Three Essays on Economic Development

A dissertation presented by Paula Luciana Méndez Errico to

The Department of Applied Economics
in partial fullfiment of the requirements for the degree of
Doctor of Philosophy in the subject of Economics
Universidad Autónoma de Barcelona

Director Director
Xavier Ramos Morilla Roxana Gutiérrez Romero

October, 2014
Aknowledgements

This thesis would not be possible without the immense support received during these
years. First, I want to thank to my advisors Xavier Ramos and Roxana Gutiérrez, who
kindly accepted me as a PhD student, and for sharing with me their knowledge and
experience, for their valuable suggestions and encouragement during this research.

In a personal dimension, my profound gratitude goes to Matías, with whom we shared

this adventure and for always encouraging me during this period. Also, to my parents
for their unconditional love, and for taking care of all those issues I could not afford
because of the distance. This thesis in part reflects their transmitted sense of what
should be fair and their concern on equity. I also want to mention to my brothers
Sebastián and Federico for their love and support, and to Fiorella for her great advices.
To my grandmother and my niece Sofia, my motives for always want to come back
home. To my grandfather, Negra and Beba for taking care of me wherever they are.
Also, I am extremely grateful to Gabriela and Luis for being unconditionally close to
me.

From this side of the ocean, to my cousin Damiana and my niece Olivia, my family and
friends in Barcelona, my strongest support in good and not so good times. To my family
in Italy, who warmly received me any time I visited them making me feeling at home.
During these six years I met great friends who became my ‘chosen’ family in
Barcelona. Special thanks go to Roberto and Ana, Dolores, Paula, Adriana, Lina,
Cristina, Jorge, Orlando and Areli for all the great moments shared. Also, I want to
specially thank to Mónica, with whom we share not only similar research interests and
rich discussions, but also a great friendship. I am also grateful to Camilo for teaching
me facing bad times with a great sense of humor. To Mamen and Marina for making me
feeling at home, and to Fausto for the saturday’s vermouths in Gràcia. To Macarena,
Natalia, Victoria, Vicky Prieto, Paula, Gonzalo and Martín, my Uruguayan friends in
Barcelona, for the mates and with whom we shared our homesickness. I would also
thank to Roger, Miquel and Sonia for sharing their culture and uses with me. To
Rodrigo and Nery, for the lovely times we spent together in Barcelona and for kindly
receiving me in Mexico.
I also would like to thank to Pilar, Monste, Imma and Miquel from the secretary of the
Applied Economics Department of the UAB who always helped me with the
bureaucracy. Special thanks go to Rosella Nicollini, for always taking care of the PhD
students at the Department. Also, to Cristina López Mayan who was my evaluator
during my PhD, for being always available to evacuate my doubts, for her insightful
comments and advices during all this period.

Finally, special thanks go to the Spanish Ministry of Science and Innovation (MICINN)
with reference project ECO2010-21668-C03-02 and reference scholarship BES-2011-
048083 for the financial support. Specific acknowledgements are shown at the end of
each essay.
Index

Introduction ...................................................................................................................... 1
1. The Impact of Social Networks on Immigrants’ Employment Prospects: The Spanish
Case 1997-2007 .............................................................................................................. 11
1.1 Introduction ...................................................................................................... 12
1.2 Data and descriptive analysis ........................................................................... 15
1.3 Methodology .................................................................................................... 19
1.3.1 Job match and social networks ................................................................. 19
1.3.2. Social networks and wages ....................................................................... 23
1.4 Empirical findings ............................................................................................ 28
1.4.1 Job match and social networks ................................................................. 28
1.4.2 Social networks and wages ....................................................................... 32
1.5 Conclusion ....................................................................................................... 37
References ................................................................................................................... 39
Tables and figures ....................................................................................................... 44
Appendix ..................................................................................................................... 53
Methodological Appendix .......................................................................................... 70
2. The Long-Term Effect of Inequality on Entrepreneurship and Job Creation............. 77
2.1 Introduction ...................................................................................................... 78
2.2 Institutions and Initial Conditions and Entrepreneurship ................................ 80
2.2.1 Banerjee and Newman’s Occupational Choice Model ............................. 82
2.2.2 Endogeneity between Credit Regulation and Entrepreneurship ............... 84
2.3 Data and Methodology..................................................................................... 86
2.3.1 Historical Income Distribution and Current Credit Regulation................ 86
2.3.2 GEM Survey ............................................................................................. 87
2.3.3 Pseudo-Panel ............................................................................................ 88
2.4. Econometric Results ..................................................................................... 90
2.4.1. Firm’s Life Cycle: Birth, Maturity and Death .......................................... 90
2.4.2 Job Creation: Firms’ Size ......................................................................... 93
2.5. Robustness Checks .......................................................................................... 96
2.6. Conclusion ....................................................................................................... 97
References ................................................................................................................... 99
 Tables and figures ..................................................................................................... 103
Appendix ................................................................................................................... 106
3. Schooling progression in Uruguay: Why some children are left behind? ................ 127
3.1 Introduction .................................................................................................... 128
3.2 Education inequality, cognitive and non-cognitive abilities .......................... 131
3.3. The Uruguayan Educational System .............................................................. 136
3.4. Data and descriptive statistics ........................................................................ 137
3.5. Methodological framework............................................................................ 140
3.5.1 A sequential model of schooling progression......................................... 141
3.5.2 Empirical strategy ................................................................................... 142
3.6 Results ............................................................................................................ 147
3.6.1 Unobserved heterogeneity and correlations ........................................... 148
3.6.2 Empirical findings .................................................................................. 149
3.6.3 Interpretation of results ........................................................................... 154
3.7 Conclusion ..................................................................................................... 158
References ................................................................................................................. 161
Tables and figures ..................................................................................................... 165
Appendix ................................................................................................................... 171
4. Conclusions ........................................................................................................... 181
Introduction

Over the last decades there has been a resurgence of interest in economic research on
economic development. The great income differences observed between as well as
within countries has turned economic science’s attention to explain why countries differ
in their economic growth, and why within countries some people may be entrapped in
poverty.
The recognition that income inequality and economic status perpetuates over
time not only in poorer countries but also in wealthier societies, and the associated costs
in different aspects of individual’s and social well-being, such as happiness, health,
education, crime, violence, corruption, among others (Wilkinson and Picket, 2011), lead
to the development of new and insightful theories in economics. This literature on the
effect of income inequality and economic growth suggests alternative mechanisms that
could cause poverty to persist, addressing both the question of how whole economies
may fail to develop, and how population subgroups within rich economies may fail to
share in overall prosperity.
I broadly identify three set of theories that explain dispersion in income across
individuals and social groups and divergence in economic growth across countries; such
as those based on (i) individual characteristics, (ii) institutional factors, and (iii) social
interactions. Although individual, social interactions and institutional factors are
interdependent, alternative explanations of poverty have different implications, both in
terms of understanding the sources of poverty and inequality as well as in terms of the
design of public policies (Durlauf, 2006).
The main objective of this dissertation is to study some of the mechanisms
suggested by the literature as factors that could prevent individuals from attaining
certain domains of well-being.
Specifically, this thesis is divided in three independent essays providing new
evidence on three issues within the field of economic development: the effect of social
networks on immigrants’ labor market outcomes (first essay), the long-lasting impact of
income inequality on entrepreneurial success and job creation (second essay), and the
importance of multiple abilities, parental educational background and race in explaining
educational gaps (third essay). Also, different cases of study are provided: immigration
issues in a developed country such as Spain, initial conditions for a broad set of

1
countries with different levels of economic development, and education in a middle-
income country such as Uruguay. Finally, different databases and econometric
techniques are properly selected to address each case of study. I explain in further detail
the goal and findings of these three essays next.

The first essay “The impact of social networks on immigrants’ employment prospects:
the Spanish case 1997-2007” analyzes the factors that could prevent or foster
immigrants’ social and economic integration in the host country. Specifically, this essay
contributes to the empirical literature on immigration and social networks by studying
the extent to which social networks affect labor market outcomes -job match and wages-
for immigrants living in Spain. To this end, I first study the impact of social networks
on the job matching process by studying the probability of keeping the first job in Spain
relative to not keeping it; namely, changing jobs, being unemployed or inactive.
Secondly, for those immigrants actually employed in the same job since arrival, we
analyze the effect of social networks on wage.
Labor market participation and conditions in terms of employment and wage, is
one of the main immigrant’s integration channel to the host country, and also an
important source of immigrant’s income. In turn, social networks have been recognized
in the literature as an important channel through which information is transmitted,
especially relevant for immigrants in the host country as it provides -among others-
information on labor market institutions and job opportunities (Calvó-Armengol and
Jackson, 2004 and 2005). But also, social networks could prevent immigrants’ to
integrate in the host country, since widespread reliance on social networks in the labor
market can lead to social stratification by limiting individuals’ opportunities to those
that their peer group can provide (Mouw, 2009). The persistent segregation of
immigrants in the labor market may affect future prospects of their offspring, leading to
the extreme case of economic immobility in which immigrants are entrapped into
poverty.
Despite the growing literature on social networks and immigrants’ labor market
outcomes, no conclusive effects of social networks on immigrants’ workers have been
found yet (Ioannides and Loury, 2004). By focusing on the effects of social networks on
immigrants’ labor market outcomes, this study contributes to the empirical literature by
addressing a less explored channel through which immigrants’ social and economic
integration could be affected.

2
 To empirically analyze the effect of social networks on job match and wages, I
use data from the National Immigrant Survey conducted in 2007. In this study two
measures of social networks are considered: the strength of the network (close and weak
ties); and the size of the network proxied by the proportion of immigrants from the same
country of origin living in the same region (Autonomous Community) on the total
immigrant population in the region of destination. It is also considered the alternative
mechanisms of job access: relatives or friends (network jobs) and formal methods (such
as public and private employment agencies, newspaper advertisements, among others).
Endogeneity issues are likely to emerge in this study, because a selection process
of immigrants in labor market statuses may take place, and because social network
formation is likely to take place among individuals with particular traits. Therefore, a
two-step procedure is applied, first for analyzing job match, and then for wage quantile
regression estimations.
Also, as individuals are more likely to socially interact if they share some
individual traits as being sociable and responsible, education or occupation, an
extensive set of exogenous variables like occupation and sector of activity in the
country of origin is included.
The findings suggest that social networks are likely to help immigrants to find a
job in the short-run, but may limit opportunities to fully integrate in the longer term. In
this sense, these findings shed light on the importance of social networks preventing
immigrants’ integration, as well as help to orientate the design of integration policies for
immigrants living in Spain.

The second essay, entitled “The Long-Term Effect of Inequality on Entrepreneurship

and Job Creation” studies the extent to which initial conditions understood as income
inequality in 1700s and 1800s, and credit market institutions, can condition
entrepreneurship and job creation to flourish over time.
This essay adds to the literature on the long-lasting effects of income inequality
on economic development by empirically testing the predictions of the model by
Banerjee and Newman (1993). This model predicts that initial conditions understood as
historical distribution of wealth, can have a long and persistent effect on development.
Specifically the model assumes that people can become either entrepreneurs or workers.
Since entrepreneurial activities require an up-front investment, they are available only to
wealthy people and to those individuals who can provide collateral to access credit.

3
Poorer and credit constrained individuals can only choose to work for a wage or to be
self-employed. Then, occupational choice will in turn give rise to a new distribution of
income by determining the returns and allocation of occupations, affecting the process
of economic development through, for instance, its effects on saving, investment, risk
bearing, and the composition of demand and production. Therefore, countries with
initially low income inequality would grow over time aided by a strong entrepreneurial
sector. A contrasting equilibrium could be reached if a country starts with a high ratio of
poor to wealthy people. In this case development runs out of steam.
Two hypotheses are derived from the model: 1) countries that have a historical
high ratio of wealthy to poor people have a lower probability of firms being created,
surviving, and of these creating jobs over time, and 2) countries that currently have
more efficient credit markets have a higher probability of people being involved in
entrepreneurship and of higher job creation.
To test the predictions of this model, a pseudo-panel of entrepreneurs across 48
countries over 2001-2009 is built using the Global Entrepreneurship Survey, and is
complemented with historical indicators of income distribution prevailing in 1700 and
1800 and current business environment, conditions that can affect the probability of
firms being created, surviving and creating jobs over time.
The methodology combines pseudo-panel techniques with instrumental
variables, given that current business environment could be affected by the proportion
of people involved in entrepreneurial activities, for instance by lobbying for certain
laws.
The findings of this essay give empirical support to the predictions of the model,
showing that historical income inequality and current credit market imperfections
prevent firms to be created and surviving over time, at the time that affect job creation
over time.
To the best of our knowledge, this article is the first one that tests the long-term
effects of inequality on occupational choice, thus giving empirical evidence on a less
studied channel through which income inequality can affect long-term development.

The third essay, entitled “Schooling progression in Uruguay: why some children are
left behind?” studies the impact of parental traits on children’s educational attainment
in Uruguay. Specifically, I analyze whether long-term parental background, crystallized
by parental educational background, race, cognitive and non-cognitive abilities, and

4
short-term family income measured by the opportunity cost of education, affect child’
schooling progression, and at what stage of the educational path they take on their
importance.
This study is motivated by the recent literature stressing the effects of multiple
abilities on persistent economic status and education inequality developed by Bowles
and Gintis (2001, 2002) and by Heckman and co-authors (Heckman et al., 2011;
Heckman and Mosso, 2014; Heckman et al., 2006). In addition, the scarcity of this type
of analysis found for less developed countries and the particularities of the Uruguayan
educational system encourages choosing Uruguay as an interesting case of study.
The empirical methodology considers a sequential probability model proposed
by Cameron and Heckman (1998, 2001), in which education attainment is the outcome
of the individual’s previous schooling decisions. Two main advantages are found in this
methodology. First, it recognizes the selection taking place across schooling, in which
more able and motivated individuals and those with better parental educational
backgrounds are more likely to attain higher levels of education. Second, it allows
identifying a direct effect of the key variables of the study on each schooling stage, and
also an indirect effect of these variables by affecting previous schooling decisions. This
analysis requires valid exclusion restrictions, thus I considered labor market conditions
at the time schooling decisions are made.
The dataset used in this study is the National Youth Survey which enables me to
construct individual’s educational path and performance, and to exploit information on
motivation and risky behavior to proxy socio-emotional endowments, as recognized by
earlier studies (Gullone and Moore, 2000; Heckman et al.,2006; Heckman e al., 2014).
The results show that parental educational background, cognitive and non-
cognitive abilities have effects of diverse magnitude across stages of the educational
path. Long-term parental background has increasing effect over the children’s schooling
progression in comparison to short-term parental income as it decreases its significance
when students progress to higher schooling stages. Specifically, cognitive ability has
increasing effects on the students’ likelihood of dropping out across the educational
path. Motivation and risky behavior measuring non-cognitive ability also influence
children’s schooling completion at early stages of education. This article finds that
despite the great supply of public education, children are being left out. The reasons, we
found, are initial conditions, understood as family background. Thus, with important
policy recommendations.

5
References

Banerjee, A., and A.F. Newman (1993). “Occupational Choice and the Process
of Development.” Journal of Political Economy, 101 (2): 363-394.
Bowles, S. and Gintis, H. (2001) “Schooling in Capitalist America Revisited”,
Sociology of Education 75(1):1-18.
Bowles, S. and Gintis, H. (2002) “The Inheritance of Inequality”, Journal of
Economic Perspectives 16 (3):3-30.
Calvó-Armengol, A., and Jackson. M. (2004). “The effects of social networks on
employment and inequality”, American Economic Review 94(3): 426-454.
Calvó-Armengol, A., and Jackson. M. (2005). “Job matching and word-of-
mouth communication”, Journal of Urban Economics 57: 500-522.
Cameron, S., and Heckman, J. (1998) “Life Cycle Schooling and Dynamic
Selection Bias: Models and Evidence for Five Cohorts of American Males”, Journal of
Political Economy 106 (2):262-333.
Cameron, S., and Heckman, J. (2001) “The dynamics of educational attainment
for black, Hyspanic and white males”, Journal of Political Economy 109 (3), 455-99.
American Economic Review, 92(4): 727–744.
Durlauf, S. (2006) “Groups, social influences and inequality”, (in) “Poverty
Traps”, (ed) Bowles, S., Durlauf, S., and Hoff, K., Princeton University Press
Gullone, E., and Moore, S. (2000) “Adolescent risky-taking and the five-factor
model of personality”, Journal of Adolescence 23:393-407.
Heckman, J. (1979). “Sample Selection Bias as a Specification Error”,
Econometrica, 47(1): 153-161.
Heckman, J., Humphries, J., Veramendi, G; and Urzúa, S. (2011) “The Effects
of Educational Choices on Labor Market, Health and Social Outcomes”, University of
Chicago WP No. 2011-002.
Heckman, J., Humphries, J., Veramendi, G; and Urzúa, S. (2014) “Education,
Health and Wages”, IZA DP No. 8027.
Heckman, J., and Mosso, S. (2014) “The Economics of Human Development
and Social Mobility”, IZA DP No. 8000.
Heckman, J.; Stixrud, J.; and Urzúa, S. (2006) “The Effects of Cognitive and
Noncognitive abilities on Labor Market Outcomes and Social Behaviour”, NBER WP
No. 12006.

6
 Ioannides, Y., and Loury, L. (2004). “Job Information Networks, Neighborhood
Effects and Inequality”, Journal of Economic Literature, 42(4): 1056-1093.
Mouw, T. (2009). “The Use of Social Networks Among Hispanic Workers: An
Indirect Test of the Effects of Social Capital”, University of North Carolina Press,
Chapel Hill.
Wilkinson, R., and Pickett, K. (2011) “The Spirit level: why greater equality
makes society stronger”, Bloomsbury Press.

7
8
 Essay 1
The impact of social networks on immigrants’
employment prospects: the Spanish case 1997-2007

9
10
 The Impact of Social Networks on Immigrants’ Employment Prospects: The
*
Spanish Case 1997-2007

Abstract
This paper studies the extent to which social networks influence the employment
stability and wages of immigrants in Spain. By doing so, we consider an aspect that has
not been previously addressed in the empirical literature, namely the connection
between immigrants’ social networks and labor market outcomes in Spain. For this
purpose, we use micro-data from the National Immigrant Survey carried out in 2007.
The analysis is conducted in two stages. First, the impact of social networks on the
probability of keeping the first job obtained in Spain is studied through a multinomial
logit regression. Second, quantile regressions are used to estimate a wage equation. The
empirical results suggest that once the endogeneity problem has been accounted for,
immigrants’ social networks influence their labor market outcomes. On arrival,
immigrants experience a mismatch in the labor market. In addition, different effects of
social networks on wages by gender and wage distribution are found.

*
This essay has been co-written with Xavier Ramos (Departament d’Economia Aplicada – Universitat
Autònima de Barcelona).

11
 1.1 Introduction
The immigrant population in Spain has largely increased over the past decade, from
2.3% of the total population in 2000 to 10% in 2007. This large immigration inflow has
turned Spain into the second largest recipient of immigrants after Germany in the
European context (OECD, 2010). The social relevance of this new phenomenon has
turned the immigration process into a key subject of social and economic research.
Different studies have focused on the assimilation process and occupational mobility of
immigrants in Spain (Izquierdo et al., 2009; Alcobendas and Rodríguez-Planas, 2009,
Simón et al., 2011; among others). However, less attention has been paid to the role of
social networks on immigrants’ labor market outcomes.
Empirical and theoretical studies point out the influence of social networks in
various areas of social and individual behavior, such as labor market performance,
education attainment, and crime among others (Jackson, 2008; Wahba and Zenou,
2005). For immigrant workers, social networks may accelerate the job finding process.
For instance, employers within an enclave may prefer to hire workers from their own
country (Borjas, 2000). However, belonging to an enclave, may in turn affect the quality
of the job offers an immigrant receive, as it influences the speed at which the immigrant
learn the skills of the host country (such as language). Therefore, strong dependence on
the social network may isolate immigrants from the native population and from the
organizations and institutions in the host country. In the long run, immigrants’ enclaves
may develop, reflecting social and economic disintegration.
In this paper, the focus is on the effects of social networks on the job quality an
immigrant finds, mainly because social and economic integration largely depends on an
immigrant’s labor market outcomes. The objective of this paper is to analyze to what
extent social networks affect immigrants’ labor market outcomes in terms of
employment stability and wages in Spain.
Theoretical literature agrees on the positive impacts of strong and weak ties on
the rate at which jobseekers receive employment offers.1 Moreover, the quality of the
members of the network influence the quality of the job an individual can find (Calvó-
Armengol and Jackson, 2004). Several empirical studies show that individuals’
probability to find a job increases with the individual social networks. For instance,

1
Close or strong ties refer to the strength of the network. Close ties include family and friends, while
weak ties are expressed in terms of a lack of overlapping in personal networks between any two agents
(e.g. professional acquaintances).

12
Munshi (2003) finds that Mexican migrants in the U.S who obtained a job through
social networks improve their labor market outcomes. Wahba and Zenou (2005) show
that, conditional on being employed, individuals’ probability to find a job through social
networks relative to formal search mechanisms, increases and it is concave with the size
of the networks. In addition, they stress that this effect is bigger for the less educated
workers. Patacchini and Zenou (2008) find that individuals’ probability of being
employed increases with the size of close and weak ties.
However, despite the growing empirical literature, no consensus of the impacts
of social networks on job quality has yet been reached (Ioannides and Loury, 2004).
Dustmann et al. (2010) show that through referrals, social networks reduce
informational deficiencies in the labor market, leading to better quality matches between
workers and firms. Some authors argue that immigrants with social resources obtain
more advantageous occupational positions, as friends and relatives sort through jobs to
reserve the better ones for their network’s members (Aguilera and Massey, 2003; Nee
and Sanders, 2001). Conversely, Bentolila et al. (2010) find that worker/job matches
tend to be poorer for jobs found through the network. In a similar line, Ottaviano and
Peri (2006) point out that job matches depend on the strength of the network. They
argue that mismatch happens if social networks are based on close ties because relatives
and friends are unrelated to the individual’s previous experience or training. Instead,
good matches can happen if job information is transmitted through professional
affiliations.
This paper aims to contribute to the empirical literature on the impact of social
networks on job quality, through studying the relationship between social networks and
job match on one hand, and the effects of social networks on wages on the other. Little
is known about the mechanisms through which social networks affect immigrants’ labor
market outcomes in Spain. We intend to provide empirical evidence of the mechanisms
through which social networks affect immigrants’ employment outcomes and thus,
contribute to the vast empirical literature on the assimilation process of immigrants in
Spain. Unlike previous studies, in this paper the focus is on the role of social networks
on immigrants’ employment outcomes, an issue not addressed before for the Spanish
case.
In contrast to other studies, we do not rely on the identification assumption that
individuals within a given group (such as ethnic group, neighborhood or firm) actually
know each other and are members of the same network. Most empirical studies of the

13
effect of social networks on immigrants’ labor market outcomes focus on indirect
measures of social interactions such as the number of other immigrant’s own country
(Munshi,2003); geographical proximity or group affiliation (e.g. Topa, 2001;Weinberg
et al., 2004; Bayer et al., 2008; Dustmann et al., 2010). The dataset used in this study,
the National Immigrant Survey (ENI, its Spanish acronym) allows us to use direct
information on social interactions provided by the immigrant such as having relatives
and friends on arrival to Spain, social participation in organizations and the job access
mechanisms used to obtain the first job in Spain.2 In addition, the richness of the ENI,
with retrospective information on individuals’ labor market characteristics and histories,
enables us to address the potential unobserved endogeneity problem controlling for
labor status and last occupation in the country of origin.
First, we study the impact of social networks on the job matching process
through studying the probability of keeping the first job relative to not keeping it;
namely, changing jobs, being unemployed or inactive. As the individuals considered in
this analysis are those with some labor experience in Spain, we estimate the
multinomial regression controlling for sample selection. Then, the effects of social
networks on wages are estimated for immigrants who keep their first jobs. We estimate
a wage equation, separately for women and men, through ordinary least squares (OLS)
and quantile regressions (QRs). We exploit a novel methodology for the study of social
network effects on wages through QRs controlling for sample selection bias. These
effects are estimated in a semi-parametric fashion using a two-step procedure similar to
that suggested by Heckman (1979).
Our results show that social networks have significant effects on the job
matching process for immigrant workers and wages. A job mismatch is observed for
immigrants upon arrival, they prefer to quickly accept a job offered through the social
network, even if it is not the most suitable given their human capital endowments. In
addition, we find positive effects of network size on job match, possibly reflecting the
existence of ethnic niches in the labor market. Finally, social networks differently
impact the wage distribution for women and men. The strength of the network (close or
weak ties) only affects men’ wages but does not affect women’s wage. Wage penalties

2
Cappelari and Tatsiramos (2010) and and Goel and Lang (2011 and 2012) also uses direct information
on social interactions in their studies of the effect of social networks on employment outcomes.
Cappellari and Tatsiramos (2010) construct a measure of the quality of the worker network based on each
respondent’s three best friends and their characteristics using the British Household Panel Survey. Goel
and Lang (2011 and 2012) use immigrants’ contacts at arrival obtained from the Longitudinal Survey of
Immigrants to Canada (LSIC).

14
are observed for both women and men who obtained the job through social networks.
This effect varies across the wage distribution between women and men. The network
size also penalizes both women’s and men’s wages.
The remainder of this paper is organized as follows. The next section describes
the data and provides summary statistics for the key variables of interest. Section 3
introduces the empirical strategy. Section 4 presents the results of the analysis. Finally,
the last section concludes.

1.2 Data and descriptive analysis

This study uses the National Immigrants’ Survey (ENI, its Spanish acronym), a single
and unique cross-sectional national representative survey on immigration conducted so
far only for 2007 by the National Statistics Institute (Instituto Nacional de Estadística).
The sample is based on the Municipal Register. In total, the original survey comprises
15,441 individuals.3 The ENI offers information on socio-demographic characteristics,
migration experience, social networks, and labor market experience.4 In particular, it
features detailed information on activity condition before migration and at the moment
of the survey and retrospective information on employment (e.g. occupation and activity
sector) at three times: in the country of origin, first job on arrival, and current job in
Spain. In addition, information on the finding methods used for the first job (social
networks or formal methods) and personal income (net amount of money employees
receive), among others, is provided.
The original sample is restricted to immigrants that arrived in Spain after 1996.
This constraint prevents selection bias in the analysis for different reasons. As Borjas
(1985, 1995) states, cross-sectional estimates of immigrant performance in the host
country could induce selection bias due to “cohort effects”, namely changes in the
composition or “quality” of immigrants arriving at different points in time or because of
nonrandom return migration or migration to a third country. In addition, the business
cycle could affect the results of labor market entrants and bias the estimation (Aslünd

3
A response rate with respect to the effective sample eligible respondents of 87.4% was obtained.
Interviews were conducted face-to-face, and for those informants unable to fill out the questionnaire in
Spanish, a telephone line was set up (in Arabic and English).
4
More detailed information on the design and contents of the ENI can be found at
http://www.ine.es/daco/daco42/inmigrantes/inmigra_meto.pdf.

15
and Rooth, 2007).5 Considering the period between 1997 and 2007 minimizes these
effects. Simón et al. (2011) also stress that during this period immigrant flows into
Spain were relatively homogeneous in relation to their regions of origin. Further, the
authors point out that the economic growth and strong job creation observed in this
period reduce the effects of the economic cycle on immigrants’ labor market situations
and the importance of return migration relative to economic downturns.
This analysis considers immigrants between 16 and 64 years old at the time of
the survey, and older than 16 and less than 57 years at the time of arrival. This selection
excludes immigrants who finished their studies in Spain, focuses only on those who
emigrated directly from their countries of birth to Spain. This leads to a final sample of
7,377 observations (8,064 observations were dropped) of which 945 individuals never
worked in Spain. After excluding those individuals who have never worked, we have a
subsample of 6,432 observations. Tables A.1 and A.2 in the Appendix detail the sample
selection and provides in-depth definitions of the variables used in this study,
respectively.
Table 1 presents summary statistics for the final sample, the subsample and the
excluded sample. For the final sample, most immigrants come from Latin America
(49%) followed by immigrants from Eastern Europe (25%), are on average 34 years old,
and have around four years of residence in Spain. In terms of educational attainment,
more than half of immigrants have at least secondary level, while approximately a
quarter of the sample reports tertiary education level. In addition, more than 75%
declares proficiency in Spanish language, and having legal residence authorization.
In order to capture the strength of the social network, two dummy variables are
created. Close ties is a dummy variable equal to one if the individual declares having
had at least one relative or friend on arrival to Spain. Weak ties are captured through
individual’s social participation in organizations. Two dummy variables are created in
order to distinguish between individuals participating in organizations devoted
exclusively to immigrants (non-mixed organizations) and those not (mixed
organizations). More than 80% of the immigrants declare having contacts at arrival
while social participation in organizations is, on average, low. Individuals participating

5
The literature addresses this issue through creating synthetic cohort of immigrants by tracking specific
immigrant waves across decennial Censuses or across Current Population Surveys (Borjas, 1994). In the
present study, the approach considered is analogous, since the ENI is a single cross-sectional database
with a 10-year period of analysis.

16
in mixed organizations represent 10% of the total sample, while 6% of the individuals
are involved in non-mixed organizations.
Columns 2 and 3 of Table 1 present summary statistics for the subsample and
excluded observations respectively. The comparison between different samples provides
a first insight of the potential sample selection bias that could happen when excluding
individuals who have never worked in the Spanish labor market. Main differences are
observed in terms of gender composition (79% are women in the excluded sample), age
(32 versus 34 years old), region of origin (30% of excluded individuals come from
North Africa) and years living in Spain (2 versus 4 years). Also, the proportion of
immigrants with proficiency in Spanish language and those with legal residence
authorization varies across different samples. In addition, differences between the
samples are observed in terms of internal mobility across municipalities (grouped as
never moved, moved once, or more than once) and in the declared motives for
migration. For instance, family regrouping motives is a dummy variable equal to one if
the immigrant declares family reunion as a motive for migration. Labor motive is a
dummy variable which refers to individual declaring job searching or looking for a
better job.6 Almost 60% of individuals in the excluded sample declare family
regrouping motives for migration, in comparison with less than 30% for the final and
sub-samples. Finally, in terms of social network variables, no differences are observed
across the different samples.
Table 2 presents the summary statistics for immigrants’ who have at least
worked once in Spain (80% of the final sample). More than 70% of them obtained their
first jobs through social networks while 30% of them got the job through formal
channels.78 Throughout this text, ‘network jobs’ and having obtained the first job

6
The ENI contains self-reported information on the reason for migration, namely due to the presence of a
family member or labor motives. As the question in the ENI allows for multiple responses, regrouping
motives considers those immigrants that declare family reunion as a motive for immigration, although
they could state another motive for migration. Labor motives is a dummy variable that is equal to one if
the immigrant declares job searching or looking for a better job as a motive for migration. Further,
migration motives were interacted with the region of origin and gender variables in the first equation and
did not change the final estimations obtained.
7
The mechanisms considered are formal methods and social networks. The translated question of the ENI
(2007) reads: By what means did you obtain your first job? Respondents can choose many options. If the
immigrant only chooses one channel, that is, getting the job through family, friends, or other contacts,
then we consider that the immigrant obtained the first job through social networks. Otherwise, it is
considered as getting the job through formal channels. In this sense, formal sources of information
include State and private employment agencies, newspaper advertisements, union hiring halls and school
and college placement services.
8
Following Goel and Lang (2011) two issues need to be noted. First, finding a job through the social
network does not necessarily imply the presence of a close tie (relative or friend on arrival). This is

17
through social networks are used interchangeably, as are ‘formal jobs’ and having
obtained the first job through formal channels. Approximately 31% remain in their first
jobs, more than 50% have changed jobs, almost 10% are unemployed, while 7% are
inactive. About half of these workers were first employed in non-skilled occupations
and a quarter in administrative jobs. The main activities in which immigrants are
involved in the first job are household activities, construction, and agriculture. In order
to explore if differences in observable characteristics exist between immigrants with
some labor experience in Spain, Panel A in Table 3 expose summary statistics for those
with and without close (columns 1 and 2); and between those with weak ties (columns 3
and 4). A priori, only slightly differences are observed. Immigrants with close ties are
on average more women than men, married, and mainly from Latin America.
Conversely, the proportion of immigrants with legal residence authorization is higher
for immigrants without close ties than for those with close ties. In terms of education
and last occupation in the country of origin, no differences are observed between those
with and without close ties. However, the proportion of those with close ties and
proficiency in the Spanish language is higher than for those without close ties. This is
also observed when analyzing immigrants with and without weak ties. Also, those with
weak ties are on average more educated. Finally, regional disparities are observed in
terms of gender composition, educational attainment, social network endowment and
occupational mobility (Tables A.3 and A.4 in Appendix, respectively). It is worth
noting that despite the low participation of immigrants in mixed organizations, the
proportion of those from Western Europe is three times that for North Africa. In
addition, immigrants from Asia and the rest of the world more than double the sample
mean of immigrants involved in non-mixed organizations.
Regarding the occupational mobility of immigrants, it is worth noting that
workers from Western Europe experience less downward mobility relative to
immigrants from other regions (Table A.3 in Appendix), thus reflecting the limited
transferability of human capital between non-Western European countries and the
Spanish labor market (Simón et al., 2011).

because immigrants may have found their job through a friend made after migrating to Spain, or a relative
or friend not living in Spain. Thus, having or not obtained the job through social networks does not imply
having or not close ties or vice versa. In addition, in contrast with other studies, we measure network use
directly, and therefore, we avoid the need to infer network use from clustering of immigrants.

18
 1.3 Methodology

This section presents the empirical approach and identification strategy. The analysis is
conducted in two steps. First, we study to what extent social networks affect the job
matching process (Section III.1). Second, we analyze whether wage differences could
arise for immigrants who maintain their first job due to the presence of close and weak
ties and job access mechanisms (section III.2).

1.3.1 Job match and social networks

We analyze the impact of social networks on the quality of the job matching process for
immigrant workers. A “good” match is considered if the skills and qualifications of the
worker are those required for the job. Then, if a good match between employers and
employees takes place, a longer duration of the worker in the same job is expected.
Duration models require information on contract job duration. However, the required
information is not provided by the ENI. Therefore, as an alternative to these models and
based on the information contained in the ENI, the quality of the matching process is
studied through the probability of keeping the first job in Spain in comparison to not
keeping it, namely relative to being employed in a different job, being unemployed, or
being inactive.9
Other studies analyze the quality of the job match through a comparison of the
last occupation in the country of origin and the occupation obtained in the country of
destination (Mahuteau and Junankar, 2008; Simón et al., 2011); or by studying
occupational mobility in the host country, comparing the first and the actual occupation
in the host country (Simón et al., 2011). However, this approach excludes from the
study those immigrants that (i) do not have previous labor experience in the country of
birth and (ii) those immigrants that despite having previous experience in the country of
birth and a first job in Spain are actually unemployed or inactive. Therefore, the
definition of the job matching process considered in this paper includes those

9 An individual is classified as “keeping the first job” if she declares that the actual job is the first
obtained in Spain. Specifically, the ENI (2007) asks for actual labor status in Spain. If the individual
declares being employed, then she is asked if this is the first job obtained in Spain. If the answer is “yes”,
the individual is considered to currently be in the first job. Otherwise, if she answers negatively, then we
consider she has had a different job since arrival. Employment stability is observed if the immigrant is
employed in the first job obtained in Spain.

19
immigrants that after having a first job in Spain are now in a different job, unemployed
or out of the labor market, thereby reflecting job mismatch.
The hypothesis to test is that the probability of keeping the first job is affected
by immigrants’ close and weak ties as well as the job search mechanisms used to obtain
the first job in Spain. Depending on the relationship (positive or negative) found
between social networks and actual labor market status, this would reflect the positive
or negative impact of social networks on the job matching process between workers and
employers.

To assess this relationship we use the following multinomial logit regression:

exp⁡(𝛽´𝑗 𝑋)⁡
𝑃(𝑌 = 𝑗|𝑋) = J
∑j=0 exp⁡(𝛽´𝑗 𝑋)

where 𝑃(𝑌 = 𝑗|𝑋) is the probability of observing the 𝑗 ∈ {0, 𝐽} outcome of the
dependent variable 𝑌 conditional on the vector 𝑋 of independent variables. 𝛽𝑗 is the
vector of regression coefficients to be estimated by the maximum likelihood method.
In this study, the dependent variable (𝑌) measures four possible labor market statuses,
namely being employed in the first job obtained in Spain, being employed in a different
job, being unemployed, or being inactive.10 The independent variables of interest are the
immigrant social networks in the host country and job access mechanisms for the first
job.
We consider different measures of the strength of immigrants’ networks. Close
ties is a dummy variable that refers to whether the immigrant had at least one relative or
friend on arrival in Spain. Endogenous network formation and the ensuing problem of
reverse causality are important empirical issues that need to be addressed in this
analysis. For instance, social networks might be affected by labor market outcomes in
that labor market status may influence social interaction and social relationships by
creating or limiting interaction opportunities. As Goel and Lang (2011) and Kahanec
and Mendola (2008) point out, contacts at arrival are largely exogenous with respect to
the individual’s subsequent labor market experience. The other two measures used in
the literature refer to weak ties: participation in social organizations distinguishing those

10
Inactive refers to those immigrants actually studying or involved in non-waged household activities,
excluding retirees.

20
devoted exclusively to immigrants and those not, and the proportion of immigrants of
the same country of birth living in the same region of the total immigrant population in
the region as a proxy of the network size (Munshi, 2003; Kahanec and Mendola, 2008).
Because the ENI is only representative at national level, the Municipal Register (Padrón
Municipal de Habitantes) for 2007 was used to calculate the share of immigrants by
country of birth in the different Autonomous Communities of Spain.11
Besides the key variables of interest, other control variables include socio-
demographic characteristics (age, gender, education, region of origin, region of
residence in Spain, proficiency in the Spanish language, legal residence authorization),
migration experience (internal migration in Spain), remittance behavior, and first job
characteristics in Spain (activity sector and occupation). In addition, variables referring
to immigrants’ labor market status and last occupation in the country of origin are
included. These variables are incorporated in order to control for potential unobserved
heterogeneity. Identifying the effect of social networks is difficult because unobserved
individual attributes such as being sociable, being ambitious, being responsible, can be
correlated with both the probability of having contacts at arrival and their own
probability of being at different labor market statuses. In addition, social interactions are
more likely to emerge among individuals that share some relevant traits, such as
education, occupation or ethnicity. Therefore, the estimated effect could be biased and
may not be attributable to a network effect. By controlling for several observable
characteristics, we are able to partially remove the potential bias arising from omitted
personality traits. A priori, it is not clear the direction of the bias. If omitted personality
traits affect both labor market outcomes and social network in the same way, neglecting
them leads to an upward bias in the coefficient, and thus an overestimation of the effects
of the networks in the multinomial regression. Otherwise, the estimated coefficients will
be downward biased. A first insight is provided in Table 3 Panel B, in which we
observe that the proportion of workers at different labor market statuses is similar
between immigrants with and without close ties, and among those with and without
weak ties. In order to disentangle the magnitude and direction of the potential bias, the
multinomial regressions are estimated with and without the skills variables such as
educational level, proficiency of the Spanish language, and previous labor experience in
the host country.

11
An Autonomous Community is a first-level political and administrative division of Spain (NUTS 2).

21
 Another source of concern could be sample selection as the individuals
considered in this analysis are those with some experience in the Spanish labor market.
In order to correct for this problem a two-step Heckman procedure adapted to logistic
regression is implemented, which consists of a two-step estimator and a maximum
likelihood estimator (Durbin and Rivers, 1990). In the first step, the probability of
having any experience in the Spanish labor market is estimated. The probability that an
individual has worked is modeled as a function of individuals’ socio-demographic
characteristics, social networks, internal mobility, and motives for migration. From this
equation, the Mills ratio is estimated. The second step estimates the probability of those
immigrants in the labor market being in one of the four outcomes stated before but
including the correction coefficient (obtained through the Mills ratio) as an additional
covariate. A key issue in this analysis is that the exclusion restriction should not be
directly related with subsequent labor market statuses.
In this study, the exclusion restriction includes two dummy variables which refer
to migration motives: family regrouping and labor motives. On the one hand,
individuals migrating for family reasons may be less prone to work (as they are
expected to engage in non-remunerated household activities). On the other hand, given
that they have at least one family member when arriving in the host country, it may be
easier for them to access job information. In Section II we observed that individuals
with and without labor experience in Spain differs in terms of motives declared for
migration. While 70% of the individuals with labor experience declare labor reasons for
migrating, 60% of those without labor experience declare family regrouping motives
(Table 1). We can expect that migration motives and immigrant’ subsequent labor status
are related, but only indirectly. A possible channel through which migration motives
may affect the quality of the job matching process is through its impact on immigrant
legal status, since having or not legal residence authorization determines whether
immigrants can freely or not search for a better job. Those who migrated for family
reasons may have already a family member with legal residence authorization who
could provide information on the legalization process, or facilitate their access to legal
status, which in turn affects the subsequent labor market status. Conversely, immigrants
declaring labor motives may quickly accept a job, because is the most direct path
towards being legalized. Thus, because of their precarious situation, they are more
prone to accept any kind of job, even if it does not match with their skills. In addition,

22
by controlling for a broad set of skill variables, we partially remove the unobserved
heterogeneity problem.
Reinforcing the exclusion restriction, Aydemir (2011) shows for the Canadian
context that immigrants’ labor market outcomes highly depend on their skill levels and
on the transferability of those skills rather than on visa categories. For the Spanish case,
Rodríguez-Planas and Vegas (2012a) find that Moroccan immigrants who declare
regrouping motives are less prone to work than immigrants declaring labor motives.
Moreover, the authors find that, once the employment decision is accounted for, no
wage differentials arise between immigrants declaring different motives for migrating.12
In sum, we can assume that migration motives are not expected to directly affect
the quality of job match. In formal terms, a good job match depends on workers’
supply-side efforts, the number of workers offering those services in the job market, and
the demand for their skills and qualifications. For instance, educational level or prior
work experience could affect the job match. For immigrant workers, language
proficiency, legal status and years living in the destination country are also important
issues.

1.3.2. Social networks and wages

Next, we test whether wage differences could arise between immigrants who maintain
their first jobs due to the strength and size of an immigrant network and the job access
mechanisms used to find the job. The hypothesis to test is that conditional on keeping
the first job in Spain, wage differentials could occur between those who found the job
through formal or informal methods and between those with or without close and weak
ties. The effect of social networks on wages is still a controversial issue in the empirical
literature. While Bentolila et al. (2010) find a wage penalty for those workers who
found their jobs through personal contacts, Pellizzari (2010) shows that the use of social
networks can lead either to a wage premium or to a wage penalty in different EU
countries.

12
These authors stress the potential endogeneity problem in studies that analyze immigrants’ labor market
outcomes with different types of visa in countries with a clear immigration policy regime in place, which
is very likely to be endogenous to the country’s social, economic, and political context, and at the same
time affect the settlement process of the different types of immigrants it receives. This issue is not present
for the Spanish case, considered as an immigrant-friendly country because of the lax implementation of
immigration laws and several generous amnesties granting legal residence to illegal immigrants (p.4).

23
 The study of social networks effects on wages consists of estimating a wage
equation of the following type:

𝑙𝑛𝑤𝑖 = (1, 𝑁𝐽𝑖 , 𝐶𝑇𝑖 , 𝑁𝐽𝑖 ∗ 𝐶𝑇𝑖 , 𝑁𝑆𝑖𝑗 , 𝑊𝑇𝑖 , 𝑋𝑖 ). 𝛾 + 𝑢𝑖 (1)

where 𝑤𝑖 is the hourly wage, network job (𝑁𝐽𝑖 ) is a dummy equal to 1 if individual i
used personal contacts to find the first job and 0 if used formal channels; while close
ties (𝐶𝑇𝑖 ) is a dummy equal to 1 if the individual had contacts on arrival and 0
otherwise. An interaction term between 𝑁𝐽𝑖 and 𝐶𝑇𝑖 is included in order to capture if
wage difference between those who found their job through its networks and those who
used formal methods is related to the presence of close ties.13 The network size (𝑁𝑆𝑖𝑗 ) is
measured by the proportion of immigrants of the country of origin of individual i living
in region j of the total immigrant population residing in region j. Weak ties⁡(𝑊𝑇𝑖 ) is
proxied by a dummy variable equal to 1 if individual i participates in social mixed
organizations, while 𝑋 is a set of demographic and socio-economic controls (the same
as in previous section except remittance behavior) and 𝛾 is a column vector with the
parameters of the equation.
Equation (1) is estimated by OLS and QR. QRs, introduced by Koenker and
Bassett (1978), estimate the conditional quantile function, namely models in which the
quantiles of the conditional distribution of the response variable are defined as functions
of observed covariates.14 QRs are used because OLS implicitly assumes no important
differences in terms of the impacts of the exogenous variables along the conditional
distribution. Instead, if exogenous variables influence the parameters of the conditional
distribution of the dependent variable other than the mean, then the analysis that
disregards this possibility will be severely weakened. Unlike OLS, QR models allow for
a full characterization of the conditional distribution of the dependent variable, bringing
much value added if the relationship between the regressors and independent variables
evolves across its conditional distribution. Second, unlike the OLS regression that is
sensitive to the presence of outliers and can be inefficient when the dependent variable
has a highly non-normal distribution, the QR estimates are more robust. Third, unlike

13
When interpreting the coefficients on close ties, network job and their interaction, it should be noted
that the omitted group is that of immigrants in formal jobs and without close ties.
14
Similar to the OLS method, the parametric QR can be presented as the solution to a minimization
problem. In this case, the asymmetrically weighted value of the residuals is considered to compute the
parameters. For more details, refer to Koenker and Bassett (1978) and Koenker and Hallock (2001).

24
OLS, QR estimators do not require existence of the conditional mean for consistency
(Cameron and Trivedi, 2005). This flexibility has thus far been precluded from social
networks’ effects on wages in empirical studies, which has left unaddressed the possible
impact of social networks upon inequality through its within-levels inequality
component.
Because the sample is restricted to those immigrants still employed in the first
job obtained in Spain, sample selection bias could emerge.15 The nature of the
underlying problem requires sample selection models since the conditional quantile of
the observed wages depend on a bias term of an unknown form, a two-stage
semiparametric method is used. Specifically, the methodology followed to address this
issue is the one proposed by Buchinsky (1998) which is similar to the one proposed for
mean regression by Heckman (1979).
This study is conducted separately for women and men in order to account for
the different factors that may influence wages by gender.16 First, we estimate the
probability of keeping the first job in Spain (the selection equation). Second, the wage
equation regression is estimated. This methodology needs at least one variable which
explains the probability of keeping the first job but not directly related with the outcome
of interest. As in many other studies, finding suitable instrumental variables is far from
straightforward, since almost any regressor that determines the probability of keeping
the first job could plausibly affect wages as well. The literature commonly uses as
exclusion restriction the number of children at home or the marital status. However,
these variables may be correlated with wages.17 Also, variables on tenant or ownership
status are used to account for possible sample selection in the decision of participation
(Rodríguez-Planas and Vegas, 2012b). In this study, the exclusion restriction is a
dummy variable that indicates whether the immigrant sends remittances to her country
of origin or not.18 This variable reflect immigrant responsibilities in the home country,

15
The sample is restricted because the ENI (2007) only provides wages for actual employment and does
not provide information about the mechanisms through which the worker obtained the job. On the
contrary, information on job access mechanisms is only given for the first job in Spain. As the aim of this
study relies on both wages and job access mechanisms, the sample is restricted to those who keep the first
job obtained in Spain.
16
As the literature on the participation of women in the labor market points out, women’s decisions to
participate have important implications on their wages.
17
There are theoretical arguments that suggest that labor supply, wages and fertility are endogenous. If
women with relatively low expected future wages had on average a high fertility, the exclusion restriction
would fail.
18
The translated question of the ENI (2007) reads: Do you sent money out of Spain? Respondents can
choose yes or no.

25
such as dependent family members or monetary debts (such as mortgage debts or
credit), or investment decisions, which may, in turn influence the individual probability
of keeping the first job, change jobs, being unemployed or inactive. 19 Moreover, as we
only consider whether the immigrant sends remittances or not instead of considering the
amount of money remitted, this variable is expected to be unrelated to current wages,
since wages strongly depend on actual labor market conditions in the host country, past
labor experience in the country of birth and on the worker’s human capital endowments.
The literature on economic integration reinforces the exclusion restriction. This
literature relates immigrants’ remittance behavior with their economic integration in the
host country.20 Studies that analyze the relationship between labor market status and
remittance behavior finds that, on the one hand, employed immigrants are more prone to
remit than unemployed or inactive immigrants (Bilgili, 2013; Al-Ali and Koser, 2001;
Holst and Schrooten, 2006). On the other hand, Holst and Schrooten (2006) find that
income has no effect on the probability to remit while it is only significant for the
amount of remittances.
The conventional Heckman correction method is applied to the OLS estimation.
However, an analysis of the distribution of the error term in the selection equation is
needed for QR because the conventional Heckman correction method assumes a
standard normal distribution of the error term in the selection equation. If this
assumption is violated, then semi-parametric methods should be applied to estimate the
first equation, because this method does not rely on a distributional assumption
(Buchinsky, 1998). This model (as the conventional Heckman procedure) highly relies
on the assumption that the variables included in the exclusion restriction are not related
to the outcome variable in the second equation.
The wage equation with semi-parametric correction for sample selection bias is
estimated following Buchinsky (1998) (See the Methodological Appendix for a detailed
description of the model).

19
Since the nature of our sample selection bias is different for the one related to the decision of working
or not, these variables may be potentially related with wages in our case, thus violating the exclusion
restriction assumption. We instead tried with alternative instruments such as home or land ownership in
the country of origin, having or not relatives in the country of birth, proving not to be useful instruments.
The estimated coefficients in the first stage were not statistically significant. Nonetheless, because of
concerns with endogeneity of our instruments we estimated the wage equation including these variables
as controls. When doing so, most of the coefficients of interest remain unaffected.
20
Economic integration of immigrants is stronger when they have higher participation rates, lower
unemployment levels, better jobs and, not directly related to labor market participation, higher income per
person at the household level (Bilgili, 2013).

26
 The quantiles of the log wage are given by:

𝑄𝜃 (𝑦|𝑥2 ) = 𝑥2′ 𝛽𝜃 + ℎ𝜃 (𝑥1 𝛾0 )⁡∀⁡𝜃 ∈ (0,1) (2)

The vector 𝑥1 is a set of observable characteristics that may affect the probability that an
individual keeps the first job obtained in Spain while 𝑥2 is a subset of 𝑥1 , which
contains labor market characteristics that could influence on the wage offer. In other
words, 𝑥1 must also contain at least one variable that is not included in 𝑥2 (the exclusion
restriction). These variable (or variables) should be uncorrelated with the log wage. The
term ℎ𝜃 (𝑥1 𝛾0 ) corrects the selection at the θth quantile. It plays the role that the Mills
ratio plays in the usual Heckman (1979) procedure, but it is quantile-specific and more
general so not to assume normality.

Buchinsky (1998) suggests a series estimator:

ℎ̂𝜃 (𝑥1 𝛾0 ) = 𝛿0 (𝜃) + 𝛿1 (𝜃)𝜆(𝑥1 𝛾0 ) + 𝛿2 (𝜃)𝜆(𝑥1 𝛾0 )2 + ⋯,

𝜙(.)
where 𝜆(. ) is the inverse Mills ratio defined as 𝜆 = Φ(.), while 𝜙(. )⁡and Φ(. ) are the

density and the c.d.f. of a standard normal variable, respectively. Thus, first 𝛾0 needs to
be estimated. As wages are only observed when the individual keeps the first job, we
only observe whether a dummy indicator D equals 1 or 0. This could be written as:

𝐷 ≡ Pr(𝐷 = 1|𝑥1 ) ≡ 𝐹(𝑥1 𝛾0 ) (3)

where 𝑥1 is a set of individual characteristics and 𝐹(𝑥1 𝛾0) is an unknown function. we

estimate 𝐹(𝑥1 𝛾0 ) using a semiparametric estimator proposed by Klein and Spady
(1993). This estimator is asymptotically efficient in the sense that it attains the
semiparametric efficiency bound (Melly and Huber, 2008).
The two-step semi-parametric method can be summarized as follows:
1. Estimate the probability of not changing the first job using the semi-parametric
index proposed by Klein and Spady (1993).
2. Estimate the parameters in the QR including an approximation of the selective
term as stated by Buchinsky (1998).

27
 1.4 Empirical findings

1.4.1 Job match and social networks

Table 4 presents the probability of having some labor experience in Spain (the first step
in Heckman’s method for binary models). Relative to the key independent variables,
close ties increase the probability of having some labor experience. However, network
size or social participation in any kind of organization has no impact on the likelihood
of having some labor experience. The exclusion restriction has an important effect on
the dependent variable and it is estimated with precision. In line with previous studies,
immigrants who declare labor motives for migration are more prone to have labor
experience in Spain, while those declaring family regrouping motives are less likely to
be involved in the labor market.
The other control variables have the expected sign. Being a woman, being
married, the number of children in the household21, are all negatively related to the
probability of having some experience in the Spanish labor market. The region of origin
has different impacts on the probability of labor experience in Spain. Immigrants from
North Africa are less prone to participate while those belonging to an Eastern European
or a Latin American country all positively influence the probability of having some
labor experience. Variables referring to the social assimilation process, such as years
living in Spain, having legal residence authorization, and proficiency in the Spanish
language, increase the probability of labor experience. Finally, a positive relation is also
observed relative to internal migration in Spain.
The estimated multinomial regression after controlling for sample selection is
shown in Table A.5 in the Appendix.22 Because the coefficients obtained through the
multinomial logit model do not measure the effect of the explanatory variables on the
outcome probability directly, we focus on the results reported in Table5, which shows
the average marginal effects of the independent variables on the probability of each of
the four labor statuses from the multinomial logit model.
In light of the hypothesis stated, the results provided below reflect a mismatch in
the labor market for immigrants on arrival, showing that upon arrival immigrants prefer
to quickly accept a job offered through the social network, even if it is not the most
suitable given their education, training, or previous experience. This mismatch is

21
We distinguish between number of children living in Spain and in the country of origin.
22
All the results of the multinomial model are interpreted in relation to the omitted labor status: being
employed in a different job from the first one obtained in Spain.

28
observed when analyzing the effects of social networks through different labor market
statuses (Table 5). For instance, immigrants with close ties are less likely to keeping the
first job while more prone to change jobs (8.9 and 5.2 percentage points, in columns 1
and 2 respectively) reflecting the importance of close ties in terms of job information
transmission or financial support when immigrants search for another job. Ottaviano
and Peri (2006) argue that job mismatch could happen because jobs found through
relatives and friends are often unrelated to the individuals’ previous experience or
training. This is the case when the social capital accumulated by the network is
restricted to a particular segment of the labor market, in which case the new
immigrant’s job prospects are limited to this segment. Therefore, this great dependence
on social networks may also reflect segmentation in the host labor market as well as a
lack of access to host labor market institutions.
In addition, a mismatch is more likely to be observed for immigrants who
obtained the first job through social networks in comparison to those who used formals
channels (6.5 percentage points in column 1). Interesting, those immigrants who have
had a first network job are more prone to being unemployed (4.6 percentage points),
thus reinforcing the negative effect of informal job access channel on the matching
process. Further, for immigrants who took less than a month to find the first job, those
who obtained the job through their social networks are more likely to be mismatched in
relation to those who used formal search methods. In line with our results, Bentolila et
al. (2010) find a mismatch for workers who access their current job through social
networks for the US and Europe. According to these authors, workers have a natural
talent for a specific occupation, which may not be the one to which their social contacts
can provide referrals. In this scenario, workers may have to accept a trade-off; they may
find it advantageous to find a job more quickly through their social networks, but they
may also work in an occupation that does not maximize their productivity.
The results in Table 5 column (1) show that the probability of keeping the first
job decreases for immigrants with weak ties (13.8 percentage points), but it is not
statistically significant on the probability of changing jobs. We find that conditional on
having obtained the first job through social networks, the probability of keeping the job
is independent of the network strength. This effect is measured through the sum of the
coefficients on close ties and this variable interacted with network jobs (almost 0). A
statistical and significant effect (positive or negative) would imply that immigrants with
close ties are better or worse workers, having different proclivities to receive network

29
and formal offers, than for those without them. The interaction term between close ties
and network job measures the causal effect of having close ties on the difference
between the expected probabilities of keeping the first job conditional on choosing
different channels to obtain the first job. Our results show that the probability of
keeping a network job is higher for immigrants with close ties than the probability of
keeping a formal job for those without close ties (8.7 percentage points).
In addition, the bigger the network size the more likely that the immigrant keeps
the first job. No statistical and significant effects of network size, weak ties or the job
access mechanisms, on the probability of changing jobs are found (column 2 in Table
5). Loury (2004) points out that differences between industries and employers may also
account for ethnic and race variations in contact effects. Ethnics groups have established
specific occupational and employment niches that facilitate employment and training of
members of their group and that limit access of outsiders. This may explain the positive
effects of network size in the probability of keeping the first job. This is also consistent
with Veira and Stanek (2011) who find ethnic niches in the Spanish labor market.
Next, we explore the effects of close and weak ties on the probability of being
unemployed (column 3 in Table 5). Contacts on arrival and network size do not
influence the likelihood of unemployment. However, the probability of unemployment
decreases for those immigrants with more years living in Spain participating in social
mixed organizations, reflecting a positive effect of individual’s social integration.
Finally, a positive effect of close ties on the probability of being inactive is observed
(3.4 percentage points), while immigrants who got the first job through social networks
and with close ties are less prone to being inactive relative to those who got the job
through formal channels and without close ties (column 4, Table 5).
While the primary interest of this study is on social networks, a brief look at the
results of the control variables is provided. The results reported in Table 5 are consistent
with previous findings in the literature. For example, being a woman increases the
probability of unemployment or being inactive, while decreases the probability of
changing jobs. Immigrants from Western Europe experience better matches in the
Spanish labor market than other immigrants groups. Different impacts of regions of
destination on labor statuses outcomes are also found, reflecting differences in labor
market conditions and opportunities for immigrant workers across Spain. Years living
in Spain decreases by almost 5 percentage points the probability of keeping the first
employment. Consistent with the idea that legal migrants can search freely in the host

30
labor market, those with legal residence authorization are more prone to change jobs
and less prone to be unemployed.
Statistical and significant effects of human capital endowment on the probability
of being in different labor market statuses are found. The probability of changing jobs
decreases for immigrants with secondary education, while immigrants with tertiary
education are less likely to be unemployed. Proficiency in the Spanish language
decreases the probability of unemployment. Immigrants with skilled occupations in the
country of origin are more likely to be mismatched upon arrival. Specifically, they are
more likely to switch jobs possibly for a better one, more in accordance with their
previous labor experience. Being a student before migration also increase the
probability of mismatch upon arrival.
Overall, our results support previous studies that stress the difficulties in
transferring immigrants’ previous labor experience and credentials. Once established in
the host country, immigrants search for a new job more in accordance with their levels
of education, previous experience, and training. Also in line with our results, Simón et
al. (2011) and Veira and Stanek (2009) find a U-shaped pattern in terms of occupational
mobility for immigrants in Spain, characterized by occupational downgrading on arrival
and a gradual improvement as the duration of residence in the host country increases.
First job characteristics also influences on the job matching probability.
Immigrants employed in qualified occupations, such as managers or skilled workers are
more likely to experience a good match, as well as those employed in any sector in
comparison to those employed in agriculture.
Finally, immigrants sending remittances to their country of birth are less prone
to keep the first job while are more likely to changing jobs. It is well addressed in the
literature that immigrants’ remittances are very important to financially support stayers
in the country of origin, namely own children, parents or other family members.
Considering these motives, immigrants probably put more effort in searching for better
jobs, more stable or with better labor conditions.
In order to be more confident in the presented results, some robustness checks
are made. First, a separate analysis is conducted for women and men. The magnitudes
of the coefficients of the key independent variables varies across gender, however the
relationship between social networks and job matches exposed above remain (Table 6).
In addition, in order to get some insight of the sign and magnitude of the potential bias
due to unobservable characteristics, we estimate the marginal effects excluding

31
measured skill variables, such as educational level, proficiency in the Spanish language,
labor status before migration and last occupation in the country of birth. As can be seen
in Table 7, the magnitude and sign of the key independent variables are similar to the
ones provided in Table 5. The estimated coefficients without controlling for these
variables would be downward biased for close and ties and informal search mechanism,
while upward biased for network size and weak ties coefficients. Finally, in order to
remove any concern with endogeneity of the variables included in the exclusion
restriction, we re-estimate the multinomial logit model and the average marginal effects
including the motives for migrating as controls. The estimated coefficients of our key
variables do not change significantly (Table 8).
While it remains possible that there is an important measure of skill that is
correlated with the immigrant’ social networks, the fact that excluding these extensive
set of variables does not alter the results in an important way, gives a reasonable level of
confidence in the presented results.

1.4.2 Social networks and wages

This subsection aims to study the impact of social networks on wages for male and
female immigrants keeping the first job.23 For this purpose, a two-step procedure is used
in order to control for the possible selection bias arising from selecting workers that
keep their first jobs in Spain (see Section III.2). This analysis is conducted separately
for women and men.
In the first step, the probability of keeping the first job, the selection equation, is
estimated using a standard probit model. After probit regression, the hypothesis of the
normality of the residuals is rejected for women but not for men (Table 9). Hence, for
women, the selection equation is estimated using the semiparametric estimator proposed
by Klein and Spady (1993). For men, a standard probit model is estimated. The
exclusion restriction is a dummy variable equal to one if the immigrant sends
remittances to her country of birth and zero otherwise (see Section III.2).
The factors influencing the probability of keeping the first job was largely
analyzed in the previous section. Therefore, for the sake of brevity, we do not describe
the results obtained for the selection equation, which are reported in Tables A.6 and A.7

23
Recall that in this case, the sample is restricted to those immigrants who keep the first job, since the
ENI provides information on the job access mechanisms only for the first job, while wage information is
provided for current job.

32
in the Appendix, for women and men respectively.24 However, it is worth noting that
the exclusion restriction, the dummy variable that indicates if the individual sends
remittances, is statically significant and negatively related to the probability of keeping
the first job for both women and men.25
The results presented in Table 10 reflect different impacts of social networks on
wages by gender and across the observed wage distribution. For instance, the job access
mechanism influences wages. Both women and men who have obtained the job through
social networks present a wage penalty in comparison to those who used formal
channels. This penalty is present across the distribution, observed for the 25th and 50th
percentile for women and among different percentiles for men. However, some
important dissimilarities between female and male appears, for instance in the
magnitude and significance of the coefficients. Having obtained the job through social
networks has a lower negative impact for women. At the 25th percentile the gap is 3.7%
and statistically significant at 10%, and this pattern is observed until the 50th percentile
where the coefficient is 0.034. For men, the wage gap is around 11.3% at the 25th
percentile, 10.8% at the 50th percentile, and 11.7% at the 75th percentile, and
statistically significant at 1% in each percentile. These estimates evidence how the
penalty for being employed in a network job has also a gender dimension that favors
female.
These results are in line with Bentolila et al. (2010) who find a wage penalty
across workers who obtained the job through informal channels. According to Pellizzari
(2010), the positive or negative effects of social networks on wages could be related to
employer characteristics, which in turn determine the context in which job search
methods operate. It could be the case that for some employers, desired applicant
characteristics may be easily discernible through formal channels rather than relying in
recommendations from trusted sources. Pellizzari (2010) finds substantial variations in
the effects of social networks on earnings. This author states that wage penalties are
likely to happen in industries where firms invest substantially in formal recruitment
activities. Firms are more likely to undertake such investments for high productivity

24
Strictly speaking, the estimated coefficients of the semi-parametric model are not comparable with the
ones obtained in the previous section through the multinomial logit regression. This is so because the
coefficients estimated in the semi-parametric model only indicate the sign of the effect, but not the
elasticity, which could be obtained through the estimation of average marginal effects.
25
An important difference between women and men, is that the probability of keeping the first job for
women decreases with the number of children in the country of birth and in Spain, while for men this is
not statistically significant. So, for men we only consider the total number of children.

33
jobs where the cost of turnover is substantial. When large investments are made,
workers found through formal channels average higher productivity than those found
through other means. An alternative explanation is that referred workers are segregated
into low wage types of jobs with respect to no referred workers. Then, workers who
access job through social networks earn less than those who used formal mechanisms.
Looking at the strength of the social network, we observe that close ties only
affect men’ but not women’ wages. For women, the estimated coefficients of close ties
and the interaction term between close ties and network job is not statistically
significant, meaning that regardless of the channel of access to employment, the
presence of close ties does not have statistically significant effect on wages. Conversely,
conditional on having found a job through formal channels, a glass ceiling effect of
close ties on men’ wages is observed. This effect refers to a wider wage gap at the top
of the distribution, suggesting that those men who obtained the job through formal
channels with close ties in the high-income jobs earn less than workers without close
ties. In other words, hourly wages decreases with close ties throughout the conditional
wage distribution For instance, the return to having close ties decreases from 8.4% to
11.5% between the 25th and 75th percentile. This could be interpreted as a negative
ability returns relationship as evidence that having close ties and ability are related,
which if true suggest that less able individuals benefit less from the presence of close
ties. However, because individuals’ abilities are unobserved for the researcher, it is
difficult to isolate the effect that drives the heterogeneous pattern of returns to personal
contacts across the wage distribution.
When interacting close ties and network job variables, the coefficient shows that
the returns to the channels of search differ for men with and without close ties. The
positive and statistically significant coefficient observed at different quintiles of the
distribution shows that immigrants who got the job through social networks and with
close ties are higher than for those who obtained the job through formal channels and do
not have close ties. In other words, a network premium (understood as the difference of
wages between network jobs and formal jobs) is observed across the wage distribution.
Moreover, this wage premium increases for higher percentiles reflecting a sticky floor
effect. This effect is observed when the gap widens at the lower percentiles of the wage
distribution.
Next, the role of weak ties on wage distribution is analyzed. The estimated
coefficients show great differences across genders. For women no statistically

34
significant effects are found. Conversely, a wage penalty is observed in the 25th and
50th percentiles of the distribution for men. For the highest percentile, the estimated
coefficient is still negative but no statistically significant. However, this penalty is
reversed as the length of time living in Spain increases possibly reflecting the positive
effects of the social integration process in the host country (Table A.9).
The network size penalizes both women and men wages. This effect is observed
for the median of the distribution for both genders, and in the 75th percentile of the
distribution only for men. This is consistent with Calvó-Armengol and Jackson (2007)
who state that in the short run, network size has a negative impact on labor market
outcomes due to competition for job information within the network, which negatively
affect immigrants’ wages. Other explanations points out the strong presence of
immigrants from the same country of origin may indicate the presence of immigrant
enclaves and, therefore, segmentation in some occupations in the labor market, which
results in wage penalties (Chiswick and Miller, 2005). This possible explanation is the
counterpart of the results exposed above, that social integration (as opposite to enclaves’
formation) in the host country positively affects wages.
In the case of the estimates of the control variables, the results reported in Tables
A.8 and A.9 in the Appendix are in the direction one would expect. Covariates referring
to socio-demographic characteristics, such as marital status and number of children
living in the immigrant’ country of birth has different impacts on wages across gender.
While being married penalizes women wages (statistically significant in the 50th
percentile of the distribution), a wage premium is observed for men across different
percentiles of the distribution. In addition, the marital status and the number of children
loses significance for higher percentiles of women’ wage distribution.
The region of origin also impact on wages. Immigrants from Western Europe
present a wage premium in comparison to other immigrant groups. This wage premium
is observed in the 50th and 75th quantiles of the distribution for women, and across the
whole distribution for men. In addition, wage differentials are observed within the
Spanish territory. This could be reflecting regional disparities in terms of productive
structures and labor markets dynamics in Spain.
In line with the literature, immigrants with legal residence authorization present
a wage premium present across the distribution. Differences in power negotiation
between immigrants with and without legal residence might explain this result. Human
capital endowments positively affect wages. Immigrants with tertiary education present

35
a wage premium at different quintiles (statistically significant for women at the 50th
percentile of the distribution, and for different percentiles for men). Men with
proficiency in Spanish language earn more than men without it. Conversely, proficiency
in Spanish language does not affect women wages.
Variables referring to actual occupation and last occupation in the country of
origin are also relevant on affecting wages. Men in skilled occupations present a wage
premium across the different percentiles. For women in professional and managerial
activities, positive and statistically significant returns on wages are observed for the
50th and 75th percentile of the distribution. Similar effects of last occupation in the
country of origin are observed for both genders. These result are not surprising, it is
expected that more qualified occupations pays better, and premiums those workers that
have the human capital endowments and previous experience required for the job.
However, for less skilled jobs, other factors such as the region of origin, the legal status,
or the years living in the host country seems to be important individual attributes and
more relevant than those referring to human capital endowments or previous experience
in the country of origin. It could be also the case that for employers these socio-
demographic factors are relevant for screening workers.
Finally, the actual sector of activity has different returns on wages and across
genders. For women, the only sector that is significant is the household activity sector in
comparison to agriculture. In this case, a wage penalty is observed for the 50th
percentile of the distribution. For men employed in construction, returns are higher than
wages in the agriculture sector, and this is observed across the wage distribution. This is
consistent with the construction boom that took place in this period in Spain, and the
consequent high labor demand of this sector. In addition, men working in industry or in
firm services present wage premiums across the wage distribution. The other activities
namely, trade, education and health services, and transportation, present a wage
premium in the 50th and 75th percentiles. The only sector that presents a wage penalty
is the household activity.

36
 1.5 Conclusion
This paper investigates the extent to which social networks influence immigrants’ labor
market outcomes in Spain. Using micro-data from the ENI, we identify the effect of
social networks by examining the effect of close and weak ties, network size and job
access mechanisms on immigrants’ labor market outcomes. The empirical strategy is
conducted in two steps. First, we study the impact of social networks on the probability
of being in different labor market statuses. Second, for those immigrants who keep the
first job, we study whether wage differentials could arise due to the presence of social
networks. Because sample selection could arise in this study, the analyses are conducted
in a two-step procedure similar to the one proposed by Heckman. In addition, a broad
set of control variables are included in order to control for potential unobserved
heterogeneity.
The findings reported in this paper indicate that a mismatch takes place in the
labor market for immigrants on arrival. Immigrants tend to quickly accept a job offered
through the social network, even if it is not the most suitable job given their levels of
education, training, and previous experience. Once established in the host country,
immigrants search for another job possibly more in accordance with their human capital
endowment. Second, different effects of social networks on wages by gender and across
the wage distribution are observed for immigrants who keep the first job. Workers who
obtained the job through social networks present a wage penalty in comparison to those
who used formal channels. This is observed for the 25th and 50th percentile for women
and among different percentiles for men. In addition, the strength of the network only
penalizes men’ wages but do not influence women’s wage. As the length of time living
in Spain increases, men’ participating in social mixed organizations present wage
premium in comparison to those not participating. The network size also penalizes both
women’s and men’s wages. Conditional on having obtained the first job through social
networks, men with close ties present wage premium in comparison to those who got
the job though formal channels and without close ties. This effect is not statistically
significant for women.
To sum up, two main factors influence immigrants’ labor market outcomes.
First, their great reliance on personal contacts as a job access mechanism is reflected in
a mismatch in the labor market and in wage penalties across the distribution for both
women and men. The positive effect of network size on job match and its negative
impact on wages may be reflecting the presence of segmentation in some occupations in

37
the labor market. Second, human capital endowment are partially transferred to the host
country, negatively affecting the matching process upon arrival.
In light of these results, some considerations are made. First, it is important to
stress that policies whose objectives are to accelerate the assimilation process or
improve the labor market outcomes of immigrants not only have to focus on the
individual (such as improving human capital endowments), but might also influence
individuals’ social backgrounds and the social networks within which an immigrant is
embedded. If this strong dependence on social networks persists over time, the
integration process of immigrants in Spain may be compromised. Second, the
adaptation process of immigrants to labor institutions and transferability of previous
experience and education should be addressed.

Acknowledgements
We appreciate the comments and suggestions made by the participants of the 2013
EEA-ESEM meeting, session “Social Netwroks II”, held at the University of
Gothenburg; participants of the “Annual Meeting on Equaliity and Poverty:
Implications and Methods”, at Universitat Autònoma de Barcelona, Spain, December,
2012; and the participants of the “Doctoral Day XTREPP” workshop, at Universidad de
Barcelona, November 2012. We are especially grateful to Cristina López Mayan for her
insightful comments and to Javier Vázquez Grenno for carefully reading this essay and
for his helpful comments and suggestions.

38
 References

Aguilera, M. (2003) “The Impact of the Worker: How Social Capital and Human
Capital Influence the Job Tenure of Formerly Undocumented Mexican Immigrants”,
Sociological Inquiry, 73(1): 52-84.
Aguilera, M. and Massey, D. (2003) “Social capital and the Wages of Mexican
Migrants: New Hypothesis and Tests”, Social Forces, 82(2): 671-701.
Al-Ali, N., Black, R.; and Koser, K. (2001) “Refugees and transnationalism: The
experience of Bosnians and Eritreans in Europe”. Journal of Ethnic and Migration
Studies, 27 (4), 615-634.
Alcobendas, M., and Rodríguez-Planas, N. (2009) “Occupational Assimilation
After a Recent Immigration Boom”, IZA DP No. 4394.
Amuedo- Dorantes, C., and de la Rica, S. (2007) “Labor Market Assimilation in
Spain”, British Journal of Industrial Relations 45(2): 257-285.
Aslund, O., and Rooth, D-O. (2007) “Do when and where matter? Initial Labor
Market Conditions and Immigrants Earnings”, The Economic Journal 117(March): 422-
448.
Aydemir, A. (2011) “Immigrant Selection and Short-Term Labor Market
Outcomes by Visa Category”, Journal of Population Economics, 24: 451-475.
Bentolila, S., Michelacci, C., and Suarez, J. (2010) “Social Contacts and
Occupational Choice”, Economica, 77: 20-45.
Bertoli, S., Fernández- Huertas, J.; and Ortega, F. (2010) “Immigration Policies
and the Ecuatorian Exodus”. IZA DP No. 4737.
Bertrand, M.; Luttmer, E.; and Mullainathan (2000) “Network Effects and
Welfare Cultures”, Quarterly Journal of Economics, 115(3): 1019-1055.
Bilgili, Ö. (2013). “The links between economic integration and remittances
behaviour of migrants in the Netherlands”, UNU-MERIT WP 037.
Borjas, G. (1985) “Assimilation, Changes in Cohort Quality, and the Earnings of
Immigrants”, Journal of Labor Economics, 3(4): 463-489.
Borjas, G. (1994) “The Economics of Immigration”, Journal of Economic
Literature, 32(4): 1667-1717.
Borjas, G. (1995) “Ethnicity, neighborhoods, and human capital externalities”,
American Economic Review, 85(3): 365-390.

39
 Borjas, G. (2000) “The Economic Progress of Immigrants”, (in) “Issues in the
Economics of Immigration”, National Bureau of Economic Research, Inc.
Buchinsky, M. (1998) “The dynamics of changes in the female wage distribution
in the USA: a quantile regression approach”, Journal of Applied Econometrics, 13: 1-
30.
Buchinsky, M. (2001) “Quantile regression with sample selection: Estimating
women’s return to education in the U.S.”, Empirical Economics, 26: 87-113.
Calvó-Armengol, A. (2004) “Job Contact Networks”, Journal of Economic
Theory, 115: 191-206.
Calvó-Armengol, A., and Jackson. M. (2004) “The effects of social networks on
employment and inequality”, American Economic Review 94(3): 426-454.
Calvó-Armengol, A., and Jackson. M. (2005) “Job matching and word-of-mouth
communication”, Journal of Urban Economics 57: 500-522.
Calvó-Armengol, A., Patacchini, E.; and Zenou, Y. (2009) “Peer effects and
social networks in education”, Review of Economic Studies 76: 1239-1267.
Cameron, C., and Trivedi, K. (2005) “Microeconometrics: Methods and
Applications”, Cambridge University Press.
Cappellari, L., and Tatsimaros, K. (2010) “Friends’ networks and job finding
rates”. CESifo WP, No. 3243.
Carrasco, R., Jimeno, J.F.; and Ortega, C. (2008) “The effect of immigration on
the labor market performance of native-born workers: some evidence for Spain”,
Journal of Population Economics, 21: 627-648.
Chiswick, B.R., and Miller, P.W. (2005) “Do Enclaves Matter in Immigrant
Adjustment?”, City and Community, 4: 5-35.
De Luca, G. (2008) “SNP and SML estimation of univariate and bivariate
bynary-choice models”, The Stata Journal 8(2): 190-220.
Dubin, J., and Rivers, D. (1990) “Selection Bias in Linear Regression, Logit and
Probit Models”, Sociological Methods and Research, 18(2 & 3): 360-390.
Dustmann, C., Glitz, A., and Schonberg, U. (2010) “Referral based Job Search
Networks”, unpublished paper, Department of Economics, University College London.
Edin, P., Fredriksson, P.; and Aslund, Ö. (2003) “Ethnic Enclaves and the
Economic Success of Immigrants. Evidence from a Natural Experiment”, Quarterly
Journal of Economics, 118(1): 329-357.

40
 Eichhorst, W., Escudero, V., Marx, P.; and Tobin, S. (2010) “The impact of the
Crisis on Employment and the Role of the Labour Market Institutions”, IZA DP No.
5320.
Elliot, J. (1999) “Social Isolation and Labor Market Isolation: Network and
Neighborhood Effects on Less Educated Urban Workers”, Sociological Quarterly, 40:
199-216.
Espinosa, K., and Massey, D. (1999) “Undocumented Migration and the
Quantity and Quality of Social Capital”, (in) “Migration and Transnational Social
Spaces. Research in Ethnic Relations”, Pries, L. (ed.) Hants, Ashgayr Publishing.
Fernández-Huertas, J. (2008) “Wealth Constraints, Skill Prices or Networks:
What Determines Emigrant Selection?”, UFAE and IAE WP 741.08, Unitat de
Fonaments de l'Anàlisi Econòmica (UAB) and Institut d'Anàlisi Econòmica (CSIC).
Fernandez Kelly, P. (1995) “Social and Cultural Capital in the Urban Ghetto:
Implications for the Economic Sociology of Immigration”, (in) “Essays on Networks,
Ethnicity amd Entrepreunership”, Portes, A. (ed.), New York, Russel Sage Foundation.
Goel, D., and Lang, K. (2011) “Social Ties and the Job Search of Recent
Immigrants”.
http://www.econ.upf.edu/docs/seminars/lang.pdf
Goel, D., and Lang, K. (2012). “Social Ties and the Job Search of Recent
Immigrants”. http://people.bu.edu/lang/network.pdf
Granovetter, M. (1973) “The Strength of Weak Ties”, American Journal of
Sociology, 78(6): 1360-1380.
Granovetter, M. (1974; 1995) “Getting a Job: A Study of Contacts and
Careers”, first edition, Harvard University Press, second edition, The University of
Chicago Press, Chicago, Illinois.
Hanson, G. (2006) “Illegal migration from Mexico to the United States”,
Journal of Economic Literature, 44(4): 869-924.
Heckman, J. (1979) “Sample Selection Bias as a Specification Error”,
Econometrica, 47(1): 153-161.
Holst, E., and Schrooten, M. (2006). “Migration and Money: What determines
Remittances? Evidence from Germany”, DIW Berlin, DP 566.
Ioannides, Y., and Loury, L. (2004) “Job Information Networks, Neighborhood
Effects and Inequality”, Journal of Economic Literature, 42(4): 1056-1093.

41
 Izquierdo, M., Lacuesta, A.; and Vegas, R. (2009) “Assimilation of Immigrants
in Spain: a Longitudinal Analysis”, Labour Economics, 16(6): 669-678.
Jasso, G., and Rosenzweig, M. (1995) “Do Immigrants Screened for Skills Do
Better than Family Reunification Immigrants?”, International Migration Review, 29:
85-111.
Kahanec, M., and Mendola, M. (2008) “Social Determinants of Labor Market
Status of Ethnic Minorities in Britain”, Centro Studi Luca d’Agliano, No. 253.
Jackson, M. (2008) “Social and Economic Networks”, Princeton University
Press.
Klein, R., and Spady, R. (1993) “An efficient semiparametric estimator of the
binary response model”, Econometrica, 61(2): 387-421.
Koenker, R., and Basset, G. (1978) “Regression Quantiles”, Econometrica, 46:
33-50.
Loury, L. (2006) “Some Contacts Are More Equal Than Others: Earnings and
Job Information Networks”, Journal of Labor Economics, 24(2): 299-318.
Llul, J. (2008) “The impacts of immigration on productivity”, CEMFI WP 0802.
Mahuteau, S., and Junankar, P.N. (2008) “Do Migrants get Good Jobs in
Australia? The Role of Ethnic Networks in Job Search”, The Economic Record,
84(Special Issue): S115-S130.
Manski, C. (1995) “Identification Problems in the Social Sciences”, Harvard
University Press.
Manski, C. (2003) “Partial Identification of Probability Distributions”,
Springer-Verlag.
Montgomery, J. (1991) “Social Networks and Labor-Market Outcomes: Toward
and Economic Analysis”, American Economic Review, 81(5): 1408-1418.
Montgomery, J. (1992) “Job Search and Network Composition: Implications of
the Strength-of-Weak-Ties Hypothesis”, American Sociology Review, 57(5): 586-596.
Munshi, K. (2003) “Networks in the modern Economy: Mexican migrants in the
US labor market”, Quarterly Journal of Economics, 549-599.
Nee, V., and Sanders, J. (2001) “Understanding the diversity of immigrant
incorporation: a forms-of-capital model”, Ethnic and Racial Studies, 24(3): 386-411.
OECD (2010) “International Migration Outlook”
Ottaviano, G., and Peri, G. (2006) “The Economic Value of Cultural Diversity:
Evidence from U.S. Cities”, Journal of Economic Geography, 6: 9-44.

42
 Patacchini, E., and Zenou, Y. (2008) “Ethnic networks and employment
outcomes”. IZA DP No. 331.
Pellizzari, M. (2010) “Do Friends and Relatives Really Help in Getting a Good
Job?”, Industrial & Labor Relations Review, 65(3), article 7.
Rees, A. (1966) “Information Networks in Labor Markets”, American Economic
Review, 56(1-2): 559-566.
Reher, D. (2008): Informe Encuesta Nacional de Inmigrantes (ENI-07), INE, DT
2-08.
Rodríguez-Planas, N., and Vegas, R. (2012a) “Moroccans’ Assimilation in
Spain: Family-Based versus Labor-Based Migration”, IZA DP No. 6368.
Rodríguez-Planas, N., and Vegas, R. (2012b) “Moroccans’, Ecuadorians’ and
Romanians’ Assimilation in Spain” IZA DP No. 6542.
Simón, H., Ramos, R.; and Sanromá, E. (2011) “Occupational Mobility of
Immigrants in a Low Skilled Economy: The Spanish Case”, IZA DP No. 5581.
Smith, S. (2000) “Mobilizing social resources: Race, ethnic, and gender
differences in social capital and persisting wage inequalities”, Sociological Quarterly,
41(4): 509-537.
Stark, O., and Wang, Y. (2002) “Migration Dynamics”, Economic Letters, 76(2):
159-164.
Topa, G. (2001) “Social Interactions, Local Spillovers and Unemployment”,
Review of Economic Studies, 68: 261-295.
Veira, A., and Stanek, M. (2009) “Occupational transitions and social mobility
at migration to Spain”, Grupo de Estudios “Población y Sociedad” (Universidad
Complutense de Madrid), DT No. 4 (III).
Veira, A., Stanek, M.; and Cachón, L. (2011) “Los determinantes de la
concentración étnica en el mercado laboral español”, Revista Internacional de
Sociología, 69(M1): 219-242.
Wahba, J., and Zenou, Y. (2005) “Density, social networks and job search
methods: theory and application to Egypt”, Journal of Development Economics 78: 443-
473.
Zenou, Y. (2009) “Urban Labor Economics”. New York: Cambridge University
Press.

43
TABLES AND FIGURES

Graph 1. Evolution of the new immigrants in Spain by year of arrival

500,000

450,000

400,000

350,000

300,000

250,000

200,000

150,000

100,000

50,000

Immigrants (millions)
Source: ENI (2007)

44
 Table 1 Descriptive statistics for socio-demographic variables
Sample Subsample Excluded
Variables
(1) (2) (3)
Female 0.57 0.53 0.79
Age (years) 34 34 32
Years since arrival 4.11 4.31 2.69
Married 0.52 0.50 0.64
Number of children 1.27 1.25 1.39
Residence authorization 0.75 0.76 0.67
Education
Primary level 0.19 0.18 0.22
Secondary level 0.55 0.57 0.44
Tertiary level 0.26 0.24 0.35
Speaks spanish 0.76 0.80 0.47
Region of origin
Western Europe 0.08 0.07 0.14
Eastern Europe 0.25 0.26 0.16
Latin America 0.49 0.52 0.31
North Africa 0.13 0.10 0.30
Asia 0.02 0.02 0.03
Rest of the world 0.03 0.03 0.05
Migration between municipalities. Frecuency (%)
1. Never moved 0.29 0.24 0.63
2. Moved once 0.35 0.37 0.22
3. More than one 0.36 0.39 0.15
Motives for migration 1
Labor motives 0.64 0.69 0.27
Family regrouping 0.27 0.22 0.59
Social networks
Contacts at arrival (Close ties) 0.83 0.83 0.86
Social participation (exclusive for immigrants) 0.06 0.06 0.06
Social participation (mixed organization) 0.10 0.10 0.09
Remmitances 0.56 0.61 0.21
Observations 7,377 6,432 945
1. More than one motive could be chosen. The options given in the ENI (2007) are: being
unemployed, search for a better job, jubilation, better quality of life, family regrouping, politic
motives, religious motives, others. Labor motives include being unemployed or search for a better
job.

45
Table 2 Descritpive Statistics. Labor outcome in Spain
Variable Freq.
Labour experience in Spain 87.19
Obs. 7,377

Dependent variables
Maintain first job 29.71
Actual job different first job 53.73
Unemployed 9.87
Inactive1 6.70
First job characteristics (dummy variables)
Job access mechanisms
Social Networks 0.70
Formal methods 0.29
Occupation
Manager 0.01
Professional 0.06
Paraprofessional2 0.27
3
Skilled workers 0.18
Unskilled workers 0.48
Sector of activity
Agriculture 0.16
Industry 0.08
Construction 0.15
Trade 0.07
Hotel sector 0.15
Transportation 0.03
Business services 0.06
Education- Health 0.06
Household activities 0.25
Public administration 0.00
Time before finding the first job (dummy variables)
Jobs proposal before migration 0.16
Less than one month 0.40
Between 1 and 3 months 0.19
Between 4 and 12 months 0.17
More than one year 0.04
Not known 0.03

Last occupation in the country of birth (dummy variables)

Manager 0.04
Professional 0.17
Paraprofessional2 0.27
3
Skilled workers 0.24
Unskilled workers 0.12
Never worked at origin 0.15
Obs. 6,432
1. Inactive excludes those immigrants thar are retired.
2. Includes administrative workers, comercial salers, personal service workers.
3. Includes qualified workers employed in industrial or agricultural activities.

46
Table 3 Observable differences across immigrants’ network strength
With CT No CT With WT No WT
Variables (1) (2) (3) (4)
Panel A
Socio-demographic characteristics
Female 0.55 0.44 0.51 0.54
Age (years) 34 35 35 34
Years since arrival 5.1 5.7 5.5 5.1
Married 0.50 0.46 0.50 0.50
Number of children 1.25 1.29 1.19 1.26
Residence authorization 0.76 0.81 0.78 0.76
Education
Primary level 0.19 0.17 0.11 0.19
Secondary level 0.58 0.55 0.56 0.57
Tertiary level 0.24 0.27 0.32 0.23
Speaks spanish 0.82 0.73 0.86 0.79
Region of origin
Western Europe 0.06 0.11 0.13 0.07
Eastern Europe 0.26 0.27 0.18 0.27
Latin America 0.54 0.39 0.57 0.51
North Africa 0.09 0.13 0.06 0.10
Asia 0.02 0.03 0.02 0.02
Rest of the world 0.02 0.07 0.04 0.03
Last occupation in the country of birth (dummy variables)
Manager 0.04 0.05 0.25 0.16
Professional 0.17 0.17 0.26 0.27
Paraprofessional2 0.27 0.27 0.20 0.25
3
Skilled workers 0.24 0.25 0.11 0.09
Unskilled workers 0.12 0.14 0.09 0.13
Never worked at origin 0.15 0.13 0.13 0.15
Panel B
Labor market status
Keep job 0.30 0.29 0.27 0.30
Change job 0.53 0.55 0.59 0.53
Unemployed 0.10 0.10 0.08 0.10
Inactive 0.07 0.05 0.07 0.07
Observations 5344 1088 656 5776

47
Table 4 Probability of labor experience in Spain. Logit regression
Variable Coefficient SE
Key independent variables
Close ties 0.679*** (0.173)
Social participation. Non mixed organizations 0.076 (0.246)
Social participations. Mixed organizations -0.041 (0.208)
Migrant proportion 0.563 (0.704)

Motives for migration (reference: other motives)

Labor 1.260*** (0.144)
Family regrouping -0.683*** (0.136)
Other controls
Female -0.915*** (0.138)
Age 0.215*** (0.045)
Age^2 -0.003*** (0.001)
Married -0.472*** (0.134)
No. children Spain -0.262*** (0.059)
No. children origin 0.100 (0.086)
Years since arrival (years) 0.387*** (0.036)
Residence authorization 0.901*** (0.143)
Educational attainment (reference: Primary level or less)
Secondary level 0.128 (0.170)
Terciary level -0.251 (0.182)
Speaks spanish 0.933*** (0.157)
Region of origin (reference: Western Europe)
Eastern Europe 1.407*** (0.247)
Latin America 0.776*** (0.214)
North Africa -0.269 (0.241)
Asia 0.053 (0.415)
Rest of the world -0.221 (0.371)
Region of destination (reference: Madrid)
Andalucía -0.389 (0.283)
Aragon -0.778*** (0.278)
Asturias -0.409 (0.417)
Balears -0.623** (0.259)
Canarias -0.715** (0.287)
Cantabria -0.802*** (0.301)
Castilla Leon -0.772** (0.312)
Castilla la Mancha -0.919*** (0.286)
Catalonia -0.799*** (0.205)
Valencian Community -0.678*** (0.232)
Extremadura -1.029*** (0.310)
Galicia -1.178*** (0.348)
Murcia -0.247 (0.239)
Navarra -0.518** (0.253)
Basque Country -0.749** (0.352)
La Rioja -0.442 (0.304)
Internal mobility (reference: never moved)
Moved once 0.768*** (0.153)
More than once 1.050*** (0.167)
Activity before migration
Unemployed at origin 0.251 (0.205)
Student at origin 0.151 (0.226)
Last occupation in the origin country (reference: unskilled worker)
Manager -0.405 (0.363)
Professional -0.501* (0.267)
Paraprofessional -0.511** (0.252)
Skilled workers -0.400 (0.276)
Never worked -1.272*** (0.253)
Constant -4.090*** (0.841)
Observations 7,377
Pseudo R2 0.404
Standard error s in parentheses
* p<0.1, ** p<0.05, *** p<0.01
48
 Table 5 Marginal effects
Keep job Different job Unemployed Inactive
(1) (2) (3) (4)
Independent interest variables
Close ties (CT) -0.089*** 0.052* 0.003 0.034**
Network job (NJ) -0.065* -0.001 0.046** 0.019
CT*NJ 0.086** -0.033 -0.013 -0.040**
Network size (NS) 0.218** -0.127 -0.039 -0.052
Weak ties (WT) -0.138*** 0.075 0.053 0.009
WT*years 0.024** -0.008 -0.015* -0.001
Time before finding the first job (less one month) 0.119*** -0.053 -0.032 -0.035*
Time before finding the first job (less one month)*NJ -0.094** 0.050 0.013 0.031

Other independent variables

Female -0.008 -0.124*** 0.036*** 0.096***
Age 0.004 0.000 -0.002 -0.003
Age^2 -0.000 -0.000 0.000 0.000
Married 0.002 -0.005 -0.004 0.007
No.of children origin -0.013 0.011 0.011** -0.009*
No. of children Spain -0.005 -0.007 0.002 0.009***
Residence authorization -0.008 0.052*** -0.037*** -0.007
Years since arrival (years) -0.048*** 0.040*** 0.003 0.005***
Educational attainment (Reference: primary level or less)
Secondary level -0.032 0.042* -0.009 -0.001
Terciary level 0.009 0.029 -0.034** -0.003
Spanish language -0.006 0.026 -0.021 0.001
Region of origin (Reference: Western Europe)
Eastern Europe -0.086*** 0.068* 0.025 -0.007
Latin America -0.082*** 0.084** 0.002 -0.005
North Africa -0.070* -0.008 0.068*** 0.009
Asia 0.098 0.080 -0.183*** 0.005
Rest of the world -0.045 -0.021 0.065** 0.000
Region of destination (Reference: Madrid)
Andalucía 0.058** -0.062* 0.007 -0.003
Aragon 0.010 -0.029 0.003 0.016
Asturias 0.094** -0.117*** 0.031 -0.008
Balears 0.044 -0.093*** 0.036* 0.013
Canarias 0.084** -0.068* -0.014 -0.001
Cantabria -0.037 -0.007 0.043 0.001
Castilla Leon 0.050 -0.065* 0.003 0.011
Castilla la Mancha 0.062* -0.095*** 0.035* -0.002
Catalonia 0.060** -0.040 -0.002 -0.019
Valencian Community 0.023 -0.040 0.027 -0.010
Extremadura 0.055 -0.104* 0.039 0.010
Galicia 0.075* -0.145*** 0.039 0.032**
Murcia 0.011 -0.057* 0.032* 0.013
Navarra 0.011 -0.008 0.018 -0.021
Basque Country 0.007 -0.051 0.043* 0.002
La Rioja 0.034 -0.014 -0.021 0.001

49
 Table 5 Marginal effects (cont.)
Keep job Different job Unemployed Inactive
(1) (2) (3) (4)

Mobility (Reference: never moved)

1. Moved once -0.125*** 0.129*** -0.002 -0.001
2. More than one -0.230*** 0.206*** 0.017 0.008
First occupation (Reference: unskilled occupation)
Manager 0.283*** -0.271*** -0.019 0.007
Professional 0.010 -0.039 0.003 0.026
Paraprofessional -0.048** 0.044* -0.001 0.005
Skilled workers 0.080*** -0.059** -0.025 0.004
Sector of activity (Reference: Agriculture)
Industry 0.168*** -0.173*** -0.003 0.007
Construction 0.158*** -0.183*** 0.017 0.009
Trade 0.198*** -0.214*** 0.020 -0.004
Hotel sector 0.135*** -0.189*** 0.030 0.024*
Transportation 0.165*** -0.138*** -0.021 -0.006
Firm services 0.243*** -0.224*** -0.003 -0.016
Education- Health 0.291*** -0.225*** -0.063** -0.004
Household activities 0.183*** -0.124*** -0.041** -0.018
Public administration 0.331*** -0.490*** 0.226*** -0.067
Mill's ratio -0.009 -0.018 0.011 0.016
Activity before migration
Unemployed at origin -0.019 -0.031 0.059*** -0.009
Student at origin -0.089*** 0.034 0.039** 0.017
Last occupation in the origin country (reference: unskilled worker)
Manager -0.128*** 0.100** 0.027 0.001
Professional -0.051* 0.063** -0.005 -0.007
Paraprofessional -0.048* 0.059** -0.003 -0.008
Skilled workers -0.030 0.061** -0.023 -0.009
Never worked at origin 0.086*** -0.073** 0.001 -0.014
Remittances -0.027* 0.058*** -0.006 -0.025***
Observations 6,432
* p<0.1, ** p<0.05, *** p<0.01

50
 Table 6 Robustness checks Marginal effects by gender
Women Men
Keep job Different job Unemployed Inactive Keep job Different job Unemployed Inactive
(1) (2) (3) (4) (1) (2) (3) (4)
Close ties (CT) -0.110** 0.026* 0.021 0.062** -0.098** 0.093** -0.003 0.008
Network job (NJ) -0.068* -0.014 0.032 0.051 -0.057* 0.008 0.056** -0.006
CT*NJ 0.067* 0.008 0.006 -0.081** 0.102* -0.065 -0.031 -0.006
Network size (NS) 0.122** -0.069 0.047 -0.101 0.283** -0.164 -0.114 -0.005
Weak ties (WT) -0.138* 0.112 0.072 -0.046 -0.112 0.059 0.034 0.018*

WT*years 0.017 -0.009 -0.016 0.007 0.024 -0.009 -0.015 -0.001

Time before finding the first
0.013 0.065 -0.041 -0.037 0.253*** -0.019 0.013 -0.246***
job (less one month)
Time before finding the first
0.004 -0.044 0.019 0.021 -0.205*** -0.016 -0.030 0.250***
job (less one month)*NJ
Observations 3429 3429 3429 3429 3003 3003 3003 3003
*Other controls used are the same as in Table 5.

Table 7 Robustness checks Marginal effects No skills variables

All Controls No controls at all
Keep job Different job Unemployed Inactive Keep job Different job Unemployed Inactive
(1) (2) (3) (4) (1) (2) (3) (4)
Independent interest variables
Close ties (CT) -0.089*** 0.052* 0.003 0.034** -0.085*** 0.052* -0.005 0.038**
Network job (NJ) -0.065* -0.001 0.046** 0.019 -0.058 -0.003 0.041* 0.020
CT*NJ 0.086** -0.033 -0.013 -0.040** 0.082** -0.036 -0.005 -0.041**
Network size (NS) 0.218** -0.127 -0.039 -0.052 0.241*** -0.146 -0.045 -0.049
Weak ties (WT) -0.138*** 0.075 0.053 0.009 -0.141*** 0.072 0.059* 0.010
WT*years 0.024** -0.008 -0.015* -0.001 0.024** -0.007 -0.016** -0.001
Time before finding the first job
(less one month) 0.119*** -0.053 -0.032 -0.035* 0.121*** -0.051 -0.034 -0.036*
Time before finding the first job
(less one month)*NJ -0.094** 0.050 0.013 0.031 -0.099** 0.054 0.012 0.033
Observations 6432 6432
* p<0.1, ** p<0.05, *** p<0.01
**Other controls used are the same as in Table 5, but excluding skill variables (educational level, proficiency in the Spanish language, labor
status before migration, and last occupation in the country of birth).

51
Table 8 Robustness checks Marginal effects (including motives for migrating)
Keep job Different job Unemployed Inactive
(1) (2) (3) (4)
Independent interest variables
Close ties (CT) -0.091*** 0.057* 0.000 0.034**
Network job (NJ) -0.063* -0.002 0.046** 0.020
CT*NJ 0.084** -0.030 -0.014 -0.040**
Network size (NS) 0.229** -0.143 -0.042 -0.044
Weak ties (WT) -0.153*** 0.073 0.071* 0.009
WT*years 0.023** -0.007 -0.016** -0.001
Time before finding the first job
0.116*** -0.051 -0.032 -0.034*
(less one month)
Time before finding the first job
-0.094** 0.047 0.013 0.034
(less one month)*NJ
Observations 6432 6432 6432 6432
* p<0.1, ** p<0.05, *** p<0.01
**Other controls used are the same as in Table 5 and adding motives for migration.

Table 9 Test for normality of the residuals

Lagrange Multiplier Test for Normality after Probit
Women Men
Chi2(2) = 5.1442 Chi2(2) = 2.7341
Prob > chi2 = 0.0764 Prob > chi2 = 0.2549

Ho: Normality
Ha: No Normality

Table 10 Wage regression. Women versus men

Dependent variable: ln(wages per hour)

QR 25 QR50 QR75 OLS
Women
Network job (NJ) -0.037* -0.034* -0.048 -0.055*
Close ties (CT) 0.015 -0.002 -0.031 -0.012
Network size (NS) -0.147 -0.194** 0.094 -0.087
Weak ties (WT) 0.074 -0.032 0.024 0.064
Observations 912 912 912 912
Men
Network job (NJ) -0.113*** -0.108*** -0.117*** -0.235***
Close ties (CT) -0.084*** -0.097*** -0.115*** -0.210***
CT*NJ 0.034* 0.078*** 0.087*** 0.195***
Network size (NS) -0.009 -0.195*** -0.105** 0.126
Weak ties (WT) -0.080*** -0.080*** -0.031 -0.096
Observations 862 862 862 862
* p<0.1, ** p<0.05, *** p<0.01
**Tables A.8 and A.9 in the Appendix present the wage regressions for full specifications for women
and men respectively.

52
APPENDIX

Table A.1 Database elaboration

Dropped observations Total
Total sample 15,441
Missing Age/ not recorded 41
Missing Year of arrival/ not recorded 212
Missing Years of residence/ not recorded 1
Subtotal 15,187
Subsample - Data restricted to:
Year of arrival>1996 5,226
Age between 16 and 65 years 242
Age at arrival (between 14 and 56 years) 411
Not finish studies in Spain 595
Missings 241
Country before migrarion: country of birth 1,095
Inactives (retired) / Missings 241
Subtotal 7,377
Without labour experience in Spain 945
Final Subsample 6,432

53
 Table A.2 Definition of independent variables
Female 1 if respondent is a woman; 0 otherwise
Man 1 if respondent is a man; 0 otherwise
Age Age in years
Age^2 Age square
Years since arrival Years
Married 1 if the respondent is married; 0 otherwise
Number of children Number of daughters and sons
Residence authorization 1 if the respondent declares having any of the following documents:
Permanent residency authorisation; temporary residency authorisation, EU
residence permit (except in the case of Romanian and Bulgarian workers who,
despite being EU citizens could not become legally contracted workers in Spain
temporarily at the time of the survey); refugee status or assylum application.
This cathegory also includes immigrants whose nationailty is Spanish, from
other EU member state (excluding Bulgaria and Romania) or from non-EU
members of thr Free Trade Association (i.e., Lichtenstein, Iceland, Switzerland
and Norway);
0 otherwise
Education level attained (dummies variables)
Primary level 1 if the respondent has primary level attained or less; 0 otherwise
Secondary level 1 if the respondent has secondary level complete or incomeplete; 0 otherwise
Tertiary level 1 if the respondent has tertiary level complete or incomeplete; 0 otherwise
Language
Speaks spanish 1 if respondent declares having spanish as her mother tongue or, if she states
can speak Spanish ‘well‘ or ‘very well‘; 0 otherwise
Region of origin
Western Europe 1 if country of birth is in Western Europe; 0 otherwise
Eastern Europe 1 if country of birth is in Eastern Europe; 0 otherwise
Latin America 1 if country of birth is in Latin America; 0 otherwise
North Africa 1 if country of birth is in North Africa; 0 otherwise
Asia 1 if country of birth is in Asia; 0 otherwise
Rest of the world 1 if country of birth is in Oceania, rest of Africa, ; 0 otherwise

54
 Table A.2 (Cont.)
Migration between municipalities. Frecuency (%)
1. Never moved 1 if respondent declares have lived in the same municipality since arrival; 0
otherwise
2. Moved once 1 if respondent declares have lived in two different municipalities; 0 otherwise
1 if respondent declares have lived in more than two different municipalities; 0
3. More than one otherwise
Motives for migration
1 if respondent declares moved because being unemployed in the country of
Labor origin or declares looking for a better job; 0 otherwise
Family regrouping 1 if respondent declares family regrouping; 0 otherwise
Social networks
Contacts at arrival (Close ties) 1 if respondent has contacts at arrival; 0 otherwise
Social participation in organizations exclusive for 1 if respondent participates in:
immigrants immigrant assistance organizations specifically to foreigners,
associations and sports clubs specifically targeting foreigners,
educational and cultural groups specifically targeting foreigners,
religious organizations and groups specifically targeting foreigners,
other groups specifically targeting foreigners;
0 otherwise
Social participation in mixed organizations 1 if respondent participates in:
NGO´s
Political organizations, unions, or neighborhood activities,
Religious groups,
Sport clubs, educational and cultural groups,
Other social groups;
0 otherwise
Migrant proportion Proportion of immigrants of the same country of birht living in the same
Autonomous Community on the total immigrant population in the Autonomous
Community (%)
Network job 1 if respondent has found the job through family and friends; 0 otherwise
Formal job 1 if respondent has found the job through State and private employment
agencies, newspapers´ advertisements, union hiring halls as well as school and
college placement services; 0 otherwise

55
 Table A.2 (Cont.)
Sector of activity
Agriculture 1 if respondent' first job is in:
Agriculture, Hunting, and Forestry
Fishing,
Minning;
0 otherwise
Industry 1 if respondent' first job is in:
Manufacture industries,
Production and distribution of electricity, gas and water;
0 otherwise
Construction 1 if respondent' first job is in Construction;
0 otherwise
Trade 1 if respondent' first job is in: Trade, repair of motor vehicles and motorcycles
and personal articles and electronic products for household;
0 otherwise
Hotel sector 1 if respondent' first job is in: Hotel sector;
0 otherwise
Transportation 1 if respondent' first job is in: Transport, storage and communications;
0 otherwise
Firm services 1 if respondent' first job is in:
Financial intermediation
Real estate, renting and business services;
0 otherwise
Education- Health 1 if respondent' first job is in:
Education,
Health and veterinary activities, social service,
Other social and community services, personal services;
0 otherwise
Household activities 1 if respondent' first job is in: Household activities;
0 otherwise
1 if respondent' first job is in: Public administration, defense and compulsory
Public administration
social security;
0 otherwise

56
 Table A.2 (Cont.)
Occupation
1 if respondent declares: Management of companies and public administrations;
Manager
0 otherwise
1 if respondent declares:
Professional
Technical and scientific professionals and intellectuals,
Technicians and associate professionals;
0 otherwise
Paraprofessional 1 if respondent declares:
Administrative workers,
Workers in catering services, personal services, protection
services, and comercial salers;
Skilled workers 0 otherwise
1 if respondent declares:
Qualified workers in fishing and agriculture activities.
Craftsmen and skilled manufacturing, construction, and mining, except plant
and machinery operators.
0 otherwise
Unskilled workers 1 if respondent declares: Unskilled occupation;
0 otherwise
Time before finding the first job Dummy variable equal to 1 if respondent declares spending less than a month
before finding the first job; 0 otherwise.
Remmitances Dummy variable equal to 1 if respondent declares sending remmitances to the
country of brith; 0 otherwise.
Notes: 1. Weak ties refer to immigrants participating in mixed organizations.
2. Migrant proportion is the network size.

57
 Table A.3 Descriptive statistics. Socio-demographic characteristics by region of origin
Western Latin Eastern North Rest of
Asia Total
Europe America Europe Africa the world
Variables
Female 0.47 0.59 0.57 0.28 0.35 0.27 0.53
Age 36 34 33 33 33 33 34
Year of arrival 2002 2002 2002 2001 2001 2001 2002
Years since arrival 4 4 4 5 5 5 4
Married 0.37 0.47 0.54 0.59 0.59 0.56 0.50
Number of children 0.89 1.49 1.02 0.98 1.09 1.15 1.25
No. children origin 0.56 0.43 0.76 0.36 0.61 1.21 0.41
No. children Spain 1.25 1.19 1.28 1.98 1.32 1.03 0.86
Residence authorization 1.00 0.74 0.70 0.86 0.90 0.85 0.76
Educational level attained (dummies variables)
Primary level 0.14 0.19 0.13 0.28 0.31 0.32 0.18
Secondary level 0.50 0.58 0.67 0.39 0.39 0.39 0.57
Tertiary level 0.36 0.23 0.20 0.33 0.31 0.29 0.24
Speaks spanish 0.64 0.98 0.64 0.55 0.39 0.50 0.80
Migration between municipalities. Frecuency (%)
1. Never moved 40.69 21.12 23.57 27.23 32.12 25.41 24.12
2. Moved once 29.65 39.99 37.59 30.99 28.47 34.05 37.31
3. More than one 29.65 38.88 29.65 41.78 39.42 40.54 38.56
1
Motives for migration
Labor motives 0.13 0.51 0.68 0.28 0.22 0.50 0.64
Family regrouping 0.40 0.31 0.27 0.39 0.48 0.30 0.27
Social networks
Contacts at arrival (Close ties) 0.75 0.87 0.82 0.78 0.76 0.58 0.83
Social participation (exclusive for immigrants) 0.04 0.06 0.05 0.07 0.14 0.17 0.06
Social participation (mixed organization) 0.18 0.11 0.07 0.06 0.10 0.14 0.10
Frecuency (region of birth) Subsample (%) 7.18 51.82 26.06 9.93 2.13 2.88 100.00
Observations 3,644 6,059 2,386 2,018 437 643 6432

58
Table A.4 Occupational mobility between actual occupation and last occupation in
Last the country of origin
occupation Actual occupation in Spain
Region
in the Manager Professional
Paraprofessional
Qualified workers
Unskilled workers Total
Manager 11.2 9.1 27.5 16.7 35.5 100
Professional 1.7 20.0 30.5 11.2 36.7 100
Paraprofessional 0.9 3.5 38.7 10.0 46.9 100
Total sample
Qualified workers 0.1 1.2 11.4 39.5 47.7 100
Unskilled workers 0.1 1.0 16.8 13.0 69.1 100
Total 1.3 5.9 25.3 19.7 47.8 100
Manager 59.5 10.8 18.9 5.4 5.4 100
Professional 6.9 60.3 25.2 3.1 4.6 100
Western Paraprofessional 7.0 17.4 55.7 4.4 15.7 100
Europe Qualified workers 1.3 6.7 17.3 62.7 12.0 100
Unskilled workers 0.0 13.3 23.3 16.7 46.7 100
Total 10.3 28.9 32.0 16.2 12.6 100
Manager 3.4 9.0 32.8 18.1 36.7 100
Professional 1.0 14.8 36.3 12.0 35.8 100
Latin Paraprofessional 0.4 2.4 43.5 9.3 44.4 100
America Qualified workers 0.0 1.4 12.9 41.3 44.4 100
Unskilled workers 0.3 0.3 23.6 12.5 63.4 100
Total 0.6 5.0 31.5 18.2 44.7 100
Manager 8.1 8.1 16.2 13.5 54.1 100
Professional 0.5 8.4 20.3 10.9 59.9 100
Eastern Paraprofessional 0.8 1.1 24.6 12.2 61.4 100
Europe Qualified workers 0.2 0.4 9.4 39.0 51.2 100
Unskilled workers 0.0 0.6 9.5 11.8 78.1 100
Total 0.6 2.0 15.4 23.3 58.7 100
Manager 0.0 5.9 23.5 23.5 47.1 100
Professional 0.0 10.8 24.3 24.3 40.5 100
North Paraprofessional 0.0 5.2 18.2 15.6 61.0 100
Africa Qualified workers 0.0 0.6 7.8 28.5 63.1 100
Unskilled workers 0.0 0.8 5.7 14.5 79.0 100
Total 0.0 2.5 11.1 21.7 64.8 100
Manager 0.0 0.0 0.0 0.0 0.0 100
Professional 6.3 18.8 31.3 12.5 31.3 100
Paraprofessional 0.0 3.3 63.3 10.0 23.3 100
Asia
Qualified workers 0.0 4.0 36.0 16.0 44.0 100
Unskilled workers 0.0 5.9 11.8 5.9 76.5 100
Total 1.1 6.8 39.8 11.4 40.9 100
Manager 0.0 12.5 12.5 37.5 37.5 100
Professional 0.0 50.0 0.0 16.7 33.3 100
Rest of the Paraprofessional 0.0 7.7 23.1 12.8 56.4 100
world Qualified workers 0.0 0.0 7.1 40.5 52.4 100
Unskilled workers 0.0 0.0 12.8 18.0 69.2 100
Total 0.0 8.9 12.3 24.0 54.8 100

59
Table A.5 Multinomial regression (base outcome: employed in a different job)
Keep job Unemployed Inactive
Ommited: Employed in a different job
Key independent variables
Close ties (CT) -0.478*** -0.081 0.591*
Network job (NJ) -0.258 0.517* 0.402
CT*NJ 0.422* -0.082 -0.751*
Migrant proportion 1.165** -0.152 -0.847
Weak ties (WT) -0.731** 0.437 0.002
WT*years 0.114* -0.149 0.011
Time before finding the first job (less one month) 0.608*** -0.247 -0.548
Time before finding the first job (less one month)*NJ -0.498** 0.036 0.488

Remmitances -0.216** -0.220* -0.538***

Control variables
Female 0.238** 0.713*** 2.236***
Age 0.017 -0.019 -0.068
Age^2 0.000 0.000 0.001
Married 0.006 -0.019 0.089
No. Children in Spain -0.006 0.044 0.203***
No. Children in origin -0.076 0.091 -0.204*
Years since arrival (years) -0.291*** -0.059* -0.009
Residence authorization -0.154 -0.525*** -0.284*
Educational attained (Reference: primary level or less)
Secondary level -0.227* -0.194 -0.131
Terciary level -0.032 -0.447** -0.142
Spanish language -0.086 -0.291* -0.052
Region of origin (Reference: Western Europe)
Eastern Europe -0.516** 0.133 -0.363
Latin America -0.531*** -0.162 -0.323
North Africa -0.270 0.780** 0.168
Asia 0.199 -2.228*** -0.114
Rest of the world -0.121 0.761* 0.120
Region of destination (Reference: Madrid)
Andalucía 0.366** 0.225 0.076
Aragon 0.097 0.097 0.382
Asturias 0.648*** 0.607 0.144
Balears 0.386** 0.610** 0.473*
Canarias 0.486** -0.005 0.115
Cantabria -0.133 0.478 -0.005
Castilla Leon 0.343 0.194 0.393
Castilla la Mancha 0.460** 0.612** 0.157
Catalonia 0.330** 0.065 -0.309
Valencian Community 0.183 0.393* -0.107
Extremadura 0.447 0.677* 0.406
Galicia 0.627** 0.769** 0.971***
Murcia 0.175 0.489** 0.435
Navarra 0.058 0.220 -0.420
Basque Country 0.137 0.593** 0.130
La Rioja 0.165 -0.196 0.035

60
 Table A.5 Multinomial regression (cont.)
Keep job Unemployed Inactive
Mobility (Reference: never moved)
1. Moved once -0.804*** -0.312* -0.336
2. More than one -1.400*** -0.273 -0.317
First occupation (Reference: unskilled occupation)
Manager 1.742*** 0.402 0.736
Professional 0.124 0.138 0.581
Paraprofessional -0.295** -0.108 -0.018
Skiled workers 0.456*** -0.143 0.200
Activity sector (Reference: Agriculture)
Industry 1.069*** 0.361 0.562
Construction 1.049*** 0.607** 0.603
Trade 1.284*** 0.712** 0.406
Hotel sector 0.975*** 0.772*** 0.972***
Transportation 0.974*** 0.074 0.203
Firm services 1.492*** 0.476 0.200
Education- Health 1.686*** -0.182 0.453
Household activities 1.027*** -0.175 -0.057
Public administration 2.447*** 3.629*** -0.332
Mill's ratio 0.018 0.201 0.377
Activity before migration
Unemployed at origin -0.001 0.716*** -0.085
Student at origin -0.429** 0.348 0.311
Last occupation in the origin country (reference: unskilled worker)
Manager -0.742*** 0.069 -0.219
Professional -0.348** -0.192 -0.296
Paraprofessional -0.329** -0.170 -0.308
Skilled workers -0.256* -0.392* -0.315
Never worked 0.511*** 0.179 -0.141
Remittances -0.246** -0.208* -0.653***
Constant 0.758 -1.071 -1.687
Observations 6432
Pseudo R2 0.159
* p<0.1, ** p<0.05, *** p<0.01

61
 Table A.6 Probability of keeping the first job. Semiparametric model. Women
Coef SE
Key independent variables
Network job (NJ) -0.575*** (0.155)
Close ties (CT) -0.201* (0.134)
CT*NJ 0.184 (0.146)
Migrant proportion 4.372*** (0.662)
Weak ties (WT) -0.011 (0.119)

Other controls
Age 0.185*** (0.039)
Age^2 -0.001*** (0.000)
Married -0.110 (0.085)
No. Children origin -0.145** (0.062)
No. Children Spain -0.166*** (0.052)
Residence authorization -0.374*** (0.082)
Years since arrival (years) -1.678*** (0.254)
Educational level attained (Reference: primary level or less)
Secondary level 0.775*** (0.154)
Terciary level 0.720*** (0.150)
Spanish language
Region of origin (Reference: Western Europe)
Eastern Europe -0.524*** (0.119)
Latin America -0.708*** (0.115)
North Africa -2.16*** (0.157)
Asia 4.123*** (0.226)
Rest of the world 5.443*** (0.244)
Region of destination (reference: Madrid)
Andalucía 0.154 (0.114)
Aragon 0.188 (0.137)
Asturias 0.305* (0.158)
Balears 0.088 (0.112)
Canarias 0.016 (0.159)
Cantabria -0.062 (0.159)
Castilla Leon 0.055 (0.139)
Castilla la Mancha 0.044 (0.133)
Catalonia 0.175* (0.100)
Valencian Community -0.007 (0.112)
Extremadura 0.023 (0.179)
Galicia 0.190 (0.158)
Murcia -0.151 (0.119)
Navarra -0.016 (0.115)
Basque Country -0.015 (0.141)
La Rioja 0.064 (0.134)

62
 Table A.6 Probability of keeping the first job. Semiparametric model. Women
(cont.)
Coef SE
First occupation (Reference: unskilled occupation)
Manager 13.682*** (2.040)
Professional 0.014 (0.136)
Paraprofessional -1.230*** (0.226)
Skilled workers -0.945*** (0.261)
Activity sector (Reference: Agriculture)
Industry -1.211*** (0.235)
Construction -1.689*** (0.321)
Trade 2.119*** (0.414)
Hotel sector 0.651*** (0.133)
Transportation 1.719*** (0.443)
Firm services 2.124*** (0.350)
Education- Health 2.089*** (0.320)
Household activities 1.539*** (0.280)
Public administration 0.950** (0.420)
Activity before migration
Unemployed at origin 0.426*** (0.100)
Student at origin -0.284*** (0.106)
Last occupation in the origin country (reference: unskilled worker)
Manager -4.502*** (0.499)
Professional -0.990*** (0.146)
Paraprofessional -1.436*** (0.179)
Skilled workers -1.524*** (0.201)
Never worked at origin 0.706*** (0.165)
Remittances -1.268*** (0.195)
Observations 3429
* p<0.1, ** p<0.05, *** p<0.01

63
Table A.7 Probability of keeping the first job. Probit model. Men
Coef SE
Key independent variables
Network job (NJ) -0.240* (0.125)
Close ties (CT) -0.209* (0.111)
CT*NJ 0.201 (0.143)
Migrant proportion 0.618* (0.323)
Weak ties (WT) -0.448** (0.194)
WT*years 0.086** (0.039)

Other controls
Age -0.013 (0.023)
Age^2 0.000 (0.000)
Married 0.012 (0.063)
Number of children -0.015 (0.028)
Residence authorization 0.147** (0.071)
Years since arrival (years) -0.178*** (0.016)
Maximum educational level attained (Reference: primary level or less)
Secondary level -0.159** (0.073)
Terciary level -0.086 (0.088)
Spanish language -0.072 (0.076)
Region of origin (Reference: Western Europe)
Eastern Europe -0.234* (0.128)
Latin America -0.204* (0.119)
North Africa -0.041 (0.137)
Asia 0.254 (0.193)
Rest of the world -0.161 (0.174)
Region of destination (reference: Madrid)
Andalucía 0.263* (0.135)
Aragon -0.167 (0.157)
Asturias 0.279 (0.214)
Balears 0.108 (0.133)
Canarias 0.445*** (0.154)
Cantabria -0.172 (0.199)
Castilla Leon 0.162 (0.158)
Castilla la Mancha 0.306** (0.138)
Catalonia 0.202* (0.110)
Valencian Community 0.128 (0.122)
Extremadura 0.069 (0.224)
Galicia 0.055 (0.209)
Murcia 0.060 (0.125)
Navarra 0.060 (0.129)
Basque Country 0.043 (0.172)
La Rioja 0.071 (0.160)

64
Table A.7 Probability of keeping the first job. Probit model. Men
Coef SE
First occupation (Reference: unskilled occupation)
Manager 1.574*** (0.251)
Professional 0.663*** (0.145)
Paraprofessional 0.457*** (0.137)
Skilled workers 0.447*** (0.071)
Activity sector (Reference: Agriculture)
Industry 0.523*** (0.109)
Construction 0.474*** (0.084)
Trade 0.427*** (0.124)
Hotel sector -0.213 (0.165)
Transportation 0.164 (0.162)
Firm services 0.339** (0.156)
Education- Health 0.600*** (0.174)
Household activities -0.569* (0.306)
Public administration 1.313*** (0.420)
Mobility (Reference: never moved)
1. Moved once -0.510*** (0.070)
2. More than one -0.880*** (0.074)
Time before finding the first job (less one month) 0.258*** (0.071)
Activity before migration
Unemployed at origin -0.090 (0.081)
Student at origin -0.317** (0.125)
Last occupation in the origin country (reference: unskilled worker)
Manager -0.341** (0.155)
Professional -0.083 (0.117)
Paraprofessional -0.132 (0.103)
Skilled workers -0.004 (0.087)
Never worked at origin 0.273** (0.129)
Remittances -0.163*** (0.061)
Constant 0.653 (0.457)
Observations 3003
Pseudo R2 0.224
* p<0.1, ** p<0.05, *** p<0.01

65
 Table A.8 Wage regression. Women
QR 25 QR50 QR75 OLS
Independent interest variables
Network job (NJ) -0.037* -0.034* -0.048 -0.055*
Close ties (CT) 0.015 -0.002 -0.031 -0.012
Network size (NS) -0.147 -0.194** 0.094 -0.087
Weak ties (WT) 0.074 -0.032 0.024 0.064
Time before finding the first job 0.090 0.063*** 0.024 0.038

Other independent variables

Age -0.008 0.001 -0.002 -0.002
Age^2 0.000 -0.000 -0.000 0.000
Married -0.025 -0.031* -0.015 -0.017
No. Children in Spain 0.005 0.006 -0.001 0.011
No. Children in origin -0.009** -0.020** -0.032 -0.029*
Residence authorization 0.102 0.061*** 0.085 0.071**
Years since arrival (years) 0.007 0.005 -0.005 0.000
Educational level attained (Reference: primary level or less)
Secondary level -0.075 -0.077*** -0.084 -0.069
Terciary level 0.080 0.060** 0.093 -0.066
Spanish language -0.065 -0.006 0.024 0.014
Region of origin (Reference: Western Europe)
Eastern Europe -0.021 -0.079** -0.101 -0.087
Latin America -0.099 -0.198*** -0.173* -0.166***
North Africa -0.176 -0.223*** -0.240* -0.257***
Asia -0.179 -0.359*** -0.258 -0.247*
Rest of the world -0.369 -0.461*** -0.263 -0.366**
Region of destination (Reference: Madrid)
Andalucía -0.039 -0.010 0.091 -0.029
Aragon -0.098 0.020 0.170 0.006
Asturias 0.028 -0.012 0.167 -0.005
Balears 0.176 0.142*** 0.169* 0.124*
Canarias 0.105 0.106** 0.152 0.100
Cantabria -0.078 -0.133** 0.010 -0.136
Castilla Leon 0.008 -0.083* 0.040 -0.053
Castilla la Mancha -0.074 -0.042 -0.003 -0.150**
Catalonia 0.199 0.166*** 0.194** 0.148***
Valencian Community -0.004 0.015 0.031 -0.071
Extremadura 0.042 -0.069 -0.071 -0.085
Galicia -0.135 -0.079 -0.010 -0.109
Murcia 0.026 0.005 0.026 -0.016
Navarra 0.154 0.107*** 0.153 0.090
Basque Country 0.031 0.116*** 0.167 0.060
La Rioja 0.154 0.076* 0.070 0.049

66
 Table A.8 Wage regression. Women (cont.)
QR 25 QR50 QR75 OLS
First occupation (Reference: unskilled occupation)
Manager 0.176 0.141** 0.615*** 0.311**
Professional 0.198 0.264*** 0.406*** 0.248***
Paraprofessional -0.026 -0.020 -0.070 -0.056
Skilled workers -0.166 -0.177 -0.118 -0.142
Sector of activity (Reference: Agriculture)
Industry -0.054 -0.083 -0.027 -0.081
Construction 0.047 -0.037 -0.104 0.002
Trade 0.124 0.007 0.025 0.070
Hotel sector 0.040 -0.031 0.022 0.015
Transportation -0.027 -0.110 0.280 0.102
Firm services 0.013 -0.039 0.209 0.087
Education- Health -0.005 -0.021 0.071 0.012
Household activities -0.173* -0.166*** -0.028 -0.139*
Public administration -0.040 -0.123 -0.178 -0.072
Activity before migration
Unemployed at origin -0.065 -0.086*** -0.066 -0.044
Student at origin -0.126 -0.004 0.008 -0.063
Last occupation in the origin country (reference: unskilled worker)
Manager -0.041 -0.049 -0.019 -0.026
Professional 0.153 0.154*** 0.157* 0.175***
Paraprofessional 0.031 0.075*** 0.045 0.077
Skilled workers -0.016 -0.008 -0.023 0.013
Never worked at origin 0.137 0.119*** 0.045 0.092
Mill's ratio 0.020** 0.019** 0.026* 0.015**
Constant 2.961 3.126*** 3.265*** 3.147***
Observations 912 912 912 912
* p<0.1, ** p<0.05, *** p<0.01

67
 Table A.9 Wage regression. Men
QR 25 QR50 QR75 OLS
Network job (NJ) -0.113*** -0.108*** -0.117*** -0.235***
Close ties (CT) -0.084*** -0.097*** -0.115*** -0.210***
CT*NJ 0.034* 0.078*** 0.087*** 0.195***
Migrant proportion -0.009 -0.195*** -0.105** 0.126
Weak ties (WT) -0.080*** -0.080*** -0.031 -0.096
WT*years 0.009** 0.020*** 0.015** 0.038*

Other independent variables

Age 0.013*** 0.019*** 0.006* 0.005
Age^2 -0.000*** -0.000*** -0.000 -0.000
Married 0.049*** 0.028*** 0.034*** 0.053*
No. of children Spain -0.008* -0.014*** -0.024*** -0.009
No. Children origin 0.010** 0.003*** -0.016*** 0.004
Residence authorization 0.055*** 0.050*** 0.052*** 0.092***
Years since arrival (years) -0.013*** 0.013*** -0.002 0.000
Educational level attained (Reference: primary level or less)
Secondary level 0.008 -0.016*** -0.015 -0.014
Terciary level 0.080*** 0.070*** 0.043*** 0.089**
Spanish language 0.047*** 0.065*** 0.038*** 0.033
Region of origin (Reference: Western Europe)
Eastern Europe -0.129*** -0.110*** -0.157*** -0.166***
Latin America -0.203*** -0.172*** -0.193*** -0.207***
North Africa -0.239*** -0.255*** -0.297*** -0.260***
Asia -0.162*** -0.218*** -0.259*** -0.219***
Rest of the world -0.294*** -0.251*** -0.312*** -0.344***
Region of destination (Reference: Madrid)
Andalucía 0.083*** 0.038*** 0.089*** 0.092
Aragon 0.014 -0.056*** -0.075*** -0.068
Asturias 0.031 -0.054*** 0.021 -0.005
Balears 0.082*** -0.026*** 0.068*** 0.100
Canarias 0.092*** -0.000*** 0.170*** 0.164**
Cantabria -0.204*** -0.187*** -0.137*** -0.260***
Castilla Leon -0.026 -0.162*** -0.102*** -0.043
Castilla la Mancha 0.038* 0.025*** 0.053** 0.035
Catalonia 0.156*** 0.094*** 0.122*** 0.124**
Valencian Community 0.014 0.068*** 0.026 0.043
Extremadura -0.219*** -0.105*** -0.152*** -0.034
Galicia -0.423*** -0.338*** -0.263*** -0.231**
Murcia 0.103*** 0.009*** -0.020 0.016
Navarra 0.132*** 0.089*** 0.091*** 0.089
Basque Country -0.015 -0.113*** -0.114*** -0.085
La Rioja 0.135*** 0.074*** 0.096*** 0.086

68
 Table A.9 Wage regression. Men (cont.)
QR 25 QR50 QR75 OLS
First occupation (Reference: unskilled
occupation)
Manager 0.596*** 0.398*** 0.651*** 0.751***
Professional 0.439*** 0.313*** 0.552*** 0.521***
Paraprofessional 0.087*** 0.088*** 0.110*** 0.159**
Skilled workers 0.156*** 0.059*** 0.127*** 0.197***
Sector of activity (Reference: Agriculture)
Industry 0.082*** 0.069*** 0.110*** 0.177**
Construction 0.181*** 0.211*** 0.226*** 0.238***
Trade 0.029 0.052*** 0.147*** 0.081
Hotel sector -0.050** 0.137*** -0.061** -0.129
Transportation 0.005 0.145*** 0.328*** 0.201**
Firm services 0.109*** 0.123*** 0.122*** 0.154**
Education- Health 0.030 0.130*** 0.216*** 0.232**
Household activities -0.195*** -0.157*** -0.372*** -0.277
Public administration 0.053 -0.031*** 0.101 0.012
Mobility (Reference: never moved)
1. Moved once -0.040*** -0.024*** -0.094*** -0.142***
2. More than one -0.077*** -0.030*** -0.144*** -0.225**
Time before finding the first job (less one month)
0.110*** 0.070*** 0.195*** 0.200***
Activity before migration
Unemployed at origin -0.144*** -0.150*** -0.135*** -0.116
Student at origin -0.173*** -0.094*** -0.124*** -0.130
Last occupation in the origin country (reference: unskilled worker)
Manager -0.118*** -0.046*** -0.012 -0.112
Professional 0.004 0.097*** 0.075*** 0.061
Paraprofessional -0.035 -0.001 -0.003 -0.041
Skilled workers -0.071*** 0.001*** 0.011 -0.044
Never worked at origin 0.036* 0.052*** 0.149*** 0.118*
Mill's ratio 0.201*** 0.025*** 0.216*** 0.402***
Constant 2.705*** 2.879*** 3.050*** 2.859***
Observations 862 862 862 862
* p<0.1, ** p<0.05, *** p<0.01

69
 METHODOLOGICAL APPENDIX

Buchinsky (1998)
Buchinsky (1998) was the first to consider the difficult problem of estimating
quantile regression in the presence of sample selection. We summarize this
methodology as if follows:

First, the reservation wage equation is considered as follows:

𝑦𝑖𝑅 = 𝑥𝑖′ 𝛼0 + 𝑣𝑖 (A.1)
The reservation wage of each individual is a function of her characteristics (𝑥1𝑖 ) in
addition to an idiosyncratic term.
The wage offer (𝑦𝑖∗ ) is assumed to be linearly dependent on a set of labor market
characteristics (𝑥2𝑖 ; a subset of 𝑥1𝑖 ) equation (2) in Buchinsky (1998) is:
𝑦𝑖∗ = 𝑐 + 𝑥2𝑖
′
𝛽0 + 𝜀𝑖 (A.2)
where 𝛽 is the vector of slope coefficients and 𝜀𝑖 is the error term.
In what follows, in order to simplify the notation, we omit the i subscript.

Equation (A.2) can be rewritten in the QR form considered by Koenker and Bassett
(1978) as:
𝑦 ∗ = 𝑐 + 𝑥2′ 𝛽𝜃 + 𝑢𝜃 ⁡⁡0 ≤ 𝜃 ≤ 1 (A.3)

where 𝑢𝜃 ≡ 𝑥2′ (𝛽0 − 𝛽𝜃 ) + 𝑢. It is assumed that the conditional quantile of 𝑦 ∗ ,

conditional on 𝑥2 , satisfies 𝑄𝜃 (𝑦|𝑥2 ) = 𝑥2′ 𝛽𝜃 so that 𝑄𝜃 (𝑢𝜃 |𝑥2 ) = 0

Since wage offer is observed only if it exceeds the reservation wage, we have 𝑦 = 𝑑 ∙
𝑦 ∗ = 𝑑(𝑥2′ 𝛽𝜃 + 𝑢𝜃 ), where 𝑑 ≡ 𝐼(𝑦 ∗ ≥ 𝑦 𝑅 ) and I(.) is the usual indicator function.

In the presence of this selection mechanism the conditional quantile of the observed
wage is given by
𝑄𝜃 (𝑦|𝑥2 ) = 𝑄𝜃 (𝑦 ∗ |𝑥2 , 𝑑 = 1) = 𝑥 ′ 𝛽𝜃 + 𝑄𝜃 (𝑢𝜃 |𝑥2 , 𝑑 = 1)

and in general 𝑄𝜃 (𝑢𝜃 |𝑥2 , 𝑑 = 1) ≠ 0. Nevertheless, if 𝑄𝜃 (𝑢𝜃 |𝑥2 , 𝑑 = 1) is only a

function of a known index f, then the observed wage equation can be written as

70
 𝑦 = 𝑥2′ 𝛽𝜃 + ℎ𝜃 (𝑓) + 𝜀𝜃 (A.4)
where ℎ𝜃 (𝑓) ≡ 𝑄𝜃 (𝑢𝜃 |𝑥1 , 𝑦 ∗ ≥ 𝑦 𝑅 |𝑥1 ) and, by construction, 𝑄(𝜀𝜃𝑖 |𝑥1 , 𝑑 = 1) = 0

The probability of working can be written as

𝑃𝑊 ≡ Pr(𝑦 ∗ ≥ 𝑦 𝑅 |𝑥1 ) = Pr⁡(𝜀 ≤ −𝑥𝑖′ 𝛼0 + 𝑥2′ 𝛽𝜃 )|𝑥1 ) (A.5)

In order to ensure that 𝑃𝑊 is only a function of 𝑓 and the representation of the equation
(A.4) holds, two additional assumptions are made by Buchinsky (1998). First, assumes
that 𝑤 ≡ (𝑣, 𝑢)′ has a continuous density; and second dependence of 𝑤 and 𝑥1 :
𝑔𝑤 (. |𝑥1 ) = 𝑔𝑤 (. |𝑓(𝑥1 ; 𝛾0 )
These assumptions on the joint distribution of these unobservables, both unconditionally
and conditional on 𝑥1 , that justifies the single-index representation.27 These
assumptions, while sufficient for the single-index representation, does not reveal the
functional form of h(.). Buchinsky (1998) suggests using the following series estimator
ℎ̂𝜃 (𝑥1 𝛾0 ) = 𝛿0 (𝜃) + 𝛿1 (𝜃)𝜆(𝑥1 𝛾0 ) + 𝛿2 (𝜃)𝜆(𝑥1 𝛾0 )2 + ⋯,

𝜙(.)
where 𝜆(. ) is the inverse Mills ratio defined as 𝜆 = Φ(.), while 𝜙(. )⁡and Φ(. ) are the

density and the c.d.f. of a standard normal variable, respectively. Thus, for appropriate
values of the δ’s ℎ̂𝜃 (𝑥1 𝛾0 ) → ℎ𝜃 (𝑥1 𝛾0 ) as the number of terms goes to infinity.
Finally, in order to estimate γ, we use the semi-parametric estimator suggested by Klein
and Spady (1993).

Klein and Spady (1993)

First, the dichotomous realization of the participation equation is specified:

′
1⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡𝑖𝑓⁡𝑣𝑖 ≥ 𝑥1𝑖 𝛾0
𝐷𝑖 = {
0⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

Taking the conditional expectation of D conditional on 𝑥1 , we obtain

′ ′
𝐸(𝐷𝑖 |𝑥1𝑖 ) = 1 × Pr(𝑣𝑖 ≥ −𝑥1𝑖 𝛾0 |𝑥1𝑖 ) = Pr(𝑣𝑖 < 𝑥1𝑖 𝛾0|𝑥1𝑖 ) = 𝐹𝑣|𝑥 (𝑥1′ 𝛾0 )

So, it is obtained:

27
Assumptions C and E in Buchinsky (1998) pp.4.

71
𝐸(𝐷𝑖 |𝑥1𝑖 ) = 𝐹𝑣|𝑥 (𝑥1′ 𝛾0 ) (1)

Klein and Spady (1993) proposes a semi-parametric estimation 𝛾0 in which assume that
the model satisfies the index restriction
𝐸(𝐷𝑖 |𝑥1𝑖 ) = 𝐸(𝑥1′ 𝛾0 ) (2)

Equations (1) and (2) implies that:

𝐸(𝐷𝑖 |𝑥1𝑖 ) = 𝐹𝑣|𝑥 (𝑥1′ 𝛾0 ) = 𝐺(𝑥1′ 𝛾0 )

Where 𝐺 is an unknown function whose range in contained in [0,1].

𝛾0 is computed by maximizing the equation (2) replacing the true but unknown
distribution 𝐹𝑣|𝑥 (. ) by 𝐺𝑛 (. ) that is a non parametric estimated of the function 𝐺(. )
which is a kernel estimate giveb by:

∑𝑗=1 𝑍𝑗 𝐾[(𝜗𝑖 − 𝜗𝑗 )⁄ℎ𝑛 ]

𝐺𝑛 (𝜗𝑖 ) =
∑𝑗=1 𝐾[(𝜗𝑖 − 𝜗𝑗 )⁄ℎ𝑛 ]

The semi-parametric estimator of 𝛾0, 𝛾0𝑆𝑃 , is obtained by maximizing the quasi

likelihood function given by
𝑁
𝐿𝑛𝐿 = ∑ 𝑍𝑖 𝐺𝑛 (𝑥1′ 𝛾0𝑆𝑃 ) + (1 − 𝑍𝑖 )(1 − 𝐺𝑛 (𝑥1′ 𝛾0𝑆𝑃 ))
𝑖=1

𝛾0𝑆𝑃 is consistent, asymptotically normally distributed and achieves the semiparametric

efficiency bound. In addition, in Monte Carlo experiments 𝛾0𝑆𝑃 performed well relative
to probit, and can, in models sufficiently perturbed from the usual probit specification
dominate the probit estimator (Klein and Spady, 1993).

72
73
74
 Essay 2
The Long-Term Effect of Inequality on
Entrepreneurship and Job Creation

75
76
 The Long-Term Effect of Inequality on Entrepreneurship and Job Creation *

Abstract
We assess the extent to which historical levels of inequality affect the probability of
businesses being created, surviving and of these creating jobs overtime. For this
end, we build a pseudo-panel of entrepreneurs across 48 countries using the Global
Entrepreneurship Monitor Survey over 2001-2009. We complement this pseudo-
panel with historical data of income distribution and current indicators of business
regulation. We find that countries with higher levels of inequality in the 1700s and
1800s, their businesses today are more likely to die young and create fewer jobs.
Our evidence support theories that argue initial wealth distribution influences
development path, thereby with important policy implications for wealth
distribution.

*
This essay has been co-written with Roxana Gutiérrez Romero (Departament d’Economia Aplicada –
Universitat Autònima de Barcelona).

77
 2.1 Introduction
To foster development it is crucial to understand the reasons why entrepreneurship
struggles or flourishes. Whilst the literature has developed complex theoretical models
on what might drive entrepreneurship over time, these theories have not been
empirically tested (Naudé, 2010). Instead, the empirical literature has focused on
analyzing separately the individual, economic or institutional factors that might affect
entrepreneurship.
We contribute to the literature by testing empirically one of the main
mechanisms highlighted in the theoretical literature that suggest affect entrepreneurship
over time. The theoretical occupational choice model proposed by Banerjee and
Newman (1993) guides our work. This model suggests that initial conditions,
understood as the historical distribution of wealth, can be detrimental for economic
development if credit constraints are such that they prevent poor individuals from
investing in profitable entrepreneurial activities. The model shows that a country can
converge to a different family of equilibriums, depending on the initial wealth
distribution. Countries that start with a high proportion of non-credit constrained people
will grow over time aided by a high share of people being able to start-up business, of
these surviving over time and with an active labor market paying high salaries. A
contrasting equilibrium could be reached if a country starts with a high proportion of
credit constrained people. In this case, only a small share of the population will be able
to start-up new businesses, whilst the rest will remain as workers, earning low wages
over time, in which there is (almost) only self-employment at small scale.
Based on this model, the main goal of this paper is to test whether initial
conditions, proxied by the income distribution prevailing in the 1700s and 1800s, and
taking into account the current business environment, have a detrimental effect on
today’s chances of businesses being created, surviving, and creating jobs over time.
Since our interest is to look at the effect of initial conditions on the dynamics of
entrepreneurship, ideally we would want to follow firms over time. Unfortunately,
empirically it is difficult to follow the same firms over time, especially if firms die in
large numbers creating substantial attrition bias and if surveys are being censored by
not representing newly created firms. We overcome these limitations by constructing a
pseudo-panel of entrepreneurs using the Global Entrepreneurship Monitor (GEM)

78
survey, the largest comparable dataset covering 70 countries over 2001-2009.1 The
GEM datasets are drawn from a new sample in each country every year. However, the
surveys include nationally representative information on how many people claimed to
be entrepreneurs, whether they are involved in nascent, young, established firms, or
have shut down businesses over the last year; as well as information on firm’s size at
each of these different stages of entrepreneurship.2 Thus, using this information we
build a pseudo-panel of cohorts of people based on their age and gender for each
country following the methodology proposed by Deaton (1985). In doing so, we are
able to track generations of people over time and assess whether initial conditions and
current business environment affect the creation, survival of firms, as well as job
creation.
We complement the GEM survey with historical data of income distribution
from the 1700s and 1800s as estimated by Morrisson and Murtin (2011) and
Bourguignon and Morrisson (2002) respectively. We also use historical indicators of
GDP per capita prevailing in the 1800s, obtained from the historical databases
estimated by Maddison. In addition, we use the index of credit protection provided by
the World Bank, which measures the degree to which laws protect the right of
borrowers and lenders, thus proxing the extent to which laws are designed to expand
access to credit.
We combine the pseudo-panel methodology with instrumental variables given
that the index of law protection of borrowers and lenders we use could be endogenously
determined by the proportion of people involved in entrepreneurial activities, who for
instance may lobby having better laws. As instrumental variables we use the legal code
of origin and the colonial origin, both variables frequently used in the literature when
dealing with the endogeneity of business regulation (La Porta, 1998; 1999). In addition,
we use the average blood pressure and cholesterol, instruments that have been found in
the literature to be correlated with the physiology responses to economic stress, such as
credit constraints (Ezzati et al., 2005; O’Neil et al., 2005).
We find that initial conditions have a detrimental effect on development, even
when taking into account current regulation in the credit market. Countries that started

1
Although the survey covers 70 countries we include in our analysis only 48 as are the ones we could
obtain data on historical income distribution.
2
Nascent firms are those recently created that have not payed wages for more than three months; young
firms have been running for up to 3.5 years and established firms have been running for more than 3.5
years.

79
with a high ratio of rich to poor people during the 1700s or 1800s currently are less
likely to open new firms, and of these to survive, and create more jobs over time.
Although several articles have tested whether inequality has a detrimental effect
on growth, our central contribution to the literature relies on testing an overlooked
mechanism as why this might be the case (Banerjee and Duflo, 2000; Benabou, 1996).
Specifically, our results suggest that high levels of inequality prevent people from
taking up business thereby affecting job creation and development in the long-run.
Our findings also suggest that improvements in the regulation of current credit
market promote the creation of both businesses and jobs. This effect however is of
lower magnitude in Africa than in other regions, perhaps because some African
households lack property rights of their land, thus prevented from providing a collateral
and accessing credit.
The article proceeds as follows. Section 2 discusses the literature on
entrepreneurship, including the model by Banerjee and Newman. Section 3 describes
the dataset and the construction of the pseudo-panel. Section 4 presents the econometric
results. Section 5 presents robustness tests. Section 6 concludes.

2.2 Institutions and Initial Conditions and Entrepreneurship

This paper is related to the large literature analyzing the factors that foster or constraint
the success of entrepreneurs. According to the interdisciplinary surveys on
entrepreneurship by Naudé (2008, 2010) the literature has experienced three important
developments over the last decades, which we describe below.
First, while the traditional research in entrepreneurship has focused on
empirically assessing the entrepreneur themselves, there has been a shift from
analyzing their personality traits and individual characteristics towards their behavior
and cognitive issues that enable them to recognize and exploit opportunities
(Blanchflower et al., 2001; Caliendo and Kritikos, 2011; Shane and Venkartarman,
2000). The same has happened in economics where there has been a shift towards
developing theoretical models of occupational choice (Evans and Jovanovic, 1989;
Lucas, 1978).
The second development in the literature has been to examine how business
environment influences the creation of firms and its relationship with long-run
development (Throton, 1999). Within this literature one can distinguish two veins. The

80
first one analyzes the extent to which historical institutions affect current ones which in
turn influence today’s entrepreneurial sector and growth. These studies, for instance,
examine the development path of former colonies.3 The second vein studies the impact
of current business regulation (such as investor protection and regulation of entry) on
entrepreneurship (Djankov et al., 2002; Glaeser et al., 2004; La Porta et al., 1998).
Within this vein, there is no consensus on whether business regulation always favors
entrepreneurship. For instance, business regulation could impose a burden on firms if
the regulation is aimed at extracting rents for the benefit of bureaucrats or certain
industries. However, the public interest theory of regulation argues entrepreneurship
can be fostered if regulation reduces market failures, by for instance allowing lenders to
seize the collateral in case borrowers default (Ardagna and Lusardi, 2008).
The third development in the literature has been the theoretical analysis on the
relationship between initial conditions, specifically wealth distribution, and
development on the long-run. This literature, within the neoclassical viewpoint,
analyzes whether initial conditions, such as country’s past inequality, can affect
entrepreneurship and economic growth in the long-run (Galor, 2011; Murphy et al.,
1989). 4 There is no consensus to the extent initial conditions can affect development.
On the one hand, the supporters of the “big push” hypothesis, argue that if there is the
possibility of coordination of investment across various sectors in the economy, which
can be promoted with public policy, countries can get out of no-
industrialization/development traps (Murphy et al., 1984; Rosenstain-Rodan, 1943). On
the other hand, other articles argue that initial conditions can determine development
path. For instance, inequality, it is argued, can have a long-term detrimental effect on
growth if the wealthier individuals lobby against changes in policies or institutions that
could distribute wealth and foster a more inclusive growth. 5 Inequality can also have a
detrimental effect on entrepreneurship if a large proportion of individuals are prevented
from taking up profitable investments, thus perpetuating inequality and low levels of
economic growth in the long-run. This negative effect of inequality on long-run
development could be enhanced whenever accompanied by credit market imperfections

3
For instance, Acemoglu et al. (2001) show that settler colonies perform better than former extractive
colonies because they inherited institutions that better protect private property rights.
4
See Benabou (1996) and Galor (2011) for a complete literature review on the effect of inequality on
development.
5
For an extensive overview of the dynamic interaction between political institutions and the development
process see Acemoglu et al. (2005).

81
(Aghion and Bolton, 1997; Banerjee and Newman, 1993; Galor and Zeira, 1993;
Ghatak and Jiang, 2002).
Within the third development in the literature, there are few empirical papers
testing the effect of wealth distribution on entrepreneurship, and among the existing
ones usually done in a static way and for a single country. Nonetheless, supportive
evidence has been found in the USA that wealthier individuals are more likely to
become entrepreneurs (Hurst and Lusardi, 2004). There is however, mixed evidence on
whether inequality affects entrepreneurship, or the other way around. For instance,
Mesnard and Ravallion (2001) show for the case of Tunisia the number of business
start-ups is an increasing function of aggregate wealth and that the greater the initial
inequality of wealth, the lower the overall rate of new business start-ups.6 In contrast,
Yanya (2012) concludes that firm establishment causes poverty and income inequality,
but not the other way around using a panel data of the 76 provinces in Thailand over
1997-2008.7

2.2.1 Banerjee and Newman’s Occupational Choice Model

In this paper we follow the theoretical model by Banerjee and Newman (1993) as it
encloses the three main developments described above in the literature: analyzing why
people choose to become entrepreneurs, and how business institutions along with initial
wealth distribution can affect entrepreneurship and development in the long-run.
Specifically, Banerjee and Newman’s model assume that because of
imperfections in the credit market, people can borrow only limited amounts. These
imperfections are derived from the possibility that borrowers may renege their debt. To
prevent this, lenders will limit borrowing and will require a collateral, such that:

𝐿 < 𝑤 + (𝜋𝐹 ⁄𝑟̅ ) (1)

where 𝐿 is the amount borrowed, w is the borrower’s wealth, 𝜋 is the probability of the
borrowers being caught if renege their debt, 𝐹 is the nonmonetary punishment of being
caught, and 𝑟̅ represents the return from a divisible safe asset which the model assumes

6
Initial wealth is captured by the amount of wealth accumulated by returned migrants from past savings
while abroad.
7
Income inequality is measured through the Gini index and poverty with the lowest income quintile at the
province level. The causal relationship is assessed using the granger causality test

82
requires no labor. The model assumes that anyone that invests only in this safe asset is
said to be idle or subsisting.
To become an entrepreneur people need to make an up-front investment. Thus,
entrepreneurship is only available to those individuals that are wealthy enough to make
this investment or provide the required collateral to accessing credit. Those poorer
individuals that do not have enough wealth to provide collateral have two occupation
choices: they can become employees, and for those individuals with individuals with
wealth between 𝑤 ∗ and 𝑤 ∗∗ they can also become self-employed. Self-employment is
assumed that requires some up-front investment but of lower level than the required to
become entrepreneur. As entrepreneurship requires an up-front investment is available
only to wealthy people or those who can provide the required collateral, whereas poorer
individuals credit constrained their choices are limited to becoming employees and if
have wealth between 𝑤 ∗ and 𝑤 ∗∗ will be able to become self-employed if they chose to.
The expected return to self-employment and subsistence are given exogenously
by the model’s parameters. Wage v, is determined endogenously in the model such that
it clears the labor market, and in turns determines the returns of entrepreneurs and
workers.
The equilibrium wage can take a low value 𝑣 if 𝐺𝑡 (𝑤 ∗ ) > 𝜇[1 − 𝐺𝑡 (𝑤 ∗∗ )]⁡, a
high value 𝑣̅ if 𝐺𝑡 (𝑤 ∗ ) < 𝜇[1 − 𝐺𝑡 (𝑤 ∗∗ )] and a value within the range [𝑣, 𝑣̅ ] if
𝐺𝑡 (𝑤 ∗ ) = 𝜇[1 − 𝐺𝑡 (𝑤 ∗∗ )].
where 𝐺𝑡 (𝑤 ∗ ) is the proportion of the population that has no other choice but to
become a worker, as does not have enough wealth to provide a collateral to become
entrepreneurs. 𝜇[1 − 𝐺𝑡 (𝑤 ∗ )] is the proportion of the population that can become
entrepreneurs. Then, the pattern of occupational choice that is generated in equilibrium
is summarized as:
1) individuals with initial wealth less than 𝑤 ∗ will be a worker unless wages are
exactly the minimum wage 𝑣 ,
2) individuals with initial wealth between 𝑤 ∗ and 𝑤 ∗∗ can become self-
employed.
3) individuals with ⁡⁡⁡𝑤 ≥ 𝑤 ∗∗ will be an entrepreneur if 𝑣 < 𝑣̅ . In the case
𝑣 = 𝑣̅ , then 1 − 𝐺𝑡 (𝑤 ∗ )/𝜇 − 𝐺𝑡 (𝑤 ∗∗ ) of them will opt becoming self-employed for the
labor market to clear.

83
 Then the pattern of occupational choice is determined by the initial distribution
of wealth, and the structure of occupational choice determines in turn, how much
people can save and leave a bequest. These factors, in turn give rise to a new
distribution of wealth affecting long-run development.
The model predicts that the fate of the economy depends on the initial wealth
distribution. Countries with an initially high proportion of non-credit constrain people
will grow over time aided by a high share of people being able to start-up business, of
these surviving over time and with an active labor market paying high salaries. A
contrasting equilibrium could be reached if a country starts with a high proportion of
credit constrained people. In this case, the process of development ends up in a
situation of low wages, in which there is (almost) only self-employment at small scale.
Based on Banerjee and Newman model, we will test the following two
hypotheses:
Hypothesis 1: Countries that have a historical high ratio of wealthy to poor
people, a proxy for being non-credit to credit-constrained, have a lower probability of
firms being created, surviving and of these creating jobs over time.
Hypothesis 2: Countries that currently have more efficient credit markets have a
higher probability of people being involved in entrepreneurship and higher job creation.

2.2.2 Endogeneity between Credit Regulation and Entrepreneurship

When testing our second hypothesis we are likely to encounter an endogenity problem.
We would expect that more efficient credit markets will benefit entrepreneurs.
However, it is also possible that the degree of imperfections in the credit market change
as response to the number of entrepreneurs in the economy, for instance if by lobbying
for better regulation (Besley and Gathak, 2010; Manski, 2000). We address this
potential endogeneity by using instrumental variables.
We use four instrumental variables across all the regressions presented in
section 4. Two of these variables (origin of country’s legal code and colonial origin),
are drawn from the institutional literature that has used these instruments to deal with
the endogeneity of current business environment. The other two instrumental variables
used (average blood pressure and cholesterol level) are drawn from the recent literature
on physiological responses to economic stress that can prove to be constrained from
credit. We explain below the literature supporting the use of these instruments.

84
 Based on the pioneer work of La Porta et al. (1998, 1999) several authors have
addressed the likely endogeneity of current business environment using as instrumental
variables the country’s historical legal origin (Ardagna and Lusardi, 2008; Djankov et
al., 2003; Gleasser et al., 2004; Levine et al., 2000). La Porta et al. show that the legal
rules protecting investors are greatly dependent on the legal traditions or origins. For
instance, they find that countries under the English common law are more protective of
investor rights and contractual enforcements than the laws originated in the French civil
code. Thus, countries with “better” legal origins are more likely to develop institutions
in which property rights are protected and less distortionary policies are implemented,
which in turn favor investment and economic growth.8 Other studies have also found
that, the colonial origin of the country is a strong predictor of current’s institutions
(Acemoglu, et al., 2001). These authors stress that different types of colonization
policies created different sets of institutions which persisted over time. In one extreme,
whenever colonizers aimed at exclusively draining resources from the colony
developed “extractive” institutions with poor emphasis on protecting private
investment.9 In contrast, whenever colonizers intended to settle in these colonies in the
long-run, they tried to replicate European institutions, protecting property rights.10
Recent literature has found that people who find hard to gain access to credit
can experience physiological responses to stress. For instance, people experiencing
financial distress are less likely to follow recommended health maintenance practices
such as eating a healthy diet, thus elevating risk of cardiovascular diseases, elevated
blood pressure, and cholesterol (O’Neill et al., 2005). Also, cardiovascular diseases and
their nutritional risk factors such as overweight and obesity, elevated blood pressure,
and cholesterol, have been predicted to rise with economic development and hence to
vary across regions, an important aspect since the credit market regulation we analyze
vary sharply across countries (Ezzati et al., 2005).

8
La Porta et al. (1998) stress that countries under the English common law have the best investor right
protection and contractual enforcements, followed by those under German or Scandinavian civil law, and
of these followed by countries with French civil law.
9
Belgian colonization in the Congo is an example of extractive institutions, whilst the Great Britain
colonization of Australia, New Zealand, United States and Canada are examples of pro-European
institutions (Acemoglu et al., 2001).
10
Acemoglu et al. (2001) argues that former British colonies prospered relative to former French,
Spanish, and Portuguese colonies because of the good economic and political institutions and culture they
inherited from Britain.

85
 2.3 Data and Methodology

2.3.1 Historical Income Distribution and Current Credit Regulation

In our regression models presented in Section 4 we control for countries’ initial wealth
per capita. For such purpose, we use the GDP per capita prevailing in the 1800
estimated by Angus Maddison’s historic income database.11
We also use the historical data on income distribution prevailing in the 1700s
and 1800s as estimated by Morrisson and Murtin12 (2011) and Bourguignon and
Morrisson (2002) respectively. These estimates provide the income share for each
decile, which we use to build different indicators to proxy the historical ratio of people
that were credit and non-credit constrained, such as the Gini coefficient and different
ratios of income shares across different deciles. We do so as Banerjee and Newman
stress that income inequality is the main factor preventing poor people to invest in
entrepreneurial activities. Moreover, previous research has shown that people in the
lower deciles are less likely to have access to credit, as they might not to have enough
wealth to provide a collateral or are living away from a banking institution, thus
affecting their chances of obtaining credit (Baliamoune-Lutz et al., 2011; Berg, 2013).
Since we are interested in assessing the impact that credit market imperfection
have on the creation of firms and jobs over time, we use indicators on credit protection
from the Doing Business database gathered by the World Bank from 2004 to 2009.13
Specifically, we use the strength of legal right index which “measures the degree of
which collateral and bankruptcy laws protect the right of borrowers and lenders and
thus facilitate lending.” The index ranks from 0 to 10, where higher scores indicating
that collateral and bankruptcy laws are better designed to expand access to credit.14
This variable is particularly suitable for our analysis as it is a proxy of the extent to
which better credit rules can enhance investment incentives by improving

11
Online data available at: Maddison Project website http://www.ggdc.net/maddison/maddison-
project/home.htm
12
We thank Fabrice Murtin for having provided us these datasets.
13
Since the Doing Business dataset covers the year 2004 until 2009, we imputed the values for the years
2001 and 2002 taking the information for the year 2004 or for the closest year we had information on. We
did so to retain as much information as possible for earlier years, and given the little change in business
environment observed for the years we have.
14
Data on the legal rights of borrowers and lenders are gathered through a questionnaire administered to
financial lawyers and verified through analysis of laws and regulations as well as public sources of
information on collateral and bankruptcy laws. A detailed description of the elaboration of this index can
be found in: http://www.doingbusiness.org/methodology/getting-credit

86
collateralizability of assets and limiting its seizing. All those aspects improve property
rights thereby reducing imperfections in the market (Besley and Gathak 2010).

2.3.2 GEM Survey

We use the Global Entrepreneurship Monitor (GEM) survey, the largest study on
entrepreneurial activity over 2001-2009.15 The surveys are representative of the adult
population and are collected annually. In developing countries the survey is conducted
in face-to-face interviews, and otherwise through random telephone interviews.
We use the definition of “entrepreneurs” proposed by the GEM network: “adults
in the process of setting up a business they will (partly) own and or currently owning
and managing and operating young businesses” (Reynolds et al., 2005 p. 209). To
study the dynamics of entrepreneurship we consider the four stages in the life-cycle of
businesses, as defined by the GEM network. In the first stage, nascent entrepreneurs,
are those actively involved in setting up a business they will own or co-own, but who
have not paid salaries, wages or any other payments to the owners for more than three
months. In the second stage are the owners of young firms, defined as those who have
paid salaries for more than three months and up to 3.5 years. In the third stage,
established firms, are those who have paid salaries or wages for more than 3.5 years. In
the fourth and last stage are firms that in the past 12 months, have been sold, shut down
or discontinued.
For our analysis, we focus on 48 countries surveyed in GEM for which we also
have gathered historic information on income distribution and GDP per capita. The
countries analysed are listed in Table A.1 in the Appendix. In total, we have 1,001,458
individuals interviewed over 2001-2009. From these, 37,136 were in nascent
entrepreneurship, 32,359 in young firms, 62,514 in established firms and 25,183 had
recently shot down their firm.
In the Appendix we report the basic descriptive statistics of the pseudo-panel
and other aggregate indicators used for the whole sample, and differentiating between
groups of countries –low-middle and high income countries (Tables A.3 and A.4). A
characterization of firms at different stages of entrepreneurship, such as size and sector,
and grouped by region are presented in Table A.6. In Table A.7, we report the

15
The chosen period of analysis refers to that for which the GEM datasets are publicly available.

87
correlation matrix among all the dependent and explanatory variables used, which show
that we have no problems of multi-colinearity.
Figure 1 shows the percentage of the population engaged in the various states of
entrepreneurship analysed over 2001-2009. The onset of the economic crises reduced
the percentage of the population involved in entrepreneurial activities across all stages
(nascent, young and established firms) particularly in 2009.

2.3.3 Pseudo-Panel
Since GEM draws new samples each year, the surveys remain representative of the
population engaged (or that were engaged) in entrepreneurial activities over time,
avoiding an attrition bias. Since a new sample is drawn each year, we cannot study the
decision of the same individuals to become or remain in entrepreneurial activities over
time. To overcome this limitation, we construct a pseudo-panel using the GEM surveys
and the methodology proposed by Deaton (1985). We describe next the construction of
the pseudo-panel.
GEM consist of a set of T independent cross-sections of i individuals that belong
to a new and most likely different set of I individuals in each period. Equation (2)
denotes the factors that affect whether a person is an entrepreneur, if we were to stack
together all the cross-section observations, typically known in the literature as pooled-
cross section.

𝑦𝑖𝑡 = 𝛽𝑥𝑖𝑡 + 𝛿𝑖 + 𝜀𝑖𝑡 (2)

where yit denotes whether the individual is engaged in an entrepreneurial stage, xit
denotes a vector of explanatory variables, i and it are the individual-specific time-
constant unobserved heterogeneity; and the unobserved idiosyncratic error that varies
over individuals and time.
OLS estimates using this pooled-cross section data will be biased and
inconsistent if the individual unobserved characteristics (such as personal traits, risk
aversion or cognitive abilities) were correlated with some or all of the explanatory
variables. To solve this potential endogeneity problem, Deaton (1985) proposed
building a pseudo-panel, which yields consistent  estimators, even when the individual
unobservables characteristics are correlated with explanatory variables. Pseudo-panels

88
have the additional advantage of avoiding attrition problem that plagues genuine panels
since data is collected from random samples drawn from cross sections.16
To build the pseudo-panel Deaton (1985) proposes to average observations with
similar characteristics that are stable over time (such as gender, year of birth) in a
sequence of repeated cross-sectional datasets. These synthetic observations can be
therefore thought as cohorts of generations being “followed” over time, just as if pure
panel surveys were available.
Following Gutiérrez-Romero (2012) who built a pseudo-panel using the GEM
survey for the case of Spain, we build the pseudo-panel by defining the cohorts within
countries in terms of gender and year of birth, as these are observable and do not
change over time.17 In total, we have nine time periods (2001-2009) and 10 cohorts in
each. Five of these cohorts are for males, and five for females. Within each gender we
further defined five cohorts of age: those who in 2001 were 28 years old or less, 29-38,
39-48, 49-58 and 58 or over.18 The average sample size for each cohort is shown in
Table A.5.
We produce the pseudo-panel by averaging observations over individuals in
each of the cohorts C described above and T periods, as shown in equation (3).

̅ + 𝜀𝑐𝑡
𝑦̅𝑐𝑡 = 𝛽𝑥̅𝑐𝑡 + 𝛿𝑐𝑡 (3)

where the bars denote the average value of all individuals in cohort c at time t. The
̅
average of the fixed effects of those members belonging to cohort c in the sample 𝛿𝑐𝑡
̅ is unobserved it might be correlated with 𝑥̅𝑐𝑡 therefore
varies over time. Since 𝛿𝑐𝑡
̅ as a fixed effect can lead to
leading to inconsistent estimates.19 In addition, treating 𝛿𝑐𝑡

16
The pseudo-panel approach is especially useful for life-cycle models, and has been recently taken in
empirical studies for which panel data is not available, largely used in social mobility analysis (Antman
and Mckenzie, 2005) and previously used for studying entrepreneurial success of the Spanish case in
Gutiérrez-Romero (2012).
17
We also define cohorts following age and gender as the literature has found evidence of the probability
of being engaged in entrepreneurial activities differs considerably with regard to these two variables and
allows to explicitly recognizing the life-cycle stage a firm is in (Bergmann and Sternberg 2007).
18
For instance, individuals are considered to belong to the first cohort of age if they were aged 30 in year
2001, 31 in 2002, 32 in 2003 and so on.
19
This is likely in our case because we consider a number of explanatory variables that might be
correlated with the error term, such as individuals’ personality traits like risk aversion and cognitive
abilities. Since these characteristics are unobservable and might be correlated with our outcome of
interest, the estimated effect could be biased.

89
an identification problem, unless it is assumed that the individual error is time invariant,
̅ = 𝛿𝑐̅ .
that is 𝛿𝑐𝑡
Baltagi (2005) argues that pseudo-panels estimations could be biased if cohorts
do not have enough observations to eliminate a potential unobserved heterogenity bias.
Verbeek and Nijman (cited by Gutiérrez-Romero, 2012) show that if each cohort has
greater than 100-200 observations, as it is our case, then the cohorts will be large
enough to eliminate the unobserved heterogeneity bias if assumed the individual error
is time invariant. In that case, equation (3) can be estimated using cohort dummy
variables yielding unbiased estimators.
To ensure that the estimators are also efficient, we control for the likely problem
of heteroskedasticity, which could occur if the number of observations per cohort varies
substantially. To correct for this we use weighted least squares (WLS) by weighting by
the square root of the number of observations in each cohort, as it is recommended in
the literature (Dargay, 2007).

2.4. Econometric Results

2.4.1. Firm’s Life Cycle: Birth, Maturity and Death

To test our two hypotheses and to deal with the potential endogeneity of the degree of
credit protection we extend equation (3) and estimate instead equation (4). We correct
for this endogeneity in a two-stage process, as it is standard in the literature, so equation
(4) represents the IV-second-stage least squares estimation.

̅ + 𝜀𝑐𝑡
𝐸[𝑦̅𝑐𝑡 |𝑍] = 𝛼 + 𝛽1 𝐼𝑁𝐸𝑄1820 +𝛽2 𝐿𝑖𝑛𝑑𝑒𝑥 + 𝛽3 𝑥̅𝑐𝑡 + 𝛽4 𝑋 + 𝛿𝑐𝑡 (4)

where 𝑦̅𝑐𝑡 measures the dependent variable in the second-stage least square, as the
proportion of individuals involved in a specific stage of entrepreneurship, namely
nascent, young, established or recently closed firm. 𝐼𝑁𝐸𝑄1820 represents the historical
ratio of wealthy people (income share of top 9th decile) to poor people (bottom 1st
decile) prevailing in 1820. We use this indicator as a proxy of the ratio of non-credit to
credit constrained people. Lindex represents the strength of legal right index20, 𝑋 is a set

20
Note that the legal right index ranks from 0 to 10, however this index is not equal to 0 for none of the
countries over the period time considered in the analysis, then being possible to make this log
transformation.

90
of characteristics, which includes GDP per capita in 1800, regional and year dummy
variables to control for unobserved regional and time effects. At cohort level, in 𝑥̅ 𝑐𝑡 we
include the proportion of people in cohort c at time t with secondary education or more,
̅ . 𝑍 is the instrument used in the first-stage least
and control for cohort fixed effects 𝛿𝑐𝑡
squares, which is a dummy variable for whether the country’s legal origin’s code is
English or not. All variables are measured in logarithms except the generation cohort,
the instrumental variable Z, regional and time dummy variables.
Table A.8 (in Appendix) shows the results of the first-stage regressions. This
table includes the coefficients associated with our instrument, whether the origin of the
legal code is English, and our endogenous variable, the legal right index. We find that
the instrument is positive and statistically significant across all models presented. We
also include the summary statistics for the first stage regressions, in which the F-
statistics test of the excluded instrument, is greater than 10 and statistically significant
across all models ran, which suggest our instrument is not weak
Table 1 presents the results of the IV-second-stage least squares. There we also
include the endogeneity test which confirm that the legal right index is endogenous
with the our dependent variable 𝑦̅𝑐𝑡 , the proportion of people involved in different
entrepreneurial stages. The Kleibergen-Paap Wald F statistic test confirms the
instrument is correlated with the endogenous variable, the legal right index.21
Our results confirm the first hypothesis. The higher the ratio of wealthy to poor
people in the 1820 the lower the probability that people were engaged in
entrepreneurial activities across all stages, nascent, young and established firms, during
the period 2001-2009 (Table 1, columns 1-4). The lower the income share of the poor
relative to the wealthy, the less the share of people involved in firms of any type. For
instance, a 1% increase in the historic ratio of wealthy to poor reduces the proportion of
people involved in nascent firms by 0.2%, the proportion of people involved in young
firms by 0.17% and the proportion of people involved in established firms by 0.08%.
We also find evidence to support our second hypothesis. The higher the index of
legal rights, a proxy we use for efficiency in the credit market, the higher the proportion
of people involved in entrepreneurial activities. Specifically a 1% increase in the legal
right index, increases the proportion of people involved in nascent firms by almost 1%,
the proportion of people involved in young firms by 0.8%, and the proportion of people

21
We do not present the exogeneity test, which test the null hypothesis that the instruments are jointly
exogenous, since this test can only be conducted with more than one (Baum, 2006).

91
involved in established firms by 0.22%. These results suggest that the strength of the
legal right index is more important in the early stages of entrepreneurship than those
already established. There are potential reasons for this. For instance, already
established firms might have had time to generate their own financial resources (from
previous profits) and had enough time to develop networks, other than with financial
markets, that could enable them to stay afloat in case of requiring prompt credit. This
argument is in line with previous research that shows small and medium firms are more
likely to be more credit constrained than larger firms (Claessens et al., 2007). For
instance, Kuntchev et al. (2013) show that the firms’ perception of being credit is
negatively correlated with firm’s size and age: smaller and younger firms tend to find
access to credit to be more of stringent constraint to carry out their operations than
larger and older firms.
We also find the higher the historical GDP per capita, the less people would be
involved in different stages of entrepreneurial activity over time. It is unclear why this
might be the case. One potential reason, and in line with the predictions of Banerjee and
Newman model, is that countries that started with higher historical GDP per capita over
time developed a more active labor market, paying higher wages. As wages rise, more
people would prefer becoming workers, instead of entrepreneurs.
The cohort effects on entrepreneurial activity show that in general, older
individuals are more prone to be involved in established business, while younger people
are engaged in young firms. This result is consistent with previous studies that show
that because knowledge, capital accumulation, and experience increase with age, over
time individuals are more likely to have an established firm (Bergmann and Sternberg,
2007).
In addition, we find evidence that the higher the proportion of people with high
school or more, the less likely would be engaged in entrepreneurial activities, for all
nascent, young and established firms during the period 2001-2009. A number of studies
have found a positive correlation between education and degree of entrepreneurship,
suggesting that education helps people identify opportunities in the market place and
provide them with needed managerial abilities (Simón-Moya et al. 2014). Our findings
instead, support the other vein in the literature that has found education to be negatively
related to the probability of being self-employed (Blanchflower, 2004; Reynolds et al.,
2003). These studies argue that education is not necessarily correlated with being an
entrepreneur as specific entrepreneurial knowledge is what matters more, managerial

92
abilities and knowledge in accounting and finance (Man et al., 2002). Other empirical
studies have found that employees in Spain and Portugal value more having higher
level of educations, whilst self-employed people have lower levels of education
(Garcia-Mainar and Montuenga-Gomez 2005).
To conclude this sub-section, we focus on the regional differences on firm’s life
cycle. We find that Africa was less likely to create firms and of these to survive over
time than firms located in the rest of the world over 2001-2009. These results might
reflect the structural and institutional differences supporting entrepreneurship in Africa
and the rest of the world.

2.4.2 Job Creation: Firms’ Size

We next move on to analyze the extent to which initial conditions and credit market
affect the number of employees hired by firms, as shown in equation (5)

̅ + 𝜀𝑐𝑡
𝐸[𝑠̅𝑐𝑡 |𝑍] = 𝛼 + 𝛽1 𝐼𝑁𝐸𝑄1820 +𝛽2 𝐿𝑖𝑛𝑑𝑒𝑥 + 𝛽3 𝑥̅𝑐𝑡 + 𝛽4 𝑋 + 𝛽5 𝐿𝑖𝑛𝑑𝑒𝑥 ∗ 𝑟𝑒𝑔𝑖𝑜𝑛 + 𝛿𝑐𝑡 (5)

where 𝑠̅𝑐𝑡 represents the average number of employees hired by firms in each stage of
entrepreneurship in the cohort c at time t. In addition, we interact the legal right index
with a regional variable (𝐿𝑖𝑛𝑑𝑒𝑥 ∗ 𝑟𝑒𝑔𝑖𝑜𝑛) to take into account for regional differences
in the credit regulation. We also add in 𝑥̅𝑐𝑡 a categorical variable denoting the sector of
the firm and a dummy variable denoting whether the firm has a medium/high level of
technological intensity, both variables provided in the GEM surveys.
We chose these explanatory variables following the literature on the
determinants of firms’ size. Our key explanatory variable affecting firm size over time
is the historical ratio of wealthy to poor. We include this variable based on the
theoretical model of Banerjee and Newman, expecting that the higher the historical
income inequality the smaller the firms will be. We also include in our regressions the
legal right index, as the literature predicts that countries with better institutions and
more access to credit to be more likely to develop larger firms (Beck et al., 2003; 2005;
Kumer et al., 2001). In addition, we control for sector fixed effects and technology
intensity as the literature has found these variables play a crucial rule on firm’s size
(Aghion et al. 2007; .Kumar et al., 2001). Finally, we take account of market size, as
the literature predicts that firms will expand in size depending on the expected profits
of the market (Lucas. 1978). Since we are interested in studying the impact of initial

93
conditions, and to avoid a potential endogenity issue with current market size, we use
as a proxy of market size the GDP per capita prevailing in 1800 and not actual GDP per
capita. We measure in logarithm our dependent variables, the ratio of wealthy to poor,
GDP per capita in 1800 and the legal right index.22
The legal right index is likely to be endogenous with the size of the firms, as
well as the interaction of this legal right index with the seven regional dummies used.23
Thus, we require at least eight instruments, one for our proxy to access to credit, and
seven for this variable interacted with the dummy regional variables. The instruments 𝑍
we use are: the country’s origin of legal code (one dummy for each legal code: English
common law, French commercial code, Socialist/Communist law, German commercial
code and Scandinavian commercial code); the colonial origin of the country (a dummy
variable equal to one if the country’s colonial origin is Spanish, and zero otherwise) and
two variables that measures blood pressure and cholesterol at country level.24
In Tables A.9.1 to A.9.3 (in the Appendix) we provide the first stage
regressions. This table includes the coefficients associated with our instruments and our
endogenous variables, the legal right index and its interaction with the regional
variables. We find that the instruments are statistically significant across all models
presented. The F-statistics test of the excluded instruments are greater than 10 and
statistically significant across all models ran, which suggest our instruments are not
weak.
In Table 2 we present the IV second-stage least squares. There we also include
the endogeneity test which shows our dependent variables are endogenous. As before,
we include the Kleibergen-Paap rank Wald F statistic test which confirms that our
instruments are not weak. All models are just identified.
We find mixed evidence to support our first hypothesis. On the one hand, the
higher the historical ratio of wealthy to poor, the bigger the nascent firms were over
2001-2009 (Table 2, column 1). On the other hand, and in line with our first hypothesis,
the higher the historical ratio, the smaller the young and established firms are over time

22
Given that firms could have no workers hired, if taking the logarithm of our dependent variable would
lose several obseravtions. To prevent this, we trasnsform our dependent variable by adding one to the
number of hired workers. Then we take the logarithm of that number, and that is the variable we use as
dependent variable.
23
The regions considered in the analysis are: Africa, Asia, Western Europe, Latin America, North
America, Oceania and Eastern Europe.
24
Table A.2 in the Appendix shows in further detail the instrumental variables definitions and data
sources.

94
(Table 2, columns 2-3). This evidence suggests that as the income share of the poor
shrinks (the higher the historical ratio of wealthy to poor) the bigger the nascent firms
aided perhaps by low salaries. But, once firms get older they shrink in size. This
apparent mixed evidence is however consistent with the predictions of Banerjee and
Newman (1993). Their model predicts that countries with high ratio of rich to poor
people will fail in the long-run building a real demand for the local market production,
thus affecting the size of firms as they mature. In similar line, Murphy, Shleifer and
Vishny (1989b) show that countries with high income inequality will have a low
demand for labor as they do not have a critical mass in their markets to justify firms of
bigger size.
We find evidence to support our second hypothesis. The higher the legal right
index the bigger the firm’s size, across at stages of entrepreneurship.25 The effect of this
index is greater for the nascent firms, and decreases for young and established firms.
Which confirms, as earlier shown, once firms are already established they might be less
dependent of external credit resources than those firms that just started.
We also find that improving credit protection increases the firm size to lesser
extent in Africa than compared to other regions. Which suggests that even if regulation
is improved in Africa, its effect will be of lesser impact than in other regions, perhaps
because fewer people in Africa will be take advantage of the improved institution if do
not have the required collateral. Thus, policy interventions aiming to reduce barriers to
access to credit should take into account the specificities of the different regions. In
some regions, the problem could be the lack of resources or competition of the banking
system, the lack of protection to lenders; while in others could be the excess of
collateral requirements. For instance, Baliamoune-Lutz et al. (2011) point out that a
major issue for African countries is the collateral needed to secure bank loans. Some
households in these countries do not have formal titles of their lands, and the constraint
is particularly severe for women-headed households.

25
We obtain the total effect of this legal index by adding up the coefficients of the legal right index and
the interactions between this variable and the regional dummies, which turned statistically significant
across all specifications in Table 2.

95
 2.5. Robustness Checks
We conducted three main robustness checks to assess the validity and consistency of
the results so far presented.
First, we re-run our IV-pseudo-panel regressions but excluding from the
analysis self-employed people, in other words, people who claimed were not hiring
workers. We do so as the model by Banerjee and Newman (1993) distinguishes
between self-employed and entrepreneurship. Table A.10 tests our two hypotheses on
the probability of people being engaged in entrepreneurial activities, and Table A.11 on
the size of the firms. Both tables confirm our previous results inequality is detrimental
for the creation of business, of these surviving and creating jobs overtime, whilst better
legal right index if beneficial.
Second, we test alternative inequality measures, four different ratios of wealthy
to poor and other indicators such as the Gini index, finding no differences with the
results so far presented.26 Tables A.12 and A.13 in the Appendix show the overall, the
detrimental effect of income inequality remained on firm’s life cycle and job creation
across the alternative indices used. For instance, when using the Gini index, we observe
that the higher this index, the less people involved in nascent and established firms.
However, we find a positive relationship between the Gini index and the proportion of
people involved in young firms, which is opposite to what is observed in Table 1.
Across all regressions presented in Table 1 and Table 2 we also tested the ratio
wealthy to poor but for 1700. This ratio yields practically identical results to once
present using the ratio 1820s, hence we omitted them.
Third, we consider different instrumental variables in our estimations, such as,
religion and language fractionalization (Alesina et al., 2002), instruments commonly
used in the literature. However, all of these variables proved to be weaker instruments
than the ones presented in our estimations. Tables A.14 to A.17 presents the estimated
coefficients of the key independent variables and a summary of first stage regressions,
weak identification test and endogeneity test. Overall, due to the weakness of the

26
These four ratios are defined as: The income share of the 1st decile to the average income (bottom 10);
income share of the 9th decile to average income (top 90); income share of the median to the average
income (middle50); the income share of the 8th decile to the income share of the bottom 2nd decile
(top20/bottom20). We also use the sum of the income shares of the 2nd, 3rd and 4th quintiles (middle).

96
instruments we obtain inconsistent estimations in comparison to the ones obtained with
strong instruments.

2.6. Conclusion
The aim of this article was to test the influence of historical income inequality along
with the current business environment on the probability of creating new businesses
and of these surviving over time and creating jobs at different stages of the firm’s life
cycle. For this purpose, we built a pseudo-panel of cohort of people across 48 countries
over 2001-2008, using the Global Entrepreneurship Monitor Survey and the pseudo-
panel methodology proposed by Deaton (1985).
We draw two main conclusions from our results. First, initial inequality,
understood as the inequality prevailing in the 1700s or 1800s, has a persistent and
detrimental effect on the creation, survival of firms as well as job creation over time.
Second, countries with worse credit markets, proxied in our analysis by an index that
measures the easiness is to lend in the market, the less likely that firms will be created,
survive and create jobs over time.
Our findings are consistent with the prediction of the model by Banerjee and
Newman (1993). This model suggests that if the initial wealth distribution, is such that
a large percentage of the population are credit constrained, then fewer firms will be
created and survive over time especially, under the presence of credit market
imperfections.
Despite the extensive research on the relationship between inequality and
economic growth, there still remains considerable disagreement about the sign of this
relationship in the literature. Banerjee and Duflo (2000) argue previous studies are far
from being conclusive of this relationship because of identification problems and data
limitations in cross-country studies. Moreover, most empirical papers have assessed the
impact of inequality by using not so distant indicators of inequality, instead of historical
ones, limiting our understanding of the extent that early inequality conditions, affect
economic development over time.
To the best of our knowledge, this is the first empirical paper that tests the
predictions of Banerjee and Newman model and other similar theoretical models that
suggest initial conditions, understood as the wealth distribution prevailing in the distant
past, can affect entrepreneurship and development in the long-run. Our results, have

97
important policy implications. Although we did not specifically test for convergence,
our findings suggest that since some countries are predisposed by their initial
conditions to be trapped into a firms-die-young equilibrium whilst others are in a
different type of equilibrium with businesses thriving over time, thus, economic
convergence across countries is unlikely to occur. Our findings, in line with the
theoretical literature, suggest that to foster the creation of jobs and businesses, policies
should focus on addressing long-standing differences in wealth within countries as well
as reducing credit constraints. Incidentally, these policies could foster convergence
across countries as well, an issue that deserves further research.

Acknowledgements
We thank Fabrice Murtin for having shared his estimations of income distribution
prevailing in the 1700s and 1800s. We are grateful to Professors Maitreesh Ghatak and
Elias Papaionnou for informal discussions on early stages of work. We thank Cristina
López-Mayan, Adam Pepelasis, and the participants of the EDIE workshop, the GEM-
Barcelona conference, UAB PhD seminar, Universidad Tecnológica Metropolitana de
Mérida, the LACEA/IADB/WB/UNDP Research Network of Inequality and Poverty
for their comments and suggestions on earlier stages of this paper. Finally, we are
grateful to Isabel Busom for her comments on an earlier version of this paper.

98
 References

Acemoglu, D., Johnson, S.; and Robinson, J. (2001) “The colonial origins of
comparative development: An empirical investigation”, American Economic Review,
91(5): 1369-1401.
Acemoglu, D., Johnson, S.; and Robinson, J. A. (2005) “Institutions as the
fundamental cause of long-run growth”, (in) P. Aghion and S. N. Durlauf (eds),
Handbook of Economic Growth, Vol. IA, Elsevier North-Holland, Amsterdam, The
Netherlands.
Aghion, P., and Bolton, P. (1997) “A theory of trickle-down growth and
development”, Review of Economic Studies 64(2): 151-172.
Aghion, P.; Fally, T., and Scarpetta, S. (2007) “Credit constraints as a barrier to
the entry and post-entry growth of firms”, Economic Policy, 22(52): 731–779.
Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S.; and Wacziarg, R.
(2002) “Fractionalization”, Journal of Economic Growth, 8(2): 155-194.
Antman, F., and McKenzie, D. (2005) “Earnings mobility and measurement
error: A pseudo-panel approach”, Stanford, United States: Stanford University WP.
Ardagna, S., and Lusardi, A. (2008) “Explaining international differences in
entrepreneurship: The role of individual characteristics and regulatory constraints”,
NBER WP 14012.
Baliamoune-Lutz, M., Brixiová, Z.; and Ndikumana, L. (2011) “Credit
constraints and productive entrepreneurship in Africa”, Political Economy Research
Institute WP No. 276.
Baltagi, B. H. (2005) “Econometric analysis of panel data” (3ed.). Chichester,
Hoboken, N.J: John Wiley & Sons.
Banerjee, A., and Duflo, E. (2000) “Inequality and growth: What can the data
say?”, NBER WP No.7793.
Banerjee, A., and Newman, A.F. (1993) “Occupational choice and the process
of development”, Journal of Political Economy, 101(2): 363-394.
Baum, C. (2006) “An introduction to modern econometrics using stata”,
StataCorp LP.
Beck, T., Demirguc-Kunt, A.; and Levine, R. (2005) “SMEs, Growth and
Poverty: Cross-Country Evidence”, Journal of Economic Growth, 10: 199-229.

99
 Beck, T., A. Demirguc¸-Kunt, and Maksimovic, V. (2003) “Financial and Legal
Institutions and Firm Size”, World Bank mimeo.
Benabou, R. (1996) “Equity and efficiency in human capital investment: the
local connection”, Review of Economic Studies 63(2): 237-264.
Berg, E. (2013) “Are poor people credit-constrained or mypic? Evidence from a
South African panel”, Journal of Development Economics, 101(3):195-205.
Bergmann, H., and Sternberg, R. (2007) “The Changing face of
entrepreneurship in Germany”, Small Business Economics, 28(2/3): 205–221.
Besley, T., and Ghatak, M. (2010) “Property rights and economic development”
(in) D. Rodrik and M. Rosenzweig (Eds.), Handbook of development economics (vol.
V, Chap. 68, pp. 4525–4595). Amsterdam: North-Holland.
Blanchflower D., Oswald, A.; and Stutzer, A. (2001) “Latent entrepreneurship
across nations”, European Economic Review, 45(4-6): 680-691.
Blanchflower, D. (2004) “Self-employment: More may not be better”, NBER
WP No. 10286.
Bourguignon, F., and Morrisson, C. (2002) “Inequality among world citizens:
1820–1992”, American Economic Review, 92(4): 727–744.
Caliendo, M., and Kritikos, A. (2011) “Searching for the entrepreneurial
personality: New evidence and avenues for further research”, IZA DP No. 5790.
Claessens, S., and Perotti, E. (2007) “Finance and inequality: Channels and
evidence”, Journal of Comparative Economics 35: 748-773.
Dargay, J. (2007) “The effect of prices and income on car travel in the UK”,
Transportation Research Part A, 41(10): 949-960.
Deaton, A. (1985) “Panel data from time series of cross-sections”, Journal of
Econometrics, 30(1-2): 109-26.
Djankov, S., La Porta, R., López-de-Silanes, F.; and Shleifer, A. (2002) “The
regulation of entry”, Quarterly Journal of Economics, 117(1):1-37.
Djankov S., R. La Porta, Lopez-De-Silanes F.; and Shleifer A. (2003) “Courts”,
Quarterly Journal of Economics, 118(2): 453-517.
Evans D., and Jovanovic, B. (1989) “An estimated model of entrepreneurial
choice under liquidity constraints”, Journal of Political Economy, 97(4): 808-827.
Ezzati M, Vander Hoorn S, Lawes C., Leach R; and James W. (2005)
“Rethinking the ‘Diseases of affluence’, Paradigm: Global Patterns of Nutritional Risks
in Relation to Economic Development”, PLoS Med 2(5).

100
 Galor, O. (2011) “Inequality, human capital formation and the process of
development”, prepared for the Handbook of the Economics of Education, North-
Holland.
Galor, O., and Zeira, J (1993) “Income distribution and macroeconomics”,
Review of Economic Studies 60(1): 35-52.
Garcia-Mainar, I, and Montuenga-Gomez, V. (2005) “Education returns of
wage earners and self-employed workers: Portugal vs. Spain”, Economics of Education
Review, 24: 161-170.
Ghatak, M., and Jiang. N.H (2002) “A simple model of inequality, occupational
choice and development”, Journal of Development Economics, 69(1): 205-226.
Glaeser, E.; La Porta, R; Lopez-de-Silanes, F.; and Shleifer, A. (2004) “Do
institutions cause growth?”, Journal of Economic Growth, 9(3): 271-303.
Gutiérrez-Romero, R. (2012) “Determinants of Spanish firms’ life cycle and job
creation: A pseudo-panel approach”, Universidad Autónoma de Barcelona WP 12.09.
Hurst E., and Lusardi, A. (2004) “Liquidity constraints, household wealth, and
entrepreneurship”, Journal of Political Economy, 112(2): 319-47.
Kuntchev, V., Ramalho, R., Rodríguez-Mesa, J.; and Yang, J. (2013) “What
Have We Learned from the Enterprise Surveys Regarding Access to Credit by SMEs?”,
World Bank WP No.6670.
La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. W. Vishny. (1998) “Law
and finance”, Journal of Political Economy 106(6): 1113-1155.
La Porta R., F. Lopez-de-Silanes, A. Shleifer, and R. W. Vishny. (1999) “The
quality of government”, Journal of Law, Economics and Organization, 15(1): 222-279.
Levine, R.; Loayza, N., and Beck, T. (2000) “Financial intermediation and
growth: Causality and causes”, Journal of Monetary Economics, 46(1):31-77.
Lucas, R.E. (1978) “On the size distribution of business firms”, Bell Journal of
Economics, 9(2): 508-523.
Man, T.; Lau, T., and Chan, K. (2002) “The competitiveness of small and
medium enterprises A conceptualization with focus on entrepreneurial competencies”,
Journal of Business Venturing 17: 123–142.
Manski, C. (2000) “Economic analysis of social interactions”, Journal of
Economic Perspectives, 14(3): 115-136.
Mesnard, A., and Ravallion, M (2001) “Is inequality bad for business?”, Policy
Research WP 2527, World Bank.

101
 Morrisson, C., and Murtin, F. (2011) “Internal income inequality and global
inequality”, Foundation pour les etudes et recherches sur le développpement
international, WP No. 26.
Murphy, K., Shleifer, A.; and Vishny, R. (1989a) “Industrialization and the big
push”, Journal of Political Economy, 97(5): 1003-1026.
Murphy, K., Shleifer, A.; and Vishny, R. (1989b) “Income Distribution, Market
Size, and Industrialization”, Quarterly Journal of Economics, 104(3): 537-564.
Naudé, W. (2010) “Entrepreneurship, developing countries, and development
economics: new approaches and insights”, Small Business Economics, 34(1): 1-12.
Naudé, W. (2008) “Entrepreneurship in economic development”, UNU-Wider
Research Paper No. 2008/20
O’Neill, B., Sorhaindo, B., Xiao, J. J.; and Garman, E. T. (2005) “Health,
financial well-being, and financial practices of financially distressed consumers”,
Consumer Interests Annual, 51.
Rajan, R.; Zingales, L.; and Kumar, K. (2001) “What Determines Firm Size?”
CRSP WP No. 496.
Reynolds, P., Bosma, N., Autio, E., Hunt, S., De Bono, N., Servais, I.; and
Lopez-Garcia, P. (2005) “Global entrepreneurship monitor: data collection design and
implementation 1998-2003’, Small Business Economics, 24(3): 205–31.
Reynolds, P., Autio, E.; and Hay, M. (2003) “Global Entrepreneurship Monitor
Report”, Kansas City, MO, US: E.M. Kauffmann Foundation
Rosenstein-Rodan, P. N. (1943) “Problems of industrialization of Eastern and
South-Eastern Europe”, The Economic Journal, 53(210/211):202-211.
Shane, S., and Venkataraman, S. (2000) “The promise of entrepreneurship as a
field of research”, Academy of Management Review; 25(1): 217-226.
Simón-Moya, V., Revuelto-Taboada, L.; and Fernández-Guerrero, R. (2014)
“Institutional and economic drivers of entrepreneurhsip: An international perspective”,
Journal of Business Research, 67:715-721.
Thornton, P. (1999) “The sociology of entrepreneurship”, Annual Review of
Sociology. 25(25):19-46
Yanya, M. (2012) “Causal relationship between entrepreneurship poverty and
income inequality in Thailand”, International Journal of Trade, Economics and
Finance, 3(6): 436-440.

102
 TABLES AND FIGURES

Figure 1 Proportion of people engaged in entreprenerual stages

8.0

7.0

6.0

5.0

4.0

3.0

2.0

1.0

0.0
2001 2002 2003 2004 2005 2006 2007 2008 2009

Nascent entrepreneurs Young firms Established firms Firms closed

Source: GEM 2001-2009

103
 Table 1 IV Second Stage Pseudo-Panel Regression: Impact of inequality on firm’s life cycle
(1) (2) (3) (4)
Nascent Young Established Closed
IV IV IV IV
Initial conditions
Log (Ratio 90/10) -0.197*** (0.005) -0.175*** (0.005) -0.087*** (0.004) -0.177*** (0.011)
Log (GDPpc1800) -0.749*** (0.006) -0.698*** (0.006) -0.500*** (0.006) -0.683*** (0.009)
Institutional environment
Log (IndexCreditProtection) 0.997*** (0.011) 0.799*** (0.011) 0.222*** (0.010) 0.707*** (0.011)
Region (reference group: Africa)
Asia 0.206*** (0.010) 1.073*** (0.011) 1.625*** (0.010) 0.727*** (0.019)
Western Europe 0.209*** (0.010) 0.664*** (0.011) 1.325*** (0.010) -0.004 (0.011)
Latin America 1.445*** (0.011) 1.541*** (0.012) 1.476*** (0.011) 1.310*** (0.012)
North America 0.892*** (0.010) 1.029*** (0.011) 1.443*** (0.010) 0.475*** (0.015)
Oceania 0.122*** (0.009) 0.570*** (0.010) 1.384*** (0.009) -0.203*** (0.010)
Eastern Europe 0.297*** (0.010) 0.433*** (0.011) 0.880*** (0.010) -0.018 (0.012)
Individual characteristics
% of individuals with high school or more (at cohort level) -0.142*** (0.006) -0.345*** (0.006) -0.317*** (0.005) -0.121*** (0.026)
Cohort (Male aged 16-28 reference group)
Male 29-38 0.138*** (0.004) 0.182*** (0.004) 1.143*** (0.004) 0.308*** (0.004)
Male 39-48 -0.105*** (0.004) -0.136*** (0.004) 1.355*** (0.004) 0.273*** (0.004)
Male 49-58 -0.570*** (0.005) -0.515*** (0.005) 1.237*** (0.004) 0.247*** (0.005)
Male 59-64 -1.456*** (0.006) -1.420*** (0.006) 0.453*** (0.005) 0.074** (0.037)
Female 16-28 -0.609*** (0.004) -0.593*** (0.004) -0.637*** (0.004) -0.384*** (0.004)
Female 29-38 -0.464*** (0.004) -0.347*** (0.004) 0.369*** (0.004) -0.117*** (0.004)
Female 39-48 -0.686*** (0.004) -0.687*** (0.004) 0.594*** (0.004) -0.186*** (0.004)
Female 49-58 -1.172*** (0.005) -1.172*** (0.004) 0.367*** (0.004) -0.375*** (0.006)
Female 59-64 -2.200*** (0.009) -2.136*** (0.009) -0.333*** (0.005) -0.553*** (0.013)
Year (reference: 2001)
2002 -0.178*** (0.005) -0.054*** (0.005) 0.105*** (0.005) -0.346*** (0.015)
2003 0.023*** (0.005) 0.171*** (0.006) 0.257*** (0.005) -0.106*** (0.020)
2004 -0.333*** (0.006) -0.072*** (0.005) 0.248*** (0.005) -0.223*** (0.018)
2005 -0.134*** (0.005) 0.004 (0.005) 0.349*** (0.005) -0.308*** (0.019)
2006 -0.064*** (0.005) 0.147*** (0.005) 0.417*** (0.005) -0.217*** (0.035)
2007 -0.114*** (0.005) 0.091*** (0.005) 0.435*** (0.005) -0.238*** (0.023)
2008 -0.116*** (0.005) 0.166*** (0.005) 0.687*** (0.005) -0.103*** (0.021)
2009 1.142*** (0.007) 1.418*** (0.007) 1.102*** (0.005)
Constant 0.772*** (0.027) 0.085*** (0.028) -1.636*** (0.026) 0.259*** (0.066)
No. Observations 959,199 942,535 973,873 914,094
R-squared 0.509 0.506 0.603 0.469
F test 31198.78*** 31095.09*** 30728.22*** 27843.20***
K-P Wald rk F statistic (weak identification test) 150,000*** 130,000*** 140,000*** 150,000***
Endogeneity test 5520*** 3866.9*** 150.286*** 2591.045***
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

104
 Table 2 IV Second Stage Pseudo-Panel Regression: Impact of inequality on job creation
Nascent Young Established
IV IV IV
Initial conditions
Log (Ratio 90/10) 0.605*** (0.126) -0.304*** (0.066) -0.165*** (0.024)
Log (GDPpc1800) -0.792*** (0.127) 0.093 (0.060) 0.087*** (0.022)
Institutional environment
Log(IndexCreditProtection) Total effect 1 7.023*** (0.703) 1.996*** (0.283) 2.224*** (0.218)
Ommited: Log(IndexCreditProtection)*Africa
Log(IndexCreditProtection) 0.529*** (0.162) 0.508*** (0.079) 0.332*** (0.051)
Log(IndexCreditProtection)*Asia 2.028*** (0.216) 0.514*** (0.090) 0.344*** (0.049)
Log(IndexCreditProtection)*Western Europe 1.354*** (0.147) 0.113* (0.058) 0.352*** (0.045)
Log(IndexCreditProtection)*Latin America 1.941*** (0.196) 0.505*** (0.066) 0.436*** (0.044)
Log(IndexCreditProtection)*North America -0.390 (0.284) 0.060 (0.085) 0.181*** (0.055)
Log(IndexCreditProtection)*Oceania 0.280** (0.136) -0.313*** (0.078) 0.118** (0.047)
Log(IndexCreditProtection)*Eastern Europe 1.282*** (0.148) 0.609*** (0.058) 0.461*** (0.044)
Individual characteristics

% of individuals with high school or more (at cohort level) -0.669*** (0.117) -0.114** (0.051) 0.066*** (0.023)
Cohort (Male aged 16-28 reference group)
Male 29-38 -0.119* (0.063) -0.073** (0.030) 0.024 (0.020)
Male 39-48 0.118 (0.079) 0.013 (0.037) 0.066*** (0.019)
Male 49-58 -0.584*** (0.122) -0.204*** (0.047) -0.043* (0.023)
Male 59-64 0.018 (0.164) -0.171* (0.091) -0.270*** (0.030)
Female 16-28 -0.548*** (0.099) -0.393*** (0.034) -0.339*** (0.025)
Female 29-38 -0.802*** (0.079) -0.308*** (0.029) -0.369*** (0.021)
Female 39-48 -0.676*** (0.072) -0.601*** (0.039) -0.337*** (0.022)
Female 49-58 0.328 (0.212) -0.359*** (0.103) -0.472*** (0.025)
Female 59-64 -0.510* (0.310) -0.620*** (0.093) -0.577*** (0.031)
Technology sector (reference: No/ Low technology sector)
Medium or high -0.003 (0.068) 0.069* (0.036) 0.025 (0.020)
Sector (reference: Extractive sector)
Transforming sector 0.095 (0.088) 0.057 (0.039) 0.064*** (0.015)
Business services 0.146 (0.092) 0.024 (0.041) 0.099*** (0.016)
Consumer oriented 0.041 (0.088) 0.030 (0.039) 0.014 (0.014)
Year (reference: 2001)
2002 -0.378 (0.245) -0.258*** (0.098) 0.146*** (0.025)
2003 0.831*** (0.270) 0.038 (0.094) 0.314*** (0.028)
2004 0.479** (0.243) -0.363*** (0.093) 0.084*** (0.023)
2005 0.465* (0.255) -0.147 (0.094) 0.188*** (0.024)
2006 0.063 (0.273) -0.316*** (0.093) 0.170*** (0.024)
2007 0.009 (0.220) -0.157* (0.093) 0.141*** (0.025)
2008 -0.615** (0.241) -0.277*** (0.095) 0.132*** (0.025)
2009 0.436*** (0.029)
Constant 2.758*** (1.046) 0.660 (0.514) 0.168 (0.169)
No. Observations 6,952 22,119 53,067
F test 933.11*** 1833.33*** 3332.82***
K-P Wald rk F statistic (weak identification test) 27.24*** 106.994*** 317.925***
Endogeneity test 28.58*** 62.53*** 489.05***
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

105
 APPENDIX

Table A.1 Independent variables

GDPpc Ratio90/10 Gini Ratio 90/10 GDPpc
Country Freq. Percent Region
1800 1800 2000 Circa2000 2000
Argentina 871.63 6.00 47.90 10.73 5472.71 18,070 1.80 Latin America
Australia 671.48 7.75 44.60 . 30869.66 12,646 1.26 Oceania
Austria 1434.51 5.42 29.20 4.46 35027.30 4,199 0.42 Western Europe
Bosnia and Herzegovina 490.82 12.00 26.00 . 2241.70 4,016 0.40 Eastern Europe
Brazil 509.20 9.38 58.61 17.84 4406.71 20,000 2.00 Latin America
Canada 1159.50 7.75 32.40 5.59 33373.13 7,008 0.70 North America
Chile 702.10 6.00 59.50 16.03 6551.57 19,005 1.90 Latin America
China 985.89 5.22 40.30 6.79 1122.26 14,443 1.44 Asia
Colombia 522.98 9.38 57.22 20.99 3074.32 8,157 0.81 Latin America
Croatia 1227.06 12.00 30.00 4.26 8141.48 16,013 1.60 Eastern Europe
Czech Republic 1622.74 5.42 23.82 3.06 10378.64 2,001 0.20 Eastern Europe
Denmark 1342.84 6.19 35.00 8.47 45339.69 26,083 2.60 Western Europe
Egypt 748.79 5.50 54.20 12.02 1140.12 2,603 0.26 Africa
Finland 1037.69 6.19 27.00 3.59 33217.07 18,044 1.80 Western Europe
France 1388.32 6.11 28.20 3.75 32392.16 17,947 1.79 Western Europe
Germany 1695.68 5.42 29.20 4.50 32662.13 58,535 5.84 Western Europe
Greece 934.41 12.00 32.30 5.33 18040.57 13,970 1.39 Western Europe
Hungary 1390.67 5.42 24.96 3.44 8810.08 17,726 1.77 Eastern Europe
Iceland 801.90 6.19 46985.73 16,025 1.60 Western Europe
India 562.88 4.80 36.00 4.76 578.22 10,751 1.07 Asia
Indonesia 514.12 4.30 36.50 4.45 1086.05 2,000 0.20 Asia
Ireland 1213.17 10.00 34.10 5.06 41953.95 15,888 1.59 Western Europe
Italy 1339.84 6.11 35.80 7.08 29872.29 20,744 2.07 Western Europe
Japan 1055.06 6.75 31.88 . 33956.81 17,270 1.72 Asia
Korea 596.01 5.22 36.90 11.06 15162.35 8,023 0.80 Asia
Macedonia 800.76 12.00 35.14 9.12 2698.55 1,971 0.20 Eastern Europe
Mexico 1053.60 9.38 53.50 15.22 7689.10 9,448 0.94 Latin America
Netherlands 2412.43 5.92 25.50 3.50 37546.78 29,625 2.96 Western Europe
New Zealand 541.62 7.75 40.20 9.63 24260.06 8,868 0.89 Oceania
Norway 950.00 6.19 27.40 3.55 60726.25 19,921 1.99 Western Europe
Peru 697.30 9.38 49.62 14.41 2309.83 9,985 1.00 Latin America
Philippines 626.97 6.47 42.20 6.47 1060.55 2,000 0.20 Asia
Poland 1198.39 12.00 34.18 7.04 6824.47 6,001 0.60 Eastern Europe
Portugal 1284.41 6.11 34.70 5.00 17891.38 5,023 0.50 Western Europe
Romania 12.00 30.27 4.76 3326.62 6,218 0.62 Eastern Europe
Russia 823.99 12.00 48.40 9.02 3870.32 9,378 0.94 Eastern Europe
Serbia 1308.87 12.00 . 4.25 2588.74 6,776 0.68 Eastern Europe
Slovenia 1357.95 12.00 24.84 3.46 15033.47 21,138 2.11 Eastern Europe
South Africa 759.05 10.83 60.10 8.93 4652.34 24,865 2.48 Africa
Spain 1443.02 6.11 32.60 4.57 23920.93 158,307 15.81 Western Europe
Sweden 1100.00 6.19 27.20 3.91 36576.19 38,786 3.87 Western Europe
Switzerland 1612.48 5.92 31.80 . 50188.18 13,632 1.36 Western Europe
Taiwan 871.27 5.22 31.90 4.51 . 2,236 0.22 Asia
Thailand 496.98 6.47 33.33 2205.78 7,043 0.70 Asia
Turkey 869.92 9.58 39.80 . 6119.23 7,217 0.72 Asia
UK 2716.87 10.00 34.60 5.17 34058.66 197,518 19.72 Western Europe
United States 1912.62 6.84 40.10 8.93 40965.03 36,848 3.68 North America
Venezuela 442.02 9.38 45.80 11.61 5255.69 7,487 0.75 Latin America
Total 1,001,458 100
Sources : Historical data on GDPpc Madisson's database, GDP per capita 2000 (constant 2005 US$) World Bank Indicators.

106
 Table A.2 Variable definitions and sources
Variable notation Definition Source

Depedent variables
Entrepreneurial stages:
Nascent firms % proportion of individuals involved in setting up a business they will own or co-own,
GEM
but has not paid any payments for more than 3 months (in natural logarithms).
Young firms % proportion of individuals that owners-manages firms, defined as having paid
GEM
salaries for more than 3 months and less than 3.5 years (in natural logarithms).
Established firms % proportion of individuals that owners-manages firms, defined as having paid
GEM
salaries for more than3.5 year (in natural logarithms)s.
Closed firms % proportion of individuals that owned-managed firms that in the past 12 months
GEM
have been sold, shut down, discontinue or quit business (in natural logarithms).
Firm size at different stages:
Nascent firms Number of employees of nascent firms (Log transformation: 1+ number of jobs) GEM
Young firms Number of employees of young firms (Log transformation: 1+ number of jobs) GEM
Established firms Number of employees of established firms (Log transformation: 1+ number of jobs) GEM

Independent variables
Historical data
The 90/10 ratio measures the income of those individuals at the 90th and those at the
Log (Ratio 90/10) 10th percentiles.Higher values of the ratio measures greatest income inequality. Bourguignon and Morrison (2002)

Log (GDPpc1800) Gross Domestic Product per capita in 1820 Angus Maddison's historic income database
Business environment
Measures the degree of which collateral and bankruptcy laws protect the right of
borrowers and lenders and thus facilitate lending. The index ranks from 0 to 10;
Log(IndexCreditProtection) World Bank
higher scores indicating that collateral and bankruptcy laws are better designed to
expand access to credit.
Dummy variable: 1 if the country is classified as low or medium income country; 0 Low-medium countries are those which mean gdp per capita
Low_medium otherwise for the considered period are below 13,000 USD dollars.
Classification according to the World Bank.

107
 Table A.2 Variable definitions and sources (cont.)
Variable notation Definition Source
Regional dummies
Africa Dummy variable: 1 Africa; 0 otherwise Own classification
Asia Dummy variable: 1 Asia; 0 otherwise
Western Europe Dummy variable: 1 Western Europe; 0 otherwise
Latin America Dummy variable: 1 Latin America; 0 otherwise
North America Dummy variable: 1 North America; 0 otherwise
Oceania Dummy variable: 1 Oceania; 0 otherwise
Eastern Europe Dummy variable: 1 Eastern Europe; 0 otherwise
Individual variables at cohort levels
% of individuals with high school Proportion of individuals in the cohort c with post-secondary level or more living in
GEM
or more (at cohort level) country i in year j
% of individual that provided
Proportion of individuals in cohort c that provided credit to others (excluding family
credit to network (at cohort GEM
members) living in country i in year j
levels)
Male aged 16-28 Proportion of males aged 16-38 years living in country i in year j
Male 29-38 Proportion of males aged 29-38 years living in country i in year j GEM
Male 39-48 Proportion of males aged 39-48 years living in country i in year j GEM
Male 49-58 Proportion of males aged 49-58 years living in country i in year j GEM
Male 59-64 Proportion of males aged 59-64 years living in country i in year j GEM
Female 16-28 Proportion of females aged 16-28 years living in country i in year j GEM
Female 29-38 Proportion of females aged 29-38 years living in country i in year j GEM
Female 39-48 Proportion of females aged 39-48 years living in country i in year j GEM
Female 49-58 Proportion of females aged 49-58 years living in country i in year j GEM
Female 59-64 Proportion of females aged 59-64 years living in country i in year j GEM
Sector
Extractive sector Dummy variable: 1 if the firm is involved in extractive activities; 0 otherwise GEM
Transforming sector Dummy variable: 1 if the firm is involved in transforming activities; 0 otherwise GEM
Business services Dummy variable: 1 if the firm is involved in business services; 0 otherwise GEM
Consumer oriented Dummy variable: 1 if the firm is involved in consumer oriented activities; 0 otherwise GEM
Dummy variable: 1 if the firm is intensive in techonology sector (medium orhigh); 0
Medium or high GEM
otherwise

108
 Table A.2 Variable definitions and sources (cont.)
Variable notation Definition Source
Instrumental variables
QOG The Quality of Government Institute (Original source: La
English Common Law Dummy variable: 1 if the country has english legal origin; 0 otherwise Porta, López-de- Silanes, Shleifer & Vishny).
http://www.qog.pol.gu.se/data/datadownloads/qogstandarddata/
QOG The Quality of Government Institute (Original source: La
French Commercial Code Dummy variable: 1 if the country hasfrench legal origin; 0 otherwise Porta, López-de- Silanes, Shleifer & Vishny).
http://www.qog.pol.gu.se/data/datadownloads/qogstandarddata/
QOG The Quality of Government Institute (Original source: La
Socialist/Communist Laws Dummy variable: 1 if the country has socialist/communist legal origin; 0 otherwise Porta, López-de- Silanes, Shleifer & Vishny).
http://www.qog.pol.gu.se/data/datadownloads/qogstandarddata/
QOG The Quality of Government Institute (Original source: La
German Commercial Code Dummy variable: 1 if the country has german legal origin; 0 otherwise Porta, López-de- Silanes, Shleifer & Vishny).
http://www.qog.pol.gu.se/data/datadownloads/qogstandarddata/
QOG The Quality of Government Institute (Original source: La
Scandinavian Commercial Code Dummy variable: 1 if the country has scandinavian legal origin; 0 otherwise Porta, López-de- Silanes, Shleifer & Vishny).
http://www.qog.pol.gu.se/data/datadownloads/qogstandarddata/
QOG The Quality of Government Institute.
colonia_spain
http://www.qog.pol.gu.se/data/datadownloads/qogstandarddata/
Blood pressure The mean SBP (Systolic Blood Pressure) of the male population, counted in mm- School of Public Health, Imperial College
Hg; this mean is calculated as if each country has the same age composition as the London.http://www1.imperial.ac.uk/publichealth/departments/eb
world population. s/projects/eresh/majidezzati/healthmetrics/metabolicriskfactors/
Colestherol The mean SBP (Systolic Blood Pressure) of the male population, counted in mm- School of Public Health, Imperial College
Hg; this mean is calculated as if each country has the same age composition as the London.http://www1.imperial.ac.uk/publichealth/departments/eb
world population. s/projects/eresh/majidezzati/healthmetrics/metabolicriskfactors/

109
 Table A.3 Summary of main variables
Year 2001 2002 2003 2004 2005 2006 2007 2008 2009
% of people involved in
Nascent firms 4.32 3.68 4.22 3.15 3.58 3.65 3.72 4.21 3.37
Young firms 2.75 2.97 3.17 2.77 3.21 3.53 3.58 3.63 3.09
Established firms 4.57 5.54 5.75 5.50 6.64 5.98 6.33 7.92 7.05
Closed firms 2.83 2.99 2.32 2.73 2.50 2.61 3.14 2.63
% of people
Education high school or more 63.26 59.94 72.91 55.87 56.93 66.82 64.75 69.13 71.65
Provided credit to the network 0.78 1.05 0.98 0.81 0.95 0.98 1.14 1.16 0.96
Firm's size by entrepreneurial
Nascent firms 2 7 3 4 3 4 11 11
Young firms 8 6 5 7 6 6 5 7
Established firms 8 13 15 9 11 9 10 10 10
Sector of activity
Extractive sector 9.15 7.98 8.99 9.93 6.24 8.75 7.23 8.56 9.97
Transforming sector 29.19 28.80 27.36 30.48 26.86 31.74 28.83 28.16 24.12
Business services 21.23 22.00 22.74 21.24 21.42 17.24 21.57 19.07 15.19
Consumer oriented 40.42 41.23 40.91 38.35 45.49 42.26 42.37 44.20 50.71
Medium/high technology intensity 7.78 7.09 7.09 7.07 7.06 4.91 5.61 5.12 3.10
Obs. 62,598 115,418 92,228 140,537 110,870 171,465 153,657 133,793 156,825

110
 Table A.4 Summary of main variables grouping by country GDP per capita
(World Bank classification)

High-income countries
Year 2001 2002 2003 2004 2005 2006 2007 2008 2009
% of people involved in
Nascent firms 3.4 3.1 3.5 2.5 3.0 2.8 3.0 3.1 2.4
Young firms 2.5 2.6 2.6 2.4 2.8 2.6 2.8 3.1 2.3
Established firms 4.5 5.6 5.5 5.4 6.9 5.1 5.9 8.0 6.7
Closed firms 2.3 2.2 1.8 1.7 1.7 1.7 1.9 1.9
% of people
Education high school or more 70.3 65.5 76.6 58.7 61.5 70.7 63.7 72.5 0.8
Provided credit to the network 0.8 1.2 1.0 0.8 1.0 0.8 0.9 0.9 7.5
Firm's size by entrepreneurial
stage
Nascent firms - 2 5 3 4 5 5 3 5
Young firms - 6 5 4 10 8 8 6 4
Established firms 3 3 5 3 4 5 6 4 7.6
Sector of activity
Extractive sector 10.73 8.97 10.71 11.16 6.98 9.17 7.9 9.26 25.2
Transforming sector 29.3 29.0 27.3 30.9 27.8 30.4 28.9 27.6 19.1
Business services 25.2 26.8 26.9 24.4 26.0 24.7 25.9 23.2 46.0
Consumer oriented 34.8 35.2 35.1 33.6 39.2 35.8 37.3 39.9 3.12
Medium/high technology intensity 9.42 9.02 7.01 7.6 7.97 5.58 6.52 5.83
Obs. 48,754 87,073 79,610 118,375 84,489 125,443 113,242 79,718 104,391

Low income countries

Year 2001 2002 2003 2004 2005 2006 2007 2008 2009
% of people involved in
Nascent firms 7.1 5.3 8.9 7.2 5.5 6.3 6.0 6.4 6.3
Young firms 3.4 3.9 6.6 4.9 4.5 6.4 6.2 4.6 5.4
Established firms 4.6 5.5 7.4 6.3 5.8 8.7 7.8 7.7 8.0
Closed firms 4.5 8.0 5.1 6.4 5.3 5.6 5.7 4.6
% of people
Education high school or more 41.3 44.8 49.8 39.3 41.2 54.4 68.1 62.2 65.5
Provided credit to the network 0.8 0.8 1.0 0.8 0.9 1.7 1.9 1.6 1.5
Firm's size by entrepreneurial
stage
Nascent firms - 1 9 2 3 3 4 8 3
Young firms - 8 7 5 6 5 6 5 8
Established firms 8 7 7 6 6 7 7 5 7
Sector of activity
Extractive sector 5.73 5.64 4.5 5.99 4.25 8.29 5.91 7.58 10.35
Transforming sector 28.9 28.3 27.6 29.1 24.3 33.3 28.7 28.9 22.8
Business services 12.7 10.7 11.9 11.2 9.3 8.9 13.0 13.2 10.1
Consumer oriented 52.7 55.4 56.0 53.7 62.2 49.5 52.4 50.3 56.8
Medium/high technology intensity 4.24 3.11 7.37 5.31 4.52 3.82 3.91 4.04 3.07
Obs. 13,844 28,345 12,618 22,162 26,381 46,022 40,415 54,075 52,434

111
 Table A.5 Number of Observations per Cohort
Cohort Freq. Percent
<29male 118,663 11.85
>28male 87,396 8.73
>38male 82,135 8.2
>48male 70,088 7.0
>58male 107,228 10.71
<29female 121,738 12.16
>28female 106,129 10.6
>38female 98,491 9.83
>48female 82,431 8.23
>58female 127,159 12.7
Total 1,001,458 100

Table A.6 Firms’ characteristics by entrepreneurial stages by region

Africa Asia Western Europe Latin America North America Oceania Eastern Europe
Nascent firms
Extractive sector 3.76 6.15 4.89 4.49 5.34 10.3 8.24
Transforming sector 23.94 22.25 24.2 23.85 22.43 24.83 30.29
Business services 11.38 9.21 26.8 10.63 27.77 27.24 21.42
Consumer oriented 60.92 62.39 44.11 61.03 44.46 37.62 40.05
Medium/High technology 4.16 5.3 7.79 5.83 7.6 11.2 6.35
Firm size 2 4 3 5 2 4 3
Young firms
extractive sector 3.72 5.2 6.25 3.49 5.15 10.41 9.03
transforming sector 25.41 25.46 26.83 28.9 25.56 26.07 31.33
business services 14.26 9.08 26.93 11.8 29.87 30.36 22.83
consumer oriented 56.61 60.26 40 55.81 39.42 33.15 36.81
High tech 5.2 4.06 7.37 5.22 6.56 9 5.09
Firm size 4 11 5 5 7 4 10
Established firms
extractive sector 5 9.79 12.37 5.33 11.36 18.35 12.99
transforming sector 31.6 27.41 30.34 32.09 24.76 30.43 34.82
business services 13.4 8.07 22.1 12.51 31.82 25.85 19.72
consumer oriented 50 54.72 35.18 50.07 32.07 25.37 32.47
High tech 2.73 3 5.87 3.89 5.38 7.5 6.24
Firm size 8 11 10 5 19 11 13

Table A.7 Pairwise correlations of main variables

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
(1) Nascent firms 1.00
(2) Young firms 0.76 1.00
(3) Established firms 0.37 0.48 1.00
(4) Firms closed 0.57 0.53 0.44 1.00
(5) Jobs nascent firms -0.01 0.09 0.19 0.03 1.00
(6) Jobs young firms 0.17 0.15 0.13 0.13 0.07 1.00
(7) Jobs established firms 0.05 0.05 0.17 0.05 0.07 0.21 1.00
(8) Informal credit 0.47 0.42 0.33 0.48 0.17 0.26 0.15 1.00
(9) Ratio 90/10 -0.05 -0.14 -0.17 -0.06 -0.12 -0.11 -0.03 -0.23 1.00
(10) Credit -0.21 -0.24 -0.13 -0.27 -0.08 0.01 0.25 -0.08 0.19 1.00
(11) GDP pc 1800 -0.37 -0.35 -0.17 -0.46 0.00 0.02 0.29 -0.26 0.17 0.59 1.00
(12) High school or more 0.09 0.04 -0.05 -0.04 -0.09 0.03 0.12 0.09 0.15 0.20 0.27 1.00
112
 Table A.8 IV First Stage Pseudo-Panel Regression:
Impact of inequality on firm’s life cycle
Nascent firms Young firms Established firms Closed firms
Log(Index Log(Index Log(Index Log(Index
CreditProtection) CreditProtection) CreditProtection) CreditProtection)
Initial conditions
Log (Ratio 90/10) -0.229*** (0.002) -0.224*** (0.002) -0.221*** (0.002) -0.239*** (0.002)
Log (GDPpc1800) 0.284*** (0.001) 0.294*** (0.001) 0.291*** (0.001) 0.284*** (0.002)
Region (reference group: Africa)
Asia -0.240*** (0.005) -0.249*** (0.005) -0.232*** (0.005) -0.244*** (0.006)
Western Europe -0.136*** (0.005) -0.146*** (0.005) -0.145*** (0.005) -0.111*** (0.005)
Latin America -0.392*** (0.005) -0.401*** (0.005) -0.389*** (0.005) -0.378*** (0.005)
North America -0.325*** (0.005) -0.332*** (0.005) -0.325*** (0.005) -0.300*** (0.006)
Oceania 0.155*** (0.004) 0.158*** (0.004) 0.163*** (0.004) 0.171*** (0.004)
Eastern Europe -0.031*** (0.005) -0.046*** (0.005) -0.035*** (0.005) -0.011** (0.005)
Year (reference: 2001)
2002 -0.013*** (0.002) -0.021*** (0.002) -0.012*** (0.002) -0.006* (0.003)
2003 -0.105*** (0.002) -0.119*** (0.002) -0.106*** (0.002) -0.092*** (0.004)
2004 -0.059*** (0.002) -0.053*** (0.002) -0.046*** (0.002) -0.046*** (0.004)
2005 -0.056*** (0.002) -0.057*** (0.002) -0.045*** (0.002) -0.040*** (0.004)
2006 -0.055*** (0.002) -0.056*** (0.002) -0.045*** (0.002) -0.050*** (0.007)
2007 -0.057*** (0.002) -0.069*** (0.002) -0.043*** (0.002) -0.036*** (0.005)
2008 0.036*** (0.002) 0.024*** (0.002) 0.052*** (0.002) 0.063*** (0.004)
2009 -0.036*** (0.002) -0.039*** (0.003) -0.033*** (0.003) 0.469*** (0.001)
Individual characteristics
% of individuals with high school or
more (at cohort level) 0.143*** (0.002) 0.159*** (0.002) 0.156*** (0.002) 0.135*** (0.005)
Cohort (Male aged 16-28 reference group)
Male 29-38 0.003 (0.002) 0.003 (0.002) 0.006*** (0.002) 0.003* (0.002)
Male 39-48 0.017*** (0.002) 0.020*** (0.002) 0.016*** (0.002) 0.015*** (0.002)
Male 49-58 0.031*** (0.002) 0.040*** (0.002) 0.033*** (0.002) 0.030*** (0.002)
Male 59-64 0.058*** (0.002) 0.059*** (0.002) 0.064*** (0.002) 0.036*** (0.007)
Female 16-28 0.002 (0.002) 0.002 (0.002) 0.001 (0.002) 0.004*** (0.002)
Female 29-38 0.002 (0.002) 0.004** (0.002) 0.006*** (0.002) 0.001 (0.002)
Female 39-48 0.020*** (0.002) 0.024*** (0.002) 0.017*** (0.002) 0.016*** (0.002)
Female 49-58 0.034*** (0.002) 0.045*** (0.002) 0.037*** (0.002) 0.038*** (0.002)
Female 59-64 0.068*** (0.002) 0.063*** (0.002) 0.077*** (0.002) 0.048*** (0.003)
Legal origin (reference: other legal origin)
English 0.458*** (0.001) 0.448*** (0.001) 0.454*** (0.001) 0.469*** (0.001)
Constant 0.229*** (0.010) 0.153*** (0.010) 0.143*** (0.010) 0.213*** (0.015)
No. Observations 959,199 942,535 973,873 914,094
R-squared 0.526 0.535 0.518 0.536
Shea Partial R2 0.1458 0.1404 0.1441 0.1507
Partial R2 0.1458 0.1404 0.1441 0.1507
F statistic test excluded instruments 150,000*** 130,000*** 140,000*** 150,000***
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

113
 Table A.9.1 IV First Stage Pseudo-Panel Regression:
Impact of inequality on job creation in nascent firms
Asia* Western Europe* Latin America* North America* Oceania* Eastern Europe*
Log(IndexCredit
Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit
Protection)
Protection) Protection) Protection) Protection) Protection) Protection)
Initial conditions
Log (Ratio 90/10) -0.114*** (0.022) -0.619*** (0.024) 0.206*** (0.027) -0.101*** (0.016) -0.192*** (0.017) 0.062*** (0.021) 0.339*** (0.027)
Log (GDPpc1800) 0.107*** (0.012) -0.059*** (0.018) 1.102*** (0.017) 0.005 (0.006) 0.260*** (0.018) -1.191*** (0.026) -0.126*** (0.015)
Individual characteristics
% of individuals with
0.176*** (0.020) 0.123*** (0.023) -0.409*** (0.025) 0.180*** (0.020) 0.094*** (0.014) -0.074*** (0.024) 0.138*** (0.019)
high school or more
Cohort (Male aged 16-28 reference group)
Male 29-38 0.017 (0.011) 0.098*** (0.016) 0.100*** (0.017) -0.011 (0.007) -0.082*** (0.013) -0.036** (0.017) -0.036*** (0.014)
Male 39-48 -0.004 (0.011) 0.085*** (0.018) 0.093*** (0.020) -0.043*** (0.009) -0.050*** (0.014) -0.025 (0.015) 0.013 (0.020)
Male 49-58 0.045** (0.018) 0.072*** (0.023) 0.101*** (0.022) 0.044*** (0.011) -0.083*** (0.014) 0.035 (0.025) -0.051*** (0.017)
Male 59-64 0.027 (0.026) 0.037 (0.029) 0.054 (0.039) -0.001 (0.013) -0.014 (0.016) -0.021 (0.027) 0.057** (0.028)
Female 16-28 0.018 (0.017) -0.025 (0.017) 0.192*** (0.023) 0.082*** (0.010) -0.087*** (0.013) -0.004 (0.018) -0.120*** (0.025)
Female 29-38 0.007 (0.012) 0.146*** (0.018) 0.178*** (0.019) 0.032*** (0.006) -0.100*** (0.019) -0.107*** (0.022) -0.062*** (0.016)
Female 39-48 0.043*** (0.011) 0.040** (0.020) 0.032 (0.020) 0.050*** (0.011) -0.051*** (0.013) 0.039* (0.021) 0.037** (0.018)
Female 49-58 -0.067*** (0.022) 0.003 (0.040) -0.034 (0.045) -0.029*** (0.011) 0.174*** (0.052) -0.087 (0.053) -0.026 (0.022)
Female 59-64 0.018 (0.028) 0.166*** (0.047) 0.189*** (0.059) -0.047** (0.021) -0.151*** (0.022) 0.049* (0.029) -0.113*** (0.033)
Technology sector (reference: No/ Low technology sector)
Medium or high -0.028** (0.014) -0.002 (0.016) -0.002 (0.019) -0.011 (0.010) 0.016 (0.015) -0.012 (0.017) -0.011 (0.017)
Sector (reference: Extractive sector)
Transforming sector -0.033** (0.015) -0.059*** (0.022) 0.008 (0.024) -0.010 (0.011) -0.018 (0.016) -0.042* (0.024) 0.066*** (0.019)
Business services -0.021 (0.015) -0.031 (0.022) 0.014 (0.025) -0.014 (0.011) -0.025 (0.017) -0.008 (0.024) 0.030 (0.020)
Consumer oriented -0.028** (0.014) -0.028 (0.021) 0.016 (0.023) -0.009 (0.010) -0.021 (0.015) -0.044** (0.022) 0.043** (0.018)
Year (reference: 2001)
2002 -0.042 (0.028) -0.214*** (0.036) -0.505*** (0.043) -0.011 (0.015) 0.132*** (0.020) -0.264*** (0.032) 0.568*** (0.032)
2003 -0.198*** (0.029) -0.248*** (0.035) -0.326*** (0.044) -0.179*** (0.021) 0.193*** (0.023) -0.142*** (0.031) 0.404*** (0.032)
2004 0.002 (0.027) -0.262*** (0.037) -0.425*** (0.045) -0.035** (0.016) 0.087*** (0.018) -0.060** (0.027) 0.503*** (0.032)
2005 0.001 (0.027) -0.357*** (0.035) -0.312*** (0.043) -0.039*** (0.015) 0.137*** (0.020) -0.194*** (0.031) 0.484*** (0.031)
2006 -0.000 (0.027) -0.220*** (0.038) -0.434*** (0.050) -0.053*** (0.016) 0.147*** (0.021) -0.161*** (0.031) 0.572*** (0.038)
2007 -0.004 (0.028) -0.104*** (0.033) -0.218*** (0.042) 0.029* (0.015) 0.076*** (0.018) -0.119*** (0.030) 0.386*** (0.035)
2008 0.311*** (0.030) -0.235*** (0.036) -0.085* (0.044) 0.206*** (0.020) 0.203*** (0.022) -0.240*** (0.029) 0.511*** (0.037)
Legal origin (reference: English)
French -0.673*** (0.012) -0.030 (0.020) 0.302*** (0.021) 0.009 (0.010) -0.247*** (0.018) -0.515*** (0.022) 0.078*** (0.015)
Socialist/Communist -0.322*** (0.023) -0.009 (0.024) -0.955*** (0.025) 0.010 (0.008) -0.139*** (0.015) -0.004 (0.023) 1.311*** (0.032)
German -0.267*** (0.016) 0.422*** (0.036) 0.412*** (0.031) -0.043*** (0.010) -0.213*** (0.017) -0.644*** (0.022) -0.100*** (0.019)
Scandinavian -0.569*** (0.012) -0.103*** (0.020) 0.731*** (0.021) -0.021** (0.009) -0.137*** (0.013) -0.760*** (0.021) -0.094*** (0.015)
Colonial origin (reference: other colonial origins or never colonized by a western oversea)
Spain -0.041*** (0.011) -0.663*** (0.015) -0.684*** (0.017) 1.351*** (0.013) 0.138*** (0.012) -0.150*** (0.012) 0.030**
Blood pressure -0.017*** (0.002) -0.051*** (0.003) 0.057*** (0.003) -0.016*** (0.001) -0.038*** (0.002) -0.055*** (0.002) 0.027*** (0.001)
Colestherol 0.425*** (0.030) -1.094*** (0.036) 0.315*** (0.029) 0.178*** (0.017) -0.015 (0.014) 1.743*** (0.036) 0.321*** (0.025)
Constant 1.568*** (0.171) 14.398*** (0.266) -15.835*** (0.289) 1.253*** (0.115) 3.710*** (0.251) 7.206*** (0.313) -5.585*** (0.203)
No. Observations 6,952 6,952 6,952 6,952 6,952 6,952 6,952
R-squared 0.722 0.799 0.893 0.914 0.301 0.646 0.772
Partial R2 of excluded
0.63 0.7021 0.7708 0.8788 0.2216 0.5490 0.7344
instruments
Shea R2 0.1942 0.1268 0.2161 0.1522 0.1107 0.2703 0.2599
F statistic test
1117.91*** 1474.49*** 1931.21*** 7744.93*** 35.89*** 262.46*** 361.21***
excluded instruments
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

114
 Table A.9.2 IV First Stage Pseudo-Panel Regression:
Impact of inequality on job creation in young firms
Asia* Western Europe* Latin America* North America* Oceania* Eastern Europe*
Log(IndexCredit
Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit
Protection)
Protection) Protection) Protection) Protection) Protection) Protection)

Initial conditions
Log (Ratio 90/10) -0.028** (0.012) -0.559*** (0.014) 0.119*** (0.019) 0.172*** (0.010) 0.038*** (0.010) -0.215***(0.011) 0.398*** (0.017)
Log (GDPpc1800) 0.193*** (0.008) -0.211*** (0.011) 0.956*** (0.012) -0.139*** (0.005) 0.329*** (0.012) -0.677***(0.020) -0.208*** (0.007)
Individual characteristics
% of individuals with high
school or more (at cohort 0.145*** (0.012) 0.065*** (0.015) -0.337*** (0.017) -0.009 (0.012) 0.201*** (0.013) -0.015 (0.017) 0.163*** (0.012)
level)
Cohort (Male aged 16-28 reference group)
Male 29-38 0.012 (0.008) 0.001 (0.011) -0.002 (0.013) 0.002 (0.007) -0.006 (0.010) 0.012 (0.012) 0.001 (0.009)
Male 39-48 0.017* (0.009) -0.039*** (0.013) 0.002 (0.014) 0.010 (0.008) 0.014 (0.013) 0.019 (0.015) 0.005 (0.011)
Male 49-58 0.042*** (0.011) -0.049*** (0.017) 0.040** (0.020) -0.020** (0.009) -0.005 (0.016) 0.061*** (0.018) -0.002 (0.014)
Male 59-64 0.037 (0.031) -0.089*** (0.022) -0.122*** (0.037) 0.026 (0.026) 0.120*** (0.040) 0.064 (0.045) 0.085*** (0.030)
Female 16-28 -0.001 (0.009) -0.026* (0.014) -0.008 (0.016) 0.012 (0.008) 0.015 (0.014) 0.012 (0.018) -0.005 (0.010)
Female 29-38 0.023*** (0.008) -0.037*** (0.011) 0.030** (0.014) 0.026*** (0.008) 0.007 (0.012) 0.022 (0.014) -0.029*** (0.011)
Female 39-48 0.042*** (0.032) -0.028** (0.023) -0.021 (0.042) 0.025** (0.036) 0.029* (0.038) 0.040** (0.045) 0.009 (0.030)
Female 49-58 0.033 (0.028) -0.056*** (0.020) -0.080*** (0.027) -0.035** (0.016) 0.079*** (0.021) 0.024 (0.019) 0.044 (0.028)
Female 59-64 0.041 (0.032) -0.143*** (0.024) -0.108*** (0.041) 0.044 (0.035) 0.178*** (0.041) 0.063 (0.044) 0.064** (0.030)
Technology sector (reference: No/ Low technology sector)
Medium or high -0.017* (0.010) 0.019 (0.013) 0.041** (0.017) -0.010 (0.009) -0.019 (0.015) 0.002 (0.016) -0.042*** (0.012)
Sector (reference: Extractive sector)
Transforming sector -0.030*** (0.011) 0.001 (0.013) -0.007 (0.016) 0.040*** (0.008) -0.007 (0.013) -0.044***(0.016) -0.032** (0.013)
Business services -0.030*** (0.011) -0.009 (0.014) -0.066*** (0.017) 0.013* (0.008) 0.036** (0.015) -0.005 (0.018) -0.020 (0.013)
Consumer oriented -0.043*** (0.011) 0.049*** (0.013) 0.011 (0.016) 0.024*** (0.008) -0.016 (0.012) -0.086***(0.016) -0.037*** (0.012)
Year (reference: 2001)
2002 -0.104*** (0.032) -0.110*** (0.024) -0.378*** (0.039) 0.022 (0.028) 0.251*** (0.045) 0.057 (0.046) 0.069** (0.030)
2003 -0.197*** (0.032) -0.072*** (0.024) -0.139*** (0.040) -0.131*** (0.035) 0.130*** (0.042) 0.118*** (0.045) -0.018 (0.029)
2004 -0.054* (0.032) -0.101*** (0.023) -0.309*** (0.039) 0.027 (0.028) 0.086** (0.039) 0.151*** (0.045) 0.124*** (0.029)
2005 -0.090*** (0.032) -0.136*** (0.024) -0.092** (0.039) -0.054* (0.029) 0.089** (0.040) 0.084* (0.045) 0.051* (0.029)
2006 0.012 (0.031) 0.018 (0.023) -0.181*** (0.039) -0.031 (0.029) 0.103** (0.041) -0.009 (0.044) 0.190*** (0.031)
2007 -0.050 (0.032) 0.081*** (0.024) -0.010 (0.039) 0.009 (0.030) -0.042 (0.041) -0.053 (0.045) 0.073** (0.030)
2008 0.094*** (0.033) -0.210*** (0.023) -0.125*** (0.039) 0.004 (0.029) 0.153*** (0.040) 0.017 (0.043) 0.245*** (0.032)
Legal origin (reference: English)
French -0.584*** (0.008) -0.234*** (0.012) 0.367*** (0.013) 0.205*** (0.006) -0.342*** (0.011) -0.417***(0.015) 0.071*** (0.007)
Socialist/Communist -0.297*** (0.016) 0.090*** (0.014) -0.689*** (0.020) 0.057*** (0.007) -0.253*** (0.012) -0.160***(0.011) 1.236*** (0.018)
German -0.196*** (0.010) 0.356*** (0.022) 0.409*** (0.020) 0.068*** (0.006) -0.389*** (0.014) -0.577***(0.018) 0.090*** (0.010)
Scandinavian -0.316*** (0.010) -0.099*** (0.013) 1.005*** (0.016) -0.008 (0.006) -0.214*** (0.011) -0.848***(0.025) -0.139*** (0.010)
Colonial origin (reference: other colonial origins or never colonized by a western oversea)
Spain -0.028*** (0.008) -0.425*** (0.008) -0.541*** (0.010) 1.059*** (0.012) 0.122*** (0.007) -0.099***(0.006) 0.004 (0.006)
Blood pressure -0.012*** (0.001) -0.049*** (0.001) 0.058*** (0.002) -0.009*** (0.001) -0.060*** (0.002) -0.011***(0.001) 0.020*** (0.001)
Colestherol 0.436*** (0.019) -0.702*** (0.019) 0.500*** (0.019) 0.184*** (0.012) -0.060*** (0.012) 0.873*** (0.026) 0.477*** (0.015)
Constant 0.089 (0.104) 13.001*** (0.160) -15.860*** (0.189) 0.761*** (0.078) 5.837*** (0.172) 2.515*** (0.177) -4.628*** (0.127)
No. Observations 22,119 22,119 22,119 22,119 22,119 22,119 22,119
R2 0.654 0.675 0.828 0.700 0.417 0.447 0.776
Partial R2 of excluded instruments 0.4664 0.5409 0.6715 0.6359 0.3438 0.3585 0.7125
Shea R2 0.1596 0.1368 0.2216 0.2294 0.1577 0.1835 0.3205
F statistic test excluded
instruments 1654.21*** 2501.37*** 4193.59*** 6516.38*** 240.8*** 193.72*** 1118.16***

Robust standard errors in parentheses

* p<0.1, ** p<0.05, *** p<0.01

115
 Table A.9.3 IV First Stage Pseudo-Panel Regression:
Impact of inequality on job creation in established firms
Asia* Western Europe* Latin America* North America* Oceania* Eastern Europe*
Log(IndexCredit
Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit
Protection)
Protection) Protection) Protection) Protection) Protection) Protection)
Initial conditions
Log (Ratio 90/10) -0.030*** (0.007) -0.487*** (0.010) 0.096*** (0.012) 0.140*** (0.006) 0.045*** (0.007) -0.161*** (0.007) 0.287*** (0.011)
Log (GDPpc1800) 0.184*** (0.005) -0.165*** (0.007) 0.940*** (0.009) -0.121*** (0.004) 0.305*** (0.008) -0.697*** (0.012) -0.193*** (0.006)
Individual characteristics
% of individuals with high school or
0.105*** 0.094*** -0.414*** -0.058*** 0.254*** -0.078*** 0.217***
more (at cohort level) (0.008) (0.010) (0.011) (0.007) (0.009) (0.011) (0.008)
Cohort (Male aged 16-28 reference group)
Male 29-38 0.009 (0.007) -0.010 (0.010) 0.003 (0.012) 0.004 (0.006) 0.005 (0.010) 0.014 (0.010) -0.009 (0.008)
Male 39-48 0.016** (0.007) 0.001 (0.010) -0.021* (0.012) 0.007 (0.006) 0.017* (0.010) 0.009 (0.010) 0.008 (0.008)
Male 49-58 0.025*** (0.007) 0.007 (0.011) -0.053*** (0.012) -0.009 (0.006) 0.028*** (0.010) 0.013 (0.010) 0.030*** (0.008)
Male 59-64 0.017* (0.010) 0.026* (0.016) -0.155*** (0.017) -0.020** (0.008) 0.055*** (0.014) 0.059*** (0.015) 0.067*** (0.010)
Female 16-28 0.009 (0.009) -0.013 (0.013) 0.017 (0.017) 0.009 (0.008) 0.016 (0.013) 0.017 (0.015) -0.021* (0.012)
Female 29-38 0.006 (0.008) -0.026** (0.011) 0.014 (0.013) 0.006 (0.007) 0.016 (0.011) 0.020* (0.011) -0.013 (0.009)
Female 39-48 0.013 (0.008) -0.030*** (0.010) -0.025* (0.013) -0.005 (0.007) 0.058*** (0.011) 0.023** (0.011) -0.004 (0.009)
Female 49-58 0.015 (0.011) -0.023* (0.012) -0.063*** (0.015) -0.010 (0.008) 0.068*** (0.012) 0.021* (0.012) 0.013 (0.013)
Female 59-64 0.026** (0.011) 0.011 (0.017) -0.186*** (0.019) -0.018* (0.009) 0.083*** (0.015) 0.096*** (0.016) 0.048*** (0.011)
Technology sector (reference: No/ Low technology sector)
Medium or high -0.018** (0.007) -0.007 (0.012) 0.046*** (0.013) -0.021*** (0.006) -0.016 (0.011) -0.005 (0.011) -0.003 (0.009)
Sector (reference: Extractive sector)
Transforming sector -0.045*** (0.005) 0.003 (0.007) -0.063*** (0.009) 0.028*** (0.004) 0.002 (0.007) -0.048*** (0.008) 0.011* (0.006)
Business services -0.052*** (0.006) -0.027*** (0.008) -0.115*** (0.010) 0.015*** (0.004) 0.042*** (0.008) -0.012 (0.009) 0.021*** (0.007)
Consumer oriented -0.059*** (0.005) 0.031*** (0.007) -0.033*** (0.008) 0.006 (0.004) -0.006 (0.007) -0.088*** (0.008) 0.017*** (0.006)
Year (reference: 2001)
2002 -0.070*** (0.008) -0.136*** (0.009) 0.047*** (0.013) -0.004 (0.005) 0.010 (0.015) 0.053*** (0.013) -0.078*** (0.009)
2003 -0.138*** (0.009) -0.107*** (0.010) 0.290*** (0.015) -0.130*** (0.010) -0.106*** (0.014) 0.134*** (0.014) -0.200*** (0.011)
2004 -0.041*** (0.007) -0.126*** (0.009) 0.050*** (0.012) -0.002 (0.006) -0.099*** (0.012) 0.124*** (0.013) -0.011 (0.008)
2005 -0.064*** (0.008) -0.182*** (0.010) 0.296*** (0.013) -0.083*** (0.007) -0.110*** (0.012) 0.086*** (0.013) -0.096*** (0.008)
2006 0.021*** (0.007) -0.027*** (0.010) 0.104*** (0.012) -0.024*** (0.005) -0.091*** (0.013) 0.019 (0.012) 0.038*** (0.010)
2007 -0.012 (0.009) -0.001 (0.011) 0.339*** (0.013) -0.009 (0.007) -0.208*** (0.012) -0.014 (0.012) -0.077*** (0.008)
2008 0.134*** (0.008) -0.238*** (0.011) 0.268*** (0.013) 0.011** (0.006) -0.059*** (0.013) 0.017 (0.010) 0.071*** (0.009)
2009 0.060*** (0.010) -0.171*** (0.014) 0.462*** (0.017) 0.029*** (0.008) -0.120*** (0.015) 0.001 (0.014) -0.148*** (0.010)
Legal origin (reference: English)
French -0.629*** (0.005) -0.210*** (0.009) 0.277*** (0.009) 0.165*** (0.003) -0.329*** (0.008) -0.413*** (0.009) 0.051*** (0.004)
Socialist/Communist -0.298*** (0.011) 0.014 (0.012) -0.859*** (0.014) 0.045*** (0.005) -0.231*** (0.008) -0.166*** (0.008) 1.328*** (0.013)
German -0.194*** (0.006) 0.563*** (0.014) 0.202*** (0.013) 0.053*** (0.003) -0.375*** (0.009) -0.565*** (0.011) 0.050*** (0.006)
Scandinavian -0.283*** (0.005) -0.060*** (0.008) 0.966*** (0.010) 0.006* (0.003) -0.202*** (0.007) -0.798*** (0.014) -0.140*** (0.006)
Colonial origin (reference: other colonial origins or never colonized by a western oversea)
Spain -0.009 (0.007) -0.477*** (0.006) -0.557*** (0.007) 1.098*** (0.009) 0.096*** (0.005) -0.098*** (0.004) 0.030*** (0.003)
Blood pressure -0.015*** (0.001) -0.048*** (0.001) 0.056*** (0.001) -0.007*** (0.001) -0.052*** (0.001) -0.014*** (0.001) 0.022*** (0.001)
Colestherol 0.417*** (0.013) -0.859*** (0.016) 0.451*** (0.014) 0.157*** (0.008) -0.051*** (0.009) 0.944*** (0.018) 0.403*** (0.012)
Constant 0.650*** (0.071) 13.199*** (0.103) -15.395*** (0.126) 0.725*** (0.057) 5.085*** (0.111) 2.690*** (0.103) -4.236*** (0.099)
No. Observations 53,067 53,067 53,067 53,067 53,067 53,067 53,067
R2 0.645 0.685 0.817 0.713 0.387 0.463 0.815
Partial R2 of excluded instruments 0.5023 0.5688 0.6787 0.6563 0.3249 0.3823 0.7674
Shea R2 0.1616 0.1533 0.1763 0.2154 0.1406 0.1959 0.2178
F statistic test excluded instruments 4580.40*** 4786.64*** 11390.30*** 12558.59*** 441.32*** 511.07*** 4015.29***
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

116
 Robustness checks

Table A.10 IV Second Stage Pseudo-Panel Regression:

Impact of inequality on firm’s life cycle excluding self-employed
(1) (2) (3) (4)
Nascent Young Established Closed
IV IV IV IV
Initial conditions
Log (Ratio 90/10) -0.175*** (0.005) -0.151*** (0.005) -0.091*** (0.004) -0.176*** (0.011)
Log (GDPpc1800) -0.747*** (0.006) -0.746*** (0.007) -0.498*** (0.006) -0.681*** (0.009)
Institutional environment
Log (IndexCreditProtection) 1.066*** (0.011) 0.829*** (0.012) 0.224*** (0.011) 0.734*** (0.011)
Region (reference group: Africa)
Asia 0.197*** (0.011) 1.141*** (0.012) 1.616*** (0.010) 0.749*** (0.021)
Western Europe 0.163*** (0.010) 0.722*** (0.011) 1.323*** (0.010) -0.002 (0.011)
Latin America 1.497*** (0.012) 1.623*** (0.012) 1.474*** (0.011) 1.348*** (0.012)
North America 0.859*** (0.010) 0.998*** (0.011) 1.440*** (0.010) 0.429*** (0.016)
Oceania 0.119*** (0.010) 0.537*** (0.010) 1.384*** (0.009) -0.212*** (0.010)
Eastern Europe 0.292*** (0.011) 0.515*** (0.011) 0.879*** (0.010) 0.013 (0.013)
Individual characteristics
% of individuals with high school or more
-0.095*** (0.006) -0.319*** (0.006) -0.307*** (0.005) -0.126*** (0.027)
(at cohort level)
Cohort (Male aged 16-28 reference group)
Male 29-38 0.151*** (0.004) 0.199*** (0.004) 1.142*** (0.004) 0.317*** (0.004)
Male 39-48 -0.082*** (0.004) -0.094*** (0.004) 1.354*** (0.004) 0.288*** (0.004)
Male 49-58 -0.551*** (0.005) -0.477*** (0.005) 1.234*** (0.005) 0.248*** (0.006)
Male 59-64 -1.440*** (0.007) -1.414*** (0.007) 0.451*** (0.005) 0.095** (0.038)
Female 16-28 -0.621*** (0.004) -0.623*** (0.004) -0.637*** (0.004) -0.392*** (0.004)
Female 29-38 -0.477*** (0.004) -0.339*** (0.004) 0.367*** (0.004) -0.122*** (0.004)
Female 39-48 -0.679*** (0.004) -0.682*** (0.004) 0.594*** (0.004) -0.190*** (0.005)
Female 49-58 -1.148*** (0.005) -1.157*** (0.005) 0.368*** (0.005) -0.383*** (0.007)
Female 59-64 -2.181*** (0.009) -2.116*** (0.009) -0.333*** (0.005) -0.551*** (0.013)
Year (reference: 2001)
2002 -0.214*** (0.005) -0.092*** (0.005) 0.106*** (0.005) -0.312*** (0.016)
2003 0.008 (0.006) 0.129*** (0.006) 0.257*** (0.005) -0.085*** (0.021)
2004 -0.466*** (0.006) -0.115*** (0.005) 0.250*** (0.005) -0.212*** (0.018)
2005 -0.157*** (0.005) -0.033*** (0.005) 0.352*** (0.005) -0.290*** (0.019)
2006 -0.112*** (0.005) 0.106*** (0.006) 0.415*** (0.005) -0.211*** (0.036)
2007 -0.112*** (0.005) 0.055*** (0.005) 0.436*** (0.005) -0.215*** (0.023)
2008 -0.123*** (0.005) 0.114*** (0.005) 0.688*** (0.005) -0.095*** (0.022)
2009 1.127*** (0.007) 1.378*** (0.008) 1.106*** (0.005)
Constant 0.571*** (0.029) 0.252*** (0.029) -1.651*** (0.027) 0.152** (0.069)
No. Observations 929,305 913,781 937,281 889,726
R-squared 0.497 0.506 0.594 0.464
F-test 30162.72*** 30154.75*** 29253.65*** 27052.57***
K-P rk F statistic (weak identification test) 140,000*** 130,000*** 120,000*** 150,000***
Endogeneity test 5636.82*** 3675.33*** 779.85*** 2451.33***
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

117
 Table A.11 IV Second Stage Pseudo-Panel Regression:
Impact of inequality on job creation excluding self-employed
Nascent Young Established
IV IV IV
Initial conditions
Log (Ratio 90/10) 0.511*** (0.130) -0.532*** (0.072) -0.341*** (0.021)
Log (GDPpc1800) -1.270*** (0.157) 0.076 (0.066) -0.039** (0.019)
Institutional environment
Log(IndexCreditProtection) Total effect 1 6.252*** (0.606) 1.984*** (0.284) 1.883*** (0.143)
Ommited: Log(IndexCreditProtection)*Africa
Log(IndexCreditProtection) 0.158 (0.175) 0.318*** (0.080) 0.314*** (0.039)
Log(IndexCreditProtection)*Asia 1.278*** (0.165) 0.396*** (0.094) 0.127*** (0.033)
Log(IndexCreditProtection)*Western Europe 1.142*** (0.107) 0.163*** (0.059) 0.306*** (0.031)
Log(IndexCreditProtection)*Latin America 1.258*** (0.151) 0.333*** (0.070) 0.192*** (0.027)
Log(IndexCreditProtection)*North America 0.888** (0.430) 0.248*** (0.093) 0.275*** (0.045)
Log(IndexCreditProtection)*Oceania 0.099 (0.125) -0.063 (0.084) 0.198*** (0.032)
Log(IndexCreditProtection)*Eastern Europe 1.430*** (0.164) 0.589*** (0.055) 0.472*** (0.030)
Individual characteristics
% of individuals with high school or more (at cohort level) -0.327** (0.134) -0.109* (0.056) -0.110*** (0.016)
Male 29-38 0.457*** (0.089) 0.024 (0.040) 0.128*** (0.012)
Male 39-48 0.257** (0.125) -0.400*** (0.058) 0.011 (0.016)
Male 49-58 0.178 (0.228) -0.155 (0.101) -0.149*** (0.031)
Male 59-64 -0.280** (0.116) -0.467*** (0.039) -0.301*** (0.016)
Female 16-28 -0.490*** (0.083) -0.270*** (0.032) -0.332*** (0.012)
Female 29-38 -0.614*** (0.080) -0.466*** (0.043) -0.300*** (0.015)
Female 39-48 -0.034 (0.280) -0.654*** (0.056) -0.465*** (0.018)
Female 49-58 -0.028 (0.380) -0.752*** (0.106) -0.450*** (0.031)
Female 59-64
Technology sector (reference: No/ Low technology sector)
Medium or high -0.001 (0.070) 0.053 (0.037) 0.014 (0.014)
Sector (reference: Extractive sector)
Transforming sector 0.055 (0.100) 0.082* (0.045) 0.037*** (0.013)
Business services 0.227** (0.106) 0.071 (0.047) 0.048*** (0.014)
Consumer oriented 0.138 (0.098) 0.041 (0.044) 0.002 (0.012)
Year (reference: 2001)
2002 -0.529* (0.306) -0.358*** (0.109) 0.085*** (0.019)
2003 0.418 (0.326) -0.214** (0.102) 0.292*** (0.020)
2004 0.343 (0.300) -0.605*** (0.102) 0.035** (0.018)
2005 0.067 (0.317) -0.298*** (0.105) 0.128*** (0.018)
2006 -0.175 (0.329) -0.411*** (0.105) 0.260*** (0.018)
2007 -0.250 (0.274) -0.205** (0.104) 0.266*** (0.018)
2008 -0.836*** (0.312) -0.403*** (0.106) 0.121*** (0.018)
2009 0.440*** (0.031)
Constant 7.128*** (1.277) 1.694*** (0.573) 1.841*** (0.155)
No. Observations 5,432 19,691 85,057
R-squared 0.63 0.78 0.89
F test 587.4*** 1581.86*** 5222.46***
K-P Wald rk F statistic (weak identification test) 18.08*** 92.39*** 563.33***
Endogeneity test 24.02*** 44.04*** 420.48***
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

118
 Table A.12 IV Second Stage Pseudo-Panel Regression:
Impact of inequality on firm’s life cycle using alternative inequality indicators
(1) (2) (3) (4)
Nascent Young Established Closed
IV IV IV IV
Initial conditions
Log (Gini) -0.495*** 0.036*** -0.401*** -0.237***
(0.017) (0.016) (0.015) (0.085)
Log (Top90) -0.678*** -0.941*** -0.385*** -0.870***
(0.010) (0.010) (0.009) (0.019)
Log (Middle 50) 0.686*** 0.358*** 1.169*** 1.052***
(0.025) (0.025) (0.022) (0.097)
Log (Bottom 10) 0.052*** 0.014*** -0.008*** -0.012
(0.004) (0.004) (0.004) (0.008)
Log (Top20/Bottom20) -0.083*** -0.018*** -0.061*** -0.033***
(0.003) (0.003) (0.003) (0.008)
Log(Middle) 1.830*** 0.699*** 1.746*** 1.116***
(0.034) (0.035) (0.031) (0.153)
No. Observations 959,199 942,535 973,873 914,094
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01
Top90 is the income share of the 9th decile relative to the income share of the 1st decile
Middle 50 is the income share of the 5th decile relative to the mean income
Bottom 10 is the income share of the 1st decile relative to the mean income
Top20/Bottom20 is the income share of the 8th decile relative to the 2nd decile
Middle is the income share of the middle class, defined as the income share of the 2nd to 4th quintiles.
Control variables as in Table 1.

Table A.13 IV Second Stage Pseudo-Panel Regression:

Impact of inequality on job creation using alternative inequality indicators
(1) (2) (3)
Nascent Young Established
IV IV IV
Initial conditions
Log (Gini) -1.385*** -1.402*** -0.989***
(0.322) (0.165) (0.069)
Log (Top90) 3.441*** 0.731*** 0.703***
(0.228) (0.136) (0.042)
Log (Middle 50) -3.288*** 2.688*** 1.956***
(0.601) (0.267) (0.136)
Log (Bottom 10) 0.382*** 0.383*** 0.321***
(0.097) (0.040) (0.017)
Log (Top20/Bottom20) 0.105 -0.310*** -0.211***
(0.082) (0.039) (0.016)
Log(Middle) 2.059*** 2.687*** 1.955***
(0.001) (0.267) (0.135)
No. Observations 6,952 22,119 53,067
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01
Top90 is the income share of the 9th decile relative to the income share of the 1st decile
Middle 50 is the income share of the 5th decile relative to the mean income
Bottom 10 is the income share of the 1st decile relative to the mean income
Top20/Bottom20 is the income share of the 8th decile relative to the 2nd decile
Middle is the income share of the middle class, defined as the income share of the 2nd to 4th quintiles
Control variables as in Table 1.

119
 Table A.14 Second Stage Pseudo-Panel Regression:
Firm’s life cycle using alternative instrumental variables
(1) (2) (3) (4)
Nascent Young Established Closed
Panel a) IV: Language
Key independent variables
Log (Ratio 90/10) -0.258*** -0.292*** -0.263*** -0.178***
Log (IndexCreditProtection) 2.028*** 2.334 *** 2.486*** 0.676***
First stage summary results
K-P Wald rk F statistic (weak identification test) 905.36*** 971.81*** 980.37*** 582.48***
Endogeneity test 1072.696*** 1412.251*** 1965.14*** 15.701***
Shea partial R2 0.0041 0.0045 0.0043 0.0057
Partial R2 0.0041 0.0045 0.0043 0.0057
Panel b) IV: Religion
Key independent variables
Log (Ratio 90/10) -0.072*** -0.871*** -0.0797*** -0.135***
Log (IndexCreditProtection) 0.684*** 0.871*** 0.164*** 0.0218***
First stage summary results
K-P Wald rk F statistic (weak identification test) 8005.066*** 8019.8*** 7955*** 4872.42***
Endogeneity test 1952.341*** 1233.588*** 143.8*** 27.58***
Shea partial R2 0.0257 0.0268 0.0254 0.0246
Partial R2 0.0257 0.0268 0.0254 0.0246
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01
Control variables as in Table 1.

Table A.15 Second Stage Pseudo-Panel Regression: Impact of Inequality

on job creation using alternative instrumental variables
(1) (2) (3)
Nascent Young Established
IV Language
Log (Ratio 90/10) -12.07 -11.9 -1.32***
Log (IndexCreditProtection) 1.88 5.92 2.45***
IV Religion
Log (Ratio 90/10) -0.160 -2.13*** -1.25***
Log (IndexCreditProtection) 4.8*** 2.61*** 2.43***
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01
Control variables as in Table 2.

120
 Table A.16 Summary results instrumental variable: Language. Job creation
Asia* Western Europe* Latin America* North America* Oceania* Eastern Europe*
Log(IndexCredit
IV- Language Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit
Protection)
Protection) Protection) Protection) Protection) Protection) Protection)
Nascent firms
Shea partial R2 0.0001 0.0027 0.0012 0.0001 0.0005 0.0003 0.0004
Partial R2 0.6535 0.5708 0.7054 0.443 0.2106 0.5451 0.7644
F test excluded instruments 1259.43 782.57 1524.8 581.64 34.86 250.78 411.78
p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000
K-P Wald rk F statistic (weak identification test) 0.013
Endogeneity test 47.05***
Young firms
Shea partial R2 0.0002 0.0009 0.0014 0.0009 0.0633 0.0004 0.0532
Partial R2 0.486 0.4802 0.631 0.3229 0.3374 0.3564 0.7437
F test excluded instruments 1547.47 2268.34 3513.56 907.55 236.86 206.85 1241.32
p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000
K-P Wald rk F statistic (weak identification test) 0.12
Endogeneity test 89.67***
Established firms
Shea partial R2 0.0036 0.0199 0.0265 0.0179 0.0653 0.0092 0.1675
Partial R2 0.515 0.5096 0.6474 0.2788 0.3206 0.3804 0.7856
F test excluded instruments 4321.54 4282.36 8815.39 1401.52 444.34 524.48 4546.76
p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000
K-P Wald rk F statistic (weak identification test) 5.607
Endogeneity test 226.05***
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

121
 Table A.17 Summary results instrumental variable: Religion. Job creation
Asia* Western Europe* Latin America* North America* Oceania* Eastern Europe*
Log(IndexCredit
Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit Log(IndexCredit
Protection)
Protection) Protection) Protection) Protection) Protection) Protection)

Nascent firms
Shea partial R2 0.0226 0.2565 0.5256 0.0806 0.0501 0.0509 0.2264
Partial R2 0.6616 0.6466 0.707 0.3622 0.2272 0.7039 0.7465
F test excluded instruments 1356.44 1634.28 1501.46 402.36 36.8 524.48 386.8
p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000
K-P Wald rk F statistic (weak identification test) 8.39
Endogeneity test 112.553***
Young firms
Shea partial R2 0.0056 0.0488 0.0636 0.0199 0.1554 0.0131 0.2274
Partial R2 0.4981 0.5336 0.6312 0.2493 0.3696 0.4997 0.7233
F test excluded instruments 2032.3 2670.9 3471.06 869.56 264.88 370.33 1047.77
p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000
K-P Wald rk F statistic (weak identification test) 5.88
Endogeneity test 68.16***
Established firms
Shea partial R2 0.0041 0.037 0.0452 0.0116 0.0906 0.0113 0.2621
Partial R2 0.5256 0.5571 0.6396 0.2281 0.3493 0.5185 0.777
F test excluded instruments 4597.68 4558.24 9061.98 1668.55 489.86 932.3 4128.77
p-value 0.000 0.000 0.000 0.000 0.000 0.000 0.000
K-P Wald rk F statistic (weak identification test) 7.52
Endogeneity test 189.66***
Robust standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

122
123
124
 Essay 3
Schooling progression in Uruguay: Why some children are
left behind?

125
126
 Schooling progression in Uruguay: Why some children are left behind?

Abstract
This study examines the factors that differently affect children’s educational path in Uruguay.
Specifically, I focus on the effects of long-term parental income crystallized by cognitive and
non-cognitive abilities, parental educational background and race, and short-term family
income proxied by the opportunity cost of education, on children’s schooling progression in
Uruguay.
For this purpose, I use a sequential probability model which allows me to analyze the factors
affecting the dynamics of the children’s educational path. The results show that long-term
parental income is the main factor influencing schooling attainment while short-term family
income has decreasing effects over the children’s education path. Specifically, parental
educational background, race, cognitive and non-cognitive abilities have effects of diverse
magnitude across stages of schooling progression. I find that cognitive ability, measured by
repetition, has long-lasting effects on children’s education attainment. Motivation and risky
behavior measuring non-cognitive ability also influence children’s schooling completion at
early stages of education.
These findings call for public intervention focused on improving cognitive and non-cognitive
abilities to enable children attaining higher education, particularly those from disadvantaged
parental backgrounds.

127
 3.1 Introduction
It is well known in the literature that children’s parental background plays a major role
in explaining educational inequality. Several studies have shown that children of well-
off parents generally receive more and better schooling and benefit from material,
cultural and genetic inheritances (Checchi, 2006). Heckman and coauthors refer to the
long-term family factors reflected by: parental educational background, children
scholastic ability, motivation, self-esteem, as important sources of disparities across
individuals’ educational attainment. In turn, these sources of disparities in education,
may well translate into other individuals’ economic outcomes, such as earnings. As long
as large differences exist in educational opportunities, individuals will have different
chances of success in life.
In turn, attaining a level of education is something that happens over a long
period of time and it is split into different schooling stages, like finishing primary
education, completing secondary level, and so on. Therefore, knowing the influence of
parental background variables at each stage of the schooling transition can give a more
complete picture of how inequality of education attainment came about. Each of the
alternative sources of inequality pointed by the literature call for specific policy
prescriptions at different stages of the schooling progression, which may well have
different effects on equity and efficiency of the education system and subsequent labor
market outcomes.
The objective of this paper is to analyze to what extent intergenerational
transmission of parental traits takes place for children’s educational attainment in
Uruguay. Specifically, this paper aims to study whether parental education, race, child’s
scholastic ability, motivation and risky behavior as measures of socio-emotional
endowments, and short-term family income proxied by the opportunity cost of
education, are key determinants of individuals’ educational path decisions and, if they
are, at what stage in the schooling process they take on their importance.1
Uruguay is a particularly interesting country to analyze this issue for many
reasons. First, it stands out in the Latin American region because of a large tradition of
publicly provided education and social inclusion. For instance, primary school was
made compulsory in 1877, universal primary schooling was achieved in the 1950s

1
In this study cognitive ability, scholastic ability and performance in different educational levels are used
as synonyms, while socio-emotional endowments and non-cognitive ability are used interchangeably.

128
(Manacorda, 2008). In addition, the system provides free access to educational
institutions in all schooling levels; in postsecondary education university is publicly
provided, students do not need to pay any fee or perform any entrance test; one feature
that distinguish Uruguay from others countries of the region. Also, the country ranks
among the highest in the region in terms of its socioeconomic indicators, presenting the
lowest poverty rate and income inequality indicators in the region (Panorama Social de
America Latina, Cepal, 2012).
However, the Uruguayan education system shows major shortcomings. In the
Latin American context, while the proportion of population aged 18 to 29 living in
urban areas with complete secondary in 2000 is less than 20% in Uruguay; this rate is
40% in Chile and 30% in Paraguay (SITEAL, 2005). Chile presents one of the highest
indicators of income inequality and is characterized by a private education system
especially at the university level; while Paraguay ranks below Uruguay in terms of the
Human Development Index. In this line, several studies stress that the Uruguayan
educational system is unable to retain a large share of students in lower high school
(Furtado, 2003; da Silveira and Queirolo, 1998), picture that worsens when educational
attainment across afro and non-afro descendants is analyzed.2 Therefore, a relevant
question is why despite the great offer of public education, children living in Uruguay
do not attain higher levels of education. This is what makes Uruguay an interesting case
study.
The contributions of this paper are twofold. First, it contributes to the recent
literature developed by Bowles and Gintis (2001, 2002) and Heckman and co-authors
by addressing the importance of cognitive and non-cognitive abilities, parental
educational background, and race, on young people’ (or their parents) educational
choices in a middle income country such as Uruguay. Indeed, empirical studies
exploring the impacts of multiple abilities on education attainment are scarce and
mainly focused on developed countries, while less usual for developing countries
mainly because of data availability. In this sense, the rich dataset used in this paper
enables me to exploit information on motivation (measured as motives reported for
secondary enrollment) and risky behavior such as adolescence use of marijuana, two
factors pointed out in the literature as important ones reflecting socio-emotional factors,

2
See Table 1.

129
and in turn affecting education attainment (see for instance Heckman et al., 2006;
Heckman et al., 2014; Gullone and Moore, 2000).
Second, by exploiting the sequential process of education attainment, it is
possible to identify different impacts of the key variables over the individual’s
educational path. Specifically, by analyzing the effect of parental educational
background, multiple abilities and race; and the opportunity cost of education at
different decision points in the schooling transition process, it is possible to distinguish
between long and short-term family income affecting schooling; to disentangle a direct
effect of these key variables on the educational level attained, but also an indirect effect
to the extent that parental background affects previous educational choices.
Therefore, this study goes beyond previous analyses on education focused on
developing countries by saying that measures of cognition are important predictors of
child’s outcomes, and by recognizing the different effects of diverse abilities across the
individual’s schooling transition in a middle-income country such as Uruguay.
This paper uses a unique micro-dataset elaborated by the Uruguayan Statistics
Institute: the Youth National Survey (ENAJ: Encuesta Nacional de Adolescencia y
Juventud), a cross-sectional national representative survey on adolescence and youth
conducted in 2008. The sample is based on the same households interviewed in the
Continuous Household Survey (ECH: Encuesta Continua de Hogares) for 2008, thus
being possible to merge the information from both surveys. Detailed information on
socio-demographic characteristics, migration trajectories, educational history, risky
behaviors, parental education, among others, is provided. In addition, the retrospective
information contained in this dataset allows me to construct educational trajectories, as
well as early behaviors of interest for theoretical ages of participation in the education
system.
The empirical strategy considers a dynamic educational model developed by
Cameron and Heckman (1998, 2001) in which schooling attainment is modeled as the
outcomes of sequential choices made at each educational level using probability models
and conditional on previous educational choices. In turn, the model accounts for
individual unobserved heterogeneity, such as ability or motivation, which may affect
individuals’ schooling progression.
The results suggest that long-term family factors greatly influences child’s
schooling transitions. Students with more favorable parental educational backgrounds
and with better performance in the educational system are more likely to survive higher

130
schooling stages. Race is an important factor preventing schooling progression for girls
and, to a lesser extent for boys. Less motivated individuals and with risky behaviors are
less likely to survive early schooling stages and therefore, to attain higher education. In
addition, short-term family income, measured as the opportunity cost of education at
each schooling level, has decreasing effects across the educational path; turning less
important -in comparison to long-term family factors- the higher we move on the
educational path.
These findings are in line with the literature, which suggests that early child’s
life cycle is a sensitive period for the formation of cognitive skills and has persistent
effects on higher stages of the schooling transition. Also, non-cognitive ability, despite
data limitations for its measurement, is seen to be an important factor affecting
schooling progression. Thus, our results call for public interventions focused on
cognitive and non-cognitive abilities at different stages of the life cycle in order to
compensate children from disadvantaged parental backgrounds.
The remainder of this paper is organized as follows. The next section presents an
overview of the literature on education, specially focusing on the literature of cognitive
and non-cognitive abilities. Section 3 describes the Uruguayan educational system.
Section 4 introduces the data and presents descriptive analysis. Section 5 describes the
econometric methods. Section 6 presents and discusses the main findings of the study.
Finally, Section 7 concludes.

3.2 Education inequality, cognitive and non-cognitive abilities

This paper is broadly related to the literature analyzing the different factors that
influence individual’s education attainment. It is well known in the literature that higher
education is positively correlated with higher income. Thus, a pertinent question is why
individuals choose different levels of education. Checchi (2006) stresses that families
are often unaware of the economic benefit of education or are prevented from sending
their children to further education by their financial needs.
The literature on education has focused on the factors preventing full access to
education and individuals’ educational choices, such as individual’s unobservable
abilities, family cultural background, family financial resources, public resources,
residential choices and social capital.3 Most of these factors exhibit intertemporal and

3
See Checchi (2006) for an exhaustive overview of the literature on Economics of Education.

131
intergenerational persistence. Therefore, alternative intergenerational transmission
channels are identified in the literature, which in turn calls for specific policy
recommendations.
Within this line of research, the literature of inequality of opportunity analyzes
the different factors influencing education attainment. The most accepted concept of
inequality of opportunity refers to the notion that inequalities which are brought about
by individual’ circumstances, like gender, ethnicity and race, place of birth, family
background, which are beyond the individual’s control, are considered ethically
unacceptable, while inequality resulting from individual’s effort and choice are ethically
accepted (Roemer, 1998). This definition requires that any inequality attributed to the
influence of exogenous circumstances should be reduced, compensated by public
interventions.
Based on this framework, several empirical studies address the alternative
mechanisms through which intergenerational transmission may operate by estimating
the relationship between one individual’s educational attainment and her parental
education, income, or occupation; controlling for other child’s circumstances like race
and gender, among others (as in Bourgignon et al., 2003; Ferreira and Gignoux, 2008;
Peragine and Serlenga, 2007, among others). Therefore, the coefficient relating parental
background and a child’s outcome measures the intergenerational transmission of an
attribute from one generation to the other. For the Uruguayan case, González and
Sanromán (2010) find persistent effects of parental educational background on
education attainment for afro and non afro-descendants. In turn, Porzecanski (2008)
studies the determinants of the educational gap between afro and non afro-descendants
in Uruguay analyzing the impact of family background on repetition in primary level,
and dropouts of adolescents in the educational system.
In this study, I follow an alternative framework developed by Heckman and
coauthors (Cameron and Heckman, 2001; Heckman and Carneiro, 2003; Cuhna and
Heckman, 2007), which considers the total effect of family background on education
attainment. Specifically, these authors refer to long-term family factors including long-
term levels of family income, reflected by parental education, scholastic ability,
motivation, time preferences, risk aversion and self-esteem, as important factors shaping
later success in life, which in turn may explain sources of disparities across individuals’
education attainment. Also, short-term family income influences individual’s education
attainment.

132
 Specifically, Cameron and Heckman (2001) find that short-term family income
effects are weakened most in the later schooling transitions, playing no role in college
entry decisions. To the extent that the influence of long-term family income measured at
a point in time is diminished by the inclusion of cognitive abilities or family
background variables, the authors conclude that long-term family factors crystallized in
these variables are the driving forces behind schooling attainment, and not short-term
credit constraints experienced in the late adolescent years.4
In turn, these authors analyze the educational level attained by one individual as
a sequential process, in which the individual chooses the educational level conditional
on having completed the previous educational level. By doing so, it is possible to
examine the different effects of variables of interest on individual’s educational
attainment, and to do so at different stages of the educational path.
Previous studies have followed this empirical strategy, mainly focused on
developed countries for which adolescent and youth panel datasets with information on
individuals’ educational path and past performance are largely available (Cameron and
Heckman, 2001 for US; Holm and Jaeger, 2011, and Blanden et al., 2002 for the UK;
multiple tracks choices in the educational path for the Danish case in Karlson, 2011, and
for Germany in Dustmann et al., 2004). Also, cross sectional data containing
information on past performance in the educational system allows Bernardi (2012) to
analyze schooling transitions in Spain.5 The one exception for Latin American countries
within this framework is found in Pal (2004) for the Peruvian case in which, using 1994
Peru Living Standards Measurement Study data, analyses the impact of parental
background and individual’s ability on individuals’ schooling transitions.
In general, these empirical studies measure ability with previous performance in
the educational system, such as repetition or test scores. These measures have been
criticized by recent literature. Indeed, the literature has recognized that abilities are
multiple in nature and that previous studies using IQ or previous performance
(repetition, test scores) measures does not properly account for ability.
For instance, Bowles and Gintis (2002) stress that “inheritance process
operating through superior cognitive performance and educational attainments of those
4
Note that this framework differentiates from the Inequality of Opportunity framework since it does not
distinguish between individuals’ circumstances and efforts. Specifically, Cameron and coauthors only
point out that abilities reflect long-term parental income.
This framework is also extended in Cappellari (2004) for the analysis of individual’s transitions between
5

the type of high school chosen (private or public) and university enrollment and school-to-work transition
using a cross sectional sample of high school leavers in Italy.

133
well-off parents, while important, explain at most half of the intergenerational
transmission of economic status. Moreover, while genetic transmission of earnings-
enhancing traits appears to play a role, the genetic transmission of IQ appears to be
relatively unimportant”. These authors conclude that empirical studies on
intergenerational transmission of economic status have over-studied education and
cognitive abilities, while other individual characteristics such as wealth, race and non-
cognitive behavioral traits have been under-studied.
Unlike other personal traits such as height or weight, personality traits cannot be
directly measured. Non-cognitive abilities, such as perseverance, motivation, risk
aversion, self-esteem, self-control, have direct effects on wages (after controlling for
schooling), schooling, performance on achievement tests, and other aspects of social
and economic life. The most widely accepted taxonomy of personality traits is the Big
Five model defined as: Conscientiousness (“the tendency to be organized, responsible
and hardworking”), Openness to Experience (“the tendency to be open to new aesthetic,
cultural, or intellectual experiences”), Extraversion (“an orientation of one interests and
energies toward the outer world of people and things rather than the inner world of
subjective experience; characterized by positive affect and sociability”), Agreeableness
(“the tendency to act in a cooperative, unselfish manner”), and Neuroticism/Emotional
Stability (Emotional stability is “predictability and consistency in emotional reactions,
with the absence of rapid mood changes”; Neuroticism is “a chronic level of emotional
instability and proneness to psychological distress”).6 Overall, observed productivities,
efforts, and actions are used to infer traits using conventional factor analysis in which
the tests are measures of different domains of personality based on observer reports or
self-report.7
Although the relationship between personality traits and education has not
received much attention, mainly due to data availability, a certain consensus emerges in
the literature. Perseverance and preferences related to an interest in learning, two traits
which are related to Consciousness and Openness to Experience, increase the likelihood
of individuals’ attaining more years of schooling (Lundberg, 2013; Almlund et al.,
2011). In turn, Heckman, Stixrud and Urzua (2006) find that locus of control and self-

6
See Table 1.3 (p45) in Almlund et al. (2011) for a comprehensive definition of the Big Five Domains,
facets and related traits.
7
The Big Five model is not without its critics. The main ones stress that the model is atheoretical; omits
individual’s motivation (what people value or desire), while other psychologists suggest that the
categories are too crude to be useful; or the luck of consensus among researchers about identifying and
organizing lower order facets of the Big Five factors (Almlund et al., 2011).

134
esteem (traits related to Neuroticism) play an important role for adolescents schooling
decisions, having different effects across schooling levels. Nonetheless, data availability
often determines which measure of non-cognitive skills is used in empirical analysis
(Brunello and Schlotter, 2011). One possibility for overcoming data limitations or
surveys without good questions on personality type is found within the psychological
literature on personality traits and adolescent risk-behavior. Gullone and Moore (2000)
identify different categories of risky behaviors traits, two of them -rebellious and
reckless risk-taking- were found to be negatively correlated with Consciousness.8
Following the psychological literature, Heckman et al. (2011) and Heckman et al.
(2014) propose to use behaviors that have proved to be strongly correlated with
Consciousness and Agreeableness, namely: violent behavior such as fighting at school
or work and hitting or threatening to hit someone, tried marijuana, daily smoking,
regular drinking, and any intercourse before age 15; measures of socio-emotional
factors that affect schooling progression.
Overall, this framework stresses those both cognitive and non-cognitive abilities,
as part of long-term parental background, jointly with parental education, race/ethnicity,
and other family characteristics, play an important role on the multiple periods in the
individuals’ life cycle. The existence of critical and sensitive periods of childhood in
skill formation and different roles played by cognitive abilities and socio-emotional
factors across an individual’s life cycle calls for different policies in time (Heckman and
Mosso, 2014). For instance, parental inputs have different effects at different stages of
the child’s life cycle with cognitive skills affecting more at early ages and non-cognitive
skills affecting more at later ages (Cunha and Heckman, 2008). In turn, both cognitive
and non-cognitive skills can be shaped by interventions and that there are effective
margins for social policy (see Heckman and Mosso, 2014; Heckman, Pinto, and
Savelyev, 2013).

8
Examples of rebellious risk-taking are drinking, smoking, and staying out at night. Examples of reckless
risk-taking are drinking and driving, having unprotected sex, and speeding.

135
 3.3. The Uruguayan Educational System
The educational system is organized in four levels: pre-school, primary education
(grades 1-6, with theoretical ages 6 to 11), secondary level which includes lower high
school (Ciclo básico, grades 7 to 9, theoretical ages 12-14) and upper high school
(Bachillerato, grades 10 to 12, theoretical ages 15-17); and tertiary level (university and
teaching training institutes). Primary and lower high school levels are compulsory.9
Lower and upper high school are offered in both liceos (non-vocational secondary
schools), and in vocational schools (UTUs). The different schooling stages are both
public and privately provided (see Figure 1).
Table 1 presenting schooling progression by gender and race for the population
aged 20 to 29 shows one of the major caveats of the educational system. While
enrollment in primary is timely and completion of primary education almost universal,
the system fails in retaining a large share of students at different schooling stages.
It is worth noting the great fall in the proportion of people completing each level
across the educational system. In particular, low enrollment rates in postsecondary
(20.5% for the total sample, in Table 1) may be explained by the low proportion of
people completing previous education levels. Note for instance the low proportion of
young people with complete lower high school or complete upper high school (64.5%
and 29.2% respectively for the final sample, in Table 1). Differences between afro and
non afro-descendants are also striking. In particular, 5% of afro-descendant males and
13.7% of afro descendant women have complete secondary education, compared to
28.8% and 36.4% for non afro-descendant men and women respectively (Table 1).
Some main features that characterize the educational system in Uruguay are
provided in Table A.1 in the Appendix. In particular, it is highlighted the great
proportion of population aged 12 to 29 who is or was enrolled in a public institution at
different levels of schooling stages. Nonetheless, notice that the proportion of students
in a private institution increases for higher levels of education. Also, students largely
choose general education institutions (Liceos or Bachilleratos).
An important feature which deserves to be highlighted is the low supply of
tertiary education institution located in the Interior of the country.10 The main
University in Uruguay is the Universidad de la República (UdelaR), which is public and

9
Since 2008 upper high school and pre-school are compulsory. Ley General de Educación No. 18.347
10
Interior is commonly used to identify the regions of the country excluding Montevideo, the capital of
Uruguay, and includes 18 Departments.

136
freely provided, meaning that students do not have to pay any fee or pass any entrance
test. But the UdelaR is mainly located in Montevideo, the capital of Uruguay, so
students wanting to enroll in college and not living in Montevideo need to migrate to the
capital. Also private colleges are mainly located in Montevideo. This may prevent many
students without financial family support to access college.11

3.4. Data and descriptive statistics

This study uses the National Youth Survey (ENAJ, Encuesta Nacional de Adolescencia
y Juventud), a cross-sectional national representative survey on adolescents and youth
conducted in 2008 by the Uruguayan National Statistics Institute (Instituto Nacional de
Estadística). The survey universe consists on all adolescents and young Uruguayans
between 12 and 29 years living in cities larger than 5,000 inhabitants. In total the
original survey comprises 4,993 individuals. Since the sample is based on the same
households interviewed in the Continuous Household Survey (ECH) for 2008,
information from both surveys can be merged.
The ECH is one of the main sources of statistical information in the country
providing socioeconomic information at the Department level, Montevideo and rest of
the country (commonly named Interior, containing 18 Departments).12 It offers detailed
information on socio-demographic characteristics, migration trajectories, labor market
participation, experience, and conditions, among others. The ENAJ incorporates rich
information about other aspects of the respondents lives, such as education attainment,
fertility, participation in crime, self-assessed health status, family background (mother
and father education), alcohol and substance use, migration, first job characteristics,
participation in social and political organizations.
Although the survey is not longitudinal, it contains retrospective information
allowing me to construct educational trajectories, as well as past performance in the
educational system (repetition in primary and secondary level), motives for attending
secondary level, and risky behaviors of interest (such as drug consumption). Moreover,
the ENAJ complements ECH in the sense that the latter suffers from major limitations
of special importance to the analyses of the role of family background on college

11
It is worth mentioning that since 2007 the UdelaR has being making great efforts in terms of territorial
decentralization in order to give major opportunities to those students living in the Interior of the country.
Also, some private universities are starting to locate in different regions of the country.
12
A Department is a first-level political and administrative division of Uruguay.

137
enrollment decisions. For instance, the ECH contains information of the family
background only for those individuals living in the origin household, while not
information is provided for those who moved out. Then, studies based on the ECH may
suffer from endogeneity issues, due to the possible sample selection of those individuals
who left the household of origin (see Francesconi and Nicoletti, 2006). In addition, the
ECH does not provide information on educational past history, such as repetition in
primary and secondary level.13 The ENAJ allows me not only to address the above
mentioned issues, but also to take into account an individual’s educational history and
exploit information on motivation and risky behaviors.
The original sample is restricted to individuals aged 20 to 29, theoretical ages for
which individuals are supposed to have completed at least secondary education. This
restriction enables me to observe different educational transitions since the child enters
the educational system until the higher attained level. After excluding observations with
missing data on key interest variables, I obtain a final sample of 2,349 individuals.
Table 2 provides summary statistics for the final sample, and by gender and race.
More than half of the sample is female (52%) while the proportion of afro-descendants
is 11%.14
A first difference is observed between afro and non-afro descendants in terms of
their parental educational backgrounds. For instance, the proportion of non-afro
descendants with high educated parents (more than 12 years) doubles afro-descendants
rate, while the proportion of afro-descendants with low educated parents is 20% higher
than for non afro-descendants.
It is worth mentioning that pre-school enrollment, despite not being compulsory
for the population considered covered a large proportion of the total sample (more than
80%). Primary education is almost universal (98% of the total sample completes this
level), however a big concern refers to the high repetition rates observed for the whole
sample (25% of children repeated at least once in primary), rate that worsens for afro-
descendants (41%) almost doubling non afro-descendants’ (22%).

13
One exception is the Extended National Household Survey (ENHA: Encuesta Nacional de Hogares
Ampliada) carried only on 2006, an extended survey with a specific module on education.
14
Afro-descendance is captured in the ECH through the following question: “Do you believe you have…
(black or afro, Asian, white, native, other) descent?”. The respondent can choose more than one option of
racial descent. For this study, individuals reporting having black or afro descent are classified as afro-
descendants. Non-afro descendants are all individuals reporting not having afro-descent (thus, including
whites, Asian, native or other). It is worth noting that almost 90% declares only white descent, while less
than 5% declares having native or other descent.

138
 A second difference arises across genders when observing performance in
primary level in which girls do better than boys (21% of girls repeated at least once
versus 27% of the males). Tables 3 and 4 present summary statistics for different
schooling levels for girls and boys respectively. Some observations can be made from
these tables.
First, the proportion of children dropping out at each educational level is mainly
from lower parental background (representing more than 70% in lower high school, and
more than 40% in upper high school) In addition, it is observed that, while the
proportion of students from disadvantaged parental educational background enrolled is
lower at higher levels of schooling, the proportion of children from better-off parental
educational background completing lower and upper high school and enrolled in
postsecondary increases. The share of children from medium parental background
enrolled and completing each level is stable across the educational path. These
frequencies suggest that in Uruguay, transitions turn more selective for boys and girls
from less advantaged parental educational background.
Second, afro-descendants are more likely to drop out in lower and upper high
school than non afro-descendants. Especially for girls, the proportion of afro-
descendants that drops out at each stage is more than twice the proportion of those
enrolled at each level. Third, worse performance in primary and secondary level seems
to prevent students from attaining higher levels of education. Note that the proportion of
students who have never repeated primary level increases across schooling levels at the
time that the proportion of repeaters decreases. A similar pattern is observed when
focusing on repetition in secondary in which those more likely to survive higher
schooling stages are those who performed better in secondary. Also, it is striking that
the proportion of students enrolled in postsecondary education who have repeated
primary is almost zero for both genders.
Differences across genders emerge across post-secondary enrollment for
repeaters in secondary level. The proportion of repeater girls enrolled in post-secondary
education is 13%, half of the rate observed for boys (24%).
Regarding to our proxies of non-cognitive ability, it is observed that the
proportion of boys who drop out lower high school with a risky behavior (tried
marijuana before age 15) is almost twice that of girls (9.5% for boys and 4.9% for girls).
For both genders it is found that the proportion of students that highly value education,

139
those more motivated to participate in secondary level, increases across the schooling
stages.
Finally, a great proportion of students dropping out from the educational system
are mainly those who attended all grade years of each stage in a public institution.
Overall, differences found across genders and among the educational path justify
a disaggregated analysis by gender, and through a sequential model, in the sense that the
educational system seems to turn more selective in boys’ and girls’ parental educational
background, past performance in schooling stages, motivation for enrollment, afro-
descendant girls especially between the first and second schooling stages, and those
receiving public education.

3.5. Methodological framework

This analysis attempts to answer the following questions: i) to what extent parental
education, cognitive and non-cognitive abilities and individual’s race influence
education attainment? and, ii) is there any differential effect of parental educational
background, individual’s cognitive and non-cognitive abilities and race at different
stages of schooling transition?
In order to answer these questions, the estimation strategy follows Cameron and
Heckman (2001) in which education attainment is analyzed through a dynamic discrete
choice model of schooling progression. This strategy recognizes that schooling
attainment is the outcome of previous educational choices, which in turn depend on
individuals’ observable characteristics, like gender, race/ethnicity, and family
background; but are also influenced by her unobservable characteristics, such as
motivation or ability. The probability that an individual enters post-secondary education
depends on upper high school graduation, which in turn depends on completing lower
high school, making the model fundamentally recursive.
Therefore, individuals moving from one educational level to the next one may
differ on their unobservable characteristics, in which the less able or motivated
individuals are less likely to succeed in the transitions to higher education stages.
Cameron and Heckman (2001) stress that at each stage the opportunity cost of schooling
attendance is different, for instance labor market opportunities and earnings may vary
according to individuals’ education. Hence, it is possible that the opportunity cost of
education increases for higher schooling stages when the system turns more selective

140
across the educational path. The intuition behind the model is that if the student
population is divided between high and low ability individuals; and in turn between
those coming from wealthier households and poorer ones; then it is expected that (i)
more able individuals are more likely to succeed in higher educational stages in
comparison to less able ones; and (ii) individuals coming from poorer households,
ceteris paribus, may be prevented to move to the next educational level because of the
household financial restrictions. Therefore, the ones surviving higher schooling stages
are a selected sample of those more able individuals and with wealthier or better-off
parental background, making important to control for the effects of such educational
selection in order to isolate the causal effects of family background variables on
education attainment.
Overall, in a dynamic framework, two factors induce biased estimations of the
effects of family background on schooling progression. The first one refers to omitted
variables (that is, not accounting for individuals’ ability or motivation), while the
second one refers to the selection taking place at different stages of the schooling
transitions.

3.5.1 A sequential model of schooling progression

Following Cameron and Heckman (2001) the model assumes that each individual make
schooling decision based on a sequential choice model. The choices available to the
individual are limited by their earlier schooling choices.
The expected utility derived from each educational level is modeled as a latent utility
∗
index𝑦𝑖𝑠 :

∗ ′
𝑦𝑖𝑠 = 𝑋𝑖𝑠 𝛽𝑠 + 𝛼𝑠 𝜃𝑖 + 𝑢𝑖𝑠 ⁡⁡𝑖 = 1, … , 𝑁; ⁡𝑠 = 1, … , 𝑆⁡ (1)

where 𝑋𝑖𝑠 is a vector of observed constraint and expectation variables relevant to

schooling decision s, 𝜃𝑖 are unobserved factors for the econometrician but known to the
agent. This is the source of the essential heterogeneity, which can reflect individual’s
ability, motivation, or preferences; while 𝑢𝑖𝑠 ⁡⁡represents an idiosyncratic error term
which is assumed to be independent of the explanatory variables (𝑋𝑖𝑠 ⁡and 𝜃𝑖 ) and is
independent across individuals. Also, it is assumed that: 𝑢𝑖𝑠 ~𝑁(0,1)

141
 ∗
1⁡⁡⁡⁡⁡if⁡𝑦𝑖𝑠 ⁡ ≥ 0⁡
Then, I can define the binary outcome 𝑦𝑖𝑠 = { ⁡ (2)
0⁡⁡⁡⁡⁡otherwise

These assumptions allow writing down the probability of making choice s as a probit
model. Conditioning on 𝜃,

′
Pr(𝑦𝑖𝑠 = 1⁡| 𝑋𝑖𝑠 , 𝜃𝑖 , 𝑦𝑖𝑠−1 ⁡) = Φ(𝑋𝑖𝑠 𝛽𝑠 + 𝛼𝑠 𝜃𝑖 ) (3)
where 𝑦𝑖𝑠−1 are the past decisions made by the individual i and Φ(. ) is the standard
normal cumulative distribution function.

The probability of any sequence of schooling choices made by the individual 𝑦𝑖𝑠 given
the observed variables and 𝜃𝑖 can be expressed as:
∏𝑠⁡𝜖⁡𝐶𝑖[ 𝑃𝑟(𝑦𝑖𝑠 = 1⁡| 𝑋𝑖𝑠 , 𝜃𝑖 , 𝑦𝑖𝑠−1 ⁡)]𝑦𝑖𝑠 [𝑃𝑟(𝑦𝑖𝑠 = 0⁡| 𝑋𝑖𝑠 , 𝜃𝑖 , 𝑦𝑖𝑠−1 ⁡)]1−𝑦𝑖𝑠 (4)

where 𝐶𝑖 is the set of decision nodes that individual i has visited.

3.5.2 Empirical strategy

In this study, I consider three education levels for girls and boys separately: lower high
school (𝑦𝑖1 ), upper high school (𝑦𝑖2 ) and postsecondary level (𝑦𝑖3 ). Then, the sequential
process for individual i consists on: first decide whether or not to complete lower high
∗
school based on the underlying and unobserved expected utility (𝑦𝑖1 ). After completing
lower high school, the individual decides whether or not to complete upper high school
∗
(y2) conditional on the expected utility (𝑦𝑖2 ). Finally, for those graduating from upper
high school, the individual chooses whether or not to enroll in postsecondary education
∗
(y3) conditional on the expected utility derived from this election (𝑦𝑖3 ).

The sequence of life cycle schooling histories can be written as follows:

s={1, 2, 3}

∗
1⁡⁡⁡⁡⁡or⁡completing⁡lower⁡high⁡school⁡if⁡𝑦𝑖1 ⁡ ≥ 0⁡
𝑦𝑖1 = { (5)
0⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡otherwise

For those individuals completing lower high school,

∗
1⁡⁡⁡⁡⁡or⁡completing⁡upper⁡high⁡school⁡if⁡𝑦𝑖2 ⁡ ≥ 0⁡
𝑦𝑖2 = { (6)
0⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡otherwise

142
Finally, for those individuals graduating for upper high school
∗
1⁡⁡⁡⁡⁡⁡⁡⁡⁡or⁡enrolled⁡in⁡postsecondary⁡if⁡𝑦𝑖3 ⁡ ≥ 0⁡
𝑦𝑖3 = { (7)
0⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡otherwise

Given the two levels of selection and the outcomes we have four types of individuals:
 Those who choose not to complete lower high school 𝑦𝑖1 = 0
 Those who complete lower high school but decide not to continue upper high school
𝑦𝑖1 = 1, 𝑦𝑖2 = 0⁡
 Those who complete upper high school but decide not to enroll in postsecondary
education: 𝑦𝑖1 = 1, 𝑦𝑖2 = 1, 𝑦𝑖3 = 0
 Those who decide to enroll in postsecondary education: 𝑦𝑖1 = 1, 𝑦𝑖2 = 1, 𝑦𝑖3 = 1

For each of the educational levels stated before, the conditional probabilities are:

Not completing lower high school:

′
Pr(yi1 = 0⁡| Xi1 , θi ⁡) = 1 − Φ(Xi1 β1 + α1 θi ) (8)

Completing lower high school and not continuing:

′
Pr(yi1 = 1⁡| Xi1 , θi ⁡) = Φ(Xi1 β1 + α1 θi ) (9)

Not completing upper high school:

Pr(yi2 = 0⁡| Xi2 , yi1 θi ⁡) =
′ ′ ′
Φ(Xi1 β1 + α1 θi ) − Φ2 (Xi1 β1 + α1 θi , Xi2 β2 + α2 θi , ρ12 ) (10)

Completing upper high school and dropping out:

′ ′ ′
Pr(yi2 = 1⁡| Xi2 , yi1 θi ⁡) = Φ2 (Xi2 β2 + α2 θi , Xi1 β1 + α1 θi ⁡, ρ12 ) − Φ3 (Xi1 β1 +
′ ′
α1 θi , Xi2 β2 + α2 θi , Xi3 β3 + α3 θi , ρ12 , ρ13 , ρ23 ⁡) (11)

Being enrolled in postsecondary education:

Pr(𝑦𝑖3 = 1⁡| 𝑋𝑖3 , 𝑦𝑖2 , 𝜃𝑖 ⁡) =
′ ′ ′
Φ3 (𝑋𝑖1 𝛽1 + 𝛼1 𝜃𝑖 , ⁡𝑋𝑖2 𝛽2 + 𝛼2 𝜃𝑖 , ⁡𝑋𝑖3 𝛽3 + 𝛼3 𝜃𝑖 , 𝜌12 , 𝜌13 , 𝜌23 ⁡) (12)

where Φ(. ) is the standard normal cumulative distribution function, Φ2 (.) is the
bivariate standard normal cumulative distribution with correlation coefficient 𝜌12 and
Φ3 (. ) is the trivariate standard normal cumulative distribution with correlation
coefficients 𝜌12 , 𝜌13 , 𝜌23 .

143
 𝜌12 = 𝑐𝑜𝑣[𝑢1 , 𝑢2 |𝑋1 , 𝑋2], 𝜌13 = 𝑐𝑜𝑣[𝑢1 , 𝑢3 |𝑋1 , 𝑋3], 𝜌23 = 𝑐𝑜𝑣[𝑢2 , 𝑢3 |𝑋2 , 𝑋3]

The model is estimated using maximum-likelihood technique of the joint trivariate

sample selection model.15

3.5.2.1 Explanatory variables

The explanatory variables included in the analysis are: race, parental educational
background (both mother and father highest educational level attained), age cohort fixed
effects, regional fixed effects, dummy variables indicating whether the child attended or
not to pre-school, and the type of institution attended in primary level.
In order to capture individuals’ cognitive ability, grade repetition in educational
level is included in the analysis. Specifically, I consider whether the child never
repeated, repeated once or more than once primary level. When considering the
probability of completing upper high school or the probability of enrollment in
postsecondary education, I also consider whether the individual never repeated, repeated
once or more than once secondary level.
Repetition takes place when the child does not obtain the knowledge and skills
necessary to think critically and solve complex problems in different areas, needed to
succeed in the society and economy where they live. These maturing mental abilities are
thought to broadly underpin learning and cognitive skills. In the Uruguayan education
system, it is the teacher's decision to assess whether the student meets at the end of the
year the minimum requirements for promotion.16 17
Overall, non-repeater students are
those who obtained the necessary knowledge and maturity to pass to the next grade.
For cognition, there is a fairly well-established set of terminologies and
conventions in the literature, for which aptitude tests and achievement test are designed
to capture the speed at which the individuals learn and the knowledge they acquire

15
This technique ensures consistent estimators (Rosenman et al, 2010).
16
Discretion may affect grading marks as teachers may have different preferences or expectations.
17
In lower high school and first grade in upper high school, students are assigned a mark for each of the
12 taught subjects based on their performance during the year. Students pass a subject if they get a mark
above a given threshold. Those who fail a subject must re-take it during subsequent exam sessions
(Manacorda, 2008; p7). For grading promotion in second and third year in upper high school, exams in
particular subjects are mandatory.

144
(Almlund, et al., 2011) .18 Thus, for the aim of this paper, repetition seems to be a good
proxy of cognition.19
In addition, two variables are used in order to proxy non-cognitive ability. First,
I consider motivation for enrollment in secondary level. Although this variable is not
explicitly recognized as a factor in the Big Five model, Almlund et al. (2011) stress that
one of the main critics received by this model is that it is silent about motivation.
However, as also pointed out in Almlund et al. (2011), some studies relate academic
motivation to Openness to Experience (p136).
The ENAJ asks individuals about the motives for secondary enrollment. Based
on the alternative responses given to this question, I categorize the enrollment motives
as: high motivation (those individuals reporting high value of education), labor motives
(individuals declaring enrollment while they find a job), and not motivated (individuals
declaring enrollment because they were “pushed to”). I expect most motivated
individuals to be more likely to complete lower and upper high school, as compared to
those who are less motivated to acquire education. Table A.4 in the Appendix provides
a detailed description of the construction of this variable.
Second, I consider a dummy variable equal to one if the individual has tried
marijuana before age 15. As was outlined in Section 2, this risky behavior was found to
be negatively related with Consciousness (Gullone and Moore, 2000) and to have a
negative influence on schooling progression (Heckman et al., 2014).
Models of educational choices also include additional choice-specific covariates.
First, I consider the type of institution attended at different levels of high school. Public
institution (both in lower or upper high school) is a dummy variable equal to one if the
individual completed all grades of the corresponding level in a public institution and
zero otherwise (those with at least one grade attending in a private institution). In
general, the choice of a school, e.g., a private (fee paying) school, may reflect parental
motivation to produce children of better quality (i.e., with higher schooling). For
instance, a private school is likely to be of a better quality than a public school in the
sense that may provide better infrastructures, better teachers, better peers, lower ratio of
students per class; possibly affecting the probability of completing a schooling level.20

18
For a deeper discussion on intelligence, see Chapter 4 in Almlund et al., (2011).
19
It is worth mentioning that cognitive ability is likely to be influenced by child’s environment, such as
parental education, issue that is controlled for in this analysis.
20
See Checchi (2006) Chapters 4 and 5 for an extensive review of the literature on the influence of supply
of education and education financing on education attainment.

145
 Also, the track chosen in secondary level is considered. While in lower high
school there are no significant differences in curricula between general education and
vocational training education, for upper high school differences turn to be important.
Vocational training education is more oriented toward job placement (but is also
possible to continue to tertiary education) than general academic education. In addition,
the track chosen may also reflect individuals’ self-selection if more able individuals
choose general education instead of vocational training.21
Finally, internal migration is considered for postsecondary enrollment. As was
stated before, universities in Uruguay (both public and private ones) are mainly located
in Montevideo, so those individuals with financial family support are more likely to
migrate to Montevideo and to attend university than poorer ones. Motive for migration
is a categorical variable that captures whether the individual did not migrate after
completing secondary level, if migrated for study motives, or migrated for other
motives.22

3.5.2.2 Exclusion restrictions

In this analysis, as is standard in the literature I use exclusion restriction variables in
order to identify the model. The model requires a subset of variables influencing the
probability of attaining a certain educational level and not directly affecting the
probability of completing the next one. Also, the exclusion restriction variables are
assumed to be independent of the model unobservables.
As in previous studies (Cameron and Heckman, 2001; Heckman et al., 2014;
Bernardi, 2012; Pal, 2004) I consider as exclusion restriction variables reflecting labor
market conditions at each schooling stage, at the time the relevant decisions are taken.
In each schooling stage, the individual chooses between completing the education level
and dropping out to participate in the labor market. The decision is made considering
the current labor market conditions and expectations on future returns to education.
These rates may account for the opportunity cost of education. Then, if the individual
continues in the educational system, the decision to attain the next schooling stage will
depend on the opportunity cost of education –on the labor market conditions– at the
time the decision of completing the level is evaluated. In other words, labor market

21
An interesting debate in the educational literature refers to the consequences of the time of tracking on
equity and efficiency of educational outcomes. See for instance van Elk et al. (2011).
22
Other motives for migration are mainly labor, health, and family motives declared for migration.

146
conditions at time t influences schooling choices at time t, and only indirectly affecting
schooling decisions of completion of the next level taken in t+1. It is clear that if the
individual decides to drop out from the system in lower high school he is indirectly
deciding not to attain upper high school, because of the sequential process of education
attainment, but the individual cannot decide completing upper high school if lower high
school was not achieved. Also, these rates are exogenous to individuals’ schooling
decisions.
A priori, the role of local labor market conditions is unclear. On the one hand, a
high probability of employment might convince students to quit school and enter the
labor market. On the other hand, the higher expected education returns could
definitively be a stimulus for acquiring further education (Moccetti, 2008).
Specifically, I consider unemployment and employment rates, which are
calculated for young people (aged 24 or less years old), by gender and at the department
level at theoretical ages in which the individual is supposed to be enrolled in each
schooling stage. Employment rates considered at each stage of the schooling
progression are the following: unskilled youth employment rate for those children
deciding whether to complete lower high school, semi-skilled youth employment rate
for those choosing to complete upper high school, and youth skilled employment rate
for individuals considering post-secondary enrollment.
Detailed information on the elaboration and classification of the variables are
provided in Table A.4, while a summary of the independent variables considered in this
analysis is provided in Tables A.5 and A.6 in the Appendix.

3.6 Results
In this section I first focus on the results related to unobserved heterogeneity and its
correlations. Next, I describe the implications of the estimates of the model by
discussing in turn, (1) the determinants of the probability of the initial schooling stage,
(2) the determinants of upper high school transitions for those who completed lower
high school, (3) the postsecondary enrollment decision for those surviving previous
schooling stages (subsection 6.1). Next, subsection 6.2 gives a more complete picture of
the educational path for boys and girls living in Uruguay.

147
3.6.1 Unobserved heterogeneity and correlations
A trivariate probit model with sample selection is estimated separately for females and
males. Before presenting the estimated results, a natural question that emerges in this
type of model is whether it is necessary to control for unobserved heterogeneity.
Estimates of the cross-equation correlations between unobservables provide insights of
the endogenous selection processes. In other words, the significance of the correlations
highlights the importance of estimating education attainment as a sequential process.
In Table 5 it is shown that for both genders, unobservables across the three
schooling levels are negatively associated although differences exist in the statistical
significance of the estimated correlations. For girls, statistical and significant
association is detected between the first and second transitions, while for boys between
the second and third transitions. Thus, results show that the three schooling stages are
differently interlinked and differ for both genders. Unobserved factors that make girls
more likely to succeed in lower high school reduce their likelihood of attaining upper
high school. For boys, unobserved heterogeneity that makes them more likely to
complete upper high reduces their chances to enroll in post-secondary education. Any
interpretation of this result is difficult. Recall that cognitive skills, motivation and risky
behavior as proxies of socio-emotional endowments, are controlled for in the model.
Therefore, these negative correlations between the residuals are capturing other
unobservables different from ability and motivation. It could be argued that cultural
factors, social pressure or labor market conditions, may induce children to achieve the
minimum educational credentials recognized by the society and, once these credentials
are obtained, children drop out from the educational system. Also, institutional and
organizational factors as well as differences in curricula and grading promotion, which
are specific of each schooling stage, could be differently affecting individuals’ decisions
of schooling. It could be speculated that these factors may influence children’s
adaptation or integration into different academic schemes.23
Tests for the ignorability of each selection mechanism were based on a Wald test
of whether every correlation connecting each equation of the model was equal to zero.
The null hypothesis of sample selection ignorability is rejected for both genders (bottom
panel of Table 5). Thus, the results provide strong evidence that not accounting for the

23
See Rama (2004) for an extensive description of the particularities of the institutional and
organizational factors in the Uruguayan educational system. Fernández-Aguerre (2010) summarizes
different empirical studies analyzing individuals’ drop out from different stages of schooling in Uruguay.

148
potential endogeneity resulting from unobserved heterogeneity would induce biased
results. This is also in line with the descriptive analysis provided in Section 3.
Tables A.5 and A.6 in the Appendix present the estimates of simple probit
models not accounting for sample selection, separately for girls and boys. The
magnitude of the bias could be observed by comparing the estimated coefficients of the
key independent variables between simple probit models and the ones obtained from the
trivariate probit estimations. Overall, it can be concluded that not accounting for
selection overestimates the effects of the key variables on education attainment.

3.6.2 Empirical findings

Next, I discuss the estimated effects of the explanatory variables from the trivariate
probit model. Average marginal effects capturing direct effects of the key variables on
the probability of different schooling stages separately for girls and boys are presented
in Tables 6 and 7 respectively.24
First, in line with the literature, parental educational background is an important
factor explaining schooling completion. For both genders it is shown that the likelihood
of attaining lower high school level for a child with highly educated parents more than
doubles the probability of those with medium educated parents (Column 1, Tables 6 and
7). Thus, at this schooling stage the system gives less opportunities of completion to
children from worse-off parental educational backgrounds.
Specifically, girls with a high educated mother are almost 12 percentage points
(pp) more likely to complete this level in comparison to a girl with a low educated
mother. For boys the opportunity gap between those with a high educated father and a
low educated father is almost 19pp. In turn, having a medium educated mother increases
the likelihood of completing this level in 5.7pp in comparison to girls from lower
parental education background. Boys with a medium educated mother or father are 5.7
and 4pp respectively more likely to attain this level than boys from a more
disadvantaged parental education.
Cognitive ability proxied by repetition in primary level decreases the probability
of graduating from lower high school. Boys repeating once or more than once are

24
Alternative specifications were also estimated not showing significant differences with the coefficients
presented in Tables 6 and 7. These estimations included interactions of: race and parental educational
background; race and motivation; motivation and parental education; repetition in both secondary and
primary with motivation; parental education and repetition; and repetition and race. None of these
interaction were statistical significant.

149
10.5pp and 15.1 pp more likely to drop out at this level than non-repeaters. Similar
effects of past performance on schooling attainment are observed for girls (10.3pp and
16.4pp respectively).
In line with what is expected in the literature, more motivated individuals are
more likely to complete lower high school. Girls and boys reporting enrollment in
secondary level because they were “pushed to” are less likely to complete this level in
comparison to those declaring high value of education (13.9 and 10.3 pp respectively).
Also, girls and boys reporting labor motives for enrollment in secondary are less likely
to complete this level than those more motivated ones, possibly putting less effort in
attaining this level because of the anticipated decision of dropping out from the system
once a job is found (4.2 and 8.3 pp respectively).
It is worth noting that at this schooling stage, while cognitive ability has similar
effects on the probability of schooling completion across genders, motives for
enrollment do not. Other things being equal, not motivated girls are more likely to leave
the system than not motivated boys.
In addition, the results point to lower opportunities for afro-descendant girls,
who are 5.1pp less likely to complete this educational level than non afro-descendants.
Conversely, race is not a significant factor preventing boys attaining this educational
level.
Next, the type of institution attended during primary level and lower high school
decreases the probability of successfully completing this level. Individuals attending all
grades in a public institution have lower chances to complete this level than those with
at least one year in a private institution (8.3pp and 11.1pp for girls, and 16 and 7.5pp for
boys, respectively for school and lower high school). Despite the heterogeneity in
quality across public and private institutions that could be found in Uruguay, the public
ones are associated in the literature with lower quality, in terms of resources and
infrastructure, number of students per teacher, peer effects, in comparison to private
ones. An alternative explanation is that private schools (mainly religious ones) are more
effective in producing more motivated students and self-disciplined students (Coleman
and Hoffer, 1983).25
It is worth mentioning that persistent effects of pre-school attendance are
observed for girls (5.3pp), while this effect vanishes for males. A possible explanation

25
Quoted in Carneiro and Heckman (2003) p39.

150
of this result is given in Apps et al. (2013). These authors stress that this result is quite
common in the international literature, and may be due to strong effect from improved
language skills (usually higher in girls), combined with the lower impact of negative
behaviors (like aggressiveness, and antisocial behaviors), which are more common in
boys (p.194).
Labor market opportunities have opposite effects on the probability of exiting
the education system across genders.26 For girls, higher unemployment rate decreases
the probability of completing lower high school. This effect could be reflecting girls’
future labor market expectations. If girls perceive that the labor market does not provide
great opportunities, then they are discouraged to invest in human capital, thus dropping
out from the system. For men, higher opportunities for unskilled workers increase the
probability of dropping out from the educational system. Both variables, which measure
opportunity cost of education, could be also measuring short-run family resource
constraints. When lack of resources in the household are observed, children are more
likely to drop out from the educational system in order to complement family’s income.
Next, I move on to analyze the determinants of upper high school attainment for
those surviving previous schooling stage (Column 2 in Tables 6 and 7). It is observed
that children with high and medium educated parents show higher probability of
graduating from secondary level relative to children from lower parental background.
Therefore, this educational stage is also found to be less supportive to children from
worse-off parental educational background giving them fewer opportunities to attain
this schooling level.
Specifically for girls, having a high educated mother or father increase the
probability of completing upper high school in comparison to girls with a low educated
parent (16.8 and 11.6pp respectively). Boys with a high educated father are 13.6pp more
likely to complete this level than those with low educated fathers. Also, boys with
medium educated fathers and girls with medium educated mothers are more likely to
complete this level in comparison to those with less educated parents (7 and 5.5pp for
boys and girls respectively).
Second, race is an important factor deterring girls’ and boys’ upper high school
completion although the effect is greater for afro-descendant girls. This is observed

26
Legal age for participating in the labor market is 14 years old in Uruguay for the period of analysis.

151
when comparing the statistical significance of both marginal effects, 9.4pp at 99% of
significance for girls and 13.1pp at 90% for boys.
Past performance in secondary level is the most important factor in explaining
students’ probability of dropping out from the system. Having repeated this level once
increases the probability of dropping out in 25pp and almost 30pp for girls and boys
respectively; while students repeating more than once are 34.5pp and 37pp less likely to
graduate from upper high school than non-repeaters (girls and boys respectively). Note
also the persistent effect of past performance in primary on the next levels of the
educational system, not only indirectly affecting the probability of dropping out the
system in an early stage but also directly decreasing the likelihood of leaving upper high
school (14.3 and 18.7pp for girls and boys respectively). Thus, consistent with Cameron
and Heckman (2001), differences in cognitive ability appear at early ages and persist
over time.
Socio-emotional factors proxied by risky behavior and motivation influence
schooling progression but play different roles across genders. For instance, motivation
for enrollment still explains girls’ but not boys’ success in attaining upper high school.
Girls who reported having been “pushed to” attend secondary level are 15.6pp less
likely to complete upper high school than more motivated ones, while non statistically
significant effects of risky behavior on upper high school completion are observed for
girls. Conversely, risky behavior has negative and statistically significant effect on
boys’ probability of completing upper high school (almost 18pp significant at 95%)
whereas motivation for secondary enrollment is not statistically significant. This is
consistent with the psychological literature stressing different adolescent personality
traits and propensity to be engaged in risky behaviors between male and female
adolescents (see Gullone and Moore, 2000).
Also, differences across genders are observed in relation to the effect of the type
of institution enrolled on upper high school completion. Girls who attended all grades
into a public institution are 6.2 pp less likely to complete this level than those with at
least one year in a private institution, while no statistical and significant effect is
observed for boys. In addition, students (or their parents) choosing a general academic
track are more likely to survive this educational stage than those tracked in vocational
training education or those with mixed tracks (those who have changed between tracks
within upper high school).

152
 Labor market conditions also influence children’s decisions on schooling
completion. For boys, higher semi-skilled employment rate when the child is aged 15
(the theoretical age for attaining first grade in upper high school) decreases the
probability of completing this level in 5.3pp. In turn, higher unemployment rate when
girls are aged 15, increases in 2.8pp the probability of dropping out from the system.
Thus, favorable labor market conditions for semi-skilled workers increase the
opportunity cost of education for boys, while less attractive labor market conditions
decrease the opportunity cost of schooling for girls.
Finally, the determinants of postsecondary enrollment are analyzed for those
students surviving previous schooling level (Column 3 in Tables 6 and 7). Two main
variables explain participation in postsecondary education for boys and girls. First,
different opportunities in postsecondary enrollment are still observed for students from
different parental educational background. For instance, boys with medium educated
fathers and high educated fathers are respectively 14.1pp and 35.6pp more likely to
attain postsecondary education than those from low educated parental background. In
turn, girls with a high educated father are more likely to be enrolled in postsecondary
education in comparison to girls with low educated father (8.6pp at 10% of
significance), while no statistical and significant effect is observed between girls with
low and medium educated parents. Therefore, this level seems to be more unequal for
boys than for girls, in the sense that parental educational background influences more in
boys’ enrollment.
Second, internal migration after finishing secondary level is an important
variable influencing individuals’ postsecondary enrollment. Those declaring study
motives for internal migration are more prone to be enrolled in this educational stage in
comparison to not migrating ones (13.1 and 20pp for girls and boys, respectively).
Internal migration for study motives could be reflecting household permanent income
on the understanding that, as far as postsecondary institutions, mainly the public
University (UdelaR) and private universities are located in Montevideo, those students
not living in the capital and wanting to continue college should move to the capital,
assuming all the related costs of this decision, like housing, food, etc. In other words,
wealthier families are more likely to invest in their children’ postsecondary education in
comparison to poorer families.
It is also worth mentioning that neither race, past performance in the educational
system, adolescent risky behavior nor motivation for secondary enrollment are

153
important direct determinants of postsecondary enrollment for any gender. This is
explained because a great proportion of afro-descendants, less able and motivated
individuals did not “survive” the previous stages and that almost all who survive and
can afford moving to Montevideo (if were living in the Interior of the country) are
enrolled in postsecondary level. This is also consistent with the descriptive analysis
presented in Section 5. Overall, this educational stage seems to be more homogeneous
in terms of individuals’ observables and unobservable characteristics, leaving aside
afro-descendants, individuals from less advantaged parental educational backgrounds
and from poorer households, and those who performed worse in previous schooling
stages, less motivated and more risky behavior.

3.6.3 Interpretation of results

This subsection describes the main findings of this study characterizing the educational
path for girls and boys separately.
In the first stage of girls’ schooling transition, parental educational level, past
performance in primary level, motivation reported for attending secondary level, race
and parental choices in terms of pre-school enrollment and types of institution attended
in primary and in lower high school, play a major role in explaining lower high school
attainment. Also, short-term income restrictions affect girls’ probability of completing
this level, but are less statistically significant than long-term family factors.
In the next educational level, the system turns more unequal in terms of
opportunity gap between girls with a high educated mother or father in comparison to
girls from worse-off parental background. This is observed when comparing the
marginal effects of different educational backgrounds on the probability of attaining this
level. Note that girls with a high educated parent (mother or father) are more likely to
graduate from upper high school than in the previous educational level, while the
marginal effect of being from a medium parental educational background decreases its
statistical significance. In the last transition, the estimated effect of parental education
decreases its magnitude and looses statistical significance, interpreted as less unequal
than previous stages in the sense that the opportunity educational gap between girls
from different parental educational background decreases.
Second, it is observed that the opportunity gap of completion for girls with
different scholastic abilities increases in comparison to previous stages. In turn,
motivation is still an important factor fostering upper high school completion but is less

154
significant than in the previous stage. This could be due to less motivated individuals
being less likely to survive the previous level and this stage is more “homogenous” in
terms of motivated individuals. Nonetheless, the decreasing effects of non-cognitive
abilities should be interpreted with caution. First, because we are measuring something
that is unobservable for the researcher, and therefore the proxies used in this kind of
studies are at best imperfect. Second, because as noticed by the psychological literature,
socio-emotional factors could be influenced over the individual’s life cycle, for instance
by schooling.27 Therefore, we can only state that those who declared enrolling in
secondary level because at this time they highly valued education are more likely to
complete this level than those who reported being “pushed to”.
Finally, postsecondary level could be seen as the less unequal schooling stage
for girls from different parental educational background, race, and abilities. It is
observed a great homogeneity in terms of girls’ characteristics in this level, mainly
explained because afro-descendant, less able and motivated girls and from worse-off
parental backgrounds are less likely to survive previous schooling stages.
As a consequence for the surviving girls, enrollment in postsecondary level is
almost determined by the possibility to migrate and to less extent, for those having a
high educated father. Therefore, the higher we move in the educational system, the more
unequal the system becomes in terms of opportunities given to girls from different
parental backgrounds. It is also observed that the opportunity cost of education has
different effects across girl’s educational path. While in the first stage of schooling
progression, fewer opportunities in the labor market increase girls’ likelihood of
schooling drop out, in the second stage worse conditions in the labor market increase
the probability of completing this level. In addition, the statistical significance of this
coefficient decreases across the educational path, possibly reflecting that opportunity
cost of education is less important at higher schooling stages.
Similar patterns of selection are observed in the schooling transitions for boys, in
the sense that more we advance in the educational path, the fewer the boys from
disadvantaged parental educational background, less motivated and with worse
performance in primary and secondary, have a chance to attain higher educational
levels. Overall, it is observed for both genders that cognitive abilities has persistent and
increasing effects in the probability of attaining higher schooling levels. Socio-

27
There is an interesting ongoing debate in the psychological literature on the permanent versus
variability in personality traits across the individual’s life cycle. See for instance Almlund et al. (2011).

155
emotional factors, while important decreases its impact across the schooling
progression.
Some differences across genders are observed. For instance, upper high school
becomes less unequal for boys from different parental educational background than in
the previous stage since the estimated coefficient decreases (for high educated father)
and looses statistical significance (medium educated father). In turn, postsecondary
level turns to be the more unequal one for boys from low and medium parental
educational background in comparison to the previous levels.
Second, race has a major role in preventing girls from graduating from lower
and upper high school than for boys, for whom race is only statistically significant in
the second stage. Since interactions between race and cognitive abilities; race and
motivation for secondary enrollment; and race and parental educational background
were not statistically significant (see footnote 18), we can rule out that the estimated
negative effect of race on schooling progression is due to differences in terms of
parental educational backgrounds, motivation or cognitive abilities. Different
interpretations are given by the literature for this negative and statistically significant
coefficient. For instance, Porzecanski (2008) stresses that this negative coefficient could
be capturing different processes of discrimination. One the one hand, it may reflect
discrimination within the educational system which in turn affect afro-descendants’
schooling decisions. On the other hand, it could be associated to discrimination in the
labor market where returns to education are lower for afro-descendants, then
discouraging afro-descendants to acquire more education.
Third, motivation and risky behavior show different effects across genders.
While motivation is an important factor deterring girls’ schooling progression, for boys
it is only important for completing lower high school. Moreover, risky behavior turns to
be an important factor in explaining boys’ upper high school graduation; but not
significant in explaining girls’ schooling attainment.
Fourth, children’s (or their parents) decisions in terms of type of institution
attended have negative and decreasing impact on boys’ and girls’ schooling completion,
but is more significant for girls than for boys (for whom in the second stage it is not
statistically significant).
The results summarized above are consistent with the recent literature that
highlights the importance of individuals’ multiple abilities across one individual’s life
cycle. This literature stresses that cognitive ability is determined early in life while non-

156
cognitive ability is more malleable later in life. Specifically, Heckman and Carneiro
(2003) point out that cognitive ability is formed relatively early in life and becomes less
malleable at later stages of child’s development. According to these authors, by age 14,
intelligence as measured by IQ tests seems to be fairly well set. Non-cognitive skills, in
turn appear to be more malleable until the late adolescent years (Heckman and Mosso,
2014) thus allowing public interventions contribute to the formation of non-cognitive
skills (Brunello and Schlotter, 2011).
Heckman and coauthors refer to long run family factors crystallized in parental
educational background, in scholastic ability and socio-emotional factors, as the driving
force behind schooling attainment, and not short-term credit constraints.
In this study, because of lack of data on family’s income or wealth data at the
time of schooling choices are made, the effect of short-term family income is reflected
by the opportunity cost of education measured by labor market variables. In line with
Cameron and Heckman (2001) and Carneiro and Heckman (2003) who show that short-
term family income is more important for high school dropout and completion than for
college enrollment decision, I find that the opportunity cost of education is significant in
explaining educational level’ attainment, but its effect is smaller in comparison to long-
term family factors and decreases along the educational path.
Finally, as was mentioned before, the public University (UdelaR) has been
making big efforts in terms of territorial decentralization since 2007. These actions
could indeed have a positive effect in terms of access to postsecondary education for
students from low and middle educational background in the Interior of Uruguay. The
literature analyzing the impact of higher education supply expansion points that any
reduction in the influence of at least one circumstance on individuals’ educational
choices can be considered as reducing inequality of opportunity in education (see for
instance Bratti et al., 2008; Peragine and Serlenga, 2007). Expanding supply in
postsecondary education institutions may be associated to a cost-reduction effect,
related to the increased supply and the possibility of enrolling at a university without
moving to a different city. Also, expansion of higher education institution is associated
to a potential increase in the expected returns of a higher schooling due to the wider and
more diverse available offer (Bratti et al., 2008). Then, if new entrants are children from
less privileged families, the effect of expansion may be the one of inclusion and
increasing equality of opportunity almost by definition. But also, this literature
recognizes that if barriers of access exist, such as fee payment, credit markets

157
imperfections, or selection tests, the effects of the supply expansion on improving
equality of educational opportunity is not so obvious.
Conversely to other educational systems, public university in Uruguay does not
rely on scholastic ability and willingness to pay. Therefore, it could be expected that
territorial decentralization may benefit students from lower family backgrounds if
policy interventions aiming to correct the selection process operating in previous stages
takes place. In other words, in order to take full advantages of this decentralization
process and the system to be inclusive in terms of less advantaged children, public
interventions in secondary level are mandatory. In particular, policies intended to
improve the environment that shape child’s multiple abilities at different levels of the
educational path will be more effective in increasing schooling progression in the long
run.

3.7 Conclusion
In this paper, I analyzed to what extent long-term family factors crystallized in parental
educational background, race, cognitive and socio-emotional endowments, as well as
short-term family income proxied by the opportunity cost of education influence child’s
schooling progression. By analyzing the impact of these key variables across different
stages of the educational path, this analysis gives a more complete overview of the
major caveats of the Uruguayan educational system and about the factors that
differentially affect girls and boys’ educational attainment and gives insights of the
inequality of acquisition in education at each stage of schooling progression.
I use the National Youth Survey containing individual information on education
achievement and performance across the educational path, risky behavior and
motivation for secondary enrollment, internal migration and schooling choices in terms
of type of institution attended, among others.
The empirical strategy considers a sequential probability model developed by
Cameron and Heckman (1998, 2001) in which schooling attainment is modeled as the
outcomes of sequential choices made at each educational level, individuals’ unobserved
heterogeneity and alternative schooling cost of attendance at different levels. By taking
into account the selection on education attainment, we obtain unbiased estimated
results. Also, this analysis provides information on the different roles played by the key
variables at different stages of schooling progression.

158
 The results of this study confirms previous analyses addressing the deficiencies
of the secondary level education in Uruguay (Aristimuñ o, 2009; Manacorda, 2008;
among others). Furthermore, it extends previous research by considering the effects of
cognitive and non-cognitive abilities, jointly with parental educational background,
race, and opportunity cost of education measuring short-term family income, on
different stages of the educational path in Uruguay.
When measuring socio-emotional endowments we encounter multiple issues
largely recognized by the literature, such as the difficulty in capturing multiple
personality traits (due to its unobservable nature), data availability that limits the
measures of non-cognitive skills that can be used; and the static dimension of our
proxies.28
Despite these limitations in measuring non-cognitive ability, the presented
results gives enough evidence on the importance of both types of abilities in schooling
progression not only directly affecting each schooling stage, but also indirectly
influencing later stages.
In particular, the estimated results identify as one major deficiency of the
Uruguayan system, the inequality in the acquisition of education for children with less
scholastic abilities, the less motivated and with riskier behaviors, afro-descendants and
from worse-off parental educational background. Also, these variables have different
impacts as the students progress to higher schooling stages. This selection is observed in
both lower and upper high school thereby affecting individual’s probability of
enrollment in postsecondary education. As was noted above, Uruguay stands-out in the
region because it provides public education at all levels of the educational path.
However, our results indicate that free education does not fully guarantee that
individuals from worse-off family backgrounds (understood as less able individuals,
poorer parental educational backgrounds) have access to high levels of education. Then,
public policies should be oriented to mitigate those factors affecting individuals’
educational decisions, especially focusing on individuals’ from lower parental
educational background, less able and motivated individuals, and afro-descendants that
because of lower expectations or discrimination in the labor market and the educational
system are more likely to drop out the educational system.

28
Recall that there is no agreement in the psychological literature regarding how changes in personality
are affected over the individual’s life cycle.

159
 In addition, in light of the results of the analysis it can be stressed that if no
actions are taken to correct the inequalities observed in lower and upper high school, the
recent decentralization process carried out by the public university will not succeed in
providing more opportunities to those students from less advantaged parental
backgrounds.
The findings presented and discussed above gives support to policy interventions
at different stages of schooling progression in order to level the playing field for
children from different parental educational backgrounds, race, scholastic and non-
cognitive abilities. In particular, policies intended to promote cognitive ability early in
life and social and behavioral skills in adolescence and youth, mainly focused on
children from more disadvantaged environments –who probably receive little
encouragement and support at home– should be explored. Finally, girls and boys
develop alternative socio-emotional abilities across their life cycle, which in turn
influence differently schooling progression across genders. Also, race is an important
factor preventing schooling transition for boys and girls. Thus, promoting cognitive and
non-cognitive abilities from a gender perspective and taking into account ethnical/ racial
diversity may have positive effects on child’s achievement of higher education. Overall,
improving educational opportunities for less advantaged children will not only have
positive impacts on future labor market outcomes, but also on other social outcomes
such as crime and health, among others.

Aknowledgements
I am particular grateful to Raúl Ramos for kindly reading an earlier version of this paper
and for his insightful comments. Also, special thanks to the participants of the Lunch
seminar organized by Regional Quantitative Analysis Group (AQR) at Universitat de
Barcelona, and participants in the Lunch Seminar organized at the Departament
d’Economia Aplicada at Universitat Autònoma de Barcelona. The comments and
suggestions received were of great value for the elaboration of this paper.

160
 References

Almlund, M., Duckworth, A.; Heckman, J, and Kautz, T. (2011) “Personality

psychology and economics”, in E. A. Hanushek, S. Machin, and L. Wößmann (Eds.),
Handbook of the Economics of Education 4:1-181. Amsterdam: Elsevier.
Aristimuño, A. (2009) “El abandono de los estudios del nivel medio en
Uruguay: un problema complejo y persistente”, REICE - Revista Electrónica
Iberoamericana sobre Calidad, Eficacia y Cambio en Educación 7 (4).
Apps, P.; Mendolia, S., and Walker, I. (2013) “The impact of pre-school on
adolescents’ outcomes: Evidence from a recent English cohort”, Economics of
Education Review 37:183-199.
Barro, R. J.; and Lee, J.L. (2001) “International Data on Educational Attainment:
Updates and Implications,” Oxford Economic Papers 3:541-563.
Bernardi, F. (2012) “Unequal transitions: Selection bias and the compensatory
effect of social background in education careers”, Research in Social Stratification and
Mobility 30:159-174.
Betts, J., and Roemer, J. (2005) “Equalizing Opportunity for Racial and
Socioeconomic Groups in the United States through Educational Finance Reform”,
University of California Economics WP Series.
Blanden, J.; Gregg, P.; and Machin, S. (2002) “Education and Family Income”,
http://personal.lse.ac.uk/machin/pdf/Education%20and%20Family%20Income%20Stok
e%20Rochford%20Final%20Version.pdf
Bourgignon, F.; Ferreira, F.; and Menéndez, M. (2003) “Inequality of Outcomes
and Inequality of Opportunities in Brazil”, William Davidson Institute WP No. 630.
Bowles, S. and Gintis, H. (2001) “Schooling in Capitalist America Revisited”,
Sociology of Education 75 (1):1-18.
Bowles, S. and Gintis, H. (2002) “The Inheritance of Inequality”, Journal of
Economic Perspectives 16 (3):3-30.
Bratti, M.; Checchi, D.; and De Blasio, G. (2008) “Does the Expansion of
Higher Education Increase the Equality of Educational Opportunities? Evidence from
Italy”, LABOUR 22 (Special Issue):53-88.
Brunello, G, and Schlotter, M. (2011) “Non Cognitive Skills and Personality
Traits: Labour Market Relevance and their Development in Education & Training
Systems”, IZA DP No. 5743.

161
 Cameron, S., and Heckman, J. (1998) “Life Cycle Schooling and Dynamic
Selection Bias: Models and Evidence for Five Cohorts of American Males”, Journal of
Political Economy 106 (2):262-333.
Cameron, S., and Heckman, J. (2001) “The dynamics of educational attainment
for black, Hyspanic and white males”, Journal of Political Economy 109 (3):455-99.
Cappellari, L. (2004) “High School Types, Academic Performance and Early
Labour Market Outcomes”, IZA WP No. 1048.
Carneiro, P., Crawford, C. and Goodman, A. (2007) “The Impact of Early
Cognitive and Non-Cognitive Skills on Later Outcomes”, CEE DP No. 92.
Cepal (2013) Anuario estadístico de América Latina y el Caribe, Santiago de
Chile.
Checchi, D. (2006) “The Economics of Education, Human Capital, Family
Background and Inequality”, Cambridge University Press.
Cunha, F.; and Heckman, J. J. (2007) “The technology of skill formation”,
American Economic Review 97 (2):31-47.
Cunha, F. and J. J. Heckman (2008) “Formulating, identifying and estimating
the technology of cognitive and noncognitive skill formation”, Journal of Human
Resources 43 (4):738–782.
Cunha, F., J. J. Heckman, L. Lochner, and D. V. Masterov (2006) “Interpreting
the evidence on life cycle skill formation”, in E. A. Hanushek and F.Welch (Eds.),
Handbook of the Economics of Education, Chapter 12, pp. 697–812. Amsterdam:
North-Holland.
Da Silveira, P. and R. Queirolo (1998) “Son nuestras escuelas y Liceos capaces
de enseñar?”, CERES WP No.7.
Dustmann, C. (2004) “Parental background, secondary school track choice, and
wages”, Oxford Economic Papers, 56(2):209-230.
Fernández-Aguerre (2010) (coord. and ed.) “La desafiliación en la Educación
Media y Superior de Uruguay: Conceptos, estudios y políticas”, Colección Art.2,
Comisión Sectorial de Investigación Científica, Universidad de la República.
Ferreira, F., and Gignoux, J. (2008) “The Measurement of Inequality of
Opportunity: Theory and an application to Latin America”, The World Bank, Policy
Research WP 4659.
Francesconi, M.; and Nicoletti, C. (2006) “Intergenerational mobility and sample
selection in short term panels” Journal of Applied Econometrics 21:1265-1293.

162
 Furtado, M. (2003), “Trayectorias Educativas de los Jóvenes: el problema de la
deserción”, Cuaderno de trabajo TEMS, No. 22, Montevideo.
González, C., and Sanromán, G. (2010) “Movilidad intergeneracional y raza en
Uruguay”, DT No.13/10, Departamento de Economía, Facultad de Ciencias Sociales
Universidad de la República.
Gullone, E., and Moore, S. (2000) “Adolescent risky-taking and the five-factor
model of personality”, Journal of Adolescence 23:393-407.
Heckman, J., and Carneiro, P. (2003) “Human Capital Policy”, IZA DP No. 821.
Heckman, J., Humphries, J., Veramendi, G; and Urzúa, S. (2011) “The Effects
of Educational Choices on Labor Market, Health and Social Outcomes”, University of
Chicago WP No. 2011-002.
Heckman, J., Humphries, J., Veramendi, G; and Urzúa, S. (2014) “Education,
Health and Wages”, IZA DP No. 8027.
Heckman, J., and Mosso, S. (2014) “The Economics of Human Development
and Social Mobility”, IZA DP No. 8000.
Heckman, J. J., R. Pinto, and P. A. Savelyev (2013) “Understanding the
mechanisms through which an influential early childhood program boosted adult
outcomes”, American Economic Review 103(6):2052-286.
Heckman, J.; Stixrud, J.; and Urzúa, S. (2006) “The Effects of Cognitive and
Noncognitive abilities on Labor Market Outcomes and Social Behaviour”, NBER WP
No. 12006.
Heckman, J. J., Urzúa, S., and E. J. Vytlacil (2006). “Understanding
instrumental variables in models with essential heterogeneity”, Review of Economic
Statistatistics 88(3):389-432.
Holm, A.; and Jaeger, M. (2011) “Dealing with selection bias in educational
transition models: The bivariate probit selection model”, Research in Social
Stratification and Mobility.
Karlson, K. (2011) “Multiple paths in educational transitions: A multinomial
transition model with unobserved heterogeneity”, Research in Social Stratification and
Mobility 29:323-341.
Lundberg, S. (2013) “Educational Inequality and the Returns to Skills”, IZA DP
No. 7595.
Manacorda, M. (2008) “The Cost of Grade Retention”, CEP Discussion Paper
No 878.

163
 Mare, R. (1980) “Social Background and School Continuation Decisions”,
Journal of American Statistics Association 75:295-305.
Moccetti, S. (2008) “Educational choices and the selection process before and
after compulsory schooling”, Temi di discussione series WP No. 691.
Pal, S. (2004) “Child schooling in Peru: Evidence from a sequential analysis of
schooling progression”, Journal of Population Economics 17:657-680.
Peragine, V. and Serlenga, L. (2007) “Higher education and equality of
opportunity in Italy”, ECINEQ WP 2007-79.
Porzecanski, R. (2008) “Raza y Desempeño Educativo en el Uruguay
Contemporáneo: Un análisis de la brecha entre afro-descendientes y blancos”, Paper
presented in the “III Congreso de la Asociación Latinoamericana de la Población”,
Cordoba, Argentina.
Roemer, J. (1998) “Equality of Opportunity”, Cambridge MA: Harvard
University Press.
Rama, G. (2004) “La evolución de la educación secundaria en Uruguay”,
REICE-Revista Electrónica Iberoamericana sobre Calidad, Eficacia y Cambio en
Educación 2(1).
Roodman, D. (2010) “Estimating fully observed recursive mixed-process models
with cmp”, Stata Journal 11(2): 159-206-
Rosenman, R.; Mandal, B.; Tennekoon, V.; and Hill, L. (2010) “Estimating
treatment effectiveness with sample selection”. Washington State University
http://faculty.ses.wsu.edu/WorkingPapers/Rosenman/WP2010-5.pdf
SITEAL (2005) “La educación superior en América Latina: acceso, permanencia
y equidad”
http://www.siteal.iipe-oei.org/sites/default/files/educacion_superior.pdf
van Elk, R.; van der Steeg, M; and Webbink, D. (2011) “Does the timing of
tracking affect higher education completion?”, Economics of Education Review
30:1009-1021.

164
 TABLES AND FIGURES

Figure 1 The Educational System in Uruguay

School cycle Grades Theoretical ages Compulsory School type
Pre-school 0 4-5 No* Centros CAIF, Guarderías
Educación primaria 1-6 6-11 Yes Escuelas
Ciclo básico 7-9 12-14 yes Liceos/UTU
Bachillerato 10-12 15-17 No Liceos/UTU
University/ Teaching Training Institutes/
Tertiary education (vocational training)
Post secondary 18-23 No
/Militar school/ Tertiary education Non
University Institutes (private institutions)
* Since 2008 this level is compulsory for children aged 4-5 years. Ley General de Educación Nº 18.437

Table 1 Schooling progression (%) by gender and ethnicity

Males Females
Total Males Females
Non- afro Afro Non- afro Afro
Finished school 96.7 96.6 96.8 96.9 94.8 97.5 95.8
Finished CBU 64.5 63.3 65.8 65.9 47.7 69.5 46.2
Finished high school 29.2 25.4 32.8 28.8 5.1 36.4 13.7
Enrolled Tertiary level 20.5 16.0 24.8 18.2 2.7 27.2 12.0
ENAJ (2008)

Table 2 Summary statistics (%)

Variable Total Female Male No afro Afro
Female 0.52 51.7 57.4
Afro-descendant 0.11 0.12 0.09
Parents' background
Mother's education
Low level 0.48 0.48 0.47 0.46 0.65
Medium level 0.37 0.36 0.37 0.38 0.28
High level 0.15 0.15 0.15 0.16 0.07
Father's education
Low level 0.52 0.54 0.51 0.50 0.69
Medium level 0.36 0.35 0.38 0.37 0.27
High level 0.11 0.11 0.12 0.12 0.04
Attended pre-school 0.83 0.83 0.84 0.84 0.75
Completed primary level 0.98 0.98 0.97 0.98 0.95
Public school (all years) 0.77 0.77 0.77 0.76 0.87
Performance in primary (Repeated)
Never 0.76 0.78 0.73 0.78 0.59
Once 0.18 0.16 0.19 0.16 0.27
More than once 0.07 0.05 0.08 0.06 0.14
Obs. 2,349 1,228 1,121 2,100 249

165
 Table 3 Summary statistics across the schooling progression for girls
Lower highschool Upper highschool Post-secondary
Variable Enrolled Drop-out Complete Enrolled Drop-out Complete Not enrolled Enrolled

Afro 0.098 0.179 0.082 0.074 0.116 0.050 0.047 0.051

Mother's edu level
Low 0.445 0.761 0.382 0.342 0.497 0.257 0.378 0.220
Medium 0.384 0.217 0.417 0.434 0.425 0.438 0.472 0.428
High 0.171 0.022 0.201 0.225 0.079 0.304 0.150 0.352
Father's edu level
Low 0.508 0.728 0.464 0.435 0.562 0.366 0.504 0.323
Medium 0.372 0.255 0.395 0.411 0.373 0.431 0.394 0.443
High 0.121 0.016 0.142 0.155 0.065 0.203 0.102 0.235
Attended pre-school 0.853 0.685 0.886 0.894 0.849 0.918 0.890 0.927
Public school (all years) 0.749 0.967 0.706 0.681 0.798 0.618 0.748 0.577
Performance in Primary
Never repeated 0.839 0.533 0.901 0.923 0.839 0.968 0.929 0.980
Repeated once 0.136 0.370 0.090 0.074 0.151 0.032 0.071 0.020
Repeated 2+ 0.024 0.098 0.010 0.004 0.010 0.000 0.000 0.000
Noncognitive ability
Tried marijuana before 15yr 0.025 0.049 0.021 0.021 0.027 0.017 0.016 0.017
Motivation to enrollment
Highly motivated 0.778 0.571 0.819 0.835 0.784 0.862 0.819 0.875
Labor motives 0.060 0.174 0.037 0.030 0.058 0.015 0.016 0.015
Not motivated 0.139 0.201 0.126 0.118 0.140 0.106 0.134 0.098
Other motives 0.023 0.054 0.017 0.017 0.017 0.017 0.031 0.012

Lower highschool vbles

Public 0.790 0.989 0.750 0.723 0.887 0.634 0.756 0.597
Private 0.184 0.000 0.221 0.245 0.075 0.338 0.205 0.379
General education (all grades) 0.884 0.701 0.920 0.950 0.901 0.978 0.961 0.983
Vocational training (UTU all grades) 0.071 0.185 0.049 0.030 0.058 0.015 0.024 0.012
Upper highschool vbles
Public institution (all yr) 0.728 0.853 0.660 0.780 0.623
General education (all grades) 0.870 0.750 0.935 0.835 0.966
Vocational training (UTU all grades)
Performance in Secondary
Never repeated 0.647 0.336 0.817 0.661 0.866
Repeated once 0.229 0.401 0.136 0.252 0.100
Repeated 2+ 0.123 0.264 0.047 0.087 0.034
Migration motives (after highschool)
Not migrated 0.638 0.616
Other motives 0.291 0.154
Study 0.071 0.230
Obs. 1,109 184 925 828 292 536 127 409

166
 Table 4 Summary statistics across the schooling progression for boys
Lower highschool Upper highschool Post-secondary
Variable Enrolled Drop-out Complete Enrolled Drop-out Complete Not enrolled Enrolled
Afro 0.089 0.150 0.073 0.067 0.098 0.040 0.060 0.031
Mother's edu level
Low 0.432 0.740 0.355 0.320 0.404 0.248 0.410 0.176
Medium 0.400 0.250 0.437 0.453 0.459 0.449 0.436 0.454
High 0.168 0.010 0.207 0.227 0.138 0.303 0.154 0.370
Father's edu level
Low 0.475 0.745 0.407 0.374 0.474 0.288 0.504 0.191
Medium 0.395 0.250 0.431 0.445 0.428 0.459 0.444 0.466
High 0.130 0.005 0.161 0.181 0.098 0.253 0.051 0.344
Attended pre-school 0.870 0.760 0.897 0.904 0.872 0.931 0.855 0.966
Public school (all years) 0.747 0.965 0.693 0.667 0.771 0.578 0.675 0.534
Performance in Primary
Never repeated 0.789 0.520 0.856 0.890 0.810 0.958 0.897 0.985
Repeated once 0.172 0.370 0.123 0.102 0.174 0.040 0.103 0.011
Repeated 2+ 0.039 0.110 0.021 0.008 0.015 0.003 0.000 0.004
Noncognitive abilities
Tried marijuana before 15yr 0.058 0.095 0.048 0.048 0.073 0.026 0.009 0.034
Motivation to enrollment
Highly motivated 0.73 0.57 0.77 0.78 0.75 0.80 0.79 0.81
Labor motives 0.09 0.16 0.07 0.06 0.07 0.04 0.08 0.03
Not motivated 0.14 0.19 0.12 0.12 0.13 0.12 0.10 0.13
Other motives 0.05 0.09 0.04 0.04 0.05 0.03 0.03 0.03
Lower highschool vbles
Public 0.775 0.960 0.729 0.701 0.829 0.591 0.778 0.508
Private 0.188 0.010 0.232 0.263 0.131 0.377 0.197 0.458
General education (Liceo all grades) 0.772 0.500 0.840 0.875 0.801 0.939 0.846 0.981
Vocational training (UTU all grades) 0.130 0.230 0.106 0.088 0.138 0.045 0.120 0.011
Upper highschool vbles
Public institution (all yr) 0.705 0.798 0.625 0.795 0.550
General education (Liceo all grades) 0.761 0.664 0.844 0.667 0.924
Vocational training (UTU all grades)
Performance in Secondary 0.795 0.550
Never repeated 0.545 0.324 0.736 0.675 0.763
Repeated once 0.252 0.346 0.172 0.205 0.156
Repeated 2+ 0.203 0.330 0.092 0.120 0.080
Migration motives (after highschool)
Not migrated 0.624 0.603
Other motives 0.085 0.248
Study 0.291 0.149
Obs. 1,005 200 805 706 327 379 117 262

167
Table 5 Estimated correlations of unobservables and test of ignorability
Girls Boys
Correlations of Estimate p-value Estimate p-value
unobservables
𝜌12 (Complete Upper HS, -0.586 0.044 -0.314 0.485
Complete Lower HS)
𝜌13 (Completing Lower -0.347 0.459 -0.395 0.469
HS, Postsec enrollment)
𝜌23 (Completing Upper -0.174 0.474 -0.591 0.028
HS, Postsec enrollment)
𝜒2 p-value 𝜒2 p-value
Wald test of ignorability
𝐻𝑜:⁡𝜌12 = 𝜌13 = 𝜌23 = 0 12.79 0.0017 26.41 0.0000
Ho: Sample selection is ignorable.

168
 Table 6 Educational path (Girls) Average marginal effects
Lower high-school Upper high-school Post-secondary
Variables
(1) (2) (3)
Afro-descendants -0.051** (0.021) -0.094** (0.048) 0.104 (0.068)
Parental education (Ommited: low level of education)
Mother's edu level medium 0.057*** (0.017) 0.055* (0.032) -0.029 (0.035)
Mother's edu level high 0.118*** (0.034) 0.168*** (0.045) 0.039 (0.045)
Father's edu level medium 0.032** (0.016) 0.040 (0.030) 0.030 (0.033)
Father's edu level high 0.057 (0.045) 0.116** (0.048) 0.086* (0.049)
Multiple abilities
Ommited variables in repetition (Never repeated)
Repeated once school -0.103*** (0.016) -0.143*** (0.053) -0.037 (0.084)
Repeated school 2+ -0.164*** (0.036) . . . .
Repeated once secondary . . -0.251*** (0.027) -0.047 (0.057)
Repeated secondary 2+ . . -0.345*** (0.036) 0.011 (0.080)
Motives for enrollment in secondary (Omitted: highly motivated)
Not motivated -0.139*** (0.025) -0.156** (0.063) -0.013 (0.098)
Labor motives -0.042** (0.019) -0.022 (0.040) -0.027 (0.047)
Other motives -0.082** (0.040) 0.009 (0.102) -0.153 (0.100)
Marijuana before 15 -0.065 (0.100)
Stage- variant variables
Lower high school
Public institution -0.111*** (0.043) . . . .
Unemployment rate -0.126** (0.058) . . . .
All years in public school -0.083*** (0.029) . . .
Attended pre-school 0.053*** (0.017) . . . .
Upper high school
Public institution . . -0.062* (0.032) . .
General education . . 0.208*** (0.038) . .
Unemployment rate_age15 . . 0.283* (0.169) . .
Unemployment rate_age16 . . -0.146 (0.174) . .
Postsecondary education
Migration motives (Omitted variable: not migrated)
Motives for migration: studies . . . . 0.131** (0.052)
Other motives for migration . . . . -0.060* (0.036)
Employment rate_skilled . . . . 0.546 (0.393)
Regional dummies Yes all stages
Cohort age dummies Yes all stages
Obs. 1109 825 536
Standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

169
 Table 7 Educational path (Boys) Average marginal effects
Lower high-school Upper high-school Post-secondary
Variables
(1) (2) (3)
Afro-descendants -0.020 (0.026) -0.130* (0.070) -0.073 (0.094)
Parental education (Ommited: low level of education)
Mother's edu level medium 0.057*** (0.017) -0.014 (0.039) 0.010 (0.054)
Mother's edu level high 0.189*** (0.048) 0.051 (0.052) 0.084 (0.068)
Father's edu level medium 0.040** (0.018) 0.070* (0.037) 0.141*** (0.052)
Father's edu level high 0.187*** (0.065) 0.136** (0.054) 0.356*** (0.078)
Multiple abilities
Ommited variables in repetition (Never repeated)
Repeated once school -0.105*** (0.019) -0.187*** (0.062) -0.226 (0.137)
Repeated school 2+ -0.151*** (0.032) -0.115 (0.237) . .
Repeated once secondary . . -0.296*** (0.032) 0.004 (0.058)
Repeated secondary 2+ . . -0.373*** (0.036) 0.019 (0.088)
Motives for enrollment in secondary (Omitted: highly motivated)
Not motivated -0.103*** (0.025) -0.004 (0.074) . .
Labor motives -0.083*** (0.020) 0.003 (0.046) . .
Other motives -0.085*** (0.031) 0.030 (0.082) . .
Marijuana before 15 . . -0.179** (0.076) 0.198 (0.148)
Stage- variant variables
Lower high school
Public institution -0.075* (0.039) . . . .
Unskilled employment rate -0.319*** (0.121) . . . .
All years in public school -0.160*** (0.031) . . .
Attended pre-school 0.032 (0.020) . . . .
Upper high school
Public institution . . -0.031 (0.037) . .
General education . . 0.192*** (0.038) . .
Semi-skilled Employment rate_age15 . . -0.527** (0.257) . .
Semi-skilled Employment rate_age16 . . 0.215 (0.261) . .
Postsecondary education
Migration motives (Omitted variable: not migrated)
Motives for migration: studies . . . . 0.199*** (0.065)
Other motives for migration . . . . -0.078 (0.049)
Unemployment rate (postsec) . . . . 0.566* (0.307)
Employment rate_skilled . . . . -0.272 (0.477)
Regional dummies Yes all stages
Cohort age dummies Yes all stages
Obs. 994 706 378
Standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

170
APPENDIX

Table A.1 Main descriptives of the Uruguayan Educational System

Variables Total Male Female
Preschool
Never attended 0.12 0.12 0.12
Public 0.54 0.55 0.54
Private 0.34 0.33 0.34
Type of school (all years)
Public 0.78 0.77 0.78
Private 0.17 0.18 0.16
Mixed 0.05 0.04 0.06
Lower highschool
Type of institution (attended al grades in)
Public 0.78 0.77 0.79
Private 0.03 0.04 0.03
Mixed 0.18 0.19 0.18
Track
General Education 0.86 0.81 0.90
Training institute 0.09 0.12 0.06
Mixed 0.06 0.08 0.04
Upper highschool
Type of institution (attended al grades in)
Public 0.72 0.70 0.74
Private 0.16 0.17 0.15
Mixed 0.12 0.13 0.11
Track
General Education 0.82 0.76 0.88
Training institute 0.06 0.10 0.03
Mixed 0.11 0.14 0.09
Postsecondary
Type of institution (last level attained in)
Public 79.5 70.78 85.01
Private 20.5 29.22 14.99
ENAJ (2008), ECH (2008)

171
 Table A.2 Simple probit Girls
Variables Lower high-school Upper high-school Post-secondary
Afro-descendants -0.069*** (0.027) -0.104** (0.048) 0.098 (0.083)
Parental education (Ommited: low level of education)
Mother's edu level medium 0.074*** (0.021) 0.065** (0.031) 0.006 (0.041)
Mother's edu level high 0.159*** (0.045) 0.180*** (0.044) 0.118** (0.050)
Father's edu level medium 0.043** (0.021) 0.047 (0.030) 0.058 (0.038)
Father's edu level high 0.066 (0.057) 0.127*** (0.049) 0.143*** (0.053)
Multiple abilities
Ommited variable in repetition (Never repeated)
Repeated once school -0.139*** (0.020) -0.181*** (0.051) -0.173* (0.095)
Repeated school 2+ -0.214*** (0.046) . . . .
Repeated once secondary -0.256*** (0.027) -0.156*** (0.045)
Repeated secondary 2+ -0.342*** (0.037) -0.144** (0.072)
Motives for enrollment in secondary (Ommited: highly motivated)
Not motivated -0.180*** (0.032) -0.165** (0.069) -0.055 (0.111)
Labor motives -0.058** (0.025) -0.026 (0.041) -0.043 (0.052)
Other motives -0.113** (0.052) 0.020 (0.102) -0.189* (0.108)
Marijuana before 15 -0.079 (0.098)
Stage- variant variables
Lower high school
Public institution -0.144*** (0.055)
Unemployment rate -0.164** (0.076)
All years in public school -0.102*** (0.036)
Attended pre-school 0.071*** (0.023)
Upper high school
Public institution -0.066** (0.032)
General education 0.177*** (0.038)
Unemployment rate_age15 0.149 (0.154)
Unemployment rate_age16 -0.019 (0.152)
Postsecondary education
Migration motives (Ommited: not migrated)
Motives for migration: studies 0.173*** (0.053)
Employment rate_skilled 0.661 (0.464)
Regional dummies Yes all stages
Cohort age dummies Yes all stages
Standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

172
 Table A.3 Simple probit Boys
Variables Lower high-school Upper high-school Post-secondary
Afro-descendants -0.030 (0.034) -0.140** (0.071) -0.072 (0.087)
Parental education (Ommited: low level of education)
Mother's edu level medium 0.078*** (0.023) 0.011 (0.038) 0.036 (0.050)
Mother's edu level high 0.265*** (0.066) 0.087* (0.049) 0.147** (0.060)
Father's edu level medium 0.054** (0.024) 0.083** (0.036) 0.156*** (0.045)
Father's edu level high 0.287*** (0.093) 0.156*** (0.052) 0.390*** (0.062)
Multiple abilities
Ommited variable in repetition (Never repeated)
Repeated once school -0.146*** (0.025) -0.237*** (0.054) -0.281** (0.113)
Repeated school 2+ -0.207*** (0.042) -0.225 (0.251) . .
Repeated once secondary -0.294*** (0.032) -0.047 (0.052)
Repeated secondary 2+ -0.378*** (0.035) -0.057 (0.075)
Motives for enrollment in secondary (Ommited: highly motivated)
Not motivated -0.121*** (0.034) 0.001 (0.065) -0.216*** (0.084)
Labor motives -0.101*** (0.028) -0.016 (0.044) -0.026 (0.065)
Other motives -0.116*** (0.042) 0.014 (0.081) 0.022 (0.094)
Marijuana before 15 -0.176** (0.079) 0.234* (0.133)
Stage- variant variables
Lower high school
Public institution -0.069 (0.043)
Unskilled employment rate -0.416** (0.167)
All years in public school -0.210*** (0.040)
Attended pre-school 0.034 (0.027)
Upper high school
Public institution -0.039 (0.036)
General education 0.182*** (0.036)
Semi-skilled Employment rate_age15 -0.544** (0.263)
Semi-skilled Employment rate_age16 0.194 (0.266)
Postsecondary education
Migration motives (Ommited: not migrated)
Motives for migration: studies 0.206*** (0.061)
Unemployment rate (postsec) -0.264 (0.486)
Employment rate_skilled 0.526* (0.310)
Regional dummies Yes all stages
Cohort age dummies Yes all stages
Standard errors in parentheses
* p<0.1, ** p<0.05, *** p<0.01

173
 Table A.4 Definition of independent variables
Variables Description Type of variable

Afro-descendant 1 if respondant declares afro descendance; 0 otherwise Dummy

Age fixed effects Age in years Categorical
Mother's highest level attained Classification according to years of education Categorical
Low Less than 9 years
Medium Between 9 to 12 years
High More than 12 years
Father's highest level attained Classification according to years of education Categorical
Low Less than 9 years
Medium Between 9 to 12 years
High More than 12 years
Cognitive ability
Performance in Primary (Repeated) Categorical
Never If the individual declares never repeated
Once if the individual declares repeated once
More than once if the individual declares repeated more than once
Performance in Secondary (Repeated) Categorical
Never If the individual declares never repeated
Once if the individual declares repeated once
More than once if the individual declares repeated more than once

174
 Table A.4 Definition of independent variables (cont.)
Variables Description Type of variable
Noncognitive ability
Motivation for secondary enrollment If the individual declares as main reason for enrollment one of the alternatives: Categorical
Highly motivated Acquisition of education
Today is essential to study
You are interested on what you are studying
Expect to improve social status through education
Labor motives If the individual declares as main reason for enrollment one of the alternatives:
In order to quickly find a job
Studies while finding a job or start a family
Not motivated If the individual declares as main reason for enrollment one of the alternatives:
Oblished to
Other motives If the individual declares as main reason for enrollment one of the alternatives:
Receive subsidies
to meet other youths
others
Tried marijuana before 15 Equal to one if the individual declares trying marijuana before age 15; 0 otherwise Dummy

175
 Table A.4 Definition of independent variables (cont.)
Variables Description Type of variable
Institutional variables
Equal to one if the individual declares attending all grades of primary level in a public
Public school (all years)
school; 0 otherwise Dummy
Attended pre-school Equal to one if the individual declares having attended pre-school; 0 otherwise Dummy
Equal to one if the individual declareshaving attended all grades of upper highschool in a
Public in lower highschool public institution; 0 otherwise Dummy
Equal to one if the individual declares attending all grades of lower highschool in a
Public in upper highschool
public institution; 0 otherwise Dummy
Equal to one if the individual declares having attended all grades of upper highschool in
Vocational education
a General academic institution; 0 otherwise Dummy
Labor market variables
Unemployment rate of population aged less than 25 by gender, department of residence
Youth unemployment rate
and different schooling stages* Numerical
Employment rates calculated at the department of residence level and different
Employment rates
schooling stages*
Unskilled employment rate Employment rate for workers with less than 9 years of education Numerical
Semi-skilled employment rate Employment rate for workers with 9 to 12 years of education
Skilled employment rate Employment rate for workers with more than 12 years of education

If the individual declares as main motives for migration (after completing upper high
Migration motives
school) Categorical
Study
Other (includes labor, health, family, and other motives
Never moved
*For example one girl living in Montevideo deciding whether or not to completing upper high school, the unemployment rates used in the model are 3 Female youth unemployment
rates in Montevideo, one for each year when the girl was aged 15, 16 and 17; theoretical ages in which girl is supposed to be in upper high school. Similar strategy was used to the
calculation of employment rates.

176
 Table A.5 Independent variables
Observed personal characteristics Race
Parental education level (mother and father) Low (less than 9 yr)
Medium (9 to 12 yr)
High (More than 12 yr)
Institutional Public school (all years)
Attended pre-school
Performance in primary (Repeated)
Never
Once
More than once
Cognitive ability
Performance in Secondary (Repeated)
Never
Once
More than once
Marijuana before age 15*
Motivation to enrollment in secondary level
Highly motivated
Non-cognitive ability
Not motivated
Labor motives
Other motives
*Tried marijuana before 15 is only included in upper high school in order to avoid endogeneity
issues in lower high school.

177
 Table A.6 Independent variables. Stage-variant regressors
Lower highschool Upper highschool Post-secondary enrollment
Region of residence (departament) Region of residence (departament) Motives for migration (at
theoretical age of attendance)
Never migrated
Study motives
Other motives (family, labor, health,
others)
Performance in secondary level Performance in secondary level
(Repeated) (Repeated)
Never Never
Once Once
More than once More than once
Labor opportunities
Unemployment youth rate (by gender, Unemployment youth rate (by gender, Skilled Employment rate (by gender,
region and for theoretical ages of region and for theoretical ages of region and for theoretical ages of
attendance) attendance) attendance)
Unskilled Employment rate (by gender, Semi-skilled Employment rate (by gender,
region and for theoretical ages of region and for theoretical ages of
attendance) attendance)
Institution type (all years in public Institution type (all years in public
institution) institution)
Unemployment youth rate (by gender, Unemployment youth rate (by gender,
region and for theoretical ages of region and for theoretical ages of
attendance) attendance)
Unskilled Employment rate (by gender, Semi-skilled Employment rate (by gender,
region and for theoretical ages of region and for theoretical ages of
attendance) attendance)
Vocational education (all yr General educ.)

178
179
180
4. Conclusions

The main aim of this thesis has been to contribute to the literature on economic development
by providing empirical evidence on three channels suggested by the literature that may cause
individuals and countries to be entrapped in poverty.

The first essay of this thesis studied the relationship between immigrants’ social networks
and their subsequent labor market outcomes in Spain for 1997-2007. For this purpose, I used
the National Immigrant Survey carried on 2007 and conducted two empirical exercises. First,
I analyzed the extent to which social networks affect immigrants’ job match. Second, for
immigrants keeping the first job in Spain, I studied to what extent social networks influence
wages. The econometric technique followed a two-step type procedure similar to the one
proposed by Heckman (1979) to control for endogeneity issues.
The main results of this essay showed a great reliance on immigrants’ social networks
for employment in the host country. Job mismatch is more likely to occur for those
immigrants that upon arrival prefer to quickly being employed in a job provided by the
network, even if it is not the most suitable one in terms of the immigrants’ human capital and
previous experience. In addition, the results confirmed a positive effect of the network size on
the probability of job matching. For those keeping the first job, network size is found to
penalize immigrants’ wages. Also, despite we found differences across the wage distribution
and gender, the strength of the network is found to penalize immigrants’ wages.
These results may be reflecting that social capital accumulated by the network is
restricted to a particular segment of the labor market and thus, limiting immigrant’s job
prospects to the network, and also depressing wages for those immigrants in segmented
occupations or sectors of activity. From this analysis we suggested that policy interventions
aiming to socially and economically integrate immigrants in Spain, should be focused on
influencing immigrant’s environment by for instance, promoting greater access to formal
institutions in the labor market and reducing immigrant’s dependence on the information
transmitted by the network.

The aim of the second essay of this dissertation was to test the predictions of Banerjee and
Newman’s model, which s, suggests that development paths are determined by countries’
initial conditions, notably wealth distribution and credit market institutions.

181
 This model predicts that countries with high historical rate of credit to non-credit
constrained people end up in a situation in which only a small share of the population might
start-up new firms, but these firms do not grow over time. In this case, the process of
development ends up in a situation of low wages, in which there is (almost) self-employment
at small scale. Conversely, countries with a low proportion of credit constrained people will
grow over time aided by a high share of people being able to start-up business, of these
surviving over time and with an active labor market paying high salaries.
To empirically test these hypotheses, we built a pseudo-panel using data from the
Global Entrepreneurship Monitor (GEM) for the period 2001-2009. The pseudo-panel was
complemented with income distribution indicators prevailing in 1700s and 1800s, and credit
protection indicators.
In order to address reverse causality between the proportion of people involved in
entrepreneurship and current business regulation, the econometric technique used
instrumental variable estimators.
The main findings of this essay support the predictions of Banerjee and Newman’s
(1993) model. We found negative and persistent effects of inequality prevailing in 1800s’ on
the likelihood of countries’ developing a healthy entrepreneurial sector, understood as firms
being created, surviving and creating jobs over time. Also, the more efficient credit markets
proxied by the legal right index are, the more likely is that countries’ have larger proportion
of people involved in entrepreneurial activities, and to these developing firms over time. In
this essay we proposed that to foster entrepreneurship to grow and create jobs over time,
countries should focus on reducing their inequality levels and improve credit market
institutions.

The third essay analyzed whether long-term parental background, crystallized by parental
educational background, race, cognitive and non-cognitive abilities, and short-term family
income measured by the non-monetary opportunity cost of education, affect child’ schooling
progression, and at what stage of the educational path they take on their importance.
To this end, I used a sequential probability model, in which education attainment is the
outcome of the individual’s previous schooling decisions. This methodology allowed me to
control for potential endogeneity issues arising from individual’s unobservable heterogeneity
and non-random selection of the sample that may occur at different educational stages.
I used the National Youth Survey and National Household Surveys conducted in 2008
from which I constructed individuals’ educational path trajectories.

182
 The main findings of this essay showed that the Uruguayan educational system is
highly stratified, only allowing those individuals with better parental educational background,
more able and motivated individuals, and non afro-descendants to attain higher educational
levels.
Short-term parental income and long-term parental factors both influence children’s
schooling progression in Uruguay although they have different impact across the educational
path. Specifically, short-term family income decreases its importance as students progress to
higher schooling stages, whereas long-term parental factors turn to be more important the
higher we move on the educational system. In particular, persistent and increasing effects of
cognitive abilities on schooling progression are found. Socio-emotional factors, proxied by
motivation in secondary level and risky behavior also influence children’ schooling
progression.
This essay supports policy interventions at different schooling stages. Policies
intended to promote cognitive ability early in life and social and behavioral skills in
adolescence and youth from a gender perspective and taking into account ethnical/ racial
diversity may have positive effects on child’s education achievement.

The thesis has overall provided evidence that initial conditions, whether immigrant’s
networks, country’s initial wealth distribution or children’s family background, affect
development in the short and long-run. The findings shown here thus contribute to the
literature and suggest important policy interventions.

183