not list that institution in their preference lists. Therefore, depending on the applicant’s ranking
position, she may have a set of institutions to select. In summary, applicants make two significant deci-
sions. First, the applicant chooses to apply to a set of institutions. Second, after the ANPEC exam result,
7According to program coordinators, most applicants within the first 25th ANPEC positions are solely admitted based on
the ANPEC ranking.
8More precisely, the applicant lists her (up to six) unordered preferred master’s programs on the ANPEC subscription form.
We expect these choices to be informative, as applicants can only choose six programs. In addition, master’s programs rarely
make offers to applicants who did not list them initially. Indeed, about 90.0% of individuals in our final sample attended a
program they listed in the ANPEC subscription. If we consider only the top four institutions’ enrollees, that number increases
to 98.1% of the applicants in the sample.
9The ANPEC Applicants Handbook provides information on the weights used by each master’s program.
5

the student decides which institution she will attend among the institutions that accepted them, i.e.,
among the admission offers she received.10 Higher reputation programs typically attract the best-ranked
applicants. However, students also sort based on location and different profiles of master’s programs.
The academic year starts in February/March for students who enroll in master’s programs. Since
master’s programs last for two years, students typically apply for Ph.D. programs abroad by the end of
their second year (in October/November) to start in September/October of the following year. However,
some students may postpone their application to the following year. Thus, our last cohort started the
master’s program in 2017 and typically applied for the Ph.D. in 2018 and 2019, beginning in the fol-
lowing year. We focus on Ph.D. programs abroad since admission to Ph.D. programs in Brazil is less
competitive, especially for students with master’s degrees from top institutions. In our sample, 76.5% of
students from the top four institutions who pursued a Ph.D. in Brazil remained in the same university.
Apart from showing the effects on Ph.D. programs abroad, we also investigate admission to particu-
larly reputable programs. Two rankings are available for Ph.D. programs abroad, as shown in Table A.1.
Amir and Knauff (2008) ranks programs based on their job market placements, considering the faculty
employed in 58 universities in 2006. We use their R3 score, which restricts hires to 1990-2006, to obtain
a recent picture of Ph.D. programs’ performance. US News (2022) also provides a ranking specific to
Ph.D. programs in Economics, including only US universities. Notably, both rankings place the same
Ph.D. programs in the first eight positions. We denote this group, including MIT, Harvard, Stanford,
Chicago, Princeton, Yale, Berkeley, and Northwestern, as the ‘top eight’ group.11
Top Four Master’s Programs
We now contextualize the top four master’s programs, describing their main features and why they are
relevant when investigating outcomes related to Ph.D. placement abroad. Table A.2 displays summary
statistics for 29 master’s programs in Brazil, including ANPEC ranking information of enrollees and
Ph.D. placements by master’s program.12
Four institutions stand out because they enroll students near the top of the ANPEC ranking and
place many in Ph.D. programs abroad.13 These institutions are PUC-Rio, FGV-EPGE, IPE-USP, and
10Institutions do not necessarily fill all vacancies simultaneously. Some programs do a second round to fill the remaining
vacancies after the first round.
11QS (2022), Times Higher Education (2022), and Shanghai Ranking (2021) provide global rankings focused on Economics
but not exclusively based on graduate programs, being therefore unfit for our purposes.
12We explain the data construction in Section 3. The sample presented in Table A.2 only includes students admitted through
the ANPEC exam. We exclude the master’s programs with less than five students admitted through ANPEC annually.
13These are also the only four institutions that received the maximum score of 7 in the Brazilian graduate programs’ evalu-
ation conducted by CAPES.
6

FGV-EESP, henceforth denoted ‘top four’ master’s programs.14
Table A.2 shows that the top four programs enroll students with a median ANPEC ranking ranging
from 19 at PUC-Rio to 69 at FGV-EESP. Undoubtedly, PUC-Rio is the most selective program in terms
of admission. Forty-two and seventy-nine percent of their enrollees were in the first 15 and 30 positions
of the ANPEC ranking, respectively. The four programs place more than half of their students in Ph.D.
programs in Brazil and abroad. When considering Ph.D. programs abroad, we see a more significant
proportion of students from PUC-Rio (40%) and FGV-EPGE (34%) who enroll in a Ph.D. abroad, as
compared to FGV-EESP and IPE-USP (20%).
The differences between PUC-Rio and FGV-EPGE institutions’ Ph.D. placements with FGV-EESP
and IPE-USP strengthen when considering higher-reputation Ph.D. programs. PUC-Rio stands out with
16% of master’s graduates in the top eight Ph.D. programs in any field. If we only consider the Ph.D.
programs in Economics, PUC-Rio and FGV-EPGE still have a large advantage: 14% and 12% enroll
in the top eight programs, respectively. At FGV-EESP and IPE-USP, only 5% and 1% of graduates
enroll in the top eight Ph.D. programs in Economics, respectively. Since PUC-Rio and FGV-EPGE
admit students with the best ANPEC ratings, we cannot attribute these differences to a program effect.
Our analysis tries precisely to deal with the selection problem in the admission process to measure the
master’s program’s impact on Ph.D. placements.
Not surprisingly, the best-ranked applicants in the ANPEC exam massively apply for these top four
programs. Table A.3 shows the percentage of individuals who applied and were admitted to the top four
programs and their subsequent Ph.D. placement based on their ANPEC ranking. These figures confirm
that most top-ranked students apply and enroll in the top four programs. Moreover, none of the other
relatively selective programs (UnB, UFRJ, UFMG, USP/RP, and UFMG) enroll a significant proportion
of top-ranked applicants. In addition, since we run our analysis on a matched sample of students who
applied and were admitted by the selective programs, including a fifth program in this context would
reduce our sample size significantly. Thus, this evidence confirms that the top four master’s programs
are the relevant set when investigating outcomes related to Ph.D. placement abroad.
14Our conclusions remain unchanged if we exclude IPE-USP from the analysis and consider only the ‘top three’ master’s
programs.
7

3 Data
Data and Sample Restrictions
Our main database comes from ANPEC. We use data on applicants’ performance in the ANPEC annual
exam from 2004 until 2017. When the applicant took the exam more than once, we consider only the
most recent application (16.0% of applicants wrote multiple exams). In this period, 11,305 applicants
took the ANPEC exam.
The dataset includes applicants’ positions according to ANPEC, each program’s ranking criteria,
and their standardized exam scores by subject. In addition, the data also hold individual background
information like gender, race, age, marital status, nationality, undergraduate degree and school, and self-
reported information on how many times the individual took the ANPEC exam before and in which year
the applicant finished her undergraduate studies. Importantly, the data include each applicant’s (up to)
six unordered choices of graduate programs.
From 2009 on, the ANPEC datasets show the MA program the applicant eventually enrolled in, but
that information was unavailable between 2004 and 2008. To obtain the 2004-2008 master’s enrollment
data, we link the ANPEC exam records to two datasets from Coordenac¸ ˜
ao de Aperfeic¸oamento de Pes-
soal de N´
ıvel Superior (CAPES).15 First, we use publicly available data on all enrollments in Brazilian
graduate programs for 2004-2019. Then, we merge the CAPES and ANPEC datasets using graduate
students’ full names and ANPEC application years, allowing us to identify the applicant’s enrollment
institution for 49.7% of the 2004-2008 sample.
For the unmatched ANPEC applicants, we also use the 2001-2020 Brazilian public catalog of the
postgraduate thesis, the Sucupira dataset, to identify where the student completed the master’s program
using the applicant’s full name.16 By including the Sucupira’s database, we additionally obtain the
master’s degree institution for 1.3% of the 2004-2008 ANPEC cohorts. As a result, we drop ANPEC
applicants without a master’s enrollment (45.1% of applicants) and those enrolled in a Ph.D. without
completing the master’s program (3.8% of applicants).
Since our analysis focuses on the most selective master’s programs, we keep in our main sample
only applicants enrolled in one of the top four master’s programs, corresponding to 903 individuals.
We then collect additional data on their subsequent Ph.D. placement for these applicants. We merge
15The Coordination for the Improvement of Higher Education Personnel, in English, is the Brazilian federal government
agency under the Ministry of Education that oversees all postgraduate institutions and centralizes information on all MA and
Ph.D. programs in Brazil.
16CAPES maintains the Sucupira’s thesis catalog, which can be accessed in CAPES (2019). All individuals who obtained
a master’s or Ph.D. in Brazil between 2001 and 2020 are in the Sucupira database. The dataset includes the thesis title, the
student’s full name, the year of the thesis defense, and the master’s degree program.
8

the CAPES and Sucupira’s datasets to ANPEC datasets using graduate students’ full names to identify
those who did a Ph.D. in Brazil in any field of study after the ANPEC application year, i.e., 2004-2017.
The CAPES dataset contains the universe of individuals enrolled in a Ph.D. program in Brazil. We find
information for 29.2% of our sample.
We search for a Ph.D. placement abroad after master’s degree enrollment for the remaining appli-
cants. For applicants who attended the master’s program at FGV-EESP and PUC-Rio, we use the public
lists containing all Ph.D. placements available on these institutions’ websites.17 For applicants who
attended the master’s program at FGV-EPGE and IPE-USP, we use placement lists provided by these
programs’ coordinators. In our sample, 28.2% of individuals did a Ph.D. abroad.18
Master’s Admission Offers
Our data allow us to observe the master’s program the applicant eventually attended but not the offers
she received, i.e., the programs where she could have enrolled. Identifying the applicants’ admission
offers is essential to implementing the method proposed by Dale and Krueger (2002). We implement a
simple method to infer applicants’ admission offers. In a nutshell, if the student was the lowest-ranked
enrolled in a program in the respective ranking, we consider that all better-ranked applicants received
offers from that same institution.
Suppose that the last student enrolled in program A was the student ranked 56. We assume that all
individuals ranked between 1 and 55 received an admission offer from institution A. We also assume
that institution A did not admit any students below 56. We discard outlier applicants who attended a
master’s program but were more than 15 places away from the nearest above-ranked admittee.19
Our method fails if the lowest-ranked student enrolled in a given program is not necessarily the last
to be accepted by that institution. Indeed, if an applicant ranked below the last enrolled did not attend a
program, we do not know if she was not admitted or opted for another program.
17These placement lists are available at https://economics-sp.fgv.br/graduate-program/placement/master-students and
http://www.econ.PUC-Rio.br/uploads/alunos doutorado exterior.pdf.
18Additionally, we collect Ph.D. placement data for the remaining applicants, i.e., those who did not enroll in a top four pro-
gram. These applicants are not in our main sample, but this extended sample allows us to characterize the top four programs
relative to the other master’s programs in Table A.2. As before, we use the CAPES and Sucupira datasets to identify Ph.D.
programs in Brazil. For Ph.D. programs abroad, we consult applicants’ Lattes Curriculum (the national register of the academic
activity of students and researchers in Brazil maintained by CNPq, Conselho Nacional de Desenvolvimento Cient´
ıfico e Tec-
nol´
ogico), LinkedIn, and personal websites. For feasibility reasons, we limit our search for Ph.D. abroad to the 250 top-ranked
applicants in the ANPEC exam in this extended sample. However, this limitation should not impact our characterization of
non-top four programs, as only 1% of applicants between 201 and 250 attend Ph.D. programs abroad, and none attended a top
eight Ph.D. programs abroad (see Table A.3).
19These outliers exist because some programs select applicants based on letters of recommendation, transcripts, or research
experience in addition to the ANPEC score. In that case, the admitted student would be in a ranking position far from most
students approved through the ANPEC exam. Outliers correspond to 0.1% applicants in our sample.
9

We check the effectiveness of our admission inference method by comparing our inference with the
actual offers for some programs. Following our request, FGV-EESP and IPE-USP provided us with
their offers list for some years. We verify that, for FGV-EESP, the actual ranking position of the last
admitted student was close to our inferred classification. For 2015, 2016, and 2017, we failed to detect
seven, three, and zero students who received admission offers, respectively. The inference for the last
students accepted in IPE-USP was also close enough: we considered two students in 2013, five in 2014,
and eight in 2015 not accepted by IPE-USP based on our method when, in fact, they received admission
offers. Thus, while we fail to account for some students below the last enrolled in a program, our
admission inference method reasonably estimates the actual admission offers. Importantly, since our
empirical strategy focuses on applicants who applied and were admitted by all the top four programs,
such imprecisions probably do not impact our results.20
Another concern is that institutions are not bound to follow the ranking strictly. Instead, they may
select some relatively worse-ranked individuals or not admit some better-ranked applicants. By exclud-
ing the outliers, we can deal with the former. However, the second issue is more complicated and may
introduce some biases in our estimation. Since cherry-picking is unlikely for top-ranked applicants, we
run our regressions considering only the individuals in the 25 first position of the ANPEC ranking. We
reach similar conclusions with this reduced sample.
Descriptive Statistics
Table 1presents detailed descriptive statistics for our main sample, consisting of students admitted to at
least one of the top four programs. Most applicants are men, white, and single. At the time of the exam,
their mean age is 24 years. Most applicants take the ANPEC exam one year after their undergraduate
studies.
Comparing the sample of students admitted and enrolled in one of the top four programs, we see
that the enrolled students have slightly higher ANPEC scores. Moreover, the ANPEC scores are even
higher in the matched sample. While admittees’ demographic characteristics are similar across the four
programs, there are some differences in the programs’ catchment areas. All top four institutions admit
many students with an undergraduate degree from the same city, ranging from 45% at PUC-Rio to 68% at
FGV-EESP. However, there are sharp differences in the proportion of these students who graduated from
the same institution, possibly reflecting variations in undergraduate program size. For example, while
20Moreover, they would not affect our analysis of the applicants ranked in the 25 first positions, as all top programs typi-
cally admit those students. However, identifying the marginal admittee would be crucial in an RDD strategy and, therefore,
unfeasible in our study.
10

only 13% of students admitted to FGV-EESP were graduates from the same institution, this number rises
to 49% in the case of IPE-USP.
Table 1: Descriptive Statistics (2004-2017) - Top Four Master’s Programs
Top four
FGV-EESP FGV-EPGE PUC-Rio IPE-USP Top four (matched sample)
Admitted Enrolled
Male 0.80 0.81 0.80 0.83 0.82 0.78 0.85
Single 0.96 0.97 0.96 0.98 1.00 0.95 0.99
Undergraduate degree in the same institution 0.34 0.30 0.13 0.23 0.27 0.49 0.28
Normalized ANPEC Score 1876.33 2086.90 1702.19 2244.53 2452.36 1948.98 2467.38
(604.05) (524.95) (606.32) (412.22) (378.61) (407.67) (360.98)
Microeconomics 1.93 2.17 1.74 2.39 2.60 1.94 2.63
(0.88) (0.81) (0.91) (0.70) (0.69) (0.70) (0.66)
Macroeconomics 1.79 1.95 1.63 2.04 2.28 1.86 2.28
(0.69) (0.65) (0.70) (0.63) (0.53) (0.57) (0.55)
Statistics 1.92 2.13 1.73 2.29 2.52 1.98 2.52
(0.77) (0.71) (0.81) (0.59) (0.59) (0.62) (0.56)
Mathematics 1.86 2.10 1.70 2.25 2.42 2.01 2.45
(0.77) (0.67) (0.65) (0.61) (0.59) (0.64) (0.55)
Observations 1,263 903 187 242 197 277 308
Age 23.56 23.16 23.70 23.11 22.71 23.17 22.88
(3.24) (2.44) (3.00) (2.65) (1.64) (2.23) (1.96)
Observations 1,053 774 157 211 166 240 285
White 0.85 0.86 0.86 0.87 0.90 0.82 0.88
Observations 1,110 788 162 206 173 247 278
Undergraduate degree in the same city 0.56 0.57 0.68 0.54 0.45 0.62 0.50
Observations 1,247 897 187 239 195 276 306
Took the ANPEC exam within a year after bachelor’s degree 0.63 0.64 0.58 0.69 0.65 0.64 0.66
Observations 1,259 902 187 242 197 276 308
Notes: The table reports the descriptive statistics of our main sample, which includes applicants enrolled in a top-four master’s program between 2004 and 2017. We
show the average of variables for all students admitted or enrolled in the top four. We also show the results for the matched sample, which contains students who applied
and were likely admitted by the top four master’s programs (Dale and Krueger,2002). The ANPEC score equally weights the Microeconomics, Macroeconomics, Mathe-
matics, and Statistics standardized scores. Undergraduate degrees in the same city and institution consider the change between college and master’s programs. Standard
deviations are shown in parentheses.
4 Empirical strategy
There are at least two potential selection issues when comparing the Ph.D. placements of different mas-
ter’s programs. The first is the institution selection bias that occurs since programs may select students
based on characteristics that are unobserved by the researcher. Then, the most skilled students are likely
to be admitted to more competitive and selective institutions. The second problem is the student se-
lection bias that occurs when the students choose institutions they want to attend. Students who attend
more selective institutions may have different unobserved abilities than those who attend relatively less
selective institutions. In both cases, the unobservable abilities may correlate with the analyzed poten-
tial outcome, which is Ph.D. enrollment abroad. Thus, comparing the placement of students who have
attended different institutions can lead to erroneous conclusions.
To address these potential selection problems, we adopt the empirical strategy proposed by Dale and
Krueger (2002), which matches students who applied to and were admitted by the same institutions.
That match deals with the institution selection problem and can at least partially address the student
11

application bias problem.21 Indeed, when applying for master’s programs, students reveal a preference
for more or less selective institutions. Moreover, this preference possibly correlates with the student’s
unobservable characteristics related to the potential outcome.
Denote the outcome by Yic, a dummy variable equal to one if the applicant iattended a Ph.D. abroad
after the master’s degree.22 To allow for distinct levels of Ph.D. reputation, we consider a Ph.D. abroad
and at a top-eight Ph.D. program in any field. We regress the outcome on binary variables of assignment,
Dj
i, which is one if student iattended the master’s program j, and zero otherwise. We control for the
applicant’s ANPEC score (i.e., the score that weights the subjects of Microeconomics, Macroeconomics,
Mathematics, and Statistics equally) using a cubic polynomial, f(Si), to allow for a non-linear impact
of ANPEC performance on placement. We also include year and undergraduate college fixed effects (δt
and ηc).23 Our main regression equation is:
Yic =
3
∑
j
βjDj
i+γf(Si) + δt+ηc+ei,(1)
where eiis the error term. In all estimates, we cluster standard errors by ANPEC year. Since we have
only 14 clusters, we also compute wild bootstrap p-values using Roodman (2015). The parameters of
interest are β1,β2, and β3. Each of these βjparameters represents the effect of attending the master’s
program jrelative to a baseline program. In this regression, we can test if the programs’ effect equals
zero among the four programs through an F-test. We do this exercise for the sample that includes all
individuals enrolled in a top-four master’s program and the matched sample of students who applied for
and were accepted by all the top-four master’s programs.
5 Main Results
Table 2presents the results for attending a Ph.D. abroad in any field of study. We first examine the sample
that includes all students enrolled in one of the top four programs. In column (1), we do not include any
control variables. The results indicate a sizable advantage for FGV-EPGE and PUC-Rio compared to
FGV-EESP, the baseline program, in placing their students in Ph.D. programs abroad. In column (2), we
add fixed effects for applicants’ undergraduate degrees, which slightly reduces the magnitude of these
21In the unlikely scenario in which enrollment choices are random among the feasible master’s programs, this approach
would completely address the student applicant bias.
22We consider Ph.D. attendance rather than conclusion since we want to measure the master’s program’s capacity to place
its students into a Ph.D. program abroad.
23We do not include controls for individual characteristics since there is low variation and a substantial amount of missing
values that would significantly reduce our sample size. Our results remain unchanged if we control for gender.
12

coefficients. Still, these results indicate large effects associated with attending FGV-EPGE and PUC-Rio
relative to FGV-EESP, corresponding to 43.0% and 70.2%, respectively. Interestingly, once we control
for applicants’ ANPEC scores, the coefficients for the two programs become very close to zero and
statistically insignificant. The coefficient for IPE-USP in column (3) is negative but close to zero and
statistically insignificant, with a wild bootstrap p-value of 0.620.
While the results in columns (1)-(3) include important controls for observables, we may worry that
applicants selecting the different top programs may differ in some relevant unobserved way. There-
fore, we restrict our analysis to the sample of students applying and receiving offers from the top four
programs in columns (4)-(6). As explained in Section 4, this matched sample allows us to at least par-
tially address selection on unobservables by making our sample more comparable on potentially relevant
characteristics.
Once we account for such unobservables, the differences between the top four programs nearly
vanish. FGV-EPGE, PUC-Rio, and IPE-USP coefficients become negative, very close to zero, and
statistically insignificant. The IPE-USP coefficient is negative and significantly reduced once we control
for undergraduate college fixed effects and ANPEC scores, with a wild bootstrap p-value of 0.621. The
reduction in the coefficient estimate in column (6) relative to column (4) suggests that part of the IPE-
USP disadvantage stands because it attracts students with relatively worse ANPEC performance.
Since institutions are similar in Ph.D. enrollment abroad, we now check whether some institutions
stand out when considering programs with higher reputations. We show the results in Table 3restricting
our outcome variable to attendance to a top-eight Ph.D. program in any field abroad.24 Focusing on
columns (1) and (2) that compare all students, we see that FGV-EPGE and PUC-Rio enroll their students
with a higher likelihood in the top eight Ph.D. programs. Controlling for the applicants’ undergraduate
college fixed effects, FGV-EPGE and PUC-Rio estimators remain significant and large, corresponding
to increases of 119.0% and 159.5% in the probability of attending a top eight Ph.D. abroad relative to
FGV-EESP, respectively.
However, if we control for the ANPEC scores or compare only students who applied and were ad-
mitted to the top four programs, FGV-EPGE and PUC-Rio coefficients become statistically insignificant.
In contrast, the coefficient for IPE-USP is negative but marginally significant at the 10% level in column
(4). All coefficients become statistically insignificant when adding undergraduate college fixed effects
and ANPEC scores as controls. However, the coefficients for FGV-EPGE, PUC-Rio, and IPE-USP are
not negligible, corresponding to increases of 34.6% and 43.6% and a decrease of 48% in the probability
24We performed the same regressions considering only Ph.D. in Economics, and the results are similar.
13

Table 2: Probability of attending a Ph.D. abroad in any field, top four institutions
All Students Same Option and Admission Students
(1) (2) (3) (4) (5) (6)
FGV-EPGE 0.138** 0.121* 0.008 -0.003 0.009 -0.037
(0.051) (0.062) (0.052) (0.081) (0.120) (0.094)
[0.028] [0.072] [0.868] [0.975] [0.940] [0.666]
PUC-Rio 0.201*** 0.198*** 0.020 -0.017 -0.008 -0.010
(0.053) (0.055) (0.039) (0.045) (0.087) (0.068)
[0.001] [0.002] [0.635] [0.697] [0.920] [0.893]
IPE-USP -0.005 0.000 -0.019 -0.128 -0.141 -0.052
(0.048) (0.046) (0.035) (0.081) (0.112) (0.092)
[0.930] [0.997] [0.620] [0.170] [0.241] [0.621]
F-test 0.000 0.003 0.872 0.262 0.079 0.906
Dependent variable (mean) 0.282 0.282 0.282 0.373 0.373 0.373
Year fixed effects Yes Yes Yes Yes Yes Yes
Undergraduate college fixed effects No Yes Yes No Yes Yes
ANPEC score (cubic polynomial) No No Yes No No Yes
Number of observations 903 903 903 308 308 308
Notes. The dependent variable is a binary variable equal to one if the applicant attended a Ph.D. program abroad in any field and zero
otherwise. FGV-EESP is the baseline institution. FGV-EPGE, PUC-Rio, and IPE-USP are dummy variables equal to one if the applicant
enrolled in the master’s program at FGV-EPGE, PUC-Rio, and IPE-USP, respectively, and zero otherwise. We control for ANPEC year
and undergraduate college fixed effects. The ANPEC score equally weights the Microeconomics, Macroeconomics, Mathematics, and
Statistics standardized scores. We control for ANPEC scores using a cubic polynomial of scores. The sample ‘All Students’ contains all
students enrolled in a top-four master’s program. ‘Same Application and Admission’ is the matched sample, which is the ‘All Students’
sample restricted to students who applied and were likely admitted by the top four programs (Dale and Krueger,2002). Cluster-robust
standard errors are shown in parentheses (at the ANPEC year level). *Significant at 10%; **significant at 5%; ***significant at 1%. Due
to the small number of clusters (i.e., fourteen), we also report wild bootstrap p-values calculated using Roodman (2015) in square brackets.
of attending a top eight Ph.D. program relative to FGV-EESP, respectively. To ensure that these statisti-
cally insignificant estimates are not due to a lack of power, we calculate the minimum detectable effects
for our matched sample with the full set of controls (i.e., the sample used in column (6) in Table 3) in
Table A.4. Considering a 5% significance level and a statistical power of 80%, we should be able to
detect both coefficient estimates for PUC-Rio and IPE-USP. If anything, we fall short of FGV-EPGE’s
coefficient estimate by a slight difference.
Moreover, our admission inference method presented in Section 3can be subject to measurement
error, as the top four institutions may not send admission offers to some ANPEC applicants that are
better ranked than the last applicant admitted. To deal with that possibility, we reestimate the equation
(1), restricting the sample to ANPEC applicants in the top 25 positions. Students ranked in the first 25
positions are typically admitted to all the top four programs they apply to. For these applicants, institu-
tions are less likely to use other admission devices, such as letters of recommendation and transcripts,
relying mostly on ANPEC rankings. For this sample, there is no advantage of FGV-EPGE and PUC-Rio
relative to FGV-EESP, even not accounting for observable and unobservable applicants’ characteristics
14

Table 3: Probability of attending a top eight Ph.D. abroad in any field, top four institutions
All Students Same Option and Admission Students
(1) (2) (3) (4) (5) (6)
FGV-EPGE 0.069** 0.100*** 0.056 -0.002 0.077 0.054
(0.028) (0.033) (0.038) (0.097) (0.099) (0.082)
[0.029] [0.008] [0.177] [0.989] [0.470] [0.521]
PUC-Rio 0.104*** 0.134*** 0.045 -0.001 0.068 0.068
(0.023) (0.029) (0.037) (0.069) (0.066) (0.086)
[0.002] [0.001] [0.273] [0.993] [0.282] [0.422]
IPE-USP -0.040* -0.030 -0.022 -0.149* -0.138 -0.075
(0.021) (0.021) (0.027) (0.081) (0.084) (0.085)
[0.075] [0.166] [0.428] [0.074] [0.072] [0.365]
F-test 0.000 0.000 0.038 0.001 0.006 0.060
Dependent variable (mean) 0.084 0.084 0.084 0.156 0.156 0.156
Year fixed effects Yes Yes Yes Yes Yes Yes
Undergraduate college fixed effects No Yes Yes No Yes Yes
ANPEC score (cubic polynomial) No No Yes No No Yes
Number of observations 903 903 903 308 308 308
Notes.The dependent variable is a binary variable equal to one if the applicant attended a top eight Ph.D. program abroad and zero oth-
erwise. The top eight Ph.D. programs are MIT, Harvard, Stanford, Chicago, Princeton, Yale, Berkeley, and Northwestern; see Table A.1.
FGV-EESP is the baseline institution. FGV-EPGE, PUC-Rio, and IPE-USP are dummy variables equal to one if the applicant enrolled
in the master’s program at FGV-EPGE, PUC-Rio, and IPE-USP, respectively, and zero otherwise. We control for ANPEC year and un-
dergraduate college fixed effects. The ANPEC score equally weights the Microeconomics, Macroeconomics, Mathematics, and Statistics
standardized scores. We control for ANPEC scores using a cubic polynomial of scores. The sample ‘All Students’ contains all students
enrolled in a top-four master’s program. ‘Same Application and Admission’ is the matched sample, which is the ‘All Students’ sample
restricted to students who applied and were likely admitted by the top four programs (Dale and Krueger,2002). Cluster-robust standard
errors are shown in parentheses (at the ANPEC year level). *Significant at 10%; **significant at 5%; ***significant at 1%. Due to the
small number of clusters (i.e., fourteen), we also report wild bootstrap p-values calculated using Roodman (2015) in square brackets.
in the top-eight Ph.D. placements abroad. Considering all students and controlling for undergraduate
college fixed effects and ANPEC scores, the estimated FGV-EPGE and PUC-Rio coefficients are -0.057
and -0.005, respectively. However, we cannot exclude a sizable disadvantage of IPE-USP relative to
FGV-EESP in placements abroad, though imprecisely estimated for top-eight Ph.D. programs, since the
estimated coefficient is -0.158.
Overall, our results suggest that controlling for the selection on observables and unobservables sub-
stantially reduces the causal effects of particular schools. In particular, we cannot exclude that the top
four master’s programs are similar regarding Ph.D. placement abroad. The results remain insignificant
for placement in the top eight Ph.D. programs, but we cannot rule out that our estimates for FGV-EPGE
are imprecise. In addition, once we restrict the analysis to ANPEC applicants ranked in the 25 first po-
sitions, a sample possibly less subject to measurement error, the estimated coefficients for FGV-EPGE
and PUC-Rio are very close to zero for the top eight Ph.D. placements. The only exception is IPE-USP,
which has a relative disadvantage in Ph.D. placements abroad, especially in the top eight programs, in
this reduced sample.
15

6 Conclusion
We investigate the impact of Brazilian top master’s programs in placing their graduates in Ph.D. pro-
grams abroad. Historically, these programs have different Ph.D. placement rates. For example, FGV-
EPGE and PUC-Rio placed more than one-third of their graduates in Ph.D. programs abroad and around
15% in the top eight Ph.D. programs abroad. However, since these programs attract better-ranked stu-
dents, in terms of the centralized graduate program admission ANPEC exam, it is unclear whether their
success is due to the programs or observable and unobservable characteristics of their students.
To address selection by institutions and students, we use the methodology suggested by Dale and
Krueger (2002) that matches students who applied to and were accepted by the same top four institu-
tions. Our results show that the impact of institutions is severely reduced once we control for students’
performance in the ANPEC exam or compare only students with the same set of options and offers. In
those cases, most programs are similar, except for IPE-USP, which has a lower relative success in Ph.D.
programs abroad, though imprecisely estimated. For placement in the top eight Ph.D. programs, we
cannot exclude that the effects of FGV-EPGE are imprecisely estimated.
Our results add to a growing literature focused on the impact of selective institutions by investigat-
ing the role of graduate programs, previously unexploited in this literature. As in the undergraduate
literature, our results suggest that the impact of selective institutions is small or nil once we account for
students’ self-selection into universities. Moreover, our study contributes to the tiny literature on the
determinants of graduate school enrollment, suggesting that individual characteristics play a major role
in such outcomes.
16

References
Altonji, J. G. and Dunn, T. A. (1996). Using siblings to estimate the effect of school quality on wages.
The Review of Economics and Statistics, 78(4):665–671.
Amir, R. and Knauff, M. (2008). Ranking Economics Departments Worldwide on the Basis of PhD
Placement. The Review of Economics and Statistics, 90(1):185–190.
Bedard, K. and Herman, D. A. (2008). Who goes to graduate/professional school? The importance of
economic fluctuations, undergraduate field, and ability. Economics of Education Review, 27(2):197–
210.
Behrman, J. R., Rosenzweig, M. R., and Taubman, P. (1996). College choice and wages: Estimates
using data on female twins. The Review of Economics and Statistics, 78(4):672–685.
Black, D. A. and Smith, J. A. (2004). How robust is the evidence on the effects of college quality?
Evidence from matching. Journal of Econometrics, 121(1):99–124. Higher education (Annals issue).
Brewer, D. J., Eide, E. R., and Ehrenberg, R. G. (1999). Does it pay to attend an elite private college?
Cross-cohort evidence on the effects of college type on earnings. The Journal of Human Resources,
34(1):104–123.
Broecke, S. (2012). University selectivity and earnings: Evidence from UK data on applications and
admissions to university. Economics of Education Review, 31(3):96–107.
CAPES (2019). Catalog of thesis and dissertations (Cat ´
alogo de Teses e Dissertac¸ ˜
oes). Available at
https://dadosabertos.capes.gov.br/dataset?groups=banco. Baseline years: 1987-2016. Accessed on 1
August 2019.
Cunha, J. and Miller, T. (2014). Measuring value-added in higher education: Possibilities and limitations
in the use of administrative data. Economics of Education Review, 42(C):64–77.
Dale, S. B. and Krueger, A. B. (2002). Estimating the Payoff to Attending a More Selective College: An
Application of Selection on Observables and Unobservables. The Quarterly Journal of Economics,
117(4):1491–1527.
Daniel, K., Black, D., and Smith, J. (1997). College Quality and the Wages of Young Men. Research
Report 9707, Department of Economics, University of Western Ontario.
Eide, E., Brewer, D. J., and Ehrenberg, R. G. (1998). Does it pay to attend an elite private college?
Evidence on the effects of undergraduate college quality on graduate school attendance. Economics
of Education Review, 17(4):371–376.
Fox, M. (1993). Is it a good investment to attend an elite private college? Economics of Education
Review, 12(2):137–151.
Grove, W. A. and Hussey, A. (2014). Returns to MBA quality: Pecuniary and non-pecuniary returns to
peers, faculty, and institution quality. Labour Economics, 26:43–54.
Hastings, J. S., Neilson, C. A., and Zimmerman, S. D. (2013). Are Some Degrees Worth More than
Others? Evidence from college admission cutoffs in Chile. Working Paper 19241, National Bureau
of Economic Research.
Hoekstra, M. (2009). The Effect of Attending the Flagship State University on Earnings: A
Discontinuity-Based Approach. The Review of Economics and Statistics, 91(4):717–724.
Hoxby, C. M. (1997). How the Changing Market Structure of U.S. Higher Education Explains College
Tuition. Working Paper 6323, National Bureau of Economic Research.
17

Johnson, M. T. (2013). The impact of business cycle fluctuations on graduate school enrollment. Eco-
nomics of Education Review, 34:122–134.
Loury, L. D. and Garman, D. (1995). College selectivity and earnings. Journal of Labor Economics,
13(2):289–308.
Mountjoy, J. and Hickman, B. (2021). The returns to college(s): Estimating value-added and match
effects in higher education. Unpublished manuscript.
QS (2022). QS World University Rankings by Subject 2022: Economics & Econometrics. Available
at https://www.topuniversities.com/university-rankings/university-subject-rankings/2022/economics-
econometrics Accessed on 12 April 2022.
Roodman, D. (2015). BOOTTEST: Stata module to provide fast execution of the wild bootstrap with
null imposed. Statistical Software Components, Boston College Department of Economics.
Shanghai Ranking (2021). 2021 Global Ranking of Academic Subjects: Economics. Available at
https://www.shanghairanking.com/rankings/gras/2021/RS0501 Accessed on 12 April 2022.
Times Higher Education (2022). World University Rankings
2022 by subject: business and economics. Available at
https://www.timeshighereducation.com/world-university-rankings/2022/subject-ranking/business-
and-economics#/page/0/length/25/subjects/3087/sort by/rank/sort order/asc/cols/stats Accessed on
12 April 2022.
US News (2022). Best Economics Schools. Available at https://www.usnews.com/best-graduate-
schools/top-humanities-schools/economics-rankings. Accessed on 12 April 2022.
Vittinghoff, E., Glidden, D. V., Shiboski, S. C., and McCulloch, C. E. (2012). Regression methods
in biostatistics: linear, logistic, survival, and repeated measures models. Statistics for Biology and
Health. Springer, second edition.
Zimmerman, S. D. (2014). The Returns to College Admission for Academically Marginal Students.
Journal of Labor Economics, 32(4):711–754.
18

A Appendix - Tables and Figures
Table A.1: Ph.D. Programs Rankings
Ranking Amir and Knauff
(2008), R3 score
US News (2022)
1 MIT Harvard U
2 Harvard U MIT
3 Stanford U Stanford U
4 U Chicago Princeton U
5 Princeton U UC-Berkeley
6 Yale U U Chicago
7 UC-Berkeley Yale U
8 Northwestern U Northwestern U
9 U Minnesota Columbia U
10 LSE U Pennsylvania
Notes. We use the top eight Ph.D. programs as our measure of
selectivity since they coincide in both rankings.
19

Table A.2: ANPEC Ranking and Ph.D. Placement of Master’s Programs Enrollees (2004-2017)
Program Enrollment
ANPEC ranking (without Brazilian Economy) Ph.D. programs
Top Bottom Median Mean 1-15 1-30 All Abroad Top 8 (Any Field) Top 8 (Only Economics)
PUC-Rio 197 1 54 19 19 0.42 0.79 0.58 0.40 0.16 0.14
FGV-EPGE 242 1 67 30 30 0.27 0.50 0.64 0.34 0.12 0.12
IPE-USP 277 1 101 51 50 0.10 0.23 0.49 0.20 0.02 0.01
FGV-EESP 187 1 219 69 71 0.10 0.19 0.60 0.20 0.05 0.05
UnB 207 3 231 104 109 0.01 0.04 0.41 0.07 0.01 0.01
UFRJ 308 9 298 136 141 0.00 0.01 0.38 0.04 0.00 0.00
UFMG 164 15 341 155 161 0.01 0.01 0.43 0.09 0.01 0.00
USP-RP 178 14 651 204 211 0.01 0.01 0.34 0.03 0.00 0.00
UFRGS 290 54 1027 259 286 0.00 0.00 0.40 0.03 0.00 0.00
UNICAMP 285 17 1028 282 408 0.00 0.00 0.57 0.04 0.00 0.00
UFF 290 66 645 294 300 0.00 0.00 0.38 0.00 0.00 0.00
UFPE 156 26 1131 298 425 0.00 0.01 0.47 0.04 0.00 0.00
UCB 119 59 892 342 389 0.00 0.00 0.22 0.00 0.00 0.00
UFJF 133 105 934 350 366 0.00 0.00 0.40 0.00 0.00 0.00
UFPR 129 50 925 352 366 0.00 0.00 0.34 0.02 0.00 0.00
UFCE 189 65 982 354 369 0.00 0.00 0.54 0.01 0.00 0.00
ESALQ 167 83 1119 378 427 0.00 0.00 0.44 0.01 0.00 0.00
UERJ 164 65 848 403 408 0.00 0.00 0.27 0.00 0.00 0.00
UFSC 175 114 886 422 435 0.00 0.00 0.27 0.00 0.00 0.00
UFV 131 161 897 461 463 0.00 0.00 0.47 0.00 0.00 0.00
UFSCAR 81 136 1153 513 529 0.00 0.00 0.16 0.01 0.00 0.00
UFPB 76 195 1001 521 550 0.00 0.00 0.45 0.00 0.00 0.00
PUC-SP 326 107 1345 530 596 0.00 0.00 0.17 0.00 0.00 0.00
UFU 159 21 1162 548 558 0.00 0.01 0.45 0.00 0.00 0.00
UFBA 172 171 1014 550 555 0.00 0.00 0.22 0.00 0.00 0.00
UNESP 171 115 1279 567 573 0.00 0.00 0.31 0.00 0.00 0.00
UEM 126 223 1148 582 584 0.00 0.00 0.39 0.00 0.00 0.00
UFES 169 90 1296 584 594 0.00 0.00 0.14 0.00 0.00 0.00
PUC-RS 98 75 1210 602 617 0.00 0.00 0.26 0.00 0.00 0.00
Notes: The table reports the descriptive statistics of our 2004-2017 sample. Enrollment is the total number of students admitted through the ANPEC
exam who attended the master’s program for all the sample years. We do not present results for master’s programs with fewer than five students en-
rolled per year. We consider the ANPEC ranking that equally weights the subjects of Microeconomics, Macroeconomics, Mathematics, and Statis-
tics (i.e., without the Brazilian Economy). The ANPEC ranking without the Brazilian Economy test is the most relevant ranking for placement in
Ph.D. abroad and is used by the top four in most years of our sample. We show the lowest, highest, median, and mean ANPEC ranking by program
and the percentage of students enrolled in the top 15 (1-15) and top 30 (1-30) positions. We also show the Ph.D. placement percentage of students en-
rolled in Ph.D. programs in any field in Brazil or abroad (All) and only abroad (Abroad). The top eight Ph.D. programs are MIT, Harvard, Stanford,
Chicago, Princeton, Yale, Berkeley, and Northwestern; see Table A.1. We consider the top eight programs in either any field or just economics.
Table A.3: ANPEC Ranking and Ph.D. Placement of Master’s Programs Enrollees
Applied for
at least one
top four
Applied
for all
top four
Admitted to at
least one top
four
Admitted to all
top four
Enrolled to a
top four
Enrolled at
UnB
Enrolled at
USP/RP
Enrolled at
UFMG
Enrolled at
UFRJ
PhD program
abroad
Top eight PhD
program abroad
1-25 93% 59% 94% 94% 90% 2% 1% 1% 1% 40% 18%
26-50 87% 49% 88% 52% 81% 3% 0% 1% 1% 18% 2%
51-100 69% 29% 61% 1% 38% 11% 2% 2% 10% 9% 1%
101-150 56% 16% 20% 0% 4% 11% 6% 8% 14% 5% 0%
151-200 52% 8% 7% 0% 2% 4% 3% 8% 12% 3% 0%
201-250 41% 5% 1% 0% 0% 1% 5% 4% 5% 1% 0%
251-300 39% 3% 0% 0% 0% 0% 4% 1% 2% - -
301-350 32% 2% 0% 0% 0% 0% 2% 0% 0% - -
Notes: The table reports the descriptive statistics of our 2004-2017 sample. We consider only the most recent ANPEC subscription and the applicants ranked in the first 350 positions with non-missing undergraduate degree informa-
tion. Application for at least one top four or all top four considers if the student chose the top four MA programs as an option in ANPEC (FGV-EPGE, FGV-EESP, PUC-Rio, and IPE-USP). Enrollment represents the sum of the number
of students who have attended the program for all the sample years. We consider the ANPEC ranking that equally weights the subjects of Microeconomics, Macroeconomics, Mathematics, and Statistics (i.e., without the Brazilian
Economy). The ANPEC ranking is the most relevant ranking for placement in Ph.D. abroad and is used by the top four in most years of our sample. We show the Ph.D. placement percentage of students enrolled in Ph.D. programs
abroad or only in the top eight Ph.D. programs. The top eight Ph.D. programs are MIT, Harvard, Stanford, Chicago, Princeton, Yale,Berkeley, and Northwestern; see Table A.1. We consider the top eight programs in anyfield.
20

Table A.4: Minimum detectable effects, Multivariate regression
Ph.D. abroad Top eight Ph.D. abroad
70 80 90 70 80 90
FGV EPGE 0.095 0.108 0.125 0.051 0.057 0.066
PUC RIO 0.087 0.099 0.114 0.047 0.053 0.061
IPE USP 0.110 0.124 0.143 0.058 0.066 0.076
Notes. The table presents each outcome variable’s minimum detectable ef-
fect sizes for our matched sample with the full set of controls (i.e., the sam-
ple with 308 observations used in columns (6) in Tables 2and 3). We calcu-
late the minimum detectable effect (MDE) using the formula (4.17) from Vit-
tinghoff et al. (2012). The MDE represents the smallest parameter value for a
given sample size for which we can reject the null hypothesis in a two-sided
test at a 5% significance level. Each column represents a different statistical
power: 70%, 80%, and 90%. In all columns, we consider the residual stan-
dard deviation in each outcome after controlling for the master’s programs
dummies, year and undergraduate college fixed effects, and a cubic polyno-
mial of ANPEC scores. We consider the MDE sizes for a unit change in
the key explanatory variables (FGV-EPGE, PUC-Rio, and IPE-USP). ‘Ph.D.
abroad’ and ‘Top eight Ph.D. abroad’ include Ph.D. programs in any field.
Table A.5: Probability of attending a Ph.D. abroad in any field, top four institutions - only ANPEC
applicants ranked in the 25 first positions
All Students Same Option and Admission Students
(1) (2) (3) (4) (5) (6)
FGV-EPGE -0.051 -0.051 -0.057 -0.084 -0.080 -0.133
(0.083) (0.103) (0.084) (0.099) (0.138) (0.109)
[0.576] [0.587] [0.441] [0.449] [0.599] [0.195]
PUC-Rio -0.081 -0.039 -0.005 -0.141 -0.111 -0.101*
(0.055) (0.068) (0.056) (0.097) (0.098) (0.054)
[0.167] [0.508] [0.902] [0.182] [0.309] [0.106]
IPE-USP -0.247** -0.228* -0.153 -0.176* -0.206* -0.158
(0.088) (0.117) (0.105) (0.099) (0.113) (0.090)
[0.030] [0.078] [0.181] [0.108] [0.075] [0.071]
F-test 0.048 0.250 0.486 0.305 0.353 0.186
Dependent variable (mean) 0.446 0.446 0.446 0.451 0.451 0.451
Year fixed effects Yes Yes Yes Yes Yes Yes
Undergraduate college fixed effects No Yes Yes No Yes Yes
ANPEC score (cubic polynomial) No No Yes No No Yes
Number of observations 314 314 314 206 206 206
Notes. We restrict our sample to ANPEC applicants in the 25 first positions according to the ANPEC ranking. The dependent vari-
able is a binary variable equal to one if the applicant attended a Ph.D. program abroad in any field and zero otherwise. FGV-EESP is
the baseline institution. FGV-EPGE, PUC-Rio, and IPE-USP are dummy variables equal to one if the applicant enrolled in the master’s
program at FGV-EPGE, PUC-Rio, and IPE-USP, respectively, and zero otherwise. We control for ANPEC year and undergraduate col-
lege fixed effects. The ANPEC score equally weights the Microeconomics, Macroeconomics, Mathematics, and Statistics standardized
scores. We control for ANPEC scores using a cubic polynomial of scores. The sample ‘All Students’ contains all students enrolled in
a top-four master’s program. ‘Same Application and Admission’ is the matched sample, which is the ‘All Students’ sample restricted
to students who applied and were likely admitted by the top four programs (Dale and Krueger,2002). Cluster-robust standard errors
are shown in parentheses (at the ANPEC year level). *Significant at 10%; **significant at 5%; ***significant at 1%. Due to the small
number of clusters (i.e., fourteen), we also report wild bootstrap p-values calculated using Roodman (2015) in square brackets.
21

Table A.6: Probability of attending a top eight Ph.D. abroad in any field, top four institutions - only
ANPEC applicants ranked in the 25 first positions
All Students Same Option and Admission Students
(1) (2) (3) (4) (5) (6)
FGV-EPGE -0.023 0.047 0.060 -0.011 0.052 0.005
(0.127) (0.112) (0.111) (0.169) (0.163) (0.140)
[0.864] [0.680] [0.605] [0.947] [0.768] [0.964]
PUC-Rio -0.070 -0.029 0.011 -0.063 -0.025 -0.026
(0.109) (0.091) (0.101) (0.150) (0.138) (0.143)
[0.556] [0.743] [0.913] [0.689] [0.843] [0.838]
IPE-USP -0.224* -0.229* -0.156 -0.217 -0.230 -0.224
(0.111) (0.111) (0.108) (0.152) (0.145) (0.148)
[0.047] [0.044] [0.205] [0.198] [0.115] [0.155]
F-test 0.005 0.007 0.005 0.025 0.073 0.078
Dependent variable (mean) 0.201 0.201 0.201 0.223 0.223 0.223
Year fixed effects Yes Yes Yes Yes Yes Yes
Undergraduate college fixed effects No Yes Yes No Yes Yes
ANPEC score (cubic polynomial) No No Yes No No Yes
Number of observations 314 314 314 206 206 206
Notes. We restrict our sample to ANPEC applicants in the 25 first positions according to the ANPEC ranking. The dependent vari-
able is a binary variable equal to one if the applicant attended a top eight Ph.D. program abroad and zero otherwise. The top eight
Ph.D. programs are MIT, Harvard, Stanford, Chicago, Princeton, Yale, Berkeley, and Northwestern; see Table A.1. FGV-EESP is the
baseline institution. FGV-EPGE, PUC-Rio, and IPE-USP are dummy variables equal to one if the applicant enrolled in the master’s
program at FGV-EPGE, PUC-Rio, and IPE-USP, respectively, and zero otherwise. We control for ANPEC year and undergraduate col-
lege fixed effects. The ANPEC score equally weights the Microeconomics, Macroeconomics, Mathematics, and Statistics standardized
scores. We control for ANPEC scores using a cubic polynomial of scores. The sample ‘All Students’ contains all students enrolled in
a top-four master’s program. ‘Same Application and Admission’ is the matched sample, which is the ‘All Students’ sample restricted
to students who applied and were likely admitted by the top four programs (Dale and Krueger,2002). Cluster-robust standard errors
are shown in parentheses (at the ANPEC year level). *Significant at 10%; **significant at 5%; ***significant at 1%. Due to the small
number of clusters (i.e., fourteen), we also report wild bootstrap p-values calculated using Roodman (2015) in square brackets.
22