What Matters for the Decision to Study Abroad?

A Lab-in-the-Field Experiment in Cape Verde*

Catia Batista† David M Costa‡ Pedro Freitas§ Gonçalo Lima** Ana B Reis††

October 2024

Abstract
Study abroad migration is the fastest growing international migration flow.
However, the college completion rates of students from low-income countries are
often modest in OECD countries, raising the hypothesis that these migrants are
poorly informed about the costs and benefits of their decision. Our work tests this
hypothesis by running a lab-in-the-field experiment where graduating high school
students in Cape Verde are faced with incentivized decisions to apply for college
studies abroad. Our results show that potential migrants react strongly to
information about the availability of financial support and about college completion
rates. Since subjects’ prior beliefs on availability of financial support are
overestimated, it is likely that study migrants need to shift their time from study to
work after uninformed migration, which likely harms their scholar performance.
Policies that inform potential migrants of actual study funding possibilities should
decrease study migration flows but may improve successful graduation.

JEL Classification: O15, F22, J61, C91

Keywords: international study migration; lab-in-the-field experiment; education;
information; uncertainty

*
The authors gratefully acknowledge comments from Toman Barsbai, Cara Ebert, Eliana La Ferrara, David
Lagakos, Ben Marx, David McKenzie, Caroline Theoaharides and other participants at several seminars and
conferences. We are grateful to our project coordinator Gonçalo Gameiro for superb work on all aspects of project
implementation and data management, as well as to Isabel Gouveia, Mariana Parente and the local team of enumerators
for their dedication and careful fieldwork. Funding was provided by the Aga Khan Foundation, grant FCT-
AGAKHAN 15417394041-2019.
†
Nova School of Business and Economics and NOVAFRICA, CReAM, IZA and JPAL. Email:
[email protected]
‡
Nova School of Business and Economics. Email: [email protected]
§
University of Oxford. Email: [email protected]
**
European University Institute. Email: [email protected]
††
Nova School of Business and Economics. Email: [email protected]
 1 Introduction

International study migration has increased faster than overall international migration in the
last decades. The stock of international study migrants went up from about 2 million in 2000 to
6.4 million in 2020, a growth of over 200%. Overall international migration, on the other hand,
grew just 62% over the same period.1
A large share of the increase in international study migration is driven by students from low-
and middle-income countries. Significantly, over 400,000 students from Sub-Saharan Africa
(SSA) migrate internationally with the purpose of studying for a college degree.2 These flows of
educated migrants have been the object of intense academic and policy debate over the years, since
Grubel and Scott (1966) and Bhagwati and Hamada (1974) first proposed that the loss of educated
individuals by a country may have negative direct consequences on key activities like research or
public services (especially healthcare and education), and governmental institutions - potentially
exacerbated by the loss of human capital externalities. However, a body of more recent empirical
evidence has highlighted the many positive impacts of educated emigration for the economic
development of the countries of origin – namely through both financial and immaterial
remittances, i.e. the positive effects on educational and health investment, the quality of political
institutions, entrepreneurship, financial sector and macroeconomic stability, innovation, FDI and
trade linkages, among others.3
In addition to the positive development impact of emigration for the countries of origin,
international study migration is likely one of the most effective pathways for individual
opportunity for students from low-income regions or countries. Individuals who earn a college
degree abroad guarantee their recognition, which facilitates access to the labor market and allows
these migrants to substantially increase the labor income they would otherwise earn in their home

1
United Nations Department of Economic and Social Affairs, Population Division (2020). International Migrant
Stock 2020.
https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/undesa_pd_2020_ims_stock_
by_sex_destination_and_origin.xlsx Last accessed on Jan 14, 2024.
2
UNESCO (2023), Outbound Internationally Mobile Students by Host Region. Available at:
http://data.uis.unesco.org/index.aspx?queryid=3807. Last accessed on January 14, 2024.
3
See, for example, Gould (1994); Edwards and Ureta (2003); Beine et al. (2008); Spilimbergo (2009); Batista
and Vicente (2011); Javorcik et al. (2011); Batista et al. (2012); Docquier et al. (2016); Batista et al. (2017); Barsbai
et al. (2017); Karadja and Prawitz (2019); Batista et al (2019); Bahar et al (2019); Bahar et al. (2022).

2
country.4
A potential problem in this virtuous cycle of beneficial international study migration is that
the performance and educational attainment of these students is often below average, potentially
limiting their access to quality education and matching employment opportunities. For instance,
in France, the overall average college completion rate of 75% drops to just 54% among students
who migrated from countries in the Maghreb.5 In Portugal, less than four in ten Cape Verdean
students successfully complete their college education, which compares to an average graduation
rate of 68%.6
In this context, it is important to understand how well-informed potential migrants are about
the costs and benefits of their study migration decisions, and what role information gaps play in
these decisions - specifically, the role of information gaps on college completion rates abroad and
on financial support availability.
Our paper addresses these research questions through an incentivized lab-in-the-field survey
experiment with relatively large stakes combined with an individual level survey. This approach
allows us to examine determinants of international study migration such as liquidity constraints
and uncertainty using within-individual variation, while assessing the heterogeneous impacts of
potentially important individual characteristics such as gender or academic ability. In this way, we
measure individual migration preferences in the origin country before actual migration decisions
take place.
We focus on a relevant international study migration corridor between Africa and Europe:
migration flows from Cape Verde to Portugal. Cape Verde is a small island country in Sub-Saharan
Africa, in which the equivalent to 38% of the resident population have emigrated, mostly to
Portugal. Most migration flows are for the purpose of studying. However, the educational
attainment of migrant students is significantly below average. Our setting is one of many corridors
of international study migrants moving to pursue college education in European countries with
former colonial ties – this applies to several African countries to France and the UK, or Latin

4
For orders of magnitude of the income gains of migration see, for example, McKenzie et al. (2010).
5
Marmolejo, F., S. Manley-Casimir and S. Vincent-Lancrin (2008). "Immigration and Access to Tertiary
Education: Integration or Marginalisation?", in Higher Education to 2030, Volume 1, Demography, OECD Publishing,
Paris. https://doi.org/10.1787/9789264040663-10-en Last accessed on Jan 14, 2024.
6
Only 36% of Cape Verdean students that started college in Portugal in the years of 2013 and 2014 completed
their bachelor’s degrees seven years later. Data on student college performance in Portugal are collected by the
Portuguese Ministry of Education and made available in the RAIDES database. Last accessed on Jan 12, 2024.

3
American countries to Spain. In many of these European countries, there are immigration policies
that favor students from former colonies, although of course each corridor has specificities and
idiosyncrasies. The context of our study is thus informative to countries whose immigrant
populations face similar challenges.
The experimental subjects in our sample are 17- and 18-year-old high school seniors in Cape
Verde, a population that has a high propensity to emigrate. Students were surveyed and played
games in our experiment about six months prior to making their college application. At that time,
90% of the respondents reported their intention to apply to college in Portugal. Individuals in our
sample present a mean expectation for average college completion rate of 48% - an overestimation,
since the actual college completion rate by Cape Verdeans in Portugal is only 36%. Similarly,
students expected that 43% of Cape Verdeans studying for college in Portugal held a scholarship,
while only 13% actually benefited from this type of financial support.
The experiment elicited incentivized hypothetical migration choices for studying abroad
under scenarios with different information sets, which varied the probability of successful college
graduation and the likelihood of obtaining financial support to study. The results showed strong
responses both in terms of magnitude and statistical significance. Cape Verdean students were 36
percentage points (pp) more willing to study for college in Portugal in response to having their
college studies abroad paid for. The increase in the probability of study migration was 22 pp when
the likelihood of graduating went from 0% to 40% (the approximate actual graduation probability
observed in the data), and 30 pp when the average probability of college completion increased
from 0% to 100%. These effects compare to a 30% base probability of migrating to Portugal with
0% probability of graduating, no financial assistance to study and an alternative of graduating from
college in Cape Verde after four years.
Our findings contribute mainly to three strands of literature. First, they contribute to the
literature on the determinants of international study migration. The early work of Rosenzweig
(2008) was followed by other research using cross-country and migrant destination data. Notable
examples are Kato and Sparber (2013) who find that more restrictive immigration policies (H-1B
visa caps in the USA) reduce the quality of prospective international undergraduate students. Beine
et al. (2014) estimate a gravity model for 13 OECD destination countries for international students
and show that high quality education and migrant networks are the most predictive of international
study migration flows. Beine et al. (2020) explore within-country data to estimate a gravity model

4
and find a negative association between tuition fees and international students in Italian
universities. More recently, Arenas (2021) uses university data in Catalonia to assess the role of
human capital portability on international student migration. Specifically, she measures the impact
of a policy change decreasing barriers to university access to international students from specific
countries, which increases both their quantity and quality. By using a laboratory experiment in the
field combined with individual level survey, our work allows us to examine determinants of
international study migration such as liquidity constraints and uncertainty using within-individual
variation, while assessing the heterogeneous impact of individual characteristics such as gender,
income or risk aversion. Our work also takes a different angle in that we measure how migration
intentions are determined by individuals still in the origin country. This allows us to measure
individual migration preferences before actual migration decisions take place – which may
eventually be hindered by (potentially endogenous) migration barriers such as visa unavailability
or changes in family circumstances.
Our paper also fits within the recent migration literature using lab experiments to measure the
determinants of international migration. However, contrary to the extant literature, our study is
specifically focused on international study migration. Most economics literature has explored the
determinants of work migration more broadly, starting with the classical models of Sjaastad (1962)
or Harris and Todaro (1970). Some studies have examined selection issues related to differences
in wage distribution and liquidity constraints to mobility, as Borjas (1987), McKenzie et al (2010)
or Grogger and Hanson (2011). Additional factors such as imperfect information and migrant
networks have been recognized and modeled as crucial determinants of international migration
(for example, by McKenzie et al., 2013, or Beine et al., 2011). Recent lab experiments, such as
Batista and McKenzie (2023), have simultaneously tested these different factors by eliciting
incentivized hypothetical migration decisions – which would not be possible to perform using
observational data given multiple endogeneity problems. Similarly, Barnett-Howell (2018) used a
migration video game laboratory experiment to investigate how individuals decide on migration
destination, revealing a significant role for imperfect information in explaining why individuals
do not migrate more often. Bah and Batista (2018) conducted a lab-in-the-field experiment to
measure how decisions regarding irregular migration to Europe were affected by the risk of travel
death and the risk of not receiving asylum upon arrival to Europe. Our approach in this paper is
very similar in that we conducted a laboratory experiment with potential migrants in order to

5
identify the most relevant determinants of international study migration – and their covariates. This
lab experiment approach to study migration follows earlier work using lab experiments to
understand migrant remittance behavior – e.g. Batista et al. (2015) and De Arcangelis et al. (2015).
Finally, this paper is closely related to the broad “brain drain” vs “brain gain” debate on
whether the emigration of the best educated of a country harms or supports development at the
country of origin, as discussed by Docquier and Rapoport (2012). Batista et al (2012) found that,
precisely for the case of Cape Verde, emigration contributed to investment in adolescent education.
Similar findings have been uncovered for other countries, such as Malawi (Dinkelman and
Mariotti, 2016), the Fiji Islands (Chand and Clemens, 2023) or the Philippines (Khanna et al.,
2022; Abarcar and Theoharides, 2024). While our findings do not speak to the same research
question, the fact that prospective migrants strongly react to variations in college completion rates
abroad - which will affect future migrant earnings – may possibly mediate and change the brain
gain effects estimated in the existing literature.

2 Setting and Experimental Design

2.1 Setting: Cape Verde

Cape Verde is a Sub-Saharan African country of nine inhabited islands, located off the West
Africa coast, and historically characterized by massive emigration. In 2020, it recorded a total
resident population of 491,233 people, and a stock of 187,558 Cape Verdeans living abroad –
equivalent to 38% of the resident population.7 According to the 2021 census, 17,868 people (or
3.6% of the resident population) had emigrated over the previous five years. The main destination
countries over this period were Portugal (62%), the United States (18%) and France (7%). Migrant
remittances to Cape Verde represented 14.1% of the GDP in 2022, one of the largest ratios in the

7
United Nations Department of Economic and Social Affairs, Population Division (2020). International Migrant
Stock 2020.
https://www.un.org/development/desa/pd/sites/www.un.org.development.desa.pd/files/undesa_pd_2020_ims_s
tock_by_sex_destination_and_origin.xlsx Last accessed on Jan 14, 2024.

6
world.8 Most relevant for our study, Cape Verde has one of the highest literacy rates in Sub-
Saharan Africa at 91% in 2022, contrasting with an average of 68% in the continent.9

According to the 2021 Cape Verde census, those emigrating with the purpose of studying
abroad represented the largest share of total emigration (39.6%). Portugal was the main receiving
country (with over 80% of Cape-Verdean migrant students). Study migration is also the main
purpose of emigration from Cape Verde to Portugal representing 53% of the flow of migrants over
the above mentioned five years period.

According to the official college statistics of Portugal, the number of Cape Verdean college
students in Portugal increased substantially over the last decade. However, the academic success
of these migrant college students is modest. Considering two cohorts of students from the academic
years of 2013/14 and 2014/15, 17% dropped out after their first year of college (compared to only
6% of the Portuguese students), and only 36% of them were able to graduate in the seven years
since their arrival (compared to 68% of the Portuguese students). 10

2.2 Experimental design and Sample

We designed a lab-in-the-field experiment to investigate the study migration decisions of

Cape Verdeans, which was implemented in multiple high schools across Cape Verde in the
academic year 2022/2023. Laboratory experiments allow for controlled variation in factors
potentially affecting migration decisions, which are otherwise difficult to disentangle in
observational data given the many dimensions of endogeneity related to migration decisions. Using
our experimental variation, we were able to elicit migration choices under different hypothetical
scenarios from each individual in our sample, which allows us to establish the impact of the main
migration drivers of interest. These migration choices made by experimental subjects in the lab are
closely related to migration intentions. They correlate well with real-world steps taken in

8
World Bank World Development Indicators (2023).
https://datacatalog.worldbank.org/search/dataset/0037712/World-Development-Indicators Last accessed on
Jan 14, 2024.
9
Ibidem.
10
Data on student college performance in Portugal are collected by the Portuguese Ministry of Education and
made available in the RAIDES database. Last accessed on Jan 12, 2024.

7
preparation for migration and are likely good predictors of actual future migration. Lab migration
choices will, however, naturally differ from actual migration decisions, which must reflect barriers
such as financial costs and immigration policies, that mediate the relationship between migration
intentions and actual migration.
Our sample includes 466 students with a high potential for study migration, namely those in
their senior high school year (12th grade) about to make migration and study decisions. Fieldwork
took place in 25 randomly selected classes among the seven largest public high schools in the
country, as measured by school-level administrative data provided by the Cape Verdean Ministry
of Education (MoE), in the two largest islands of the country, Santiago and Santo Antão. Data
were collected over a period of 10 weeks, between the months of September and December 2022.
The implementation was coordinated with the MoE, which authorized both the school visits and
the data collection efforts, but no representatives of the MoE were directly involved in the school
visits or in implementing the survey. All students’ parents were informed about the study, and
participation was entirely voluntary. Surveys were conducted by enumerators in schools, using
Qualtrics software in electronic tablets. Each session of the lab-in-the-field was conducted outside
the students’ academic schedule.
The lab-in-the-field elicited students’ incentivized hypothetical study migration choices in a
specific host country---Portugal---which is the main out-migration destination for high school
graduates from Cape Verde. Experimental subjects played 13 rounds corresponding to different
hypothetical scenarios, where in each round subjects were asked to make a migration choice
between staying in Cape Verde or migrating to study in Portugal. In each round, we varied three
parameters, made explicit to the subjects. First, we presented a different successful college
graduation rate for Cape Verdeans students in Portugal (0%, 40%, 100%).11 Second, we stated
whether the subject would receive a full scholarship covering all costs during their study years in
Portugal or not. Third, we described whether the option to stay in Cape Verde entailed working
full-time with an upper secondary degree or studying for college in Cape Verde and working full-

11
Students were informed about college graduation probabilities in Portugal according to the unconditional
proportion of Cape Verdean students that hypothetically graduate from college in Portugal, not the subjective
expectation of graduation. From the three scenarios, two have no uncertainty: one in which no one graduates, and thus
the student necessarily fails to attain a bachelor’s degree while incurring the cost of studying; and another one in which
the student attains a college-degree. The third central scenario, with graduation uncertainty, approximates the
proportion of Cape Verdean individuals, who completed high school in Cape Verde and enrolled in Portuguese college
institutions, that graduate from college in Portugal over a period of seven years (36%, according to college
administrative data from Portugal).

8
time only after successfully completing college in Cape Verde. The combination of these three
parameters and respective alternatives (3x2x2) results in 12 rounds. Figure 1 depicts the
combinations of parameters in each of these rounds in a schematic way. Following Batista and
McKenzie (2023), we introduced one additional round (henceforth, the “home bias” round). In this
round, students were faced with precisely the same pecuniary conditions in both Cape Verde and
Portugal to assess the relative preference of each destination based on other amenities, which were
not presented in the game.

Figure 1. Hypothetical Scenarios in each Round

Round Stay in Cape Verde Migrate to Portugal
1 0% Graduate College
2 Without Scholarship 40% Graduate College
3 100% Graduate College
Work with High School Degree
4 0% Graduate College
5 With Scholarship 40% Graduate College
6 100% Graduate College
7 0% Graduate College
8 Without Scholarship 40% Graduate College
9 100% Graduate College
Work after Completing College
10 0% Graduate College
11 With Scholarship 40% Graduate College
12 100% Graduate College

In each lab session, field enumerators described the general framing and practical
instructions of the experiment to subjects (see Appendix A1). Subjects were also given the
opportunity to read the instructions themselves. After all the details of the experiment were made
clear, subjects played two practice rounds, before proceeding to the actual choices. For each
round, students were presented with a hypothetical endowment of 110000 Cape Verdean
escudos (ECV), about EUR 1000, which could be used to either cover the costs of migrating to
Portugal to pursue a college degree or be kept if the choice was to stay in Cape Verde.
Experimental subjects were told about the costs and wage returns associated with each choice,
in each round, and they were told to consider a time horizon of 10 years, no unemployment and
no return migration before the end of the 10-year horizon. Under the scenarios of no scholarship
in Portugal, students faced an annual cost of EUR 5220. When the scenario awarded students with
9
a full scholarship, players would be exempt from living costs for the first three years in Portugal.
In Cape Verde, annual costs were set to EUR 3768.12 For wages in Portugal, we set the mean
annualized wage of the Cape Verdean employed population in Portugal, separately by levels of
education: a base wage of EUR 7524 for the case where no-one graduates from college; and EUR
9724 for the case where all students graduate from college. For the case where only 40% graduate
from college, we sum to the base wage the expected value of the education premium (7524 + 2200
x 0.4 = EUR 8404). For the Cape Verde option, the wage was assumed to be EUR 5772 for
scenarios with no college education degree, and EUR 7140 in the case of college graduation.13 In
the scenario of studying in Cape Verde, we considered zero income in the four initial years and
the EUR 7140 wage for the following years. All costs and returns were presented year-by-year,
with no time discounting. A summary of the parameters and payoff scheme for each game round
is presented in Appendix Table A1.
All information was presented in the electronic tablet through showcards providing graphical
depictions of each round’s parameters (see Appendix A2).
The order in which the subjects were presented with each round was block-randomized as
follows. Students were first randomly presented with either the “Work with High School in Cape
Verde” or the “Work with College in Cape Verde” set of scenarios, in which every other scenario
in Portugal was randomized within each of these sets. For each subject, the final round was
always the “home bias” round.
Choices were incentivized with a relatively high stake. The average payout to subjects was
EUR 4.77, which was about the daily minimum wage. After the last round, each student’s payoff
was determined by randomly selecting one of the twelve rounds played14 and applying an exchange

12
All expected costs were computed according to annual living costs of students in Portugal, including housing
and tuition fees. Due to lack of data on students’ living costs in Cape Verde, the cost of living for the country was
estimated according to a parity purchasing index.
13
The computations of these annual earnings resulted from an analysis of the Cape Verdean employed population
available in the Portuguese Private Sector Employment Survey (Quadros de Pessoal), in the year 2017. For the case of
labor returns in Cape Verde, we used administrative data on public sector employees to infer wages for both high school
and college graduates. Because wages in both countries are, on average, higher in public sector jobs, the mean wage
assumed for Portugal (Cape Verde) is a lower (upper) bound of the unconditional mean wage across private and public
sectors.
14
Excluding the “home bias” round.

10
 rate from the hypothetical scenario to real units, namely EUR 1 in mobile phone credit for each
hypothetical EUR 3400. Negative hypothetical returns represented no mobile phone credit.
Before the experiment, students answered a short survey including questions on their age, parental
background, and household socioeconomic status. We collected further information on each subjects’
family migration history, individual propensity to migrate, as well as subjective priors on the college
graduation probabilities of Cape Verdean students both in Cape Verde and Portugal. Finally, we
elicited each students’ risk aversion through a hypothetical lottery. Subjects took an average (and
median) of about 53 minutes to complete the survey, including the migration choice elicitation game
rounds.

2.3 Descriptive statistics and subjective beliefs

Table 1 presents descriptive statistics for the sample of 12th grade Cape Verdean students used
in our empirical analysis. On average, subjects are 18 years old and close to 65% of them are female.
We measured the socio-economic status (SES) of participants using the number of items owned
by the household from a list of 14.15 On average, subjects had fewer than 5 of those items at home,
although there is considerable variation in the sample. As a proxy of mathematical ability, we measured
each subject’s proficiency in answering a set of four simple arithmetic questions. We coded as high
ability those that answered correctly to all four questions. According to this measure, slightly above
one quarter of the participants are high ability. Reflecting the out-migration propensity of the Cape
Verdean population, 18 percent of participants reported having close family---defined as at least
one of the parents or siblings---living in Portugal. Students in our sample displayed an intermediate
level of risk aversion measured using a hypothetical lottery – they reported an average willingness
to gamble 39% of the marginal earnings on a windfall on a fair lottery, which is slightly higher
than the same measure for a sample of graduating college students in Portugal and Kenya reported
by Batista and McKenzie (2023).
We also surveyed students on their preferences and subjective expectations regarding college
and study migration. Almost all subjects report a willingness to apply to college (99%), with 90%

15
The items in our SES index are: a desk to study; individual bedroom; calm place to study; computer; internet;
books; electricity; calculator; dictionary; car/motorbike/bicycle; tv; dishwasher; washing machine; smartphone.

11
reporting a willingness to apply to college in Portugal. These values provide a good illustration of
how prone our sample is to continue studying beyond upper secondary education. Interestingly, in
line with the recent survey of evidence by Bursztyn and Yang (2022), experimental subjects have
beliefs about the proportion of high school graduates willing to apply to college in Portugal that
are significantly different from their own average propensity (52% vs. 90%, respectively, as shown
in Table 1).
When asked how likely they were (on a scale from 0 to 10) to live, for at least six months, in
Portugal in the next two years, the mean reported value was relatively high (7.5) and more than half
of students reported a value equal or higher to eight with almost 40% being certain of taking that
migration decision. We observe this high propensity to migrate despite a low expectation to graduate
from college in Portugal. Among the surveyed students only 48% believed Cape Verdean students
would be able to graduate from college in Portugal and around 60% of those graduates would find a
job in Portugal, whereas only 54% of those same graduates would find a job in Cape Verde after
returning to the country. Of course, when contrasting own aspirations and prior beliefs, it is important
to recall that these priors are elicited for the average population of migrant students and not for their
own individual probability of graduation abroad, which are likely to differ substantially, again in
line with Bursztyn and Yang (2022).
When contrasting the subjective expectations of the students in our sample with our best
estimates for the population, we find that the former are often inaccurate. Students typically
overestimate college graduation rates and scholarship availability, and underestimate rates of job
finding. As shown in Table 1, sampled students report beliefs that 43% of Cape Verdean college
students in Portugal hold a scholarship, while only 13% do. They believe 48% of these students are
able to graduate from college degrees in Portugal, whereas only 36% do. The same happens with
beliefs on college graduation rates in Cape Verde, which are 58% compared to an actual graduation
rate of 42%. In contrast, sampled students hold underestimated beliefs regarding employment rates
of Cape Verdeans who graduate from college in Portugal: they expect that 60% find a job, whereas
85% of them do. Similarly, students in our sample believe that only 39% of graduates from Cape
Verdean colleges can find a job, whereas 90% of them do.
The high propensity to migrate displayed by the experimental subjects in our sample is in line
with the historical patterns observed in Cape Verde and described in the previous section, and
also with the choices made in our experiment. Indeed, on average, students chose to migrate in 67%

12
of the rounds and 15% of the students chose to migrate in all game rounds, even in some of the
rounds where costs of migration are higher than returns. Given this high propensity to migrate we
also collect information on the actual steps students already took regarding preparation for
migration. Following Bah et al. (2023) we use a set of five questions which capture different
preparatory steps to migration: a) talked with someone living in Portugal; b) looked for information
related with migration to Portugal; c) started saving money to migrate; d) started taking care of the
visa to migrate to Portugal; e) already knows where to stay after migrating to Portugal. Each
question is coded as 1 if student took the respective step, and 0 otherwise. We report a variable
which corresponds to the sum of the binary answers in the five step questions, and which by
construction is lower bounded by zero and upper bounded by five.

13
Table 1: Descriptive statistics for baseline survey and choices in lab games
(1) (2) (3) (4) (5) (6) (7)
Official Estimates (%) Mean (%) Std. Dev. (%) 10th Pctl. Median 90th Pctl. N

Female (%) 65.66 47.54 0 100 100 463

Age 18.12 1.157 17.05 17.73 19.86 460
SES Index (0-14) 4.532 4.371 0 4 12 466
High Ability (%) 27.90 44.90 0 0 100 466
Parents or Siblings Living in Portugal (%) 17.60 38.12 0 0 100 466
Risk Aversion measured using Hypothetical Lottery (0-10) 6.126 2.903 1 6 10 452
Wants to apply to college (%) 98.82 10.81 100 100 100 424
Wants to apply to college in Portugal (%) 89.86 30.22 0 100 100 424
Wants to migrate to Portugal (0-10) 7.44 2.75 4 8 10 290
Sum migration steps 0.82 1.16 0.00 0.00 3.00 466

Subjective expectation on college application rates by high-school graduates in CV (%) 20.81 17.48 0 20 50 459
Subjective expectation on college application rates to study in CV by high-school graduates in CV (%) 26.52 16.53 0 30.00 50 460
Subjective expectation on college application rates to study in PT by high-school graduates in CV (%) 51.77 23.96 20 50 80 457
Subjective expectation on probability of holding scholarship to study in PT (%) 13 42.53 19.82 20 40 70 446
Subjective expectation on college graduation rate in PT (%) 36 48.15 21.15 20 50 80 443
Subjective expectation on graduates from PT finding job in PT (%) 85 60.39 27.65 20 60.00 100 441
Subjective expectation on graduates from PT finding job in CV (%) NA 54.32 24.49 20 50 90 437
Subjective expectation on college graduation rate in CV (%) 42 57.69 18.42 40 60.00 80 451
Subjective expectation on college graduates from CV finding job in CV (%) 90 38.81 17.22 20 40 60.00 447

Fraction of rounds where migration was chosen (%) 67.45 21.48 41.67 66.67 100 466
Experimental subjects always choosing to migrate (%) 15.45 36.18 0 0 100 466
Experimental subjects never choosing to migrate (%) 0.43 6.54 0 0 0 466

Notes: Official estimates in column (1) are shown for comparison with data from survey in column (2). The probability of holding a scholarship to study in PT was calculated using a 2021/2022 survey of Cape
Verdean students in Portuguese universities by Batista et al (2023). The college graduation rate in PT was calculated from the Portuguese Ministry of Education administrative data RAIDES(2023). The rate of
graduates from PT finding a job in PT is based on a survey of Cape Verdean immigrants residing in Portugal by Batista et al (2022). The college graduation rate in CV was obtained from the Cape Verde Ministry
of Education DGES/ARES database. The rate of graduates from CV finding a job in CV is based on a survey of the Cape Verde National Statistics Institute - Inquérito Multi-objectivo Contínuo (2020).

14
3 Empirical analysis

3.1 Econometric specifications

We estimate the following Linear Probability Model (LPM):

Mir = 𝛼 + α1 Pg40%
ir
+ α2 Pg100%
ir
+ α3 Sir + α4 Wir + θi + μir (1)

The dependent variable Mir is a binary variable taking value 1 if student i chose to migrate in
game round r. We measure the effect that the probability of graduating from college abroad has
in the experimental subjects’ willingness to migrate through the indicator variables Pg40%
ir
and

Pg100%
ir
, which correspond to the rounds where the proportion of migrant students graduating
from college abroad is 40% and 100%, respectively. The variable Sir indicates whether student i
received a scholarship for studying abroad in game round 𝑟. Finally, we also include a binary
variable taking value 1 when the alternative to migrating is to work in the home country without
a college degree (Wir ) - as opposed to studying for college in the home country (this is the
omitted category). For precision and to absorb variation from unobserved individual
characteristics, we include individual fixed effects (𝜃 ) in our econometric model. To account
for correlation in choices within classrooms, we cluster the standard errors at the class level.
We are interested in estimates of 𝛼 , which measure the extent to which each of the
different corresponding factors affects the willingness to migrate of the experimental subjects
in our sample.

We extend model (1) to test for heterogeneous effects across a set of group characteristics.
We estimate the following model:

Mir = 𝛼 + α1 Pg40%
ir
β1 Pg40%
ir
* 𝑋 + α2 Pg100%
ir
+ β2 Pg100%
ir
*𝑋

+ α3 Sir + β3 Sir * 𝑋 + α4 Wir + β4 Wir * 𝑋 + θi + 𝜗ir

(2)

15
 Each main regressor of interest is now interacted with a specific individual characteristic (𝑋 ),
and we are interested in estimating the marginal effect of this interaction with our original main
variables of interest. These new estimates are captured by the set of coefficients 𝛽 .

3.2 Main estimation results

Table 2 presents our main results. We find that information about the academic success of
migrant students in the destination country, as well as receiving a full scholarship covering all studying
costs, strongly predicts willingness to migrate. Increasing the unconditional probability of college
graduation abroad from 0% to 40% increases the probability of choosing to migrate by 22 percentage
points (p.p.). We also find that removing uncertainty about college graduation in the destination
country increases the probability of migration by 36.1 p.p. Furthermore, fully covering living and
tuition costs would increase the probability of study migration by 29.6 p.p. Interestingly, we find that
subjects are not responsive to changes in pecuniary conditions in the home country. Although subjects
slightly prefer to migrate (by 1.6 p.p.) when the alternative option is to work without a college degree
in the home country, we cannot reject that the effect is null even at a 10% significance level. In this
context, pull factors seem more relevant for migration to study than push factors, as out-migration is
relatively inelastic to home country conditions.16
Overall, the evidence suggests a strong preference for migration, even when there are no college
returns. For hypothetical scenarios where no migrant students graduate from college abroad and have
no scholarship support, more than 30% of participants reveal a preference for migration. Given the
hypothetical payoffs in the game, this implies that about one third of the sample would be willing to
forgo EUR 4500 over a period of 10 years to migrate without obtaining a college degree abroad,
indicating that there are non-pecuniary benefits of migration to the individuals in our sample which are
not explicitly captured by our experimental design.

16
For the 10-year horizon considered in the experiment, the option to work without a college degree in the home
country entails a net gain of EUR 14 880 in the context of the experiment.

16
 Table 2: Lab decisions to migrate
(1)
Chose Migration

Prob. Graduation 40% 0.220***

(0.021)
Prob. Graduation 100% 0.361***
(0.032)
Holding Scholarship 0.296***
(0.022)
Working in Home Country as Alternative to Migration 0.016
(0.010)
Constant 0.323***
(0.023)

Observations 5,512
Adjusted R-squared 0.353
Notes: The dependent variable is a binary variable taking value 1 if the student
chooses to migrate. Omitted category has 0% graduation rate abroad, no
scholarship available, and study in home country if not migrating. LPM estimates
include individual fixed effects. Standard errors clustered at the class level in
parentheses. Significance levels: *** p<0.01 ** p<0.05 * p<0.1.

It is relevant to measure if these empirical results are purely driven by income

maximization choices, disregarding preferences for other migration relevant parameters, such as
holding a scholarship or uncertainty regarding college degree completion.
Figure 2 plots the net payoff from choosing migration in each round against the share of
students who make the migration choice in that round. The positive slope of the correlation
between lab payoffs and in the lab choices to migrate points to students understanding and
responding to monetary incentives. But it is also clear that students are doing more than just
choosing the highest payoffs in each lab round.
We show that different shares of students choose the migration option for similar payoffs.
For example, for rounds around the zero-payoff threshold, we observe that the average share of
students who choose to migrate ranges between 50% and 90%, depending on the different values
of the parameters. The lowest percentage of migration choices (30%) happens when there is no
scholarship, the probability of graduation is zero and the alternative to migration is studying at

17
higher education in Cape Verde. Increasing the probability of graduation to 100% but changing
the alternative to migration to Portugal to working as a high-school graduate in Cape Verde implies
a small change in the net pay-off of migration (from EUR -5,692 to EUR -5,032), but a large
increase in the percentage of students that choose the migration option in the lab, to almost 80%.
When the probability to graduate is kept at 0% but the scholarship increases to cover 100% of
costs in the first 3 years in Portugal, the net pay-off is again very similar (EUR -4,912) but the
percentage of students that choose to migrate is only around 60%. This range of migration choices
shows that students react differently to the different parameters.
Figure 2 also shows that for negative payoffs of migration---mostly scenarios without
scholarship availability---the share of students choosing to migrate decreases substantially. This
pattern also indicates that students understand the stakes of the game and the experimental payoffs.
These results illustrate how, across the different rounds, the subjects’ choices in the lab
were not being purely driven by a payoff maximization behavior and that the migration parameters
weighed in students’ lab choices.

18
 3.3 Heterogeneous responses by gender and academic ability

We estimated heterogeneous responses to potentially relevant individual student

characteristics, like gender and academic ability.17
As displayed in Table 3, female students are less sensitive than male students to changes in
the graduation rate, with a difference of 7.8 p.p. and 8.7 p.p. in the response to increases in college

17
We also examined differential lab migration choices with respect to socioeconomic status (SES), network at
destination country and risk aversion. These estimation results are shown in the Appendix Tables A2-A4. Students
with higher levels of risk aversion appear to make similar lab migration choices than those with lower levels of risk
aversion, except that more risk loving individuals seem slightly less sensitive to the alternative of working (as opposed
to studying) in the home country. Interestingly, the students’ choices in the lab are unaffected by their socioeconomic
status or by their network at destination, i.e., the existence of close family living at destination. This result may be
related to the very high baseline propensity to migrate among the surveyed students.

19
graduation rates in Portugal. Also, female students increase their likelihood to migrate by 6.8 p.p.
in a migration alternative scenario of working (instead of studying) in the home country while
male students do not change their choices. However, we find no differences in responsiveness to
financial support across gender groups.

Table 3: Lab decisions to migrate interacted with gender

(1)
Chose Migration

Prob. Graduation 40% (α1) 0.271***

(0.037)
Prob. Graduation 40% * Female (β1) -0.078**
(0.037)
Prob. Graduation 100% (α2) 0.418***
(0.051)
Prob. Graduation 100% * Female (β2) -0.087*
(0.048)
Holding Scholarship (α3) 0.269***
(0.030)
Holding Scholarship * Female (β3) 0.041
(0.034)
Working in Home Country as Alternative to Migration (α4) -0.030
(0.019)
Working in Home Country as Alternative to Migration * Female (β4) 0.068**
(0.030)
Constant 0.323***
(0.022)
p-value (α1 + β1=0) 0.000
p-value (α2 + β2=0) 0.000
p-value (α3 + β3=0) 0.000
p-value (α4 + β4=0) 0.026
Observations 5,476
Adjusted R-squared 0.356
Notes: The dependent variable is a binary variable taking value 1 if the student
chooses to migrate. Omitted category has 0% graduation rate abroad, no scholarship
available, and study in home country if not migrating. LPM estimates include
individual fixed effects. Standard errors clustered at the class level in parentheses.
Significance levels: *** p<0.01 ** p<0.05 * p<0.1.

20
 Table 4 shows that high ability students (i.e., students who showed higher proficiency levels
in arithmetic problems) are more sensitive to both more certain college graduation rates and to
holding a full scholarship than lower ability students, with coefficients being statistically different
at the 5% level. As shown in Table 4, when faced with a 40% (100%) increased probability of
college graduation abroad, highly skilled students depict a stronger increase on the probability to
migrate of 8.1 p.p. (10 p.p.), reacting more to this change than low skilled students. Highly skilled
students also show a 7.1 p.p. higher probability to choose to migrate when holding a full
scholarship. We find no difference between high and low ability students on the impact on
migration arising from working (instead of studying) in the home country being the alternative to
migration.

21
 Table 4: Lab decisions to migrate interacted with ability
(1)
Chose Migration

Prob. Graduation 40% (α1) 0.197***

(0.026)
Prob. Graduation 40% * High Ability (β1) 0.081**
(0.035)
Prob. Graduation 100% (α2) 0.333***
(0.039)
Prob. Graduation 100% * High Ability (β2) 0.100**
(0.043)
Holding Scholarship (α3) 0.276***
(0.024)
Holding Scholarship * High Ability (β3) 0.071**
(0.031)
Working in Home Country as Alternative to Migration (α4) 0.024**
(0.011)
Working in Home Country as Alternative to Migration * High Ability (β4) -0.031
(0.026)
Constant 0.323***
(0.021)
p-value (α1 + β1=0) 0.000
p-value (α2 + β2=0) 0.000
p-value (α3 + β3=0) 0.000
p-value (α4 + β4=0) 0.770
Observations 5,512
Adjusted R-squared 0.356
Notes: The dependent variable is a binary variable taking value 1 if the student chooses to
migrate. Omitted category has 0% graduation rate abroad, no scholarship available, and study in
home country if not migrating. LPM estimates include individual fixed effects. Standard errors
clustered at the class level in parentheses. Significance levels: *** p<0.01 ** p<0.05 * p<0.1.

3.4 Foreign bias

Following Batista and McKenzie (2023), we introduced an experimental game round

designed to capture “home bias”, i.e. the preference for staying in the home country as opposed to
migrating abroad when faced with the same outcomes at home and abroad. Contrary to Batista and
McKenzie (2023), we do not find a home bias, but rather a “foreign bias". Experimental subjects

22
are significantly more likely to choose to migrate than what indifference would imply: 59.9% [95%
CI: 54.5; 65.2] state a preference for migration in a round where equivalent financial conditions
are provided at home and abroad.
To investigate the motives underlying the foreign bias measured in our lab experiment, we
collected qualitative evidence from focus groups.18 72% of the participants reported that the motive
to study in Portugal was the better quality of college education (especially in the area of science
and technology) and better work opportunities, compared to low-prestige college degrees in Cape
Verde.
In terms of heterogeneity of migration choices in the lab, Table 5 shows that home biased
students (defined as those who decide not to migrate when faced with equal outcomes at home and
abroad) are significantly (at the 1% level) more responsive to scholarship provision, increasing
their choices to migrate by 15.5 p.p. more than foreign biased students when they are provided
with a full scholarship. They are, however, less responsive to changes in the college graduation
probability than foreign biased students. This is in line with the idea that students that choose to
stay in Cape Verde do so mainly due to financial costs of migrating, as mentioned in the focus
groups.

18
We organized three focus groups in three of the schools where the original lab experiment took place in
December 2023. There were 18 focus groups participants, who were 12th grade students in the academic year 2023/24.

23
 Table 5: Lab decisions to migrate interacted with home biased
(1)
Chose Migration

Prob. Graduation 40% (α1) 0.242***

(0.025)
Prob. Graduation 40% * Home Biased (β1) -0.054**
(0.022)
Prob. Graduation 100% (α2) 0.386***
(0.039)
Prob. Graduation 100% * Home Biased (β2) -0.060*
(0.034)
Holding Scholarship (α3) 0.235***
(0.025)
Holding Scholarship * Home Biased (β3) 0.155***
(0.024)
Working in Home Country as Alternative to Migration (α4) -0.005
(0.011)
Working in Home Country as Alternative to Migration * Home Biased (β4) 0.053*
(0.027)
Constant 0.321***
(0.022)
p-value (α1 + β1=0) 0.000
p-value (α2 + β2=0) 0.000
p-value (α3 + β3=0) 0.000
p-value (α4 + β4=0) 0.049
Observations 5,407
Adjusted R-squared 0.363
Notes: The dependent variable is a binary variable taking value 1 if the student chooses to
migrate. Omitted category has 0% graduation rate abroad, no scholarship available, and study
in home country if not migrating. LPM estimates include individual fixed effects. Standard
errors clustered at the class level in parentheses. Significance levels: *** p<0.01 ** p<0.05 *
p<0.1.

3.5 Do lab choices translate into real-world behavior?

It is a possible concern that the migration choices made in the context of our lab experiment
may not translate into real world migration decisions. To mitigate this concern, we make two points

24
about our setting: (1) migration choices in the lab are strongly correlated to the migration intentions
of experimental subjects; (2) migration intentions are a good predictor of real-world migration.
To support the proposition that migration choices in the lab reveal real world behavior, we
correlate the migration choice in each one of the lab rounds with a migration preparation index
adding up five binary variables corresponding to five different questions regarding steps taken to
prepare for migration to Portugal, as described in Section 2.3.
Table 6 reports the simple LPM regressions of the choice of migrating to Portugal in the lab
on the migration preparation index. We find a significant positive correlation between steps taken
to prepare for migration to Portugal and the probability of choosing to migrate in the lab. As shown
in Table 6, one additional step taken towards preparing migration by the experimental subjects is
associated with a 2 percentage points higher likelihood of choosing to migrate to Portugal in the
lab rounds.

Table 6. Correlation between migration steps and choices in the lab

(1)
Chose Migration
Sum migration steps 0.019**
(0.005)
Constant 0.657***
(0.008)
Observations 5,512
Notes: The dependent variable is a binary variable taking value 1 if the student
chooses to migrate. "Sum migration steps" is a variable corresponding to the
sum of five binary variables corresponding to different steps taken (or not) in
prepaprtion of migration. LPM estimates. Standard errors clustered at the class
level in parentheses. Significance levels: *** p<0.01 ** p<0.05 * p<0.1.

Having established a strong correlation between migration choices in the lab and migration
intentions at the individual level, as proxied by steps taken towards migration, we claim that
migration intentions can be taken as good predictors of actual migration behaviors. However, we
recognize that, using within-individual variation, revealed information affecting lab decisions
might not necessarily translate to changes in real‐world decision‐making. Nevertheless, extant
evidence supports a strong association between migration intentions and eventual migration
behavior at both the macro level (e.g. Tjaden et al, 2019, Docquier et al, 2014, Bertoli and Ruyssen,

25
2018), and micro level (e.g. Chort, 2014; Creighton, 2013; Van Dalen and Henkens, 2013). For
example, Tjaden et al. (2019) find that a 1 per cent increase in migration intentions in the Gallup
World Poll data is associated with a 0.8 per cent increase in actual bilateral migration, while Chort
(2014) finds that migration intentions in the 2002 wave of the Mexican Family Life Survey predict
subsequent moves collected in the second wave (2005–06).
An additional concern relates to the decision-making process at the household level. Even if
elicited migration preferences in the lab can measure migration intentions for individual students,
the relevant decision-making unit for college migration could be the household. Intra-household
bargaining on the migration decision can determine the actual migration outcome. Such concern
is particularly important in settings with children or teenagers, for which parents may play an
important role in decisions. Unfortunately, we did not collect data for our sample that would allow
us to directly test this particular hypothesis. However, we shed some light on this concern by using
additional data from a separate, larger survey of 4049 12th grade Cape Verdean students collected
in the same year as our survey experiment. More than half (58%) of the surveyed individuals did
not live with their father, who is an immigrant in Portugal in only 4% of the cases. Additionally,
among fathers who live in the same household as their children in the sample, only 11% have
completed secondary education, the same being true for 19% of mothers. This evidence depicts a
context in which 12th grade students in Cape Verde, who are on average 18 years old, most often
live in homes without their fathers to actively inform or support their decision to apply to college
abroad. In addition, parents are also overwhelmingly less educated than these students – and
presumably less qualified to acquire information about international college migration and to
contribute to informed decisions about the college education of their young adult children.
Therefore, in this setting, it seems reasonable that the young adults in our sample are the primary
decision-makers regarding their migration choices, as they are more educated and do not typically
have close parental support for their choices.
In light of this evidence, we argue that the migration choices made by experimental subjects
in the lab correlate well with real-world steps taken in preparation for migration, and that this is
likely a good predictor of actual future migration. We must, however, acknowledge that the actual
migration of individual students in our sample will also be the result of many other factors
mediating the relation between migration intentions and actual migration. These barriers include
information and support conveyed by individuals in their personal networks (such as parents,

26
teachers or peer students), financial migration costs or lack of access to student visas. The exact
contribution of these factors in the transformation of migration intentions into actual migration is
an interesting topic for future research.

4 Concluding Remarks

This paper examines the importance of information gaps for international study migration
decisions. Namely, it measures the role of information about college completion rates abroad and
about financial support availability on migration intentions. For this purpose, we implemented an
incentivized lab-in-the-field experiment with relatively large stakes. By combining this experiment
with individual level survey, our work allows us to examine determinants of international study
migration such as liquidity constraints and uncertainty, while assessing the heterogeneous impact
of individual characteristics.
Overall, our findings show a strong preference for migration, only partially explained by
information gaps of potential migrants - who overestimate both average college completion rates
abroad and also the availability of financial support for college studies abroad.
We find that information about the academic success of migrant students in the destination
country, as well as receiving a full scholarship covering all studying costs, strongly predicts willingness
to migrate. Female students are less responsive than males to increases in college graduation rates
abroad, but there are no differences in responsiveness to financial support across gender groups. High-
ability students are more sensitive to both certain college graduation rates and to holding a full
scholarship than lower ability students. Beyond gender and academic ability, we only observe marginal
differences across different levels of risk aversion, socioeconomic status or network at destination.
In terms of external validity, we believe our findings have some degree of portability to
international study migration corridors to Europe from countries with former colonial ties – this
applies, for example, to several African countries with students moving to France and the UK, or
Latin American countries to Spain. The results of our study should therefore be informative to
countries in these corridors whose immigrant populations face similar academic success
challenges.
Given that the baseline potential migrants have prior beliefs on availability of financial

27
support that are overoptimistic, it is likely that study migrants need to shift their time from study
to work after uninformed migration, which likely harms their scholar performance. Policies that
inform potential migrants of actual study funding possibilities and actual college graduation rates
may decrease study migration flows but are likely to improve successful graduation.

28
References
Abarcar, P. and C. Theoharides (2024). Medical Worker Migration and Origin-Country Human
Capital: Evidence from U.S. Visa Policy, Review of Economics and Statistics, 106 (1): 20–35.

Arenas, A. (2021). Human capital portability and international student migration, Journal of
Economic Geography, 21 (2): 195–229.

Bah, T. L. and C. Batista (2018). Understanding Willingness to Migrate Illegally: Evidence from a
Lab in the Field Experiment. NOVAFRICA Working Paper 1803, 1–38.

Bahar, D., and H. Rapoport (2019). Migration, Knowledge Diffusion and the Comparative
Advantage of Nations, Economic Journal, 128 (July): F273–F305.

Bahar, D., A. Hauptmann, C. Ozguzel, and H. Rapoport (2022). Migration and knowledge diffusion:
The effect of returning refugees on export performance in the former Yugoslavia. Review of
Economics and Statistics, 1-50.

Barsbai, T., H. Rapoport, A. Steinmayr, and C. Trebesch (2017). The Effect of Labor Migration on
the Diffusion of Democracy: Evidence from a Former Soviet Republic, American Economic
Journal: Applied Economics, 9 (3): 36-69.

Batista, C., D. Costa, P. Freitas, G. Lima, and A.B. Reis (2023). Mentoring the Success of Immigrant
College Students. Nova School of Business and Economics.

Batista, C., J. Gazeaud, and J. Seither (2022). Integrating Immigrants as a Tool for Broad
Development: Experimental Evidence for Portugal and Cape Verde, NOVAFRICA Working
Paper 2205.

Batista, C., A. Lacuesta, and P. C. Vicente (2012). Testing the ‘brain gain’ hypothesis: Micro
evidence from Cape Verde, Journal of Development Economics, 97(1): 32-45.

Batista, C., T. McIndoe-Calder, and P. Vicente (2017). “Return Migration, Self-Selection and
Entrepreneurship.” Oxford Bulletin of Economics and Statistics, 79 (5): 797-821.

Batista, C., and D. J. McKenzie (2023). Testing Classic Theories of Migration in the Lab. Journal
of International Economics, 145: 103826.

Batista, C., J Seither, and P. Vicente. (2019). Do migrant social networks shape political attitudes
and behavior at home? World Development, 117(C): 328-343.

29
Batista, C., D. Silverman, and D. Yang (2015). "Directed giving: Evidence from an inter-household
transfer experiment," Journal of Economic Behavior & Organization, 118(C): 2-21.

Batista, C., and P. C. Vicente (2011). Do Migrants Improve Governance at Home? Evidence from a
Voting Experiment, World Bank Economic Review, 25(1): 77-104.

Beine, M., M. Delogu, and L. Ragot, (2020) The role of fees in foreign education: evidence from
Italy, Journal of Economic Geography, 20: 571–600.

Beine, M., F. Docquier, and C. Ozden (2011). Diasporas, Journal of Development Economics, 95
(1): 30-41.

Beine, M., F. Docquier, and H. Rapoport (2008). “Brain Drain and Human Capital Formation in
Developing Countries: Winners and Losers.” Economic Journal, 118(528): 631-652.

Beine, M., R. Noel, and L. Ragot (2014) Determinants of the international mobility of students,
Economics of Education Review, 41: 40–54.

Bertoli, S., J. Fernandez-Huertas Moraga, F. Ortega (2013). Crossing the border: Self-selection,
earnings, and individual migration decisions. Journal of Development Economics, 101(1): 75–
91.

Bhagwati, J., Hamada, K. (1974). The brain drain, international integration of markets for
professionals and unemployment: a theoretical analysis. Journal of Development Economics,
1(1), 19-42.

Bursztyn, L., D. Yang (2022) Misperceptions about Others, Annual Review of Economics, 14, 425-
452.

Chand, S., and M. Clemens (2023). Human capital investment under exit options: Evidence from a
natural quasi-experiment. Journal of Development Economics, 163: 103112.
Chort, I. (2014). Mexican migrants to the US: What do unrealized migration intentions tell us
about gender inequalities?. World Development, 59, 535-552.
Creighton, M. J. (2013). The role of aspirations in domestic and international migration. The Social
Science Journal, 50(1), 79-88.

De Arcangelis, G., M. Joxhe, D. McKenzie, E. Tiongson, and D. Yang (2015). Directing remittances
to education with soft and hard commitments: Evidence from a lab-in-the-field experiment and
new product take-up among Filipino migrants in Rome. Journal of Economic Behavior &

30
 Organization, 111: 197-208.

Dinkelman, T., Mariotti, M. (2016). The long-run effects of labor migration on human capital
formation in communities of origin. American Economic Journal: Applied Economics, 8(4),
1-35.

Docquier, F., E. Lodigliani, H. Rapoport, and M. Schiff (2016). Emigration and Democracy. Journal
of Development Economics, 120: 209-223.

Docquier, F., and H. Rapoport (2012). Globalization, brain drain and development. Journal of
Economic Literature, 50 (3), 681-730.

Edwards, A., and M. Ureta (2003). International Migration, Remittances and Schooling: Evidence
from El Salvador. Journal of Development Economics, 75: 429-461.

Gould, D. (1994). Immigrant links to home country: empirical implications for U.S. bilateral trade
flows. Review of Economics and Statistics, 76: 302-316.

Grogger, J. and G. H. Hanson (2011). Income maximization and the selection and sorting of
international migrants. Journal of Development Economics, 95 (1), 42–57.

Grubel, H., and A. Scott (1966). The International Flow of Human Capital. American Economic
Review, 56(1/2), 268–274.

Harris, John, and Michael Todaro (1970). Migration, Unemployment, and Development: A Two-
sector Analysis. American Economic Review, 60(1): 126-42.

Javorcik, B., C. Ozden, M. Spatareanu, and C. Neagu (2011). Migrant Networks and Foreign Direct
Investment. Journal of Development Economics, 94: 231-241.

Khanna, G., E. Murathanoglu, C. Theoharides, and D. Yang (2022). Abundance From Abroad:
Migrant Income And Long-Run Economic Development. NBER Working Paper 29862.

Karadja, M., and E. Prawitz (2019). Exit, voice, and political change: Evidence from Swedish mass
migration to the United States. Journal of Political Economy, 127 (4): 1864-1925.

Kato, T., and C. Sparber (2013). Quotas and quality: the effect of H-1B visa restrictions on the pool
of prospective undergraduate students from abroad. Review of Economics and Statistics, 95:
109–126.

McKenzie, David, John Gibson and Steven Stillman (2010). “How Important is Selection?

31
 Experimental vs. Non-Experimental Measures of the Income Gains from Migration”, Journal
of the European Economic Association 8 (4): 913–45.

McKenzie, David, John Gibson and Steven Stillman (2013). “A land of milk and honey with streets
paved with gold: Do emigrants have over-optimistic expectations about incomes abroad?”,
Journal of Development Economics 102: 16-127.

Rosenzweig, M. (2008) “College and International Migration in Asia: Brain Circulation”. In Annual
World Bank Conference on Development Economics, Washington, DC: World Bank, pp. 59–
100.

Sjaastad, L. (1962) The Costs and Returns of Human Migration. Journal of Political Economy,
70(5): 80–93.

Spilimbergo, A. (2009). Democracy and Foreign Education. American Economic Review, 99 (1):
528-43.
Tjaden, J., Auer, D., & Laczko, F. (2019). Linking migration intentions with flows: Evidence and
potential use. International Migration, 57(1), 36-57.
Van Dalen H, Hendrik P, Henkens K (2013) Explaining emigration intentions and the behavior in
the Netherlands, 2005-10. Population Studies: A Journal of Demography, 67:225-241

32
Appendix

A1. General Prompt (Translated to English from original Portuguese version)

Suppose that you have received 1000 Euros (110 000 ECV) and that you now have to decide
what to do with that amount.

You have two alternatives:

1) You can keep the money for yourself and stay in Cape Verde; or
2) You can use that money to study in college in Portugal.

In this game we will present you with 13 different scenarios that you should consider carefully.
For each scenario, we ask you to answer if you would choose to stay in Cape Verde or to study for
college in Portugal.

One of the scenarios will be selected at random and, depending on your decision in that
scenario, you will receive a real amount in mobile phone credit. Therefore, answer sincerely and
carefully to each question.

33
A2. Game Showcards

Figure A1: Showcard with Cape Verde and Portugal detailed Scenario

Figure A2: Showcard with Cape Verde and Portugal choice options

34
A3. Rounds payoff

Table A1: Lab rounds description (parameters and payoff scheme) and Income-Maximizing Migration Choice
Probability of Net Payoff of % students
Probabilty of Alternative to No Income-Maximizing
Round College Migration to Portugal choose
Full Scholarship Migration Migration Choice
Graduation (in EUR) migrate
1 0% 0% Work in CV -20,572 NO 32.39%
2 0% 40% Work in CV -14,356 NO 53.91%
3 0% 100% Work in CV -5,032 NO 75.70%
4 100% 0% Work in CV -4,912 NO 61.10%
5 100% 40% Work in CV 1,304 YES 89.69%
6 100% 100% Work in CV 10,628 YES 95.65%
7 0% 0% College in CV -5,692 NO 30.43%
8 0% 40% College in CV 524 YES 51.11%
9 0% 100% College in CV 9,848 YES 72.17%
10 100% 0% College in CV 9,968 YES 67.60%
11 100% 40% College in CV 16,184 YES 85.59%
12 100% 100% College in CV 25,508 YES 92.61%

35
A4. Additional Estimation Results

Table A2: Lab decisions to migrate interacted with risk aversion

(1)
Chose Migration

Prob. Graduation 40% (α1) 0.228***

(0.025)
Prob. Graduation 40% * High Risk Aversion (β1) -0.015
(0.024)
Prob. Graduation 100% (α2) 0.382***
(0.032)
Prob. Graduation 100% * High Risk Aversion (β2) -0.038
(0.027)
Holding Scholarship (α3) 0.302***
(0.025)
Holding Scholarship * High Risk Aversion (β3) -0.007
(0.023)
Working in Home Country as Alternative to Migration (α4) 0.034*
(0.017)
Working in Home Country as Alternative to Migration * High Risk Aversion (β4) -0.036*
(0.020)
Constant 0.322***
(0.022)
p-value (α1 + β1=0) 0.000
p-value (α2 + β2=0) 0.000
p-value (α3 + β3=0) 0.000
p-value (α4 + β4=0) 0.861
Observations 5,347
Adjusted R-squared 0.356

Notes: The dependent variable is a binary variable taking value 1 if the student chooses to migrate.
Omitted category has 0% graduation rate abroad, no scholarship available, and study in home country if
not migrating. LPM estimates include individual fixed effects. Standard errors clustered at the class level in
parentheses. Significance levels: *** p<0.01 ** p<0.05 * p<0.1.

36
Table A3: Lab decisions to migrate interacted with close family abroad
(1)
Chose Migration

Prob. Graduation 40% (α1) 0.219***

(0.020)
Prob. Graduation 40% * Close Family Abroad (β1) 0.006
(0.034)
Prob. Graduation 100% (α2) 0.353***
(0.032)
Prob. Graduation 100% * Close Family Abroad (β2) 0.050
(0.046)
Holding Scholarship (α3) 0.298***
(0.023)
Holding Scholarship * Close Family Abroad (β3) -0.009
(0.031)
Working in Home Country as Alternative to Migration (α4) 0.016
(0.011)
Working in Home Country as Alternative to Migration * Close Family Abroad (β4) -0.001
(0.027)
Constant 0.323***
(0.023)
p-value (α1 + β1=0) 0.000
p-value (α2 + β2=0) 0.000
p-value (α3 + β3=0) 0.000
p-value (α4 + β4=0) 0.549
Observations 5,512
Adjusted R-squared 0.353

Notes: The dependent variable is a binary variable taking value 1 if the student chooses to migrate. Omitted
category has 0% graduation rate abroad, no scholarship available, and study in home country if not
migrating. LPM estimates include individual fixed effects. Standard errors clustered at the class level in
parentheses. Significance levels: *** p<0.01 ** p<0.05 * p<0.1.

37
Table A4: Lab decisions to migrate interacted with SES
(1)
Chose Migration

Prob. Graduation 40% (α1) 0.238***

(0.023)
Prob. Graduation 40% * Low SES (β1) -0.037
(0.026)
Prob. Graduation 100% (α2) 0.395***
(0.037)
Prob. Graduation 100% * Low SES (β2) -0.070*
(0.037)
Holding Scholarship (α3) 0.299***
(0.024)
Holding Scholarship * Low SES (β3) -0.005
(0.024)
Working in Home Country as Alternative to Migration (α4) 0.011
(0.015)
Working in Home Country as Alternative to Migration * Low SES (β4) 0.010
(0.020)
Constant 0.323***
(0.022)
p-value (α1 + β1=0) 0.000
p-value (α2 + β2=0) 0.000
p-value (α3 + β3=0) 0.000
p-value (α4 + β4=0) 0.134
Observations 5,512
Adjusted R-squared 0.354

Notes: The dependent variable is a binary variable taking value 1 if the student chooses to
migrate. Omitted category has 0% graduation rate abroad, no scholarship available, and study
in home country if not migrating. LPM estimates include individual fixed effects. Standard
errors clustered at the class level in parentheses. Significance levels: *** p<0.01 ** p<0.05 *
p<0.1.

38