ABSTRACT
Title of Dissertation: CAFFEINATED DEVELOPMENT AND
OTHER ESSAYS IN LATIN AMERICAN
ECONOMIC HISTORY
Mateo Uribe-Castro
Doctor of Philosophy, 2020
Dissertation Directed by: Professor John J. Wallis
Department of Economics
This dissertation consists of three essays. The first one focuses on Colom-
bia after 1850 and measures the impact of the expropriation of Church?s assets
on political violence. With yearly data on the number of battles per municipal-
ity, archival information on the reform, and difference-in-differences, the paper
documents a reduction of political violence in places where the Church?s assets
were expropriated. The paper contests the traditional idea of the expropriation of
Church?s real estate as a source of political violence. It highlights changes in polit-
ical competition after the alliance between Conservative factions and the Church
was weakened. Specifically, it shows the reduction in political violence was con-
centrated inmunicipalities with high political competition andwhere the Conser-
vative Party was relatively weak.
The second essay studies the effect of the first wave of globalization on de-
veloping countries? structural transformation, using data fromColombia?s expan-
sion of coffee cultivation. Counties engaged in coffee cultivation in the 1920s de-
veloped a smaller manufacturing sector by 1973 than comparable counties, de-
spite starting at a similar level in 1912. My empirical strategy exploits variation in
potential coffee yields, and variations in the probability to grow coffee at differ-
ent altitudes. This paper argues that coffee cultivation increased the opportunity
cost of education, which reduced the supply of skilled workers, and slowed down
structural transformation. Using exogenous exposure to coffee price shocks as in-
strument, I show that reductions in cohorts? educational attainment led to lower
manufacturing activity in the long-run. The effect is driven by both a decrease in
demand for education and reductions in public goods. Finally, coffee cultivation
during the early 20th Century had negative long-run effects on both individual
incomes and poverty rates.
The third essay explores how changes in commodities? prices can have dif-
ferential effects on school enrollment according to characteristics of crop?s pro-
duction functions. It compares schooling outcomes in counties that specialize in
sugar (a land intensive crop with economies of scale) or coffee (mostly produced
in small farms) in Puerto Rico between 1900 and 1930. Sugar price increases lead
to increases in enrollment in sugar counties, while coffee price changes have a
negative relationship with enrollment in coffee regions.
CAFFEINATED DEVELOPMENT AND OTHER ESSAYS
IN LATIN AMERICAN ECONOMIC HISTORY
by
Mateo Uribe-Castro
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2020
Advisory Committee:
Professor John J. Wallis (Chair)
Professor Allan Drazen
Professor Sebastian Galiani
Professor Ethan Kaplan
Professor Mark Lichbach (Dean?s Representative)
?c Copyright by
Mateo Uribe-Castro
2020
Preface
Countries in Latin America became independent around the same period
than British colonies in North America. Though standards of living were rela-
tively similar between them around 1800, Latin America had fallen behind the
United States by the beginning of the 20th century (Acemoglu et al., 2002; Prados
de la Escosura, 2007). To explain these differences in economic performance, some
influential theories propose that the answer lies on the relationship between in-
stitutions and factor endowments (geography, climate, etc.). Colonial institutions
that persisted over time created an environment that did not promote innovation
and investments in physical and human capital (Engerman and Sokoloff, 1997;
Acemoglu et al., 2005; Coatsworth, 2008). Other hypothesis, perhaps more pop-
ular within Latin America, highlight the interaction between endowments and
international trade. The region had a bad draw in what Bulmer-Thomas (2003)
calls ?the lottery of natural resources? and ended up specializing in primary com-
modities, neglecting industrialization (Cardoso and Faletto, 1979). Whatever the
case may be, these theories are somewhat silent or vague on the specific mecha-
nisms connecting institutions, endowments, and international trade to economic
performance over the long run.
This dissertation presents empirical evidence from three different settings
to provide nuance and context to development theories about why Latin America
fell behind. First, the institutional theory implies that replacing colonial institu-
tions that benefitted a narrow elite for more inclusive ones, though hard, would
ii
have generated incentives to invest and innovate. However, the struggle for in-
stitutional change generated widespread political violence in the region during
the 19th century. How did Latin American countries move toward more stable
polities? Chapter 1 studies a particularly contested institutional change, property
rights for the Catholic Church, and suggests that in fact the move towards more
stability comes from institutional changes that benefit the elite (an idea related
to North, Wallis, and Weingast (2009)). In other words, better outcomes can be
achieved even when inclusive institutions are not sustainable.
Second, the dependency theory explains that international trade prevented
Latin American countries from industrializing because of incentives to special-
ize in primary commodities. While this theory emphasizes the role of terms of
trade, Chapter 2 explores how abundance of a demanded commodity prevented
industrialization through a more fundamental mechanism: reductions in human
capital accumulation. Together with Chapter 3, they show that the cost of attend-
ing school increased when workers had the option to work in profitable export
agriculture sector, especially when crops are amenable to employ child labor.
Though there are some common threads across the three settings, perhaps
the most salient common theme between the three chapters is the focus on rig-
orous empirical work and causal identification. Each of the chapters uses newly
digitized historical data and combines them with available data from census of
population. Chapter 1 uses archival data and compiles information on political
violence at the local level from a book from the early 20th century. For Chapter 2,
I digitized early coffee cultivation census and historical census of population. Fi-
iii
nally, Chapter 3 uses data from digitized government reports.
Chapter 1 studies a specific institutional change in late 19th century Colom-
bia: the expropriation of the Catholic Church?s real estate in 1861. The Church?s
land was then sold in auctions, which meant that rich members of society bought
it and consolidated their landholdings. Commentators at the time and current day
historians thought the expropriation increased political violence through retalia-
tion by Conservative factions allied with the Church against the Liberal govern-
ment. However, I show that the disentailment reform reduced political violence,
especially in counties (municipios) where the Conservative party had electoral ad-
vantage. I compare counties where the government expropriated at least one of
the Church?s real estate properties with municipalities where the Church did not
own real estate under the mortmain land tenure. Before 1862, political violence
evolved similarly in both types of municipalities but in places where the Catholic
Church?s real estate was expropriated the level of political violence decreased af-
ter the disentailment reform. One common theme across Latin American coun-
tries during the 19th century was the high frequency of civil wars and political
violence. For instance, while the United States had one civil war, Colombia had
around 16 (Centeno, 1997). This chapter shows that getting rid of colonial insti-
tutions can promote stability precisely because the institutions that replace them
are not egalitarian in nature.
Chapter 2 focuses on Colombia?s expansion of coffee cultivation during the
first two decades of the 20th century. The four-fold coffee production expansion
between 1905 and 1921 is comparable to the largest expansion of modern agri-
iv
cultural exports (palm oil in Indonesia (Edwards, 2019)). The main finding is
that coffee cultivation had a negative impact on the structural transformation of
Colombian counties during the 20th century. However, identifying the causal re-
lationship between coffee cultivation and structural transformation is challenging.
For example, regions that hadmore difficulty importing capital goods might have
seen a profitable opportunity in coffee bean production since it was transportable
bymules. More generally, counties thatwould not have developed a strongmanu-
facturing sector in the early 20th centurymight have taken up coffee cultivation as
an alternative. I use two instrumental variables approaches to deal with omitted
variable bias problem. The assumption behind both instruments is that climatic
conditions specific to coffee trees only affect industrialization through coffee cul-
tivation.
The adoption of coffee cultivation effectively increased land productivity in
places used to grow staple goods like beans and corn. Moreover, coffee production
was labor-intensive, given the need to pick and classify coffee cherries by hand.
Therefore, the expansion of coffee cultivation increased the opportunity cost of
education, shifting the type of skills workers accumulated. The chapter shows
that in coffee counties, the labor force accumulated less years of education and
specialized in agricultural skills, which slowed down the process of industrializa-
tion relative to other counties with a higher supply of workers with skills valued
outside agriculture.1 While the capacity to export primary goods to the devel-
1This argument was made theoretically by Caselli and Coleman II (2001) and Acemoglu and
Guerrieri (2008). Recently, Porzio and Santangelo (2019) presented causal evidence across coun-
tries and within Indonesia in favor of this mechanism.
v
oped world was beneficial in the short run, labor-intensive commodities gener-
ated strong incentives for households and landowners in developing countries to
reduce investments in human capital. Therefore, the supply of non-agricultural
workers decreased, causing the process of structural transformation to be slower,
and reducing income as a result.
Chapter 2 compares counties that produced an export crop, coffee, with
other counties that were not suitable to produce a demanded commodity in in-
ternational markets. Coffee counties developed a less educated labor force be-
cause coffee production increased household?s opportunity cost of investing in
children?s education. However, thatmight not necessarily be the case for all export
crops. Chapter 3 uses data from Puerto Rico between 1900 and 1930 to compare
places suitable to produce coffee and places suitable to produce sugar. Both crops
have considerably different production functions, especially in terms of economies
of scale, that translate into large differences in farm sizes. While coffee cultivation
happens in small, family farms, sugar cane cultivation takes place in larger plan-
tations. Exploiting changes in international prices of both crops, Chapter 3 shows
that increases in the price of coffee reduce enrollment in schools, in line with the
results presented in Chapter 2 for Colombia. Moreover, it shows that when sugar
price increases, school enrollment in sugar areas increases as well. In other words,
when households get relatively richer because wages are going up driven by in-
ternational prices, kids are more likely to attend school. Both effects, for coffee
and sugar regions, are stronger in rural schools relative to urban ones. Moreover,
the effect from coffee prices is as strong for women than for men, but the effect
vi
of sugar prices is concentrated on men?s enrollment. That is consistent with the
fact that women are employed more in coffee cherries? picking season than in the
sugar cane production process.
vii
Acknowledgments
Finishing this dissertation during a global pandemic has been a sobering
experience and a reminder of the importance of social science, economic history,
and rigorous empirical work. It has also put in serious perspective how privileged
I am for being able to do research for a job. That privilege does not come frommy
own ability or effort. It comes from a long series of lucky events and by invaluable
help and support from many people that took a kid from Medell?n to graduate
from one of the top economics departments in the world.
I would like to thank John Wallis, for his professional guidance, our conver-
sations, but especially for trustingmy ability to find answerswhen others only saw
dead ends; Ethan Kaplan, for pushing me to be a much more rigorous economist;
and Allan Drazen, for showing me how to be a sharp and constructive critic with-
out losing humor. I also received considerable support from Sebastian Galiani,
Felipe Saffie, and Sergio Urz?a, who were always open to new ideas and helped
me connect my interest in Latin American economic history with broader topics
within the profession. I am also grateful to Cindy Clement, John Shea, Vickie
Fletcher, Amanda Statland, and Jessica Gray, for their support during the past six
years at the University of Maryland.
I have been lucky to meet several people in the profession who have be-
come important sources of support and guidance. I would like to thank Ran
Abramitzky, Gustavo Canavire, M?nica Ospina, Andr?s Ram?rez Hassan, Jorge
A. Tamayo, Felipe Valencia, and especially Pablo Querub?n and James Robinson
viii
who unknowingly changedmy career path after I took their class in 2013 and have
provided me with generous advice over the years.
The contents of this dissertation improved significantly due to discussions
with Dan Bogart, Irina Espa?a, Leopoldo Fergusson, Martin Fizsbein, Felipe Gon-
z?lez, Judy Hellerstein, Maria L?pez-Uribe, Javier Mej?a, Kris Mitchener, Michele
Rosenberg, Jared Rubin, Felipe S?enz, Fabio S?nchez, Lesley Turner, Juan F. Var-
gas, and participants at EHA Meetings, Cliometrics, NBER-SI DAE, EH-Clio Lab
at PUC Chile, IOEA, LSE-Stanford-Uniandes Latam Development, RIDGE Eco-
nomic History, and LACEA. I gratefully acknowledge funding for assisting to
these conferences from the Gruchy Fellowship in Economics from UMD, the In-
stitute of Humane Studies, and Shealtiel Group.
I would like to thank my grad school family for being the best source of mo-
tivation, support, and lively academic discussions. Some of them are part of the
UMD Economics community: Prateik Dalmia, Juan Echenique, Camila Galindo-
Pardo, Alejandro Graziano, Fernando Saltiel, Matt Staiger, Temp, Cody Tuttle,
and Santiago V?lez. Others are not: Christian G?mez, Rafael Hern?ndez, Daniel
P?rez, Lizz Reeves, and especially Manuela V?squez. I thank you all for putting
up with my half-baked paper ideas and my inability to complete a f?tbol game
without a yellow card.
I would not have made it here without my friends from home. They bring
me perspective by not caring about rankings or top-5s and are too many to men-
tion by name. Finally, my loving family, who I need tomention by name regardless
of how many they are because I am indebted to all of them more than they can
ix
imagine. My siblings: Miguel, Sara, Susana, David, andMar?a, plus Daniel, Santi-
ago, and Pedro, who have also served as outstanding research assistants (though
I wonder what says about me as a researcher that none of them decided to study
economics). And my parents, Gabriel Uribe and Ana Mar?a Castro. This disser-
tation is dedicated to Hern?n Castro, Marielena Mej?a, and Rosa Garc?a de Uribe
(q.e.p.d.).
x
Table of Contents
Preface ii
Acknowledgements viii
Table of Contents xi
List of Tables xiii
List of Figures xiv
1 Expropriation ofChurchWealth andPolitical Conflict in 19thCenturyColom-
bia 1
1.1 Historical background . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.1.1 The Catholic Church and Political Violence . . . . . . . . . . 9
1.1.2 Disentailment reform . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Estimating the Effect of the Disentailment Reform on Political Vio-
lence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.2 Measuring the Disentailment Reform . . . . . . . . . . . . . 17
1.2.3 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3 Disentailment Reform Contributed to Reduce Political Violence . . 22
1.3.1 Political Competition as an Explanation for the Negative Ef-
fect of Disentailment Reform in Political Violence . . . . . . 27
1.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2 Caffeinated Development: Exports, Human Capital, and Structural Trans-
formation in Colombia 48
2.1 Exports and Structural Transformation in Colombia . . . . . . . . . 56
2.1.1 Coffee in Colombia: Historical Background . . . . . . . . . . 58
2.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3 Coffee Cultivation and Structural Change in Colombia . . . . . . . 67
2.3.1 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . 71
2.4 The Effect of Coffee Cultivation on Structural Transformation . . . 78
2.5 Channel: Human Capital . . . . . . . . . . . . . . . . . . . . . . . . . 82
xi
2.5.1 Growing Up During Coffee Price Booms Reduces School-
ing and Employment in Manufacturing for Cohorts Born in
Coffee Counties . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.5.2 Supply and Demand of Schooling . . . . . . . . . . . . . . . 88
2.6 Channel: Linkages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.7 Long Term Effects on Urbanization and Income . . . . . . . . . . . . 92
2.8 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3 Commodities? Production Function andHumanCapitalAccumulation: Ev-
idence from Early 20th Century Puerto Rico 115
3.1 Data and Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . 119
3.1.1 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . 123
3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.2.1 Commodity Price Shocks and School Enrollment . . . . . . 126
3.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
A Appendix to Chapter 1 137
B Appendix to Chapter 2 139
B.1 Data Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
B.2 Supporting Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
B.3 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
B.3.1 Sensitivity to Excluding Younger Cohorts . . . . . . . . . . . 146
Bibliography 148
xii
List of Tables
1.1 Comparison of Municipalities without and with Disentailment . . . 40
1.2 Correlates of the disentailment reform . . . . . . . . . . . . . . . . . 41
1.3 Difference-in-differences estimator: Effect of the disentailment re-
form on political violence . . . . . . . . . . . . . . . . . . . . . . . . 42
1.4 Continuous measures of the disentailment reform . . . . . . . . . . 43
1.5 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
1.6 Robustness checks: Dynamics . . . . . . . . . . . . . . . . . . . . . . 45
1.7 Only one pre and post period . . . . . . . . . . . . . . . . . . . . . . 46
1.8 Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.1 Coffee Cultivation and Economic Structure, 1973 . . . . . . . . . . . 104
2.2 Coffee Cultivation and Manufacturing in 1945 . . . . . . . . . . . . 105
2.3 Test for Other Discontinuities . . . . . . . . . . . . . . . . . . . . . . 106
2.4 Effect of Coffee Cultivation on Structural Transformation, 1973 . . . 107
2.5 1945 IV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
2.6 Effect of Coffee Price Shocks on Schooling by Cohort, 1973 . . . . . 109
2.7 Effect of Coffee Price Shocks on Economic Structure and Income,
1973 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
2.8 Coffee Shocks and Structural Transformation by Inequality . . . . . 111
2.9 Effect of Coffee Cultivation on Number of Schools, 1951 . . . . . . . 112
2.10 Linkages in Coffee Production and Structural Transformation, 1973 113
2.11 Effect of Coffee Cultivation on Long Term Income, 2005 . . . . . . . 114
3.1 Average Crops? Characteristics: Coffee and Sugar Production 1910. 133
3.2 Descriptive Statistics: 1910 Census and Governor Reports . . . . . . 134
3.3 Effect of Coffee and Sugar Income on Total Enrollment (Governor
Reports) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
3.4 Effect of Coffee and Sugar Income on Total Enrollment (Census Data)136
A.1 Standard errors: different clustering methods . . . . . . . . . . . . . 138
B.1 1945 IV by sector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
B.2 Effect of Coffee Price Shocks on Economic Structure, 1912, 1938,
and 2005 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
B.3 Balance Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
B.4 Coffee Shocks and Structural Transformation by Inequality . . . . . 145
B.5 Measuring 1920 Coffee Cultivation with Extensive Margin . . . . . 146
xiii
List of Figures
1.1 Number of purchases per year . . . . . . . . . . . . . . . . . . . . . 33
1.2 Share of municipalities with disentailed property by decile of Con-
servative party vote share . . . . . . . . . . . . . . . . . . . . . . . . 33
1.3 Share of municipalities with disentailed property by decile of po-
litical competition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.4 Average area disentailed in treated municipalities by decile of Con-
servative vote share . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.5 Average area disentailed in treatedmunicipalities by decile of polit-
ical competition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
1.6 Average battles per year by Disentailment Reform status . . . . . . 35
1.7 Residual average battles per year by Disentailment Reform status . 36
1.8 Placebo test: diff-in-diffs estimator (? in eq. 1.1) by cut-off year . . . 37
1.9 Empirical distribution of ? in eq. 1.1 when randomizing treatment
units and year . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.10 Empirical distribution of ? in eq. 1.1 when randomizing treatment
units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.1 Patterns of Coffee Exports and Manufacturing in Colombia . . . . . 94
2.2 Correlation Coffee Cultivation and Structural Transformation . . . 95
2.3 Coffee vs. Non-Coffee Counties Comparison . . . . . . . . . . . . . 96
2.4 Coffee Potential Yield . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2.5 Discontinuity in the Probability of Coffee Cultivation at 2,400mts . 98
2.6 Coffee Promotion Pamphlet, 1880 . . . . . . . . . . . . . . . . . . . . 98
2.7 Counties in IV and FRDD Samples . . . . . . . . . . . . . . . . . . . 99
2.8 Comparison Counties Above and Below 2,400 mts. of Altitude . . . 100
2.9 The Effect of Coffee Cultivation on Structural Transformation . . . 101
2.10 Effect of Coffee Cultivation on Industrialization by Human Capital
Requirements in 1945 . . . . . . . . . . . . . . . . . . . . . . . . . . 102
2.11 Differences in Cohorts? Schooling between Coffee and Non-Coffee
Counties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.1 Weather, Geography, and Main Crops . . . . . . . . . . . . . . . . . 128
3.2 Sugar and Coffee Areas by County . . . . . . . . . . . . . . . . . . . 129
3.3 Location of Coffee and Sugar Counties . . . . . . . . . . . . . . . . . 129
3.4 Real Coffee and Sugar Prices (1900=1) . . . . . . . . . . . . . . . . . 130
3.5 Enrollment Rates in Public Schools by Type of County . . . . . . . . 130
xiv
3.6 Land Allocated to Coffee and Sugar as Share of Cultivated Land,
1910 and 1920 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3.7 Comparing Enrollment Data from Reports and Census . . . . . . . 132
A.1 Map of the main sample . . . . . . . . . . . . . . . . . . . . . . . . . 137
A.2 Parallel trends assumption test . . . . . . . . . . . . . . . . . . . . . 138
B.1 Example from 1928 Coffee Census (Municipality of Anza, Depart-
ment of Antioquia) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
B.2 Sensitivity of results to different cohorts in sample . . . . . . . . . . 147
xv
Chapter 1: Expropriation of Church Wealth and Political Conflict
in 19th Century Colombia
Political instability and violence are one of the main causes of Latin Amer-
ica?s poor relative economic performance during the nineteenth century. Multiple
factions fought constantly over privileges, rights, and resources that became avail-
able after independence. Institutional changes shifted the balance between pow-
erful groups that frequently contested established authority (North et al., 2000;
Coatsworth, 2008; Centeno, 1997). Among these powerful groups there was the
Catholic Church. Politically influential, it also benefited from economic rents de-
fined by the land tenure system established by the Spanish Crown, known as
mortmain. Under this system the Church?s real estate was inalienable, free of
taxes, and owned in perpetuity. In this article, I explore the evolution of political
violence after an institutional reform that radically changed the Catholic Church?s
property rights: the disentailment of mortmain.
I focus on the case of Colombia, which abolished the mortmain institution
in 1861 and redistributed the Church?s real estate after 1862. Given the Church?s
influence and wealth, it is possible that the disentailment reform fueled political
A version of this chapter was published in Explorations in Economic History, vol. 73, 2019.
1
instability and violence by generating grievances between Conservative factions
allied with the Church and Liberal factions who wanted to reduce the Church?s
role in society. However, the disentailment reform may have led to less conflict if
it helped consolidate ?secular? elites or if it generated better economic outcomes.
How the expropriation of the Church?s real estate in Colombia affected political
violence in the second half of the nineteenth century is therefore an empirical
question and the main goal of this article.
I estimate the impact of the disentailment reform on political violence us-
ing archival records from the reform and a difference-in-differences estimation
strategy. Specifically, I compare municipalities (counties) where the government
expropriated at least one of the Church?s real estate properties withmunicipalities
where the Church did not own real estate under the mortmain land tenure. Be-
fore 1862, political violence evolved similarly in both types of municipalities but
in places where the Catholic Church?s real estate was expropriated the level of
political violence decreased after the disentailment reform. The effect is sizable:
political violence in the municipalities where Church land was expropriated falls
by about 29%. This effect is robust to different specifications, sample restrictions,
controls for the dynamics of political conflict, and to various standard error ad-
justments. Moreover, the disentailment reform had a negative effect on political
violence evenwhen focusing exclusively onmunicipalities where the government
expropriated the Church?s real estate.
A limitation of my approach is that there are only records of auctions of ex-
propriated properties but no estimates at the local level of the Church?s wealth be-
2
fore 1861. One concern is that the government could have targeted the Church?s
assets only in certain areas in order to gain political advantage. I use data on
the 1856 presidential elections to measure support for the Conservative Party and
electoral competition at the municipality level before the disentailment reform. I
find no correlation between political forces and the probability or extent of ex-
propriation. Moreover, the average impact of the disentailment reform on polit-
ical violence hides interesting heterogeneity. First, the disentailment reform had
a smaller, but still negative, effect on places where the Conservative Party had
widespread support compared to the Liberal party. Second, the reform was more
powerful in reducing violence in electorally contested municipalities. Given the
dynamics of political competition at the time, these results point to a political ex-
planation of why the redistribution of the Church?s real estate had a pacifying
effect during the second half of the nineteenth century. Simply, Conservative fac-
tions lost most of the appeal to support the Catholic Church?s preferred policies,
which was the item that led them to compete with Liberal factions the most. In
every other dimension, especially in economic policy, there was not much dis-
agreement between the factions. After the Church?s real estate was redistributed
to other rich members of society, the benefits of having the Church?s support di-
minished, which led to less competition. The decrease in competition did not
have much effect in municipalities which were already Conservative strongholds
or where elections were not very contested before the 1860s.
Theremight be other reasonswhypolitical violence decreased after theChurch?s
real estate was redistributed. Importantly, the disentailment reform not only auc-
3
tioned the Church?s assets off, but also changed the land tenure system estab-
lished in the Colonial period to a more modern one. Recent empirical literature
has shown that the expropriation of the Church?s wealth in Europe had economic
consequences in the sixteenth century. The dissolution of monasteries in England
(Heldring et al., 2015) and the Holy Roman Empire (Cantoni et al., 2017) affected
the long-run allocation of physical and human capital, leading to structural trans-
formation and industrialization. Later on, during the French Revolution, the re-
definition of the Catholic Church?s property rights also had a positive impact on
agricultural productivity and economic performance (Finley et al., 2017). It is
plausible that the disentailment reform in Colombia had an effect on political vi-
olence through productivity increases. While I do not have data on productivity
at the local level, I explore the economic hypothesis by estimating the relationship
between political violence and the share of a municipality?s area that was expro-
priated from the Church and changed land tenure system. I do not find empirical
support for this relationship. Even though the reform might have had effects on
productivity in the long run, the short run effect on political violence does not
seem to be driven by changes in the land tenure system.
The redefinition of the Church?s property rights was common in Europe
and the Americas during late eighteenth- and nineteenth-century. Motivated by
the French Revolution and Spain?s disentailment reform, most of the countries in
LatinAmerica carried out similar reforms in the decades after 1820 (Bazant, 2008).
Chile?s reform in the 1820s, Mexico?s in 1856, and Colombia?s in 1861 stand out as
important examples because of the central role the Church played in the politics
4
of these countries. However, most studies of the economic effects of the expropri-
ation of Church wealth in Latin America focus on the revenue collected by gov-
ernments because that was the most cited motivation for such reforms (Jaramillo
andMeisel, 2009; Bazant, 2008). This chapter studies the consequences of the dis-
entailment reform beyond fiscal dimensions.
The effect of land redistribution on violence has been studied both theoreti-
cally and empirically, especially for cases where reforms aim at solving the prob-
lem of unequal land distribution (e.g., Grossman (1994), Domenech andHerreros
(2017)). However, the Colombian disentailment reform did little to improve ac-
cess to land due to both its focus on revenue collection and the way the auction
process took place. Even though there was discussion at the time about using the
Church?s land to reduce inequality, the reform ended up only redistributing land
within elites (Fazio and S?nchez, 2010).
Precisely for that reason, the Colombian disentailment reform has tradition-
ally been viewed as a catalyst of conflict. Shortly after the disentailment decree
went into effect, an American diplomat in Colombia wrote: ?the war has virtu-
ally become one of religion, the Liberals against the Church, and the most intense
fanaticism against anything that may be proposed by them.? He added, ?when
I commenced preparing the accompanying papers for the Department [of State],
it would have appeared almost certain that the controversies to which they re-
late would soon involve the unfortunate country in another Civil War? (Shaw,
1941). That notion has been carried on to Colombia?s historiography. For instance,
Jaramillo and Meisel (2009) claim that the antagonism between the Church and
5
the Liberal party reached its peak after the 1860s. However, the relationship be-
tween the disentailment reform and political violence has not been rigorously ex-
plored.
I offer empirical support for a different interpretation: by reducing the eco-
nomic power of the Church and reallocating its real estate properties, the reform
changed the incentive of powerful groups to engage in conflict and helped lessen
political violence. In particular, factions organized in the Conservative Party lost
rewards from supporting the clergy because the Church was considerably im-
poverished. Secular elites who purchased the Church?s land and increased their
landholdings had less incentive to promote and engage in political violence after
the reform. Conflict typically increased wages, crowded out production inputs,
and made expropriation and pillage more likely (Safford and Palacios, 2002). In
otherwords, the consolidation of landholdings by secular elitesmay have reduced
their interest in violence as a profitable political strategy. This idea is related to the
view of how elites allocate rents and privileges to solve the problem of violence
(North et al., 2009) and more closely to Bazant?s (2008) depiction of Spanish dis-
entailment reform, in which landowners were sympathetic to the Church?s causes
until they bought its expropriated properties.
1.1 Historical background
After achieving independence in 1819, Colombia experienced a century of
political chaos and instability. Several factions competed for legitimate rule through
6
the electoral process. Between 1830 and 1930, 25 of the 27 presidential elections
were highly competitive, constitutional changes were frequent, and violence was
embedded in the political process (Posada-Carb?, 1995; Deas, 1996). ?The record
includes nine national civil wars, local revolts, mutinies and pronunciamientos,
material destruction equivalent to several years of economic output, and at least
250,000 deaths due to political violence? (Mazzuca and Robinson, 2009). Political
factions used violence as a tool to compete for the rents of power (Fergusson and
Vargas, 2013).
Several factors contributed to the country?s violent political competition.
First, the executive had a considerable discretion to exclude factions from the po-
litical arena. It appointed ministers and provincial governors, controlled the mili-
tary, and could allocate monopoly rights. Second, the country?s geography made
it very hard for the State to establish effective control over the territory it claimed.
Additionally, the brief experience with dictatorship in the 1820s led to a reduction
in the size of the military. Third, it was easy for political factions to raise bands of
civilians due to the country?s relative poverty.1 These guerrilla forces were often
larger than the national army (Hartlyn, 1988).
Despite their fierce competition, political factionswere relatively similar and
homogenous in their socio-economic composition. Hartlyn (1988) defines the
parties as ?loose confederations of large landowners and merchants who pos-
sessed considerable autonomy in their region rather than tightly knit organiza-
1In 1850, Colombia?s per capita GDP was around 20% of that of the US, while Argentina?s was
only 63% and Chile?s 35% of U.S. GDP (Kalmanovitz, 2011; Coatsworth, 2008).
7
tions.? Safford and Palacios (2002) describe the political elite as, ?men who were
born into the upper class and/or whose social position was confirmed by mar-
riage, through achievement in education and at the bar, in economic enterprise,
or by rising through the ranks of the military or the clergy. Most were university-
educated professionals or hadmilitary careers; in either case, theywere also likely
to own land and quite possibly also engage in commerce.? Political factions or-
ganized in the 1850s at the national level in the Conservative and Liberal parties.
However, the conventional notion of conservatives as landowners andmembers of
the military, and liberals as merchants and lawyers is not useful when describing
Colombian political and economic elites.
The similarity between the two partiesmade it relatively easy to reach agree-
ments over economic policy. Most of the liberal reforms pushed from 1845, such
as eliminating state monopolies, instituting civil marriage and universal male suf-
frage, or shifting tax revenues to regional governments, did not find organized
opposition from the Conservative Party. Even the abolition of slavery, which took
almost 30 years to complete, was peacefully resolved by compensating slave own-
ers (Tovar, 2007). The relatively peaceful way of undertaking economic reforms
led Bushnell (1993) to conclude that ?economic policy was not an area of clear-
cut differences between the parties.? Political violence frequently took place, not
because of the policies that elites took while in power, but because of the winner-
takes-all feature of the political process.
An issue that dividedpolitical factionswas their attitudes toward theCatholic
Church. Even though both members of the Liberal and Conservative parties were
8
practicing Catholics, the latter defended the central role the Church had in pre-
serving social order. Conservative elites viewed the Church as a powerful ally to
their causes. The support the Church could bring to the political process was not
only ideological but also material, which made Church related political and ideo-
logical differences ?sharp and strident? after 1850 (Safford and Palacios, 2002, p.
156).
1.1.1 The Catholic Church and Political Violence
During the conquest and colonial periods, especially in the seventeenth cen-
tury, the Catholic Church received numerous land grants and donations due to
its relationship with the Spanish Crown (Coatsworth, 2006). The Church also re-
ceived pious donations and inherited estates. By the end of the colonial period, it
was the largest landowner in the country (Fazio and S?nchez, 2010). Such prop-
erty was held in mortmain. It was inalienable, not subject to taxes, and owned in
perpetuity (Jaramillo and Meisel, 2009). It also held a monopoly on education,
controlled tithes, and was so embedded in the bureaucratic structure that becom-
ing a priest was sometimes a young man?s only available option to climb the so-
cial ladder. As a consequence, the Catholic Church had an immense influence on
Colombian society that persisted after independence.
That influence materialized in different ways. First, local priests were very
influential figures in the countryside, able to mobilize the masses against their
enemies. Second, assets held inmortmain allowed theChurch to generate revenue
9
and maintain a patron-client network. Third, the Church acted as a monopolist
on the market for mortgage loans, which it allocated to ?wealthy notables with
good political connections? (Coatsworth, 1988). Finally, it was also common for
themain political families to have representation in theCatholic Church hierarchy.
One of themost striking examples is theMosquera family. Manuel JoseMosquera
was Bogot??s Archbishop, while his older brother Tom?s Cipriano de Mosquera
served as president four times.2 The Catholic Church also had representation in
the legislative body. In 1834, for instance, one-third of the Senate and one-fourth
of the House of Representatives members were priests.
Not only was the Church involved in politics, but it also agitated the masses
to violent uprisings evenwhen its direct interests were not threatened. During the
1839-1842 civil war, which was instrumental in the formation of political parties,
James Semple, an American diplomat placed in Bogot?, described how:
?The Archbishop issued a proclamation calling on all the faithful, from the highest to
the lowest, to turn out and defend the city of the Holy Faith [Bogot?]. A solemn procession
was formed, and an oration delivered by one of the most eloquent of the clergy, closing
with a prayer to the Virgin Mary to protect the Holy City. This operation had a great
effect, many men of all classes went to the barracks and took arms? (Semple to Forthsyt,
November 21, 1840, as quoted by Shaw (1941))
The Church also participated directly in violent confrontations. President
Mosquera complained to Pope Pius IX about how several priests joined the revolu-
2Banco de la Republica: http://www.banrepcultural.org/blaavirtual/biografias/
mosqtoma.htm
10
tion, incited the masses to rebel against the constitutional government, provided
funds for weapons, and died in combat while heading guerrilla groups.3 Ortiz
(2010) also documents that ?parish priests participated in different war activities
in almost every region of the country?s interior. Bogot??s guerrillas recruited 35
priests, while in Antioquia most of the 150 priests preached, helped recruit sol-
diers, provided support both in kind and in cash to the Conservative troops, and
put together relief funds for women and children affected by violence.?4
1.1.2 Disentailment reform
General Tom?s Cipriano de Mosquera became president after taking over
Bogot? in 1861. He was once a conservative figure and had changed political
sides several times.5 Shortly after taking over the presidency, Mosquera decreed
the disentailment of the mortmain. Four sources of motivation were behind the
disentailment reform.
First was the pressing fiscal situation due to the accumulation of debt from
wars dating back to 1810 (D?az, 1977). Both Safford and Palacios (2002) and
Jaramillo and Meisel (2009) highlight the fiscal motive as being the most im-
portant. The latter document how profitable the disentailment reform was for
the government despite popular belief at the time. Second, Mosquera?s govern-
ment wanted retaliation against the Catholic Church for aligning with the Con-
3Credencial Historica Magazine, ed. 153.
4The translation is mine.
5Allan Burton, a U.S. diplomat in Colombia, commented about President Mosquera: ?He was
once the idol and worshipper of the very men, or class of men he now pursues, until he saw more
inviting fields of ambition among his ancient adversaries.?
11
servative Party during the war. Third, the government aimed at stimulating local
economies by changing the institutional framework under which those properties
were owned. Finally, there was the issue of land redistribution. The Secretary of
Finance, Rafael Nu?ez (1962) wrote: ?this is about solving with disentailment to
the greatest extent possible, the arduous and immense problem of the egalitarian
distribution of property.?6
The 1861 decree created the Disentailment Agency, which had offices in the
different states, and oversaw the operation of public auctions. The decree allowed
for payments in bonds, but a minimum of 10% of the property?s value had to be
paid in cash. It also tried to divide larger estates into smaller parcels (Fazio and
S?nchez, 2010). Finally, to prevent the Catholic Church from hiding its properties,
the Disentailment Agency offered 10% of the property?s value to informants that
denounced hidden estates.
Despite the Church?s natural resistance and logistical problems, the disen-
tailment reform progressed quickly. Figure 1.1 shows the pace of reform. By 1876,
the government estimated itwas stillmissing several properties valued in $247,000
pesos. However, it had already auctioned off properties for roughly $7million pe-
sos (Jaramillo and Meisel, 2009). Most of the auctions took place between 1862
and 1868. Seventy-eight percent of the value disentailed from 1862 to 1881 had
been auctioned off by 1868. After 1871, disentailment reform stalled because the
remaining estates had low value and the Disentailment Agency was moved to the
Secretary of the Treasury, which had different priorities. Some unsold proper-
6Quoted by Fazio and S?nchez (2010)
12
ties were given to the state governments and the rest was returned to the Church
in 1887 after the Concordato deal between the Pope and the Colombian govern-
ment had been signed. However, the reformwas not reversed during the period of
Conservative rule (1885-1898). Jaramillo andMeisel (2009) estimate that the gov-
ernment made over $8 million pesos from disentailment reform after accounting
for the annuity agreed upon with the Holy See.
The pressingfiscal situationwas alleviatedwith the reform, aswas the change
in property rights and the increase of land circulating in themarket. However, the
resulting distribution of land was far from democratic. Fazio and S?nchez (2010)
show the disentailed real estate had a higher gini coefficient than the lots traded in
1857 in Bogot?. They argue that the auctioned land went to the hands of already
powerful elites that consolidated their estates or acquired new ones in different
places.
The lower estimates for the value of the reformare around 16%ofColombia?s
1860 GDP. As a reference, the Mexican reform undertaken from 1856 to 1875 ac-
counted for 23% ofMexico?s GDP.However, Colombia?s reform valuewas thirteen
times as large as the central government?s revenue, while inMexico it was only six
times as large as federal government expenditures. The Church was stronger and
richer in Mexico than in Colombia, but the Colombian government was weaker
than Mexico?s government (Jaramillo and Meisel, 2009).
13
1.2 Estimating the Effect of the Disentailment Reform on Political
Violence
1.2.1 Data
To estimate the effect of the disentailment land reform on conflict and vio-
lence in Colombia I rely on four main data sources. First and foremost, data from
the disentailment reform from the Colombian National Archives (Archivo Gen-
eral de la Nacion) (Fazio and S?nchez, 2010). These archives contain information
on all the properties that were sold at public auction, including size, value ap-
praised by surveyors, total price paid, buyer?s name, and year of purchase. With
this source, I build the main independent variable: a dummy that equals one for
those municipalities in which land was expropriated from the Church and sold
at public auction. I also calculate the extent of the reform by measuring, at the
municipality level, the total acreage sold, total revenue collected, as well as the
original value appraised by the Disentailment Agency officials. Finally, I calcu-
late the share of the municipality?s area that was expropriated from the Church. I
use, however, a twentieth century measure of total municipality area, which will
lead to measurement error since some municipalities split and transformed their
boundaries. To alleviate these concerns, whenever I use the share of total area
measure, I will drop out of the sample municipalities with shares higher than
the 95th percentile, equivalent to 19.7% of municipality area. In this manner, I
omit municipalities where it looks like the Church lost by expropriation more
14
than 100% of the area in the twentieth century.
Second, I code the information from Riascos Grueso (1950) book ?Colom-
bia?s War Geography? to create a yearly panel with the number of battles on each
municipality from the late 18th century to 1902. This is the main dependent vari-
able. Notice it is only a measure of political violence, not of the general level of
insecurity. Even though Riascos Grueso?s book is the most comprehensive mea-
sure of political violence in the nineteenth century, it has not been widely used in
the literature.7 As a robustness check, and to deal with plausible measurement
error, I also use a dummy variable that equals one for each municipality that had
a positive number of violent confrontations in a given year. Mymain sample uses
observations from the period between the 1853 and 1886 Constitutions (1854 to
1885), but the results do not change for a sample centered at 1862, or when using
all the years from 1850 to 1900.
Third, I use the data from Bushnell (1970) on the 1856 presidential elections
to measure political support. Using data from this election is useful for three
reasons: (1) it was the first direct presidential election in Colombia; (2) the 1853
Constitution eliminated property and literacy restrictions to vote, extending the
franchise to all men older than 21 years old or younger and married; and (3) the
turnout for the election was around 41%.
In 1856, the race was decided among three candidates: Mariano Ospina, for
the Conservative Party; Manuel Murillo Toro, for the Liberal party; and Tom?s C.
7To my knowledge, only Fergusson and Vargas (2013) use the same source to study the effect
of increasing the size of the franchise in political violence.
15
de Mosquera, as an independent candidate representing his own National Party.
The election was won by the Conservative Party, and represented a relatively
peaceful transition of power from Liberal rule. Ospina won the election with 47%
of total votes. Murillo came in second place with 37.3% and Mosquera received
14.5%.8
I also calculated the level of political competition using an index ranging
from 0, where one candidate gets all the votes, to 1, where the two first candidates
split the votes evenly (Fergusson and Vargas, 2013). Let v1i , v2i be respectively the
vote shares of the winning candidate and the runner up in municipality i. The
political competition index is given by:
( )
v1 ? v2
PoliticalCompetition = 1? i ii
v1i + v
2
i
Notice it does not provide information onwhichpartywas relatively stronger.
It only measures how competitive the elections were, with 1 being the most com-
petitive.
Finally, I collected geographic and historical information for eachmunicipal-
ity from the Municipalities Panel dataset from Universidad de los Andes? Center
for Economic Development Studies (CEDE). In particular, I collected measures
of altitude, soil quality, distance to the department?s capital, distance to the main
food market, and distance to Bogot?, as well as indicators for indigenous popu-
lation after 1535, and for Spanish settlements from 1510 to 1561. I use these as
8The remaining votes (1.2%) were cast for ?other candidates.?
16
control variables.
1.2.2 Measuring the Disentailment Reform
The Colombian National Archives hold a rich section on the disentailment
reform. The General Office for Disentailed Estates had agents in the largest cities,
and their records contain useful information on the process of the reform. How-
ever, there were not preliminary estimates of the total amount of the Church land
at the municipality level (there were some at the state level), so the main measure
of disentailment is based only on the outcome of the reform, aggregating at the
municipality level the total value and area disentailed.
The Church had accumulated properties over the years, and it was willing
to defend its property rights (D?az, 1977; Coatsworth, 2006). Therefore, the re-
form?s outcome, measured either by total area or value, is not only a function of
land the Church had, but also of the capacity it had to hide or deter the govern-
ment from knowing exactly what it did own. In other words, there is non-random
measurement error in the continuous assessments of the reform (area or value).
It is not random because it is correlated with the capacity of the Church to hide
its estates, which at the same time can be conceived as influencing the level of
political violence.
To alleviate measurement error in this sense, I focus on a discrete measure
of the reform: a dummy variable that equals one if at least one of the Church?s
properties had been expropriated in a given municipality. In other words, even
17
though the total value of the mortmain disentailed per municipality may be sys-
tematically biased by the Church?s relative power, I assume the Church did not
own land in mortmain in any municipality with no records of the disentailment
reform.
The assumption is plausible for two reasons: one, the incentives the gov-
ernment offered to informants (Jaramillo and Meisel, 2009); and two, the gov-
ernment?s ability to expropriate Church property in each municipality should in-
crease according to the amount of land owned by the Church. To put it differently,
the systematic measurement errormay exist in the intensivemargin of the reform,
but not on the extensive margin.
Table 1.1 compares municipalities with reform to those without the reform
on several dimensions. First, notice they differ in the expected ways. Property
held in mortmain was a legacy of the Spanish empire. Therefore, the Catholic
Church held land in places that were founded earlier andwhere it wasmore likely
to find indigenous groups, which is in the temperate areas, high in the Andes
mountains, and closer to Bogot? (Acemoglu and Robinson, 2012).
However, the two groups do not differ in other geographical variables, such
as the distance to the department?s capital or the soil quality index. Using data
from the 1856 elections, I find that the Conservative Party had a higher vote share
where the Church had estates, but the average level of political competition was
not different between the two groups of municipalities.
Figure 1.2 presents additional evidence that the geographical variation of
reform obeys factors other than political competition. It divides municipalities in
18
deciles according to the Conservative Party 1856 vote share, and shows that the
share of ?treated? municipalities does not increase systematically with Conserva-
tive Party support. For instance, municipalities in the third decile of conservative
vote share were as likely to be treated as municipalities where 100% of the votes
went to the conservative candidate (tenth decile). A similar conclusion can be
drawn from Figure 1.3 for the case of political competition.
In order to explore more rigorously the correlates of disentailment reform,
Table 1.2 shows the results for regressions on both the intensive (columns 1 and 2)
and extensive (column 3) margins of reform. When I include all relevant covari-
ates to try to explain what type of municipalities experienced the disentailment
reform, I find a robust correlation among the following factors: foundation year
(older municipalities, with greater exposure to the colonial economy); distance to
Bogot? (municipalities closer to the colonial center, where Spanish first settled in
large numbers); and amount and value of more land expropriated and the proba-
bility of being treated. These relationships hold even when I focus only on treated
municipalities in columns 4 and 5.
Interestingly, there is not a robust relationship between political indicators
before the reform and the extent of the disentailment reform. As further evidence
of this finding, consider Figures 1.4 and 1.5. They show the intensive margin of
reform, amount of area expropriated, only for treated municipalities. There does
not exist a relationship between political support for theConservative Party before
the 1860s and the location or severity of the expropriation of the Church?s real
estate.
19
1.2.3 Empirical strategy
To estimate the effect of the disentailment reform on political conflict I ex-
ploit the geographic variation of the reformand its timing in a difference-in-differences
setting. I compare the change in political violence, before and after 1862, in mu-
nicipalities where the Church?s real estate was expropriated with that of munici-
palities where the Church did not own land in mortmain. The main specification
is given by:
?K
B = ?(d1863 ?DR ) + ? (xj ? d1863it t i j i ) + ?i + ?t + it (1.1)
j=1
WhereBit is the number of battles in year t in municipality i;DRi is an indi-
cator equal to one if therewas a record of the disentailment reform inmunicipality
i, d1863t is a dummy variable equal to one from the year of the first recorded pur-
chase onwards, Xi = [x1i , ..., xKi ] is a set of controls, ?t is a full set of year fixed
effects, and ?i is a set of municipality fixed effects9, to control for both national
trends in conflict and common factors at the municipality level, respectively. The
coefficient of interest is ?.
Xi include municipalities characteristics, in particular: distance to Bogot?
and the State?s capital, altitude, soil quality index, an indicator for the location
of indigenous groups around 1540, and an indicator of early Spanish settlements
(1510 to 1560). To control for the level of political partisanship that may drive po-
litical violence I also include the Conservative party vote share in the 1856 presi-
9The results are the same when I used a full set of province fixed effects.
20
dential election.
Given the time invariant differences between places that received the reform
and those which did not, I include a set of interactions between the control vari-
ables in Xi and the dummy for the post period of the reform. In this fashion, I
flexibly control for the concern that the underlying (observable) characteristics in
Xi may be driving the results and not the treatment indicator. In particular, I con-
trol for the 1856 election results, to account for underlying ideology and relative
power of both parties.
I also estimate Equation 1.1 changing the dummy variable DRi for various
continuousmeasures of the success of the reform: total amount of land, total value
of the land, and share of the municipality?s area that was disentailed. This last
?intensive margin? measure has to be taken with a grain of salt. I use as a denom-
inator the municipality?s area in 1980, which is the earliest measure of total area
I have. Therefore, the measure would be inaccurate for the youngest municipal-
ities, whose boundaries were not well defined in the 1860s. This is the best I can
do, given data availability.
Finally, as the level of violence is serially correlated for each municipality,
unless otherwise noted, I estimate standard errors clustered at the municipality
level.
21
1.3 Disentailment ReformContributed toReduce Political Violence
The basic logic of the empirical strategy is to compare the change in the
level of violence before and after the disentailment reform between municipal-
ities where the Church?s real estate was expropriated and municipalities where
the Church did not own properties in mortmain. Figure 1.6 shows the raw data
for comparison, where indeed it appears that after the disentailment reform, the
level of violence fell by a greater amount in the municipalities where properties
were taken away from the Catholic Church and sold at public auction. Figure 1.7
shows the residual level of violence after controlling for municipality character-
istics and municipality and year fixed effects. Here it is even clearer that munic-
ipalities where the Church?s real estate was expropriated became more peaceful
after 1863 when compared to control municipalities.
The difference in the level of conflict can only be interpreted as being gener-
ated by the disentailment reform under the assumption that, absent the reform,
political violence in treated municipalities would have followed a similar path
than in control municipalities (Angrist and Pischke, 2008a). While this is impos-
sible to prove definitively, the figures also show that the trends in political conflict
between municipalities in which the reform took place and those in which it did
not were relatively similar before 1863.
The raw data for Figures 1.6 and 1.7 provide a first pass test on the parallel
trends assumption. However, I test it more formally following Autor (2003). In
22
particular, I estimate:
?m ?q ?K
Bit = ?m + ?
j
t + ??tDRi + ?+tDRi + ?j(xi ? d1863) + it (1.2)
t=0 t=0 j=1
If the trends on political violence before the reform are not different between
the treatment and control group, the coefficients for the years leading to the re-
form should not be statistically different, but the coefficients on the years after the
reform should be10. It is a validation of the interpretation of the main results as
coming from the disentailment reform and not from chance or mean regression.
The negative effect of the disentailment reform is sizable: it represents a
decrease of 29%of the average number of battles per year in themunicipalities that
received the reform. It is also robust to the inclusion of controls. Table 1.3 presents
the main results. I estimate Equation 1.1 adding controls step by step. In every
column, standard errors are clustered at the municipality level. The coefficient of
interest, ?, is the interaction between a dummy variable equal to one for treated
municipalities and a dummy variable equal to one for the period after 1862. It is
negative and significant at the 95% level in all the specifications, including column
5 where several controls for differential trends after the reform were included.
Another way to understand the results is to recognize that results in column 5 are
the numbers underlying Figure 1.6.
To corroborate these results, I performed a placebo testwhere I estimated the
10A graphic representation of this test is presented in Appendix A.2.
23
equation in column 5 from Table 1.3 as if the reform had taken place in different
years. If there were differential changes in political violence between municipal-
ities that experienced disentailment reform and those that did not during years
when the reform did not take place, that would cast serious doubt on the conclu-
sion that the reduction in violence was due to the disentailment reform and not
to other factors. Figure 1.8 plots the coefficient point estimate and the clustered
standard error for the placebo regressions. The coefficient is only statistically sig-
nificant for the year the reform took place and two or three years later (depending
on the significance level used).
So far I have only used the extensive margin definition of the reform. This
analysis lumps into one category municipalities where the Church owned little
real estate and others where it had large landholdings. If disentailment reform
had an impact on political violence because it weakened the Church?s status, we
should see a more pronounced effect on municipalities where the Church was
hit the hardest. Table 1.4 reproduces the estimation of Equation 1.1, replacing
the dummy variable of the reform for a continuous measure of the extent of the
disentailment reform. Columns 1 and 4 use the natural log of the total area dis-
entailed, columns 2 and 5 use the natural log of the total appraisal of the prop-
erties disentailed, and columns 3 and 6 use the share of land expropriated from
the municipality. While columns 1, 2, and 3 use the whole sample of municipali-
ties, assigning the value of zero to those that did not have records of the reform,
columns 4, 5, and 6 restrict the sample to those which had records of the reform.
The results for the area and value expropriated accord with those in Ta-
24
ble 1.3. The take-away is in the same direction: where the reform had a stronger
impact on the Church?s estates, political violence decreased faster. This is true
even when considering only ?treated? municipalities. However, I do not find a
significant effect using as an intensive measure of the reform the share of the mu-
nicipality?s area expropriated. Moreover, when I consider only treated munici-
palities, the coefficient changes its sign, from negative in column 3 to positive in
column 6.
Results in Table 1.4 help to alleviate concerns about measurement error for
the treatment variable expressed in section 1.2.2. However, the dependent vari-
able, political violence, may suffer from measurement error since it was coded
using historical accounts of political violence in the nineteenth century. In Ta-
ble 1.5, I compare the results whenmeasuring political violence using the number
of battles (column 1) or simply an indicator variable equal to one for years when
there was a positive number of battles (column 2). Results in column 2 show that
after 1863 the probability of an event of political violence dropped considerably
for municipalities where the Church?s real estate was expropriated compared to
municipalities were the disentailment reform had no impact.
I selected the main period of analysis based on Colombian political history,
however, I could have chosen the time frame in many different ways. In columns
3 and 4 of Table 1.5, I show the results are robust to changes in the sample years.
Column 1 shows the main results for comparison. Column 3 extends the period
to cover the second half of the nineteenth century, but leaves out the period from
1899 to 1902, where the longest civil war took place. Results in column 4 use a 25
25
year sample period centered at 1863. The effect of disentailment reform on politi-
cal violence remains negative and significant. Finally, capital cities from the nine
states may be different in many regards to the typical municipality, from popula-
tion density to political activity. Therefore, I estimate the main results removing
them from the sample in column 5 and find capital cities do not drive the main
results.
Given the yearly nature of the data and the nature of political conflict, I ex-
tend the main results to account for the dynamics of conflict in Table 1.6. Column
1 reproduces the main results for comparison. In column 2 I control for lagged vi-
olence and show that political conflict is not serially correlated once I control for
year fixed effects, as can be seen from the coefficient on the first lag. In columns 3
and 4, I flexibly control for interactions of the treatment variable (Disentailment)
with different years around (column 3) and after (column 4) disentailment re-
form took place (Mora and Reggio, 2012). The coefficient remains negative and
statistically significant in column 4, but is not significant in column 3, although
the interaction with the previous year dummy is now significant.
Showing that serial correlation of political violence is not strong is impor-
tant since Bertrand et al. (2004) raise concerns about the calculation of standard
errors in difference in differences settings when the dependent variable is serially
correlated. Even though this is not the case for the number of battles, I perform
two of their suggested corrections for calculating standard errors more accurately.
First, I collapse the data in two periods, before (1854 to 1862) and after (1863
to 1885) disentailment reform, and replicate the analysis from Table 1.3. Table 1.7
26
shows the results for two different measures of violence: the average of the num-
ber of battles per year (columns 1, 2, and 3) and the share of years with at least
one battle (column 4). The coefficient remains negative, statistically significant,
and shows the same magnitude.11
Second, I simulate the empirical distribution of ? by randomizing both the
year of the reform and themunicipalities thatwere treated. Then I estimate ?? from
Equation 1.1. Figure 1.9 shows the empirical distribution when I randomize year
and treatment 5,000 times. The empirical p-value is 0.0013. In Figure 1.10, I show
the empirical distribution when I only randomize treatment, but not the reform
year (N=800). The empirical p-value is smaller than 3%. The vertical line in both
graphs shows where ?? lies in the distribution.
1.3.1 Political Competition as an Explanation for the Negative Ef-
fect of Disentailment Reform in Political Violence
Political violence in Colombia decreased in places where the Church had
properties, compared to places where it did not after the government expropri-
ated them and auctioned them off. Even focusing only on places where Church
holdings were expropriated, the disentailment reform had a more negative effect
in violence where the Church held more land. It is important to analyze why that
was the case.
First, the historical evidence for the Church?s direct participation in violence
11As a final check, Appendix A.1 compares the standard errors under three different clustering
methods: (1) by municipality, (2) two-way year x municipality, and (3) by province using wild-
bootstrap.
27
(Ortiz and Javier, 1840; Ortiz, 2010) points to one potential explanation: when
the clergy lost economic power, it reduced its direct participation in political vio-
lence. Second, the disentailment reform also changed the property rights regime
to which Church estates had obeyed. These changes might have shaped the in-
centive structure, increasing productivity and economic performance. This chan-
nel would be consistent with evidence from Europe presented by Heldring et al.
(2015) and Finley et al. (2017). While the ?economic performance? channel is
plausible, Figures 1.6 and 1.7 show an immediate negative effect of disentailment
reform. One could argue that it might take some time for productivity to increase
and for the people to enjoy a higher standard of living that would lead to lower
levels of political violence.
In order to explore the economic performance mechanism, columns 3 and 6
of Table 1.4 introduce a continuous measure of the reform: the share of the mu-
nicipality?s area that was expropriated from the Church. If the economic channel
were very important, the disentailment reform would have had more impact on
places where a higher share of their land changed land tenure systems. Results in
Table 1.4 do not support this hypothesis. When I consider the whole sample, the
coefficient of the interaction between the extent of the reform and the post-period
(column 3) is negative and larger thanwhen Imeasure the reform using total area
(column 1), but the standard errors increase evenmore, making it not statistically
significant. Interestingly, when I focus only on ?treated? municipalities (column
6), the coefficient has the opposite sign and is not statistically significant. If there
is an economic performance effect, it might not be the main explanation to the
28
rapid decrease in violence in places where the Church lost most of its economic
power.
Given the historical context presented in section 1.1, another potential ex-
planation comes from the patterns of political competition between elite groups,
specifically, the political parties. Given their similar economic composition (Saf-
ford and Palacios, 2002), the Liberal and Conservative parties disagreed the most
with respect to the Church?s role in society. It is plausible that, as soon as the
Church lost its economic power, the Conservative Party lost some of its incentive
to support the causes that made them compete with the Liberal Party the most.
In the main results from column 5 of Table 1.1, there is an evident difference
of the effect of the reformbetween very conservative and very liberal places. Over-
all, consider two municipalities, one where the Conservatives won all the votes in
1856 and the otherwhere theConservative Party got zero votes. If bothmunicipal-
ities were treated, the reform had a negative effect on the former one that is only
19.6% of the reform?s effect on the latter one. In places where the Conservative
Party had great support, the fact that the Church?s real estate was expropriated
did not significantly change the dynamics of violent political competition. How-
ever, where the Conservative Party did not enjoy great support, once the Church
becomes weaker, violence diminishes sharply.
In Table 1.8, I present evidence that argues formore in-depth study of the re-
lationship between political competition and the effects of disentailment reform.
Column 1 shows the main results for comparison. Column 2 uses the triple in-
teraction between Conservative support, post period, and the indicator for the
29
disentailment. These results corroborate the idea that the reform?s effect on the
Conservative Party?s strongholds was very mild. A simpler way to show this is to
split the sample in two according to Conservative support. Column 6 only consid-
ers municipalities where the Conservative Party got more than 43% of the votes
(the median vote share). For that group the reform did not impact political vi-
olence. Column 7 considers the other 297 municipalities. The Liberal Party was
stronger there, and the reform had a strong negative impact on political violence.
It is important to remember that these results are not coming from a differential
incidence of the reform according to political support (see Figures 1.2 and 1.4 and
Table 1.2). The Church was as likely to have estates that were expropriated in
Conservative as in Liberal municipalities.
Not only was the negative effect concentrated in more Liberal places, it was
also concentrated in politically contested municipalities, that is, in municipalities
where the Liberal and Conservative parties split the votes relatively evenly. Col-
umn 3 shows the result from adding a triple interaction between political com-
petition and the dummies for the post-period and treatment groups. Columns 4
and 5 indicates the same results in a simpler way. The negative effect of the disen-
tailment reform is only present in places with a political competition index above
the median (column 3). Again, Figures 1.3 and 1.5 and Table 1.2 show that polit-
ically contested municipalities were as likely to be part of the ?treatment group?
as non-contested places.
Interestingly, the effect of disentailment reform on municipalities where po-
litical competition was low can be seen as indirect evidence favoring political
30
rather than economic mechanisms. In other words, in places with low political
competition, the land tenure system also changed. If the economic mechanism
was the main driver, there should also be an effect in those places.
The disentailment reform weakened the Catholic Church considerably. I ar-
gue that it led to a change in the dynamics of political competition in Colombia,
where Conservative factions did not have a strong incentive to keep backing the
Church politically since the Church had less to offer them in terms of economic
support. As a consequence, the level of political violence decreased in places
where the Church was weakened. I showed that this effect was weaker for Con-
servative strongholds and for municipalities with very low political competition.
1.4 Conclusion
Not only inColombiawas violence endemic to the system. Countries through-
out Latin America ?virtually collapsed under the weight of what historians refer
to as ?state building?? during the nineteenth century (North et al., 2000). In order
to solve the problem of violence, elites had to develop ways to distribute rents
to generate a more peaceful equilibrium. Political rules, tax systems, trade poli-
cies, labor contracts, and land property rights, which were changing constantly
at the time, are examples of how contentious issues could either enhance stability
or promote violence. This chapter illustrates how institutional changes like dis-
entailment reform can be used as natural experiments to study civil war and the
process of state building in Latin America.
31
In this chapter, I explore one particular reform that was common in the re-
gion, the disentailment of mortmain land, using data from Colombia. Contrary
to some traditional interpretations of the disentailment reform, I show political
violence decreased after the government expropriated the Church?s assets and
distributed them among other powerful members of the society. The negative ef-
fect of the reform was stronger in municipalities with high political competition
or low support for the Conservative Party. These results do not imply that the
Church ceased to be involved in the political process. Rather, they point out that
the relationship between Conservative factions and the Church was weakened,
leading to less political violence.
Other countries in the region experienced similar reforms. As the Colom-
bian experience shows, these reforms might have led to consolidate secular elites
and reduce the direct involvement of the Catholic Church in political conflicts. As
long as expropriating the Church?s properties or other institutional reforms had
an impact on the formation political coalitions, they could contribute to reduce vi-
olence and conflict. Given that civil wars and instability hurt long run economic
growth (Blattman and Miguel, 2010), this is an important research agenda for
understanding Latin American development during the nineteenth century and
afterwards.
32
Figures and Tables
Figure 1.1: Number of purchases per year
Note: The figure shows the number of purchases of disentailed property by year, from
1864 to 1884. Source: Fazio and S?nchez (2010).
Figure 1.2: Share of municipalities with disentailed property by decile of
Conservative party vote share
Note: the figure plots the share of municipalities, in each decile of Conservative vote
share in 1856, which belong to the treatment group, i.e. municipalities where the
Church?s real estate was expropriated. The Conservative party registered more than
99% of the votes in the 10th decile and 0% of the votes in the first decile.
33
Figure 1.3: Share of municipalities with disentailed property by decile of
political competition
Note: the figure plots the share of municipalities, in each decile of the measure of po-
litical competition, which belong to the treatment group, i.e. municipalities where the
Church?s real estate was expropriated. Elections in municipalities in the 10th decile
were the most competitive.
Figure 1.4: Average area disentailed in treated municipalities by decile of
Conservative vote share
Note: the figure plots the average area disentailed in treated municipalities in each
decile of Conservative vote share in 1856. The Conservative party registered more
than 99% of the votes in the 10th decile and 0% of the votes in the first decile.
34
Figure 1.5: Average area disentailed in treated municipalities by decile of political
competition
Note: the figure plots the average area disentailed in treated municipalities in each
decile of the measure of political competition. Elections in municipalities in the 10th
decile were the most competitive.
Figure 1.6: Average battles per year by Disentailment Reform status
Note: The figure plots the average number of battles per year by disentailment status.
The triangles represent municipalities where the Church?s real estate was expropri-
ated (treated), while the circles represent the control group. Similarly, the lines show a
fifth degree local polynomial smoothing. The dashed line belongs to the control group.
35
Figure 1.7: Residual average battles per year by Disentailment Reform status
Note: The figure plots the average of the residuals after regressing the number of
battles per year on year fixed effects and a vector of controls interacted with a dummy
equal to one after 1962. Controls include: area, altitude, distance to Bogot?, distance
to the department capital, a dummy for indigenous population in 1550, a dummy
for Spanish occupation in 1550, and a soil quality index. The triangles represent the
average residual for municipalities where the Church?s real estate was expropriated
(treated), while the circles represent the average for the control group. Similarly, the
lines show a fifth degree local polynomial smoothing. The dashed line belongs to the
control group.
36
Figure 1.8: Placebo test: diff-in-diffs estimator (? in eq. 1.1) by cut-off year
Note: the figure plots the coefficient of interest in equation 1.1, ?, and its confidence
interval when varying the year t from 1857 to 1874. The coefficient of interest comes
from the interaction between a dummy variable equal to 1 for the treatment group and
a dummy variable equal to 1 from year t onwards. Each point in the figure represents
a different regression. The dashed line highlights the ?true? year of the reform (the
period after 1862), it is equivalent to the coefficient in column (6) from table 1.3.
37
Figure 1.9: Empirical distribution of ? in eq. 1.1 when randomizing treatment
units and year
Note: the figure is based on regressions of the number of battles per year on an interac-
tion between an indicator of treatment and an indicator for the period after the reform
as indicated in equation 1.1. I jointly randomized both of these indicators within my
sample. That is, I randomly chose a group of 149 municipalities as treated and one
year as the reform year, and I estimated ??. I repeated these steps 5,000 times. The fig-
ure plots the distribution of the estimates, and the vertical line shows the value of the
coefficient when using the actual treatment group and year. 2.98% of the estimated
coefficients are smaller than the ?true? coefficient.
38
Figure 1.10: Empirical distribution of ? in eq. 1.1 when randomizing treatment
units
Note: the figure is based on regressions of the number of battles per year on an interac-
tion between an indicator of treatment and an indicator equal to 1 from 1863 onwards,
as indicated in equation 1.1. I randomly chose a group of 149municipalities as treated,
and estimated ??. I repeated these steps 5,000 times. The figure plots the distribution
of the estimates, and the vertical line shows the value of the coefficient when using
the actual treatment group. 0.13% of the estimated coefficients are smaller than the
?true? coefficient.
39
Table 1.1: Comparison of Municipalities without and with Disentailment
Control Treatment Control-Treatment
Variable N Mean N Mean t-statistic
SE SE
Dependent variable
Battles per year (1854-1885) 14,624 0.021 4,896 0.043
0.152 0.237
Municipality characteristics
Foundation year 448 1833.634 149 1759.698 7.467
100.740 115.844
Altitude (mts over sea level) 448 1219.337 149 1796.718 -4.621
931.986 2096.082
Distance to Bogota (km) 448 338.366 149 195.817 8.559
188.670 131.057
Indigenous population (1535-1540) [dummy] 448 0.482 149 0.584 -2.157
0.500 0.495
Spanish occupation (1510-1560) [dummy] 448 0.438 149 0.624 -3.995
0.497 0.486
Distance to Department Capital (km) 448 71.983 149 71.293 0.153
48.306 45.416
Distance to main market (km) 448 111.825 149 99.611 1.897
73.228 49.387
Soil erosion index (2005) 448 1.990 149 2.060 -0.689
1.070 1.098
Long run outcomes
Land owned by religious groups (2005, hm2) 384 22.642 146 21.346 0.136
108.965 61.630
Public land (2005, hm2) 384 7658.755 146 5294.451 0.460
58672.713 33109.284
Land inequality (2005) 428 0.713 147 0.707 0.750
0.091 0.095
La Violencia (fights between 1948-1953) [dummy] 448 0.116 149 0.134 -0.589
0.321 0.342
Land disputes (1901-1931) [dummy] 448 0.121 149 0.087 1.114
0.326 0.283
Land disputes (1901-1917) [dummy] 448 0.167 149 0.121 1.359
0.374 0.327
Vote share 1856 elections
Conservative (Mariano Ospina) 448 0.434 149 0.531 -2.616
0.399 0.374
Liberal (Manuel Murillo) 448 0.355 149 0.390 -1.043
0.361 0.353
Independent (Tomas C. Mosquera) 448 0.211 149 0.078 4.546
0.340 0.182
Political competition 448 0.326 149 0.363 -1.221
0.316 0.333
Note: Control: No records of disentailed property. Treatment: Records of disentailed property
40
Table 1.2: Correlates of the disentailment reform
(1) (2) (3) (4) (5)
Dependent variable: Area exprop. Appraisal Treatment=1 Area Appraisal
Sample: All Treated (Disentailment=1)
Cons. Vote share -0.1325 -0.2132 -0.0354 -0.1402 0.3238
(0.253) (0.352) (0.046) (0.618) (0.450)
Political Comp. Index 0.0781 0.2711 0.0443 -0.5035 -0.0392
(0.283) (0.395) (0.051) (0.623) (0.454)
Foundation year -0.0036??? -0.0063??? -0.0007??? -0.0014 -0.0037???
(0.001) (0.001) (0.000) (0.002) (0.001)
Area -0.0073 0.0075 0.0001 -0.1818 -0.0801
(0.094) (0.132) (0.017) (0.245) (0.178)
Altitude -0.0279 0.0315 0.0056 -0.3483? 0.1546
(0.066) (0.091) (0.012) (0.201) (0.147)
Distance to Dept. Capital -0.0785 -0.1311 -0.0098 -0.0620 -0.1702
(0.091) (0.127) (0.017) (0.176) (0.128)
Distance to Bogota -0.8806??? -1.3468??? -0.1659??? -0.3479 -0.3467?
(0.137) (0.191) (0.025) (0.253) (0.184)
Spanish ocup. 1560=1 0.4025?? 0.6180?? 0.0711?? 0.5131 0.3576
(0.198) (0.276) (0.036) (0.456) (0.332)
Indigenous pop. 1534=1 0.0930 0.1094 0.0185 -0.0321 -0.1028
(0.190) (0.264) (0.034) (0.429) (0.312)
R2 0.1532 0.2030 0.1781 0.0432 0.1532
Municipalities 610 610 610 153 153
This table shows some correlates of the extent of disentailment reform. The dependent variables are: Area ex-
propriated (in log) (columns 1 and 4), total value as appraised by the Disentailment Agency (in log) (columns
2 and 5), and a dummy equal to 1 for municipalities where the Church?s real estate was expropriated (column
3). Columns 1, 2, 3 consider the full sample, imputing a value of zero on the dependent variable for control
municipalities. Columns 4 and 5 focus only on treated municipalities.
? p < 0.1, ?? p < 0.05, ??? p < 0.01
41
Table 1.3: Difference-in-differences estimator: Effect of the disentailment reform
on political violence
(1) (2) (3) (4) (5)
Dependent variable: # Battles per year
Sample years: 1854-1885
d1863 -0.0149??? -0.0149???
(0.003) (0.003)
Disentailment 0.0356??? 0.0356???
(0.011) (0.011)
d1863 x Disentailment -0.0221?? -0.0221?? -0.0221?? -0.0221?? -0.0214??
(0.009) (0.009) (0.009) (0.009) (0.011)
d1863 x Cons. vote share 0.0172??
(0.008)
d1863 x Control X
Municipality FE X X X
Year FE X X X
R2 0.0061 0.0505 0.0861 0.1305 0.1385
N 22,048 22,048 22,048 22,048 19,104
Municipalities 689 689 689 689 597
Standard errors clustered at the municipality level in parentheses. The dependent variable is the
number of battles per year, built from Riascos Grueso (1950). Disentailment is a dummy variable
equal to 1 for municipalities where the Church?s real estate was expropriated. dt is a dummy
equal to 1 for years ? t. Conservative vote share (1856) is the share of total votes won by Mariano
Ospina, the Conservative party candidate in the 1856 presidential election. Control variables are
defined at the municipality level and include: altitude, total area of the municipality, distance
to the State?s capital (log), distance to Bogota (log), distance to the closest main market (log),
a dummy indicating early indigenous settlements (by 1534), a dummy indicating early Spanish
settlements (by 1560), and soil quality index.
? p < 0.1, ?? p < 0.05, ??? p < 0.01
42
Table 1.4: Continuous measures of the disentailment reform
(1) (2) (3) (4) (5) (6)
Dependent variable: # Battles per year
Sample years: 1854-1885
Sample: All Disentailment=1
Cons. vote share x d1863 0.018?? 0.017?? 0.018?? 0.021 0.026 0.023
(0.008) (0.008) (0.009) (0.024) (0.023) (0.025)
d1863 x ln(Area Disent.) -0.004??? -0.003
(0.002) (0.004)
d1863 x ln(Appraisal) -0.003??? -0.011?
(0.001) (0.006)
d1863 x Share of Mun. Area -0.033 0.037
(0.092) (0.127)
d1863 x Control X X X X X X
Municipality FE X X X X X X
Year FE X X X X X X
R2 0.139 0.139 0.135 0.169 0.171 0.166
N 19,104 19,104 18,880 4,768 4,768 4,544
Municipalities 597 597 590 149 149 142
Standard errors clustered at the municipality level in parentheses. The dependent variable is the number of battles per
year, built from Riascos Grueso (1950). Disentailment is a dummy equal to 1 for municipalities where the Church?s
real estate was expropriated. Area Disent. and Appraisal are, respectively, the natural log of one plus the total size
in hectares and total value in pesos of the properties expropriated from the Church in the disentailment reform.
Share Mun. Area is the total area disentailed divided by the municipality?s area (note: the denominator is the area
measured in 2005, not in the 1860s. Therefore, I drop out of the sample municipalities with a share higher than the
95 percentile of the distribution of share, as explained in section 1.2.2). Conservative vote share (1856) is the share
of total votes won by Mariano Ospina, the Conservative party candidate in the 1856 presidential election. Control
variables are defined at the municipality level and include: altitude, total area of the municipality, distance to the
State?s capital (log), distance to Bogota (log), distance to the closest main market (log), a dummy indicating early
indigenous settlements (by 1534), a dummy indicating early Spanish settlements (by 1560), and soil quality index.
? p < 0.1, ?? p < 0.05, ??? p < 0.01
43
Table 1.5: Robustness checks
(1) (2) (3) (4) (5)
Dependent variable # Battles 1(Battles > 0) # Battles # Battles # Battles
Sample years 1854-1885 1854-1885 1850-1898 1850-1875 1854-1885
Sample municipalities All All All All No capitals
d1863 x Disentailment -0.0214?? -0.0208?? -0.0167?? -0.0221?? -0.0219??
(0.011) (0.009) (0.007) (0.009) (0.009)
d1863 x Cons. vote share 0.0172?? 0.0054 0.0120?? 0.0185?? 0.0174??
(0.008) (0.008) (0.006) (0.007) (0.008)
d1863 x Control X X X X X
Municipality FE X X X X X
Year FE X X X X X
r2 0.1385 0.1420 0.1162 0.1302 0.1338
N 19,104 19,104 28,656 15,522 18,848
Municipalities 597 597 597 597 589
Standard errors clustered at themunicipality level in parentheses, at the province level in brackets. The dependent variable
is either the number of battles or a dummy equal to 1 for years with a positive number of battles (1(Battles > 0)), built
from Riascos Grueso (1950). Disentailment is a dummy variable equal to 1 for municipalities where the Church?s real estate
was expropriated. dt is a dummy equal to 1 for years ? t. Conservative vote share (1856) is the share of total votes won
by Mariano Ospina, the Conservative party candidate in the 1856 presidential election. Control variables are defined at
the municipality level and include: altitude, total area of the municipality, distance to the State?s capital (log), distance
to Bogota (log), distance to the closest main market (log), a dummy indicating early indigenous settlements (by 1534), a
dummy indicating early Spanish settlements (by 1560), and soil quality index.
? p < 0.1, ?? p < 0.05, ??? p < 0.01
44
Table 1.6: Robustness checks: Dynamics
(1) (2) (3) (4)
Dependent variable: # Battles per year
Sample years: 1854-1885
# Battles (t-1) 0.012
(0.018)
# Battles (t-2) -0.056???
(0.012)
# Battles (t-3) -0.042??
(0.017)
d1860 x Disentailment -0.000
(0.047)
d1861 x Disentailment 0.018
(0.041)
d1862 x Disentailment -0.075?
(0.042)
d1863 x Disentailment -0.021?? -0.022?? -0.013 -0.032??
(0.011) (0.011) (0.017) (0.015)
d1864 x Disentailment 0.002 0.002
(0.019) (0.019)
d1865 x Disentailment -0.009 -0.009
(0.016) (0.016)
d1866 x Disentailment 0.029 0.029
(0.021) (0.021)
R2 0.139 0.142 0.140 0.139
N 19,104 19,104 19,104 19,104
Municipalities 597 597 597 597
Standard errors clustered at the municipality level in parentheses. The dependent
variable is the number of battles per year. Disentailment is a dummy equal to 1 for mu-
nicipalitieswhere theChurch?s real estatewas expropriated. dt is a dummy equal 1 for
years? t. All results include: interaction between controls and d1863, andmunicipal-
ity and year fixed effects. Control variables are defined at the municipality level and
include: 1856 presidential elections Conservative vote share, altitude, total area of the
municipality, distance to the State?s capital (log), distance to Bogota (log), distance
to the closest main market (log), a dummy indicating early indigenous settlements
(by 1534), a dummy indicating early Spanish settlements (by 1560), and soil quality
index.
? p < 0.1, ?? p < 0.05, ??? p < 0.01
45
Table 1.7: Only one pre and post period
(1) (2) (3) (4)
Dependent variable: Average # of battles % of years with
per year at least one battle
Sample years: Pre= 1854-1862; Post= 1863-1885
Post=1 -0.015??? -0.015??? 0.084 0.098??
(0.004) (0.003) (0.057) (0.040)
Disentailment 0.036???
(0.006)
Post x Disentailment -0.022?? -0.022?? -0.024?? -0.023???
(0.009) (0.009) (0.010) (0.008)
Constant 0.031???
(0.003)
Post x Control X X
Municipality FE X X X
R2 0.046 0.717 0.731 0.782
N 1,378 1,378 1,350 1,350
Municipalities 689 689 597 597
Standard errors clustered at the municipality level in parentheses. The dependent variable
for columns (1), (2), and (3) is the average number of battles per year. The dependent variable
for column (4) is the percentage of years with at least one battle. Both are built from Rias-
cos Grueso (1950). There are two periods for each municipality, the pre-period goes from
1854 to 1862 and the post-period is from 1863 to 1885. Post is a dummy variable equal to 1
for the post-period. Disentailment is a dummy variable equal to 1 for municipalities where
the Church?s real estate was expropriated. dt is a dummy equal to 1 for years ? t. Control
variables are defined at the municipality level and include: altitude, total area of the munic-
ipality, distance to the State?s capital (log), distance to Bogota (log), distance to the closest
main market (log), a dummy indicating early indigenous settlements (by 1534), a dummy
indicating early Spanish settlements (by 1560), and soil quality index.
? p < 0.1, ?? p < 0.05, ??? p < 0.01
46
47
Table 1.8: Mechanisms
(1) (2) (3) (4) (5) (6) (7)
Dependent variable: # Battles per year, 1854-1885
Political Comp. Vote in 1856
Sample All High Low Conservative Liberal
d1863 x Disentailment -0.021?? -0.030? -0.001 -0.035?? -0.000 -0.002 -0.038??
(0.010) (0.017) (0.011) (0.015) (0.012) (0.009) (0.019)
d1863 x Cons. vote share 0.017?? 0.013 0.015? 0.005 0.021?? 0.044?? -0.048
(0.008) (0.009) (0.008) (0.018) (0.010) (0.019) (0.046)
d1863 x Cons. vote share x Disent. 0.017
(0.022)
d1863 x Pol. Comp. -0.004
(0.010)
d1863 x Pol. Comp. x Disent. -0.055??
(0.025)
d1863 x Control X X X X X X X
Municipality FE X X X X X X X
Year FE X X X X X X X
R2 0.139 0.139 0.139 0.161 0.110 0.128 0.153
N 19,104 19,104 19,104 9,536 9,568 9,600 9,504
Municipalities 597 597 597 298 299 300 297
Standard errors clustered at themunicipality level in parentheses. The dependent variable is the number of battles per year, built fromRiascos Grueso
(1950). Political Competition is measured using an index distributed between 0 and 1, where 0 means one candidate got all the votes and 1 means
the candidates split the votes perfectly. Columns (4) and (5) use, respectively, municipalities where Political Competition was higher and lower
than the median. Columns (6) and (7) use, respectively municipalities where the Conservative candidate?s vote share was higher and lower than
the median. Disentailment is a dummy variable equal to 1 for municipalities where the Church?s real estate was expropriated. dt is a dummy equal
to 1 for years ? t. Conservative vote share (1856) is the share of total votes won by Mariano Ospina, the Conservative party candidate in the 1856
presidential election. Control variables are defined at the municipality level and include: altitude, total area of the municipality, distance to the
State?s capital (log), distance to Bogota (log), distance to the closest main market (log), a dummy indicating early indigenous settlements (by
1534), a dummy indicating early Spanish settlements (by 1560), and soil quality index.
? p < 0.1, ?? p < 0.05, ??? p < 0.01
Chapter 2: CaffeinatedDevelopment: Exports, HumanCapital, and
Structural Transformation in Colombia
The first wave of globalization at the dawn of the 20th century allowed coun-
tries that had not yet industrialized to expand their agricultural production to
supply world demand (O?Rourke and Williamson, 2002). Were these export op-
portunities leveraged for expanding the industrial sector? Or, on the contrary, did
those places focus on agriculture and delay industrialization? Whether the rise in
agricultural exports helped the development of manufacturing and services -the
process of structural transformation- is a central question on development eco-
nomics and has been debated for decades (Rosenstein-Rodan, 1943; Lewis, 1955;
Schultz, 1964; Kuznets, 1966). Theoretical contributions highlight potential mech-
anisms in both directions.1 The debate has influenced political views about glob-
alization as well as trade and industrial policy in developing countries since the
post-war period (Cardoso and Faletto, 1979; Wallerstein, 2011). But the direction
of the change in structural transformation resulting from the expansion of agri-
cultural exports is context-specific and, ultimately, an empirical question.
1The direction may depend on the degree of trade openness (Matsuyama, 1992), income elas-
ticity of demand for manufacturing goods (Murphy et al., 1989), changes on terms of trade (Pre-
bisch, 1950), depth of linkages with the rest of the economy (Hirschman, 1958), or features of
crops? production function (Engerman and Sokoloff, 1997; Vollrath, 2011).
48
This chapter provides new evidence on the effect of the first wave of glob-
alization on developing countries? processes of structural transformation on the
long run. Specifically, I study the effect of Colombia?s expansion of coffee culti-
vation on industrialization and economic development.2 A long peaceful period
after 1902 and the construction of the Panama Canal in 1914 allowed the country
to increase its participation in global trade by introducing a new labor-intensive
crop, coffee, to areas mostly used to produce maize, beans, and other staples for
local consumption (Parsons, 1949). Colombia?s broken geography generated a set
of local economies relatively isolated from one another and comparable in terms
of size and population. Rich variation in climatic conditions within the country
provides a good setting to study how the opportunity to produce an agricultural
export good impacted long-run development.
This chapter shows the expansion of coffee cultivation deterred industrial-
ization. Counties producing coffee beans around 1920 developed aweaker manu-
facturing sector through the 20th century. Thoughmanufacturing employment in
1912 was at similar levels in coffee-cultivating counties and non-coffee-cultivating
counties, the expansion of agricultural exports had a negative and sizable effect on
manufacturing employment in 1938, 1973, and 2005, reaching its peak in 1973. By
2005, coffee cultivation?s effect on manufacturing employment had halved, which
follows the pattern of Colombia?s structural transformation established in Fig-
ure 2.1. Consequently, I show that counties producing coffee beans around 1920
2The four-fold coffee production expansion between 1905 and 1921 is comparable to the largest
expansion of modern agricultural exports (Palm oil in Indonesia (Edwards, 2019)).
49
had lower population density and higher poverty rates as of 2005.
Identifying the causal relationship between coffee cultivation and structural
transformation is challenging. Counties that would not have developed a strong
manufacturing sector through the 20th century could have taken up coffee culti-
vation as an alternative. For instance, regions that had more difficulty importing
capital goods might have seen a profitable opportunity in coffee bean production
since it was transportable by mules. What would appear to be a negative effect of
coffee bean cultivation on industrialization, could, in fact, be driven by geography
or location.
In this chapter, I exploit two different sources of variation related to climatic
conditions to address endogeneity concerns. The assumption behind both instru-
ments is that climatic conditions specific to coffee trees only affect industrializa-
tion through coffee cultivation. The first instrument for 1920 coffee cultivation
is the average potential coffee yield from FAO?s Global Agro-Ecological Zones
project. FAO-GAEZ estimates potential coffee bean yields at a high-resolution
level using a combination of local climatic conditions and coffee?s growth cycle.
The second instrument exploits a discontinuous reduction in the probability that a
county grew coffee trees at 2,400 meters above sea level (7,874ft). The discontinu-
ity is explained by both low temperatures in counties above the altitude threshold
and the dissemination of information regarding coffee cultivation in the late 19th
century. Optimal temperatures to grow coffee trees ranged between 16 and 24 de-
grees Celsius (60 to 75 degrees Farenheit). Given Colombia?s tropical location, the
temperature bandwidth mapped directly to an altitude bandwidth between 400
50
and 2,400meters. Moreover, 19th century pamphlets promoting coffee cultivation
explicitly identified towns just below and just above the upper altitude threshold
as a reference due to lack of easily available thermometers3 (Saenz, 1892). This
fuzzy regression discontinuity strategy compares counties with average altitudes
higher and lower than 2,400 meters. The main specification restricts the sample to
include counties above 1,800 meters to guarantee an equal number on each side
of the threshold.
The expansion of coffee cultivation inColombiawas effectively a land-augmenting
technical change. A simple two-sector model with land-augmenting productiv-
ity can explain employment reallocation from manufacturing into agriculture, as
Bustos et al. (2016) showusing data fromBrazil after 1990. However, a theory that
explains coffee?s negative effect onmanufacturing employment must also account
for the fact that around 80% of Colombia?s labor force was employed in agricul-
ture in 1912. It must explain the differential evolution of industrialization between
coffee and non-coffee counties. Such a theory would apply more generally to de-
veloping countries during the first wave of globalization, before the proliferation
of industrialization.
Using historical and present-day data at the local level, the empirical spec-
ifications compare structural transformation patterns throughout the 20th cen-
tury between places that did and did not produce coffee beans around 1920. This
approach is relevant for two reasons: first, Colombian counties during the early
20th century are characterized by low labor mobility and connected product mar-
3As illustrated in Figure 2.6.
51
kets. Since they behave as small open economies, empirical evidence from local
units can be tied to insights from theoretical models (e.g. Foster and Rosenzweig
(2004), Bustos et al. (2016) and Fiszbein (2017)). Second, it highlights the distri-
butional consequences of trade across local economies within countries. Though
these consequences are well documented for late 20th century globalization (Au-
tor et al., 2016; Goldberg and Pavcnik, 2007), evidence is scarce for the early 20th
century. If the effect of trade on structural transformation depends on an econ-
omy?s stage of development, evidence from more recent periods might not be as
informative.
Given coffee?s labor-intensive production function, the expansion of coffee
cultivation increased the opportunity cost of education. Therefore, the supply of
skilled workers in coffee-cultivating counties increased at a slower pace relative
to other counties, which in turn slowed growth in the manufacturing sector. The
argument connecting human capital and structural transformation is formalized
explicitly by Caselli and Coleman II (2001) and indirectly by Acemoglu andGuer-
rieri (2008). Porzio and Santangelo (2019) use data across countries and within
districts in Indonesia to provide causal evidence of the positive role of schooling in
industrialization through increases in availability of workers for non-agricultural
sector. This chapter adds to the empirical evidence on supply-side mechanisms,
specially related to education, as mediators in the process of industrialization.
Two pieces of evidence support the human capital mechanism. First, the
difference in manufacturing employment between coffee and non-coffee counties
is concentrated in sectors intensive in human capital, classified according to Ci-
52
ccone and Papaioannou (2009). Second, I present difference-in-differences evi-
dence from comparing adults in the 1973 census of population born between 1902
and 1952. Different cohorts were exposed to different world coffee prices while
they were of school age, which determined the opportunity cost of dropping out
of school. Cohorts born in coffee counties and exposed to higher coffee prices
during school age accumulate fewer years of education by 1973.
These results contribute to a growing empirical literature on how export
booms reduce human capital accumulation (e.g. Bobonis and Morrow (2014);
Atkin (2016); Sviatschi (2018)). Moreover, they complement Carrillo (2019), who
finds a negative, though smaller, effect of coffee price shocks on education using
data from the second half of the 20th century4 This chapter looks at coffee price
shocks during the first half of the century, when industrialization first spread, and
focuses on coffee cultivation?s effect on structural transformation.
The fact that coffee cultivation leads to lower levels of education could be a
consequence of both household decisions and changes in supply of schooling. For
instance, landowners in coffee regions might oppose the construction of schools
or limit funding to existing ones in order to guarantee supply of agricultural work-
ers (Galor et al., 2009; Galiani et al., 2008). I leverage data on county level land
inequality between coffee farms to provide suggestive evidence of both educa-
tion supply and demand channels. In other words, coffee-bean-producing coun-
ties develop a less skilled labor force due to individual?s decisions to drop out of
4The effect I present in this chapter is almost twice as large as Carrillo (2019) findings. The
difference might be due to reduction in transportation costs, changes in education?s rate of return
or better enforcement of child labor and mandatory elementary school laws between the first and
second half of the century.
53
school and a lower number of schools per capita by 1951.
Coffee price shocks also had a negative direct effect on employment in man-
ufacturing in 1973. The share of the labor force employed in manufacturing in
1973 is lower for cohorts born in coffee counties who were exposed to higher cof-
fee prices during school age. The effect is similar in magnitude to the effect on
education. Mediation analysis Dippel et al. (2019b) suggests around 70% of the
effect of 1920 coffee cultivation on 1973 manufacturing employment is mediated
by the effect of coffee cultivation on education. This result is only suggestive of the
importance of the effect because it relies on one strong assumption: the sources
of omitted variable bias present when estimating coffee?s effect on cohort?s edu-
cation are identical to the ones that would bias the estimation coffee?s effect on
cohort?s employment in manufacturing.
Finally, this chapter explores other potential mechanism cited in the Colom-
bian economic history literature: linkages between coffee cultivation and manu-
facturing (e.g. Ocampo (1984)). I exploit variation within coffee-bean-producing
counties in terms of linkages with non-agricultural sectors by exploring one cru-
cial stage in coffee bean exports: threshing, or removing the husk from the coffee
bean. Threshing machines needed reliable energy sources that were also useful
for manufacturing activities. Since they were imported from Britain, the presence
of threshing machines also signals connection with international trade. Thresh-
ing also benefited smelting businesses that provided parts to constantly repair
them. I find, however, that the effect of 1920 coffee cultivation on manufactur-
ing employment in 1973 does not depend on the presence of threshing machines.
54
Stronger linkages do not prevent coffee cultivation from having a negative effect
on structural transformation.
This chapter contributes to the empirical literature on the effect of agricul-
ture on structural transformation and local development through productivity
increases (Foster and Rosenzweig, 2004; Hornbeck and Keskin, 2015; Moscona,
2018; Bustos et al., 2016) or other factors (Fiszbein, 2017; Droller and Fiszbein,
2019). By highlighting human capital as a relevant mechanism, my findings re-
late to studies looking at differences in living standards at the subnational level
that result from productivity gaps between agricultural and non-agricultural em-
ployment (Acemoglu andDell, 2010; Gennaioli et al., 2013; Gollin et al., 2014; Her-
rendorf and Schoellman, 2018).
This chapter?s argument about the role of human capital on the onset of in-
dustrialization in developing countries complements scholarship about Europe?s
Industrial Revolution (Galor and Moav, 2004; Squicciarini and Voigtl?nder, 2015;
Franck and Galor, 2017; de la Croix et al., 2018). Similarly, this chapter fits in
with recent works on Latin American economic history which highlight the role
of human capital in the process of structural transformation either directly (Va-
lencia Caicedo, 2019) or indirectly (Perez, 2017). This chapter adds to the study of
the adoption of coffee cultivation in Colombian history. As (McGreevey, 1971, p.
198) put it: ?No other substantive economic change in Colombian economic his-
tory can have been of such overriding social importance.? This chapter brings
comprehensive data and modern econometrics to an old debate in Colombian
economic history. It revisits an established literature studying the relationship
55
between coffee cultivation and industrialization that mostly rely on comparative
studies or time series data. The next section describes this literature inmore detail.
Afterwards, I turn to the empirical analysis. Section 2.2 describes the main
datasets used in later sections. Section 2.3 presents main correlations between
coffee cultivation and structural transformation. It also discusses the main ob-
stacles for identification and presents the empirical strategies used in Section 2.4.
Sections 2.5 and 2.6 discuss potential mechanisms. Finally, 2.7 discusses the long
term effects of coffee cultivation on income and urbanization.
2.1 Exports and Structural Transformation in Colombia
Countries in LatinAmerica started their processes of industrialization around
the first two decades of the 20th century. There was considerable heterogeneity in
the path and timing of structural transformation across the region (Salvucci, 2006;
Duran et al., 2017). While some countries like Argentina or Mexico had devel-
oped manufacturing industry by 1900, smaller countries struggled to consolidate
industrial activities (Williamson, 2011). Development economists and economic
historians have argued that differences in the features of the export sector help
to explain the diverse experiences with industrialization. What Bulmer-Thomas
(2003) called ?the lottery of commodities? has explanatory power to understand
the development of manufacturing in the region.
Demand for commodities from the world economy might help develop the
non-export economy through increases in income that increase demand for lo-
56
cally produced manufacturing. This is more likely to happen if the export sector
benefits a large fraction of the population and if transportation costs for imported
manufactured goods are high (Murphy et al., 1989; Matsuyama, 1992). Addition-
ally, different export products had different degrees of connection with other eco-
nomic activities. Linkages or complementarities of exports are cited as a reason
for successful development of manufacturing (Bulmer-Thomas, 2003; Hirschman,
1958).
These conditions were not met, for instance, for crops like bananas, pro-
duced in enclaves with limited population, or for mining activities performed in
isolation from themain centers of population Bulmer-Thomas (2003). On the con-
trary, successful episodes of industrial growth, like Argentina around 1900, have
traditionally been explained by the presence of agricultural activities like wheat
or the exporting of processed meat that were not available in other countries in
the region. Recent empirical evidence by Droller and Fiszbein (2019) support the
hypothesis that linkages in agricultural activities generate industrial growth.
Colombia did not consolidate its export sector until coffee cultivation took
off around 1910. During the 19th century gold was consistently the main export,
with a couple of short experiments with tobacco and quinine (Ocampo, 1984).
Even though coffee was relatively new in the country, a long period of peace after
1902 and two coffee price booms (1906 and in the 1920s) allowed coffee to grow
until it represented more than 80% of exports by 1940 (Nieto Arteta, 1971). Co-
incidentally, manufacturing took off around the 1930s. It had been relegated to
cottage industry during the first two decades of the century, but more modern es-
57
tablishments appeared during the 30s and 40s (Ocampo and Montenegro, 2007).
Historians and economic historians have interpreted this coincidence in tim-
ing as evidence of the causal positive effect of coffee cultivation on the develop-
ment of manufacturing, though the claim has been subject to extensive debate.5
Some features of coffee cultivation fit the two theories explained above. Coffee di-
rectly employed 18% of the labor force at its peak (McGreevey, 1971). Moreover,
its production and exporting connected an extensive area and requiredmachinery
and manufacturing products like sacks.
Proponents of the positive link between coffee cultivation and manufactur-
ing back their claims with time series or Department level data. In this chapter I
collect a wealth of historical data at both the county and individual level to em-
pirically estimate the connection between coffee cultivation and structural trans-
formation.
2.1.1 Coffee in Colombia: Historical Background
Colombia went from producing around 230 thousand bags per year in 1900
to 3.2 million in 1932. Figure 2.1a shows the evolution of exports during the first
half of the 20th century. At the end of the 19th century, the Eastern part of the
5Some version of this claim is discussed in the main economic history textbooks. The argu-
ment starts with Ospina V?squez (1955) and Parsons (1949). McGreevey summarizes the argu-
ment saying: ?the rapid growth of a new export product raised income levels and generated new
demands for imported and locally produced goods of all kinds? (McGreevey, 1971, p.198). Brew
(1973), Nieto Arteta (1971) and Palacios (2002) studied coffee cultivation and its social impacts to
Colombia?s and Antioquia?s societies. Arango (1981) focused exclusively on the direct connection
between coffee and manufacturing. Bejarano (1980) summarizes the literature up to 1980 and
Ocampo and Botero (2000), Ocampo (2015) discuss new developments from the past 40 years.
More modern literature on Colombia?s industrialization downplays the role of coffee cultivation
using network data on entrepreneurs and elite members (Mejia, 2018).
58
country produced most of the coffee. The crop made its way to Colombia?s West
and South West in the first two decades of the 20th century, well after the frontier
closed (Parsons, 1949). By 1930, the East only produced around 30%of total coffee
exports.
Early adopters of the crop wrote several pamphlets around 1880 to inform
potential investors of the opportunities that coffee cultivation provided. Those
pamphlets were collected in the book Memorias sobre el cultivo del cafeto (Saenz,
1892). They provide information about the different features of coffee?s produc-
tion function at the turn of the 20th century. In this chapter, I highlight four of
them.
First, producing coffee was labor intensive. Coffee trees had two large crops
during the year, but it was possible to collect coffee cherries all year round. Even
when labor was not required to pick the cherries, coffee farms demanded con-
stant labor for other purposes like weeding, pruning, and pest control. Second,
the pamphlets highlighted that a lot of the tasks involved in the collection and
classification of coffee were ideal for children. I argue in this chapter that those
two features of coffee production function shaped incentives to accumulate hu-
man capital and ultimately affected coffee counties? process of structural transfor-
mation and development.
Third, the production of coffee required heavy machinery to remove the fi-
nal grain for exporting from its husk. This process known as threshing6 used
imported machines, generally owned by farmers? cooperatives. Not every coffee
6In Spanish: trilla.
59
producing county had threshing machines. They were in strategic locations, not
necessarily in the main production centers. In this chapter, I argue coffee cultiva-
tion in counties with threshing machines had stronger linkages to the non-export
economy. I use this fact to test whether the effect of coffee cultivation on manu-
facturing depended on linkages.
Finally, coffee was ideally produced at medium altitude. Those pamphlets
consistently pointed out that coffee could be produced between 24 and 16 degrees
Celsius (76 to 60 Fahrenheit). Given that climate in Colombia is determined by
altitude, early coffee adopters provided reference points in terms of altitude to
decide which terrains were feasible to produce the crop. Figure 2.6 shows one
of those instances. It highlights that coffee could be produced near Rionegro,
located at an altitude of 2,200 meters, but could not be produced near Sons?n or
Santa Rosa, at altitudes of 2,500 and 2,450 meters respectively. In general, authors
of the pamphlets recognized there was an altitude bandwidth inside which coffee
cultivation was suitable. I use the upper threshold of the bandwidth in order to
identify the causal effect of coffee cultivation on structural transformation.
2.2 Data
The empirical analysis spans several decades and uses information fromvar-
ious sources. Moreover, as this chapter estimates the effect of exports on local de-
velopment, it is crucial to consistently define the unit of observation. Colombia?s
population was distributed in 18 Departamentos during most of the 20th century.
60
There were also a handful of Intendencias, where population density was lower
and most of the land was unsettled. The country?s smallest political division are
municipios, equivalent to US counties. They were generally comprised of a town
(Cabecera) and a rural area. I refer to them as ?counties.? They are the main unit
of observation, as each one of them represents a local economy.
I digitized county-level data from Colombia?s first coffee census (published
in 1927) and 1912 and 1938 census of population. Additionally, I use 1945 First
Census of Manufacturing. I match 1927, 1938, and 1945 counties to the set of 741
counties reported in 1912 Census. Whenever I could not match by name, I used
historical sources to match a county created after 1912 to its ?parent? 1912 county.
This procedure yields a set of 734 counties with observations in 1912, 1927, 1938,
and 1945. Figure 2.7 shows population patterns in 1912 and highlights the main
sample.
I also use 1973 and 2005 Census of Population, available from IPUMS Inter-
national (Ruggles et al., 2003). IPUMS homogenizes counties over time by merg-
ing small counties in terms of population and pooling them together into larger
units. I call those units ?IPUMS-county?. There are 564 in 2005 Census. The av-
erage IPUMS-county contains 1.9 actual counties (municipios). However, 57% of
IPUMS-counties only contain one actual county. 84.4%of IPUMS-counties contain
one or two actual counties. Moreover, out of the 564 counties, only 495 counties
can be traced to be part of a 1912 county. The other 69 counties are located in land
that was colonized after 1950.
For each set of results, I explicitly define the unit of observation it uses, be-
61
tween counties and IPUMS-counties. I do this for two reasons: first, counties bet-
ter represent local economies for the first part of the 20th century. I use IPUMS-
counties for results for the second half of the 20th century, where larger units
capture better the idea of a local economy. Second, even though there are some
differences, there is significant overlap between both definitions. Results using
counties look qualitatively similar as those using IPUMS-counties, but since the
sample size is smaller, power tends to be lower.
Coffee cultivation before 1920
I measure coffee cultivation at the beginning of the 20th century with the
number of coffee trees used in production by county. This measure comes from
the first coffee producers? census: Monsalve?s 1927 book, ?Colombia Cafetera.?
Monsalve was an agricultural engineer who led Colombia?s Propaganda and In-
formation Office between 1920 and 1924. During that period, he surveyed coffee
farms around the country and put together a 950-page book describing Colom-
bia?s coffee industry. In 1924, Colombia?s government bought the book?s rights.
The goal was to promote coffee exports by ?distributing the book to foreign mar-
kets, giving it out for free to public offices, and charging only the production cost
to private individuals.? Since coffee trees take around 5 years to start produc-
ing coffee cherries, the number of coffee trees registered in Monsalve?s census is
likely to represent trees that were planted in the 1910s, even though the book was
eventually published in 1927. Therefore, I interpret the number of coffee trees as
62
a measure of early exposure to coffee cultivation. For robustness, I also use the
extensive margin, a dummy equal to one for counties with a positive number of
coffee trees planted before 1924.
The average county had 427 thousand coffee trees, equivalent to around 95
hectares, but the distribution is skewed to the left. 43% of counties had no early
coffee production. 50% of counties had less than 20,000 trees, which is equivalent
to less than half a hectare. These figures show how even though coffee was taking
off during 1910s and 20s, it still represented a small share of counties? land. For
instance, Fredonia (Antioquia) had the largest number of coffee trees used for
production in 1920. Its 8.3 million trees were equivalent to 1,800 hectares or 7% its
total area. As a comparison, using data from 2005 coffee census, 22% of counties
use more than 7% of their area to produce coffee. Chinchina (Caldas) was the
county with highest concentration of coffee trees in 2005. It devoted 44% of its
area to the crop.
I use the coffee census to measure land inequality between coffee landown-
ers. I calculate the ratio between the average and themedian farm for each county
with a positive number of coffee trees. This ratio was 1.9 for the average coffee
county. Appendix B.1 describes the calculation in more detail. A typical coffee
county had around 85 coffee farms and 5,600 inhabitants in 1912. A typical farm
had 10 to 30 thousand trees. At a rate of 2 pounds per tree per year, a typical coffee
3
farm could produce between 110 and 330 60-pound bags per year.
63
Economic structure
I measure population, population in the labor force, and shares of labor
force employed in manufacturing, agriculture, and services in 1912, 1938, 1973,
and 2005. I digitized 1912 and 1938 Census of Population at the county level. I
aggregated IPUMS International?s Census samples (Ruggles et al., 2003) to build
measures at the county level for 1973 and 2005. Additionally, I estimated shares of
population who could read and write (1912, 1938, 1973, and 2005), average years
of schooling of adult population by county (1973 and 2005), and created house-
hold incomemeasures using Filmer and Pritchett (2001)methodology to summa-
rize information about housing quality and durable goods (1973 and 2005).
1912 and 1938 Census of Population provide headcounts for different ?Pro-
fessions and Occupations? at the county level. 1912 census counted the ?Active
Population? and divided it between occupations.7 I consider Agriculture as the
combination between Agriculture and Cattle Raising. Manufacturing sector is
given by the ?Crafts and Manufacturing? category, while Services adds up Lib-
eral Professions, Commerce, and Transportation. 1938 Census was also a series of
headcounts at the county level, but the division between occupations was more
detailed. Occupations were divided between Primary Production, Transforma-
tion Industries, Services, Liberal Activities, and Other. I define Agriculture as
Primary Production employment not in ?Extractive Activities? such as mining.
71912 Occupations are: Liberal Professions, Arts, Crafts andManufacturing, Priests and Nuns,
Public Employees, Military, Policemen, Agriculture, Cattle Raising, Commerce, Transportation,
and Domestic Employees.
64
Manufacturing employment is given by employment in Transformation Indus-
tries excluding ?Construction and Buildings? Finally, Services is its own category
formed by Transportation, Commerce, and Banking subdivisions.
I build measures of economic structure at the county level for 1973 and 2005
using individual level data from IPUMS International. To make it comparable
with 1912 and 1938 figures, I calculate share of population in the labor force. Then
I build counts of people employed in Agriculture, Manufacturing, and Services
to calculate shares of labor force employed in each category. Additionally, I fo-
cus on population between 18 and 65 years old to estimate household income
measures. I follow Filmer and Pritchett (2001) and use the first vector out of a
Principal Component Analysis using information on housing quality (roof and
floor materials, number of rooms, connection to electricity and sewage) as well
as durable goods consumption (washing machine, radio, refrigerator). Through-
out the calculations explained in this paragraph, I weight individuals according
to their sample weight provided by IPUMS. Further details are explained in Ap-
pendix B.1.
Adifferentmeasure of economic structure comes fromColombia?s FirstMan-
ufacturing Census in 1945. This Census measures more established type of man-
ufacturing than using data from employment out of Census of Population. Plants
with five or more employees provided information about employment, wages,
and financial status (Santos Cardenas, 2017). The census contains information
for 458 municipalities. It divides the establishments in 15 different sectors. Fol-
lowing Ciccone and Papaioannou (2009) and Valencia Caicedo (2019), I classified
65
the sectors in three groups according to their human capital requirements (high,
medium, low). I measure the share of population working in industrial establish-
ments with five or more employees, as well as shares of employment in each of
the three human capital groups. I interpolate 1938 and 1951 census of population
to obtain 1945 population data at the county level.
Human capital
Themain measure of human capital comes from 1973 Census of Population.
This is the first available census with individual-level data that reports county of
birth. I use this information to build a panel at the gender by cohort by county-
of-birth level for individuals born between 1900 and 1951. That is cohorts that are
between 73 and 21 years old in 1973. I measure cohorts? average year of schooling,
share of cohort-county-of birth who is literate, and occupations shares of the la-
bor force as well as labor force participation information and average household
income.
I combine 1973 cohort by county-of-birth panel data with information about
international coffee prices. I assign to each cohort the series of real international
coffee prices in Colombian pesos before they turn 18 years old. I use nominal ex-
change rate between Colombian pesos and US dollars and Colombia?s price index
before 1972 (GRECO, 2002) to estimate real international coffee prices between
1900 and 1972.
Additionally, I calculate literacy rates at the county level from 1912, 1938,
66
and 1951 Census of Population. The 1951 population census also reported the
number of schools per county.
County Characteristics
Finally, I compile a set of county fixed characteristics from different sources.
I calculate 1912 counties and IPUMS-counties? area and average altitude using
GIS software and shape files with current counties? boundaries. Similarly, I cal-
culate average terrain ruggedness using Nunn and Puga (2010) data. To estimate
connection to markets, I measure Euclidean distance from each county centroid
to Bogota, the Department?s capital, and the second largest town in 1912 differ-
ent than the Department?s capital. Climatic data comes from Dube and Vargas
(2013), who calculate long term averages of rainfall and temperature. As mea-
sures of state capacity and institutions, I use an indicator for whether each county
had Native communities in 1560 (Acevedo and Bornacelly, 2014) and the number
of land disputes between 1901 and 1931 from LeGrand (1986).
2.3 Coffee Cultivation and Structural Change in Colombia
This section presents evidence of the negative relationship between coffee
cultivation and structural transformation for Colombian counties. It documents
the correlation between coffee cultivation at the beginning of the 20th century
and labor force participation, employment in manufacturing and employment in
agriculture during different years throughout the century.
67
The main specification is given by the following equation:
yjm = ?CoffeeTrees
1920
m + ?Xm + ?d + m (2.1)
Where yjm is an outcome for county m measured in year j. Outcomes are
share of labor force employed in manufacturing and agriculture as well as share
of population in the labor force. Xm is a vector of county-level controls including
population (log), a dummy variable for Department?s capitals, linear distance to
Department?s capital, and distance to closest largest county other than the capital.
?d are Department fixed effects and m is the error term. ? is the coefficient on
coffee cultivation, measured as log of one plus the number of coffee trees in county
m around 1920.
Counties adjacent or close to one anothermight have similar shocks. In order
to account for correlated shocks across space, I adjust standard errors using arbi-
trary clustering as proposed by Colella et al. (2019), who build on Conley (1999)
to adjust for spatial correlation in 2SLS settings. My preferred specification al-
lows for decaying correlation between errors of units inside a circle with 100km
radius.8 This distance allows the spatial cluster drawn around each county to in-
clude close to 30 other counties. Moreover, 100km is roughly half of the distance
between Bogota and Medellin, Colombia?s two largest cities.9
Table 2.1 shows the relationship between coffee cultivation and the outcomes
8I implement it using the acreg command in Stata, version Beta June 2019 (1.0.1) (Colella et al.,
2019). Results are similar using 50km and 200km distance cut-offs.
9Another possibility would be to cluster standard errors on arbitrary squares from a grid over-
laid on Colombia?s map (Bester et al., 2011; Bazzi et al., 2017). Results are qualitatively similar.
68
of interest in 1973 using several specifications. 1973 is a relevant year since around
this time employment in manufacturing peaked in the country. Panel A focuses
on the share of labor force employed in manufacturing, Panel B, on the share of
labor force employed in agriculture, and Panel C, on the share of adult popula-
tion in the labor force. Column 1 shows results only controlling for population
and subsequent columns expand controls to include geographic characteristics
and Department fixed effects. Starting in Column 4, I remove counties containing
the Department capital from the sample. Those counties are less likely to grow
coffee and tend to be more urban, which could drive the results. My preferred
specification is given by Column 4. It includes geographic controls and Depart-
ment fixed effects but exclude counties containing capitals. In Columns 5 and 6, I
present differential results for men and women.
Coefficients on 1920 coffee cultivation are stable across different specifica-
tions. In general, an increase of 1% in the number of coffee trees is associated with
a decrease of 0.4 percentage points in 1973 manufacturing employment share and
with an increase of 0.6 percentage points in 1973 agricultural employment share.
These changes are equivalent to, respectively, -2% and 1.6% with respect to the
means of 19.8% and 37%. Additionally, the correlation with labor force participa-
tion is not different from zero.
These correlations mask some interesting heterogeneity across gender. The
relationship between coffee cultivation and men?s employment in both manufac-
turing and agriculture is stronger than for women. However, on average women
report lower levels of participation in the labor force and lower levels of employ-
69
ment in agriculture. This could be measurement error if domestic labor is not
registered properly on the census.
I repeat the analysis using data from 1912, 1938, and 2005. Figure 2.2 plots
OLS estimates of the correlation between coffee cultivation in 1920 and employ-
ment in manufacturing and agriculture. All estimates are equivalent to Column 4
of Table 2.1. The correlation starts out very small for 1912, only a decade after the
beginning of the expansion of coffee cultivation. For manufacturing it decreases
(becomes more negative) throughout the century, peaking in 1973 and increasing
(but still negative) in 2005. For agriculture the peak happens faster, with correla-
tions in 1938 and 1973 being almost identical.
Results discussed so far come fromCensus of Population. They include self-
reported occupation and lump together all types of manufacturing activity. In or-
der to isolate the effect of coffee cultivation during the early 20th century on struc-
tural transformation, I look at data from Colombia?s first manufacturing census,
collected in 1945. It surveyed establishments with more than five employees. It
is therefore a measure of more modern type of manufacturing. Using the same
specification described above, I focus on two different outcomes: employment
and number of establishments per county. I measure each outcome in logs and
divided by total population. Table 2.2 shows correlations using the same structure
as Table 2.1.
Panels A and B show the negative correlation between coffee cultivation in
1920 and manufacturing employment in large establishments in 1945. Panels C
and D show the negative correlation between coffee cultivation in 1920 and the
70
number of industrial establishments. The correlation is not driven by the main
centers of industrial production. Column 4 does not include Departments? capi-
tals and shows almost identical results than Column 3, which does include large
cities. Panels A and C measure dependent variables in logs, while Panels B and
Dmeasure them as shares of population and are therefore more relevant to inter-
pret. An increase of 1% in the number of coffee trees in 1920 is correlated with a
reduction of 0.03 industrial workers per 100 inhabitants in 1945. This is around
6%with respect to themean. Similarly, a 1% increase in the number of coffee trees
in 1920 is correlated with a reduction of 0.02 industrial establishments per 1,000
inhabitants in 1945. That is equivalent to around 5% with respect to the mean.
In the remaining parts of this section, I discuss why these correlations, while
illustrative, cannot be considered causal and propose different instrumental vari-
able strategies to estimate the effect of coffee cultivation on structural transforma-
tion.
2.3.1 Empirical Strategy
The negative correlation between coffee cultivation in early 20th century and
employment in manufacturing later in time could be the result of omitted county-
level characteristics that deterred the rise of manufacturing and at the same time
encouraged production of coffee. For instance, counties with a poor geographic
location might have a hard time importing capital goods to set up manufacturing
firms, which might drive them to take up economic activities that suffer less from
71
transportation costs. One of such activities at the beginning of the 20th century
was coffee production. Coffee was suitable to be transported by mules, which
were ideal to overcome Colombia?s difficult geography. Under that scenario, a
negative correlation between coffee and structural transformationmight be driven
by geography rather than by the expansion of the export sector.
Another story with similar implications would be one where the only coun-
ties which produce coffee are those with low domestic market access, since cof-
fee was primarily exported, while manufacturing entrepreneurs located close to
main population centers. One could also be worried coffee counties start out the
20th century with lower levels of public goods or lower state capacity, given the
colonization patterns described in Section 2.1. With these ideas in mind, the pre-
vious OLS results controlled for geographic characteristics intended to capture
market access and exposure to the State. I showed the negative correlation be-
tween coffee production and manufacturing did not change when those controls
were included. Moreover, the correlation did not changewhen biggest population
centers were excluded from the sample.
Finally, while I am estimating the effect of the exposure to the expansion of
coffee cultivation on structural transformation, my measure of coffee cultivation
is taken from the 1920s and potentially suffers from measurement error. For in-
stance, some counties might have expanded coffee cultivation in the 1920s when
prices were relatively high but went back to a lower level after the Great Depres-
sion. To partially deal with measurement errors concerns, Appendix B.3 repro-
duces the main analysis using only the extensive margin of coffee cultivation- i.e.
72
a dummy equal to one for counties with more than one coffee tree in 1920.
Before turning to the main empirical strategies, Figure 2.3 illustrate some of
the dimensions over which coffee counties differed from the rest. The figure plots
standardized coefficients (and 95% confidence intervals based on robust standard
errors) out of OLS regressions of variables in y-axis over a dummy for coffee coun-
ties. In 1912, coffee countieswere, on average,more literate and employed a higher
share of labor force in agriculture, however there were no differences in the share
of labor force employed inmanufacturing or the level of population density, which
might alleviate some of the concerns described above.
Geographically, however, there are considerable differences between the two
groups of counties. Specifically, coffee counties are located at a higher altitude and
their terrain is considerably more rugged. They are closer to Colombia?s capital,
Bogota, and to the Department capital. Interestingly, there are no differences in
terms of patterns of colonization on average. Coffee counties are as likely as other
counties to have had presence of native population when the Spanish arrived
around 1560. Places with native population were generally settled first, while the
frontier around 1600 took at least two centuries to be settled. Finally, there were
around the same number of land disputes during the first three decades of the
20th century, which might be indicative of the security of property rights and the
quality of institutions at the time.
These results highlight that features related to transportation costs and geog-
raphy, rather than market access or state presence, are the main source of omitted
variable bias. To deal with it, I exploit two exogenous sources of variation in a
73
county?s suitability for growing coffee. The main idea is that by exploiting cof-
fee suitability, I isolate the effect of coffee exporting on structural transformation,
rather than the effect of location or transportation costs.
Climate and Attainable Yields
The first source of variation is given by local climatic conditions that make
some counties more productive at growing coffee. I use two different but related
approaches. First, I use data from FAO?s Global Agro-Ecological Zones project
(FAO-GAEZ). The project produces information on maximum attainable yields
for different crops at high geographical resolution by combining data on climate
and crop-specific features. These potential yields do not depend on actual produc-
tion and are calculated for different levels of inputs. I use rain-fed Coffee Maxi-
mum Attainable Yield with intermediate inputs and aggregate it to the county
level using area-weighted averages. Then I normalize yields from 0 to 100 by di-
viding by the maximum value. Figure 2.4 shows the variation on the instrument
across the country.
The first stage equation is given by:
CoffeeTrees1920m = ?1Pot.Yieldm + ?Xm + ?d + ?m (2.2)
Where Pot.Yieldm is FAO maximum attainable yield and ?d is a set of De-
partment fixed effects.
Second, I follow Dube and Vargas (2013) and instrument coffee cultivation
74
with long term averages of rainfall and temperature levels at the county level. In
theory, these two approaches are identical to one another with the only difference
that the rainfall and temperature instrument does not rely on a climatic model like
the one used to calculate attainable yields. The first stage is given by:
CoffeeTrees1920m = ?1rainm + ?2tempm + ?3rainm? tempm +?Xm + ?d + ?m (2.3)
Fuzzy Regression Discontinuity in Altitude
The previous approach is useful to isolate coffee cultivation motivated by
productivity reasons. However, since it only uses climatic conditions some of the
concerns about location and geography might still apply. In other words, the IV
strategies described above could compare counties with high suitability located
close to the oceanwith places in the interiorwith the same climate. Therefore, I in-
troduce another identification strategy that does not rely directly on weather. The
strategy isolates more comparable counties in terms of geographic characteristics.
Figure 2.5 plots CoffeeTrees1920m for counties in different altitude bins. The
figure focuses on counties above 1,800 meters above the sea level. The vertical
line is located at 2,400 meters. The slope of coffee cultivation is negative below
the 2,400 meters cut-off and flat above. Moreover, there is a downward jump at
2,400m of altitude. Around that altitude temperature at nights gets sufficiently
cold that coffee does not grow as well as a couple hundred meters below. The
discontinuity might also be due to information pamphlets distributed by late 19th
75
century investors who were aiming at getting more landowners into the coffee
business. These recommendations, compiled in the bookMemorias sobre el cultivo
del Caf? (Saenz, 1892), provided temperature bounds for the optimal production
of the crop. In the 19th century those bounds were between 24 and 16 degrees
Celsius. Since temperature in Colombia is driven by altitude, temperature bounds
translate directly into altitude bounds between 400 and 2,400meters above the sea
level. Moreover, some of the pamphlets directly provided information on altitude
and temperature of specific towns to make it easy for landowners to figure out
whether their land was located inside the altitude bandwidth.
Figure 2.6 shows one of those instances in a pamphlet written by Mariano
Ospina Rodriguez in 1880. Ospina was Colombia?s president in 1857 and is con-
sidered one of the pioneers of coffee cultivation in Colombia. He started growing
coffee in his family?s farmwell before the expansion in the first two decades of the
20th century.
From Figure 2.5, some counties below the threshold did not grow coffee in
the 1920s and some counties above the threshold had a positive number of coffee
trees. Therefore, the setting is not one of a sharp regression discontinuity. Rather,
I use the discontinuous fall in the probability of growing coffee in 1920 as an in-
strument for actual coffee cultivation. In otherwords, I instrument CoffeeTrees1920m
with a dummy variable equal to one for counties above the altitude threshold and
a simple polynomial in altitude. This allows for coffee cultivation to fall with alti-
tude and even for the slope to change below and above the threshold. Identifica-
tion comes from a discontinuous jump at the threshold.
76
In this fuzzy regression discontinuity design (FRDD) (Angrist and Pischke,
2008b), the first stage is given by:
CoffeeTrees1920m = ?1abovem+?2altitudem+?3above x altitudem+ ?d+ ?m (2.4)
Where above is a dummy variable equal to one for counties above 2,400 me-
ters of altitude. Altitude enters the equation centered at 2,400meters, both linearly
and interacted with above.
The benefits of using fuzzy regression discontinuity design are evident once
we compare counties above and below the threshold. Figure 2.8 plots coefficients
fromOLS regressions of variables on the y-axis on abovem dummy variable. Most
of the coefficients are very close to zero, with the exceptions of literacy rate in
1912 and terrain ruggedness (both lower for counties above). Altitude is higher
by construction.
Notice, however, that those means test are not necessarily all that is needed
for using the fuzzy regression discontinuity as instrument for coffee cultivation.
Importantly, the identifying assumption is that no other factor should change dis-
continuously at 2,400 meters. Only the probability of growing coffee. To test for
discontinuities in other county characteristics, Table 2.3 shows the results from
OLS estimation using specifications identical to equation 2.4, but plugging in as
dependent variable all the factors represented in Figure 2.8. There are no dis-
continuities for most characteristics. For manufacturing employment in 1912, the
coefficient on above is marginally significant and on the opposite direction than
77
expected: places above, which do not grow coffee, employ a slightly lower share
of population in manufacturing in 1912.
The twodifferent approaches (IV andFRDD)present a clear trade-off. While
IV strategies (FAOdata and rainfall, temperature polynomial) use all the available
counties, the FRDD strategy potentially gives amore reliable estimate since it uses
very comparable ?treatment? and ?control? groups. Figure 2.7 illustrate the trade-
off in terms of sample size. Figure (a) shows the 759 counties populated in 1928
and classify them by the number of coffee trees per square kilometer. Figure (b)
highlights counties above 1,800 meters of altitude and classifies them by their lo-
cation with respect to the altitude threshold.
2.4 The Effect of Coffee Cultivation on Structural Transformation
This section estimates how exposure to coffee exports at the beginning of the
20th century shaped local development and structural transformation in Colom-
bian counties. Using a sample of 550 IPUMS-counties, I document a sharp pattern
of divergence in economic structure between coffee producer counties and other
counties. In particular, the share of employment inmanufacturing increased faster
during the first part of the 20th century, up to 1973, in counties which did not pro-
duce coffee. Meanwhile, counties which produced coffee remained mostly spe-
cialized in agriculture. I focus first on results using 1973 Census of Population.
Then I show how did coffee cultivation affect economic structure at various peri-
ods during the century. The following section expands on an important mecha-
78
nism to explain this divergence: human capital accumulation. Finally, Section 2.7
estimates the long-term impact of coffee cultivation on income and urbanization.
Coffee cultivation during the early 20th century had a negative and sizable
effect on structural transformation throughout the century. Results fromTable 2.4,
Panel A show an increase of 10% in the number of coffee trees planted before
1920 led to a reduction of around 0.05 percentage points in the share of labor
force employed in manufacturing in 1973. This effect is equivalent to a reduction
of 0.2% with respect to the average county. To put it differently, going from the
median level of coffee cultivation to the 75th percentile in 1920would decrease the
1973 share of labor forceworking inmanufacturing by 1.6 percentage points or 8%
relative to the mean. Going from the median level to the 99th percentile would
decrease manufacturing employment by 15.2% relative to the average county.
Panel B shows the positive effect of coffee cultivation on agricultural em-
ployment. An increase of 10% in coffee cultivation prior to 1920 would increase
employment in agriculture in 1973 by 0.08 percentage points, or 0.21% relative to
the mean of 37% of the labor force. The estimated magnitude of these effects is
the same regardless of the method used to instrument for coffee cultivation be-
fore 1920. Column 2 shows results using FAO coffee attainable yields. Column
3 shows results using a simple polynomial in rainfall and temperature. Results
from these two methods are expected to be similar since they use the same set of
counties and exploit the same source of variation (climate). Column 4, however,
restricts the sample to counties with average altitude higher than 1,800meters and
instruments coffee cultivation using the fuzzy regression discontinuity approach.
79
In other words, the set of counties ?treated? by coffee availability are between
1,800 and 2,400 meters above the sea level. The fact that the results are similar
between columns 2 and 3 and column 4 is evidence that the average treatment
effect of coffee on structural transformation might be homogeneous, at least on
dimensions related to altitude, market access, and transportation costs.
The effect of coffee cultivation on structural transformation in terms of em-
ployment could be driven by differences in labor force participation. I alleviate
those concerns first by measuring employment as shares of labor force. More im-
portantly, Panel C shows coffee cultivation had no significant effect on labor force
participation. Even for the case where there is a statistically significant positive ef-
fect, it is tiny. Column 2 in Panel C implies that an increase of 10% on the number
of coffee trees planted before 1920would increase labor force participation in 1973
by 0.07% with respect to the mean. This effect is at least an order of magnitude
smaller than the effects in Panels A and B for employment by sector.
The estimates discussed above rely on exogenous variation provided by three
different sets of instruments. Panel D presents evidence on the relevance of the
three sets of instruments. It regresses the log of one plus the number of coffee
trees in 1920 on coffee attainable yield (column 2), on a polynomial on rainfall
and temperature (column 3), and on altitude, a dummy equal to one for counties
above 2,400 meters of altitude, and an interaction between the ?above? dummy
and altitude (column 4). All instruments are significant and sizable. Moreover,
the table also shows the F statistic for the excluded instruments on each one of the
first stages. I do not find evidence of weak instruments.
80
The effect of coffee exports on employment by sector varied through the cen-
tury, as Figure 2.9 shows. The figure complements Figure 2.2 by adding IV esti-
mates from 2SLS using attainable yields from FAO (like Column 2 in Table 2.1)
and FRDD estimates (like Column 4). In 1912, there was no difference in sectoral
employment between coffee counties and non-coffee counties. In other words,
there was no effect of potential to cultivate coffee on the share of labor force em-
ployed in manufacturing, though there was some small positive effect on agricul-
ture. By 1938, the effect became large and significant. It was negative for man-
ufacturing and positive for agriculture. This effect remained through 1973, as
explained earlier. Employment in manufacturing peaked in the country in the
1970s. After that most of the population shifted to services. Consequently, the ef-
fect of coffee cultivation in 1920 on manufacturing employment in 2005 is smaller,
though significant.
In addition to using data collected from employees, coming out of Census of
Population, I use data from employers from Colombia?s first manufacturing cen-
sus from 1945. After the rapid growth in manufacturing in the 1930s, the country
surveyed all industrial establishments with more than five workers. It is there-
fore a more formal sample of manufacturing establishments whichmight provide
more information on the effect of coffee cultivation on the process of structural
transformation.
Table 2.5 estimates the effect of early coffee cultivation on industrial employ-
ment and the number of industrial establishments in 1945. Column (1) repro-
duces results from Table 2.4, Column 4, Panels B and D as reference. Columns 2
81
to 4 follow the same order than results presented in Table 2.4. Once again, results
across different instruments have similar magnitudes and directions despite their
underlying differences in sample size and composition.
An increase of 10%on coffee trees in 1920, decreases the number of industrial
establishments with more than five employees in 1945 by 1.7%with respect to the
mean (Panel B). It would also decrease the share of population working in indus-
trial establishments by 0.08 percentage points, or 16% with respect to the mean.
Panel D presents the three different first stage estimations. All specifications in-
strument coffee cultivation with instruments that are not weak. Importantly, Col-
umn 4 shows evidence of the negative discontinuous jump in the probability of
growing coffee above 2,400 meters of altitude.
2.5 Channel: Human Capital
Colombia?s firstmanufacturing census provides a good setting to studywhether
human capital had something to do with the effect of coffee cultivation on man-
ufacturing. I adapt Ciccone and Papaioannou (2009) classification for industrial
sectors in the United States to the Colombian context and divide 15 sectors com-
piled by 1945 Census into three groups according to their human capital intensity.
High human capital sectors include Beverages, Instruments, Arts (Printing), and
Chemicals. Medium human capital sectors include Tobacco, Minerals, chapter,
Rubber, and Metal. Finally, Low human capital sectors include Leather, Textiles,
Clothes, Wood, and Food. I use specification given by equation 2.1 for employ-
82
ment and number of establishments in each of the three groups. I plot the coef-
ficients on coffee cultivation and their standard errors. Coefficients come from a
2SLS estimation using Coffee attainable yields from FAO as an instrument, like
Table 2.5, Column 2.
Figure 2.10 shows the effect of coffee cultivation in 1920 on (a) employment
in manufacturing and (b) number of industrial establishments in 1945 by human
capital intensity. While there is not a difference between coffee and non-coffee
counties in terms of employment in more basic sectors like textiles or food, the
largest difference shows up for sectors with high intensity of human capital. In
other words, the effect coffee cultivation has on structural transformation seems
to be concentrated in activities that require a more educated labor force.
This evidence goes in linewith the hypothesis described above and themod-
els introduced byPorzio and Santangelo (2019) andCaselli andColeman II (2001).
Figure 2.11 complements the insight from the previous evidence. It shows the dif-
ference between coffee and non-coffee counties in terms of education of the labor
force in 1973. Each point represents the difference in average number of years of
schooling in coffee counties minus the average number of years of schooling in
non-coffee counties for people born in each cohort. Though it only shows data
for cohorts born from 1900 onward, there is a clear trend: the labor force in coffee
counties becomes relatively less educated in the first decades of the 20th century.
This reduction in the level of education seems to be negatively correlated with the
pattern of coffee production shown by the short-dashed line.
Using individual level data from 1973 Census of Population, I show that in
83
fact coffee cultivation reduced schooling and made the labor force more biased
toward staying in agriculture. Therefore, manufacturing appeared in counties
which did not produce coffee.
Empirical Strategy
The 1973 Census of Population allows me to observe in which county and
year of birth for a 10% sample of Colombians. People born in the first half of the
20th century in counties suitable to produce coffee were more exposed to the first
large scale exporting industry in the country. For them, the opportunity cost of at-
tending school was higher. Moreover, the specifics of coffee?s production function
discussed on section 2.1 increased parent?s opportunity cost of sending their chil-
dren to school. These incentives away from education were potentially stronger
during years when the coffee price was higher.
I test these hypotheses by estimating the effect of coffee price during school
age on kids? education and occupation as adults. I estimate the following equa-
tion:
ymcg = ?CoffeeTrees1920 ? Price5,16m c + ?g + ?m + ?c + mcg (2.5)
Where ymcg is average education or occupation outcome for gender g, cohort c,
born in countym. The coefficient of interest is ?, which captures the effect of coffee
price shocks between a given cohort is 5 and 16 years old. Price5,16c is average log
real coffee price in New York between years c+ 5 and c+ 16. ?g, ?m, ?c are gender,
county, and cohort fixed effects. The unit of observation is a gender by county-of-
84
birth by cohort cell. Each cell is weighted by the inverse of its variance.
Comparable with the empirical strategies described in section 2.3, I instru-
ment Coffee trees1920 using three different instrumental variables: coffee attain-
able yield, rainfall and temperature polynomial, and a fuzzy regression disconti-
nuity in altitude.
2.5.1 Growing Up During Coffee Price Booms Reduces Schooling
and Employment in Manufacturing for Cohorts Born in Cof-
fee Counties
Table 2.6 shows the effect of coffee price shocks on educational attainment
in 1973. Panel Ameasures coffee shocks as the interaction between the number of
coffee trees in 1920 and the log average coffee price between a cohort is 5 and 16
years old. Panel B changes the number of coffee trees by a simple dummy equal to
one for counties with some coffee cultivation in 1920. Column 1 presents the OLS
results. Columns 2 to 4 show results from instrumenting coffee shocks using the
variables detailed in Panel C. Interpretation of results in Panel B is more straight-
forward. For simplicity, I discuss results from instrumenting coffee shocks with
attainable yields interactedwith average school age price (Column 2): cohorts ex-
posed to average coffee price 10% higher, born in coffee counties accumulate 0.7%
fewer years of education. Another way to put it would be to compare 1910 and
1940 cohorts. The latter cohort experienced average coffee princes during school
age 140% higher than the former due to differences on coffee prices. Therefore,
85
the cohort born in 1940 acquired on average 9% fewer years of education.
Panel C shows the reduced form, that is the effect of the instrument on av-
erage years of education. Interestingly, even though the second stage is not sig-
nificant and small for column 4, the reduced form shows a coefficient with a very
similar magnitude than the 2SLS results for columns 2 and 4, Panel B, but with the
opposite sign, as expected. Counties above the threshold did not produce coffee
and therefore are not hit by shocks to prices.
Table 2.7 shows related results but looking at economic occupation and in-
come. In general, children born in coffee counties during periods with high coffee
prices not only accumulated less human capital but also were less likely to work
in manufacturing as adults. Take the result from Panel A, Column 2. Comparing
again cohorts born in 1910 and 1940, the latter cohort is almost 10% less likely to
work in manufacturing due to the availability of coffee cultivation that made that
individual drop out of school.
Similarly, though less precisely estimated, I find a positive effect of coffee
shocks during school age on the probability of employment in agriculture as adults.
Finally, in Panel C, I show that cohorts born in coffee counties, exposed to higher
coffee prices during school age have lower household income, as measured by the
first vector out of a Principal Component Analysis on a matrix of house character-
istics and durable goods.
Notice the decline inmanufacturing employment is very similar than the de-
cline in schooling. Since schooling decisions are taken earlier (or simultaneously)
than occupation decisions, I explore the effect of coffee shocks on employment in
86
manufacturing that is channeled through education by doing a mediation anal-
ysis proposed by Dippel et al. (2019b). In other words, I estimate the effect of a
cohort?s schooling on the share of its members working in manufacturing in 1973
by instrumenting education using the interaction term between coffee attainable
yields and average prices during school age, conditional on coffee shocks (dummy
for coffee interacted with average coffee prices during school age). This approach
provides an idea of howmuch of the effect of coffee prices on occupation is acting
through human capital accumulation and what fraction is going through other
channels. Specifically, I estimate the following equation:
Occup. = ? Education +?CoffeeTrees1920?Price5,16mcg 1 mcg m c +?g+?m+?c+mcg (2.6)
And instrument schooling using the following equation:
Education = ? Attn. Yield ? Price5,16+?CoffeeTrees1920mcg 1 m c m ? Price5,16c + ...
?g + ?m + ?c + ?mcg
(2.7)
This approach, however, relies on one very strong assumption: the concerns
about endogeneity between coffee shocks and manufacturing are the same as the
concerns about endogeneity between coffee shocks and schooling (Dippel et al.,
2019a). Using this approach, the total effect is given by table 2.7. The share of the
effect of coffee shocks on manufacturing that goes through schooling decisions
87
is between 80% and 96%.10 The share of the effect depends on the measure of
education used (cohort?s average years of education or literacy rate) and on the
measure of coffee cultivation (continuous or dummy). Again, these results only
hold under the assumption about the sources of endogeneity being the samewhen
estimating the effect of coffee shocks on education thanwhen estimating the effect
of coffee shocks on occupation.
2.5.2 Supply and Demand of Schooling
So far, I have showed evidence on the negative effect of coffee cultivation
on education and manufacturing employment. My main conjecture is that coffee
cultivation increases opportunity cost of attending school. Differences in the op-
portunity cost of schooling generates differences in cohorts? levels of education.
As a consequence, counties producing coffee developed lower supplies of non-
agricultural workers. According to Caselli and Coleman II (2001) and Porzio and
Santangello (2019), these differences shaped the process of structural transforma-
tion by reducing the availability of skilled workers for manufacturing.
In the past section I showed evidence that cohorts exposed to higher coffee
prices acquired lower levels of education. However, these effects of coffee culti-
vation and shocks on education could come from both supply and demand. One
potential explanation is that people?s demand for schooling goes down with the
possibility of producing coffee. However, another possibility is that simply the
supply of schooling in coffee counties goes down when prices go up with respect
10Detailed results in Appendix B.2.
88
to non-coffee counties. This might occur if, for instance, landowners benefit from
lower wages and a readily available labor force (Galor et al., 2009; Galiani et al.,
2008).
To explore the sources of differences in schooling between coffee and non-
coffee counties, I exploit historical data on land inequality using the First Coffee
Census. I observe the mean and median farm size in terms of coffee trees for
each county with positive number of coffee trees in 1920. My conjecture is that in
places with higher land inequality within coffee landowners, the higher the mar-
ket power of landowners. This would have two consequences: first, they would
have perhaps more political power to block funding and construction of schools.
Second, they could potentially behave like a monopsony and wages would not be
as responsive to changes in international price as counties with lower inequality.
In other words, if the effect of coffee shocks on education is stronger for counties
with high inequality than for those with low inequality, that would be suggestive
of the effect of coffee on education being mainly driven by supply of schooling.
Table 2.8 shows the effect of coffee shocks on schooling, literacy, employment
in manufacturing, employment in agriculture, and household income in 1973 for
different samples. Column 1 shows the full sample for comparison. Column 2
restricts the analysis only to counties with at least one coffee tree in 1920. Column
3 and 4 splits the sample on Column 2 according to the median of the level of land
inequality (mean/median farm).
The negative effect of coffee shocks on education and income and the pos-
itive effect on employment in agriculture are concentrated in counties with low
89
land inequality within coffee landholders. This result is consistent with the hy-
pothesis that the effect of coffee cultivation on education is coming from changes
in the demand for education. Of course, this is only suggestive evidence given
that differences in inequality might be correlated with some of the forces behind
differences in education or occupation.
I explore data on the number of schools per 10,000 inhabitants in 1951 to
get at more direct evidence on the effect of coffee cultivation on the provision of
education. Unfortunately, that is the only year with readily available information
on provision of education at the local level for the first half of the 20th century.
Table 2.9 shows the effect of coffee cultivation on the number of schools, instru-
menting coffee cultivation with FAO attainable yields data. Panel A measures
coffee cultivation with a dummy variable. Column 1 shows OLS results while
Column 2 shows 2SLS results from using the full sample. Coffee counties have
1.05 less schools per 10,000 inhabitants than non-coffee counties in 1951. This ef-
fect is equal to 135% of themean of 0.64 schools per 10, 000 inhabitants on average.
From Panel B, an increase of 10% in the number of coffee trees decreases the num-
bers of schools per 10,000 inhabitants in 1%. Moreover, the effect has relatively
the samemagnitude when focusing only on counties with some coffee cultivation
(Column 3). But perhaps it is more interesting that the negative effect of coffee
cultivation in 1920 on the number of schools in 1951 is only present and stronger
(thought only significant at 15%) for counties with high levels of inequality.
In other words, these results taken together suggest that the negative effect
of coffee cultivation on human capital and structural transformation comes both
90
from supply and demand of education. On one hand, coffee counties have fewer
schools, especially in high inequality counties, where landowners havemore power
to block schooling and guarantee a higher supply of agricultural workers. On the
other hand, coffee shocks have a stronger negative impact in counties with low in-
equality, where landowners have potentially less power to fixwages they fluctuate
more with international prices.
2.6 Channel: Linkages
I have discussed evidence of the effect of coffee cultivation on industrial-
ization focusing on the role of human capital accumulation, which increased the
supply of non-agriculture workers in counties not suitable to produce coffee. One
concern is that the effect is coming from counties isolated from the rest of the
economy. In that scenario places with coffee cultivation should be better off spe-
cializing on coffee in the long run even if they end up being poorer than other
counties because they would not develop manufacturing otherwise.
Though this concern is alleviated by the comparison between coffee and
non-coffee counties in Figure 2.3, I explore if the effect of coffee cultivation varies
with respect to how connected with other sectors in the economy a county is. One
way I can study the level of linkages is by using the presence of threshingmachines
in some counties. Threshing is the last process coffee cherries undergo before be-
ing exported. It is a process through which by tumbling in a large machine, cof-
fee cherries lose their covering husk or pergamino. These machines were used by
91
owners of different farms, required heavy machinery, and expertise. Therefore,
counties with this part of the coffee industry are potentially more connected with
other sectors. For example, through transportation networks, through workers
who can operate the machines, or banks who can fund their purchase, and so on.
Out of the 710 counties, 440 had some coffee cultivation in 1920. Around 120
coffee counties also had threshing establishments. These counties had around 0.4
machines per farm. The county with the highest number of machines had 65.
I estimate separate coefficients for the effect of producing coffee without a
threshing machine and producing coffee with threshing machines. I argue the
difference between both would give an estimate of the effect of linkages if the
only difference between counties with and without threshing is the fact that some
have linkages. Table 2.10 shows the results. Columns 1 and 2 replicate OLS and
2SLS estimates from section 2.4. Columns 3 and 4 split the dummy for coffee
cultivation in 1920 in two, according to their linkages. I find that the effect is very
close to one another. If anything, the effect of coffee onmanufacturing for counties
with threshing machines is larger. The fact that threshing may not be exogenous
should give us pause to put too much weight on the coefficients.
2.7 Long Term Effects on Urbanization and Income
Finally, I replicate the main results using household income data for 2005
adults and counties? poverty rate as calculated by the Social Prosperity Depart-
ment. Table 2.11 show the main results for all the different instruments for coffee
92
cultivation. An increase in the number of coffee trees in 1920 by 10% would in-
crease poverty rate of a county in 2005 by 10% increases poverty rate in 2005 by
0.1% with respect to the mean. In other words, going from the median level of
coffee cultivation in 1920 to the 75th percentile would increase a county?s poverty
rate in 2005 by 3.1%.
2.8 Concluding Remarks
This chapter illustrate how the opportunity to trade certain agricultural com-
modities had negative effects on structural transformation and development in
the long run. Using data from Colombian counties, it shows that counties pro-
ducing coffee in 1920 had slower growth in manufacturing sector than other com-
parable counties. Coffee cultivation had a negative long-run effect on income and
poverty rates. Additionally, this chapter provides empirical support to Caselli and
Coleman II (2001) theory about supply-side mechanisms behind structural trans-
formation. In particular, it highlights that slower growth in the supply of skilled
workers can also slow down structural transformation. Given coffee?s labor inten-
sive production function, this chapter complements evidence fromCarrillo (2019)
about the negative effect of coffee cultivation on human capital accumulation.
The evidence discussed in this chapter suggests evaluating the effects of the
first wave of globalization depends on the context. Specifically, it depends on
features of commodities? production function which shape incentives to accumu-
late human capital and select high productivity occupations. Countries which
93
exported more than one important agricultural product might be ideal settings to
sort out which features of commodities? production functions mediate the effect
of agricultural exports on structural transformation.
Figures
Figure 2.1: Patterns of Coffee Exports and Manufacturing in Colombia
(a) Value of Main Exports
400 Coffee
Oil
Banana
Gold
300
200
100
0
1900 1910 1920 1930 1940 1950
Year
(b) Employment in Manufacturing and Services (% of labor
force)
.4
Manufacturing Employment
Services Employment
.3
.2
.1
0
1900 1912 1938 1973 1993 2000
Year
Note: Figure 2.1a shows the real value of Colombia?s main exports in 1951 US dollars. It uses data from
GRECO (2002). Figure 2.1b shows shares of labor force employed in manufacturing and services. It uses data
from Census of Population. 1912 and 1938 Census were digitized for this chapter. 1973 and 1993 Census are
available in IPUMS-international.
94
Share of Labor Force Exports in Million Real US$
Figure 2.2: Correlation Coffee Cultivation and Structural Transformation
(a) Dep. Variable: Manufacturing Employment
(% of Labor Force)
0
-.4
-.8
-1.2
1912 1938 1973 2005
(b) Dep. Variable: Agricultural Employment (%
of Labor Force)
.8
.6
.4
.2
0
1912 1938 1973 2005
Note: The figures plot the coefficients of the relationship between coffee cultivation in 1920
and employment shares in manufacturing and agriculture during the 20th century. Esti-
mates control for gender and Department fixed effects and geographic controls. Capital
cities are excluded. Lines represent 95% confidence intervals based on Conley (1999) stan-
dard errors as described in section 2.3.
95
Figure 2.3: Coffee vs. Non-Coffee Counties Comparison
Literacy, 1912
Share Manufacturing, 1912
Share Agriculture, 1912
Pop. Density, 1912
Altitude
Distance Bogota
Distance Capital
Ruggedness
Soil Quality
Native Pop., 1560
Land Disputes, 1901-31
-1 -.5 0 .5 1
Note: Figure plots standardized coefficients on an indicator variable equal to one if the county had
some coffee trees in 1920. Dependent variables are detailed on the vertical axis. For instance, the
first coefficient means coffee counties had higher literacy rate in 1912 on average than non-coffee
counties. Lines represent 95% confidence intervals based on robust standard errors.
96
Figure 2.4: Coffee Potential Yield
82.83 - 100.00
72.84 - 82.83
64.92 - 72.84
55.95 - 64.92
48.81 - 55.95
30.10 - 48.81
0.00 - 30.10
0.00 - 0.00
No data
Note: Map shows the averagemaximumattainable yield at the county level using data
from FAO-GAEZ. Yields are estimated using rain-fed conditions with intermediate
level of inputs. Darker shades represent higher yields.
97
Figure 2.5: Discontinuity in the Probability of Coffee Cultivation at 2,400mts
10
8
6
4
2
0
1800 2200 2600 3000
Altitude (mts.)
Note: Figure plots average log of one plus number of coffee trees for counties grouped in equal sized bins in
terms of altitude. The figure is restricted for counties abouve 1,800 meters above sea level.
Figure 2.6: Coffee Promotion Pamphlet, 1880
Note: Excerpt of a pamphlet promoting coffee cultivation written byMariano Ospina in 1880. Ospina suggests
coffee grows in places at or below 17 degrees celsius and then benchmarks that temperature with different
towns. Rio-negro is located at 2,200 meters and average temperature was 17 degrees. Santa-Rosa de Osos and
Sonson are located at 2,500 and 2,450 meters of altitude, respectively.
98
log 1+ Coffee Trees (1925)
Figure 2.7: Counties in IV and FRDD Samples
(a) IV Sample: Counties by Coffee trees (b) FRDD Sample: Counties above 1.8km
per sq. km, 1920 by Side of Discontinuity
32.4 - 437.0
7.3 - 32.4
2.1 - 7.3
0.2 - 2.1 Below 2,400m
0.0 - 0.2
0.0 - 0.0 Above 2,400m
No data No data
Note: The maps illustrate the main samples used in each empirical strategy. Figure (a) shows the number of
coffee trees per square km in 1920 for counties with positive population in 1912. Figure (b) shows counties
above 1,800 meters classified by whether or not they are above 2,400 meters. Both maps use current county
borders.
99
Figure 2.8: Comparison Counties Above and Below 2,400 mts. of Altitude
Literacy, 1912
Share Manufacturing, 1912
Share Agriculture, 1912
Pop. Density, 1912
Altitude
Distance Bogota
Distance Capital
Ruggedness
Soil Quality
Native Pop., 1560
Land Disputes, 1901-31
-1 -.5 0 .5 1
Note: Figure plots standardized coefficients on an indicator variable equal to one if for counties
with altitude higher than 2,400 meters. Dependent variables are detailed on the vertical axis. For
instance, the first coefficient means counties above 2,400 meters had lower literacy rate in 1912 on
average than counties between 1,800 and 2,400 meters of altitude. Lines represent 95% confidence
intervals based on robust standard errors.
100
Figure 2.9: The Effect of Coffee Cultivation on Structural Transformation
(a) Dep. Variable: Manufacturing Employment (% of Labor
Force)
0
-.4
-.8
OLS
IV
RDD
-1.2
1912 1938 1973 2005
(b) Dep. Variable: Agricultural Employment (% of Labor
Force)
2
1.5
1
.5
0
OLS
IV
RDD
-.5
1912 1938 1973 2005
Note: The figure illustrate the effect of coffee cultivation on employment in manufacturing
and agriculture for different years. Squares represent estimates coming from OLS regres-
sions. Circles represent estimates from 2SLS regressions using coffee potential yields as in-
strument for coffee cultivation in 1920. Diamonds display estimates using fuzzy regression
discontinuity in altitude, focusing on counties above 1.8km of altitude. All specifications
control for gender and Department fixed effects and geographic controls. Capital cities are
excluded. Lines represent 95% confidence intervals based onConley (1999) standard errors,
as described in section 2.3.
101
102
Figure 2.10: Effect of Coffee Cultivation on Industrialization by Human Capital Requirements in 1945
(a) Dep. Variable: Manufacturing Employment by sector (b) Dep. Variable: Industrial Establishments per 1,000
(% of population) inhab. by sector
High HK High HK
Medium HK Medium HK
Low HK Low HK
-.15 -.1 -.05 0 .05 -.15 -.1 -.05 0
Note: Each circle represents the coefficient of log coffee trees in 1920 (a) industrial employment and (b) industrial establishments in 1945. Each
subfigure shows three different models, one for each group of industrial sectors according to their human capital intensity (high, medium, low).
Coefficients stem from a 2SLS regression where log coffee trees is instrumented by coffee attainable yields. All specifications control for population
in 1938 (log), distance to department?s capital, distance to second largest market, and Department fixed effects. All specifications exclude capital
cities. Lines represent 95% confidence intervals based on Conley (1999) standard errors, as described in section 2.3.
Figure 2.11: Differences in Cohorts? Schooling between Coffee and Non-Coffee
Counties
.4 60
Difference Schooling Coffee - Non Coffee counties
Smoothed Average
Avg. Coffee Price, 5 to 16 y.o.
50
.2
40
0
30
-.2
20
-.4 10
1900 1910 1920 1930 1940 1950
Cohort
Note: The circles show differences in average schooling in coffee counties and non-coffee counties
for individual cohorts born in 1900 and 1950, using data from 1973 Census of Population. The long
dashed line shows the smoothed average over time of schooling differences. The short dashed
line (right axis) plots the average real coffee price between 5 and 16 years old for each cohort born
between 1900 and 1950.
103
Years of Schooling
Real Coffee Price, Cents per pound
Tables
Table 2.1: Coffee Cultivation and Economic Structure, 1973
(1) (2) (3) (4) (5) (6)
Sample restriction: Men Women
Panel A: Dep. var.: Share of Labor Force in Manufacturing, 1973
log Coffee trees1920 -0.002??? -0.002??? -0.004??? -0.004??? -0.006??? -0.003???
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Mean Dep. Var. 0.201 0.201 0.201 0.198 0.191 0.205
r2 0.082 0.104 0.201 0.198 0.387 0.279
Panel B: Dep. var.: Share of Labor Force in Agriculture, 1973
log Coffee trees ???1920 0.005 0.005??? 0.006??? 0.006??? 0.011??? 0.002?
(0.001) (0.001) (0.001) (0.001) (0.002) (0.001)
Mean Dep. Var. 0.363 0.363 0.363 0.370 0.635 0.108
r2 0.763 0.765 0.795 0.804 0.443 0.335
Panel C: Dep. var.: Share of Population in Labor Force, 1973
log Coffee trees -0.001?1920 -0.000 -0.000 -0.000 0.002??? -0.002???
(0.0003) (0.0003) (0.0003) (0.0003) (0.0004) (0.0006)
Mean Dep. Var. 0.531 0.531 0.531 0.530 0.891 0.169
r2 0.967 0.967 0.973 0.974 0.294 0.540
Counties 563 563 563 550 548 548
Geo Controls X X X X X
Department FE X X X X
Capitals X X X
Note: Each Panel estimates the correlation between coffee trees in 1920 and measures of economic structure in
1973. All specifications control for population in 1973 (log). Geo controls include: distance to department?s
capital, distance to second largest market, and a dummy for capital cities. Conley (1999) standard errors as
described in section 2.3 in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
104
Table 2.2: Coffee Cultivation and Manufacturing in 1945
(1) (2) (3) (4)
Panel A: log Industrial workers, 1945
log Coffee trees1920 -0.034??? -0.057??? -0.065??? -0.066???
(0.011) (0.012) (0.015) (0.015)
r2 0.343 0.395 0.443 0.360
Panel B: Industrial workers per 100 inhab., 1945
log Coffee trees1920 -0.011 -0.023?? -0.034??? -0.033???
(0.010) (0.009) (0.011) (0.011)
Mean Dep. Var. 0.553 0.553 0.550 0.495
r2 0.059 0.093 0.120 0.074
Panel C: log Industrial Establishments, 1945
log Coffee trees -0.023??? -0.031???1920 -0.033??? -0.033???
(0.006) (0.006) (0.008) (0.008)
r2 0.378 0.443 0.489 0.359
Panel D: Industrial Establishments per 1,000 inhab., 1945
log Coffee trees ??? ??? ??? ???1920 -0.019 -0.023 -0.022 -0.022
(0.005) (0.006) (0.008) (0.008)
Mean Dep. Var. 0.431 0.431 0.429 0.397
r2 0.050 0.090 0.154 0.096
Counties 734 734 730 707
Geo controls Y Y Y
Dept. FE Y Y
Dept. Capitals Y Y Y
Note: Each Panel presents the correlation between coffee trees in 1920 and some mea-
sure of industrial activity in 1945. All specifications control for population in 1938
(log). Geo controls include: distance to department?s capital, distance to second
largest market, and a dummy for capital cities (except for column (4) where capitals
are excluded). Conley (1999) standard errors as described in section 2.3 in parenthe-
sis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
105
106
Table 2.3: Test for Other Discontinuities
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
From 1912 census: Geographic controls
Dep. Variable: Literacy Employment in Population Distance to... Terrain Soil Native Pop. Land
Rate Manufact. Agricult. Density Bogota Dept. Cap. Ruggedness Quality in 1560=1 Disputes
Altitude> 2, 400m -0.034 -0.016? -0.031 -15.047 -92.356? 24.137 -0.263 0.447 0.314 -0.086
(0.045) (0.009) (0.025) (10.969) (51.295) (15.493) (0.385) (0.382) (0.190) (0.065)
F-stat 2.508 2.809 1.815 0.818 2.293 1.421 6.537 0.700 1.905 1.285
r2 0.055 0.061 0.040 0.019 0.050 0.032 0.131 0.016 0.042 0.029
Counties 134 134 134 134 134 134 134 131 134 134
Note: The table tests for discontinuous jumps in counties? characteristics at 2,400 meters. It shows results from regressing dependent variables on a dummy equal to one for
counties above 2,400meters over the sea level, altitude, and altitude? dummy for altitude> 2, 400m. Results restrict the sample to counties above 1,800meters. See appendix B.1
for variables definitions. Standard errors clustered at 60-by-60 miles grid squares in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01.
Table 2.4: Effect of Coffee Cultivation on Structural Transformation, 1973
(1) (2) (3) (4)
Estimator: OLS 2SLS
Panel A, Dep. var.: Share of Labor Force in Manufacturing, 1973
log Coffee trees ???1920 -0.004 -0.005??? -0.006??? -0.005?
(0.001) (0.002) (0.001) (0.003)
Mean Dep. Var. 0.198 0.198 0.196 0.205
r2 0.198 0.103 0.087 0.159
Panel B, Dep. var.: Share of Labor Force in Agriculture, 1973
log Coffee trees ??? ??? ??? ???1920 0.006 0.008 0.009 0.010
(0.001) (0.003) (0.002) (0.003)
Mean Dep. Var. 0.370 0.371 0.377 0.397
r2 0.804 0.795 0.804 0.833
Panel C, Dep. var.: Share of Population in Labor Force, 1973
log Coffee trees -0.000 0.004???1920 -0.001 0.000
(0.0003) (0.0011) (0.0008) (0.0013)
Mean Dep. Var. 0.530 0.530 0.528 0.521
r2 0.974 0.970 0.978 0.983
Panel D, First Stage Dep. var.: log Coffee trees1920
Coffee attainable yield (FAO) 0.062???
(0.009)
Rainfall 0.015???
(0.002)
Temperature 1.208???
(0.139)
Rainfall ? Temperature -0.001???
(0.000)
Altitude> 2, 400m -2.150?
(1.172)
F-stat Excluded Inst. 51.761 30.675 43.183
r2 0.508 0.546 0.617
Observations 1,100 1,096 1,056 270
Counties 550 548 528 135
Note: This table shows the effect of coffee cultivation on structural transformation using data at the IPUMS-
county level. Column (1) shows OLS results (equivalent to Table 2.1, Column 4). Columns (2) to (4) of
Panels A, B, and C show results from 2SLS using instruments detailed in Panel D. Column (4) instruments
coffee cultivation using altitude, a dummy equal to one for counties above 2,400 meters of altitude (Altitude>
2, 400m), and an interaction between both. All specifications control for population in 1973 (log), gender
fixed effects, distance to Department?s capital, distance to second largest market, and Department fixed effects.
Conley (1999) standard errors as described in section 2.3 in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
107
Table 2.5: 1945 IV
(1) (2) (3) (4)
Estimator: OLS 2SLS
Panel A: Industrial workers per 100 inhab., 1945
log Coffee trees -0.033??? -0.080?? -0.050? -0.088???1920
(0.011) (0.035) (0.027) (0.029)
Mean Dep. Var. 0.495 0.495 0.467 0.399
r2 0.074 0.022 0.029 0.038
Panel B: Industrial Establishments per 1,000 inhab., 1945
log Coffee trees -0.022??? -0.071???1920 -0.061??? -0.084???
(0.008) (0.020) (0.020) (0.031)
Mean Dep. Var. 0.397 0.397 0.400 0.505
r2 0.096 -0.034 -0.006 0.003
Panel C: First Stage Dep. var.: log Coffee trees1920
Coffee attainable yield (FAO) 0.065???
(0.007)
Rainfall 0.010???
(0.001)
Temperature 0.970???
(0.101)
Rainfall ? Temperature -0.000???
(0.000)
Altitude> 2, 400m -2.335??
(1.098)
F-stat excluded inst. 97.464 35.960 31.581
r2 0.570 0.585 0.426
Counties 707 706 689 250
Note: This table shows the effect of coffee cultivation on on some measure of industrial activity
in 1945 using data at the county level. Column (1) shows OLS results (equivalent to Table 2.2,
Column 4). Columns (2) to (4) of Panels A and B show results from 2SLS using instruments
detailed in Panel C. Column (4) instruments coffee cultivation using altitude, a dummy equal to
one for counties above 2,400 meters of altitude (Altitude> 2, 400m), and an interaction between
both. All specifications control for population in 1938 (log), distance to department?s capital,
distance to second largest market, and Department fixed effects. All specifications exclude capital
cities. Conley (1999) standard errors as described in section 2.3 in parenthesis. ? p < 0.1, ??
p < 0.05, ??? p < 0.01
108
Table 2.6: Effect of Coffee Price Shocks on Schooling by Cohort, 1973
(1) (2) (3) (4)
Estimator: OLS 2SLS
Dep. Variable Average Years of Education, 1973
Panel A: Continuous measure Coffee, 1920
log Coffee trees ? Price5,16 -0.007??? -0.017??? -0.022???1920 c -0.003
(0.002) (0.006) (0.005) (0.009)
F-stat Excluded Inst. 131.886 60.195 17.244
A-R p-value 0.002 0.000 0.233
Panel B: Discrete measure Coffee, 1920
(Coffee trees1920 > 0)? Price5,16c -0.099??? -0.238??? -0.293??? -0.032
(0.032) (0.078) (0.065) (0.114)
F-stat Excluded Inst. 97.286 54.511 14.941
A-R p-value 0.002 0.000 0.233
Panel C: Reduced Form
Coffee attainable yield ? Price5,16c -0.0018???
(0.0006)
Rainfall ? Price5,16c -0.4147???
(0.0879)
Temperature ? Price5,16c -0.0270???
(0.0068)
Rain. ? Temp. ? Price5,16c 0.0156???
(0.0038)
Altitude> 2, 400m ? Price5,16c 0.2597?
(0.1436)
Altitude ? Price5,16c -0.0003
(0.0003)
Altitude> 2, 400m ? Altitude ? Price5,16c -0.0002
(0.0004)
N 44,826 44,826 43,072 10,300
Counties 431 431 414 98
Mean Dep. Variable 2.881 2.881 2.873 3.131
Note: This table estimates the effect of coffee price shocks on schooling using data at the gender x cohort x
county-or-birth level, for cohorts born between 1901 and 1951. The dependent variable for all specifications is
average years of education. Panel A shows results measures coffee in 1920 with a continuous variable. Panel B
measures coffee in 1920 with a dummy variable. Price5,16c is log average real coffee price for cohort c between
5 and 16 years old. Column (1) shows OLS results. Columns (2) to (4) of Panels A and B show results from
2SLS using instruments detailed in Panel C. Panel C shows reduced form estimates. All specifications control
for gender, cohort, and county-of-birth fixed effects. F statistic from Kleinberg and Paap tests and p-values
from Anderson and Rubin tests are presented to test for weak instruments. Cohorts born in capital cities are
excluded. Standard errors clustered at the county-of-birth level in parenthesis. ? p < 0.1, ?? p < 0.05, ???
p < 0.01
109
Table 2.7: Effect of Coffee Price Shocks on Economic Structure and Income, 1973
(1) (2) (3) (4)
Estimator: OLS 2SLS
Instrument Attn Yield Rain x Temp FRDD
Panel A: Dependent Variable: Share of Cohort Employed in Manufacturing, 1973
(Coffee trees1920 > 0)? Price5,16c -0.004 -0.027?? -0.009 -0.037??
(0.005) (0.012) (0.010) (0.016)
Mean Dep. Variable 0.259 0.258 0.258 0.268
F-stat Excluded Inst. 96.135 53.614 14.569
A-R p-value 0.018 0.001 0.019
Panel B: Dependent Variable: Share of Cohort Employed in Agriculture, 1973
(Coffee trees1920 > 0)? Price5,16c 0.003 0.012 0.015? 0.001
(0.004) (0.010) (0.009) (0.013)
Mean Dep. Variable 0.311 0.311 0.313 0.293
F-stat Excluded Inst. 96.135 53.614 14.569
A-R p-value 0.230 0.107 0.978
Panel C: Dependent Variable: Household Income, 1973
(Coffee trees1920 > 0)? Price5,16c -0.022 -0.117??? -0.100??? -0.036
(0.018) (0.045) (0.034) (0.062)
Mean Dep. Variable -0.192 -0.191 -0.193 0.035
F-stat Excluded Inst. 97.255 54.535 14.941
A-R p-value 0.009 0.000 0.790
N 37,416 37,325 35,804 8,743
Counties 359 358 344 84
Note: This table estimates the effect of coffee price shocks on economic structure and income in 1973
using data at the gender x cohort x county-or-birth level, for cohorts born between 1901 and 1951. The
dependent variable is given at the top of each panel. (Coffee trees1920 > 0) is a dummy equal to one
for counties with a positive number of coffee trees in 1920. Price5,16c is log average real coffee price
for cohort c between 5 and 16 years old. Column (1) shows OLS results. Columns (2) to (4) show
results from 2SLS using instruments detailed at the top of the table. Column (2) uses coffee attainable
yields from FAO. Column (3) uses a polynomial on rainfall and temperature. Column (4) uses a fuzzy
regression discontinuity design (FRDD) on altitude. All specifications control for gender, cohort, and
county-of-birth fixed effects. Cohorts born in capital cities are excluded. Standard errors clustered at the
county-of-birth level in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
110
Table 2.8: Coffee Shocks and Structural Transformation by Inequality
(1) (2) (3) (4)
Sample Counties: All Coffee trees> 0 Low Ineq. High Ineq.
Panel A, Dep. Var.: Average Years of Schooling, 1973
log Coffee trees ? Price5,16 -0.0073???1920 c -0.0005 -0.0159 0.0220
(0.0025) (0.0106) (0.0144) (0.0161)
Mean Dep. Var. 2.8814 2.9576 2.9311 3.0287
Panel B, Dep. Var.: Literacy Rate, 1973
log Coffee trees ? Price5,16 -0.2351??? -0.6238??? -1.0410???1920 c 0.1914
(0.0418) (0.1658) (0.2184) (0.2839)
Mean Dep. Var. 72.7947 75.7333 75.1425 77.3802
Panel C, Dep. Var.: Share of Cohort Employed in Manufacturing, 1973
log Coffee trees1920 ? Price5,16c -0.0003 0.0009 0.0016 0.0010
(0.0004) (0.0013) (0.0019) (0.0024)
Mean Dep. Var 0.2586 0.2600 0.2518 0.2661
Panel D, Dep. Var.: Share of Cohort Employed in Agriculture, 1973
log Coffee trees 5,16 ? ??1920 ? Pricec 0.0004 0.0021 0.0037 -0.0010
(0.0003) (0.0012) (0.0016) (0.0024)
Mean Dep. Var 0.3106 0.3135 0.3180 0.3062
Panel E, Dep. Var.: Average Household Income, 1973
log Coffee trees1920 ? Price5,16c -0.0023? -0.0100? -0.0173?? 0.0089
(0.0014) (0.0057) (0.0085) (0.0079)
Mean Dep. Var. -0.1918 0.0334 -0.0080 0.1374
N 37,416 22,973 10,963 10,495
Counties 359 220 112 108
Note: This table shows correlations between coffee price shocks and outcomes in 1973 using data at the gender
x cohort x county-or-birth level. It uses different county samples. Column (1) uses all counties. Column (2)
restricts to counties with some coffee trees in 1920. Columns (3) and (4) restrict the sample further to coffee
counties with lower and higher (respectively) land inequality than the median coffee county. All specifica-
tions control for gender, cohort, and county-of-birth fixed effects. Cohorts born in capital cities are excluded.
Standard errors clustered at the county-of-birth level in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
111
Table 2.9: Effect of Coffee Cultivation on Number of Schools, 1951
(1) (2) (3) (4) (5)
Dep. Variable: Schools per 10,000 inhabitants, 1951
Sample: All counties
Estimator: OLS 2SLS
Coffee trees > 0 -0.436?? -1.053??1920
(0.176) (0.415)
Counties 491 476
Mean Dep. Var. 0.678 0.643
F-stat Excluded Inst. 49.871
r2 0.154 -0.018
Sample: All counties Coffee trees> 0 Low Ineq. High Ineq.
Estimator: OLS 2SLS 2SLS 2SLS 2SLS
log Coffee trees ?? ??1920 -0.036 -0.073 -0.089 0.015 -0.341
(0.014) (0.029) (0.084) (0.079) (0.214)
Counties 491 476 316 147 167
Mean Dep. Var. 0.678 0.643 0.627 0.530 0.715
F-stat Excluded Inst. 72.457 37.934 35.803 8.747
r2 0.155 0.008 0.019 0.028 -0.105
Note: The tables shows the effect of coffee cultivation on the number of schools per 10,000 inhabitants
in 1951 using data at the IPUMS-county level. All specifications estimated using 2SLS instrument coffee
cultivation with Coffee Attainable Yields from FAO. Coffee trees1920 > 0 is a dummy equal to one for
counties with positive number of coffee trees in 1920. Column (3) restricts the sample to only coun-
ties with coffee cultivation in 1920. Columns (4) and(5) further divide coffee counties by level of land
inequality. All specifications control for population in 1951 (log), distance to Department?s capital, dis-
tance to second largestmarket, andDepartment fixed effects. Conley (1999) standard errors as described
in section 2.3 in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
112
Table 2.10: Linkages in Coffee Production and Structural Transformation, 1973
(1) (2) (3) (4)
Dep. Variable: Share of Labor Force in Manufacturing, 1973
Panel A: OLS and Second Stage
Estimator: OLS 2SLS OLS 2SLS
Coffee trees ??? ??1920 > 0 -0.049 -0.069
(0.009) (0.027)
Coffee trees1920 > 0, No Threshing -0.042??? -0.064?
(0.009) (0.034)
Coffee trees1920 > 0, Threshing -0.068??? -0.071???
(0.011) (0.023)
N 1,100 1,096 1,100 1,096
Counties 550 448 550 448
r2 0.192 0.092 0.197 0.096
F-stat Excluded Inst. 38.546 11.769
Threshing = No Threshing? (F-stat) 6.799 0.205
Panel B: Reduced Form
Coffee attainable yield (FAO) -0.0003?? -0.0002?
(0.0001) (0.0001)
Coffee attainable yield (FAO) ? Threshing -0.0004??
(0.0002)
N 1,096 1,096
Counties 448 448
r2 0.17183 0.17641
Note: This table shows the effect of coffee cultivation and coffee threshing on share of labor force employed
in manufacturing (1973) using data at the IPUMS-county level. Column (1) shows OLS results (equivalent
to Table 2.1, Column 4). ?Coffee trees1920 > 0, (No) Threshing? is a dummy equal to one for counties with
positive number of coffee trees in 1920 and (no) threshing machines in 1920. All specifications control for
population in 1973 (log), gender fixed effects, distance to Department?s capital, distance to second largest
market, and Department fixed effects. Conley (1999) standard errors as described in section 2.3 in parenthesis.
? p < 0.1, ?? p < 0.05, ??? p < 0.01
113
Table 2.11: Effect of Coffee Cultivation on Long Term Income, 2005
(1) (2) (3) (4)
Estimator OLS 2SLS
Instrument: Att. Yields Rain ? Temp. FRDD
Panel A: Poverty Rate, 2005
log Coffee trees 0.003??? 0.005?? 0.005???1920 0.003
(0.001) (0.002) (0.001) (0.003)
Mean Dep. Var. 0.520 0.519 0.519 0.473
r2 0.701 0.374 0.349 0.226
Panel B: Average Household Income, 2005
log Coffee trees1920 -0.031??? -0.002 -0.059??? -0.054??
(0.009) (0.035) (0.012) (0.021)
Mean Dep. Var. -0.367 -0.359 -0.369 -0.068
r2 0.670 0.410 0.411 0.492
N 472 464 449 91
F stat Excluded Inst. 12.117 20.931 12.479
Note: This table shows the effect of coffee cultivation on 2005 poverty and income using data
at the IPUMS-county level. Poverty rate comes fromDANE.Household income is calculated
from a principal components analysis using household characteristics and durable goods, as
described in 3.1 Column (1) shows OLS results. Columns (2) to (4) show results from 2SLS
using instruments detailed at the top of the table. All specifications control for population
(log), distance to Department?s capital, distance to second largest market, and Department
fixed effects. Conley (1999) standard errors as described in section 2.3 in parenthesis. ?
p < 0.1, ?? p < 0.05, ??? p < 0.01
114
Chapter 3: Commodities? Production Function and Human Cap-
ital Accumulation: Evidence from Early 20th Century
Puerto Rico
Countries rich on natural resources tend to be poor because short term
gains from exploiting those resources tend to crowd out productive investments
that are associated with long-term growth. That general statement has been stud-
ied at different levels, from Innis (1999) staple growth theory to Engerman and
Sokoloff (1997) ?endowments? hypothesis. The relationship between resource
richness and growth was exacerbated by international trade that pushed towards
specialization to exploit precisely those natural resources (Stokey, 1991; Galor,
2005). Among the potential explanations lies the role of human capital accumu-
lation. For rural households, the possibility to work at primary sector increases
the opportunity cost of investing in education, which is productive in the long run
(Porzio and Santangelo, 2019). That effect can be especially strong when interna-
tional demand for primary products generates increases in crops? prices. How-
ever, those price increases can also make budget constraints less binding for rural
households, allowing them to invest in the education of their children. When does
This chapter is a short version of a paper with the same title, joint with Matthew Curtis (UC
Davis).
115
the opportunity cost effect dominates the income effect is a relevant question for
both current social policy and our understanding of how endowments historically
affected the development process.
In this chapter, we show that the effect of changes in commodity prices on
the process of human capital accumulation depends on features of crops? produc-
tion function. We use data from Puerto Rico at the beginning of the 20th century,
which provides us with a compelling case study for three reasons. First, the is-
land?s geography is such that by 1900 different parts of the island specialized in
either sugar or coffee production (see Figure 3.2). The first one is characterized
by strong economies of scale which generated larger farms than the second one,
which is produced in small family farms (Bobonis and Morrow, 2014). Second,
farmers in the island are price takers in the coffee and sugar markets, which allow
us to use changes in the international world price of both commodities asmeasure
of changes in income. Finally, because the United States colonized Puerto Rico in
1899 the island counts with better historical information than other similar coun-
tries. For this chapter, we use decennial Census of population (1910, 1920, and
1930). Moreover, we digitize a series of detailed yearly reports from the island?s
appointed Governor to the President of the United States.
The Governor reports include detailed information on enrollment in pub-
lic schools and area cultivated by crop for the 76 municipios (counties). We use
them to build a panel database of counties between 1907 and 1920, which allow
us to compare how school enrollment responded to increases in crop?s prices in
a county that specializes in sugar production versus one that specializes in coffee
116
production and viceversa. Additionally, we use census data to provide an alter-
native measure of schooling. Importantly, since one of the main goals of the US
government in the island was to increase education (Bobonis and Toro, 2007),
census data does not rely on official reporting from bureaucrats in the island in
charge of the school system to the central government.
In sugar counties, increases in the price of sugar led to increases in enroll-
ment rates. However, increases in the price of coffee reduced school enrollment in
counties that specialize in the crop. Why did increases in agricultural income have
such different effects on investments in human capital? We propose it is due to
the differences between sugar and coffee production function that lead to income
effect of higher prices dominating in sugar areas, while the labor effect dominates
for coffee areas. Sugar farms tend to be large plantations with a lot of workers and
considerable economies of scale in land. When prices increase, workers involved
in sugar production have higher wages so they can afford to send their children
to school. While this effect also happens in coffee areas, coffee farms tend to be
smaller farms with no economies of scale. Moreover, coffee is more amenable
for child labor than sugar, since it requires to pick the cherries as opposed to cut
the sugar cane, an activity more suitable for adult males. Therefore, when coffee
prices increase, the opportunity cost for a family of not sending their children to
school also increases. This chapter points out that this effect dominates for coffee
areas, while the income effect dominates for sugar areas.
We provide more support for this hypothesis by showing that the effect is
concentrated in rural schools, as opposed to grade schools. One way to interpret
117
this result is that children going to rural schools are from families involved in
the primary sector and therefore they are the most affected by changes in prices.
Additionally, we use census data from 1910, 1920 and 1930 and show the main
results also hold when looking at data reported by households and not by the
government. Importantly, the effect of coffee prices on enrollment is stronger for
women than formenwhile the effect of sugar price shocks onwomen is essentially
zero. Since women participate directly in coffee collection and cultivation but less
so in the process of sugar production, this result provides support to the main
hypothesis.
Our results are consistent with Bobonis and Morrow (2014) evidence for
Puerto Rico in the mid-nineteenth century. Using a similar empirical strategy,
they find coffee price shocks have a negative effect on literacy rates when no coer-
cive institutions exist. In this chapter we observe schooling directly, as opposed to
literacy rates, andwe compare coffee areaswith areas specialized in sugar produc-
tion. The fact that coffee is a labor-intensive crop that generates negative effects
on schooling is not unique to Puerto Rico. It has been explored, for instance, for
Colombia by Carrillo (2019) and Uribe-Castro (2020). Moreover, Franck and Ga-
lor (2017); Atkin (2016) also propose that economic activities that increase the
opportunity cost of acquiring education can shift the population outside of hu-
man capital accumulation, leading to lower rates of industrialization. Finally, this
chapter is related to a literature on how products? production functions shape
economic performance over time through many channels other than education
(Droller and Fiszbein, 2019; Dell and Olken, 2020; Vollrath, 2011)
118
While the results presented in this chapter are suggestive, more work is re-
quired to rule out alternative hypothesis. Importantly, the Governor reports in-
clude information on the number of teachers and schools per county and year. We
plan to digitize these data to rule out that the differential effects of price shocks
on enrollment depend on someway of government intervention, for instance pro-
viding more schooling in sugar areas than in coffee ones.
3.1 Data and Empirical Strategy
This chapter uses information about Puerto Rican counties (municipios) dur-
ing the first three decades on the 20th century. It collects data from several sources,
most importantly from a series of reports to the President of the United States,
written by the Governor of Puerto Rico from 1901 until 1920. In the reports, the
Governor accounted for the fiscal situation on the island and documented a wide
set of issues at the county level. The reports are rich in terms of topics, ranging
from crime and social order to taxation. In this chapter, we focus on education
data, digitizing a yearly panel of enrollment totals for different types of schools.
We also use data from the reports on cultivated land by crops to measure local
economic activity. Finally, we use data on census of population samples in 1910,
1920, and 1930. This section describes the data collected and explains how it is
used to estimate the effect of commodity price shocks on human capital accumu-
lation.
119
Commodities and Endowments
From the reports, we extract information on total cultivated area allocated
to different crops every year. The main crops produced are coffee, sugar, tobacco,
cotton, and fruits (mostly oranges and pineapples). Though there is some year to
year variation, the rank of each county?smost important crop does not change over
the years. For themost important crops, coffee and sugar, Figure 3.6 compares the
share of cultivated land allocated to each crop in 1910 and 1920. Evidently, the
observations are close to the 45-degree line for both crops. Counties that by 1920
had a large share of their productive land used for coffee production, also had a
proportionally large share in 1910.
Figure 3.1 illustrates why there is such a persistence in the crop mixture.
Counties located in the central and western parts of the island have a mixture
of high elevation (Figure 3.1a) and high average rainfall (Figure 3.1b) that make
them suitable to grow coffee trees. The lowlands in the east and south-west parts
of the island have better suitability for planting sugar cane, as can be seen in Fig-
ure 3.1d.
However, even counties that are primarily producing sugar, also have some
coffee production. Figure 3.2 shows the mix between coffee and sugar cane plan-
tations for counties in 1920. The horizontal and vertical dashed lines show the
averages for each crop. That is, dots to the right of the vertical line have more
area cultivated in sugar cane than the average county in 1920. Dots above the
horizontal dashed line have more area cultivated in coffee trees than the average
120
county. Moreover, with circles we show counties that were above average in terms
of coffee cultivation at least twice in 1901, 1910 and 1920. Similarly, squares denote
counties that were above average in terms of sugar cane cultivation in at least two
years. We will refer to these two groups of counties as Coffee and Sugar counties,
respectively, as seen in the figure. Finally, there are some counties that either do
not have any of the crops or that are hard to classify using the quick rule of thumb
described above. We will refer to those counties as ?Ambiguous? and they are
represented by diamonds in the figure. Figure 3.3 shows the geographic location
of Coffee and Sugar counties.
Education data
We build a yearly panel at the county level with total enrollment in elemen-
tary and secondary schools. Starting on 1907, the reports provide information on
the number of students in every county that were enrolled in a public school by
March 1. During the first few years of our sample, the reports focus on ?Com-
mon? schools, divided in ?Graded? and ?Rural? schools. After 1912, the reports
also include information on enrollment in secondary schools. For some years, the
data is disaggregated at the race and gender level.
Census data
We use the census of population samples for 1910, 1920, and 1930 available
in IPUMS-International (Ruggles et al., 2003). Since the data is at the individual
121
level, we aggregate it at the county level to generate population counts for differ-
ent characteristics. We collect information on total population, urban shares, labor
force participation rate, literacy rate, school attendance, number of farm owners,
and occupations. Table 3.2 illustrates some characteristics of PuertoRican counties
taken from the 1910 Census of population. In terms of occupation, 1910 and 1920
census contain information on the share of working age population employed in
coffee and sugar production, as well as the number of coffee and sugar farm own-
ers. The table illustrates in which dimensions Coffee and Sugar counties differ.
Crucially, we calculate population of school age as the number of people
between 5 and 15 years old. We use these to calculate enrollment rates. We do so
by linearly interpolating between census to get the number of kids between 5 and
15 years old every year. For years before the 1910 Census we extrapolate using
the rate of growth of school age population between 1910 and 1920. We follow
the same interpolation process for total population and urban population.
Sincewe canmeasure school enrollment rates using the census in 1910, 1920,
and 1930, we compare them with the enrollment rates for elementary school col-
lected from reports. For both census there is a strong, positive correlation, though
enrollment rates from the Governor reports are generally lower because they only
focus on elementary school.
122
Other Geographic Characteristics
We compile geographic information at the county level from two different
sources. From the United States Geological Survey we collect information on av-
erage altitude (Figure 3.1a). We also use information on latitude and longitude
to calculate distance of each county to San Juan, the capital and largest city, lo-
cated in the northeast. and Mayaguez, the largest eastern city and a major port.
Finally, we use data from weather stations available from the National Oceanic
and Atmospheric Administration on average yearly precipitation between 1981
and 2010. We further averaged the measures from the stations contained on each
county. Since 34 counties do not have their own weather station, we averaged
measures from stations within 20 km radius, weighting then inversely according
to distance to the county?s centroid.
3.1.1 Empirical Strategy
To explore the effect of commodity prices, we estimate the following model
for two sets of measures of school enrollment take from two sources (Governor
reports and Census of Population):
y = ? + ? psct t s t ? sugarc + ?
c
cpt ? coffeec + ?Xct + ?Zc + ct (3.1)
Where yct is a measure of schooling for county c on year t. The coefficients
of interest are ?j . Those are coefficients on interactions between a dummy equal
123
to one for county specialized in crop j (sugar or coffee) and pjt , crop j?s inter-
national price. The specification includes year fixed effects (?t), controls for the
shares of cultivated land devoted to coffee and sugar in 1920 (Zc), and a series
of time varying controls (Xct) comprised of logs of total population, school age
population, and urban population. In order to use the full sample from the Gov-
ernor reports, we interpolate population figures linearly between census. If some
county has missing population number for one or more of the years for which we
have enrollment data, we assume population growth would have been equal to
the average population growth and impute those figures.
To capture schooling, we use two different measures: total enrollment in
logs and total enrollment divided by the number of people of school age. For
data from the Governor reports, we can also disaggregate total enrollment into
?Graded? and ?Rural? schools. The main sample consists of 76 counties between
1907 and 1920. Census data does not have detailed information on type of school
but it allow us to differentiate enrollment by gender. The census sample consists
on 76 counties and three years: 1910, 1920, 1930.
Finally, for all specifications we cluster standard errors at the county level to
account for serial auto-correlation within counties.
124
3.2 Results
Crops? Production Functions
The idea that coffee and sugar cane cultivation have very different produc-
tion functions is easily shown combining the data described above. Table 3.1
shows several characteristics of crops, averaging at the county level for all coun-
ties, counties that specialize in coffee, and counties that specialize in sugar (columns
1 to 3, respectively). It shows the number of acres dedicated to coffee and sugar
cane cultivation on average from the reports, the number of farm owners, and the
number of workers, both coming from the 1910 census.
Three statistics help drive home the point that sugar production has more
economies of scale and therefore is produced in large farms, with lower demand
for child labor. First, the number of workers per farm in sugar plantations is be-
tween 5 and 9 times larger, on average, than in coffee farms. Additionally, the size
of farms is considerably larger in sugar plantations than in coffee plantations.
Some caveats apply when interpreting these figures. First, if sugar workers
are exclusively working in sugar farms while coffee farmers are working in coffee
and other sectors, or if women work more informally, the numbers may be biased
downwards for coffee workers but not for sugar. Second, we assume the number
of farms is equal to the number of farm owners in the census. However, that
might not be true. One single farm owner can own several properties. If that
is more likely to happen in the sugar sector, where there are more economies of
125
scale, than in the coffee sector, that would bias the interpretation of these table
upwards. That is, we would be overestimating the number of farm owners.
3.2.1 Commodity Price Shocks and School Enrollment
The behavior of coffee and sugar prices shown in Figure 3.2, togetherwith 3.5
provides the main intuition for the preliminary results discussed on this chapter.
The latter figure shows the enrollment rate (removing within-year averages) for
sugar and coffee counties. The enrollment rate decreased for coffee counties right
around 1910, when the real coffee price was increasing, and it came back down at
the end of the period. The following analysis formalizes this insight and translates
it into regression form, depicted in Equation 1.
Tables 3.3 and 3.4 have the same basic layout. The dependent variable is
total enrollment (log), for columns (1) to (3), and total enrollment divided by the
number of people between 5 and 15 years old, for columns (4) to (6). Different
panels considers different dimensions of total school enrollment. In Table 3.3,
Panel A shows enrollment in all schools, while Panels B and C use enrollment in
only rural and urban schools, respectively.
Two main takeaways come out of Table 3.3: first, on average, increases in
sugar prices had a positive effect of enrollment in sugar counties, while the op-
posite seems to be true for coffee counties (negative or zero effect). Second, most
of the effect is driven by rural schools, which are arguably the most exposed to
changes in crop?s prices.
126
The first takeaway can be confirmed moving to Table 3.4. Changes in sugar
prices are positively associated with increases in enrollment in sugar counties. At
the same time, positive changes in the coffee price are associated with a reduction
in enrollment rates. However, given the reduction on sample size, none of the
effects are statistically significant.
3.3 Conclusions
This chapter shows preliminary evidence that the effect of changes in com-
modity prices on the process of human capital accumulation depends on specific
features of crops? production function. For crops with more economies of scale,
larger and fewer farms, increases in crop?s price can translate into more schooling
since rural household do not have a strong incentive on using family or child la-
bor. However, for labor intensive crops with limited economies of scale, increases
in price might increase the opportunity cost of education, leading to lower enroll-
ment rates.
Future work will collect more years of information from the Governor Re-
ports to fully exploit the increase in coffee prices during the 1920?s. Moreover, we
will collect information on the supply of education for each county (number of
school building and teachers). Having these data will allow us to support our
hypothesis that the difference between coffee and sugar counties? responses to in-
creases in prices is not driven by the response, for instance, by the government
funding less schools in some counties but not in others.
127
Figures
Figure 3.1: Weather, Geography, and Main Crops
(a) Altitude (feet)
(b) Average Precipitation Level (inches)
(c) Area Planted in Coffee (Share of Cultivated Land)
(d) Area Planted in Sugar (Share of Cultivated Land)
Note: Data comes from USGS, NOAA, 1920 Puerto Rico Governor report to the President of the US..
128
Figure 3.2: Sugar and Coffee Areas by County
Note: Data comes from 1920 Puerto Rico Governor report to the President of the US.
Figure 3.3: Location of Coffee and Sugar Counties
129
Figure 3.4: Real Coffee and Sugar Prices (1900=1)
Source: Jacks (2019).
Figure 3.5: Enrollment Rates in Public Schools by Type of County
Note: Average enrollment rate by type of county between 1907 and 1920.
Total enrollment comes from Governor Reports. School age population
comes from 1910 and 1920 Census of Population.
130
Figure 3.6: Land Allocated to Coffee and Sugar as Share of Cultivated Land,
1910 and 1920
(a) Coffee
(b) Sugar
Note: Share of cultivated land in coffee and sugar as a share of total cultivated land in 1920. Source: Governor
Reports, 1920.
131
Figure 3.7: Comparing Enrollment Data from Reports and Census
132
Tables
Table 3.1: Average Crops? Characteristics: Coffee and Sugar Production 1910.
Measure Crop (1) (2) (3)All Counties Coffee Counties Sugar Counties
Coffee 2,274.4 4,756.6 595.7
Land (Acres) (3,589.0) (4,610.2) (928.7)
Sugar 2,671.4 799.5 4012.6(2,972.8) (1,565.7) (2689.7)
Coffee 78.53 155.6 26.39
Farm Owners (135.4) (178.4) (52.87)
Sugar 72.94 26.73 112.1(90.39) (47.75) (101.7)
Coffee 479.9 970.6 124.6
Workers (887.4) (1156.4) (290.8)
Sugar 1,189.1 398.5 1734.7(1,199.7) (571.8) (1038.1)
Coffee 5.282 5.574 3.935
Workers per Farm (3.662) (2.780) (3.912)
Sugar 33.94 19.19 34.94(52.76) (21.01) (46.30)
Coffee 0.206 0.167 0.235
Workers per Acre (0.383) (0.114) (0.510)
Sugar 1.649 3.008 0.739(5.080) (7.836) (0.992)
Coffee 38.29 43.80 28.39
Acres per Farm (36.56) (35.50) (37.42)
Sugar 77.61 30.32 85.80(153.6) (28.07) (159.6)
Counties 68 26 36
Note: The table presents cross-county means and standard deviations (in brackets) for different
crops? characteristics in 1910. Land acres comes from Governor Reports. Number of Farm owners
and number of Workers comes from the census of population. Workers per Farm is equal to the
number of workers over the number of farm owners. More details in section 3.1.
133
Table 3.2: Descriptive Statistics: 1910 Census and Governor Reports
(1) (2) (3)
Full Sample Coffee Counties Sugar Counties Source
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
Enrollment rate, Census data 0.39 0.11 0.35 0.08 0.41 0.09 1910 Census
Enrollment rate, Governor reports 0.29 0.07 0.28 0.06 0.30 0.08 Governor Reports
Literacy rate 0.23 0.09 0.20 0.06 0.23 0.06
Population (thousands) 17.13 12.85 16.20 9.52 15.58 8.49
Urban share 0.21 0.17 0.14 0.10 0.23 0.14
Black share 0.36 0.15 0.27 0.13 0.42 0.14 1910 Census
Labor Force Participation 0.61 0.04 0.59 0.03 0.62 0.04
Women?s Labor Force Participation 0.21 0.07 0.19 0.06 0.22 0.06
Employment in Manufacturing 0.08 0.06 0.05 0.03 0.09 0.06
Livestock Heads (thousands) 3.23 2.08 2.09 1.39 4.11 2.12 Gov. Reports
Coffee Cultivation
Coffee (% of cultivated area) 22.22 23.37 45.02 21.03 7.47 8.37 Gov. Reports
Employment (share work age pop.) 0.05 0.10 0.11 0.13 0.01 0.02
# Coffee Farm Owners 78.53 135.40 155.58 178.42 26.39 52.87 1910 Census
Workers per farm owner 5.28 3.66 5.57 2.78 3.94 3.91
Sugar Production
Sugar Cane (% of cultivated area) 34.60 30.95 8.75 12.45 55.97 24.97 Gov. Reports
Employment (share work age pop.) 0.15 0.12 0.04 0.06 0.23 0.10
# Sugar Cane Farm Owners 72.94 90.39 26.73 47.75 112.08 101.71
Workers per farm owner 33.94 52.76 19.19 21.01 34.94 46.30 1910 Census
# Refinery Workers 8.09 13.77 3.85 9.09 11.11 16.52
Geographic Characteristics
Avg. Elevation (ft) 552.46 549.32 1091.68 570.44 239.23 133.77 USGS
Avg. Yearly Precipitation (inches) 67.39 15.40 68.74 12.87 68.32 16.83 NOAA
Distance to San Juan (km) 59.11 31.88 62.99 29.13 54.00 32.89 USGS
Distance to Mayaguez (km) 81.34 47.00 61.17 34.59 96.36 47.47 USGS
Counties 76 28 41
Note: The table shows descriptive statistics from 1910 census and other sources. It divides the
sample in three groups: Full Sample, only Coffee, and only Sugar counties according to details in
section 3.1.
134
Table 3.3: Effect of Coffee and Sugar Income on Total Enrollment (Governor
Reports)
(1) (2) (3) (4) (5) (6)
Dependent Variable: (log) Total Enrolled Enrollment rate
Panel A: All Common Schools
Sugar Price ? Sugar County 0.171?? 0.170??? 0.055?? 0.059???
(0.076) (0.061) (0.026) (0.022)
Coffee Price ? Coffee County -0.078 -0.001 -0.016 0.011
(0.087) (0.075) (0.028) (0.025)
Mean Dep. Var. 7.267 7.267 7.267 0.348 0.348 0.348
R2 0.919 0.918 0.919 0.721 0.717 0.722
Panel B: Only Rural Schools
Sugar Price ? Sugar County 0.385??? 0.488??? 0.067??? 0.075???
(0.134) (0.168) (0.025) (0.023)
Coffee Price ? Coffee County 0.035 0.257 -0.014 0.020
(0.147) (0.187) (0.026) (0.026)
Mean Dep. Var. 6.809 6.809 6.809 0.286 0.286 0.286
R2 0.713 0.705 0.723 0.629 0.619 0.629
Panel B: Only Graded Schools
Sugar Price ? Sugar County 0.007 -0.004 -0.013 -0.020
(0.094) (0.099) (0.014) (0.017)
Coffee Price ? Coffee County -0.026 -0.027 -0.009 -0.018
(0.105) (0.112) (0.014) (0.017)
Mean Dep. Var. 6.120 6.120 6.120 0.119 0.119 0.119
R2 0.895 0.895 0.895 0.733 0.734 0.738
Observations 983 983 983 983 983 983
Years 1907 to 1920
Counties 76
Note: The sample is an unbalanced yearly panel for 76 counties between 1907 and 1920. Dependent
variable for columns (1) to (3) is total enrollment per year (log) and for columns (4) to (6) is total
enrollment divided by school age population (between 5 and 15 years old). Coffee and Sugar
Prices are measured as log of real price. All specifications control for share of cultivated land
allocated to coffee and sugar. Population controls include: total population, urban population
and school age population (all in logs). Standard errors clustered at the county level in brackets.
? p < 0.1, ?? p < 0.05, ??? p < 0.01
135
Table 3.4: Effect of Coffee and Sugar Income on Total Enrollment (Census Data)
(1) (2) (3) (4) (5) (6)
Dependent Variable: (log) Enrollment Enrollment rate
Panel A: Total Enrollment
Sugar Price ? Sugar County 0.088 0.053 0.030 0.014
(0.060) (0.063) (0.026) (0.028)
Coffee Price ? Coffee County -0.235? -0.214? -0.104? -0.099?
(0.126) (0.128) (0.055) (0.056)
Mean Dep. Var. 7.611 7.611 7.611 0.443 0.443 0.443
R2 0.894 0.894 0.895 0.334 0.337 0.338
Panel B: Women?s Enrollment
Sugar Price ? Sugar County 0.098 0.051 0.017 -0.001
(0.081) (0.083) (0.031) (0.033)
Coffee Price ? Coffee County -0.310? -0.290? -0.112? -0.112?
(0.166) (0.168) (0.063) (0.065)
Mean Dep. Var. 6.803 6.803 6.803 0.410 0.410 0.410
R2 0.853 0.853 0.853 0.337 0.340 0.340
Observations 217 217 217 217 217 217
Years 1910, 1920, 1930
Counties 76
Note: The sample is an unbalanced panel for 76 counties in 1910, 1920, 1930 census of popula-
tion. Dependent variable for columns (1) to (3) is number of people attending school (log) and
for columns (4) to (6) is number of people attending school divided by school age population
(between 5 and 15 years old). Coffee and Sugar Prices are measured as log of real price. All speci-
fications control for year fixed effects, share of cultivated land allocated to coffee and sugar in 1920,
and a vector of population controls that includes: total population, urban population and school
age population (all in logs). Standard errors clustered at the county level in brackets. ? p < 0.1,
?? p < 0.05, ??? p < 0.01
136
Appendix A: Appendix to Chapter 1
Figure A.1: Map of the main sample
Note: the map shows the geographical distribution of the main sample using current
municipality boundaries. Lighter municipalities belong to the control group, darker,
to the treatment group, and blank municipalities are not on my sample, that comes
from Bushnell (1970) 1856 elections data.
137
Figure A.2: Parallel trends assumption test
Note: the figure is based on a regression of the number of battles per year on the
interaction between a dummy equal to 1 for the treatment group and dummies equal
to 1 for each year t from 1854 to 1885, as noted in equation 1.2. The figure shows the
interactions? coefficients and their standard error. Year 0 is 1863, where the period
after the reform started.
Table A.1: Standard errors: different clustering methods
(1) (2) (3) (4)
Dependent variable: # Battles per year (1854-1885)
Coefficient Standard errors clustered at
Municipality Year*Municipality Province
d1863 x Disentailment -0.021 (0.0104) (0.0141) [0.0103]
d1863 x Cons. vote share 0.017 (0.0083) (0.0113) [0.078]
Number of Clusters 597 32 x 597 22
d1863 x Control X
Municipality FE X
Year FE X
R2 0.139
N 19,104
Municipalities 597
Columns (2) and (3) report, in brackets, standard errors clustered at municipality level and two-way
year*municipality, respectively. Column (4) reports, in square brackets, the p-value corresponding to
the F-statistic estimated using wild bootstrap. The dependent variable is the number of battles per year,
built from Riascos Grueso (1950). Disentailment is a dummy variable equal to 1 for municipalities where
the Church?s real estate. dt is a dummy equal to 1 for years ? t. Conservative vote share (1856) is the
share of total votes won by Mariano Ospina, the Conservative party candidate in the 1856 presidential
election. Control variables are defined at the municipality level and include: foundation year, altitude,
total area of the municipality, distance to the State?s capital (log), distance to Bogota (log), distance to
the closest main market (log), a dummy indicating early indigenous settlements (by 1534), a dummy
indicating early Spanish settlements (by 1560), and soil quality index.
138
Appendix B: Appendix to Chapter 2
B.1 Data Appendix
Outcome Variables
Share of population in labor force: 1912 and 1938 census already included
these numbers for every county. For 1973 and 2005 census, I calculate the county-
level number of adults between 18 and 65 years old who answered affirmatively
to the question about labor force participation. I then divide by total population,
taken from each census.
Share of labor force employed in manufacturing and agriculture: for 1912
and 1938 census already included the number of people by occupation. For 1973
and 2005, I use the ISCO-68 3 code classification of occupation to identify worker?s
employment sector in three broad categories: Agriculture (Occupations in the
600 ISCO-68 code), Manufacturing (Occupations in the 700 ISCO-68 code except
for ?Miners and quarrymen? (711) and ?Mineral and stone treaters? (712). Oc-
cupations in the 800 code and from codes 901 to 989 (inclusive)), and Services
(Occupations in the 400 and 500 ISCO-68 code).
Number of manufacturing establishments per capita: Number of manu-
facturing establishments with more than 5 employees divided by population in
1945. From 1945 Industrial Census (Santos Cardenas, 2017). I interpolate county
level population using data from 1938 and 1951 census of population.
Share of population employed in industrial establishments: Number of
workers in manufacturing establishments with more than five employees divided
by population in 1945. From 1945 Industrial Census (Santos Cardenas, 2017).
I interpolate county level population using data from 1938 and 1951 census of
population.
Human capital: I measure education in 1973. I use two variables available
for individuals older than 5 years old. Literacy: a dummy equal to one if the
individual can read and write, zero otherwise. Years of schooling: Highest year
of education completed by the individual. Ranges from 0 to 18. I then aggregate
at the county-of-birth x cohort level using population weights.
Household income: I calculate a measure of income for households in 1973
139
and 2005. I extract the first vector of a principal components analysis on a series
of variables containing information on house quality. I use data on household
characteristics to build a measure of household wealth. They are:
Electricity: a dummy equal to one if the dwelling is connected to electricity.
Sewage: a dummy equal to one if the dwelling is connected to a drainage sewer
system. Water supply: a dummy equal to one if the dwelling is connected to piped
water supply. Toilet: a dummy equal to one if the dwelling has either flush toilet
or latrine. Floor material: a dummy equal to one if the dwelling?s floors are made
of cement, tile, brick, wood, or plastic. It equals zero if the dwelling has unfinished
or no floor. Roof material: a dummy equal to one if the dwelling?s roof is made of
reinforced concrete or clay tile. It equals zero if the roof ismade of zinc, tin, thatch,
or discarded material. Rooms per person: number of rooms the household uses
divided by the number of people in the household.
Poverty rate: share of households living below poverty line in 2005. From
Acevedo and Bornacelly (2014).
Coffee Cultivation
The census was part of a larger project, the book ?Colombia Cafetera? by
Dario Monsalve, commissioned by Colombia?s Department of Commerce to pro-
mote coffee exports abroad. The book includes general information about the
country, infographics, pictures, and a detailed account of coffee farms at the mu-
nicipality level. I digitized information on farms? names, owners when available,
and size, measured by the number of coffee trees used in production. For some
of the smaller plantations, the census does not include the owner?s name. For
some counties, there are only counts of small farms and their size. In some rare
cases, the census pools together an unknown number of plantations in a single
category (?Some? or ?Varias? in Spanish). The documentation on how the census
was collected is not very comprehensive and I cannot say with confidence why
some counties report their information pooling the smaller farms in this fashion.
Figure B.1 shows the census records for the municipality of Anz? in the de-
partment ofAntioquia, an instancewhere the threeways of reporting the informa-
tion appear. Overall, the census reports information for 37,689 farms, containing
242 million coffee trees.
I use the 1928 coffee census records to build four measures at the county
level. First, the total land used for coffee plantations, which is simply the sum of
the individual farms? sizes. Second, the total number of farms. I measure this
in two different ways to deal with the pooled category ?Some:? one, assuming
?Some? is equal to one farm (lower bound), and two: assuming the farms are
equal to the smallest plantation in the county for which I have information (up-
per bound). Using the example from figure B.1, the number of farms in ?Some?
will be, respectively, 1 and 10. Third, the number of farms allows me to measure
the average and the median farm in every county. Finally, I calculate the Gini
coefficient at the county level.
Ideally, I would like to use the information on the owners? name, because
140
there are instances where the same name appears as the owner of more than one
plantation. There are, however, two obstacles for doing so. One, naming conven-
tions in Colombia use two last names system. The first last name is the father?s
first last name, and the second last name is the mothers? first last name. How-
ever, the data only includes one last name. Since last names in Colombia are very
common, I might identify two different people with the same name as the same
individual. Two, as pointed out above, there are a significant number of farms
for which I do not know the owner?s name. For the time being, I will assume the
number of plantations is equal to the number of owners in order to calculate two
measures of Gini coefficient, one for every estimate for the number of owners.
Figure B.1: Example from 1928 Coffee Census (Municipality of Anza,
Department of Antioquia)
Note: Columns correspond to: Plantation name, owner?s name, number
of trees.
141
B.2 Supporting Results
Table B.1: 1945 IV by sector
(1) (2) (3) (4) (5) (6)
Dependent Variable: Industrial workers Industrial Establishments
per 100 inhab., 1945 per 1,000 inhab., 1945
Instrument for Coffee Yields R ? T. FRDD Yields R ? T FRDD
Panel A: High Human Capital Sectors
log Coffee trees1920 -0.084?? -0.051??? -0.061?? -0.120??? -0.115??? -0.044
(0.037) (0.011) (0.026) (0.024) (0.021) (0.032)
Mean Dep. Var. 0.057 0.057 0.073 0.066 0.067 0.104
r2 -0.032 0.016 -0.002 -0.140 -0.119 0.067
Panel B:Medium Human Capital Sectors
log Coffee trees1920 -0.016 -0.011 -0.056 -0.023 -0.009 -0.024
(0.019) (0.023) (0.036) (0.014) (0.012) (0.029)
Mean Dep. Var. 0.106 0.102 0.077 0.047 0.046 0.035
r2 0.024 0.021 0.009 0.030 0.030 0.036
Panel C: Low Human Capital Sectors
log Coffee trees ??1920 -0.034 -0.021 -0.066 -0.070??? -0.061?? -0.083???
(0.021) (0.014) (0.028) (0.024) (0.024) (0.031)
Mean Dep. Var. 0.332 0.307 0.249 0.284 0.287 0.366
r2 0.013 0.016 0.032 -0.023 -0.004 -0.000
Counties 706 689 250 706 689 250
Note: Table shows effects of coffee cultivation on industrial employment and number of establishments in
1945 by sector, according to Human Capital requirement. Columns (1) and (4) use Coffee attainable yields
from FAO as instrument for coffee cultivation in 1920. Columns (2) and (5) use a polynomial on rainfall
and temperature. Columns (3) and (6) use a fuzzy regression discontinuity design (FRDD) on altitude. All
specifications control for population in 1938 (log), distance to department?s capital, distance to second largest
market, and Department fixed effects. All specifications exclude capital cities. Conley (1999) standard errors
as described in section 2.3 in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
142
143
Table B.2: Effect of Coffee Price Shocks on Economic Structure, 1912, 1938, and 2005
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Year: 1912 1938 2005
Estimator: OLS 2SLS 2SLS OLS 2SLS 2SLS OLS 2SLS 2SLS
Instrument Attn. Yield FRDD Attn. Yield FRDD Attn. Yield FRDD
Panel A, Dep. Var.: Share of Labor Force Employed in Manufacturing
log Coffee trees -0.002??? -0.002??? -0.001 -0.003??? -0.006??1920 -0.005?? -0.001??? -0.001 -0.003??
(0.001) (0.001) (0.001) (0.001) (0.003) (0.002) (0.000) (0.001) (0.002)
Mean Dep. Var. 0.035 0.034 0.033 0.097 0.096 0.137 0.024 0.024 0.034
r2 0.204 0.029 0.001 0.450 0.025 0.013 0.128 0.046 0.087
Panel B, Dep. Var.: Share of Labor Force Employed in Agriculture
log Coffee trees ??1920 0.002 0.002 0.003?? 0.006??? 0.008??? 0.008?? 0.004??? -0.011 0.009
(0.001) (0.002) (0.002) (0.002) (0.003) (0.003) (0.001) (0.007) (0.006)
Mean Dep. Var. 0.172 0.172 0.167 0.825 0.825 0.798 0.205 0.205 0.257
r2 0.218 0.033 0.071 0.267 0.073 0.039 0.476 0.315 0.408
Observations 719 698 245 713 695 250 934 898 182
F stat Excluded Inst. 28.638 11.645 35.752 12.331 12.706 39.242
Note: This table presents supporting results for figure 2.9. It shows the effect of coffee cultivation on structural transformation
using data at the IPUMS-county level. Columns (1) (4) and (7) show OLS results (equivalent to Table 2.1, Column 4).
Columns (3) (6) and (9) instrument coffee cultivation using altitude, a dummy equal to one for counties above 2,400 meters
of altitude (Altitude> 2, 400m), and an interaction between both. All specifications control for population (log), gender
fixed effects, distance toDepartment?s capital, distance to second largestmarket, andDepartment fixed effects. Conley (1999)
standard errors as described in section 2.3 in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
144
Table B.3: Balance Tests
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
From 1912 census: Geographic controls
Dep. Variable: Literacy Employment in Population Altitude Distance to... Terrain Soil Native Pop. Land
Rate Manufact. Agricult. Density (km) Bogota Dept. Cap. Ruggedness Quality in 1560=1 Disputes
Panel A: Difference in Means test Coffee vs. Non-coffee, 1925
Coffee trees1920 > 0 0.044??? -0.005 0.036??? 1.308 0.352??? -123.438??? -7.793? 1.284??? -0.392??? -0.043 -0.060?
(0.009) (0.005) (0.009) (2.999) (0.078) (16.602) (4.543) (0.107) (0.106) (0.047) (0.036)
Mean Coffee trees1920 = 0 0.108 0.035 0.120 29.332 1.001 375.607 72.661 1.360 3.021 0.562 0.225
r2 0.043 0.002 0.034 0.000 0.040 0.101 0.006 0.229 0.027 0.002 0.006
Counties 494 494 494 494 494 494 494 492 489 494 494
Panel B: Difference in Means test Above vs. Below 2,400mts
Altitude> 2, 400m -0.054?? 0.004 -0.022? -1.561 0.593??? 1.624 1.169 -0.691??? -0.026 0.089 0.014
(0.021) (0.004) (0.012) (5.089) (0.027) (24.200) (7.239) (0.180) (0.177) (0.089) (0.030)
Mean Counties Below 0.181 0.030 0.162 40.287 2.040 216.503 65.683 3.072 2.298 0.488 0.024
r2 0.048 0.005 0.025 0.001 0.785 0.000 0.000 0.101 0.000 0.008 0.002
Counties 134 134 134 134 134 134 134 134 131 134 134
Note: Panel A shows results from regressing dependent variables on a dummy equal to one for counties with positive Coffee Trees in 1925. Panel B
shows results from regressing dependent variables on a dummy equal to one for counties above 2,400 meters over the sea level, restricting the sample
to counties above 1,800 meters. Panel B restrict the sample to counties above 1,800 meters. See appendix B.1 for variables definitions. Conley (1999)
standard errors as described in section 2.3 in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01.
Table B.4: Coffee Shocks and Structural Transformation by Inequality
(1) (2) (3) (4) (5)
OLS 2nd. Stage OLS 2nd. Stage Red. Form
Panel A, Dep. Var: Sh. Cohort Employed in Manufacturing, 1973
Avg. Years of Schooling 0.0032?? 0.1244?
(0.0015) (0.0636)
Literacy Rate 0.0006??? 0.0065??
(0.0001) (0.0027)
Coffee yield ? Price5,16c -0.0002??
(0.0001)
Mean Dep. Var. 0.2587 0.2586 0.2585 0.2585 0.2585
F-stat Excluded Inst. 8.2063 10.6217
A-R test p-value 0.0179 0.0177
Panel B, Dep. Var: Sh. Cohort Employed in Agriculture, 1973
Avg. Years of Schooling -0.0478??? -0.0558
(0.0013) (0.0440)
Literacy Rate -0.0018??? -0.0029
(0.0001) (0.0021)
Coffee yield ? Price5,16c 0.0001
(0.0001)
Mean Dep. Var. 0.3092 0.3105 0.3096 0.3110 0.3110
F-stat Excluded Inst. 8.2063 10.6217
A-R test p-value 0.2308 0.2305
N 37,558 37,269 37,558 37,269 37,269
Counties 361 358 361 358 358
Note: This table shows the effect of education on occupation in 1973, instrumenting education using
coffee prices during school age for cohorts born between 1901 and 1951. It uses data at the gender x
cohort x county-or-birth level. All specifications control for gender, cohort, and county-of-birth fixed
effects. Cohorts born in capital cities are excluded. Standard errors clustered at the county-of-birth level
in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
145
B.3 Robustness Checks
Table B.5: Measuring 1920 Coffee Cultivation with Extensive Margin
(1) (2) (3) (4) (5) (6) (7) (8)
Year: 1912 1938 1973 2005
Instrument Attn. Yield FRDD Attn. Yield FRDD Attn. Yield FRDD
Panel A, Dep. Var.: Share of Labor Force Employed in Manufacturing
Coffee trees ?? ?? ? ? ??1920 > 0 -0.010 -0.018 -0.076 -0.059 -0.069 -0.063 -0.019 -0.040
(0.011) (0.012) (0.033) (0.026) (0.038) (0.036) (0.013) (0.019)
Mean Dep. Var. 0.035 0.033 0.096 0.137 0.198 0.205 0.025 0.034
Panel B, Dep. Var.: Share of Labor Force Employed in Agriculture
Coffee trees1920 > 0 0.050??? 0.045?? 0.108??? 0.092?? 0.111?? 0.133??? -0.157 0.151?
(0.015) (0.018) (0.039) (0.038) (0.044) (0.032) (0.105) (0.075)
Mean Dep. Var. 0.239 0.218 0.825 0.798 0.377 0.397 0.210 0.257
Panel C, Dep. Var.: Labor force participation
Coffee trees1920 > 0 -0.0020 -0.0064 0.0405 0.0219 0.063??? 0.002 -0.050 -0.041
(0.0250) (0.0178) (0.0311) (0.0236) (0.023) (0.013) (0.053) (0.035)
Mean Dep. Var. 0.411 0.382 0.551 0.587 0.530 0.521 0.571 0.596
Counties 719 245 713 250 547 135 467 141
F stat Excluded Inst. 26.182 11.019 31.436 11.523 14.542 12.244 11.014 9.203
Note: This table presents the effect of coffee cultivation on structural transformation over the 20th century
using a discrete value equal to 1 for counties with positive coffee production in 1920, 0 otherwise. Columns
(1) (3) (5) and (7) instrument coffee cultivation using attainable coffee yields. Columns (2) (4) (6) and (8)
instrument coffee cultivation using altitude, a dummy equal to one for counties above 2,400 meters of altitude
(Altitude> 2, 400m), and an interaction between both. All specifications control for population (log), gender
fixed effects, distance to Department?s capital, distance to second largest market, and Department fixed effects.
Conley (1999) standard errors as described in section 2.3 in parenthesis. ? p < 0.1, ?? p < 0.05, ??? p < 0.01
B.3.1 Sensitivity to Excluding Younger Cohorts
In order to test the sensibility of results to specific cohorts, I estimate results
presented in Section 2.5.1 removing different cohorts a at a time. The following
graphs show estimates of ? from equation 2.5 including cohorts up to years de-
picted in the x-axis. The preferred specification includes cohorts between 1902
and 1052.
146
Figure B.2: Sensitivity of results to different cohorts in sample
(a) Years of Schooling
-.2
-.3
-.4
-.5
-.6
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62
Cohorts born between 1900 and 19...
(b) Agriculture 1973
.08
.06
.04
.02
0
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62
Cohorts born between 1900 and 19...
(c) Manufacturing 1973
0
-.01
-.02
-.03
-.04
-.05
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62
Cohorts born between 1900 and 19...
(d) Labor force participation
.01
0
-.01
-.02
-.03
40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62
Cohorts born between 1900 and 19...
Note: the scaling of the y-axis differs for all the figures.
147
Coefficient: Coffee Suit x Price (10-16) Coefficient: Coffee Suit x Price (10-16) Coefficient: Coffee Suit x Price (10-16) Coefficient: Coffee Suit x Price (10-16)
Bibliography
Acemoglu, D. and Dell, M. (2010). Productivity Differences between and within
Countries. American Economic Journal: Macroeconomics, 2(1):169?188.
Acemoglu, D. and Guerrieri, V. (2008). Capital Deepening andNonbalanced Eco-
nomic Growth. Journal of Political Economy, 116(3):32.
Acemoglu, D., Johnson, S., and Robinson, J. A. (2002). Reversal of fortune: Geog-
raphy and institutions in the making of the modern world income distribution.
The Quarterly journal of economics, 117(4):1231?1294.
Acemoglu, D., Johnson, S., and Robinson, J. A. (2005). Institutions as a fundamen-
tal cause of long-run growth. Handbook of economic growth, 1:385?472. Publisher:
Elsevier.
Acemoglu, D. and Robinson, J. A. (2012). Why nations fail: the origins of power,
prosperity and poverty. Crown, New York.
Acevedo, K.M. and Bornacelly, I. D. (2014). PanelMunicipal del CEDE. Technical
Report 012223, Universidad de los Andes - CEDE.
Angrist, J. D. and Pischke, J.-S. (2008a). Mostly harmless econometrics: An empiri-
cist?s companion. Princeton University Press.
148
Angrist, J. D. and Pischke, J.-S. (2008b). Mostly harmless econometrics: An empiri-
cist?s companion. Princeton University Press.
Arango, M. (1981). Caf? e industria, 1850-1930.
Atkin, D. (2016). Endogenous Skill Acquisition and Export Manufacturing in
Mexico. American Economic Review, 106(8):2046?2085.
Autor, D. H. (2003). Outsourcing at will: The contribution of unjust dismissal
doctrine to the growth of employment outsourcing. Journal of labor economics,
21(1):1?42.
Autor, D.H., Dorn, D., andHanson, G.H. (2016). TheChina shock: Learning from
labor-market adjustment to large changes in trade. Annual Review of Economics,
8:205?240.
Bazant, J. (2008). Alienation of Church Wealth in Mexico: Social and Economic Aspects
of the Liberal Revolution 1856-1875, volume 11. Cambridge University Press.
Bazzi, S., Fiszbein, M., and Gebresilasse, M. (2017). Frontier Culture: The Roots
and Persistence of "Rugged Individualism" in the United States. Technical Re-
port w23997, National Bureau of Economic Research, Cambridge, MA.
Bejarano, J. A. (1980). Los estudios sobre la historia del caf? en Colombia. Cuader-
nos de econom?a, 1(2):115?140.
Bertrand, M., Duflo, E., and Mullainathan, S. (2004). How much should we
149
trust differences-in-differences estimates? The Quarterly journal of economics,
119(1):249?275.
Bester, C. A., Conley, T. G., and Hansen, C. B. (2011). Inference with dependent
data using cluster covariance estimators. Journal of Econometrics, 165(2):137?
151.
Blattman, C. and Miguel, E. (2010). Civil war. Journal of Economic literature,
48(1):3?57.
Bobonis, G. J. and Morrow, P. M. (2014). Labor coercion and the accumulation of
human capital. Journal of Development Economics, 108:32?53.
Bobonis, G. J. and Toro, H. J. (2007). Modern colonization and its consequences:
the effects of US educational policy on Puerto Rico?s educational stratification,
1899-1910. Caribbean Studies, 35(2):31?76. Publisher: Instituto de Estudios del
Caribe.
Brew, R. (1973). The economic development of Antioquia from 1850-1920.
Bulmer-Thomas, V. (2003). The economic history of Latin America since independence.
Cambridge University Press.
Bushnell, D. (1970). Elecciones presidenciales colombianas, 1825-1856. Compendio
de estad?sticas hist?ricas en Colombia. Bogot?: Universidad Nacional de Colombia.
Bushnell, D. (1993). The making of modern Colombia: A nation in spite of itself. Univ
of California Press.
150
Bustos, P., Caprettini, B., and Ponticelli, J. (2016). Agricultural Productivity and
Structural Transformation: Evidence from Brazil. American Economic Review,
106(6):1320?1365.
Cantoni, D., Dittmar, J., and Yuchtman, N. (2017). Religious competition and real-
location: The political economy of secularization in the protestant reformation.
Technical report, National Bureau of Economic Research.
Cardoso, F. H. and Faletto, E. (1979). Dependency and development in Latin America
(Dependencia y desarrollo en Am?rica Latina, engl.). Univ of California Press.
Carrillo, B. (2019). The Value of Time and Skill Acquisition in the Long Run:
Evidence from Coffee Booms and Busts. Journal of Labor Economics. In Press.
Caselli, F. and Coleman II, W. J. (2001). The U.S. Structural Transformation
and Regional Convergence: A Reinterpretation. Journal of Political Economy,
109(3):584?616.
Centeno, M. A. (1997). Blood and debt: War and taxation in nineteenth-century
latin america. American Journal of sociology, 102(6):1565?1605.
Ciccone, A. and Papaioannou, E. (2009). Human Capital, the Structure of Pro-
duction, and Growth. The Review of Economics and Statistics, 91(1):66?82.
Coatsworth, J. H. (1988). Patterns of rural rebellion in latin america: Mexico in
comparative perspective. Riot, Rebellion, and Revolution: Rural Social Conflict in
Mexico, pages 21?62.
151
Coatsworth, J. H. (2006). Political economy and economic organization. The Cam-
bridge Economic History of Latin America, 1:237?73.
Coatsworth, J. H. (2008). Inequality, institutions and economic growth in latin
america. Journal of Latin American Studies, 40(03):545?569.
Colella, F., Lalive, R., Sakalli, S. O., and Thoenig, M. (2019). Inference with Ar-
bitrary Clustering. SSRN Scholarly Paper ID 3449578, Social Science Research
Network, Rochester, NY.
Conley, T. G. (1999). GMM estimation with cross sectional dependence. Journal
of Econometrics, 92(1):1?45.
de la Croix, D., Doepke,M., andMokyr, J. (2018). Clans, Guilds, andMarkets: Ap-
prenticeship Institutions and Growth in the Preindustrial Economy. The Quar-
terly Journal of Economics, 133(1):1?70.
Deas, M. (1996). The role of the church, the army and the police in colombian
elections, c. 1850?1930. In Elections Before Democracy: The History of Elections in
Europe and Latin America, pages 163?180. Springer.
Dell, M. andOlken, B.A. (2020). The development effects of the extractive colonial
economy: The Dutch cultivation system in Java. Review of Economic Studies,
page 51.
D?az, F. D. (1977). La desamortizaci?n de bienes eclesi?sticos en Boyac?, volume 12.
Universidad Pedag?gica y Tecnol?gica de Colombia, Secretar?a de Investiga-
ciones y Extensi?n Universitaria.
152
Dippel, C., Ferrara, A., and Heblich, S. (2019a). ivmediate: Causal mediation
analysis in instrumental variables regressions.
Dippel, C., Gold, R., Heblich, S., and Pinto, R. (2019b). Mediation Analysis in IV
Settings With a Single Instrument.
Domenech, J. and Herreros, F. (2017). Land reform and peasant revolution. evi-
dence from 1930s spain. Explorations in Economic History, 64:82 ? 103.
Droller, F. and Fiszbein, M. (2019). Staple Products, Linkages, and Development:
Evidence from Argentina. Technical Report 0898-2937, National Bureau of Eco-
nomic Research.
Dube, O. and Vargas, J. F. (2013). Commodity Price Shocks and Civil Conflict:
Evidence from Colombia. The Review of Economic Studies, 80(4):1384?1421.
Duran, X., Musacchio, A., and Paolera, G. d. (2017). Industrial Growth in South
America: Argentina, Brazil, Chile, and Colombia, 1890?2010. Oxford University
Press.
Edwards, R. B. (2019). Export agriculture and rural poverty: evidence from In-
donesian palm oil.
Engerman, S. L. and Sokoloff, K. L. (1997). Factor endowments, institutions, and
differential paths of growth among new world economies. In Haber, S., edi-
tor, How Latin America Fell Behind: Essays on the Economic Histories of Brazil and
Mexico, 1800-1914. Stanford University Press.
153
Fazio, A. and S?nchez, F. (2010). Educational effects of 19th century disentailment
of catholic church land. Journal of Iberian and Latin American Economic History,
pages 1?27.
Fergusson, L. and Vargas, J. F. (2013). Don?t make war, make elections-franchise
extension and violence in XIXth-century Colombia. Documento CEDE.
Filmer, D. and Pritchett, L. H. (2001). Estimating Wealth Effects without Expen-
diture Data-or Tears: An Application to Educational Enrollments in States of
India. Demography, 38(1):115?132.
Finley, T., Franck, R., and Johnson, N. (2017). The effects of land redistribution:
evidence from the French Revolution. Working paper.
Fiszbein, M. (2017). Agricultural Diversity, Structural Change and Long-run De-
velopment: Evidence from the U.S. Working Paper 23183, National Bureau of
Economic Research.
Foster, A. D. and Rosenzweig, M. R. (2004). Agricultural Productivity Growth,
Rural Economic Diversity, and Economic Reforms: India, 1970?2000. Economic
Development and Cultural Change, 52(3):509?542.
Franck, R. and Galor, O. (2017). Technology-Skill Complementarity in Early
Phases of Industrialization. Technical Report w23197, National Bureau of Eco-
nomic Research, Cambridge, MA.
Galiani, S., Heymann, D., Dab?s, C., and Tohm?, F. (2008). On the emergence
154
of public education in land-rich economies. Journal of Development Economics,
86(2):434?446.
Galor, O. (2005). The Demographic Transition and the Emergence of Sustained
Economic Growth. Journal of the European Economic Association, 3(2-3):494?504.
Publisher: Oxford Academic.
Galor, O. and Moav, O. (2004). From Physical to Human Capital Accumulation:
Inequality and the Process of Development. The Review of Economic Studies,
71(4):1001?1026.
Galor, O., Moav, O., and Vollrath, D. (2009). Inequality in Landownership, the
Emergence of Human-Capital Promoting Institutions, and the Great Diver-
gence. Review of Economic Studies, page 37.
Gennaioli, N., La Porta, R., Lopez-de Silanes, F., and Shleifer, A. (2013). Hu-
man Capital and Regional Development. The Quarterly Journal of Economics,
128(1):105?164.
Goldberg, P. K. and Pavcnik, N. (2007). Distributional Effects of Globalization in
Developing Countries. Journal of Economic Literature, 45(1):39?82.
Gollin, D., Lagakos, D., and Waugh, M. E. (2014). The Agricultural Productivity
Gap. The Quarterly Journal of Economics, 129(2):939?993.
GRECO (2002). El crecimiento econ?mico colombiano en el siglo XX. Banco de la
Rep?blica.
155
Grossman, H. I. (1994). Production, appropriation, and land reform. TheAmerican
Economic Review, 84(3):705?712.
Hartlyn, J. (1988). The politics of coalition rule in Colombia. Cambridge University
Press.
Heldring, L., Robinson, J. A., and Vollmer, S. (2015). Monks, gents and industri-
alists: The long-run impact of the dissolution of the English monasteries. Tech-
nical report, National Bureau of Economic Research.
Herrendorf, B. and Schoellman, T. (2018). Wages, Human Capital, and Barriers to
Structural Transformation. American Economic Journal: Macroeconomics, 10(2):1?
23.
Hirschman, A. O. (1958). The strategy of economic development. Yale University
Press.
Hornbeck, R. and Keskin, P. (2015). Does Agriculture Generate Local Economic
Spillovers? Short-Run and Long-Run Evidence from the Ogallala Aquifer.
American Economic Journal: Economic Policy, 7(2):192?213.
Innis, H. A. (1999). The fur trade in Canada: An introduction to Canadian economic
history. University of Toronto Press.
Jacks, D. S. (2019). From boom to bust: a typology of real commodity prices in
the long run. Cliometrica, 13(2):201?220.
Jaramillo, R. L. andMeisel, A. (2009). M?s all? de la ret?rica de la reacci?n. an?lisis
156
econ?mico de la desamortizaci?n en colombia, 1861-1888. Revista de Econom?a
Institucional, 11(20):45?81.
Kalmanovitz, S. (2011). Nueva historia econ?mica de Colombia. Taurus.
Kuznets, S. (1966). Modern economic growth: findings and reflections. Nobel foun-
dation.
LeGrand, C. (1986). Frontier expansion and peasant protest in Colombia, 1850-1936.
University of New Mexico Press.
Lewis, A. (1955). The Theory of EconomicDevelopment. Allen andUnwin, London.
Matsuyama, K. (1992). Agricultural productivity, comparative advantage, and
economic growth. Journal of Economic Theory, 58(2):317?334.
Mazzuca, S. and Robinson, J. A. (2009). Political conflict and power sharing in the
origins ofmodern colombia. Hispanic AmericanHistorical Review, 89(2):285?321.
McGreevey, W. P. (1971). An economic history of Colombia, 1845-1930. Cambridge
University Press.
Mejia, J. (2018). Social networks and entrepreneurship. evidence from a historical
episode of industrialization.
Mora, R. and Reggio, I. (2012). Treatment effect identification using alternative
parallel assumptions. Technical report, Universidad Carloss III Madrid, Work-
ing paper.
157
Moscona, J. (2018). Agricultural development and structural change, within and
across countries.
Murphy, K. M., Shleifer, A., and Vishny, R. (1989). Income Distribution, Market
Size, and Industrialization. The Quarterly Journal of Economics, 104(3):537?564.
Nieto Arteta, L. E. (1971). El caf? en la sociedad colombiana.
North, D. C., Summerhill, W., and Weingast, B. R. (2000). Order, disorder and
economic change: Latin america vs. north america. In deMesquita, B. and Root,
editors, Governing for prosperity. Yale University Press, New Haven.
North, D. C., Wallis, J. J., and Weingast, B. R. (2009). Violence and social orders: a
conceptual framework for interpreting recorded human history. Cambridge Univer-
sity Press.
Nunn, N. and Puga, D. (2010). Ruggedness: The Blessing of Bad Geography in
Africa. The Review of Economics and Statistics, 94(1):20?36.
Ocampo, J. A. (1984). The Colombian Economy in the 1930s. In Thorp, R., editor,
Latin America in the 1930s: The Role of the Periphery in World Crisis, St Antony?s
Series, pages 117?143. Palgrave Macmillan UK, London.
Ocampo, J. A. (2015). Caf?, industria y macroeconom?a: ensayos de historia econ?mica
colombiana. Fondo de Cultura Econ?mica.
Ocampo, J. A. and Botero, M. M. (2000). Coffee and the Origins of Modern Eco-
nomic Development in Colombia. In C?rdenas, E., Ocampo, J. A., and Thorp,
158
R., editors, An Economic History of Twentieth-Century Latin America: Volume 1 The
Export Age: The LatinAmerican Economies in the LateNineteenth and Early Twentieth
Centuries, pages 55?84. Palgrave Macmillan UK, London.
Ocampo, J. A. and Montenegro, S. (2007). Crisis mundial, protecci?n e industrial-
izaci?n. Editorial Norma.
O?Rourke, K. H. and Williamson, J. G. (2002). When did globalisation begin?
European Review of Economic History, 6(1):23?50.
Ortiz, L. J. (2010). Obispos, cl?rigos y fieles en pie de guerra: Antioquia, 1870-1880.
Universidad de Antioquia.
Ortiz, L. J. and Javier, L. (1840). Guerras civiles e iglesia cat?lica en colombia en la
segunda mitad del siglo xix. Grupo de Investigaci?n Religi?n, Cultura y Sociedad.
Ganarse el cielo defendiendo la religi?n. Guerras civiles en Colombia, 1902.
Ospina V?squez, L. (1955). Industria y protecci?n en Colombia. Medell?n: ESF.
Palacios, M. (2002). Coffee in Colombia, 1850-1970. Cambridge University Press.
Parsons, J. J. (1949). Antioque?o colonization in western Colombia. University of
California Press.
Perez, S. (2017). Railroads and the rural to urban transition: Evidence from 19th-
century Argentina. Technical report, University of California, Davis.
Porzio, T. and Santangelo, G. (2019). Does Schooling Cause Structural Transfor-
mation? In Barcelona GSE Forum Working Paper.
159
Posada-Carb?, E. (1995). Civilizar las urnas: conflicto y control en las elecciones
colombianas, 1830-1930. Bolet?n Cultural y Bibliogr?fico, 32(39):3?25.
Prados de la Escosura, L. (2007). When did Latin America fall behind? In The
decline of Latin American economies: Growth, institutions, and crises, pages 15?58.
University of Chicago Press.
Prebisch, R. (1950). The economic development of Latin America. ECLAC Think-
ing, Selected Texts (1948-1998). Santiago: ECLAC, 2016. p. 45-84.
Riascos Grueso, E. (1950). Geograf?a guerrera colombiana. Cali: Imprenta Bolivar-
iana.
Rosenstein-Rodan, P. N. (1943). Problems of Industrialisation of Eastern and
South-Eastern Europe. The Economic Journal, 53(210/211):202?211.
Ruggles, S., King, M. L., Levison, D., McCaa, R., and Sobek, M. (2003). IPUMS-
international. Historical Methods: A Journal of Quantitative and Interdisciplinary
History, 36(2):60?65.
Saenz, N. (1892). Memoria sobre el cultivo del cafeto. Imprenta de la Luz.
Safford, F. and Palacios, M. (2002). Colombia: Fragmented land, divided society (Latin
American histories). Oxford Univ. Press, Oxford.
Salvucci, R. (2006). Export-Led Industrialization. In Bulmer-Thomas, V.,
Coatsworth, J., and Cortes-Conde, R., editors, The Cambridge Economic History
of Latin America.
160
Santos Cardenas, D. (2017). From Skirts to Slacks: Female Workers and Wage
Gap in the Colombian Industry in 1945. SSRN Scholarly Paper ID 3040722,
Social Science Research Network, Rochester, NY.
Schultz, T. W. (1964). Transforming traditional agriculture. Yale University Press.
Shaw, C. (1941). Church and state in colombia as observed by americandiplomats,
1834-1906. The Hispanic American Historical Review, 21(4):577?613.
Squicciarini, M. P. and Voigtl?nder, N. (2015). Human Capital and Industrial-
ization: Evidence from the Age of Enlightenment. The Quarterly Journal of Eco-
nomics, 130(4):1825?1883.
Stokey, N. L. (1991). Human Capital, Product Quality, and Growth. The Quarterly
Journal of Economics, 106(2):587?616. Publisher: Oxford Academic.
Sviatschi, M. M. (2018). Making a Narco: Childhood Exposure to Illegal Labor
Markets and Criminal Life Paths.
Tovar, J. A. (2007). La manumisi?n en Colombia: 1821-1851: un an?lisis cuantitativo.
Universidad de los Andes, Facultad de Econom?a, CEDE.
Valencia Caicedo, F. (2019). The Mission: Human Capital Transmission, Eco-
nomic Persistence, and Culture in South America. The Quarterly Journal of Eco-
nomics.
Vollrath, D. (2011). The agricultural basis of comparative development. Journal of
Economic Growth, 16(4):343?370.
161
Wallerstein, I. (2011). The modern world-system I: Capitalist agriculture and the ori-
gins of the European world-economy in the sixteenth century, volume 1. Univ of
California Press.
Williamson, J. G. (2011). Industrial Catching Up in the Poor Periphery 1870-1975.
Working Paper 16809, National Bureau of Economic Research.
162