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.