Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
ESSAYS ON ECONOMIC DEVELOPMENT
BY
FEDERICO DROLLER
B.A., UNIVERSIDAD TORCUATO DI TELLA, 2002
M.A., BROWN UNIVERSITY, 2008
A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF THE
REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
IN THE DEPARTMENT OF ECONOMICS AT BROWN UNIVERSITY
PROVIDENCE, RHODE ISLAND
MAY 2013
c© Copyright 2013 by Federico Droller
This dissertation by Federico Droller is accepted in its present form
by the Department of Economics as satisfying the
dissertation requirement for the degree of Doctor of Philosophy.
Date
David Weil, Adviser
Recommended to the Graduate Council
Date
Pedro Dal Bo, Reader
Date
Ross Levine, Reader
Approved by the Graduate Council
Date
Peter Weber, Dean of the Graduate School
iii
Vita
Federico Droller was born on August 16, 1979 in Buenos Aires, Argentina. He earned
his Bachelor’s degree from Universidad Torcuato Di Tella in 2002. He was awarded
Highest Honors for his undergraduate thesis in Economics. He started his graduated
studies in Argentina and completed all the course work for the M.A. in Economics.
He enrolled in Brown University’s Economics Ph.D. program in 2007 and obtained
his M.A. in Economics in 2008. In the course of the program he was awarded a Craig
M. Cogut Dissertation Fellowship and a Merit Dissertation Fellowships. He received
a Ph.D. in 2013 and will continue his research in Economics as an Assistant Professor
at Universidad de Santiago de Chile.
iv
Acknowledgements
I am deeply indebted to my advisors David Weil, Pedro Dal Bo and Ross Levine for
their guidance, advice and support throughout the years of work on this dissertation.
I specially thank Pedro Dal Bo for his dedication and support during the process of
working on a new idea and write a paper. I am also grateful to other faculty members
in the Department of Economics, Oded Galor whose questions and comments helped
me to improve the scope of this project, Blaise Melly and Vernon Henderson whose
advice was crucial for the execution of my work. This dissertation would not have
been possible without the help provided by the Brown University Library. I big
thank goes to my friends at Brown who contributed to my well being throughout the
PhD program, also to Angelica Vargas who provided an enormous help with all the
administrative issues.
I would never have made it into and through graduate school without the love and
support from Flor. Words are not enough to thank her for all the years we spent
together and for all the projects we pursued together.
v
Contents
List of Figures viii
List of Tables x
1 Migration and Long-run Economic Development: Evidence fromSettlements in the Pampas 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The History of the Fertile Plains . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 The Conquest of the Plains: the Desert . . . . . . . . . . . . . 6
1.2.2 Settlement of the Fertile Plains . . . . . . . . . . . . . . . . . 7
1.3 Data and Summary Statistics . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Estimation Strategy and Results . . . . . . . . . . . . . . . . . . . . . 12
1.4.1 Instrumental Variable Approach . . . . . . . . . . . . . . . . . 14
1.4.2 The long-run effect of European immigration . . . . . . . . . . 20
1.4.3 The effect of European immigration: the channels of persistence 22
1.4.4 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . 28
1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2 Beliefs in Market Economy and Macroeconomic Crises while Young 51
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
vi
2.2 Data Description & Methodology . . . . . . . . . . . . . . . . . . . . 56
2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.4 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3 Population Composition and Human Capital Creation: the Raise inEducation in the U.S. 95
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.3.1 Individual Level Data . . . . . . . . . . . . . . . . . . . . . . . 99
3.3.2 County Level Data . . . . . . . . . . . . . . . . . . . . . . . . 102
3.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Bibliography 119
vii
List of Figures
1.1 Correlation between current log per-capita GDP and the share of Eu-ropean population in 2000. . . . . . . . . . . . . . . . . . . . . . . . . 31
1.2 Correlation between log per-capita GDP in 1994 and the share of Eu-ropean population in 1914, in Argentina. . . . . . . . . . . . . . . . . 32
1.3 Advancement of the frontier, 1810 - 1828. . . . . . . . . . . . . . . . . 33
1.4 Advancement of the frontier, 1852 - 1876. . . . . . . . . . . . . . . . . 34
1.5 Immigration Time Series, 1857 - 1914. . . . . . . . . . . . . . . . . . 35
1.6 Cumulative Net-Immigration and Area for settlement, 1857 - 1914. . 36
1.7 1st Stage correlation between the share of European population andthe constructed share of European immigration. . . . . . . . . . . . . 37
1.8 1st Stage correlation between the share of European population andthe constructed share of European immigration, control variables andfixed effects included. . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1 Mean Beliefs by Year of Birth and Country. . . . . . . . . . . . . . . 70
2.2 Economic Crises in the Age Period 22-25, by Year of Birth and Country. 71
2.3 Economic Crises in the Age Period 18-21, by Year of Birth and Country. 72
2.4 Economic Crises in the Age Period 26-29, by Year of Birth and Country. 73
2.5 Economic Crises in the Age Period 30-33, by Year of Birth and Country. 74
2.6 Economic Crises in the Age Period 34-37, by Year of Birth and Country. 75
2.7 Economic Crises in the Age Period 38-41, by Year of Birth and Country. 76
viii
2.8 Economic Crises by Age Periods and Year of Birth for the Whole Sample. 77
3.1 Immigration Time Series, 1820 - 1920. . . . . . . . . . . . . . . . . . 107
ix
List of Tables
1.1 Summary Statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.2 OLS Results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.3 First Stage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.4 IV Results, log per-capita GDP, 1994. . . . . . . . . . . . . . . . . . . 42
1.5 IV Results, share of population with higher education, 2001. . . . . . 43
1.6 IV Results, share of population with high skilled occupations, 2001. . 44
1.7 Ownership and Industrial Workers. . . . . . . . . . . . . . . . . . . . 45
1.8 IV Results, early Industrial Indicators. . . . . . . . . . . . . . . . . . 46
1.9 Literacy Rates by Contry of Birth. . . . . . . . . . . . . . . . . . . . 47
1.10 IV Results, Literacy Rates and Number of Schools, 1914. . . . . . . . 48
1.11 Robustness Checks I. . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
1.12 Robustness Checks II. . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.1 Summary Statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.2 Beliefs and Economic Crisis at Different Age Periods. . . . . . . . . . 79
2.3 Effect of Economic Crisis on Beliefs with Country Fixed Effects. . . . 80
2.4 Effect of Economic Crisis on Beliefs, Adding Controls. . . . . . . . . . 81
2.5 Effect of Economic Crisis on Beliefs, Adding Controls, Cont.. . . . . . 82
2.6 Testing the Impressionable Years Hypothesis. . . . . . . . . . . . . . 83
x
2.7 Addressing Cohort Effects, Sample of Oldest individuals. . . . . . . . 84
2.8 Addressing Cohort Effects, Sample of Oldest individuals, Cont.. . . . 85
2.9 Addressing Cohort Effects, Adding Cohort Dummies. . . . . . . . . . 86
2.10 Linear Probability Model. . . . . . . . . . . . . . . . . . . . . . . . . 87
2.11 Logistic Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.12 Appendix: Summary Statistics by Country. . . . . . . . . . . . . . . . 89
2.13 Appendix: Summary Statistics by Country, Cont.. . . . . . . . . . . . 90
2.14 Appendix: Summary Statistics by Country and Age. . . . . . . . . . 91
2.15 Appendix: Summary Statistics by Country and Age, Cont.. . . . . . . 92
2.16 Appendix: Cross Country correlation, Real GDP Growth Rate. . . . 93
2.17 Appendix: Effect of Economic Crisis with Controls and Fixed Effects. 94
3.1 Share of Immigrants by Country of Birth. . . . . . . . . . . . . . . . 108
3.2 Summary Statistics. . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
3.3 Probit and OLS Results. . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.4 Results for Individuals aged 14 to 17. . . . . . . . . . . . . . . . . . . 111
3.5 Results for Individuals aged 7 to 17. . . . . . . . . . . . . . . . . . . . 112
3.6 Results for Individuals aged 7 to 17, with dummy young. . . . . . . 113
3.7 Results for Individuals aged 14 to 17, Goldin - Sample. . . . . . . . . 114
3.8 Results for Individuals aged 14 to 17, County-level. . . . . . . . . . . 115
3.9 Results for Individuals aged 14 to 17, including Fractionalization, byyear. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.10 Results for Individuals aged 14 to 17, including Fractionalization, allyear. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.11 Results for Individuals aged 14 to 17, including Fractionalization, GoldinSample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
xi
Chapter 1
Migration and Long-run Economic Devel-opment: Evidence from Settlements in thePampas
1.1 Introduction
Understanding the fundamental causes of the large differences in income per-capita
across countries has led economists to examine the effect of historical events on eco-
nomic development. Of particular importance is the process of settlement and popu-
lation that countries followed during and after the colonial period. Places with more
European settlements in the past tend to outperform in the present in various mea-
sures of development (Easterly and Levine 2009), and even today there is a positive
correlation between current per-capita GDP and places were Europeans live (see figure
1). Different theories have been proposed to understand how historical events per-
sisted and shaped current economic conditions resulting in a growing literature.1 One
of the first ones to formalize the importance of history were Engerman and Sokoloff, in
their research program (Engerman and Sokoloff 1997, 2002) they analyzed the effect
1See Nunn (2009) for a review of the literature.1
2of initial endowments on its distribution, inequality, political power and the resulting
institutions that were in place. By comparing colonies in north and south America,
they show that ares with a higher native population and/or potential for valuable ex-
ports generated unequal land holdings and concentrated political power on the elites.
These differences created, in turn, rent-seeking institutions that where less conducive
to economic growth in the long-run. A similar argument was propose by Acemoglu
et al. (2001, 2002), they follow this line of research by focusing on the importance of
colonial institutions for economic development. Another view was pushed forward by
Sachs, he emphasizes that levels of development (per-capita income, economic growth,
and other economic and demographic dimensions) are strongly correlated with geo-
graphical and ecological variables. In his view geographic and climate endowments
(such as latitude, disease ecology or distance from the coast) have a direct effect on
development (Gallup, Sachs, and Mellinger 1998 and 2000, Gallup and Sachs 2001
and Sachs and Malaney 2002). Glaeser et al. (2004) highlighted a different aspect
of population: knowledge and know-how, or human capital in a broader sense. They
argued that human capital was brought by European settlers, and these past differ-
ences in human capital across societies/countries explain a greater part of current
differences in economic growth, a point also stressed by Easterly and Levine (2009).
In the process of settlement and population immigration played an important role,
the short and medium run effects of immigration have been extensively analyzed,
with seminal studies like Borjas (1994) and Card (1990). But how population com-
position can affect a country’s economic performance in the long-run remains an open
question. Putterman and Weil (2010) recognized the importance of historic migra-
tion and how it altered population composition. They construct a matrix that links
current population to population in source countries, and show how adjusting by the
history of population’s ancestors improves the prediction of current GDP by historic
indicators of development across countries.
3
The complexity in understanding the impact of population composition lies in its non
random nature. Individuals that end up living in a certain place may have decided to
migrate, in the first place, and their destination. Therefore empirically assessing the
effect of population composition and disentangling it from other confounding factors
is a challenge for most studies. In this paper exploit the history of the settlement in
the Pampas, in Argentina, to identify the causal effect of historical population com-
position on long-run development. In Argentina the process of settlement was greatly
influenced by the arrival of European immigrants, areas were exposed differently to
European immigration that resulted in a great variation in the composition of pop-
ulation across counties. The characterization of European settlements in Argentina
resembles that of the world: areas differ in the intensity of European population. In
figure 2 I replicate figure 1 for this time for counties in Argentina. The positive correla-
tion between the share of Europeans and per-capita GDP is also present in this figure,
counties in Argentina look similar to countries in the world. The case of Argentina
offers the possibility to understand the long-run effect of European immigration in a
contained setting: focusing on a single country, with common macro-institutions and
similar geographic endowments. The nature of the European immigration process
in Argentina makes it relevant to the understanding of the long-run effects of the
composition of population on development.
I will first establish the causal effect of population composition in the late nineteenth
century on current GDP, education and skilled labor. I measure population compo-
sition as the share of European born immigrants. I show that there is a strong and
positive effect of the share of immigration on these variables. Then I propose two
channels through which the effect persisted over time. To overcome the problem of en-
dogenous sorting of migrants I use an exogenous measure of the share of immigration
in a given region as an IV. The IV is constructed from a simple model of settlement
and demographic growth. The IV exploits variation over time in the incorporation of
4
land to the country interacted with variation in the net-immigration of Europeans.
This empirical setting benefits from two important conditions: First, by focusing on
a single country macro-political-institutions are common across regions. Second, the
uniform geographical characteristics of the fertile plains allows me to compare a cross
section of counties that are close to equal in all geographical endowments. Holding
institutions and geography constant across counties allows me to focus on the compo-
sition of population, in particular given the heterogeneous population characteristics
that arise from the process of migration. Moreover, once institutions and geographic
endowments are accounted for in the analyzes, any effect coming from the population
composition can only be attributed to the population characteristics.
The empirical analysis exploits a particular historical setting in the fertile plains
of Argentina, the Pampas, an area originally occupied by native population, over
which the Argentine government struggled to gain power. The availability of the
fertile plains to those willing to settle varied over time depending on the civil and
international conflicts and on the success of military campaign to conquer the plains.2
European migration to Argentina was restrictive over the colonial period and only
started years after independence, with peaks by the end of the nineteenth century and
before the First World War. Between 1857 and 1914 close to 5.5 million Europeans
migrated to Argentina.3 The fertile plains, otherwise an area with geographically
similar characteristics and common political institutions, were shocked in varying
intensity by European immigrants. The shock to the population was not negligible,
areas ranged in the intensity of treatment, the percent of European population after
the shock, from 0% to 30%.
2The process of settling the Pampas drastically contrasted to what happened in the US, while inArgentina settlers arrived after the government conquered the land, in the US colonizers precededthe military.
3The Argentine government started recording statistics for immigration in 1857 and in 1914 thegovernment conducted a census.
5
Using this predicted measure of the share of European population as an instrumental
variable for the actual share of European population, I compare counties in the fertile
plains and estimate that an increase of 11% (one standard-deviation) in the share of
European population raises per-capita GDP by 60% in the long-run (0.77 standard
deviations). Similar results hold for education: areas with higher share of European
immigration in 1914 have a higher share of population with higher education in 2001.
After establishing the long-run effects of immigration on development, I investigate
two channels through which initial differences in the composition of population per-
sisted over time: industrialization and human capital.
Census data shows that industrial establishments were owned predominantly by Eu-
ropeans. Consistent with this fact I find that measures of industrial development such
as the number of industrial establishment, the employment of high- and low- skilled
industrial workers and the usage of energy for industry, where substantially higher in
regions where the intensity of immigration was higher. This suggest that industrial-
ization was a path through which differences in development arose and persisted over
time. Regarding human capital, I show that areas where Europeans accounted for a
higher share of the population had higher literacy rates in the past. The evidence sug-
gests that immigrants not only contributed with their higher literacy, but generated
a positive externality on the population, raising early levels of human capital.
The results I present in this paper show the importance of people themselves for
economic development. The setting I exploit allows me to abstract from the classical
institutional view, as well as from the geographic endowment hypothesis. These
results demonstrates that people matter, and that they matter for reasons related to
their knowledge: European immigrants are associated with greater industrialization
and higher literacy for the population at large, and that the initial difference in the
composition of the population has a long-lasting effect on development.
6
This paper is organized as follows, Section 2 reviews the conquest of the fertile plains
and the European immigration to Argentina. I provide an historical account of the
reasons that motivated military campaigns to the Pampas and timing of these cam-
paigns. Further, I describe the process by which the plains were settled and how
migration to Argentina resembles the migration pattern to the USA. Section 3 de-
scribes the data, its sources, the unit of observation and how geo-referenced data was
computed for this study. Section 4 develops the empirical strategy and shows the re-
sults. In the beginning of section 4 I show OLS estimates and in section 4.1 I proceed
to develop the instrumental variable approach. In section 4.2 I implement my IV and
show the causal effect of migrants on long-run development. Next in section 4.3 I
show two channels of persistence: industrialization and human capital. In section 4.4
I perform a series of robustness checks: I consider variations to the parameters of the
demographic model. Section 5 concludes.
1.2 The History of the Fertile Plains
1.2.1 The Conquest of the Plains: the Desert
It was not until end of the nineteenth century that the Argentinean government
gained political power over the whole territory that nowadays is Argentina. During
colonial times and after independence from the Spanish Empire in 1816 most of
the fertile plains where settled by several indigenous tribes that did not recognize
the Argentinean government. Relationships between Argentineans and indigenous
tribes were characterized by mistrust and violence. By the time of independence the
situation was such that Argentineans used to dispute land and wild livestock to the
indigenous tribes, while indigenous people organized assaults into settlements and
7
cities, stealing livestock, goods and kidnapping people. Indigenous raids attacking
cities and military excursions into indigenous settlements, both ending in destruction
and deaths, were common. The Argentinean government and main tribes often agreed
on peace treaties, but the Argentinean government never recognized that area as an
independent state, nor did it recognize indigenous people as legal owners of the land.
The threat of indigenous tribes over Argentinean settlements was not the only concern
of the government regarding the national territory. For Argentina to consolidate as
a nation it was necessary to delimit its frontiers, which turned necessary to occupy
Patagonia, an area also claimed by neighboring country Chile (Lacoste 2002). But
it was not until the end of the civil war in 1862 that a unified national government
developed systematic plans to conquer the rest of the territory, starting in 1870 until
1885.
Previous to 1870, military campaigns developed with many years of interruption
and loss of domain, in particular during episodes of civil war and the war against
Paraguay. Detailed information on the military campaigns and its effect on how the
frontier between Argentineans and the indigenous tribes changed over time has been
documented by Walther (1964). Figures 3-4 depict maps showing the frontier between
Argentina and the indigenous tribes in 1779, 1823, 1826, 1828, 1852, 1860, 1864 and
1876. Gains of territory by the Argentinean army and losses of domain over these
years were a consequence of the limited resources the government had for the multiple
military conflicts it faced (Luna 1993).
1.2.2 Settlement of the Fertile Plains
The end of the civil war and the re-unification and pacification of the country started
a period of European migration to Argentina in the second half of the nineteenth
8
century. Immigrants were granted the same legal rights as Argentineans, without need
to naturalize or acquire citizenship. The flow of immigrants to Argentina resembles
the flow of immigrants to the USA, Canada and Australia.
Figure 5 shows the time series of immigration and net immigration of Europeans to
Argentina. The series starts in 1857 when the national government started recording
statistics on the arrival of immigrants to its ports. The flow of migration is far from
constant, nor it is a monotonic function of time.
Immigrants settled in cities, urban areas and in the countryside, and were occupied
both as skilled labor or unskilled labor. Activities were diverse, ranging from farmers
to construction workers, merchants and craftsmen. As of 1895, 41 percent of the
European immigrants (males, aged 15 or above) were living in urban areas, while 32
percent devoted their time to farming and 28 percent to non-farm skilled labor.
The ultimate conquest of the Pampas was possible between 1870 and 1895, once mil-
itary resources were not longer used in civil or international wars. At the same time,
the peace achieved in the country and the economic conditions in Europe motivated
Europeans to migrate to Argentina. Between independence and the reunification of
the country, a period close to fifty years, civil war prevented many Europeans of mi-
grating to Argentina45. Although the decision to conquer the plains was unrelated to
the immigration patterns, the timing of the expansion of the frontier over the plains
overlaps with the arrival of the first European immigrants to the country, as shown
in Figure 6. Concerns might be raised on Europeans migrating to Argentina because
of the growing availability of land. The data doesn’t point to this conclusion, the
correlation between the time series of immigration and the amount of land in the
fertile plains under the political power of the government over time is close to 0.5,
4In contrast to the US, which experienced large migration from northern Europe over this period.5.
9
and a regression of immigration on the amount of land yields and R2 of 20%. Eu-
ropeans were attracted by a peaceful place to live, prospects of a work and the legal
protection of its rights. Temporary and permanent workers migrated mostly to the
fertile plains, some of then coming back to Europe after the harvest in the southern
hemisphere (right before the harvest in the northern hemisphere) and some of them
settling down and bringing the rest of their families over time. Progress and well
being among immigrants was not instant, but not hard to achieve.
1.3 Data and Summary Statistics
This study combines current data on economic development (per-capita GDP, higher
education rate and share of skilled workers) with historical data on economic and
social conditions (population density, productive uses of land, etc.). The unit of
observation is at the county level. The sample covers the four provinces that hold
the fertile plains: Buenos Aires, Santa Fe, Cordoba and Entre Rıos. The southwest
section of the fertile plains lays in the state of La Pampa, which is not included
in the sample. It was not until 1952 that La Pampa became a province, before
that it was a national territory, i.e. a territory ruled by the national government,
with appointed officials and no state constitution. Statistical information is not as
exhaustive for national territories as it is for states. Moreover, the state of La Pampa
changed all the county boundaries over the period of time considered in this study.
Working with four states allows me to control for unobservable fixed variables at the
state level. Though county boundaries have slightly changed over time, it is still
possible to match older counties to new counties. New counties were mostly founded
on previously unoccupied land, but there were cases where old counties split into
two or more counties. When a new county can not be linked to an old county, the
10
observation is dropped from the sample. There are 197 counties in the sample, where
31 are new counties not linked to an old county. From the remaining 166 counties, 25
are capital cities or large urban areas and 5 are counties without current information
on economic outcomes. Excluding capital cities and the urbanized greater Buenos
Aires, the final sample has 136 counties in four states.
Historical information comes from four sources: the 1895 and 1914 Argentinean cen-
suses, the Argentine Office of Migration and Walther (1964). Both censuses contain
detailed information at the county level on population characteristics and economic
activities. I digitalized data on all variables used from the censuses: total population,
foreign born population and population living in urban areas. Moreover, the 1914
census includes an agricultural and livestock census, which was used to construct a
variable on the economic activities performed at the county level. Somoza and Lattes
(1967) computerized representative samples of historical 1895 census microdata, from
which individual level data on nationality, age, sex and occupation can be obtained.
The Argentine Office of Migration records since 1857 all non-Argentine incoming and
outgoing population. Detailed data on the number of migrants and country of origin
since 1857 until 1914 was digitalized for this study.
Data on the territory under the political power of the Argentine government comes
from Walther (1964). Walther’s detailed description of the military campaigns are
summarized with a series of maps that show for different years the actual frontier
between the territory under the Argentinean government and the native tribes’ ter-
ritory. Walther’s work is based on military and historical documents. I complement
these maps with Gallo (1983) and Tell (2008) who provide more detailed information
for the states of Cordoba and Santa Fe.
The Argentinean Statistical Office (INDEC) computes GDP at the national and state
11
level, but not at the county level. In 1994 INDEC conducted the National Economic
Census (CNE) censing all business at the county level, except for the agricultural
sector, recording the value of production, costs, investment, etc. Per-capita GDP is
constructed by combining CNE’s gross product data with yearly agricultural output
estimates from the Ministry of Agriculture (see Appendix). For the states of Buenos
Aires and Santa Fe state-statistical offices compute GDP at the county level. For
these two states, the correlation between CNE’s gross product with state’s GDP at
the county level is 95%, the correlation between CNE’s gross product augmented by
the agricultural output estimates and state’s GDP is also 95%. The regression of
state’s GDP on the CNE’s gross product augmented by agricultural output has an
R2 of 90.34. I will use CNE’s gross product augmented by agricultural output as a
proxy for GDP at the county level.
Further, I will use data from the 1935 Industrial Census, which documents the number
of industrial establishments, the value of the production, the number of workers and
the usage of energy at the county level.
Data on higher education rates and share of skilled workers is from the 2001 Popu-
lation Census and is publicly available from the Argentine Statistical Office. Finally,
geo-referenced data on the quality of the soil comes from the National Institute for
Agriculture and Livestock Technology (INTA) (Cruzate et al. 1990). INTA provides
geo-referenced detailed data on the quality of the soil and elaborates an index that as-
signs a greater value to better soils. This index of land quality refers the geographical
conditions of the soil (like ground composition and rain) and not to the technologies
used for cultivation. I combine the geo-referenced data provided by INTA with the
county boundaries and compute an area weighted average of the land-quality index.
Geographical information on the average rain and temperature comes from World-
12
clim,6 data on elevation from the National Oceanic and Atmospheric Administration
(NOAA) and U.S. National Geophysical Data Center and data on ruggedness of the
terrain from Nunn and Puga (2012). All the geographical variables are geo-referenced
data which I combined with county boundaries to compute county averages. The
availability of railroads in a given county is computed as the average railroad density
in a radius of 5 km, data on railroads comes from ATLAS de Suelos de la Republica
Argentina.7
Table 1 shows the summary statistics for the variables used in this study. As a
measure of the intensity of European immigration I construct the share of European
population, defined as the fraction of European born population in 1914. The average
(and median) share of European population is 23% (16%) and a standard deviation
of 11%, with counties ranging from less than 1% to 47% of its population of European
origin. GDP per capita averages slightly above 6.700 dollars, where the bottom 25%
of the counties have less than 3.560 dollars and top 25% of the counties have a per-
capita GDP above 9.000 dollars. On average 10.4% of the population 25 years of
age and older have completed more than 12 years of education (completed secondary
school and started or finished tertiary or university degrees). Of those individuals
reporting an occupation in 2001, on average 18% work in high skilled jobs.
1.4 Estimation Strategy and Results
I will compare log per-capita GDP, higher education rates and the share of skilled
workers today between counties with different population composition in the past.
I start by running a regression of the dependent variable on the share of European
6See http://www.worldclim.org/formats.7See Cruzate et al. (1990).
13
population and other controls:
yi = α + βSEi +Xiγ + ηp + εi (1.4.1)
Where yi is the dependent variable in county i, SEi is the share of European pop-
ulation in county i in 1914, Xi are controls for county i characteristics in 1914, and
ηs are state fixed effects. County characteristics include population density, share
of the population living in urban areas (2000 or more inhabitants), share of produc-
tive land used for agriculture, land-quality and (log) distance to the city of Buenos
Aires.8 I also control for geographical characteristics (mean temperature, rainfall and
ruggedness) and for the availability of railroads.
Table 2 documents OLS results of regressing log per-capita GDP in 1994 on the share
of European population in 1914, equation (3.3.1). Column 1 only controls for state
fixed effects, column 2 adds controls for the distance to the city of Buenos Aires,
density of railroads, the share of productive land used for agriculture, population
density and urbanization rate. Column 3 adds geographical controls (rain, tempera-
ture, elevation, ruggedness and land quality). The basic OLS regression shows that
the share of Europeans in 1914 has a positive and significant coefficient. In column
3 distance to Buenos Aires has a coefficient statistically not different from zero and
density of railroads has a positive coefficient. Land quality has a positive (though not
different from zero) effect on development, and the share of productive land used for
agriculture enters positively. Population density enters negatively, while urbanization
has a positive coefficient but not statistically different from zero.
Following column 3, the preferred specification, a one standard deviation in the share
of Europeans increases per-capita GDP by 0.55 standard deviation. As this result
8The city of Buenos Aires is the capital city of the country, the main port of entry for tradedgoods and immigrants, and the most densely populated city. Proximity to this political and economicrelevant city may have independent effects on development.
14
shows, European immigration positively correlate with economic development in the
long-run, since close to eighty years after the arrival of European immigrants dif-
ferences in economic performances can be found across counties depending on the
pattern of settlement. The evidence presented in table 2 is based on correlations,
and its interpretation has to be taken with caution. If European immigrants selected
themselves into the counties depending on an omitted characteristic or an unobserv-
able variable, the results would be biased. To deal with this potential problem I will
use variation in the availability of land for settlement in the years of immigration
as instrumental variables to account for the possible endogeneity in the selection of
Europeans to the different counties.
1.4.1 Instrumental Variable Approach
European migration to the different counties in the fertile plains may not have been
random. Immigrants may have had information in hand to choose one destination in
favor of another, for example, previously settled immigrants may have sent letters or
went back to the home country to attract the rest of the family to the newly settled
area across the ocean. Even differences in infrastructure, access to railroad or size of
the cities in the plains may have played a role for immigrants when deciding where to
settle. To account for the possible endogeneity in where European immigrants settled
once they arrived to Argentina, I will construct an exogenous measure of the share of
immigrants in each county and use it as an instrumental variable for the actual share
of immigrants in a given county.
In order to construct an exogenous measure of the share of immigrants in a given
county I will exploit two sources of variation: a) changes in the frontier between
Argentina and the indigenous tribes. And b) changes in immigration to Argentina
15
between 1857 and 1914. As will be discussed below, a simple demographic model will
exploit the variation in both, available land for settlement and arrival of immigrants,
to allocate immigrants (depending on the year of arrival) and Argentineans to coun-
ties and construct an exogenous share of European population.
The History of the Instrument
Using historical information on the military campaigns followed by the Argentine
government, I am able to assign to each county a year in which (at least half of) the
land was available to settlers.
From historical records (Walther 1964) I am able to trace the area under the political
power of the Argentine government for this period. Walther (1964) documents for
several years the end result of military excursions and the boundary that resulted
of these expeditions between the Argentine government and the indigenous tribes,
in a series of maps, Figures 3-4 being two examples of it. By 1884 the Argentine
government controlled the rest of the fertile plains. I assume that no land is conquered
or lost until the next military campaign, an assumption very close to the actual
events. I overlap county boundaries to these maps and establish the date in which
the boundary moved such that a county started to be on the Argentinean side.9
The second source of variation comes from the time series of immigration to Argentina.
The migration pattern to Argentina resembles that of the USA, when comparing the
two time series the correlation of migration to Argentina and the USA is 0.795.10
9The date a county enters Argentina has not to be confused with the date in which a county isofficially founded, usually years after it was under the Argentinean power
10Data on USA migration from Historical Statistics of the United States, Millennial Edi-tion On Line, edited by Susan B. Carter, Scott Sigmund Gartner, Michael R. Haines,Alan L. Olmstead, Richard Sutch, and Gavin Wright, Cambridge University Press 2006.http://hsus.cambridge.org/HSUSWeb/toc/tableToc.do?id=Ad1-2
16
An ideal experimental setting would consist of regions (counties) that are equal in
all respects, and have a given number of Argentinean population. These regions are
then randomly shocked with European population in different intensities. I could
analyze economic and social development in these regions in the long run, and see
whether there are differences to be explained by the share European population, the
only variable that varies across regions. The actual empirical setting I am analyzing
approximates very closely my ideal experiment: it consists of regions that are geo-
graphically uniform, had an initial stock of Argentinean population and were shocked
by European population in different degrees. The key difference is that Europeans
were not randomly distributed as they choose where to settle. The IV I am proposing
consists of randomly distributing Europeans across counties, using variation in the
timing of seizure of land from the indigenous tribes and the timing of arrival of Euro-
peans, combined with a demographic model. In particular, for the shock of European
population to be random in my analysis I need that Europeans decided to migrate to
Argentina for reasons unrelated to the success or failure of the military campaigns in
conquering new land, and that the decision by the government to conquer these vast
tracks of land was independent of the arrival of European immigrants to the country.
History shows that this appears to be the case, as discussed above, military and safety
issues prompted the government to take power of this region, starting years before
the first wave of European immigrants arrived; the military campaigns in the fertile
plains ended by 1884, when slightly less than 900,000 immigrants had arrived to Ar-
gentina, in comparison to circa 3million net-immigrants immigrants that arrived by
1914. Finally, for the identifying assumption to be correct, the constructed share of
European immigration has to affect the dependent variable (per capita GDP, higher
education, etc.) only through the actual share of European immigration, while having
no effect through other variables.
17
The Instrument
The instrument is constructed by assigning Argentinean and European population to
each county and simulating the process of population growth, given the fertility and
mortality rates, over the years 1857 to 1914.
For the construction of the IV, starting in 1857 Europeans will be distributed uni-
formly across counties. The quantity of immigrants each county is assigned varies
by year of arrival, according to the time series. Argentineans, on the other hand,
are initially present in counties under the political power of the Argentine govern-
ment by 1857, but not in counties conquered after 1857. The population growth of
Argentineans and Europeans is given by the fertility rate and the mortality rate.
Europeans arrive every year and move uniformly to any county that is under the
political power of Argentina, and once they settled they never move again. Euro-
peans die at rate δ and reproduce at rate ρ, although children born to Europeans in
Argentina are considered as Argentineans.11
The initial Argentinean population in 1857 comes from the 1869 census, adjusted by
the population growth rate to the year 1857. Argentineans die at rate δ and reproduce
at rate ρ. There is a fraction φ of Argentineans that each year decides to move to a
new county. I assume they move equally to all the counties that belong to Argentina.
The mortality rate, the fertility rate and the fraction of Argentineans that move
each year are computed from the 1869, 1895 and 1914 censuses. The mortality rate
is computed to be equal to 2.2%.12 The fertility rate is computed to be equal to
11From 1857 until 1914.12I compare the stock of Europeans in 1914 with the flow of Europeans from 1857 to 1914 and
assuming that Europeans die at a constant rate δ I solve for δ such that∑1914
t=1857(1− δ)1914−t · xt =X1914, where xt is the number of Europeans that arrived at time t, and X1914 is the stock ofEuropeans in 1914.
18
5.3%.13 The moving rate for Argentineans, φ, is computed to be equal to 1.95%.14
The first stage and the analyses in the coming section are robust to changes in the
parameters of the demographic model, as well as changes in the assumption on the
initial Argentinean population. All these possibilities will be considered in Section
4.4.
The number of Europeans in each county in 1914 is defined as:
CEi =1914∑
t=1857
1
Nt
(1− δ)1914−tet · 1i{t ≥ Di}. (1.4.2)
The number of Argentineans in each county in 1914 is defined as:
CAi = CAi1857(1− δ+ ρ−φ)57 +1914∑
t=1857
1
Nt
(1− δ+ ρ−φ)1914−t(φat + ρet) ·1i{t ≥ Di},
(1.4.3)
13Given the Argentinean population from 1869 and 1914 censuses, and given that children ofEuropeans are considered Argentineans, I solve for ρ such that:w1870 = (1− δ + ρ) · w1869 + ρx1869,w1871 = (1− δ + ρ) · w1870 + ρx1870 = (1− δ + ρ)2 · w1869 + (1− δ + ρ) · ρx1870 + ρx1869,...w1914 = (1− δ + ρ)1914−1869 · w1869 +
∑1914−1t=1869 (1− δ + ρ)1914−1−t · ρxt,
where wt is the number of Argentineans at time t.14Using individual-level data from 1895 census I estimate the fraction of Argentineans living in a
different province than the one in which they were born (since there is no county level information).Define πi,a as the fraction of people aged a born in county i, who still live in county i.
πi,a =pii,a∑j p
ij,a
,
where pii,a is the number of people born in county i who live in county i, and pij,a is the number ofpeople born in county i who live in county j.Then,
πii,a = (1− φa)a.
I will compute φa for all ages and then compute the average φ weighting by the fraction of peoplein each cohort.
φ =
I∑i=1
99∑a=1
pi,a∑i
∑a pi,a
· (1− π1/ai,a ),
where pi,a/∑
i
∑a pi,a is the fraction of a years old in the population.
19
where CEi and CAi are the constructed number of Europeans and Argentineans in
county i in 1914, respectively. et is the number of Europeans that arrived in year
t, and at is the number of Argentineans that move to a different county in year t.
CAi1857 is the initial number of Argentineans in a given county. 1i{·} is an indicator
whether county i belongs to Argentina, and D is the year in which county i started
to be under the political power of the Argentine government. Nt =∑
i nit is number
of counties under the Argentinean political power at time t and nit equals 1 if county
i belongs to Argentina at time t, 0 otherwise.
The constructed share of Europeans population is defined as CSEi = CEi/(CEi +
CAi), and is used as IV for the actual share of European population. Variation in
both CEi and CAi will induce variation in the constructed share. CEi varies across
counties i depending on the year in which county i started to be under the political
power of the Argentine government, Di, and also on the number of immigrants, et,
that arrived at time t. Variation in CAi not only depends on Di, the number of
Argentineans moving, φat, and the children of Europeans, ρet, but also on the initial
stock of Argentinean population, CAi1857. Since CAi1857 is not a random variable
and depends on observed and unobserved characteristics, I will show that results hold
under a different assumption. In particular, in Section 4.4 I assume that instead of the
actual population all counties will be assigned the same initial stock of Argentineans:
CAi1857 = ¯CA1857 if CAi1857 > 0, and CAi1857 = 0 otherwise. Also I will consider
the case in which all counties are assigned the same initial stock of Argentineans,
Wi1857 = W1857.
As mentioned earlier, the conquest of the plains ended up generating 8 waves of land
incorporation: 1779, 1823, 1826, 1860, 1864, 1869, 1876 and 1884; figure 6 shows the
distribution of the counties over time, 66 counties already existed at the independence,
while six were conquered in 1860, seven in 1864, eleven in 1869, eleven in 1876 and
20
five in 1884.
1.4.2 The long-run effect of European immigration
I run the following specification for the first stage:
SEi = α + ψCSEi +Xiγ + ηp + εi (1.4.4)
Where CSEi is the constructed share of European immigration.
Figure 7 shows the first-stage relation between the share of European population
and the constructed share of European population. Figure 8 shows the first-stage
correlation when control variables and fixed effects are included. Both figures show a
strong positive correlation between the two variables.
Table 3 shows the first-stage regression, equation (1.4.4). In column 1 controls for
Xi and no geographical controls are included, column 2 adds geographical controls,
while in column 3 standard errors are clustered at the year of incorporation, Di. The
coefficient on the constructed share of immigration remains positive and significant
across specifications, confirming the result presented in figures 7 and 8. An F-test of
the coefficient ψ shows a strong first-stage with a statistic greater than 30 for the full
specification in column 3, and weak identification is ruled out by the Kleibergen-Paap
test of 34.1.
Table 4-6 show results for three different dependent variables, where the constructed
share of immigration is used as instrumental variable for the actual share of European
population. I report results for three specifications discussed above: not including
geographical controls (column 1), controlling for all variables (column 2) and cluster-
ing standard errors at the year of conquest level, D, (columns 3). In table 4 columns
21
1-3 the dependent variable is per-capita GDP in 1994. The coefficient on the share of
Europeans in columns 1-3 shows a long-run effect of the share of European popula-
tion on per-capita GDP, one standard deviation in the share of European population
increases per-capita GDP by 0.77 standard deviations.15 The point estimate of 5.49
is slightly higher than the OLS estimate of 3.91, suggesting a negative bias in the
selection of Europeans to counties and/or measurement error. The effect of having
relatively more European has an important effect in the long-run, an increase in the
share of Europeans of 5% raises per-capita GDP by one third of a standard devia-
tion. For a county like Rıo Cuarto with a share of Europeans of 20%, increasing the
share to 25% would raise per-capita GDP from 6912 dollars to 9097. Certainly an
economically significant effect.
Columns 1-3 of Table 5 examine census data on higher education in 2001. Results
also show a positive and significant effect of European immigration on this variable.
One standard deviation in the share of European immigration raises the share of
population with higher education by 0.49 s.d., an effect significant at the 5% level.
Table 6 columns 1-3 repeats the analyzes for the share of workers in high skilled
occupations. Results show a positive effect: one standard deviation in the share of
European immigration raises the share of workers in high skilled occupations by 0.51
standard deviations, a result significant at the 1% level. The results in tables 4-6 show
an important causal effect of European immigration over the long-run: Europeans
affected the degree of economic development as measured by GDP, higher education
and skilled workers. The intensity of European migration appears to have created a
divergence in the paths of economic development across counties. I will be examine
the channels through which development diverged and persisted over time in the next
section.
15One standard deviation in the share of Europeans equals 0.11 (11%), a 50% increase in theshare of Europeans for an average county
22
1.4.3 The effect of European immigration: the channels of
persistence
Why did Europeans affect economic outcomes close to a century after their arrival?
How did their initial effect on the economy propagate and persist over time? To
answer these questions I will next investigate two channels through which the effect
of European immigration created differences in the paths of economic development
over time: Industrialization and Human Capital. Both channels are linked together
and show two different aspects of the process of development.
Industrialization
Industrialization has been widely understood as an important factor in a country’s
development, countries that industrialized earlier rank higher in todays development,
per-capita income and living standards. Since the Industrial Revolution higher stan-
dards of development have been closely tight to the degree of industrialization of the
economy, where the terms industrialized nation or developed nation and advanced
economy have been used interchangeably to describe it. In the case of Argentina, in-
dustrialization arose in some counties more than in others, and cities that developed
more were also cities that experienced higher industrialization in the beginning of the
twentieth century. Why industrialization arose in the first place is an open question,
but from the industrial census in 1895, 1914 and 1935 we know that the process of
industrialization was tightly linked to immigrants and their ability and willingness to
set up and operate industrial establishments. In this sense industrialization operates
as a vehicle that propagates development over time, and long-term differences across
regions emerge between more and less industrialized counties.
Table 7 examines the nationality of the owners and workers of industrial establish-
23
ments in Argentina in 1895, 1913 and 1935. In 1895 81% of these establishments were
owned by foreigners, while 59% of the workers employed were immigrants. Close to
twenty years later, in 1913, 65% of the industrial establishments were run by foreign-
ers and workers of foreign origin made up 49% of the employment. Industry at that
time was mostly centered around the production of garment, food, wooden, metal
and chemical products, and construction. Table 6 also shows that still in 1935, 58%
of the industrial establishments were under the ownership of foreign citizens.
Below I investigate the relationship between the structure of the industrial sector in
1935 and the share of Europeans twenty years earlier.16The 1935 industrial census
records information at the establishment level and at the county level. My outcome
variables are the number of establishments per person, percent of skilled workers
in the population, per-capita value of production17 and energy in horse power per
person.18 In table 8 I examine the effect of the share of European immigration on
these variables, using IV for the share of European population. The share of European
population has a positive and significant effect on all industrial variables. Following
columns 1-4, one standard deviation (SD) in the share of European population raises
the value of industrial production by 0.66 SD, the share of skilled workers by 0.85
SD, the number of factories per person by 1.04 SD and the energy in horse power per
person by 0.64 SD. For a county like Rıo Cuarto, having a share of Europeans of 25%
instead of 20% would have raised the value of industrial production in 1935 by 41%.
Tables 7 and 8 show the importance of the European population in the process of
industrialization, in 1895, 1914 and 1935 the fraction of industrial firms owned by
Europeans was above 50%, industrial workers were mostly of European origin and
counties that happened to have a greater share of their population of European ori-
161935 is the first industrial census for which data at the county level is available17In 1935 peso currency.18For the per person variables I consider the 1914 population, since it is the closest population
census.
24
gin experienced greater industrial output and assigned more resources to industry:
workers and investment in energy production.
Consistent with the results presented in the previous section, counties where the share
of European population is greater experienced more industrial output, had a higher
share of skilled workers and greater investments in installed energy in 1935.
Human Capital: Literacy rates in 1914
Human capital is an important factor in the process of economic growth (Galor and
Weil 1999, 2000 and Galor 2005), as it is directly related to technological progress,
increases productivity and contributed to the rapid growth of per-capita GDP. Con-
temporary differences in human capital have been shown to affect development at the
macro- and micro-level, but evidence pointing to the effect of historic differences in
human capital on development in the long-run is scarce. Glaeser et al. (2004) find
evidence for human capital as a channel for growth and better political institutions
and Easterly and Levine (2009) point out that human capital was an important in-
termediating channel through which colonial settlement affected development in the
long-run. I will add to the literature providing evidence for migration generating dif-
ferences in the initial levels of human capital and on current levels of human capital.
European immigrants had a positive impact on literacy rates and the effect lasted for
more than eighty years.
The level of human capital at the end of the nineteenth century, beginning of the
twentieth century was drastically altered by the inflow of more educated immigrants.
Literacy rates vary more within Europeans than between Europeans and Argentini-
ans. Table 9 examines literacy rates in 1914 by nationality for immigrants in Ar-
gentina: while the Argentinean population is on average 63.2% literate, Germans are
25
88.2% literate and immigrants from Italy, Spain and France are 59.6%, 67.4% and
79.3% respectively. When weighted by population, on average Europeans are 64.2%
literate and the population as a whole is 63.3% literate. Europeans migrating to the
Pampas were on average more literate than locals, but the difference does not seem
important at first sight. What was the effect, if any, of a population with higher
human capital on development? Did Europeans also foster the acquisition of human
capital by the population at a large?
In table 10 I examine the relationship between the literacy rate in 1914 at the county
level and the share of European population, column 1 shows IV estimates. As column
1 shows once the endogenous distribution of immigrants is accounted for, the share
of European immigration has a positive and significant effect on literacy rates, the
coefficient of the IV regression is 0.07. This coefficient implies that one standard
deviation in the share of European population rises literacy rates by 0.15 SD. Contin-
uing with our example on Rıo Cuarto, if the share of Europeans would have been 5%
higher, the literacy rate would have been 0.35% higher, raising from 57.1% to 57.5%.
The question that tables 9 and 10 raise is what explains this difference in literacy rates
across counties? Can this difference be explained by a composition effect, namely by
substituting a less literate Argentinean by a more literate European? Or is the effect
of immigration on literacy the consequence of an increase in the acquisition of human
capital? As documented in table 9 on average Europeans are 1.1% more literate than
Argentineans, implying that switching 1% European population for 1% Argentinean
population will automatically raise literacy by 1.1%. The effect of 7% shown in table
10 column 1 is far greater than 1.1%. The composition effect can explain part but not
the whole difference in literacy rates across counties. Beyond the composition effect,
immigration has a positive externality on literacy rates on the rest of the population.
There are several potential explanations for this: it may be that Europeans provide
26
more education to their offspring, it may also be related to Europeans demanding
more schools in the places were they settled and afterward schools provide educa-
tion to all citizens, or the Argentinean government providing education to the newly
arrived immigrant, or it may also be the case were economic progress generated a
demand for more skilled labor, providing higher incentives to acquire human capital.
In accordance to the results provided in the previous section, places were Europeans
accounted for a higher share of the population had higher literacy rates in 1914, partly
due to more literate immigrants and partly due to a positive externality on the rest
of the population (their children and others). In the next section I will investigate if
more education was provided in areas with a higher share of European immigrants.
European Immigration and Human Capital formation in 1914
I analyze whether more education was provided in areas with higher shares of Euro-
pean immigrants. Were counties with a higher share of European population more
literate because of school availability? Did the Argentinean government promote ed-
ucation in areas with more Europeans to assimilate them to the native population?
Are counties with higher literacy the results of public financed education, or the result
of private financed education?
Since mid-eighteenth century schools were built through the country by the govern-
ment, offering free public education to all individuals in school-age (6 to 14 years
old). These schools were mostly in urban areas or highly densely populated areas.
Private schools were also present and offered religious learning and/or were present
in areas without public schools. Given that the government followed an active policy
of educating the population, it is plausible that counties with a higher share of Euro-
peans experienced more public financed education. However, the opposite is actually
true, areas with a higher share of European immigrants are associated with a higher
27
number of private schools per schooling age population and a lower number of public
schools.
In table 10, columns 2 and 3 I regress the number of public schools and private schools
per 1000 school-age population on the share of European immigrants, controlling for
county characteristics. Census data on schools in 1914 lists schools’ location’s and the
school-age population in each county, from which I construct the number of schools
per 1000 school-age children, on average there are 5.3 public schools and 0.85 private
schools in each county per 1000 school-age population, with a standard deviation of
2.32 and 0.71, respectively. In column 2, I regress the number of public schools per
school-age population on the share of European population, the share of European
population has a negative and significant effect on the number of public schools. One
standard deviation in the share of European population reduces the number of public
schools by 0.61 standard deviations, a magnitude equivalent to reducing close to one
and a half schools. Column 3 shows IV estimates of regressing the number of private
schools on the share of Europeans, results show a positive, although not significant,
effect of immigrants on the quantity of schools, one standard deviation in the share
of immigrants increases by 0.38 SD the number of private schools per school-age
population.
These findings show that government educational policy was not targeted to areas
where Europeans concentrated, quite the opposite, an increase by 0.11 in the share
of Europeans is associated with a reduction of 1.5 public schools. On the other hand,
the share of Europeans has a positive but not significant effect on the number of
private schools. The evidence points to literacy rates being higher in areas with more
Europeans not because of educational policies pursued by the national government,
but because of individual decisions of the citizens of these counties.
28
1.4.4 Robustness Checks
The results are robust to a series of variations in the specification and construction
of the IV: I consider changes on the assumptions of the demographic model, as well
as alternative explanations for the divergence in economic growth. In table 11 I
consider 6 variations to the parameters of the demographic model presented in section
4: column 1 shows results when initial stock of Argentineans is fixed among counties
with Wi,1857 > 0, and 0 otherwise, namely I assume Wi,1857 = 6269, the average
number per county of Argentineans in 1857. In column 2 all counties have an average
initial number of Argentineans equal to 3600. I also consider arbitrarily high (double)
values for the parameters of the model as follows: in column 3 the moving rate φ
equals 6%, in column 4 the fertility rate ρ equals 10%, in column 5 the mortality rate
δ equals 6% and in column 6 φ = 6%, ρ = 10% and δ = 6% simultaneously. Columns
1-6 in table 10 show that results remain consistent with my main results, changes in
the assumptions of the model do not alter the effect on per-capita GDP and literacy
rates (results for all the other variables considered in this study are also robust to
these changes).
In table 12 I consider alternative explanations to the divergence in the paths of eco-
nomic development: land inequality and access to a highly valuable export crop:
wheat. Columns 1 and 2 show that adding these variables to the analyses do not
alter the statistical relevance of the share of Europeans in explaining economic devel-
opment. Finally in column 3 I repeat the main regressions of the paper weighting by
the population of the county. Relative differences in the population size of a county
may be relevant to assess the effect of the population composition on development.
As column 3 shows, weighting for the population does not change the results.
In sum, the regressions shown in the previous sections are robust to the inclusion
29
of other potential relevant variables, changes in the parameters of the model and
weighting by population.
1.5 Conclusion
The period between mid eighteen hundred and the First World War saw an unprece-
dented flow of European immigrants to Argentina, mostly to the rural and urban
areas across the fertile plains. Areas where Europeans accounted for a greater share
of the total population developed more than areas with fewer Europeans, as measured
by GDP close to one hundred years later.
Why were areas with a higher share of European immigrants able to develop more
than areas where Europeans represented a fewer share of the population? As I have
discussed above, the Pampas provides an area of study where political institutions are
common across counties and geographical conditions are uniform, therefore differences
in development are found in the role played by immigration and human capital.
When compared to Argentineans, Europeans were engage in industrial production
complementary to human capital, knowledge or skills. Europeans started most of the
industrial activities and provided for most of the industrial (skilled and unskilled)
workers.
Moreover, where Europeans accounted for a greater share of the population, the
population had higher literacy rates. This higher literacy rates cannot be explained
by differences in literacy of Europeans and Argentineans alone, Europeans had a
positive effect on literacy rates beyond what can be attributed to a composition effect.
Higher literacy rates cannot be explained by an effort of the national government to
educate and assimilate immigrants, since public schools were less available in counties
30
were Europeans accounted for a higher share of the population. Private schools were
created either by Argentineans or immigrants, and although there is no statistical
significant effect of Europeans on the availability of private schools, the data shows a
positive correlation between private schools availability and the share of Europeans.
Europeans generated a positive externality on the society as a whole, generating
greater literacy rates.
These results point to the importance of people themselves in the process of economic
development. This study of the fertile plains of Argentina, an area with equal political
institutions and uniform geographical characteristics, shows that there is a long-term
impact of initial differences in the composition of the population and human capital
on economic development.
31
Figure 1: Correlation between current log per-capita GDP and the share of European population in 2000.
AFG
ALBDZA
AGO
ARG
ARM
AUS AUT
AZE
BHR
BGD
BLR
BEL
BLZ
BEN
BTN BOL
BIH
BWABRA
BGR
BFA
BDI
KHM
CMR
CAN
CPV
CAFTCD
CHL
CHN
COL
COM
ZAR
COG
CRI
CIV
HRV
CUB
CYP
CZE
DNK
DOMECU
EGY
SLV
GNQ
ERI
EST
ETH
FJI
FINFRA
GAB
GMB
GEO
DEU
GHA
GRC
GTM
GINGNB
GUY
HTI
HND
HKG
HUN
ISL
IND
IDN
IRN
IRQ
IRL
ISRITA
JAM
JPN
JOR
KAZ
KEN
KOR
KWT
KGZLAO
LVALBN
LSO
LBR
LBY LTU
LUX
MKD
MDGMWI
MYS
MLI
MLT
MRT
MUSMEX
MDAMNG
MAR
MOZ
NAM
NPL
NLD
NZL
NIC
NER
NGA
NOR
OMN
PAK
PAN
PNG
PRY
PER
PHL
POL
PRTPRI
QAT
ROMRUS
RWA
WSM
STP
SAU
SEN
YUG
SLE
SGP
SVK
SVN
ZAF
ESP
LKA
VCT
SDN
SWZ
SWECHE
SYR
TJKTZA
THA
TGO
TON
TTO
TUN
TUR
TKM
UGA
UKR
AREGBR
USA
URY
UZB
VEN
VNMYEMZMBZWE
46
81
01
2(l
og)
per
-ca
pita
GD
P(m
ean
199
5-2
010)
0 .2 .4 .6 .8 1Share of Europeans in 2000
32
Figure 2: Correlation between log per-capita GDP in 1994 and the share of European population in 1914, in Argentina.
56
78
91
0(l
og)
per
-ca
pita
GD
P (
1994
)
0 .1 .2 .3 .4 .5Share of European Population (1914)
33
Figure 3
34
Figure 4
35
Figure 5: Immigration Time Series.
-50
0
50
100
150
200
250
300
350
1857
1859
1861
1863
1865
1867
1869
1871
1873
1875
1877
1879
1881
1883
1885
1887
1889
1891
1893
1895
1897
1899
1901
1903
1905
1907
1909
1911
1913
Net
-Im
mig
ratio
n, in
Tho
usan
dNet Immigration Immigration
36
Figure 6: Cumulative Net-Immigration and Area for settlement.
0
0.5
1
1.5
2
2.5
3
1857
1859
1861
1863
1865
1867
1869
1871
1873
1875
1877
1879
1881
1883
1885
1887
1889
1891
1893
1895
1897
1899
1901
1903
1905
1907
1909
1911
1913
Cum
mul
ativ
e Im
mig
ratio
nin
Mill
ions
300
350
400
450
500
550
600
650
Are
a in
Tho
usan
ds k
m2
Cummulative Net-Immigration Total Area
37
Figure 7: 1st Stage correlation between the share of European population and the constructed share of European immigration.
-.2
-.1
0.1
.2S
har
e o
f Eur
ope
an P
opu
latio
n (
191
4)
-.2 -.1 0 .1 .2Constructed Share of European Population (1914)
coef = .80743583, (robust) se = .09312186, t = 8.67
38
Figure 8: 1st Stage correlation between the share of European population and the constructed share of European immigration, control variables and fixed effects included.
-.1
0.1
.2S
har
e o
f Eur
ope
an P
opu
latio
n (
191
4)
-.2 -.1 0 .1 .2Constructed Share of European Population (1914)
coef = .46400939, (robust) se = .07047107, t = 6.58
39
Tabl
e 1:
Sum
mar
y St
atis
tics
Shar
e of
Eur
opea
n po
pula
tion,
191
40.
230.
110.
16G
DP
per-
capi
ta, 1
994
6754
4190
3560
log
GD
P pe
r-ca
pita
, 199
48.
590.
788.
18Sh
are
of p
op. w
/hig
her e
duca
tion,
200
10.
10.
020.
09Sh
are
of sk
illed
wor
kers
, 200
10.
180.
040.
15lo
g in
dust
rial o
utpu
t pe
r-ca
pita
, 193
54.
41.
143.
87Sk
illed
wor
kers
per
-100
0 in
divi
dual
s, 19
351.
992.
060.
89N
umbe
r of f
acto
ries p
er-1
000
indi
vidu
als,
1935
3.69
2.16
2.16
Ener
gy in
H.P
. per
-cap
ita, 1
935
0.1
0.14
0.05
Lite
racy
rate
, 191
40.
630.
050.
58N
umbe
r of p
rivat
e sc
hool
s per
-100
0 sc
hool
age
pop
.0.
850.
710.
35N
umbe
r of p
uclic
scho
ols p
er-1
000
scho
ol a
ge p
op.
5.33
2.32
3.63
Num
ber o
f sec
onda
ry sc
hool
s per
-100
0 in
divi
d. 2
007
0.89
0.45
0.63
Perc
ent o
f Lan
d us
ed fo
r Agr
icul
ture
0.28
0.23
0.07
Popu
latio
n D
ensi
ty6.
675.
532.
78U
rban
Rat
e0.
330.
180.
22N
umbe
r of o
bser
vatio
ns: 1
36
Var
iabl
eM
ean
Stan
dard
Dev
iatio
n50
th P
erce
ntile
40Table 2: OLS
Dependent Variable:(1) (2) (3)
5.668*** 4.403*** 3.914***(0.632) (0.732) (0.796)
Distance to BA City -0.010 0.079(0.114) (0.151)
Land Quality 0.004(0.004)
Railroad Density 0.069*** 0.052*(0.026) (0.029)
0.715*** 0.644**(0.248) (0.293)
Population Density in 1914 -0.037*** -0.028**(0.009) (0.011)
Urban Rate in 1914 0.684** 0.557(0.335) (0.341)
Geographic Controls no no yesProvince Fixed Effects yes yes yesObservations 136 136 136R-squared 0.507 0.561 0.596
European population / total population, 1914
Percent of Land used for Agriculture in 1914
log per capita GDP, 1994
Note: Ordinary least squares regressions with robust standard errors in parentheses. Dependent variable inall columns is log per-capita GDP in 1994. In column 1 only province fixed effects are included. Column2 includes all control variables except for the geographical controls. In column 3 all control variables areincluded. *** p<0.01, ** p<0.05, * p<0.1.
41Table 3: First Stage
Dependent Variable:(1) (2) (3)
0.450*** 0.464*** 0.464***(0.084) (0.070) (0.069)
Distance to BA City 0.040*** 0.018 0.018(0.010) (0.012) (0.012)
Land Quality 0.000 0.000(0.000) (0.000)
Railroad Density 0.001 0.002 0.002(0.003) (0.003) (0.003)
0.221*** 0.184*** 0.184***(0.027) (0.024) (0.025)
Population Density in 1914 0.003* 0.003** 0.003***(0.001) (0.001) (0.001)
Urban Rate in 1914 0.122*** 0.086** 0.086***
(0.033) (0.040) (0.021)
Geographic Controls no yes yes
Province Fixed Effects yes yes yesCluster SE at year of conquest no no yesObservations 136 136 136Adjusted R-squared 0.768 0.805 0.805
Percent of Land used for Agriculture in 1914
Constructed European population / total population
European population / total population
Note: Ordinary least squares regressions with robust standard errors in parentheses. Dependent variablein all columns is the Share of European Population in 1914. In column 1 includes all the controlvariables except for the geographical controls. In column 2 all control variables are included and incolumn 3 standard errors are clustered at the year of incorporation. *** p<0.01, ** p<0.05, * p<0.1.
42Table 4: IV Results
Dependent Variable:(1) (2) (3)
5.564*** 5.493*** 5.493***(1.451) (1.514) (0.688)
Distance to BA City -0.085 0.000 0.000(0.149) (0.162) (0.050)
Land Quality 0.004 0.004(0.004) (0.002)
Railroad Density 0.067*** 0.047* 0.047**(0.025) (0.028) (0.014)0.419 0.291 0.291
(0.395) (0.360) (0.309)Population Density in 1914 -0.038*** -0.031*** -0.031***
(0.009) (0.011) (0.005)
Urban Rate in 1914 0.541 0.437 0.437
(0.344) (0.330) (0.374)
Geographic Controls no yes yes
Province Fixed Effects yes yes yesCluster SE at year of conquest no no yesObservations 136 136 136Adjusted R-squared 0.553 0.583 0.432
Percent of Land used for Agriculture in 1914
European population / total population
log per capita GDP, 1994
Note: Instrumental Variable regressions with robust standard errors in parentheses. Dependent variablein all columns is log per-capita GDP in 1994. In column 1 includes all the control variables except forthe geographical controls. In column 2 all control variables are included and in column 3 standarderrors are clustered at the year of incorporation. *** p<0.01, ** p<0.05, * p<0.1.
43Table 5: IV Results
Dependent Variable:(1) (2) (3)
0.074* 0.089** 0.089**(0.044) (0.041) (0.037)
Distance to BA City -0.004 -0.006* -0.006*(0.004) (0.003) (0.003)
Land Quality -0.000** -0.000**(0.000) (0.000)
Railroad Density 0.002** 0.002*** 0.002***(0.001) (0.001) (0.001)-0.020 -0.012 -0.012(0.013) (0.011) (0.010)
Population Density in 1914 0.000 -0.000 -0.000(0.000) (0.000) (0.001)
Urban Rate in 1914 0.026** 0.034*** 0.034**
(0.011) (0.009) (0.012)
Geographic Controls no yes yes
Province Fixed Effects yes yes yesCluster SE at year of conquest no no yesObservations 136 136 136Adjusted R-squared 0.295 0.472 0.224
Percent of Land used for Agriculture in 1914
European population / total population
share of population with higher education, 2001
Note: Instrumental Variable regressions with robust standard errors in parentheses. Dependent variablein all columns is the share of population age 25 and above with higher education in 2001. In column 1includes all the control variables except for the geographical controls. In column 2 all control variablesare included and in column 3 standard errors are clustered at the year of incorporation. *** p<0.01, **p<0.05, * p<0.1.
44Table 6: IV Results
Dependent Variable:(1) (2) (3)
0.174*** 0.184*** 0.184(0.067) (0.066) (0.105)
Distance to BA City 0.006 0.002 0.002(0.006) (0.006) (0.007)
Land Quality -0.000 -0.000(0.000) (0.000)
Railroad Density 0.003*** 0.004*** 0.004***(0.001) (0.001) (0.001)0.035* 0.036* 0.036(0.021) (0.019) (0.040)
Population Density in 1914 -0.002*** -0.002*** -0.002**(0.001) (0.001) (0.001)
Urban Rate in 1914 0.065*** 0.064*** 0.064***
(0.018) (0.015) (0.016)
Geographic Controls no yes yes
Province Fixed Effects yes yes yesCluster SE at year of conquest no no yesObservations 136 136 136Adjusted R-squared 0.675 0.738 0.484
Percent of Land used for Agriculture in 1914
European population / total population
share of population with high skilled occupations, 2001
Note: Instrumental Variable regressions with robust standard errors in parentheses. Dependent variablein all columns is the share workers in high-skilled occupation in 2001. In column 1 includes all thecontrol variables except for the geographical controls. In column 2 all control variables are includedand in column 3 standard errors are clustered at the year of incorporation. *** p<0.01, ** p<0.05, *p<0.1.
45Table 7: Ownership and Industrial Workers
year
1895 0.811913 0.651935 0.58
1895 0.591913 0.49
Share of Foreigners
Ownership
Workers
46
Tab
le 8
: IV
Res
ults
(1)
(2)
(3)
(4)
6.88
5***
16.0
25**
*20
.381
***
0.81
7**
(2.4
98)
(6.0
91)
(5.5
27)
(0.3
23)
Dis
tanc
e to
BA
Cit
y-0
.235
-0.8
47-0
.725
0.03
4(0
.299
)(0
.714
)(0
.547
)(0
.030
)L
and
Qua
lity
-0.0
22**
*-0
.056
***
-0.0
01-0
.003
***
(0.0
08)
(0.0
19)
(0.0
15)
(0.0
01)
Rai
lroa
d D
ensi
ty0.
000
-0.1
89*
0.01
20.
009
(0.0
53)
(0.1
12)
(0.0
76)
(0.0
07)
-0.8
70-2
.956
**-4
.402
***
-0.1
87*
(0.6
72)
(1.3
26)
(1.5
48)
(0.1
10)
Pop
ulat
ion
Den
sity
in 1
914
0.03
5*0.
217*
**0.
019
0.00
7***
(0.0
20)
(0.0
47)
(0.0
46)
(0.0
03)
Urb
an R
ate
in 1
914
0.21
5-0
.914
1.65
30.
006
(0.7
90)
(1.2
38)
(1.0
57)
(0.0
70)
Geo
grap
hic
Con
trol
sye
sye
sye
sye
s
Pro
vinc
e F
ixed
Eff
ects
yes
yes
yes
yes
Obs
erva
tion
s13
613
613
613
6A
djus
ted
R-s
quar
ed0.
190
0.24
30.
344
0.08
4N
ote:
Inst
rum
enta
lV
aria
ble
regr
essi
ons
wit
hro
bust
stan
dard
erro
rsin
pare
nthe
ses.
Dep
ende
ntva
riab
les
inco
lum
ns1-
4ar
eth
eva
lue
ofin
dust
rial
prod
ucti
on,
the
num
ber
ofsk
ille
dw
orke
rspe
r10
00in
divi
dual
s,th
enu
mbe
rof
fact
orie
spe
r10
00in
divi
dual
s an
d th
e en
ergy
in h
.p. p
er p
erso
n. E
ach
colu
mn
incl
udes
all
the
cont
rol v
aria
bles
. ***
p<
0.01
, **
p<0.
05, *
p<
0.1.
Per
cent
of
Lan
d us
ed f
or A
gric
ultu
re
in 1
914
Eu
rop
ean
pop
ula
tion
/ to
tal
pop
ula
tion
ener
gy in
h.p
. per
pe
rson
log
valu
e of
in
dust
rial
pro
duct
ion
Dep
ende
nt V
aria
ble:
skil
led
wor
kers
per
-10
00 in
divi
dual
sfa
ctor
ies
per-
1000
in
divi
dual
s
47Table 9: Literacy Rates by Contry of Birth
Nationality Literacy rate
Argentina 63.2%Average European 64.2%Average Population 63.3%
Austria 69.2%France 79.3%Germany 88.2%Great Britain 90.9%Italy 59.6%Spain 67.4%Switzerland 86.9%
48
Tab
le 1
0: I
V R
esul
ts
(1)
(2)
(3)
(4)
0.07
0**
-12.
817*
**2.
430
1.48
4**
(0.0
35)
(4.5
36)
(1.8
50)
(0.6
66)
Dis
tanc
e to
BA
Cit
y-0
.011
***
-1.5
22**
*-0
.119
0.04
1(0
.003
)(0
.436
)(0
.169
)(0
.064
)L
and
Qua
lity
-0.0
00*
-0.0
04-0
.003
-0.0
04*
(0.0
00)
(0.0
12)
(0.0
04)
(0.0
02)
Rai
lroa
d D
ensi
ty0.
000
0.03
20.
019
0.00
5(0
.001
)(0
.087
)(0
.033
)(0
.011
)-0
.021
**-1
.142
0.18
5-0
.218
(0.0
09)
(1.2
68)
(0.4
94)
(0.1
75)
Pop
ulat
ion
Den
sity
in 1
914
-0.0
01**
*-0
.015
-0.0
06-0
.030
***
(0.0
00)
(0.0
31)
(0.0
14)
(0.0
06)
Urb
an R
ate
in 1
914
0.00
1-0
.783
0.37
20.
019
(0.0
07)
(1.2
32)
(0.3
64)
(0.1
96)
Geo
grap
hic
Con
trol
sye
sye
sye
sye
s
Pro
vinc
e F
ixed
Eff
ects
yes
yes
yes
yes
Obs
erva
tion
s13
613
613
613
6A
djus
ted
R-s
quar
ed0.
945
0.49
00.
226
0.66
1N
ote:
Inst
rum
enta
lV
aria
ble
regr
essi
ons
wit
hro
bust
stan
dard
erro
rsin
pare
nthe
ses.
Dep
ende
ntva
riab
les
inco
lum
ns1-
3ar
eth
esh
are
ofli
tera
tepo
pula
tion
in19
14,
the
num
ber
ofpu
blic
scho
ols
per
1000
scho
ol-a
gepo
pula
tion
and
the
num
ber
ofpr
ivat
esc
hool
spe
r10
00sc
hool
-age
popu
lati
on.
Eac
hco
lum
nin
clud
esal
lth
eco
ntro
lva
riab
les.
***
p<0.
01,
**p<
0.05
,*
p<0.
1.
Per
cent
of
Lan
d us
ed f
or A
gric
ultu
re
in 1
914
Eu
rop
ean
pop
ula
tion
/ to
tal
pop
ula
tion
Sec
onda
ry S
choo
ls
x 10
00 in
divi
dual
s,
2007
shar
e of
lite
rate
po
pula
tion
Dep
ende
nt V
aria
ble:
Pub
lic
Sch
ools
x
1000
sch
ool-
age
popu
lati
on
Pri
vate
Sch
ools
x
1000
sch
ool-
age
popu
lati
on
49
Tabl
e 11
: Rob
ustn
ess C
heck
sD
epen
dent
Var
iabl
e:A
ssum
ptio
ns:
(1)
(2)
(3)
(4)
(5)
(6)
7.24
9***
5.38
7**
4.81
0***
7.02
5***
5.49
2***
5.30
0***
(1.8
26)
(2.5
01)
(1.7
74)
(1.6
63)
(1.6
66)
(1.5
71)
Dis
tanc
e to
BA
City
-0.0
870.
006
0.03
4-0
.076
0.00
00.
010
(0.1
72)
(0.1
92)
(0.1
73)
(0.1
52)
(0.1
69)
(0.1
64)
Land
Qua
lity
0.00
40.
004
0.00
40.
004
0.00
40.
004
(0.0
04)
(0.0
04)
(0.0
04)
(0.0
04)
(0.0
04)
(0.0
04)
Rai
lroad
Den
sity
0.04
00.
047*
0.04
9*0.
041
0.04
7*0.
047*
(0.0
29)
(0.0
27)
(0.0
28)
(0.0
28)
(0.0
28)
(0.0
28)
-0.1
010.
315
0.44
4-0
.051
0.29
10.
335
(0.4
30)
(0.5
85)
(0.3
92)
(0.4
52)
(0.3
72)
(0.3
66)
Popu
latio
n D
ensi
ty in
191
4-0
.034
***
-0.0
31**
-0.0
30**
*-0
.034
***
-0.0
31**
*-0
.031
***
(0.0
12)
(0.0
13)
(0.0
11)
(0.0
12)
(0.0
11)
(0.0
11)
Urb
an R
ate
in 1
914
0.30
30.
445
0.48
90.
320
0.43
70.
451
(0.3
81)
(0.3
59)
(0.3
26)
(0.3
67)
(0.3
31)
(0.3
29)
Geo
grap
hic
Con
trols
yes
yes
yes
yes
yes
yes
Prov
ince
Fix
ed E
ffec
tsye
sye
sye
sye
sye
sye
sO
bser
vatio
ns13
613
613
613
613
613
6A
djus
ted
R-s
quar
ed0.
538
0.58
50.
592
0.54
60.
583
0.58
6N
ote:
Inst
rum
enta
lV
aria
ble
regr
essi
ons
with
robu
stst
anda
rder
rors
inpa
rent
hese
s.D
epen
dent
varia
bles
inco
lum
ns1-
6ar
edi
ffer
enta
ssum
ptio
nsfo
rth
eco
nstru
ctio
nof
the
IV.E
ach
colu
mn
incl
udes
allt
heco
ntro
lvar
iabl
es.*
**p<
0.01
,**
p<0.
05,*
p<0.
1.
Perc
ent o
f Lan
d us
ed fo
r A
gric
ultu
re in
191
4
Eur
opea
n po
pula
tion
/ tot
al
popu
latio
n
log
per c
apita
GD
P, 1
994
If in
itial
A
rg.>
0,
Arg
_0=6
300
Arg
_0=3
600
for a
ll co
untie
s m
ovin
g ra
te
=6%
ferti
lity
rate
=1
0%m
orta
lity
rate
=6
%as
sum
ptio
ns
(3),
(4) a
nd (5
)
50Table 12: Robustness Checks
Dependent Variable:
Assumptions:
(1) (2) (3)
5.451*** 5.389*** 4.067***(1.533) (1.444) (1.389)0.136
(0.516)Land-gini in 1914 -1.058*
(0.590)
Distance to BA City -0.004 0.059 0.036(0.164) (0.158) (0.141)
Land Quality 0.004 0.003 0.004(0.004) (0.004) (0.003)
Railroad Density 0.045 0.042 0.037(0.028) (0.029) (0.025)
0.257 -0.051 0.379
(0.349) (0.452) (0.275)
Population Density in 1914 -0.030*** -0.027** -0.026***
(0.011) (0.012) (0.010)
Urban Rate in 1914 0.428 0.490 0.407
(0.325) (0.325) (0.318)Geographic Controls yes yes yesProvince Fixed Effects yes yes yesObservations 136 136 136Adjusted R-squared 0.580 0.592 0.582
Obs. Weighted by population
log per capita GDP, 1994
Note: Instrumental Variable regressions with robust standard errors in parentheses. Dependentvariables in columns 1-3 is log per-capita GDP in 1914. In column 1 the percent of land used for wheatproduction is included as a regressor. In column 2 the land gini is included as a regressor. In column 3observations are weighted by the population. Each column includes all the control variables. ***p<0.01, ** p<0.05, * p<0.1.
European population / total populationPercent of Land used for Wheat
Percent of Land used for Agriculture in 1914
Wheat Land-gini
Chapter 2
Beliefs in Market Economy and Macroeconomic Crises
while Young
2.1 Introduction
In this paper I analyze how and when beliefs are formed. In particular I analyze
beliefs in market economy and explore how economic crises affect these beliefs. I also
analyze whether the effect of crises on beliefs differs between old and young people
and whether crises have a long lasting effect on beliefs.
Why is it important to study belief’s formation? Why does it make any difference to
form beliefs at younger or older ages? One of the reasons to study the formation of
beliefs refers to the literature on long term persistence of institutions. Tabellini (2007)
argues how distant political and economic history shape the functioning of current
institutions, where slow moving individual values, beliefs and convictions provide an
explanation to the persistence of institutional outcomes. He goes further and sets
a research agenda around the questions of how values and beliefs influence political
outcomes, and how these values evolve over time. Building up from this last question,
51
52
I analyze whether beliefs are more prone to being revised during early adulthood than
in latter years and whether beliefs are preserved almost unaltered for the rest of our
lives.
Different theories have been proposed to explain how and when beliefs are formed,
but the literature on belief formation is not vast, mainly because we do not observe
beliefs. In general we observe decisions, but behind the choice-making behavior there
are underlying beliefs upon which decisions have been based. Some theories of how
beliefs are formed argue that beliefs are engrained in the culture inherited by individ-
uals and therefore are difficult to change in the short run, evolving slowly over long
periods of time. Other theories stress that beliefs are formed according to present
conditions faced by individuals and therefore are a response to the current environ-
ment and endowments. In between these extreme positions is the work by Giuliano
and Spilimbergo (2009), who propose two hypotheses derived from the psychological
literature: the Impressionable Years Hypothesis (IYH) and the Increasing Persistence
Hypothesis (IPH). According to the IYH what matters in individual belief formation
are the circumstances experienced by people in their early adulthood, a period of
mental plasticity where core attitudes, values and beliefs are formed. According to
this hypothesis these attitudes and beliefs then remain largely unaltered through-
out the remaining adult years. The Increasing Persistence Hypothesis (IPH) states
that individuals are flexible while young, but as they age their flexibility gradually
decreases.
According to the first view, beliefs are part of the culture of the country, and if they
change, this will only happen over long periods of time, meaning that current events
(such as economic crisis or unemployment) should have little or no impact. Quite the
opposite would be true according to the second view. It is the current environment,
the endowments individuals have and their own experience that determine beliefs.
53
Evidence pointing to this effect has been found by Di Tella, Galiani and Schargrodsky
(2007), who find that individuals beliefs changed when land titles were randomly given
to squatters in Buenos Aires. In between Giuliano and Spilimbergo (2009) analyze
how recessions affect preference for redistribution in the US and find evidence that
individuals growing up during a recession tend to believe that success in life depends
more on luck than on effort and therefore support redistributive policies.
The relationship between beliefs and economic outcomes has been stressed by numer-
ous authors, most notably in the literature on income redistribution. In a seminal
paper, Piketty (1995) shows that people prefer more redistribution to the poor if
they believe that poverty is caused by circumstances beyond individual control. If
people believe in self-determination of income, i.e. that income is a consequence of
effort, they will believe that outcomes are determined by factors that are within indi-
vidual control, such as a willingness to work hard. Those who believe in exogenous-
determination of income, place more importance on factors beyond individual control,
such as luck or lack of opportunity. Alesina and Angeletos (2005) employ a similar
argument where beliefs on whether effort pays or not will translate into different tax
choices and tax choices will ultimately reinforce these beliefs. In their model two
equilibriums arise: one with low taxes and high effort and a second one with high
taxes and low effort. Further, this result has been shown by Fong (2000): beliefs on
whether income depends on luck or effort have large and significant effects on peoples
support for redistribution.
Another theory on income redistribution relates to beliefs on upward mobility and
was advanced by Benabou and Ok (2001). These authors stress the Prospect of Up-
ward Mobility hypothesis (POUM), in their work individuals hold rational beliefs on
upward mobility and therefore do not vote redistributive policies while being poor,
since they expect to have an income above average in the future. Further, work on
54
upward mobility and redistribution has been made by Alesina et al. (2001), Alesina
and Angeletos (2005) and Alesina and La Ferrara (2005). In Alesina and Angele-
tos (2005) different beliefs about how fair social competition is and what determines
income inequality influenced the redistributive policy chosen democratically in a so-
ciety. In equilibrium social beliefs will be self-fulfilled, where a society that believes
that individual effort determines income, will have in equilibrium a low-tax policy
and effort will be high. If, on the other hand, a society believes that income is a con-
sequence of exogenous conditions to the individual, the equilibrium tax policy will be
greater, more redistribution will take place and effort will be low. The authors argue
that this self-fulfilling mechanism can explain the different perceptions on income and
inequality between the US and continental Europe choices of redistributive policies.
In another study on the formation of beliefs, Di Tella, Galiani and Schargrodsky
(2007) show how peoples report beliefs closer to those that favor the workings of a free
market when randomly assigned property rights were given (i.e. whether individuals
are more or less materialist and individualist after they hold property rights). Further,
in a study on the relationship between beliefs and macroeconomic variables, Di Tella
et al. (2007) analyzes how macro volatility affects political beliefs in Venezuela and
finds that real shocks have a role in the determination of beliefs. They show that
high levels of volatility affects the reward to effort, and this in turn affects people’s
beliefs about the degree of regulation and taxation that is required for their society.
In this paper I will address the question of whether individuals facing a macroeco-
nomic shock during early adulthood formed beliefs differently, and if these beliefs
persisted over time. I exploit cross country variation in macroeconomic crises to
identify the effect of these events on the beliefs people hold, and will asses at which
ages economic crises affect these beliefs.
This paper differs from previous work in several dimensions: the beliefs under anal-
55
yses, the time span where beliefs are formed (or modified) and the macroeconomic
shock under consideration. First I analyze cross-country data from countries in Latin
America. Second, the relationship between macroeconomic crises and beliefs has not
been analyzed before. Previous works studied how economic recessions or macroe-
conomic volatility influenced the formation of beliefs, but the effect of an important
event, such as an economic crisis, was lacking. An economic crisis can be regarded
as episode that strongly affects peoples economic relationships, generating an inter-
esting environment to study the formation of beliefs in market economy. Further, I
introduce a new outcome variable, beliefs in market economy. My work is related
to Giuliano and Spilimbergo (2009), since they study the relationship between reces-
sions and beliefs on whether success in life depends more on luck than on effort, for
the US. Also, close to my work is Di Tella et al. (2007) who analyze the effect of
macroeconomic volatility on beliefs.
My main result is that economic crises that happened during early adulthood, between
22 and 25 years old, have a negative impact on the probability of believing in the
market economy and these beliefs remain fairly stable over life. Further, there is
evidence (subject to the caveats I discus below) to support the Impressionable Years
Hypothesis (IYH)1.
The next section describes the data used in this paper and the hypothesis to be
tested. Section 3 discusses the main results, section 4 the robustness checks and the
last section summarizes.
1As defined by the period between 18 and 25 years old.
56
2.2 Data Description & Methodology
Latinbarometro is an annual public opinion survey that involves some 19,000 inter-
views in 18 Latin American countries, produced by Latinobarometro Corporation,
a non-profit NGO based in Santiago, Chile. Latinobarometro surveys cover ques-
tions on democracy and economies as well as societies, using indicators of opinion,
attitudes, behavior and values. The survey has been conducted since 1995, but its
questionnaire has changed over the years. Many questions can be tracked for most
years, but not all of them. For the current study I am interested in questions related
to beliefs on Market Economy, which can be found since 2003 onwards. The data is
available for free, with the exception of the last three years (2006, 2007 and 20082).
Unfortunately the survey changed its questionnaire in 2004, therefore I will focus on
a cross section of countries for the year 20053.
The question under study in this paper is about beliefs on market economy and reads:
“For a country to develop, it is necessary to have a market economy” Individuals had
to choose between one of the following answers: “I strongly agree, I agree, I do not
agree, I disagree.”
The importance of this question relates to the fact that countries in Latin America
are in the process of being develope and different views prevail within citizens on
which economic institutions channel resources towards economic growth. Anecdotal
evidence suggests that a market economy is far from obvious a solution to the eco-
nomic backwardness that prevails in many countries. Successions of pro-market and
pro-regulation governments over the years in many countries are just an example of
how much beliefs differ across population.
22009 is not yet available.3I was not able to get the funding to buy the data for the years 2006, 2007 and 2008.
57
For my paper I will distinguish between those who agree and those who do not
agree with the market economy being necessary for economic growth, therefore I
dichotomize the variable setting it equal to 1 if he/she agrees and 0 otherwise.
Figure (I) shows the probability of believing in the market economy by year of birth
and country. Two facts stand out from this figure: First, there are no clear trends
across countries, nor within countries. Second, within countries beliefs vary consid-
erably across ages. An important fact that comes out of the graphs is that there is a
non-monotonic relationship between beliefs and age4
With respect to economic crises, I construct a variable that identifies if an individual
faced a systemic banking crisis or a currency crisis during a specific age-span. I
consider crises as defined by Laeven and Valencia (2008), who list every banking
and currency crisis for each country since 1970 onwards. Thus, in order to back up
the complete economic history of each individual, I only consider those who were 18
years old or older in 1970 and also those who are older than 21 in 2005. For each
person I back up his crises history and create a dummy variable that is equal to 1 if
the individual suffered an economic crisis at a given age span. In my ideal setting I
would like to define a cohort for each age, unfortunately the data does not allow me
to work with such a disaggregate level. Therefore I will consider a time spans of 4
years and define 6 dummy variables for crises at different ages: 18-21, 22-25, 26-29,
30-33, 34-37 and 38-41. Thus, I create one dummy variable for each period, equal
to 1 if the individual faced at least one economic crisis during that age period, and
0 otherwise. Moreover, to test the IYH I will use a 8 years time-span and define 3
dummy variables: 18-25, 26-33 and 34-415. There are two reasons that justify the
conformation of 4-years cohorts: First it allows my to compare my results directly to
4A monotonic relation between beliefs and age would imply a direct cohort effect in the formationof beliefs.
5The 8 years time span follows Giuliano and Spilimbergo (2009), as the impressionable years arethose between 18 and 25 years.
58
the ones obtained by Giuliano and Spilimbergo (2009). Second, the economic crises
databse records the date a crisis begins, but not when it ends. Since the length of
a crisis is at most 4 years a cohort of 4 years tracks the crisis database, since a fifth
year is recorded as a new crisis. Latinborometer’s survey also has information on
individual’s characteristics, such as sex, age, religion, etc, as well as information on
education and income.
Figure (II) depicts the probability that a crisis occurs in the period when the indi-
vidual is aged 22 to 25, by year of birth and country, and figures (III-VII) depict the
same graph for a crisis that occurs when the individual is aged 18 to 21, 26 to 29, 30
to 33, 34 to 37 and 38 to 41. Figures (II-VI) reveal that every country in the sample
faced at least one economic crisis, but that these crises happened at different times
across countries (with few exceptions, as the debt crisis in 1982). However countries
differ in the number of crises they experienced. For instance, countries like Argentina
experienced on average a crisis every 10 years or less, while Chile experienced crises
every 4 years until 1982 and none afterwards. Countries like Guatemala, Honduras,
El Salvador or Panama experienced only 1 economic crisis during this period. In sum,
there exists within and between country variation in the occurrence of crisis.
Finally figure (VIII) shows the probability of facing a crisis in the age periods 18-
21, 22-25, 26-29, 30-33, 34-37 and 38-41, by year of birth, for the whole sample of
countries6. A comparison from the different crisis dummies reveals that older cohorts
suffered relatively more crisis while old than when they were young. Moreover, those
who were born around the ’60 suffered on average the same number of crises over
their life.
As seen from figures (II-VIII) there is variation across countries and ages in the timing
6Due to the 1982 debt crisis which hit almost every country in Latin America, the probabilityof facing a crisis in the age period 18 to 25 is close to 1 (> 0.96) for those born in 1963 and 1964.
59
of the crises, an important fact for the empirical strategy, since (as I discuss below)
I will exploit this cross age and country variation, and justify that the estimated
correlations are not entirely due cohort effects. Table (I) shows the summary statistics
for the variables in these study: Panel (A) shows the summary statistics for the whole
sample, while panel (B) shows the summary statistics for beliefs for each country.
The main message from the summary statistics is that the sample is balance across
countries, and no country displays extreme values that might drive the results7.
My baseline model assumes a Probit model for binary response of the form:
P (y = 1|X) = G(Xβ) = p(X), (2.2.1)
where
Xβ = α + γ1 · crisis[t1 − t2]ij + xijγ2 + γ3FEj,
and G(·), is a standard normal cumulative distribution function.
The dependent variable, y, is the belief on market economy as a source of growth,
crisis refers to a dummy equal to 1 if the individual suffered an economic crisis
between ages t1 and t2, xij are controls for age, gender, income and education and FE
refers to country fixed effects. For all specification I will use Maximum Likelihood
estimation, and since G(·) is a standard normal (by assumption), β is the probit
estimator (unless otherwise stated), marginal effects are always reported. Further, I
will also test the Linear Probability model and the Logistic model.
The identification of the effect of economic crises on beliefs exploits two facts: the
variation of economic crises across countries and ages. As seen in figure (II-VII)
countries mostly experienced crises in different periods of time and at different fre-
quencies. This difference in the timing of the crises across countries allows me to
7Appendix tables (I-II) show summary statistics for the remaining variables by country.
60
identify the effect of crises on beliefs, as long as crises are not related to a particular
cohort. Also, for cohort to be comparable across countries, the age-profile of beliefs
must be the same across countries, which might be regarded as a strong assumption.
Assuming that there are no “sharp” differences between the age-profiles of different
countries is a less restrictive assumption that permits me to identify the effect by
incorporating country fixed effects. In sum, the combination of smooth difference
between age-profiles of beliefs and different timing of the crises across countries ables
me to capture the coefficient of interest8.
My baseline specification starts with none or minimum controls and later I control for
individual characteristics such that possible omitted variables that are correlated both
with economic crises and beliefs are taken into account. Since economic crises have
a large impact on the economy, on the labor supply and demand, endowments, etc,
there might be other indirect channels through which crises influence beliefs. The
included controls for individual characteristics are intended to capture these other
indirect channels and address possible omitted variable bias.
With regard to my empirical strategy there is a second issue worth discussing: cohort
effects. Beliefs may differ across ages not only because of people changing their beliefs
depending on what they lived or just due to aging, but also because new cohorts may
form systematically different beliefs, i.e. individuals may have different beliefs just
because they belong to a different cohort. It would be possible to control for this
cohort effect in a repeated cross section, but it is less obvious how to deal with it in
a single cross section of countries.
As I mention before, I rely on the variation in crises experienced at different ages in
8Further, there might exist concerns on a common macro trend to all countries, appendix table(3) shows the cross country correlation of growth rates of the GDP (PWT 6.3) for the countries inthe sample. A brief inspection of these numbers reveals that over the whole period of time underscrutiny, no common pattern can be found in the business cycles of these countries.
61
different countries, to capture the effect of crises on beliefs. Unfortunately, I can not
directly control for cohort effects at the country level, since there is little variation
within a cohort in the number of individuals who suffered a crisis in a single country.
But if the difference in the age profiles (between countries) is constant or smooth,
controlling for country fixed effects will allow me to identify the coefficient of interest.
Moreover, even if I am not able to fully control the effect of cohorts, I rely on the
equality of my estimates on several samples (different cohorts) to justify that the
effect I found it is not entirely due to a cohort effect. I will not be ruling out the
objection of cohort effects driving my results, but I will push the analysis as close as
the data allows me to minimize the probability of such an argument.
To deal with the cohort effects I proceed as follows: First I estimate equation (2.2.1)
for the whole sample, i.e. all persons greater or equal to 22 years old. Later I restrict
the sample to those greater or equal to 26, 30, 34, 38 and 42. For each subsample
I estimate the coefficient of interest, as well as the effect of economic crisis at older
ages (26, 30 and 34 years old). Finally I test for the Impressionable Years Hypothesis.
2.3 Results
Table (II) shows the basic correlation between beliefs in market economy as a source
of economic growth and economic crises at different ages. Each column is the basic
probit regression of beliefs on a dummy variable that captures economic crisis at a
given age period. Column (1) shows that suffering an economic crisis in the age period
18 to 21 has a positive correlation with beliefs (i.e. economic crisis at young ages
favors beliefs on market economy), a result that appears counter-intuitive. Columns
(2-6) show a non-significant correlation between beliefs and crises at other ages. The
sample used in these regressions differs across columns, since for every specification
62
I used all individuals for which there exists data on their history. Therefore, when
considering whether a crisis at age 30 matters, I restrict the sample to those who are
at least 30 years old.
Assuming that there may be cross-country time invariant characteristics in beliefs
might be a more reasonable assumption. Table (III) columns (1-6) repeats the re-
sults from the previous table but allowing for country fixed effects. Results differ
considerably between the two tables: crisis at ages 18-21 are not longer statistically
significant, the perverse positive sign does not show up once I control for the diverse
age profiles across countries. Further, column (2) shows that suffering an economic
crisis in the age period 22-25 appears to have a negative effect on beliefs later on in
life; experiencing a crises in other age intervals does not seem to have an effect on
beliefs.
The result in column (2), a negative effect of crises on beliefs formation at early
adulthood, is the main finding of this paper. This finding implies that experiencing
an economic crisis in the age period 22-25 reduces the probability of believing in the
market economy as a source of growth by about two percentage points, in compar-
ison to those who have not suffered such a crisis. In columns (7-12) I add to each
specification the complete past economic history9, so in each column is it possible
not only to individually test whether the coefficient is different from zero (t-value)
but also the equality of coefficients for crises experienced at different ages. The only
crisis dummy that is statistically different from 0 is (as before) crisis between 22-25
years. Tests for equality of coefficients in each column reject the null. Note, however,
that for the sub-sample of individuals older than 34 years crises between 26-29 years
is also negative and statistically significative. I am cautious in interpreting this last
result, since once I control for individual characteristics this result not longer holds.
9Whenever it is possible.
63
In table (III), the coefficient of crisis at age 22-25 ranges from -0.018 to -0.034, still
when the sample is changing and the number of observations is getting smaller, this
result persists across samples and the coefficient of interest has a nearly constant
value across the columns.
Next I address omitted variable bias. In Table (IV) I add controls for age, sex,
education, income, marital status and employment status. Columns (1-6) adds one
control at the time and in the next two columns I run the complete specification.
All regressions consider the effect of experiencing a crisis in the age period 22-2510.
Controlling for variables that are correlated with beliefs, such as age, education and
income, appear not to change the coefficient of interest. If one has the prior that
the baseline regression ails from omitted variable bias, then these results display that
OVB seems not to be an important problem. It might be the case that there is no
omitted variable bias, or that I have enough variance in my explanatory variable such
that biases are mitigated.
Columns (9-13) show that experiencing an economic crisis in the age period of 22-25
continues to be statistically significative. Further, when I add past economic history
the effect of a crisis at age 26-29 is no longer statistically significant once I control
for personal characteristics, moreover for the sub-sample of those older than 34 years,
the t-value for crisis at age 26-29 is 0.14 (while crisis at age 22-25 has in most cases
a t-value of 0.015). In what follows Table (IV) column (7) is going to be my main
specification11.
Impressionable Years Hypothesis
10Results for all other age intervals are shown in the appendix table IV.11Appendix table (IV) shows results for crisis in all other age periods excluding 22-25, none of
the crisis dummies is statistically significative.
64
Next, I proceed to test the Impressionable Years Hypothesis. According to the IYH
early adulthood is a period of mental plasticity where core attitudes, values and
beliefs are formed and remain largely unaltered throughout the remaining adult years.
Therefore, beliefs are greatly influenced by the environment people faced at that age,
which is broadly defined to be between 18 years and 25 years. Thus, using an age
span of 8 years, I define crisis-dummy variables at ages 18-25, 26-33 and 34-41.
Table (5) presents the main results, where in each column I regress beliefs on crisis
dummies. In columns (1-2) none of the crisis dummies appear to explain differ-
ences in beliefs. After restricting the sample to those older than 34 and 42, columns
(3-4), crisis at the impressionable years have an effect on beliefs, namely the prob-
ability of believing in market economy as a source of growth is lower by 4% to 5%.
Looking closely at the crisis variable, figure (8), reveals that there is not enough vari-
ation within and between countries, since the time period under consideration is long
enough to capture an economic crisis for every cohort12. When truncating the sample
to the oldest cohorts I get more variation in the independent variable, which drives
the encountered results. In sum, for the subsample of individuals 34 and 42 years
old or older, columns (3-4) show that suffering an economic crisis at ages 18-25 has
a negative impact on beliefs on market economy. Moreover, no other age-interval is
relevant for beliefs formation, as predicted by the IPH.
2.4 Robustness Checks
Addressing Cohort Effects
As discussed previously, working with data on a single cross section of countries does
12Economic crisis happened every 10 years or less for many countries in the sample.
65
not allow me to completely address concerns about possible cohort effects. However
I can perform a series of robustness checks that will reinforce my previous results.
As a first approach, I restrict my sample to the oldest people. Given a median
age in the sample of 34 years (the 25th percentile is 27 years), it might be argued
that the encountered effect of crisis on beliefs is being mostly driven by the young
population. For those aged 34 or less the crisis dummy at ages 22-25 might capture
the effect of recent crises, rather than crises at a given age. Therefore in Table (VI)
I run the complete specification restricting the sample to those older than a given
age. The regressions also include all past history on crisis, so each column adds an
additional crisis dummy at the time. Columns (1-3) shows the results for those older
than 26, columns(4-8) for those older than 34, columns(9-14) for those older than
38 and finally columns (15-21) repeats the exercise for those older than 42. There
are two important results: First, an economic crisis at age 22-25 reduces beliefs in
the market economy between 2% and 3.8%, changing the sample varies (a little) the
coefficient but continues to be statistically significative. Second, economic crises in
other age periods do not explain beliefs in market economy, even when the sample
is being changed. These results show that macroeconomic instability continues to
exert an impact on the beliefs individuals hold as far as twenty years later. Columns
(15-21) further restricts the sample to those older than 42 years. The small sample,
3325 observations, causes the standard errors to increase (by twofold) but this is not
enough to make my estimates statistically insignificant.
In sum, the results appear not to be driven by the young individuals in the sample,
it is not the recent economic history that is driving my results, but an effect of crises
during early adulthood. Limiting the sample to those older than the median age
(equal to 34 years) does not vanish the encountered effect of lower beliefs in market
economy (by 2%) for those who suffered an economic crisis while young.
66
In a second attempt to address the cohort effects I control directly for each cohort
in the regression. I create cohort dummies in the same time-intervals as the crises
dummies, i.e. there are cohort dummies for each of the age-groups: 18-21, 22-25,
26-29, 30-33, 34-37, 38-41 and 42-45 (those older than 46 are the omitted group).
Next, in Table (7) I regress beliefs on the crisis dummy at ages 22-25 and the cohort
dummies. I repeat this setting for different samples: all individuals, older than 26,
older than 30, older than 34, older than 38 and older than 42.
Results again support the proposed hypothesis: even limiting the sample to those
above the median age and directly controlling for the cohort effects does not alter the
significance of macroeconomic crisis at early adulthood. When the sample is reduced
to those 42 years old or greater, the coefficient of interest is no longer statistically sig-
nificant. As before, the standard errors almost double which might be a consequence
of working with less than 25% of the sample.
The previous exercises were centered on two pivotal issues: Fist, are young people in
the sample driving the results? Second, am I capturing pure cohort effects? With
regard to the first question, Table (VI) provides evidence that young people are not
driving my results. Limiting the sample (even by half!) to the oldest individuals
does not vanish the results. Further, directly controlling for cohort effects, as well
as removing young people from the sample, does not change my results: suffering
an economic crisis in the age period 22 to 25 has an impact on the beliefs on mar-
ket economy as a source of economic growth, in particular the suggest effect in my
estimations is a reduction of about 2% in the probability of believing in markets as
a source of growth, in comparison to those who have not experience a crisis at that age.
Linear Probability Model
67
In the previous sections, I have estimated my model using a Probit estimator. In this
section I analyze the results obtained from estimating a simple Linear Probability
Model (LPM). Estimating a LPM is appealing because of its computational easiness,
the straightforward interpretation of the results and most important as a test for bias
in my probit estimator due to the inclusion of fixed effects. Including fixed effects in
a Probit model might bias the results, as suggested by Wooldridge (2002), therefore
I test for this possibility through the comparison of my estimates between the LPM
and the probit estimator.
Given my binary response variable y, the LPM is specified as:
P (y = 1|X) = α + γ1 · crisis[t1 − t2]ij + xijγ2 + γ3FEj.
My variable of interest, economic crisis, is also a binary variable and so the coefficient
of interest is just the difference in the probability of believing in market economy as
a source of growth when the individual suffered a crisis or not (i.e. crisis = 1 or
crisis = 0). In most of the cases the LPM is not a good description of probability
model, but it can be a good approximation to the underlying response probability.
In Table (VIII) I estimate the LPM for several specifications using OLS. Column
(1) shows the basic correlation of beliefs and crisis between 22-25 years, which is
statistically equal to zero. Column (2) adds country fixed effects and column (3)
shows the complete specification controlling for individual characteristics. Notably
the sign and values obtained are similar to those obtained using a probit estimator.
Having a crisis at age 22-25 reduces the probability of believing in market economy as
a source of growth by almost 2%. Further, columns (2-3) show that adding controls
for age, education and income does not change by much the coefficient of interest.
Finally, the similarity of these results with my previous estimates suggests that fixed
68
effects are not causing any visible bias in the probit estimation.
Next, columns (4-8) repeat this exercise for crisis at other ages. The results show that
crises at ages other than 22-25 do not explain beliefs. This result further supports my
hypothesis: economic instability has it greatest effect on beliefs when they happen
while young.
Logistic Model
There is at least one reasons for estimating a logistic model: it might be a proper
assumption for G(·) to be a standard logistic distribution function, rather than a
standard normal. Table (9) column(2) shows the main specification, where crisis at
ages 22-25 is the variable of interest. In columns (1, 3-7) the variable of interest
is crisis at other ages. Two results are worth mentioning: First, crisis at ages 22-
25 continues to be statistically significative and with a coefficient of equal magnitude
than before. Second, all other crisis histories do not explain beliefs in market economy.
2.5 Conclusion
In this paper, I study how banking and currency crises impact beliefs in market
economy at different ages. I match every individual in the sample to the country’s
economic history, and analyze how crises impact the beliefs they hold. Using a cross
section of countries in Latin America I exploit cross country variation in the timing of
economic crisis and find that crises only have an impact on beliefs in market economy
when they happened in the age period 22 to 25. In particular, suffering a crisis at these
ages lowers the probability of believing in the market economy as a source of economic
69
growth by about 2%. This result has been confirmed by several robustness checks:
assuming different probabilistic models and controlling for individual characteristics
and (as much as possible) for cohort effects. Results consistently show an effect of
crises at early adulthood on beliefs and no effect when crises happened later in life.
Has the estimated effect an economic meaning? One can argue that a reduction by
2% in the probability of believing in market economy lacks strong economic conse-
quences. Besides the magnitude of the change, the main message of this paper is that
beliefs held by individuals indeed differ depending on the economic circumstances ex-
perienced during their lives, as well as how these differences in beliefs persisted over
time. I provide evidence for beliefs being influenced by economic shocks at younger
ages, and not later in life. I test for the Impressionable Years Hypothesis an find
supporting evidence for it.
This paper contributes to the literature on belief’s formation by providing evidence
on beliefs being shaped during early adulthood. Moreover, I show how economic
crises at older ages do not have an effect on beliefs. These results also shed some light
on the long term persistence of institutions, by providing empirical evidence on how
beliefs react to economic conditions and when beliefs are permeable to the economic
environment. Future research on this topic should incorporate a time series aspect
to control for the cohort effects, which might add to the understanding of beliefs
formation.
70
Figu
re I
Mea
n B
elie
fs b
y Y
ear o
f Birt
h an
d C
ount
ry
71
Figu
re II
Econ
omic
Cris
es in
the
Age
Per
iod
22-2
5, b
y Y
ear o
f Birt
h an
d C
ount
ry
72
Figu
re II
IEc
onom
ic C
rises
in th
e A
ge P
erio
d 18
-21,
by
Yea
r of B
irth
and
Cou
ntry
73
Figu
re IV
Econ
omic
Cris
es in
the
Age
Per
iod
26-2
9, b
y Y
ear o
f Birt
h an
d C
ount
ry
74
Figu
re V
Econ
omic
Cris
es in
the
Age
Per
iod
30-3
3, b
y Y
ear o
f Birt
h an
d C
ount
ry
75
Figu
re V
IEc
onom
ic C
rises
in th
e A
ge P
erio
d 34
-37,
by
Yea
r of B
irth
and
Cou
ntry
76
Figu
re V
IIEc
onom
ic C
rises
in th
e A
ge P
erio
d 38
-41,
by
Yea
r of B
irth
and
Cou
ntry
77
Figu
re V
III
Econ
omic
Cris
es b
y A
ge P
erio
ds a
nd Y
ear o
f Birt
h fo
r the
Who
le S
ampl
e
78
Variable Obs Mean Std. Dev. Min Max
Beliefs 13257 0.749 0.434 0 1Age 13257 35.183 8.914 22 53Woman 13257 0.512 0.500 0 1Education 12651 8.119 4.519 0 16Income Ladder 13021 3.698 1.794 1 10Married 13207 0.718 0.450 0 1Self-employed 13257 0.477 0.499 0 1
Crisis in the Age Period18-21 13257 0.400 0.490 0 122-25 13257 0.352 0.478 0 126-29 10896 0.358 0.480 0 130-33 8850 0.326 0.469 0 134-37 7037 0.298 0.457 0 138-41 5417 0.229 0.420 0 142-45 3500 0.212 0.409 0 1
Freq. Mean Std. Dev. Min MaxCountryArgentina 687 0.696 0.460 0 1Bolivia 786 0.790 0.408 0 1Brasil 766 0.769 0.422 0 1Colombia 868 0.781 0.414 0 1Costa Rica 659 0.783 0.413 0 1Chile 746 0.820 0.384 0 1Ecuador 799 0.762 0.426 0 1El Salvador 682 0.603 0.490 0 1Guatemala 658 0.669 0.471 0 1Honduras 666 0.758 0.428 0 1México 799 0.756 0.430 0 1Nicaragua 626 0.808 0.394 0 1Panamá 671 0.687 0.464 0 1Paraguay 799 0.721 0.449 0 1Perú 806 0.756 0.430 0 1Uruguay 690 0.842 0.365 0 1Venezuela 871 0.755 0.430 0 1República Dominicana 678 0.698 0.460 0 1
Whole Sample (A)
Summary Statistics for Beliefs by Country (B)
TABLE ISummary Statistics
79
\(1) (2) (3) (4) (5) (6)
Crisis in the Age Period18-21 0.0155**
(0.00765)22-25 -0.0101
(0.00792)26-29 0.00515
(0.00866)30-33 0.00109
(0.00983)34-37 0.000953
(0.0113)38-41 0.00416
(0.0140)
Observations 13257 13257 10896 8850 7037 5417
*** p<0.01, ** p<0.05, * p<0.1
TABLE IIBeliefs and Economic Crisis at Different Age Periods
Beliefs in Market Economy
Notes: Marginal Effects reported from a probit estimator. Robust standard errors in
80
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
Cri
sis i
n th
e Ag
e Pe
riod
18-2
10.
0043
30.
0043
30.
0031
50.
0056
4-0
.001
16-0
.009
71-0
.007
45(0
.008
17)
(0.0
0817
)(0
.008
19)
(0.0
0891
)(0
.010
1)(0
.011
7)(0
.013
9)22
-25
-0.0
180*
*-0
.017
8**
-0.0
218*
*-0
.028
2***
-0.0
370*
**-0
.033
9**
(0.0
0860
)(0
.008
62)
(0.0
0943
)(0
.010
4)(0
.012
3)(0
.014
3)26
-29
-0.0
0352
-0.0
0539
-0.0
137
-0.0
256*
*-0
.021
2(0
.009
54)
(0.0
0965
)(0
.010
9)(0
.012
6)(0
.015
2)30
-33
0.00
457
-0.0
0094
8-0
.009
11-0
.007
64(0
.010
6)(0
.011
0)(0
.012
9)(0
.015
4)34
-37
0.00
241
-0.0
0313
-0.0
0155
(0.0
127)
(0.0
133)
(0.0
160)
38-4
1-0
.010
2-0
.011
1(0
.015
9)(0
.016
8)
Obs
erva
tion
1325
713
257
1089
688
5070
3754
1713
257
1325
710
896
8850
7037
5417
TAB
LE II
IEf
fect
of E
cono
mic
Cris
is o
n B
elie
fs w
ith C
ount
ry F
ixed
Eff
ects
Bel
iefs
in M
arke
t Eco
nom
y
Not
es:
Mar
gina
l Eff
ects
repo
rted
from
a p
robi
t est
imat
or. R
obus
t sta
ndar
d er
rors
in p
aren
thes
es. C
ount
ry fi
xed
effe
cts a
re in
clud
ed in
the
regr
essi
on. *
** p
<0.0
1, *
* p<
0.05
, * p
<0.1
81
(1) (2) (3) (4) (5) (6)Crisis in the Age Period22-25 -0.0180** -0.0180** -0.0181** -0.0188** -0.0193** -0.0188**
(0.00895) (0.00860) (0.00883) (0.00869) (0.00867) (0.00861)18-21
26-29
30-33
34-37
38-41
Age 2.19e-06(0.000442)
Woman 0.0111(0.00754)
Secondary Edu -0.0271***(0.00918)
Higher Edu -0.0499***(0.0116)
Income Dummy2 -0.0449***
(0.0160)3 -0.0146
(0.0139)4 -0.0368***
(0.0140)5 -0.0237*
(0.0131)6 -0.0408**
(0.0186)7 -0.0410*
(0.0241)8 0.0308
(0.0346)9 0.184***
(0.0447)10 0.0302
(0.0575)Married 0.0117
(0.00854)Self-employed 0.0238***
(0.00776)
Observations 13257 13257 12651 13021 13207 13257
Notes: Marginal Effects reported from a probit estimator. Robust standard errors in parentheses. Country fixed effects are included in the regression. Secundary Education is a dummy variable equal to 1 if the individual completed between 7 and 12 years of education. Higher Education is a dummy variable equal to 1 if the individual completed more than 12 years of education. *** p<0.01, ** p<0.05, * p<0.1
TABLE IVEffect of Economic Crisis on Beliefs, Adding Controls
Beliefs in Market Economy
82
(7) (8) (9) (10) (11) (12) (13)
-0.0186** -0.0190** -0.0184** -0.0240** -0.0275** -0.0318** -0.0265*(0.00925) (0.00928) (0.00928) (0.00995) (0.0110) (0.0126) (0.0146)
0.00181(0.00852)
-0.00690 -0.0123 -0.0208 -0.0136(0.0110) (0.0121) (0.0140) (0.0162)
-0.00342 -0.00751 0.000791(0.0128) (0.0149) (0.0170)
0.00119 0.00749(0.0155) (0.0186)
-0.0140(0.0191)
-1.29e-05 -0.000265 -2.37e-05 0.000441 0.000190 0.000301 -0.000386(0.000460) (0.000480) (0.000462) (0.000623) (0.000861) (0.00125) (0.00171)
0.0106 0.0109 0.0106 0.0123 0.00655 0.0109 0.00696(0.00781) (0.00785) (0.00781) (0.00861) (0.00949) (0.0107) (0.0122)
-0.0244*** -0.0235** -0.0244*** -0.0250** -0.0253** -0.0262** -0.0191(0.00939) (0.00946) (0.00939) (0.0104) (0.0115) (0.0129) (0.0148)
-0.0472*** -0.0435*** -0.0472*** -0.0532*** -0.0559*** -0.0736*** -0.0765***(0.0120) (0.0121) (0.0120) (0.0133) (0.0149) (0.0170) (0.0196)
-0.0369** -0.0379** -0.0369** -0.0327* -0.0215 -0.0156 -0.0183(0.0163) (0.0164) (0.0163) (0.0176) (0.0189) (0.0207) (0.0235)-0.00884 -0.00836 -0.00888 -0.00387 0.00393 0.0160 0.0214(0.0142) (0.0142) (0.0142) (0.0154) (0.0166) (0.0181) (0.0204)-0.0277* -0.0256* -0.0277* -0.0269* -0.0245 -0.0184 -0.00787(0.0144) (0.0144) (0.0144) (0.0157) (0.0171) (0.0189) (0.0212)-0.0106 -0.0104 -0.0107 -0.00595 0.00800 0.0234 0.0341*(0.0135) (0.0135) (0.0135) (0.0147) (0.0158) (0.0173) (0.0195)-0.0287 -0.0267 -0.0287 -0.0164 -0.00245 0.0207 0.00655(0.0190) (0.0190) (0.0190) (0.0207) (0.0224) (0.0244) (0.0284)-0.0190 -0.0200 -0.0190 -0.0212 -0.00126 -0.00775 -0.0180(0.0242) (0.0244) (0.0242) (0.0270) (0.0292) (0.0328) (0.0371)0.0433 0.0440 0.0433 0.0910*** 0.0859** 0.0874** 0.0775
(0.0348) (0.0346) (0.0348) (0.0346) (0.0384) (0.0415) (0.0493)0.219*** 0.218*** 0.219*** 0.212*** 0.200*** 0.197*** 0.186***(0.0314) (0.0318) (0.0313) (0.0393) (0.0479) (0.0526) (0.0643)0.0240 0.0248 0.0241 0.0286 -0.0138 -0.0284 -0.0264
(0.0594) (0.0593) (0.0594) (0.0688) (0.0798) (0.0932) (0.110)0.0125
(0.00929)0.0202**(0.00823)
12441 12394 12,441 10,266 8,385 6,672 5,132
Beliefs in Market Economy
Cont. TABLE IVEffect of Economic Crisis on Beliefs, Adding Controls
83
Older than Older than
34 42(1) (2) (3) (4)
Crisis in the Age Period18-25 -0.0138 -0.0152 -0.0387*** -0.0598**
(0.00916) (0.0105) (0.0142) (0.0279)26-33 -0.00201 -0.0249 -0.0338
(0.0130) (0.0170) (0.0281)34-41 -0.0157 0.00866
(0.0177) (0.0310)42-49 0.0116
(0.0319)Controls for Age, Sex,Income and Education: yes yes yes yes
Country Fixed Effects: yes yes yes yes
Observations 12,441 10,266 6,672 3,325
*** p<0.01, ** p<0.05, * p<0.1Notes: Robust standard errors in parentheses.
TABLE VTesting the Impressionable Years Hypothesis
Older than
26
84
(1)
(2)
(3)
(9)
(10)
(11)
(12)
(13)
(14)
Cri
sis i
n th
e Ag
e Pe
riod
18-2
10.
0072
00.
0053
80.
0047
6-0
.007
26-0
.011
0-0
.013
6-0
.013
8-0
.012
9-0
.012
3(0
.009
14)
(0.0
0916
)(0
.009
23)
(0.0
146)
(0.0
147)
(0.0
149)
(0.0
150)
(0.0
150)
(0.0
150)
22-2
5-0
.022
3**
-0.0
234*
*-0
.026
2*-0
.029
0**
-0.0
297*
*-0
.028
1*-0
.028
6*(0
.009
75)
(0.0
1000
)(0
.013
6)(0
.013
8)(0
.014
6)(0
.014
8)(0
.014
8)26
-29
-0.0
0622
-0.0
174
-0.0
182
-0.0
138
-0.0
158
(0.0
110)
(0.0
143)
(0.0
153)
(0.0
162)
(0.0
164)
30-3
3-0
.002
490.
0027
1-0
.000
347
(0.0
153)
(0.0
167)
(0.0
172)
34-3
70.
0129
0.00
686
(0.0
166)
(0.0
187)
38-4
1-0
.013
1(0
.019
1)
Obs
erva
tions
1026
610
266
1026
651
3251
3251
3251
3251
3251
32
TAB
LE V
IA
ddre
sing
Coh
ort E
ffec
ts, S
ampl
e of
Old
est i
ndiv
idua
ls
Old
er th
an 2
6O
lder
than
38
85
(4)
(5)
(6)
(7)
(8)
(15)
(16)
(17)
(18)
(19)
(20)
(21)
Cri
sis i
n th
e Ag
e Pe
riod
18-2
1-0
.005
56-0
.008
59-0
.011
7-0
.012
0-0
.012
00.
0032
1-0
.013
4-0
.016
9-0
.014
1-0
.014
9-0
.014
1-0
.013
9(0
.012
0)(0
.012
0)(0
.012
2)(0
.012
2)(0
.012
3)(0
.022
8)(0
.025
1)(0
.025
9)(0
.025
9)(0
.026
0)(0
.026
0)(0
.026
3)22
-25
-0.0
276*
*-0
.031
4***
-0.0
337*
**-0
.033
7***
-0.0
311
-0.0
337*
-0.0
378*
-0.0
359*
-0.0
361*
-0.0
361*
(0.0
117)
(0.0
120)
(0.0
125)
(0.0
128)
(0.0
196)
(0.0
201)
(0.0
205)
(0.0
206)
(0.0
206)
(0.0
207)
26-2
9-0
.020
3-0
.023
2*-0
.023
3-0
.010
2-0
.017
5-0
.011
5-0
.012
5-0
.012
4(0
.012
7)(0
.013
6)(0
.014
2)(0
.017
7)(0
.018
6)(0
.020
2)(0
.020
3)(0
.020
5)30
-33
-0.0
0849
-0.0
0859
-0.0
247
-0.0
165
-0.0
194
-0.0
191
(0.0
137)
(0.0
150)
(0.0
191)
(0.0
223)
(0.0
233)
(0.0
240)
34-3
7-0
.000
249
0.01
600.
0117
0.01
22(0
.015
5)(0
.022
2)(0
.024
8)(0
.026
9)38
-41
-0.0
0833
-0.0
0763
(0.0
222)
(0.0
272)
42-4
50.
0012
3(0
.028
3)
Obs
erva
tions
6672
6672
6672
6672
6672
3325
3325
3325
3325
3325
3325
3325
Not
es:
Mar
gina
l Eff
ects
repo
rted
from
a p
robi
t est
imat
or. R
obus
t sta
ndar
d er
rors
in p
aren
thes
es. C
ount
ry fi
x ef
fect
s and
indi
vidu
al's
char
acte
ristic
s are
incl
uded
in
***
p<0.
01, *
* p<
0.05
, * p
<0.1
Old
er th
an 4
2O
lder
than
34
Con
t. TA
BLE
VI
86
(1) (2) (3) (4) (5) (6)Crisis in the Age Period22-25 -0.0185* -0.0233** -0.0250** -0.0254** -0.0227* -0.0252
(0.00959) (0.0101) (0.0109) (0.0122) (0.0136) (0.0180)Cohort Dummies22-25 0.101
(0.0623)26-29 0.0741 0.0731
(0.0560) (0.0607)30-33 0.0723 0.0714 0.0869*
(0.0463) (0.0501) (0.0519)34-37 0.0564 0.0554 0.0681 0.0756*
(0.0378) (0.0405) (0.0419) (0.0454)38-41 0.0524* 0.0527* 0.0624** 0.0680** 0.0566
(0.0287) (0.0304) (0.0316) (0.0340) (0.0372)42-45 0.0158 0.0161 0.0223 0.0247 0.0172 0.0206
(0.0221) (0.0230) (0.0234) (0.0245) (0.0259) (0.0278)
Observations 12441 10266 8385 6672 5132 3325
*** p<0.01, ** p<0.05, * p<0.1
TABLE VIIAddresing Cohort Effects, Adding Cohort Dummies
Notes: Marginal Effects reported from a probit estimator. Robust standard errors in parentheses.
87
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Cri
sis i
n th
e Ag
e Pe
riod
22-2
5-0
.010
1-0
.017
8**
-0.0
180*
*(0
.007
92)
(0.0
0844
)(0
.009
06)
18-2
10.
0037
4(0
.008
35)
26-2
9-0
.001
24(0
.010
6)30
-33
0.00
531
(0.0
117)
34-3
70.
0073
7(0
.014
0)38
-41
-0.0
175
(0.0
165)
Con
trols
for A
ge, S
ex,
Inco
me
and
Educ
atio
n:no
noye
sye
sye
sye
sye
sye
s
Cou
ntry
Fix
ed E
ffec
ts:
noye
sye
sye
sye
sye
sye
sye
s
Obs
erva
tions
1325
713
257
1244
112
441
1026
683
8566
7251
32
***
p<0.
01, *
* p<
0.05
, * p
<0.1
TAB
LE V
III
Line
ar P
roba
bilit
y M
odel
Not
es:
Rob
ust s
tand
ard
erro
rs in
par
enth
eses
.
88
(1) (2) (3) (4) (5) (6) (7)Crisis in the Age Period18-21 0.00373
(0.00849)22-25 -0.0185**
(0.00925)26-29 -0.00101
(0.0107)30-33 0.00533
(0.0117)34-37 0.00739
(0.0140)38-41 -0.0183
(0.0172)42-45 0.00351
(0.0228)
Observations 12,441 12,441 10,266 8,385 6,672 5,132 3,325
*** p<0.01, ** p<0.05, * p<0.1
TABLE IXLogistic Model
Notes: Robust standard errors in parentheses. Logistic estimation.
89
Vari
able
Freq
.M
ean
Std.
Dev
.M
inM
axFr
eq.
Mea
nSt
d. D
ev.
Min
Max
Cou
ntry
Arg
entin
a68
737
.26
9.23
2253
687
0.53
0.50
01
Bol
ivia
786
35.1
18.
7422
5378
60.
520.
500
1B
rasi
l76
635
.30
8.92
2253
766
0.52
0.50
01
Col
ombi
a86
834
.57
8.51
2253
868
0.52
0.50
01
Cos
ta R
ica
659
35.6
68.
7822
5365
90.
530.
500
1C
hile
746
37.0
09.
0122
5374
60.
520.
500
1Ec
uado
r79
935
.09
9.06
2253
799
0.50
0.50
01
El S
alva
dor
682
34.0
49.
1422
5368
20.
490.
500
1G
uate
mal
a65
834
.21
8.78
2253
658
0.51
0.50
01
Hon
dura
s66
634
.35
8.93
2253
666
0.49
0.50
01
Méx
ico
799
34.3
88.
7622
5379
90.
520.
500
1N
icar
agua
626
33.5
58.
7422
5362
60.
500.
500
1Pa
nam
á67
135
.51
8.71
2253
671
0.50
0.50
01
Para
guay
799
35.3
18.
8222
5379
90.
510.
500
1Pe
rú80
634
.47
8.81
2253
806
0.53
0.50
01
Uru
guay
690
36.8
39.
0922
5369
00.
490.
500
1V
enez
uela
871
35.8
78.
7622
5387
10.
510.
500
1R
epúb
lica
Dom
inic
ana
678
34.6
48.
8322
5367
80.
510.
500
1
Age
Wom
an
App
endi
x TA
BLE
ISu
mm
ary
Stat
istic
s by
Cou
ntry
90
Vari
able
Freq
.M
ean
Std.
Dev
.M
inM
axFr
eq.
Mea
nSt
d. D
ev.
Min
Max
Cou
ntry
Arg
entin
a68
79.
873.
580
1668
74.
361.
601
10B
oliv
ia78
69.
464.
760
1678
63.
711.
491
8B
rasi
l76
65.
024.
690
1476
63.
661.
801
10C
olom
bia
868
8.17
4.52
016
868
3.67
1.87
110
Cos
ta R
ica
659
7.91
3.91
016
659
4.32
1.92
110
Chi
le74
611
.09
3.04
016
746
4.17
1.43
18
Ecua
dor
799
8.51
4.06
016
799
3.41
1.54
19
El S
alva
dor
682
6.41
4.81
016
682
3.41
1.97
19
Gua
tem
ala
658
4.80
4.52
016
658
3.93
1.60
110
Hon
dura
s66
65.
654.
470
1666
63.
161.
981
10M
éxic
o79
98.
434.
410
1679
94.
321.
821
10N
icar
agua
626
6.31
4.71
016
626
2.34
1.83
110
Pana
má
671
8.58
4.16
016
671
3.66
1.79
110
Para
guay
799
8.11
3.79
016
799
3.64
1.59
110
Perú
806
9.30
4.32
016
806
3.36
1.60
110
Uru
guay
690
9.20
3.39
016
690
3.96
1.49
19
Ven
ezue
la87
19.
004.
040
1687
13.
961.
741
10R
epúb
lica
Dom
inic
ana
678
8.15
4.63
016
678
3.29
2.09
110
Educ
atio
nIn
com
eC
ont.
App
endi
x TA
BLE
I
91
Cri
sis i
n th
e Ag
e Pe
riO
bsM
ean
Std.
Dev
.O
bsM
ean
Std.
Dev
.O
bsM
ean
Std.
Dev
.O
bsM
ean
Std.
Dev
.C
ount
ryA
rgen
tina
687
0.68
90.
463
687
0.72
30.
448
687
0.67
90.
467
687
0.59
20.
492
Bol
ivia
786
0.44
90.
498
786
0.33
20.
471
786
0.36
80.
483
786
0.25
50.
436
Bra
sil
766
0.74
80.
434
766
0.62
40.
485
766
0.67
70.
468
766
0.63
10.
483
Col
ombi
a86
80.
379
0.48
586
80.
294
0.45
686
80.
327
0.46
986
80.
367
0.48
2C
osta
Ric
a65
90.
478
0.50
065
90.
466
0.49
965
90.
500
0.50
065
90.
357
0.48
0C
hile
746
0.35
80.
480
746
0.20
80.
406
746
0.15
10.
358
746
0.02
00.
140
Ecua
dor
799
0.33
50.
472
799
0.23
30.
423
799
0.28
60.
452
799
0.33
30.
472
El S
alva
dor
682
0.21
00.
407
682
0.18
60.
390
682
0.18
70.
390
682
0.22
50.
418
Gua
tem
ala
658
0.14
60.
353
658
0.06
80.
253
658
0.09
50.
294
658
0.12
50.
331
Hon
dura
s66
60.
134
0.34
166
60.
129
0.33
666
60.
095
0.29
466
60.
144
0.35
1M
éxic
o79
90.
394
0.48
979
90.
313
0.46
479
90.
308
0.46
279
90.
221
0.41
5N
icar
agua
626
0.33
10.
471
626
0.26
70.
443
626
0.22
20.
416
626
0.20
20.
402
Pana
má
671
0.15
50.
362
671
0.14
60.
353
671
0.10
80.
310
671
0.12
70.
333
Para
guay
799
0.54
80.
498
799
0.52
20.
500
799
0.58
80.
493
799
0.58
70.
493
Perú
806
0.38
50.
487
806
0.28
00.
449
806
0.22
70.
419
806
0.17
80.
383
Uru
guay
690
0.47
70.
500
690
0.43
50.
496
690
0.53
10.
499
690
0.37
20.
484
Ven
ezue
la87
10.
513
0.50
087
10.
582
0.49
387
10.
561
0.49
787
10.
548
0.49
8R
. Dom
inic
ana
678
0.37
20.
484
678
0.45
70.
499
678
0.38
00.
486
678
0.42
00.
494
App
endi
x TA
BLE
IISu
mm
ary
Stat
istic
s by
Cou
ntry
18-2
122
-25
26-2
930
-33
92
Cri
sis i
n th
e Ag
e Pe
riO
bsM
ean
Std.
Dev
.O
bsM
ean
Std.
Dev
.O
bsM
ean
Std.
Dev
.C
ount
ryA
rgen
tina
687
0.64
20.
480
687
0.51
20.
501
687
0.59
00.
493
Bol
ivia
786
0.23
30.
423
786
0.14
20.
350
786
0.07
80.
269
Bra
sil
766
0.60
00.
491
766
0.48
50.
501
766
0.37
30.
485
Col
ombi
a86
80.
128
0.33
486
80.
206
0.40
586
80.
332
0.47
2C
osta
Ric
a65
90.
336
0.47
365
90.
236
0.42
665
90.
085
0.28
0C
hile
746
0.00
00.
000
746
0.00
00.
000
746
0.00
00.
000
Ecua
dor
799
0.33
30.
472
799
0.25
50.
437
799
0.35
70.
480
El S
alva
dor
682
0.16
30.
370
682
0.00
00.
000
682
0.00
00.
000
Gua
tem
ala
658
0.03
20.
177
658
0.00
00.
000
658
0.00
00.
000
Hon
dura
s66
60.
147
0.35
566
60.
046
0.20
966
60.
000
0.00
0M
éxic
o79
90.
208
0.40
779
90.
270
0.44
579
90.
144
0.35
2N
icar
agua
626
0.12
40.
330
626
0.06
80.
253
626
0.00
00.
000
Pana
má
671
0.09
90.
299
671
0.00
00.
000
671
0.00
00.
000
Para
guay
799
0.55
90.
497
799
0.50
30.
501
799
0.45
90.
499
Perú
806
0.09
70.
297
806
0.00
00.
000
806
0.00
00.
000
Uru
guay
690
0.40
00.
490
690
0.34
70.
477
690
0.28
70.
453
Ven
ezue
la87
10.
595
0.49
187
10.
436
0.49
787
10.
407
0.49
2R
. Dom
inic
ana
678
0.44
80.
498
678
0.33
00.
471
678
0.33
10.
472
38-4
142
-45
34-3
7
Con
t. A
ppen
dix
TAB
LE II
Sum
mar
y St
atis
tics b
y C
ount
ry
93
12
34
56
78
910
1112
1314
1516
1719
11.
002
0.21
1.00
30.
240.
241.
004
0.32
0.30
0.58
1.00
50.
530.
320.
430.
341.
006
0.31
0.18
-0.0
20.
380.
371.
007
0.18
0.47
0.55
0.48
0.20
-0.0
81.
008
0.29
0.47
0.19
0.23
0.57
0.16
0.29
1.00
90.
240.
600.
550.
500.
530.
280.
560.
281.
0010
0.05
0.32
0.20
0.55
0.05
0.21
0.28
0.12
0.31
1.00
110.
050.
360.
150.
220.
100.
130.
37-0
.19
0.52
0.23
1.00
120.
080.
020.
11-0
.12
0.05
-0.0
80.
110.
020.
06-0
.31
-0.0
21.
0013
0.07
0.08
0.21
0.16
-0.0
9-0
.02
0.24
-0.2
70.
20-0
.01
0.41
0.40
1.00
14-0
.09
0.09
0.18
0.34
0.11
0.14
0.27
-0.1
60.
400.
420.
32-0
.42
0.06
1.00
150.
390.
200.
480.
290.
140.
060.
240.
100.
190.
050.
110.
280.
34-0
.18
1.00
160.
580.
320.
270.
520.
340.
400.
150.
140.
220.
250.
38-0
.07
0.34
0.22
0.40
1.00
170.
490.
340.
330.
450.
260.
240.
480.
360.
230.
280.
110.
010.
14-0
.09
0.37
0.44
1.00
190.
270.
130.
300.
250.
320.
050.
140.
190.
390.
220.
230.
150.
060.
020.
240.
150.
201.
00
App
endi
x TA
BLE
III
Cro
ss C
ount
ry c
orre
latio
n: R
eal G
DP
Gro
wth
Rat
eC
ount
ry
94
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
Cri
sis i
n th
e Ag
e Pe
riod
18-2
10.
0034
80.
0071
2-0
.000
280
-0.0
0703
-0.0
0727
0.00
560
(0.0
0849
)(0
.009
17)
(0.0
104)
(0.0
122)
(0.0
148)
(0.0
237)
26-2
9-0
.001
76-0
.000
942
-0.0
0485
-0.0
116
-0.0
0644
-0.0
0179
(0.0
107)
(0.0
107)
(0.0
118)
(0.0
135)
(0.0
157)
(0.0
197)
30-3
30.
0057
60.
0042
40.
0045
40.
0113
-0.0
104
(0.0
118)
(0.0
124)
(0.0
140)
(0.0
160)
(0.0
234)
34-3
70.
0077
90.
0070
90.
0124
0.01
90(0
.014
0)(0
.015
3)(0
.018
4)(0
.026
6)38
-41
-0.0
184
-0.0
111
-0.0
0420
(0.0
171)
(0.0
191)
(0.0
270)
42-4
50.
0036
00.
0053
5(0
.022
6)(0
.027
9)In
com
e an
d Ed
ucat
ion:
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
Cou
ntry
Fix
ed E
ffec
ts:
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
Obs
erva
tions
12,4
4110
,266
8,38
56,
672
5,13
23,
325
10,2
668,
385
6,67
25,
132
3,32
5N
otes
: R
obus
t sta
ndar
d er
rors
in p
aren
thes
es**
* p<
0.01
, **
p<0.
05, *
p<0
.1
App
endi
x TA
BLE
IVEf
fect
of E
cono
mic
Cris
is w
ith C
ontro
ls a
nd F
ixed
Eff
ects
Chapter 3
Population Composition and Human Capital Cre-
ation: the Raise in Education in the U.S.
3.1 Introduction
The United States preceded all other nations in the provision of secondary education
to its citizens. Beginning in the twentieth century and until 1940 the U.S. saw an
unprecedented increase in access to secondary education and its citizens acquired
levels of education that were not achieve until years later by other industrializing
countries (Goldin (1994, 1998), Goldin and Katz (1998, 2008)).
During the nineteenth and twentieth century the U.S. saw a constant flow of immi-
grants, mostly from Europe. These newly arrived immigrants helped populate the
U.S. and changed the composition of the population. The settlements’ patterns of
these immigrants provided for differences in the origin of individuals in each commu-
nity.
This study answers the question of how the composition of the population affected
school enrollment at the beginning of the twentieth century. The provision of educa-
95
96
tion, like other public goods, can largely depend on the ability of a group of individuals
to agree on its provision (Alesina et al. 1999), and individuals’ diversity can greatly
affect these decisions. In general, heterogeneity in preferences will make it less likely
for individuals to agree on the value of a public good, with respect to its cost.
One important dimension in which diversity can be viewed is the country of birth.
Individuals born to a same country have an array of common characteristics and share
the same language, culture, and in many cases even religion. In this study I analyze
the composition of population that resulted from the European mass migration to
the U.S.. I will measure diversity by the country of birth of individuals, study the
implications of having a population of varying origins versus another where individuals
were born in the same country.
Ideally I would like to look at the ability of a community to provide education to its
members, but the data only provides me with the number of individuals in a given
community that decided to attend any school. Thus I can not observe the decision to
provide education by a community (i.e. the decision to built schools, hire teachers,
etc.), I observe whether members of the community attend any school.
A community may have a lower share of its school-age-population attending school
for more than one reason, it can be related to demand of education, the supply of
education, or both. The opportunity cost of attending school can differ for different
communities, making it more or less attractive for young individuals to acquire ed-
ucation. On the other hand the community itself may, or may not, be able to agree
on providing education and provide for the resources needed to built schools and hire
teachers.
First I will show how attendance to school varies across counties depending on indi-
vidual’s characteristics. In particular I will study how the decision to attend school
97
is affected when individuals are born in a foreign country, or when their parents are
born in a foreign country. Children born in the U.S. to foreign parents may also
decide differently about schooling, with respect to children born to U.S. parents.
After showing results at the individual level, I will analyze enrollments rates at the
county level. The analysis at the county level allows me to understand heterogeneity
from a different perspective. In particular, besides studying the effect of the share of
foreign born population on enrollment rates, I will compute a fractionalization index
that measures diversity in the country of birth.
This paper is organize as follows, section 2 describes the data and the summary
statistics, section 3 presents the empirical results for the individual level data and for
the county level data. Finally, section 4 concludes.
3.2 Data
Individual level data comes from a representative sample of census records from
IPUMS. Following the analysis of Goldin (1994, 1998) I use data from 1900, 1910,
1920 and 1930. Census records provides information on age, sex, race, country of
birth, country of birth of the parents, school attendance and occupation, and the
place where the individual is living.
Individual data is collapse at the county level for the county level analysis. I com-
plement it with data on county characteristics from ICPSR. County characteristics
include share of individuals born in a foreign country, share of females, share of
population in rural areas, average urban population in the county and the share of
manufacturing output.
98
The analysis focuses on secondary-school-age population, namely individuals aged 14
to 17. The census data does not specify the type of school an individual is attending,
therefore I can not distinguish those attending secondary school from those attending
elementary school. Focusing on those aged 14 to 17 narrows the group of individuals
that may be attending secondary school, but does not assure that is the case. In this
respect, results for a broaden population, individuals aged 7 to 17, are also shown.
I will consider two samples: first I consider all individuals for which there is in-
formation available for these years. Second I will focus on individuals living in New
England-, Middle Atlantic-, West North Central- and Pacific-regions, following Goldin
(1998).
Starting in the nineteenth century and spanning into the twentieth century the U.S.
received a large flow of migration, mostly from countries in Europe. Figure 1 shows
a time series of the number of immigrants arriving to the U.S. for each year, the
time series starts in 1820 with 8400 immigrants arriving and continues until past
the First World War, achieving the highest number in 1907 with 1.3 million arrived
immigrants. On average 333,000 immigrants arrived every year between 1820 and
1920. Unfortunately there are no records on the net number of immigrants, since
those turning back to their home countries were not counted.
Table 1 summarizes for each nationality the share of immigrants, for the year 1910.
Nationals of Germany, Austria, Soviet Union, Denmark, Ireland, Italy, Canada and
United Kingdom represent the highest shares of foreigners in the U.S., summing up
to 13% of the population. Column 3 specifies, for each nationality, the maximum
share that they achieved in any county, ranging from 15% for citizens from France,
up to 80% for citizens from Switzerland.
Table 2 shows summary statistics for the share of population enrolled in any school,
99
the share of foreign born population, the share of female population, the share of
African-Americans and the share of the population living in rural areas, for the years
1900, 1910, 1920 and 1930. Summary statistics are shown for the samples of individ-
uals aged 14 to 17 in all counties. On average enrollment rates increase over time,
while the share of individuals born outside the U.S. decreases over time. The share
of African Americans is slightly lower in later years, while the share of individuals
living in rural areas also decreases over time.
3.3 Results
3.3.1 Individual Level Data
For the empirical analysis I start investigating the determinants of school attendance
for the secondary-school-age population across counties. My dependent variable is a
dummy variable that equals one if the individual attends any school, zero otherwise.
I will run a regression of the school attendance dummy on individual characteristics,
as represented by the following linear equation:
yi = α + βforeign+Xiγ + ηc + εi, (3.3.1)
where yi is a dummy variable equal to one if the individual is attending any school,
zero otherwise. foreign is a dummy variable equal to one if the individual was born
outside the U.S., Xi is a set of individual characteristics including age, sex, race, and
two dummy variables that are equal to one if the father or the mother were born in
a foreign country.1 Controls also include whether the location is a rural area and the
1The race dummy variable equals one if the individual reported being black.
100
urban population. County fixed effects are included, η, and in all specifications, and
standard errors are clustered at the county level.
Table 3 reports regression results for equation (3.3.1), column 1 shows marginal effects
from a probit regression and column 2 results from OLS.2 Results in columns 1 and 2
do not vary much, suggesting that a linear probability model does not pose a strong
assumption. In columns 3 and 4 I repeat the analysis but expanding the sample to
all individuals age 7 to 17, potentially capturing enrollment at the elementary and
high-school level. Coefficients do not vary between columns 3 and 4, suggesting again
that a linear probability model is not a strong assumption.
Table 3 columns 1-4 show that there is a negative effect of being a foreigner when
attending school. Females and younger individuals are more likely to attend school,
while African-Americans attend less school. Having a father or a mother born in a
foreign country also hinders the likelihood of attending any school. Living in a rural
area appears to have a different effect when considering older students, rather than
every school-age individual. Columns 1 and 2 show a positive effect of living in rural
areas, while when all individuals age 7 to 17 are considered, the effect is negative.
In particular, the variable of interest: individuals born in a foreign country, has a
negative effect on schooling, consistent across specifications. Comparing the coeffi-
cients between columns 1 and 3 or 2 and 4 shows that including young individuals in
elementary-school-age tends to lower the coefficient on the foreign dummy. Individ-
uals born abroad tend to go less to school, but even more at older ages. Females are
also less likely to attend school at older ages, with respect to males.
Table 4 reports in columns 1-4 coefficients from OLS regressions, where each column
represents a different year: 1900, 1910, 1920 and 1930, respectively. The effect of
2In these regressions the sample comprises all individuals aged 14 to 17 (high-school age popu-lation) in 1910. These regressions do not include county fixed effects.
101
being a foreigner is negative and significant. A foreigner is on average 20% less likely
to attend any school. The effect is not constant across years, in early years children
born in a foreign country attend less school, than in later years.
In column 5 and 6 I consider a repeated cross-section of children age 14 to 17, con-
trolling for year fixed effects and in column 6 also for the interaction between the year
dummies and foreign. Column 5 shows that on average the effect of being born in a
foreign country is negative and significant. Moreover, the year dummies show that
for later years the likelihood of attending any school is higher.3 Column 6 adds the
interaction between being a foreigner and the year dummies. The interactions show a
striking result, the negative effect of being born outside the U.S. tends to be smaller
for later years. In particular, an individual age 14 to 17 in 1930 are 18% more likely
of attending any school than an individual in the same age thirty years before.
Tables 5 and 6 repeat this same analysis for the sample of individuals age 7 to 17.
Table 5 shows similar results to those in table 4, the coefficient on the foreign dummy is
negative and significant but of a smaller magnitude. Table 6 adds a dummy variable
for those individuals age 7 to 13. As expected, the dummy variable for younger
individuals shows that they more likely to attend any school. Consistent with the
results in table 3, including in the sample those in elementary-school-age tends to
lower the coefficient on foreign.
Table 7 repeats the analysis of table 3 but only for individuals living in New England-,
Middle Atlantic-, West North Central- and Pacific-regions, following the regions cov-
ered in Goldin (1994, 1998). Results are qualitatively the same to table 4. Individuals
born overseas are on average 26% less likely to attend any school. The interaction
between the year-dummies and foreign shows the same result as before, individuals
in 1930 are 17% more likely to attend school than an individual in 1900.
3Year dummies are not shown in the table.
102
In sum, consistent with the results presented by Goldin (1994, 1998) and Goldin and
Katz (1998) tables 3-7 show that attendance to any school is less likely for foreign
individuals, and more likely for females and younger individuals. For individuals
living in rural areas the evidence is mixed, for early years there is a positive effect of
rural areas on school attendance, but in later years I not longer find this result.
3.3.2 County Level Data
For the county level analysis I collapsed the individual level data at the county level,
and complemented it with data on county characteristics from ICPSR.4 For each
county I compute the attendance rate, the share of young individuals born in a foreign
country, the share of females, the share of individuals living in rural places and the
mean urban population. County characteristics also include the share of manufacture
in total output.5
I compute an index of fractionalization to measure diversity in the country of birth.6
Fractionalization measures the probability that two randomly drawn individuals from
a given county were born in different countries. The index is defined as:
fractc = 1−N∑i=1
π2i,c,
where πi,c is the share of the county population born in country i. The N countries
of birth include all countries specified in the census, except for the following cases:
individuals born in the U.S. are divided between Afro-Americans and non-Afro Amer-
icans, individuals born in England, Scotland and Wales are referred to as born in the
4I collapse the data at the county level for those individuals aged 13 to 17.5The share of manufacture is defined as manufacture output divided by the sum of manufacturing
and farm output, in 1900 U.S. dollars.6See Easterly and Levine (1997), Fearon (2003).
103
United Kingdom, individuals born in Austria and Hungary are referred to as born in
the Austro-Hungarian Empire, and individuals born in Norway, Sweden and Finland
are referred to as born in Scandinavia.
Similar to the preceding analysis, I run the following specification:
yc = α + βforeign+ δfract+Xcγ + ηs + ψt + εc, (3.3.2)
where foreign is the share of individuals born in a foreign country, fract is the frac-
tionalization index, Xc are county characteristics, ηs are state fixed effects and ψt are
time fixed effects.
First I start by estimating equation (3.3.2) without including the fractionalization
index. Table 8 shows results replicating the exercise of the previous section at the
county level, columns 1 to 4 each represent a different year: 1900, 1910, 1920 and
1930, respectively, and column 5 pools all years together. Results are similar to those
in table 4, individuals born in a foreign country are less likely to attend any school,
one standard deviation (s.d.) in foreign (8%) decreases enrollment rates by 2.2%, or
0.11 s.d.. In general, by aggregating at the county level, some information is lost, but
results are qualitatively the same.
Next, table 9 shows results when the fractionalization index is included as an inde-
pendent variable. The effect of fractionalization varies across years, it is negative and
significant for 1900 and 1910 (columns 1 and 2), but its effect is less clear for later
years (columns 3 and 4). In table 10 column 1 I consider a repeated cross-section
including year dummies, under this specification fractionalization has a negative and
significant coefficient. The coefficient implies that one s.d in fractionalization (0.17)
would decrease the enrollment rate by 1.6%, or 8% of a s.d..
104
The negative effect of fractionalization on enrollment rates shows how relevant diver-
sity is for the understanding of enrollment decisions. Not only the share of foreign
born individuals help to explain differences in enrollment rates, but also how het-
erogeneous a society is. Diversity plays an important role in explaining why some
communities are more successful in providing education to their children. In diverse
communities not only foreign individuals are less likely o attend school, every young
individual is less likely to attend school.
Column 2 adds to the regression the interactions between fractionalization and the
years dummies, in order to capture how the negative effect of fractionalization varies
over time. Consistent with the results in table 9, fractionalization has a negative and
significant coefficient, and the interactions shows that this effect diminishes over time.
One standard deviation in fractionalization decreases enrollment rates by 3.3% (16%
of a s.d.), but this effect is almost 0 in 1930 (with respect to 1900).
Next I examine whether the negative effect of fractionalization differs between rural
and urban counties. To the specification above I also include the interaction of frac-
tionalization with the share of the population living in rural places. Column 3 shows
results for this regression. As expected fractionalization continues to be negative and
significant, as well as th share of foreign population. What is remarkable is the coeffi-
cient on the fractionalization-rural interaction, it is positive and significant (at 10%).
Raising fractionalization by one s.d. lowers enrollment rates in all counties, but less
in rural counties. Fractionalization has less implications for enrollment rates in rural
than in urban counties.
Finally, table 11 replicates results from table 9 and 10, but only for the sample of
individuals living in the regions analyzed by Goldin (1998). As it was the case in the
previous section with individual level data, results do not vary much when restricting
to sample to this regions, though the coefficient for fractionalization increases in
105
absolute value. The interaction between fractionalization and the year dummies also
shows that the negative effect of a more diverse society diminishes over time. Column
6 shows that with this constraint sample the share of foreign individuals is no longer
significant. The negative effect of diversity is only captured by the fractionalization
index.
This county-level analysis provides evidence for the understanding of diversity and
its implication for education and human capital formation. Heterogeneous societies,
as measured by the country of birth, were less successful in providing education for
their children. Not only children born outside the U.S. were less likely to attend any
school, but every children living in these diverse communities were less likely to attend
school. The long-run implications of these differences in enrollment rates appear to
be mitigated by the fact that this effect diminishes over time.
3.4 Concluding Remarks
The beginning of the twentieth century was both a period of rapid rise in educational
attainment and of mass migration from Europe to the U.S.. The result on develop-
ment of these two forces is an interesting topic of study that has been under constant
analysis by researchers.
In this study I further contribute to the understanding of the differences in school
enrollment across communities and provide for a novel explanation. The process of
migration and settlement generated heterogeneous communities, people born in differ-
ent countries, that speak many languages and have different culture and backgrounds
started to live together and to participate in collective decisions: provision of public
goods and in particular education. Diversity, as measure by an index of fractionaliza-
106
tion in the country of origin, had a negative effect on enrollment rates. Individuals
aged 14 to 17 were 20% less likely to attend any school during this period. Diverse
communities had on average lower enrollment rates.
These negative effects from foreigners and heterogeneous communities tend to dissi-
pate over time. A foreign individual in 1930 was 18% more likely to attend school
than a foreigner in 1900. The effect of fractionalization in 1930 was almost negligible,
in comparisson to the effect in 1900.
Diversity can have multiple effects on development. This research shows the short-
and medium-run effects on enrollment rates in a period were education played a key
role in the economy. Further research is needed in order to understand the different
mechanisms through which diversity affects the process of economic growth.
107
0
200,
000
400,
000
600,
000
800,
000
1,00
0,00
0
1,20
0,00
0
1,40
0,00
0
1820182318261829183218351838184118441847185018531856185918621865186818711874187718801883188618891892189518981901190419071910191319161919
108Table 1Year: 1910
Germany 2.71% 25%Austria 1.78% 26%USSR 1.66% 50%Denmark 1.52% 50%Ireland 1.48% 18%Italy 1.41% 42%Canada 1.35% 30%United Kingdom 1.35% 25%Mexico 0.26% 75%France 0.14% 15%Swizterland 0.14% 80%Japan 0.14% 44%Netherlands 0.13% 22%Less than 1/1000 0.73% -
Share in Total Population
Max Share in Any County
109Table 2
Variable year Mean Std. Deviation Obs.
1900 0.47 0.50 615371910 0.67 0.47 729881920 0.68 0.47 773511930 0.72 0.45 93145
Foreign 1900 0.06 0.241910 0.05 0.221920 0.05 0.211930 0.03 0.16
Female 1900 0.50 0.501910 0.50 0.501920 0.50 0.501930 0.50 0.50
African-American 1900 0.13 0.331910 0.12 0.331920 0.11 0.321930 0.11 0.31
Rural area 1900 0.65 0.481910 0.59 0.491920 0.56 0.501930 0.51 0.50
Attending any School
110Table 3
Dependent Variable:Year: 1910 Probit OLS Probit OLSSample:
(1) (2) (3) (4)
Foreign -0.254*** -0.228*** -0.129*** -0.123***(0.009) (0.008) (0.004) (0.005)
Female 0.030*** 0.028*** 0.012*** 0.013***(0.004) (0.003) (0.002) (0.002)
Age -0.148*** -0.137*** -0.032*** -0.034***(0.002) (0.001) (0.000) (0.000)
African-American -0.226*** -0.195*** -0.227*** -0.206***(0.006) (0.006) (0.003) (0.003)
Foreign Father -0.061*** -0.055*** -0.015*** -0.015***(0.006) (0.006) (0.003) (0.002)
Foreign Mother -0.071*** -0.064*** -0.015*** -0.017***(0.006) (0.006) (0.003) (0.003)
Rural 0.063*** 0.057*** -0.015*** -0.008***(0.004) (0.004) (0.002) (0.002)
Urban Population -0.000*** -0.000*** -0.000*** -0.000***(0.000) (0.000) (0.000) (0.000)
Constant 2.815*** 1.273***(0.021) (0.004)
Observations 72,988 72,988 202,699 202,699Adjusted R-squared 0.149 0.117Robust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
Attending any School
Age: 14 - 17 Age: 7 - 17
111Table 4: Individuals aged 14 to 17
Dependent Variable:Year: 1900 1910 1920 1930 1900-1930 1900-1930
(1) (2) (3) (4) (5) (6)
Foreign -0.276*** -0.215*** -0.124*** -0.096*** -0.185*** -0.266***(0.012) (0.013) (0.011) (0.010) (0.009) (0.013)
Foreign * 1910 0.039***(0.011)
Foreign * 1920 0.128***(0.011)
Foreign * 1930 0.183***(0.016)
Female 0.041*** 0.028*** 0.043*** 0.003 0.027*** 0.027***(0.004) (0.004) (0.004) (0.003) (0.002) (0.002)
Age -0.129*** -0.138*** -0.162*** -0.151*** -0.146*** -0.146***(0.002) (0.004) (0.003) (0.003) (0.002) (0.002)
African-American -0.175*** -0.151*** -0.110*** -0.095*** -0.126*** -0.127***(0.010) (0.009) (0.008) (0.008) (0.005) (0.005)
Foreign Father -0.066*** -0.061*** -0.049*** -0.043*** -0.050*** -0.050***(0.008) (0.006) (0.007) (0.006) (0.004) (0.004)
Foreign Mother -0.087*** -0.065*** -0.059*** -0.031*** -0.054*** -0.054***(0.008) (0.007) (0.006) (0.006) (0.004) (0.004)
Rural 0.019*** 0.013** -0.021*** -0.062*** -0.011*** -0.012***(0.007) (0.006) (0.005) (0.005) (0.003) (0.003)
Urban Population -0.000*** -0.000*** -0.000 -0.000*** 0.000*** 0.000***(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Constant 2.530*** 2.863*** 3.212*** 3.118*** 2.762*** 2.770***(0.029) (0.059) (0.046) (0.041) (0.032) (0.033)
Year Dummies no no no no yes yesObservations 61,537 72,988 77,351 93,145 305,021 305,021Number of Clusters 2,744 2,902 3,003 3,048 3,116 3,116Adjusted R-squared 0.121 0.138 0.171 0.156 0.172 0.172
Attend any School
Notes: Individual-level regressions. Standard errors clustered at theCounty-level. Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.
112Table 5: Individuals aged 7 to 17
Dependent Variable:Year: 1900 1910 1920 1930 1900-1930 1900-1930
(1) (2) (3) (4) (5) (6)
Foreign -0.230*** -0.130*** -0.091*** -0.068*** -0.128*** -0.190***(0.008) (0.007) (0.007) (0.007) (0.005) (0.007)
Foreign * 1910 0.058***(0.009)
Foreign * 1920 0.081***(0.010)
Foreign * 1930 0.119***(0.009)
Female 0.019*** 0.013*** 0.017*** 0.004*** 0.012*** 0.012***(0.002) (0.002) (0.002) (0.002) (0.001) (0.001)
Age -0.026*** -0.035*** -0.039*** -0.032*** -0.033*** -0.033***(0.002) (0.001) (0.001) (0.001) (0.001) (0.001)
African-American -0.171*** -0.150*** -0.101*** -0.073*** -0.121*** -0.121***(0.007) (0.006) (0.005) (0.005) (0.004) (0.004)
Foreign Father -0.028*** -0.027*** -0.019*** -0.014*** -0.018*** -0.018***(0.005) (0.003) (0.003) (0.003) (0.002) (0.002)
Foreign Mother -0.038*** -0.025*** -0.022*** -0.010*** -0.021*** -0.021***(0.005) (0.003) (0.003) (0.003) (0.002) (0.002)
Rural -0.013*** -0.007** -0.020*** -0.028*** -0.015*** -0.015***(0.005) (0.003) (0.002) (0.002) (0.002) (0.002)
Urban Population -0.000*** -0.000*** -0.000* -0.000*** 0.000 0.000(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Constant 1.001*** 1.283*** 1.349*** 1.267*** 1.061*** 1.064***(0.019) (0.017) (0.011) (0.008) (0.012) (0.012)
Year Dummies no no no no yes yesObservations 178,111 202,699 229,724 264,394 874,928 874,928Number of Clusters 2,793 2,935 3,051 3,080 3,131 3,131Adjusted R-squared 0.050 0.106 0.140 0.092 0.131 0.131
Attend any School
Notes: Individual-level regressions. Standard errors clustered at the County-level. Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.
113Table 6: Individuals aged 7 to 17
Dependent Variable:Year: 1900 1910 1920 1930 1900-1930 1900-1930
(1) (2) (3) (4) (5) (6)
Foreign -0.229*** -0.129*** -0.090*** -0.065*** -0.127*** -0.189***(0.008) (0.007) (0.007) (0.007) (0.005) (0.007)
Foreign * 1910 0.057***(0.009)
Foreign * 1920 0.081***(0.010)
Foreign * 1930 0.122***(0.009)
Female 0.018*** 0.013*** 0.017*** 0.004*** 0.012*** 0.012***(0.002) (0.002) (0.002) (0.002) (0.001) (0.001)
Age 0.018*** -0.011*** -0.017*** -0.014*** -0.007*** -0.007***(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Dummy Age 7 to 13 0.351*** 0.185*** 0.182*** 0.144*** 0.204*** 0.204***(0.008) (0.005) (0.004) (0.004) (0.003) (0.003)
Control Variables yes yes yes yes yes yesYear Dummies no no no no yes yesObservations 178,111 202,699 229,724 264,394 874,928 874,928Number of Clusters 2,793 2,935 3,051 3,080 3,131 3,131Adjusted R-squared 0.090 0.123 0.160 0.104 0.150 0.150
Attend any School
Notes: Individual-level regressions. Standard errors clustered at the County-level. Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.
114Table 7: Individuals aged 14 to 17
Dependent Variable:Year: 1900 1910 1920 1930 1900-1930 1900-1930
(1) (2) (3) (4) (5) (6)
Foreign -0.269*** -0.212*** -0.108*** -0.087*** -0.173*** -0.263***(0.013) (0.014) (0.011) (0.011) (0.010) (0.014)
Foreign * 1910 0.053***(0.011)
Foreign * 1920 0.160***(0.011)
Foreign * 1930 0.169***(0.016)
Female 0.034*** 0.013** 0.037*** -0.016*** 0.014*** 0.014***(0.006) (0.005) (0.005) (0.004) (0.003) (0.003)
Age -0.148*** -0.170*** -0.186*** -0.158*** -0.166*** -0.166***(0.003) (0.004) (0.003) (0.004) (0.003) (0.003)
African-American -0.083*** -0.055** -0.038* -0.044** -0.049*** -0.049***(0.024) (0.024) (0.019) (0.018) (0.011) (0.011)
Foreign Father -0.072*** -0.066*** -0.049*** -0.038*** -0.052*** -0.052***(0.009) (0.008) (0.008) (0.007) (0.005) (0.005)
Foreign Mother -0.091*** -0.074*** -0.064*** -0.029*** -0.058*** -0.058***(0.009) (0.008) (0.007) (0.007) (0.005) (0.005)
Rural 0.024*** 0.023*** -0.028*** -0.060*** -0.006 -0.007(0.009) (0.008) (0.008) (0.006) (0.005) (0.005)
Urban Population -0.000*** -0.000*** -0.000 -0.000*** 0.000*** 0.000***(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Constant 2.858*** 3.360*** 3.580*** 3.268*** 3.056*** 3.069***(0.045) (0.069) (0.057) (0.070) (0.038) (0.039)
Year Dummies no no no no yes yesObservations 29,784 35,461 37,497 47,789 150,531 150,531Number of Clusters 776 782 786 783 797 797Adjusted R-squared 0.154 0.190 0.212 0.181 0.217 0.218
Attend any School
Notes: Individual-level regressions. Sample of counties included in Goldin (1998). Standard errors clustered at the County-level. Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1.
115Table 8
Dependent Variable:year: 1900 1910 1920 1930 1900-1930
(1) (2) (3) (4) (5)
Foreign -0.301*** -0.356*** -0.207*** -0.092 -0.269***(0.055) (0.083) (0.074) (0.093) (0.039)
Female 0.125*** 0.056** 0.071*** 0.003 0.072***(0.033) (0.027) (0.027) (0.028) (0.015)
Rural -0.010 0.009 -0.008 -0.070*** -0.020**(0.024) (0.018) (0.020) (0.017) (0.009)
Urban Population -0.000*** -0.000*** -0.000*** -0.000** -0.000***(0.000) (0.000) (0.000) (0.000) (0.000)
-0.115*** -0.089*** -0.151*** -0.063** -0.104***(0.024) (0.018) (0.037) (0.026) (0.011)
Constant 0.517*** 0.726*** 0.720*** 0.781*** 0.548***(0.032) (0.025) (0.026) (0.025) (0.014)
Year Dummies no no no no yesObservations 2,677 2,595 2,834 2,547 10,653Adjusted R-squared 0.195 0.229 0.134 0.192 0.310Notes: County-level regressions. Standard errors clustered at the State-level.Robust standard errors in parentheses*** p<0.01, ** p<0.05, * p<0.1
Share of Manufature Output
Enrollment Rates
116Table 9
Dependent Variable:year: 1900 1910 1920 1930
(1) (2) (3) (4)
Fractionalization -0.154*** -0.139*** -0.040 -0.049*(0.033) (0.028) (0.029) (0.027)
Foreign -0.241*** -0.281*** -0.189** -0.073(0.058) (0.091) (0.076) (0.093)
Female 0.126*** 0.058** 0.071*** 0.003(0.033) (0.027) (0.027) (0.028)
Rural -0.021 -0.002 -0.011 -0.074***(0.024) (0.018) (0.020) (0.017)
Urban Population -0.000*** -0.000*** -0.000** -0.000(0.000) (0.000) (0.000) (0.000)
-0.092*** -0.072*** -0.144*** -0.060**(0.024) (0.018) (0.037) (0.026)
Constant 0.560*** 0.766*** 0.731*** 0.794***(0.034) (0.026) (0.027) (0.026)
Year Dummies no no no noObservations 2,677 2,595 2,834 2,547Adjusted R-squared 0.203 0.239 0.134 0.193
*** p<0.01, ** p<0.05, * p<0.1
Enrollment Rates
Share of Manufature Output
Notes: County-level regressions. Standard errors clustered at the State-level. Robust standard errors in parentheses
117Table 10
Dependent Variable:(1) (2) (3)
Fractionalization -0.093*** -0.190*** -0.154***(0.015) (0.026) (0.037)
Fract * 1910 0.046(0.030)
Fract * 1920 0.187***(0.031)
Fract * 1930 0.166***(0.030)
Fract * Rural 0.070*(0.042)
Foreign -0.227*** -0.213*** -0.224***(0.040) (0.040) (0.041)
Female 0.072*** 0.072*** 0.072***(0.015) (0.015) (0.015)
Rural -0.028*** -0.023** -0.050***(0.009) (0.010) (0.016)
Urban Population -0.000** -0.000*** -0.000(0.000) (0.000) (0.000)
-0.093*** -0.082*** -0.093***(0.011) (0.012) (0.011)
Constant 0.577*** 0.596*** 0.596***(0.015) (0.016) (0.018)
Year Dummies yes yes yesObservations 10,653 10,653 10,653Adjusted R-squared 0.313 0.317 0.313
*** p<0.01, ** p<0.05, * p<0.1
Enrollment Rates
Share of Manufature Output
Notes: County-level regressions. Standard errors clustered at the State-level. Robust standard errors in parentheses
118Table 11
Dependent Variable:year: 1900 1910 1920 1930 1900-1930 1900-1930
(1) (2) (3) (4) (5) (6)
Fractionalization -0.228*** -0.203*** -0.124** -0.094* -0.160*** -0.248***(0.068) (0.058) (0.054) (0.056) (0.030) (0.049)
Fract * 1910 0.039(0.051)
Fract * 1920 0.137***(0.053)
Fract * 1930 0.184***(0.054)
Foreign -0.039 -0.238*** 0.075 -0.116 -0.087 -0.059(0.103) (0.090) (0.070) (0.127) (0.067) (0.068)
Female 0.153** 0.097* 0.130*** 0.067 0.122*** 0.121***(0.064) (0.051) (0.049) (0.050) (0.029) (0.029)
Rural -0.016 0.032 0.013 -0.074*** -0.021 -0.013(0.037) (0.026) (0.038) (0.027) (0.016) (0.016)
Urban Population -0.000*** -0.000 -0.000 -0.000 -0.000 -0.000**(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
-0.147*** -0.071** -0.122* -0.021 -0.126*** -0.109***(0.041) (0.031) (0.073) (0.049) (0.020) (0.021)
Constant 0.613*** 0.756*** 0.699*** 0.804*** 0.608*** 0.615***(0.056) (0.045) (0.054) (0.048) (0.027) (0.028)
Year Dummies no no no no yes yesObservations 771 763 774 734 3,042 3,042Adjusted R-squared 0.243 0.282 0.173 0.157 0.376 0.379
*** p<0.01, ** p<0.05, * p<0.1
Enrollment Rates
Share of Manufature Output
Notes: County-level regressions. Standard errors clustered at the State-level. Sample of counties included in Goldin (1998). Robust standard errors in parentheses
Bibliography
Chapter 1
[1] Acemoglu, D., S. Johnson, and J.A. Robinson (2001): “The colonial originsof comparative development: an empirical investigation,” American EconomicReview, 91:1369–401.
[2] Acemoglu, D., S. Johnson, and J.A. Robinson (2002): “Reversal of fortunes:geography and institutions in the making of the modern world income distribu-tion,” Quarterly Journal of Economics, 117:1231–94.
[3] Banerjee, A., and L. Iyer (2005): “History, institutions and economic perfor-mance: the legacy of colonial land tenure systems in India,” American EconomicReview 95: 1190–213.
[4] Borjas, George J. (1994): “The economics of immigration,” Journal of EconomicLiterature 32 (4): 1667-717.
[5] Card, David (1990): “The impact of the Mariel boatlift on the Miami labormarket,” Industrial and Labor Relations Review 43 (2): 245-57.
[6] Cruzate, G., L. Gomez, M. J. Pizarro, P. Mercuri, S. Banchero, G. Angueira,D. Prieto, J. Lopez and G. Ibarraza (1990): “Suelos de la Republica Argentina,Version 1.0.” Version digital corregida y ajustada mediante tecnicas actuales deingenierıa SIG en base a la informacion original vectorizada a partir de los mapasde suelos provinciales que integran el Atlas de Suelos de la Republica Argentina,INTA.
[7] Dell, M., (2010): “The Persistent Effects of Peru’s Mining Mita,” Econometrica78(6), 1863–1903.
[8] Direccion General de Inmigracion (1925): Resumen Estadıstico del MovimientoMigratorio en la Republica Argentina. Anos 1857 - 1924. Buenos Aires: TalleresGraficos del Ministerio de Agricultura de la Nacion.
119
120
[9] Easterly, W., and R. Levine(2009): “The European Origins of Economic Devel-opment,” mimeo, Brown University.
[10] Engerman, S.L., and K.L. Sokoloff (1997): Factor endowments, institutions, anddifferential paths of growth among New World economies: a view from economichistorians of the United States. In How Latin America Fell Behind, ed. S Harber,pp. 260–304. Stanford, CA: Stanford Univ. Press.
[11] Engerman, S.L., and K.L. Sokoloff (2002): Factor endowments, inequality, andpaths of development among New World economies. NBER, Working Paper 9259.
[12] Gallo, E. (1983): La Pampa gringa: la colonizacion agrıcola en Santa Fe (1870-1895). Buenos Aires: Editorial Sudamericana, 1983.
[13] Gallup, John Luke, and Jeffrey Sachs D. with Andrew D. Mellinger (1998):“Geography and Economic Development,” Annual World Bank Conference onDevelopment Economics 1998 (April), The World Bank: Washington, DC.
[14] Gallup, John Luke and Jeffrey D. Sachs (2001): “The Economic Burden ofMalaria,” The Supplement to The American Journal of Tropical Medicine &Hygiene Vol. 64, no. 1, 2, January/February.
[15] Galor, O. and David N. Weil (1999):“From Malthusian Stagnation to ModernGrowth,” American Economic Review, 89, 150-154.
[16] Galor, O. and David N. Weil (2000):“Population, Technology and Growth: FromMalthusian Stagnation to the Demographic Transition and Beyond,” AmericanEconomic Review, 90, 806-828.
[17] Galor, O., O. Moav and D. Vollrath (2009): “Inequality in land ownership, theemergence of human capital promoting institutions and the great divergence,”Review of Economic Studies 76: 143–79.
[18] Gennaioli, N., R. La Porta, F. Lopez-de-Silanes and A. Shleifer (2011): “HumanCapital and Regional Development,” NBER Working Paper No. 17158.
[19] Glaeser, E.L., R. La Porta, F. Lopez-De-Silanes and A. Shleifer (2004): “Doinstitutions cause growth?” Journal of Economic Growth, 9: 271–303.
[20] Lacoste, Pablo (2002): “La guerra de los mapas entre Argentina y Chile: unamirada desde Chile,” Historia (Santiago), ISSN 0717–7194, vol. 35, pages 211–249.
[21] Luna, Felix (1993): “Breve Historia de los Argentinos,” Buenos Aires: Planeta.
[22] Mellinger, Andrew D., Jeffrey D. Sachs and John L. Gallup (2000): “Climate,Coastal Proximity, and Development,” Oxford Handbook of Economic Geogra-phy. Clark, Gordon L., Maryann P. Feldman, and Meric S. Gertler, eds. OxfordUniversity Press.
121
[23] National Oceanic and Atmospheric Administration (NOAA) and U.S. NationalGeophysical Data Center, TerrainBase, release 1.0 (CD-ROM), Boulder, Col-orado.
[24] Nunn, Nathan (2009): “The Importance of History for Economic Development,”The Annual Review of Economics, 1:65–92.
[25] Nunn, Nathan and D. Puga (2012): “Ruggedness: The blessing of bad geographyin Africa”, Review of Economics and Statistics, 94(1), 20-36.
[26] Putterman, L. and D. Weil (2010): “Post-1500 Population Flows and the LongRun Determinants of Economic Growth and Inequality”, Quarterly Journal ofEconomics 125(4): 1627-1682.
[27] Republica Argentina (1938): Censo Industrial de 1935. Buenos Aires: DireccionGeneral de Estadıstica de la Nacion.
[28] Republica Argentina (1898): Segundo Censo de la Republica Argentina, Mayo10 de 1895. Buenos Aires: Taller Tipografico de la Penitenciaria Nacional.
[29] Republica Argentina (1916): Tercer Censo Nacional de la Republica Argentina,Levantado el 1ro de Junio de 1914. Buenos Aires: Talleres Graficos de L. J. Rossoy Cıa..
[30] Sachs, Jeffrey D. and Pia Malaney (2002): “The Economic and Social Burden ofMalaria,” Nature Insight, Vol. 415, no. 6872, Feb. 7.
[31] Somoza, J. L., and A. E. Lattes (1967): Muestras de los dos primeros censosnacionales de poblacion, 1869 y 1895. Buenos Aires: Instituto Torcuato Di Tella,Centro de Investigaciones Sociales, Documento de Trabajo no 46.
[32] Tell, S. (2008): Cordoba rural, una sociedad campesina (1750-1850). BuenosAires: Prometeo Libros, Asociacion Argentina de Historia Economica, 2008.
[33] Walther, J. C. (1964): La Conquista del Desierto. Buenos Aires, Cırculo Militar,1934.
Chapter 2
[34] Alesina, A., Glaeser, E., Sacerdote, B., (2001): “Why doesn’t the United Stateshave a European style welfare state?,” Brookings papers on Economic Activity,187 – 277 (Fall).
[35] Alesina, A. and E. La Ferrara (2005): “Preference for Redistribution in the Landof Oportunities,” Journal of Public Economics 89, 897 – 931.
122
[36] Alesina, A. and G. Angeletos (2005): “Fairness and Redistribution,” AmericanEconomic Review, Vol. 95 No. 4.
[37] Benabou, R. and E. Ok (2001): “Social mobility and the demand for redistri-bution: The POUM hypothesis,” The Quarterly Journal of Economics, 116(2):447-487.
[38] Di Tella, R., S. Galiani and E. Schargrodsky (2007): “The Formation of Be-liefs: Evidence from the Allocation of Land Titles to Squatters,” The QuarterlyJournal of Economics, February.
[39] Di Tella, R., Javier Donna, and Robert J. MacCulloch (2007): “Oil, MacroVolatility and Crime in the Determination of Beliefs in Venezuela,” MIMEO.
[40] Fong, C. (2000): “Social preferences, self-interest, and the demand for redistri-bution,” Journal of Public Economics.
[41] Giuliano, P. and A. Spilimbergo (2009): “Growing up in a Recession: Beliefsand the Macroeconomy,” IZA, DP No. 4365.
[42] Laeven, L. and F. Valencia (2008): “Systemic Banking Crises: A new Database,”IMF Working Papers, WP/08/224.
[43] Piketty, T. (1995): “Social Mobility and Redistributive Politics,” The QuarterlyJournal of Economics 110, 551-584.
[44] Roland, G. (2004): “Understanding Institutional Change: Fast-Moving andSlow-Moving Institutions,” Studies in Comparative International Development,38(4), 109 – 131.
[45] Tabellini, G. (2007): “Institutions and Culture,” Presidential Lecture, EuropeanEconomic Association, August.
[46] Wooldridge, J. M. (2002): Econometric Analysis of Cross Section and PanelData. MIT Press Books, The MIT Press.
Chapter 3
[47] Alesina, Alberto, Reza Baqir and William Easterly (1999): “Public Goods andEthnic Divisions,” The Quarterly Journal of Economics, November.
[48] Easterly William and Ross Levine (1997): “ Africa’s Growth Tragedy: Policiesand ethnic Divisions,” The Quarterly Journal of Economics, November.
[49] Fearon, James (2003): “Ethnic and Cultural Diversity by Country,” Journal ofEconomic Growth, 8.
123
[50] Goldin, Claudia (1994): “How America Graduated from High School: 1910 to1960,” NBER, Working Paper No. 4762.
[51] Goldin, Claudia (1998): “America’s Graduation from High School: The Evolu-tion and Spread of Secondary Schooling in the Twentieth Century,” The Journalof Economic History, Vol. 58, No. 2.
[52] Goldin, Claudia and Lawrence F. Katz (1998): “Human Capital and SocialCapital: The Rise of Secondary Schooling in America, 1910 to 1940,” NBER,Working Paper No. 6439.
[53] Goldin, Claudia and Lawrence F. Katz (2008): The race between education andtechnology. Cambridge, Mass.: Belknap Press of Harvard University Press.