essays in comparative economic development alberto...

Essays in comparative economic development

Alberto Basso

http://www.ua.es/

http://www.eltallerdigital.com/

Essays in comparative economic development

Alberto Basso

Supervisor: Dr. David Cuberes

A thesis submitted to

the Departamento de Fundamentos del Análisis Económico

Universidad de Alicante

In partial fulfilment of the requirements for the degree of

Doctor of Philosophy

March 2013

To my family

Acknowledgements

I am grateful to my advisor, David Cuberes, for his support and advice throughout my graduatestudies. I am indebted also to Carl-Johan Dalgaard for his guidance during my visiting period

at the Department of Economics of the University of Copenhagen in the Fall of 2012.

Many thanks go to the faculty members of the Departamento de Fundamentos del AnálisisEconómico at the University of Alicante: in particular, Sonia Oreffice, Climent

Quintana-Domeque, Pedro Albarran, Asier Mariscal, and Marc Teignier.

I thank Jacob Weisdorf, Fabrice Murtin, the participants to the "7th Sound Economic HistoryWorkshop" in Tampere and to the MEHR seminar in Copenhagen for helpful comments and

suggestions.

Many thanks go to the staff of the Departamento de Fundamentos del Análisis Económico, inparticular Marilo Rufete and Josefa Zaragoza.

My appreciation and gratitude go to my classmates for their help, company and friendship: inparticular, Nathan, Serafima, Xavier, Victor, MJosé, Danilo and Fernando.

Contents

Introducción ix

0.1 Transición de la fecundidad y el compromiso cantidad-calidad: evidencia históricaen España . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

0.2 La cantidad afecta a la calidad: fecundidad, educación, y género en la España de1887 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi

0.3 Capital humano, cultura y el comienzo de la transición de la fecundidad . . . . . xviii

1 Fertility transition and the quantity-quality trade-off: historical evidence fromSpain 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Conceptual framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.3 Data description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.4 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.4.1 Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

1.4.2 Instrument choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.5 Evidence on fertility transition and quantity-quality trade-off . . . . . . . . . . . 18

1.5.1 Panel analysis: OLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.5.2 Panel analysis: 2SLS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.5.3 Long-time differences: 2SLS, 3SLS and robustness checks . . . . . . . . . 24

1.5.4 Long-time differences: spatial diffusion . . . . . . . . . . . . . . . . . . . . 26

1.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

0 Contents

2 Quantity affects quality: fertility, education, and gender in 1887 Spain 33

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.2 Data and empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.2.2 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

2.3 Quantity and quality of children: results . . . . . . . . . . . . . . . . . . . . . . . 41

2.3.1 Allowing for spatial dependence . . . . . . . . . . . . . . . . . . . . . . . . 48

2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3 Human capital, culture and the onset of the fertility transition 53

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3.3 Data and methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.3.1 Baseline analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.3.2 Bilateral analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.3.3 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.4 Results: genetic distance to the technological frontier and the onset of fertilitytransition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.4.1 Baseline analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.4.2 Bilateral analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.4.3 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.5 Verification of the mechanism: genetic distance to technological frontier, educationand fertility transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

Bibliography 79

viii

Introducción

Esta Tésis Doctoral está compuesta de tres capítulos que tratan temas relacionados con el estu-

dio de las diferencias en los niveles de desarrollo entre areas geográficas. La literatura económica

sobre el desarrollo comparativo es amplia y propone diferentes mecanismos basados en factores

culturales, institucionales, geográficos, climáticos, etc., para explicar diferentes patrones de de-

sarrollo. En este ámbito está creciendo el interes sobre el papel jugado por el proceso de transición

de fecundidad y la interacción entre la cantidad y la calidad de los hijos, es decir, el concepto de

quantity-quality trade-off. Como consecuencia de esto, el estudio de las causas de los procesos de

transición demográfica, en particular de la transición de fecundidad, en economía ha crecido en

importancia en los últimos años. La identificación de los factores que explican este fenómeno es

importante para mejorar nuestra comprensión del proceso de desarrollo entre los países y dentro

de ellos mismos. Además, nos ayuda a entender por qué algunos países han tenido éxito y otros

no. Los economistas se han centrado en el análisis de varios factores para explicar el descenso de

la fecundidad: entre los más estudiados estan el aumento de la inversión en la educación de los

hijos (es decir un incremento en la calidad), el descenso de la mortalidad infantil, y el cambio

del papel jugado por las mujeres en la sociedad. De estos factores, el incremento de la inversión

en la calidad de los hijos aparece como el más interesante, debido al reciente desarrollo de la

teoría unificada del desarrollo (unified growth theory) por Galor y Weil (2000). Esta teoría está

basada en un modelo teórico que describe el proceso de desarrollo y el cambio de una fase de

subsistencia (pre-malthusiana) a una de crecimiento económico (post-malthusiana). El mecan-

ismo fundamental de esta teoría se centra en el papel jugado por el progreso tecnológico en el

0 Introducción

fomento de la demanda de capital humano. Debido a la mayor remuneración de la educación, los

hogares reaccionan invirtiendo más recursos en la calidad de su hijos, reduciendo así su cantidad

para mantener el equilibrio presupuestario. Desde el trabajo de Galor y Weil, varios economistas

han analizado los procesos de transición en la fecundidad. Algunos ejemplos de análisis histórico

de los determinantes del descenso de la fecundidad son Bleakley y Lange (2009), Murphy (2010)

y Murtin (próximamente). Además, varios estudios se han centrado específicamente en la carac-

terización del compromiso cantidad-calidad en contextos históricos como, por ejemplo, Becker,

Cinnirella y Woessmann (2010), Klemp y Weisdorf (2011).

El objetivo de esta tésis es ampliar el análisis empírico sobre los procesos de transición de fe-

cundidad y el compromiso cantidad-calidad teniendo en cuenta los resultados de la investigación

económica reciente. Sin embargo, uno de los principales problemas a los que se enfrentan las

investigaciones empíricas que estudian el desarrollo comparativo es la dificultad en la contabi-

lización de toda la heterogeneidad que hay, entre las areas geográficas utilizadas como unidades

de análisis, y que podría ser responsable de los diferentes patrones de desarrollo. Una manera

de resolver parcialmente esta cuestión es explotar la variación en el tiempo y así poder tener en

cuenta todos los factores históricos, culturales, climáticos y geográficos que se pueden considerar

constantes en un período de tiempo. El primer capítulo de esta tésis - titulado Transición de la

fecundidad y el compromiso cantidad-calidad: evidencia histórica en España - utiliza esta estrate-

gia para analizar las causas del comienzo del proceso de transición de la fecundidad en España.

El estudio se centra en las primeras dos décadas del siglo XX y está realizado usando datos a

nivel provincial. El análisis se caracteriza por el papel jugado por los aumentos de la inversión en

la educación de los hijos como uno de los factores claves para explicar el descenso en la fecundidad.

Una segunda manera para afrontar el problema mencionado anteriormente es centrarse en analizar

zonas geográficas bastante homogéneas, como por ejemplo explotando las diferencias que hay en-

tre áreas dentro de cada país. El segundo capítulo de esta tésis, titulado La cantidad afecta a

la calidad: fecundidad, educación, y género en la España de 1887, analiza y caracteriza el com-

promiso cantidad-calidad de los hijos utilizando datos a nivel de distrito (partido judicial) en el

año 1887. En particular, se estudia el impacto de la fecundidad sobre los niveles de educación

x

0 Introducción

infantil distinguiendo el efecto por género.

Finalmente, el tercer capítulo de esta tésis, titulado Capital humano, cultura y el comienzo de

la transición de la fecundidad, en colaboración con David Cuberes analiza la interacción entre

factores culturales, acumulación de capital humano y el comienzo del proceso de transición de la

fecundidad a lo largo de diferentes países del mundo. En particular, la evidencia empírica sugiere

que una gran distancia genética, utilizada como medida de diferencias culturales, con respecto

a la frontera tecnológica (Reino Unido o Estados Unidos) retrasa el comienzo de la transición

demográfica de un país. Este resultado se puede explicar con un mecanismo indirecto que opera

a través de la difusión tecnológica sugerido por Spolaore y Wacziarg (2009). Una mayor distancia

cultural de la frontera tecnológica tiende a retrasar la adopción de tecnología, disminuyendo la

demanda de capital humano. En consecuencia, este patrón conduciría a un inicio tardío en la

transición de la fecundidad. El mecanismo sigue el trabajo de Galor y Weil (2000) que argumen-

tan que los avances tecnológicos aumentan la demanda de capital humano y, debido a la mayor

remuneración de la educación, los hogares tienden a intercambiar la cantidad por la calidad de

los hijos. Cuando una fracción significativa de las familias decide tener menos hijos pero más

educados, tiene lugar el inicio de la transición de la fecundidad. A continuación se describe de

manera más detallada el análisis realizado en cada capítulo.

0.1 Transición de la fecundidad y el compromiso cantidad-calidad:

evidencia histórica en España

El estudio de los fenomenos de transición de fecundidad ha proporcionado diferentes factores

explicativos. En este capítulo se estudia la relación entre los cambios en la educación, es decir,

la alfabetización, de los hijos y los cambios en la fecundidad de los padres utilizando datos a

nivel provincial en España durante el período 1900-1920. En este estudio se considera un período

alrededor del inicio de la transición demográfica en España para comprender sus factores des-

encadenantes. En particular, nos centramos en un mecanismo específico: la interacción entre

la calidad y la cantidad de los hijos. De acuerdo con este mecanismo, un incremento en la es-

xi

0 Introducción

colarización de los hijos afecta a la decisión de los padres sobre su fecundidad. Varios factores

podrían inducir a los padres a invertir más en la educación de sus hijos. Entre estos factores

se encuentra un incremento en la demanda de capital humano que aumentaría la remuneración

de la educación y por lo tanto la asistencia escolar. Sin embargo, las reformas educativas y los

cambios en las leyes de escolarización obligatoria también afectan a las decisiones de los hogares

sobre la educación de sus hijos. A principios del siglo XX, en España tuvo lugar una amplia

reforma del sistema escolar, que incluye la creación del Ministerio de Educación Pública en 1900

y la extensión de la edad de escolarización obligatoria de 6-9 a 6-12 años en 1909. El nuevo

sistema, llamado Escuelas graduadas, separaba a los estudiantes en clases diferentes según la

edad y el nivel educativo. El sistema anterior, llamado Escuelas unitarias, agrupaba a los estudi-

antes independientemente de su edad y capacidad. Sin embargo, el desarrollo del nuevo sistema

fue lento debido a los limitados recursos financieros y a las presiones provenientes de sectores

tradicionales que trataron de evitar cambios radicales. Explotando la variación regional entre

provincias en la demanda y en la oferta local de educación, estrechamente relacionados con la

reacción a las reformas educativas, se estudia si los cambios en la educación de los hijos estan

relacionados con los cambios en la fecundidad.

Uno de los primeros estudios sobre la natalidad, el Princeton European Fertility Project (EFP en

lo sucesivo), identificó factores culturales y sociológicos como clave para el proceso de reducción

de la fecundidad en Europa (por ejemplo, Coale y Watkins 1986). El propósito del EFP era

caracterizar la reducción de las tasas de natalidad que se inició en Europa en los siglos XIX y

XX. Sus conclusiones finales sugieren que las variables socioeconómicas desempeñaron un papel

de escasa importancia en el desencadenamiento de las transiciones de fecundidad en los países

europeos. No obstante, estudios posteriores han señalado varios defectos en el análisis utilizado

en el EFP, que podrían ser la causa de este hallazgo. Entre estos estudios, Brown y Guinnane

(2007) ponen de relieve los principales problemas estadísticos relacionados con la metodología

del EFP. En primer lugar, según los autores las unidades estadísticas de análisis están muy

agregadas. En segundo lugar, y más importante desde el punto de vista de este estudio, la mod-

elización del cambio de la natalidad en el tiempo. El enfoque del EFP no parece en consonancia

xii

0 Introducción

con el concepto de transición de la fecundidad, debido a que cambios en el comportamiento de

la fecundidad deberian ser causados por cambios en las variables explicativas. Varios estudios

que han examinados las causas de los cambios de fecundidad han encontrado que las variables

económicas han desempeñado un papel importante (por ejemplo, Galloway, Lee y Hammel 1994;

Brown y Guinnane 2002). Además, Brown y Guinnane (2007) subrayan que la mayoría de los

estudios del EFP se basan en simples correlaciones, por lo que adolecen de varios problemas tales

como la causalidad inversa y el sesgo de variables omitidas.

Estudios recientes, que abordan varios de los defectos mencionados anteriormente, han propor-

cionado evidencia histórica que sugiere la existencia de un compromiso cantidad-calidad y han

destacado su papel en el desencadenamiento del descenso de la fecundidad. Entre estos Bleakley

y Lange (2009) exploran el efecto causal de la educación infantil sobre la fecundidad mediante la

explotación de la política de erradicación de la anquilostomiasis en los estados del sur de América

del Norte en 1910. Su estudio sostiene que esta erradicación aumentó la remuneración de la ed-

ucación y por lo tanto redujo el precio de la calidad de los hijos. Como consecuencia, tuvo lugar

un aumento de la asistencia escolar y una reducción de la fecundidad. Murphy (2010) encuentra

evidencia que los factores económicos y culturales han afectado a los cambios en la fecundidad a

lo largo de los departamentos francéses a finales del siglo XIX. En particular, la asistencia a la es-

cuela primaria está asociada negativamente con la fecundidad. Becker, Cinnirella y Woessmann

(2010) utilizan datos de los condados de Prusia en 1849 e identifican una relación negativa entre

la cantidad de hijos y la asistencia a la escuela en un contexto en el que la transición demográfica

aún no ha tenido lugar. También destacan que el nivel inicial de la educación es un buen pre-

dictor de la transición de la fecundidad que se produjo en Prusia durante el período 1880-1905.

Klemp y Weisdorf (2011) encuentran un impacto negativo significativo del tamaño de la familia

sobre la alfabetización de los hijos a partir de datos procedentes de los registros parroquiales

anglicanos en Inglaterra durante el período 1700-1830. Por último, Fernihough (2011) encuentra

evidencia de compromiso cantidad-calidad a partir de datos censales para Irlanda en 1911. En

concreto, utilizando datos de Belfast y Dublín, encuentra que el aumento de la fecundidad redujo

la probabilidad de asistencia escolar.

xiii

0 Introducción

En la literatura económica no se encuentran estudios centrados en analizar el papel desempeñado

por los cambios provocados por la educación de los hijos sobre la fecundidad en España a partir

de datos históricos a nivel provincial. Los estudios previos sobre los determinantes de los niveles

de fecundidad en las provincias españolas en la primera parte del siglo XX proporcionan una

imagen desconcertante (Leasure 1963; Reher y Iriso Napal-1989). Por un lado, los diferentes

contextos culturales y linguísticos parecen importantes en determinar los niveles de la fecundi-

dad en las distintas provincias. Por otro lado, el papel jugado por los factores socioeconómicos

es incierto. Concretamente, con respecto a la educación, no hay una relación inequívoca en-

tre los niveles de educación y los niveles de fecundidad. Los estudios empíricos recientes sobre

transiciones de fecundidad y el compromiso calidad-cantidad se han centrado principalmente en

analizar los países de Europa septentrional y central. Este trabajo contribuye a ampliar el análi-

sis e incluir el sur de Europa, concretamente España, que es un área periférica con respecto al

proceso histórico de industrialización. Además, este trabajo afronta tres cuestiones principales

que la literatura económica reciente ha señalado. En primer lugar, se aborda el posible sesgo

debido a la omisión de las diversidades culturales y de las características históricas utilizando

datos de panel y estimando un modelo empírico con efectos fijos (por ejemplo, Galloway, Lee y

Hammel 1994). La omisión de las características culturales e históricas está relacionada tambien

con un problema conceptual. Con el fin de captar los factores responsables de la transición de la

fecundidad, este estudio se centra en explicar los cambios de la fecundidad en lugar de los niveles.

Esto significa implícitamente que el análisis tiene en cuenta los factores históricos y culturales

que son constantes en cada provincia y que pueden afectar tanto a la educación de los hijos como

a la fecundidad de los padres. En cuanto a la alfabetización, varias características específicas de

cada provincia pueden ser responsables de los niveles educativos infantiles tales como, por ejem-

plo, los sistemas de cultivo y las prácticas agrícolas. Estas características son particularmente

importantes ya que afectan a la productividad agrícola y a la demanda de trabajo infantil. A

su vez, éstas dependen fundamentalmente de las condiciones geográficas y climáticas que pode-

mos considerar como constantes en el tiempo. Entonces comparar simplemente los niveles de

educación y de fecundidad entre las provincias llevaría a ignorar algunas de las fuerzas que son

xiv

0 Introducción

responsables de la alfabetización infantil y de la fecundidad. En segundo lugar, se aborda el

sesgo de la endogeneidad debido a errores de medición, a variables omitidas y al problema de

causalidad inversa utilizando estimaciones de variables instrumentales (por ejemplo, Brown y

Guinnane 2002). El problema de endogeneidad puede tener varios orígenes. Entre estos se en-

cuentra la causalidad inversa que existe entre la fecundidad y la educación de los hijos. Además,

el error de medición, de los datos históricos y de censo utilizados en este estudio, es probable

que afecte los valores tomados por las variables. Finalmente, el problema de variables omitidas

debido a la falta de datos también es susceptible de afectar al análisis empírico. Para hacer

frente a estos problemas y para establecer una relación causal entre cambios en la alfabetización

infantil y cambios en la fecundidad este estudio explota diferentes estrategias de variables instru-

mentales (IV). En concreto, se instrumenta la alfabetización de los hijos con medidas directas

e indirectas de apoyo local a la educación. La medida indirecta de apoyo local a la educación

está definida como la cuota de propietarios de ganado, de tamaño medio-grande, en 1865. Esta

medida es parecida a otras ya utilizadas en estudios recientes (Galor, Moav y Vollrath 2009;

Becker, Cinnirella y Woessmann 2010). Según Galor, Moav y Vollrath (2009), la desigualdad en

la distribución de la propiedad de la tierra retrasa el desarrollo de instituciones que promuevan

la acumulación de capital humano. Este fenomeno es debido al hecho de que los grandes ter-

ratenientes no se beneficiarían de la acumulación de capital humano ya que este último no es

complementario a la tierra en su función de producción. Siguiendo este razonamiento, Becker,

Cinnirella y Woessmann (2010) utilizan la proporción de grandes proprietarios como instrumento

para la asistencia escolar infantil para estimar el efecto causal de la educación de los hijos sobre la

fecundidad en la Prusia del siglo XIX. Desafortunadamente, no existen datos sobre la propiedad

de la tierra a nivel provincial para España en el siglo XIX. Sin embargo, en 1865 se llevó a cabo

un Censo de Ganadería. Suponiendo que la propiedad del ganado, especialmente de los animales

empleados principalmente en agricultura, va junta, o, por lo menos, está bien correlacionada,

con la propiedad de la tierra, ésta puede ser utilizada para construir un instrumento para los

cambios en la educación infantil en el período 1900-1920. La medida directa de apoyo local a

la educación está definida como la inversión en educación financiada por las autoridades locales

xv

0 Introducción

dividida por el número de hijos en la edad comprendida entre 5-15. Es factible pensar que la

inversión financiada localmente afecte a las decisiones de fecundidad sólo a través de su efecto

sobre la decisión de los padres de enviar a sus hijos a la escuela, es decir, lo que desencadena el

compromiso cantidad-calidad. Finalmente, se tiene en cuenta el papel jugado por el proceso de

difusión espacial (por ejemplo, Murphy 2010). La cuestión de la difusión (o dependencia) espacial

se refiere a la presencia de patrones geográficos en los descensos de fecundidad. La difusión de

nuevas normas sociales y culturales puede ser responsable de tales patrones espaciales, los cuales

reflejan, por ejemplo, nuevas actitudes hacia las prácticas de control de la natalidad inducidas

por un proceso de modernización.

0.2 La cantidad afecta a la calidad: fecundidad, educación, y

género en la España de 1887

En 1857 la Ley Moyano estableció en España la asistencia escolar obligatoria para los hijos de

edades comprendidas entre los 6 y 9 años. Sin embargo, treinta años después la alfabetización

infantil era aún relativamente baja. En 1887 la alfabetización infantil, definida como la propor-

ción de hijos que sabian leer y escribir de edad entre los 5 y 15 años, era alrededor del 24%.

Las diferencias de género eran evidentes: la alfabetización masculina (29%) era superior a la

femenina (19%).

Los niveles educativos de los hijos son generalmente utilizados para medir su calidad. La for-

malización de la teoría de la demanda de hijos y del compromiso cantidad-calidad como un

mecanismo económico se remonta a la teoría de la familia de Gary Becker (por ejemplo, Becker

y Lewis 1973; Becker 1981). La mayoría de los estudios que investigan el efecto del tamaño de la

familia (cantidad) sobre la calidad de los hijos utiliza datos modernos (por ejemplo Angrist, Lavy

y Schlosser 2005; Black, Devereux y Salvanes 2005). Este estudio está relacionado estrechamente

con la literatura económica reciente que se centra en el análisis del compromiso cantidad-calidad

en un contexto histórico. El objetivo de este capítulo es estudiar la relación entre el nivel de

educación de los hijos y el nivel de fecundidad de los padres utilizando datos a nivel de distri-

xvi

0 Introducción

tos en España a finales del siglo XIX. Los datos utilizados en el análisis empírico provienen del

censo de población de 1887 que proporciona datos a nivel de distrito (o partido judicial). Estas

unidades, mucho más pequeñas en tamaño que las provincias, permiten lograr una muestra de

más de 400 observaciones. El uso de estas unidades estadísticas de análisis reduce los problemas,

típicos de la macroeconomía, relacionados con unidades de análisis demasiado agregadas. Para

medir la cantidad de hijos, la variable utilizada es la relación hijo-mujer calculada de dos modos

diferentes: como el número de hijos de edad 0-5 dividido por el número de mujeres de edad 16-45

y como el número de hijos de edad 6-15 dividido por el número de mujeres de edad 21-50. La

segunda medida incluye a los hijos de edad comprendida entre 6 y 15 años con el fin de eliminar

el impacto de las tasas de mortalidad infantil sobre la fecundidad, de manera que se captura

exclusivamente el número de hijos supervivientes. La calidad de los hijos es representada por la

proporción de hijos alfabetizados, es decir, capaces de leer y escribir, de edad comprendida entre

5 y 15 años. Éste parece ser un buen indicador tanto de la asistencia a la escuela primaria y

de su terminación, ya que la alfabetización es uno de los principales resultados de la enseñanza

primaria y la enseñanza obligatoria en 1887 se limitaba a hijos de edad entre los 6 y 9 años. Se

consideran tres diferentes medidas de alfabetización infantil: una que incluye tanto a los hijos

como a las hijas, una que incluye sólo a las hijas, y otra que considera exclusivamente a los hijos.

El análisis empirico está desarrolado en tres partes. La primera parte estudia la correlación entre

el nivel de fecundidad y el nivel de educación de los hijos usando estimaciones de MCO (Mini-

mos Cuadrados Ordinarios). La segunda parte aborda el sesgo de la endogeneidad debido a los

errores de medición, a las variables omitidas y a la causalidad inversa mediante estimaciones de

variables instrumentales. La variable utilizada para construir el instrumento para los niveles de

la fecundidad es la relación entre mujeres y hombres (RMH de aquí en adelante) en la población

adulta, es decir, de edad comprendida entre 21 y 50 años. La RMH en la población adulta quiere

identificar una variación exógena en los niveles de fecundidad. Esta estrategia está empleada en

otro trabajo reciente, Becker, Cinnirella y Woessmann (2010), que estudia el mismo fenómeno

pero a lo largo de los condados de Prusia en el siglo XIX. La evidencia empírica sugiere que la

fecundidad tuvo un efecto negativo sobre las tasas de alfabetización de los hijos, mientras este

xvii

0 Introducción

efecto parece menos importante en el caso de las hijas. Por un lado, el impacto significativo de la

cantidad de hijos sobre la educación de los varones confirma la existencia de un compromiso entre

cantidad y calidad en un contexto histórico como España a finales del siglo XIX. Por lo tanto,

cuando los padres tenian familias más numerosas, los hijos varones eran más propensos a ser

analfabetos. Por otro lado, este resultado es consistente con la existencia de una heterogeneidad

cultural entre las areas geográficas de España en el papel jugado por la mujer en la sociedad.

Este resultado sugiere que la fecundidad, a través de un mecanismo presupuestario y económico,

no estaba entre los principales determinantes de la escolarización de las hijas.

0.3 Capital humano, cultura y el comienzo de la transición de la

fecundidad

La transformación de una economía de una fase de estancamiento malthusiano a una fase de crec-

imiento está fundamentalmente vinculada al proceso de transición de la fecundidad. Cambiando

la relación entre ingresos y fecundidad de positiva a negativa, esta transición tiene un papel

clave en el fomento de la inversión en el capital humano y en el crecimiento económico a largo

plazo (por ejemplo Galor y Weil 1999, 2000). Como consecuencia, se observa que los países que

experimentaron primero el inicio de la transición de la fecundidad son relativamente más ricos y

más desarrollados que los que lo experimentaron más tarde o que aún no lo han experimentado.

El comienzo de la transición en la fecundidad difiere ampliamente entre países. Reher (2004) ha

estimado los años en los cuales los diferentes países alcanzaron su transición demográfica. Según

estas estimaciones, la mayoría de los países que experimentaron la transición a finales del siglo

XIX y a principios del siglo XX se encuentran en Europa occidental. En cambio, la mayoría

de los países que experimentaron una transición tardía, es decir, después de 1950, pertenecen a

Asia, África y América Latina.

Una parte reciente de la literatura económica enfatiza el papel jugado por los factores culturales

para explicar los diferentes niveles de desarrollo económico entre países (Guiso, Sapienza y Zin-

gales 2006; Spolaore y Wacziarg 2009). El trabajo de Spolaore y Warcziarg (2009) muestra que

xviii

0 Introducción

una fracción significativa de las diferencias de ingresos entre los países se puede explicar utilizando

su distancia genética en relación con el país que se ecuentra en la frontera tecnológica. Según la

opinión de los autores esta medida debería captar las barreras en la adopción y difusión de nuevas

tecnologías desde el país que se encuentra en la frontera tecnológica. La medida de distancia

genética pretende capturar una relación general y cultural entre poblaciones. Las poblaciones

que se encuentran más cercanas en términos de distancia genética tienen menores diferencias en

los rasgos y en las normas sociales tales como, por ejemplo, creencias y hábitos. Por otro lado,

la literatura destaca también el papel histórico jugado por la acumulación de capital humano en

el proceso de desarrollo de un país. La expansión de la educación es a menudo considerada como

uno de los factores fundamentales en el desarrollo económico. Análisis comparativos sugieren que,

entre varios factores, las diferencias históricas en el capital humano pueden ser responsables de

los diferentes patrones de desarrollo observados durante y después del período de la colonización.

Por ejemplo, Glaeser, La Porta, Lopez-de-Silanes y Shleifer (2004) sostienen que los colonos eu-

ropeos trajeron consigo su capital humano donde se establecieron en gran número, fomentando

así el progreso tecnológico, el crecimiento económico y la formación de mejores instituciones.

Varios mecanismos han sido propuestos para explicar el fenomeno de descenso en la fecundidad.

Entre estos se encuentran un aumento de la inversión en la calidad de los hijos debida a un

incremento de la demanda del capital humano por el efecto de los avances tecnológicos (Galor y

Weil 2000), un aumento de los ingresos durante el período de industrialización (Becker y Lewis,

1973; Becker 1981), una reducción de la tasa de mortalidad infantil que reduce los motivos pre-

ventivos y de reemplazo (Coale 1973; van de Walle 1986; Sah 1991; Galloway, Lee y Hammel

1998; Eckstein, Mira y Wolpin 1999; Kalemli-Ozcan 2002; Angeles 2010); una reducción de las

brechas de género que provoca un cambio en el papel jugado por las mujeres en la sociedad

(Galor y Weil 1996; Goldin 1990; Lagerlöf 2003). Trabajos recientes como aquellos realizados

por Guinnane (2011) y Galor (2012) analizan en detalle las teorías y los estudios empíricos que

se han centrado en explicar los factores detrás de las transiciónes en la fecundidad.

En este estudio se explotan las variaciones en los factores culturales y en el capital humano entre

países para determinar los factores explicativos del comienzo de las transiciones de la fecundidad

xix

0 Introducción

en el mundo. El objetivo es explorar la contribución de una variable específica en el proceso de

transición demográfica y racionalizar el mecanismo a través del cual ha operado. En concreto, el

análisis se centra en la relación cultural de un país con la frontera tecnológica y en mostrar que

su impacto puede atribuirse a su efecto sobre la acumulación de capital humano. Tras el estudio

de Warcziarg y Spolaore (2009), que utilizan la distancia genética con respecto a Reino Unido

(UK) y a Estados Unidos (EE.UU.) como una medida de la relación cultural con la frontera

tecnológica, este trabajo estudia si la distancia genética de Reino Unido (o EE.UU.) ha sido un

factor importante en determinar las diferencias en el comienzo de la transición de la fecundidad

entre países. Este hecho se puede explicar con un mecanismo indirecto que opera a través de

la difusión de tecnologías como el sugerido por Spolaore y Wacziarg (2009, 2011). Una mayor

distancia cultural con respecto a la frontera tecnológica retrasaría la adopción de tecnología y

disminuiría la productividad y la demanda del capital humano. Como consecuencia, este patrón

conduciría a un inicio tardío de la transición en la fecundidad. El mecanismo empleado aquí sigue

el trabajo de Galor y Weil (2000). Éstos argumentan que los avances tecnológicos aumentan la

demanda del capital humano y, debido a la mayor remuneración de la educación, los hogares

intercambian la cantidad por la calidad de los hijos. Cuando una fracción significativa de las

familias decide tener menos hijos pero más educados, tiene lugar el inicio de la transición en la

fecundidad. Por lo tanto, los factores culturales y las instituciones informales, al afectar a los

incentivos para innovar y acumular capital humano, podrían haber afectado al comienzo de la

transición de la fecundidad y, en consecuencia, la actual distribución de los ingresos entre los

países del mundo. El razonamiento es que la distancia genética de Reino Unido (o EE.UU.), a

través de su efecto sobre la adopción de tecnología y sobre la acumulación de capital humano,

ha facilitado el inicio de la transición. Sin embargo, esto no significa necesariamente que el país

situado en la frontera tecnológica tenga que ser el primero en experimentar dicha transición.

Existen otros factores que son importantes para explicar la aparición de las transiciones en la

fecundidad.

A lo largo del análisis se tiene en cuenta el efecto de las características geográficas y climáticas,

como por ejemplo la latitud de cada país y el índice de malaria. Se tiene también en cuenta

xx

0 Introducción

factores históricos como la densidad de población en el año 1400 y los años pasados desde la

revolución neolítica (es decir, la transición agrícola). También se controla por el tipo de orígen

en las leyes del país y por medidas de similitudes linguísticas y religiosas. Además, se controla

por diferentes medidas históricas de la calidad institucional, como por ejemplo el nivel de democ-

racia. Algunas de las fechas del comienzo de la transición en la fecundidad proporcionadas por

Reher (2004) pueden ser criticadas porque a veces asignan un año relativamente tardío, como

por ejemplo en el caso de Francia. Para tener en cuenta esto, se utilizan tambien las fechas

proporcionadas por otros estudios, en particular Coale y Watkins (1986) y Bailey (2009). Estas

fechas alternativas conciernen al comienzo de las transiciones en la fecundidad en países que

experimentaron este proceso más allá en el pasado. También siguiendo el enfoque de Spolaore

y Wacziarg (2009), se evalúa si la distancia cultural de la frontera tecnológica es un factor de-

terminante del inicio de la transición demográfica con un análisis bilateral, es decir, analizando

países de dos en dos. Una ventaja de este enfoque es que permite aumentar el conjunto de datos

de manera significativa y, en consecuencia, ayuda a aumentar la precisión de las estimaciones.

En concreto, se estima una regresión de la distancia en el inicio de la transición de fecundidad

entre cada par de países sobre la distancia genética en relación al Reino Unido (o EE.UU.) y un

conjunto de variables de control muy similares a las utilizadas por Spolaore y Wacziarg (2009),

con el objetivo de capturar diferencias o similitudes geográficas, climáticas, e históricas.

Los principales hallazgos de este estudio se pueden resumir de la siguiente manera. En primer

lugar, una distancia genética grande con respecto al Reino Unido (EE.UU.), es decir, una difer-

encia cultural grande retrasa el comienzo de la transición en la fecundidad de un país. Este

efecto perdura despues de controlar por varios determinantes del desarrollo a largo plazo y de la

productividad sugeridas por la literatura económica. Segundo, se encuentra un efecto causal del

capital humano sobre la fecha de comienzo de la transición en la fecundidad, como el predecido

por Galor y Weil (2000). Este resultado se obtiene instrumentando los niveles de la escolarización

de un país en el año 1870 con la distancia genética con respecto al Reino Unido y con otra medida

que captura los incentivos a acumular capital humano, en concreto una medida de la difusión

del protestantismo.

xxi

Chapter 1

Fertility transition and the

quantity-quality trade-off: historical

evidence from Spain

1.1 Introduction

This paper studies the relationship between children’s education and parents’ fertility using

provincial level data in Spain during the first decades of the 20th century. We focus on the

mechanism through which children’s schooling affects parents’ fertility choice: the interaction

between quality and quantity of children. Different factors might induce parents to invest more

in the education of their children. Among these, an increase in the demand for human capital

that would raise returns to schooling and hence school attendance. However also educational

reforms and changes in compulsory schooling laws affect households’ schooling decisions. One of

the first steps to implement a system of primary school at the national level in Spain dates back

to the Ley Moyano of 1857. This established compulsory schooling attendance between 6 and 9

years old, that could be voluntarily extended till the age of 12. However during the 19th century

school attendance was relatively low on average in Spain, especially compared to other European

1 Fertility transition and the quantity-quality trade-off: historical evidence from Spain

countries as it can be noticed looking at illiteracy rates.1 Primary school enrolment was generally

free in public schools, the latter being financed at the municipality level through income-based

taxes (Nuñez 2005a).2 During the whole 19th century the system of financing for primary school

was decentralized, meaning that the burden was on municipalities that had to collect resources

from households. The financing of public education depended heavily on local funding till the

end of the second decade of the 20th century, meaning that large inequalities in education related

expenditures across localities persisted (or even increased). The share of investments funded by

local authorities over total investments was 0.53 in 1900, 0.35 in 1910 and 0.38 in 1920 while

it decreased substantially in the following years.3 At the beginning of the 20th century, Spain

witnessed the onset of a broad reform of the schooling system including the establishment of

the Ministry of Public Education in 1900 and the extension of compulsory schooling age from

6-9 to 6-12 in 1909. However, the development of the new system (called escuelas graduadas)

was slow due to limited financial resources and pressures coming from traditional sectors that

tried to avoid radical changes.4 Exploiting regional variation across provinces in local demand

and supply for education - closely related to the reaction to the educational reforms - we study

whether changes in children’s education are related to changes in fertility. This study considers

a period around the onset of the demographic transition at the country level (i.e. 1900-1920),

so to understand its triggers. Reher (2004) provides estimates of the year of the onset of the

demographic transition for a large group of countries. Reher (p. 21) explains the criterion

for choosing the year of the onset of the transition: "It has been set at the beginning of the first

quinquennium after a peak, where fertility declines by at least 8% over two quinquennia and never

increases again to levels approximating the original take-off point". Accordingly, Spain started

the transition in 1910. This means that - at the country level - the time period we consider can

1According to Morrisson and Murtin (2007), the illiteracy rate in Spain in 1900 was 0.59. Compared to otherwestern European countries it stands out as relatively high: 0.05 in Austria, 0.19 in Belgium, 0.16 in France, 0.18in Ireland.

2Even if primary school enrolment was essentially free, sending children to school entailed indirect costs interms of foregone income due to child labour.

3One of the reasons why our analysis focuses on the period 1900-1920 is that our identification assumption inthe instrumental variable strategies relies on the fact that the local environment played a major role in triggeringthe process of education expansion.

4The new system separated students in different classes according to age and level of education. The previoussystem (called escuelas unitarias) pooled students together independently of their age and ability.

2


10

20

30

40

cbr

1880 1900 1920 1940year

Spain Portugal

England and Wales France

Switzerland

Figure 1.1: Crude birth rate in selected European countries: 1880-1940 (Source: Mitchell, 2007)

be divided in a pre-transition (1900-1910) and a post-transition (1910-1920) decade. Figure 1.1

shows the time-series of the crude birth rates in selected European countries for the period 1880-

1940. We can notice that fertility was higher at the beginning of the period in the two southern

countries (Spain, Portugal) compared to those belonging to continental Europe (England and

Wales, France and Switzerland). This difference persisted until 1940 due to a later onset of a

(sustained) fertility decline.

Recent studies have provided historical evidence suggesting the existence of a quantity-quality

trade-off and have emphasised its role in triggering fertility declines. Among these Becker et

al. (2010) for 19th century Prussia, Murphy (2010) for late 19th century France, Klemp and

Weisdorf (2011) for 18th century England and Fernihough (2011) for early 20th century Ire-

land. To our knowledge there are no specific studies focusing on the role played by increases

in children’s education in triggering fertility declines in historical Spain using provincial level

data. This study then adds to the literature by providing evidence in a peripheral European

country that experienced the transition later compared to other western countries. Previous

studies on the determinants of fertility levels across Spanish provinces in the first part of the

3


20th century highlight a puzzling picture (Leasure 1963; Reher and Iriso-Napal 1989). While it

is stressed that cultural and linguistic contexts are important in shaping fertility profiles across

provinces, the role of socio-economic factors is uncertain. In particular, regarding education,

no unambiguous negative relationship between education (i.e. illiteracy in the total population)

and fertility levels is found. Regarding the latter finding, Reher and Iriso-Napal (1989) state

(p. 410): "Regional differences in literacy probably date back as far as the sixteenth century and

were relatively impervious to social and economic change. In other words, the index need not

necessarily be interpreted as a sign of modernization or of changing value structures. Thus, while

in the northeastern part of the peninsula high literacy indicates development, in the northwest

relatively high literacy may well be a sign of traditional rather than innovative behaviour."

The contribution of this paper is to study the role played by increases in children’s education

in triggering fertility declines in Spain in the early 20th century, tackling three main issues that

the recent literature pointed out. First, we address the potential bias due to the omission of

cultural and historical characteristics using fixed-effects in a panel framework (e.g. Galloway et

al. 1994). Second, we address the endogeneity bias due to measurement error, omitted variables

and reverse causality using IV estimation (e.g. Brown and Guinnane 2002). Finally we account

for the role of spatial dependence or diffusion (e.g. Murphy 2010).5 The main finding of the

paper is that there is evidence of a negative association between children’s education and parents’

fertility across Spanish provinces in a period around the onset of the fertility transition. This

is consistent with those theories - as unified growth theory - arguing that increases in children’s

quality are important in explaining fertility declines.

The paper is structured as follows. Section 1.2 motivates the analysis and reviews the literature

on the determinants of fertility transitions focusing on a specific mechanism, the quantity-quality

trade-off. Section 1.3 describes the data and the main variables used in the analysis while Section

1.4 introduces the baseline empirical strategy. Section 1.5 displays the results of the empirical

analyses including panel and long-time differences frameworks. Finally, Section 1.6 concludes.

5These problems are described in more details in Section 1.2.

4


1.2 Conceptual framework

As mentioned in the introduction we study the relationship between changes in children’s educa-

tion and changes in fertility tackling three main issues: the potential bias due to the omission of

cultural and historical characteristics, the endogeneity bias due to measurement error, omitted

variables and reverse causality and the role of spatial dependence. In this section we motivate

why it is important to account for these problems and we review the related literature.

1.2.1 Motivation

The first issue - the omission of cultural and historical characteristics - is related to a concep-

tual problem: in order to capture the factors responsible of the fertility transition we focus

on explaining changes in fertility rather than levels. This implicitly means that the analysis

takes into consideration province-specific cultural and historical factors that might affect both

children’s education and parents’ fertility. Regarding literacy, several province-specific charac-

teristics might be responsible of educational levels such as, for example, farming systems and

agricultural practices. These are particularly important as they shape agricultural productivity

and the demand for child labour: also, they depends fundamentally on geographic and climatic

conditions that we can regard as constant over time. Simply comparing levels of education and

fertility across provinces would lead to ignore some of those forces that are responsible of literacy

and fertility behaviour. To understand why this is the case let’s consider a simple variation of the

framework commonly used to characterize fertility choice and its interaction with investments in

children’s quality (Galor 2012). Assume an hypothetical household enjoys utility from consuming

an amount of generic goods, c, the quantity of children, n and their human capital, h according

to the following utility function:

U = (1− γ) ln c+ γ[ρi lnn+ β lnh] (1.1)

5


The term ρi represents cultural, social and historical factors that should captures heterogeneity

in preferences over quantity of children. The household budget constraint, naming household’s

income y, takes the form:

yn[τq + τee] + c ≤ y (1.2)

As in Galor (2012) the term in square brackets in Equation 1.2 is the time cost of raising a child

with education e, where τq is the fraction of time endowment necessary to take care of a child

while τe is the fraction of time endowment necessary for one unit of education per child. Solving

for the optimal level of children yields:

n =γρi

γρi + (1− γ)

1

τq + τee(1.3)

Even if the quantity-quality trade-off tells us that quantity and quality of children are negatively

related, in some cases evidence of such association might be weak. As it can be noticed from

Equation 1.3 even if parents intend to invest few resources in children’s education, they might

have relatively few children if their preferences do not place a high weight on quantity of children

(i.e. low ρi).

The second issue - endogeneity - might have several sources. The simple model developed above

tells us that reverse causality between fertility and children’s education has to be taken into

account.6 Measurement error using historical and census data is likely to affect the values taken

by our variables. Omitted variables - due to data unavailability - are also likely to affect our

analysis. To cope with these issues and establish a causal link, we exploit alternative instrumental

variable (IV) strategies by instrumenting children’s literacy with direct and indirect measures of

local support to education. The direct measure is constructed using investments in education

6This is because the optimal choice in terms of children’s education is itself a function of the quantity ofchildren parents decide to have over their lifetime.

6


funded by local authorities and it allows to perform IV estimation in a panel framework.7 Locally

financed investments can be regarded as a direct measure of local support to education and should

affect fertility decisions only through their effect on the decision of parents to send their children

to school, thus triggering the quantity-quality trade-off. In addition, we use as additional (and

alternative) instrument an indirect measure of local support to education following recent works

(Galor et al. 2009; Becker et al. 2010): a time-invariant measure capturing the share of medium-

large livestock owners in 1865 that is used in a long-time differences set-up.

The third issue - spatial diffusion - relates to the presence of geographical patterns in fertility

declines. The diffusion of new social and cultural norms might be responsible of such spatial

patterns, reflecting for example new attitudes towards birth control practices induced by a mod-

ernization process. We account for this possibility and show that our main result is robust to

controlling for such phenomena.

1.2.2 Literature review

The literature analysing the determinants of fertility choice and demographic transitions across

and within countries has proposed several possible explanations (Guinnane 2011; Galor 2012).

This paper focuses on a specific one: the role played by increases in children’s education in

triggering fertility declines. The formalization of the theory of the demand for children and of the

quantity-quality trade-off as an economic mechanism date back to Becker’s theory of the family

(e.g. Becker and Lewis 1973; Becker 1981). Accordingly the main trigger of the fertility decline

is a change in the relative price of quantity and quality of children. While this change might have

several causes, the one originally suggested as crucial is rising income, under the assumption that

as income increases parents shift their preference from quantity to quality of children (Becker

and Lewis 1973). This effect takes place assuming that the substitution effect is larger than the

income effect.8 Other recent works have focused instead on the role of technological progress in7Both supply and demand factors are important in explaining educational attainments. An increase in local

investments in education (i.e. in the supply of school services) is likely to be driven by - and will be more effectivein fostering school attendance where there is - higher demand for education.

8Higher incomes, besides shifting parents’ focus towards quality, might redirect expenditure towards otherconsumption goods as suggested by Guzman and Weisdorf (2010). They argue that, assuming children and other

7


fostering the demand for human capital. Due to higher returns to education, households would

react by investing more in the quality of their offspring, thus reducing their quantity to keep their

budget balanced (e.g. Galor and Weil 2000). Even if the factors fostering higher educational

attainments can be multiple - including educational reforms as the extension of compulsory

schooling age - the ultimate effect of increases in children’s education would be a reduction in

the optimal number of children.

One of the first comprehensive studies, the Princeton European Fertility Project (EFP here-

after), identified cultural and sociological factors as key in the process of reduction of fertility

across Europe (e.g. Coale and Watkins 1986). The purpose of the EFP was to characterize

the reduction in fertility rates that started in Europe during the 19th and early 20th centuries.

Its final conclusions suggest that socio-economic variables played a minor role in triggering the

demographic transitions across European countries.9 Subsequent studies pointed out several

flaws in the analyses used within the EFP framework that might be the cause of such finding.

Among these, Brown and Guinnane (2007) stress two main statistical problems related to the

EFP methodology. First, the statistical units of analysis, that according to them are too aggre-

gated. Second, and most important from the perspective of this study, "the way that it modelled

change over time" (p. 585). Basically the approach of the EFP is not in line with the concept

of fertility transition, that is where changes in fertility behaviour should be caused by changes

in the explanatory variables. Several studies that looked at the causes of fertility changes have

found that economic variables played an important role (e.g. Galloway et al. 1994; Brown and

Guinnane 2002). Also, Brown and Guinnane (2007) stress that most of these studies use simple

cross-section and bivariate correlations, so they suffer of several issues such as reverse causality

and omitted variables bias.

Recent empirical studies - tackling several of the flaws just mentioned - support the existence of

a negative relationship between children’s schooling attainments and parents’ fertility in histor-

consumption items are normal goods and substitutes to each other, an increase in the variety of consumptiongoods might lead to a lower demand for children.

9An exception is van de Walle (1980) that, using educational data from military recruit tests in Switzerland,finds a negative correlation between marital fertility and education in a period around 1900 and a general tendencyof fertility to decline as educational attainments increased.

8


ical contexts.10 Bleakley and Lange (2009) explore the causal effect of education on fertility by

exploiting the eradication policy of the hookworm disease in southern states of North America

in 1910. Their study argues that this eradication increased the return to schooling and hence

reduced the price of child quality, thus increasing school attendance and reducing fertility. Mur-

phy (2010) finds that both economic and cultural factors affected fertility changes across French

department in the late 19th century. In particular, enrolment in primary schools is found to

be negatively associated to fertility. Becker et al. (2010) use data on Prussian counties in 1849

and identify a negative relation between child quantity and enrolment at school in a context in

which the demographic transition has not yet taken place. They also highlight that the initial

level of education is a good predictor of the fertility transition that occurred in Prussia during

the 1880-1905 period. Klemp and Weisdorf (2011) find a significant negative impact of family

size on children’s literacy using data from Anglican parish registers in England over the period

1700-1830. Finally, Fernihough (2011) finds evidence of the quantity-quality trade-off using cen-

sus data for Ireland in 1911. Specifically, using data for Belfast and Dublin, he finds that higher

fertility (measured by the number of siblings) reduced the probability of school attendance.

1.3 Data description

This paper studies the effect of quality on quantity of children in Spain using historical provincial

level data for the period 1900-1920. Data is taken from population censuses (carried out in 1900,

1910, 1920) and several other sources including the anuarios (i.e. yearly statistical issues).11

The main variables used in the analysis are related to parents’ fertility and children’s education.

To measure fertility we use an index of marital fertility - specifically the Ig index - which is a

measure of fertility within marriage. As a robustness check we also use the total fertility rate

(TFR).12 To capture quality of children we use the share of children (aged 5 to 15) that can read

10The literature includes also cross-country historical analysis of the determinants of demographic transitions(e.g. Murtin, forthcoming).

11Population censuses and yearly statistical issues are available at www.ine.es.12Both measures are taken from Delgado (2009). Marital fertility is computed following the methodology in

Coale and Watkins (1986).

9


Álava

Albacete

Alicante

Almería

Ávila

BadajozBaleares

Barcelona

Burgos

Cáceres

Cádiz

Canarias

Castellón de La Plana

Ciudad RealCórdoba

La Coruña

Cuenca

Gerona

Granada

Guadalajara

Guipúzcoa

Huelva

Huesca

Jaén

León

Lérida

Logroño

Lugo

Madrid

Málaga

Murcia

Navarra

Orense

Oviedo

Palencia

Pontevedra

Salamanca

SantanderSegoviaSevilla

Soria

Tarragona

Teruel

ToledoValencia

Valladolid

VizcayaZamora

Zaragoza

.2.4

.6.8

1P

rim

ary

school attendance in 1

908

0 .2 .4 .6 .8Children’s literacy in 1910

Corr=0.73***

Figure 1.2: Children’s literacy and primary school attendance around 1910

and write. Literacy should proxy for school attendance and basic educational attainment as it

is the main output of primary school. To provide some evidence suggesting that this is actually

the case, we look at the correlation between primary school attendance (for children aged 6-12)

in 1908 and the share of literate children (aged 5-15) in 1910.13 As displayed in Figure 1.2 the

correlation between these two variables is high and significant, suggesting that children’s literacy

is indeed a good proxy for primary school attendance.

A specific issue regarding our measure of children’s education has to be taken into account. As

fertility and fertility changes affect the population age structure, provinces might differ in the

age distribution of children between 5 and 15 years old. Since younger children would tend to

be declared illiterate, there is a mechanical effect from fertility to the share of educated children

via the age structure of children aged 5 to 15. Hence to make the share of literate children

comparable across provinces and over time, we construct an age-adjusted measure of children’s

education. First, we compute age-specific literacy rates for children aged 5, 6, 7, 8, 9, 10 and

11-15:14

13School attendance is not available for the whole period under consideration, so it cannot be used as measureof children’ education.

14Census data allows for this disaggregation of children’s literacy by age.

10


sharelitj = # literate children in age group j# children in age group j ∗ 100

where j=1,2,3,4,5,6,7 represents age groups 5,6,7,8,9,10,11-15 respectively.

Using as weights the share of children in each of these categories in the U.S. population (aged 5 to

15) in 2000 (here used as standard population) we define the age-adjusted measure of children’s

education as follows:

adjeduc=7∑

j=1(sharelitj ∗ wj) where

7∑j=1

wj = 1

Finally, several control variables are included in the empirical specifications. The share of adult

men (aged 20-60) that work in the agriculture or fishery sector and the share of adult individuals

(aged 20-60) employed in the industry sector; the share of population living in urban areas

defined as the fraction of individuals living in towns with more than 10000 inhabitants. These

variables aim at measuring the stage of development and the economic structure of the province.

Population density is used to capture the degree of interaction between individuals and the

consequent sharing of information (for example on birth control practices, mortality events,

etc.). Child (infant) mortality and life expectancy at 15 to measure the mortality environment

during both childhood and adulthood.15 To account for the effect of inter-provincial migration

flows, we use measures of permanent and temporary in-migration.16 Table 1.1 lists the variables

used in our analysis and their sources.

1.4 Empirical strategy

This paper studies the association between children’s education and parents’ fertility controlling

for province-fixed characteristics, so to get rid of all unobservable factors that can be assumed

constant within provinces over time.17 This methodology is consistent with the concept of fertility15Several theories emphasize the importance of precautionary and replacement motives to explain fertility

behaviour (e.g. Kalemli-Ozcan 2003).16Permanent in-migration is measured by the number of individuals born in another province over total popu-

lation. Temporary in-migration is taken from Silvestre (2007).17The implicit assumption is that a province fixed effect should capture those geographical, climatic and cultural-

historical factors that are constant over the time period considered.

11


Table

1.1:Variables

anddata

sourcesChildren’s

education(aged

5-15)Author’s

computation

usingpopulation

censuses(1900,1910,1920)

Age-adjusted

children’seducation

(aged5-15)

Author’s

computation

usingpopulation

censuses(1900,1910,1920)

andUSPopulation

Census

(2000)Index

ofmaritalfertility

Ig

Delgado

(2009)Totalfertility

rate(T

FR)

Delgado

(2009)Crude

birthrate

(CBR)

Author’s

computation

usingstatisticalyearbooks

(1912,1915,1930)and

populationcensuses

(1900,1910,1920)Crude

birthrate

(nationaltime-series)

Mitchell(2007)

Sharein

agriculture,men

(aged20-60)

Author’s

computation

usingpopulation

censuses(1900,1910,1920)

Sharein

industry(aged

20-60)Author’s

computation

usingpopulation

censuses(1900,1910,1920)

Shareurban

Author’s

computation

usingpopulation

censuses(1900,1910,1920)

Population

densityStatisticalyearbooks

(1921-22,1931)Child

mortality

rateDopico

andReher

(1999)Infant

mortality

rateDopico

andReher

(1999)Life

expectancyat

15Dopico

andReher

(1999)Perm

anentin-m

igrationAuthor’s

computation

usingpopulation

censuses(1900,1910,1920)

Tem

poraryin-m

igrationSilvestre

(2007)Realinvestm

entsin

educationfunded

bylocalauthorities

Mas

Ivarsand

Cucarella

Torm

o(2009)

Prim

aryschoolattendance

(childrenaged

6-12)in

1908Author’s

computation

usingstatisticalyearbook

(1912)Nuptiality

indexDelgado

(2009)Share

wom

en(aged

16-45)Author’s

computation

usingpopulation

censuses(1900,1910,1920)

Sharemarried

wom

en(aged

16-45)Author’s

computation

usingpopulation

censuses(1900,1910,1920)

12


transition which entails the characterization of fertility changes over time.

1.4.1 Framework

The relationship between children’s education and parents’ fertility is characterized as follows:

ferti,t = γ1 educi,t + γ2Xi,t + ρi + θt + ψi,t (1.4)

where t=1900, 1910, 1920 and i=1,..49, ferti,t is parents’ fertility at time t in province i, educi,t

is the share of literate children aged 5-15 at time t in province i, Xi,t includes control variables,

ρi captures province-specific characteristics and θt include time effects.18

Equation 1.4 is first estimated by OLS. To cope with endogeneity issues, we use an instrumental

variable strategy where children’s literacy is instrumented with a time-varying measure of local

support to education. Then, the estimation framework is reduced to a simple cross-section

relating changes in children’s education to changes in fertility in the period 1900-1920:

∆ferti = γ1∆educi + γ2∆Xi + const+ εi (1.5)

In such a way we can rely also on a time invariant instrument for the change in children’s

education between 1900 and 1920.

18Throughout the analysis we apply the log transformation to the measures of fertility and children’s educationso to provide easy to interpret coefficients’ estimates (i.e. elasticities) and reduce the impact of outliers.

13


1.4.2 Instrument choice

The period considered in this study is characterized by the beginning of a process of transforma-

tion of the educational system that took place with large heterogeneity across the country. Hence,

it is reasonable to think that part of the changes in schooling achievements across provinces are

due to the different support that citizens gave to this process. To assess whether the causal effect

of changes in children’s education on changes in parents’ fertility is indeed present, changes in

children’s literacy are instrumented with both a direct and an indirect measure of local support to

education expansion: per child investments in education funded by local authorities and the share

of medium-large livestock owners in 1865. Regarding the former, this is available for the whole

period under consideration and it allows to implement an IV estimation in a panel framework.

This measure can be regarded as a valid instrument if it satisfies the exclusion restriction, that

is it does not affect fertility directly but only through children’s education. While expenditure

financed by the central government is determined by several factors, locally funded expenditure

(i.e. the local supply of school services) should reflect the local demand for education. A bet-

ter supply of education - responding to the higher demand - affects the decision of parents to

send their children to school by increasing the incentives of parents to educate their children.

Our reasoning is that in provinces where local efforts to promote and develop schooling were

higher, the (current and/or expected future) demand for human capital was higher too. House-

holds recognizing that school attendance would bring future returns (or learning this by looking

at the behaviour of their peers) would then reduce the number of children to afford schooling

expenditures. If one accepts this reasoning, our measure of local support to education should

affect fertility decisions only through its effect on the decision of parents to send their children

to school, thus triggering the quantity-quality trade-off. We provide some evidence in favour of

the validity of the exclusion restriction in Section 1.5.3 where we use multiple instruments.

The second instrument, an indirect measure of local support to education, is the share of medium-

large livestock owners in 1865. It is a time-invariant measure that should capture exogenous

variation in the support to education expansion following Galor et al. (2009). According to

14


the latter, inequality in the distribution of land ownership delays the implementation of human-

capital promoting institutions because large landowners would not gain from the accumulation

of human capital since the latter is not complementary to land in production. Following this

reasoning, Becker et al. (2010) use land ownership inequality as instrument for education to

estimate the causal effect of children’s education on parent’s fertility choice in Prussia using

cross-county data. Unfortunately, to our knowledge, no data on land ownership at the province

level is available for Spain in the late 19th century. However, in 1865 a livestock census (Censo

de Ganaderia) was carried out. Assuming that livestock ownership, especially of those animals

employed mainly in agriculture, goes along (or, at least, is well correlated) with land ownership,

it can be used to construct an instrument for changes in educational attainments.

Demand factors seem particularly important in explaining cross-province differences in education

in historical Spain. Nuñez (2005b) suggests that (p.132):"...in regions where peasant owners were

more numerous, however, demand for primary schooling was also higher and the school calendar

was frequently adapted to the agricultural one to make work compatible with schooling. Day-

labourers put a very low premium on schooling and education while peasants expected higher

rewards and were thus more committed to it".19 Hence provinces with a relatively large amount

of small land owners would be characterized by a higher local support (both in terms of demand

and supply factors) for education and by a higher propensity to react positively to the incentives

provided by educational reforms, especially the change in the compulsory schooling age. On the

other hand, provinces with large land owners (and consequently many day-labourers) would react

slowly or not react since the majority of individuals (both land owners and day-labourers) would

not gain by educating their children.20 Our measure of livestock ownership aims at capturing

this source of heterogeneity that seems plausibly exogenous in our context. The identification

assumption is that, if in a given province there is a sufficiently high number of small livestock

owners, demand and supply factors will favour schooling in response to an educational reform

19According to Núñez (2005b) these rewards might come from reduced transaction costs related to changes inland property-rights and to other market-related elements.

20The presence of large land owners in a given province affect the opportunity cost of the time spent in school.In fact, as large farms need a high number of daily labourers, parents would face higher incentives to send theirchildren to work rather than to school.

15


aiming to expand education. On the other hand human capital accumulation would be slower

where large land owners and day-labourers sum up to a large share, as they would not (or to a

lower extent) support education expansion. Hence the reforms affected differently the decision of

parents to send their children to school, and through this channel also the decision about their

fertility.21

The Censo de Ganaderia of 1865 provides information on how many livestock owners were enti-

tled a certain amount of units of different type of animals. Since this information is available at

the province level, it can be used to construct a measure of the share of medium-large livestock

owners. It also provides the allocation of each animal according to the task it was assigned.

Overall there are five possible destinations: consumption, agricultural work, machines’ move-

ment, transportation and reproduction (including production of dairy products, etc.). Among

all type of animals the ones that were assigned, among others, to agricultural tasks are the follow-

ing: cattle (cows, oxen), mules, donkeys and horses. Analysing the distribution across Spanish

provinces, two main features characterize the allocation of these animals according to the above

tasks (see Table 1.2). First, the animal that within its type is used mostly in agriculture is the

mule followed by donkeys, bovine animals and horses. Second, by looking only at the animals

used in agriculture the most used is cattle followed by mules, donkeys and horses. Average

mules, donkeys, horses and cattle per owner are 1.9, 1.4, 1.8 and 4.8 units, respectively. Hence,

to capture medium-large livestock owners a lower bound of 3 units is used for mules, donkeys

and horses while of 5 units for cattle. In order to check whether this measure is a good proxy

of the share of large land owners, we look at the correlation between the share of medium-large

livestock owners in 1865 and a measure of land ownership concentration in 1924 for 27 (out of 49)

provinces. Land ownership inequality is the share of land owners with more than 100 hectares

of land in 1924. Despite the 60-years time period passed since 1865, the correlation between this

measure and our proxy of livestock ownership concentration is high (0.7) as it can be noticed by

looking at Figure 1.3.

21Large land owners might have had a role in marriage decisions, but as Becker et al. (2012) suggest it isunlikely that they could affect decision about (changes and levels of) marital fertility. Hence since we use anage-adjusted indicator of marital fertility, we can exclude a direct effect from our instrument to the dependentvariable.

16


Tab

le1.2:

LivestockCensusof

1865

Distributionof

each

type

across

activities

Con

sumption

Agriculture

Machines

Transpo

rtReprodu

ction

Cattle

0.07

0.56

00.03

0.34

Mules

00.62

0.01

0.26

0.1

Don

keys

00.48

00.37

0.14

Horses

00.35

0.01

0.31

0.33

Distributionof

each

type

inagricultu

reCattle

Mules

Don

keys

Horses

Agriculture

0.43

0.27

0.23

0.07

Owne

rshipsize

Cattle

Mules

Don

keys

Horses

Per

owner,

average

4.8

1.9

1.4

1.8

Owne

rshipdistribution

Mean

Std.

dev.

Min

Max

Shareof

largeliv

estock

owne

rs0.14

0.07

0.02

0.31

(mules,c

attle,

donk

eys,

horses,),a

vg.

Dataon

49Sp

anishprovincescollected

from

theliv

estock

census

of18

65.

17


Albacete

Alicante

Almería

Ávila

BadajozCáceres

Cádiz


Ciudad RealCórdoba

CuencaGranada

Guadalajara Huelva

Jaén MadridMálagaMurcia

Palencia

Salamanca

Segovia

Sevilla

Soria

Toledo

ValenciaValladolid

Zamora

0.0

2.0

4.0

6Landow

ners

hip

inequalit

y in 1

924

0 .36Share of medium−large livestock owners in 1865

Figure 1.3: Livestock and land ownership inequality

1.5 Evidence on fertility transition and quantity-quality trade-off

1.5.1 Panel analysis: OLS

Table 1.3 displays some descriptive statistics that characterize our sample. As it can be noticed

in the period 1900-1920, on average, there is a decline in parents’ fertility while the share of

literate children increases. As Figure 1.4 shows, there is a negative association between changes

in children’s education and changes in fertility across Spanish provinces. Our analysis tries to

establish a causal link between these two phenomena. As already mentioned, not fully accounting

for cultural and historical factors that are province specific might lead to find an unclear rela-

tionship between children’s education and parents’ fertility. To check for this possibility we first

look at the association between parents’ fertility and children’s education described in Equation

1.4 in a panel framework without including province-fixed effects.

18


Table 1.3: Province level data: descriptive statistics(1) (2) (3) (4)

Mean Std. dev. Min MaxYear 1900Share of literate children (aged 5-15) 0.31 0.14 0.12 0.57Age-adjusted share of literate children (aged 5-15) 0.32 0.14 0.13 0.59Marital fertility (Ig) 0.67 0.09 0.46 1.01Total fertility rate (TFR) 4.91 0.67 3.19 6.17Share in agriculture, men (aged 20-60) 0.71 0.12 0.33 0.88Share in industry, all (aged 20-60) 0.07 0.04 0.03 0.21Share urban (>10000) 0.25 0.2 0 0.81Population density 45.38 30.9 14.59 143.79Infant mortality rate 206.53 40.08 110 290Child mortality rate 207.34 51.14 100 310Life expectancy at 15 43.1 1.47 40 48.33Permanent in-migration 0.07 0.07 0.01 0.42Temporary in-migration 2.85 1.54 0.3 6.6Local investments in education (per child, aged 5-15) 1.4 1.3 0.22 6.8Nuptiality 0.57 0.07 0.41 0.66

Year 1910Child mortality rate 153.26 38.75 90 230Infant mortality rate 163.47 30.72 90 210Life expectancy at 15 45.8 1.28 43.03 49.92

Year 1920Share of literate children (aged 5-15) 0.43 0.16 0.18 0.69Age-adjusted share of literate children (aged 5-15) 0.43 0.16 0.18 0.7Marital fertility (Ig) 0.61 0.09 0.38 0.78Total fertility rate (TFR) 4.06 0.71 2.59 5.25Share in agriculture, men (aged 20-60) 0.6 0.17 0.14 0.86Share in industry, all (aged 20-60) 0.08 0.05 0.02 0.34Share urban (>10000) 0.3 0.22 0 0.83Population density 52.47 40.13 14.7 189.43Infant mortality rate 162.65 32.58 90 230Child mortality rate 155.1 38.35 80 230Life expectancy at 15 45.92 1.52 41.77 50.63Permanent in-migration 0.08 0.07 0.01 0.4Temporary in-migration 2.12 1.21 0.2 5.2Local investments in education (per child, aged 5-15) 1.46 1.28 0.25 6.57Nuptiality 0.51 0.07 0.34 0.62

Data on 49 Spanish provinces. Temporary in-migration is not available for the Canary Islands.

19


Álava

Albacete

Alicante

Almería

Ávila Badajoz

Baleares

Barcelona

Burgos

Cáceres

Cádiz

Canarias


Ciudad RealCórdoba

La CoruñaCuenca

Gerona

Granada

Guadalajara

Guipúzcoa

Huelva

Huesca

Jaén

León

Lérida

Logroño

Lugo

Madrid

Málaga

Murcia

Navarra

Orense

OviedoPalencia

Pontevedra

Salamanca

Santander

SegoviaSevilla

Soria

Tarragona

Teruel

Toledo

Valencia

Valladolid

VizcayaZamora Zaragoza

−.4

−.3

−.2

−.1

0.1

Change in (

log)

marita

l fe

rtili

ty (

1900−

20)

0 .2 .4 .6 .8Change in (log) age−adjusted children’s education (1900−20)

Figure 1.4: Change in children’s literacy and fertility change (1900-1920)

Table 1.4 shows the results of this exercise using the age-adjusted measure of children’s educa-

tion.22 As estimation results point out, when not accounting for province specific characteristics

(i.e. estimating a pooled OLS regression), the relationship between parents’ fertility and chil-

dren’s education is positive and not significant, consistently with the above mentioned puzzle

(column 1). However, the association of children’s education with parents’ fertility turns nega-

tive and significant when introducing province-fixed effects (column 2). This seems to indicate

that once taken into account unobserved heterogeneity, the negative association between chil-

dren’s schooling and fertility is re-established. Also, it suggests that using a methodology more

in line with the concept of fertility transition, that is looking at changes rather than levels,

socio-economic variables as children’s education did matter.

The next step of the analysis assesses whether this relationship is robust to the inclusion of

several control variables. Column 3 adds the shares of adult men employed in agriculture and

of adult individuals employed in the industry sector while column 4 includes urbanization and22Using the simple (i.e. age-unadjusted) measure of children’s literacy, we obtain similar results.

20


population density. Column 5 adds child mortality and life expectancy at 15.23 Column 6

controls for permanent in-migration flows as heterogeneity in the support to the schooling reform

might have contributed to such phenomena and temporary in-migration to capture seasonal

migration flows.24 In column 7 and 8 we exploit the long-time variation between 1900 and 1920.

As estimation results point out the coefficient associated to children’s literacy is negative and

significant in all specifications. In particular when focusing on the long-time difference we find

a stronger association: this is in line with the fact that the effect of the increase in compulsory

schooling age in 1909 - together with the gradual innovation introduced by the educational reform

- in triggering fertility declines is likely to appear in the second decade of the 20th century.

1.5.2 Panel analysis: 2SLS

We deal with the potential bias of OLS estimates by instrumenting children’s literacy with

(per child) investments in education funded by local institutions.25 The latter should capture

the average local support to education expansion. Table 1.5 shows the first (columns 1-2) and

second stage (columns 3-4) estimates. We also use as measure of fertility the total fertility rate:

as the latter is an age-adjusted measure of total (including non-marital) fertility, we include an

index of nuptiality among the regressors. The instrument’s coefficient is positive and significantly

different from zero in all specifications and the F statistic is above 10. Second stage estimates

show that children’s literacy is negatively related to fertility independently of the measure of

fertility. As IV estimates are larger in size (in absolute value) if compared to the corresponding

OLS estimates in Table 1.4, the latter are likely to be biased downward due to reverse causality,

measurement error and omitted variables problems.

23In 1918-20 took place a temporary increase in mortality due to the influenza pandemic known as the Spanishflu. To fully account for this our analysis controls for child mortality - which shows an increase both at the countrylevel and in some provinces in the year 1920 compared to 1910 - and for adult mortality as well.

24Data on temporary in-migration from Silvestre (2007) is not available for the Canary Islands.25Our sample size ranges from 147 to 27 observations depending on different empirical models. Instrumental

variable estimates might suffer of bias due to this relatively small sample. However, rather than presenting onlyresults based on least squares estimates, we prefer to provide also IV estimates based on alternative instruments.If the latter satisfy the two requirements to be valid (i.e. being well correlated with the endogenous variable andfulfilling the exclusion restrictions), their use can be regarded as an added value to OLS estimates.

21


Table

1.4:Determ

inantsof

fertilitydeclines:

OLS

panelevidenceDependent

variable(Log)

Maritalfertility

t

(1)(2)

(3)(4)

(5)(6)

(7)(8)

(Log)Age-adjusted

0.0346-0.0915*

-0.0918*-0.0896*

-0.0837*-0.0903**

-0.1793**-0.1744***

children’seducationt

[0.0406][0.0489]

[0.0484][0.0482]

[0.0480][0.0442]

[0.0675][0.0637]

Shareinagriculture

t(m

en)0.0386

0.03730.0448

0.0190-0.0402

-0.0362[0.0603]

[0.0605][0.0658]

[0.0711][0.1128]

[0.1216]Sharein

industryt

0.22530.3597

0.34420.1800

0.1640-0.0027

[0.2297][0.2971]

[0.2966][0.2423]

[0.3837][0.3581]

Shareurbant

0.14410.1172

0.15220.0537

-0.0471[0.1927]

[0.2043][0.2217]

[0.3136][0.2940]

(Log)Population

densityt

-0.1731-0.1683

-0.0989-0.0509

0.0293[0.1497]

[0.1518][0.1240]

[0.1678][0.1702]

(Log)Child

mortality

ratet

0.02800.0288

0.0579[0.0359]

[0.0349][0.0509]

(Log)Infantmortality

ratet

0.1684**[0.0733]

Lifeexpectancyat15

t-0.0039

-0.0039-0.0064

-0.0047[0.0082]

[0.0081][0.0083]

[0.0070](Log)

In-migration

t-0.0328

0.01340.0122

[0.0536][0.0719]

[0.0594](Log)

Tem

poraryin-m

igrationt

-0.0060-0.0085

-0.0140[0.0082]

[0.0097][0.0085]

Constant

-0.3677***-0.5225***

-0.5663***0.0182

0.0291-0.3116

-0.4380-1.3445

[0.0589][0.0600]

[0.0756][0.4997]

[0.6751][0.5755]

[0.7654][0.8356]

Tim

edum

mies

yesyes

yesyes

yesyes

yesyes

Province-fixed

effectsno

yesyes

yesyes

yesyes

yesProvinces

4949

4949

4948

4848

Observations

147147

147147

147144

9696

***,**,*denote

statisticalsignificanceat

1%,5%

and10%

levels,respectively.The

dependentvariable

is(log)

maritalfertility

(Ig )

attim

et.

Children’s

educationis

theshare

ofchildren

aged5-15

thatcan

readand

write.

t=1900,1910,1920

exceptcolum

ns7-8

(t=1900,1920).

Robust

standarderrors

clusteredat

theprovince

levelreportedin

parentheses.

22


Tab

le1.5:

Determinan

tsof

fertility

declines:2S

LS.F

irst

andsecond

stageestimates

Depen

dent

variable

(Log)Age-adjustedchild

ren’seducation t

(Log

)Marital

(Log)

fertility

tTFR

t

(1)

(2)

(3)

(4)

First

stage

Second

stag

e

(Log)A

ge-adjustedchild

ren’se

ducation

t-0.401

8**

-0.343

1**

[0.154

9][0.1310]

Localsup

portto

education t

0.2747***

0.2742***

[0.0736]

[0.0733]

Shareinagriculture t

(men)

0.0625

0.0631

0.05

16-0.064

8[0.1461]

[0.1480]

[0.096

6][0.082

1]Sh

areinindu

stry

t-0.4746

-0.4768

0.01

07-0.175

9[0.4198]

[0.4255]

[0.324

5][0.333

4]Sh

areu

rban

t-0.2049

-0.2065

0.10

83-0.028

6[0.8350]

[0.8385]

[0.383

0][0.348

5](L

og)C

hildmortalityrate

t-0.0732

-0.0729

-0.011

30.01

28[0.1175]

[0.1185]

[0.063

3][0.061

0]Lifeexpe

ctan

cyat

15t

0.0001

0.0000

-0.003

1-0.014

4*[0.0134]

[0.0134]

[0.008

1][0.008

3](L

og)P

opulationdensity t

0.5371

0.5459

-0.033

10.01

75[0.3226]

[0.3323]

[0.150

8][0.174

7](L

og)In-m

igration

t0.2251*

0.2279*

0.04

510.05

33[0.1249]

[0.1206]

[0.064

8][0.064

0](L

og)Tem

porary

in-m

igration

t0.0116

0.0114

-0.004

0-0.012

3*[0.0172]

[0.0175]

[0.007

9][0.0066]

Nup

tiality t

-0.0950

1.24

90**

*[0.5946]

[0.454

0]

F-statistic

1ststage

13.94

13.98

Tim

edu

mmies

yes

yes

yes

yes

Province-fix

edeff

ects

yes

yes

yes

yes

Provinces

4848

4848

Observation

s144

144

144

144

***,

**,*

deno

testatisticals

ignifican

ceat

1%,5

%an

d10%

levels,r

espe

ctively.

Children’seducationis

theshareof

child

renag

ed5-15

that

canread

andwrite.Lo

cals

uppo

rtto

educationis

investments

ineducation(per

child

)fund

edby

localinstitution

s.t=

1900

,191

0,19

20.

Rob

uststan

dard

errors

clusteredat

theprovince

levelr

eportedin

parenthe

ses.

23


Álava

AlbaceteAlicante

Almería Ávila

Badajoz

Baleares

Barcelona

Burgos

CáceresCádiz

Canarias


Ciudad Real

Córdoba

La Coruña

CuencaGerona

Granada

GuadalajaraGuipúzcoa

Huelva

Huesca

Jaén

LeónLérida

Logroño

Lugo

Madrid

Málaga

Murcia

Navarra

Orense

Oviedo

Palencia

Pontevedra

Salamanca

Santander

Segovia

Sevilla

Soria

Tarragona

Teruel

Toledo

Valencia

Valladolid

Vizcaya

Zamora

Zaragoza

0.2

.4.6

.8C

hange in (

log)

age−

adj child

ren’s

education (

1900−

20)

0 .34Share of medium−large livestock owners in 1865

Figure 1.5: Livestock ownership inequality and change in children’s education

1.5.3 Long-time differences: 2SLS, 3SLS and robustness checks

Moving to the long-time differences set-up described above, we now estimate Equation 1.5 using

three alternative IV strategies. First, we use as only instrument for changes in children’s literacy

the share of medium-large livestock owners in 1865. Figure 1.5 points out a negative relationship

between this measure and changes in children’s education in the period 1900-20. Second, we use

both livestock ownership concentration and changes in (per child) local investments in education

as instruments for changes in children’s literacy. This allows to investigate the validity of the

exclusion restrictions. Third, since our reasoning is that provinces characterized by a substantial

share of medium-large livestock owners would support to a lower extent education expansion,

we use a 3SLS procedure where the share of medium-large livestock owners in 1865 is used

as instrument for the change in local support to education in the period 1900-20. The latter

procedure is more in line with our reasoning outlined in Section 1.4.2, as livestock ownership

concentration represents a plausibly exogenous source of variation in the local support to human

capital accumulation.

24


Table 1.6 shows the first (columns 1-3) and second stage (columns 4-6) estimates of the 2SLS

regressions. We notice that the share of medium-large livestock owners in 1865 is negatively

correlated to changes in children’s education between 1900 and 1920 both when used as unique

instrument and when used together to the change in locally funded (per child) investments in

education (the latter being positively related to changes in the endogenous variable). When

using both instruments we notice from the first stage estimates that there is some collinearity

between them, leading to lower first stage F statistics: this is consistent with the reasoning

that our measure of livestock ownership concentration explains (part of) the local support to

education expansion. Looking at second stage estimates, the negative impact of changes in

children’s education on changes in parents’ fertility is significant in all cases, thus confirming our

previous findings.26 Also, the Hansen J test suggests that we cannot reject the null hypothesis

that the instruments satisfy the exclusion restrictions, thus providing some evidence in favour

of our reasoning(s) about their validity. As in the panel estimation, 2SLS estimates are larger

in size (in absolute value) with respect to OLS ones, suggesting the latter are biased downward.

As we show in the section 1.5.4 a spatially lagged dependent variable is likely to be one of the

omitted variables that, being positively related to changes in fertility, would cause such bias of

OLS estimates.

Table 1.7 shows the results from the 3SLS procedure including several robustness checks.27 We

first notice that - in line with our reasoning - the share of medium-large livestock owners in

1865 is negatively related to changes in the support to education expansion (first-stage), the

latter being positively related to changes in children’s literacy (second-stage). While column 1

replicates the results obtained using the 2SLS procedure, columns 2-6 present evidence to rule

out the possibility that our results are driven by the Spanish flu. We start by dropping provinces

most affected by this phenomenon: specifically, in columns 2 and 3 we drop provinces where child

26As shown in columns 7-8, this result holds when replacing the child mortality rate with the infant mortalityrate.

27First and second stage panels display only the estimates of the variables of interest.

25


mortality increased from 1910 to 1920. We observe that - despite the smaller sample size - the

relationship between changes in children’s education and fertility is still negative and significant.

Columns 4 and 5 replicate the analysis using the TFR as measure of fertility. Finally, we use

as dependent variable the change in (log) crude birth rate (CBR) between 1900 and 1917, that

is the year before the influenza pandemic (column 6): this measure of fertility change is then

completely unaffected by the Spanish flu.28 As we notice the estimation results confirm the

negative relationship previously pointed out, meaning that our results are not driven by the

influenza pandemic.

1.5.4 Long-time differences: spatial diffusion

The diffusion of new social and cultural norms has been considered one of the drivers of fertility

transitions, especially after the conclusions reached by the EFP.29 This has also been suggested

for the case of Spain: Reher and Iriso-Napal (1989) argue that a process of diffusion of new ideas

favourable to birth control was actually taking place across Spanish provinces at the beginning

of the 20th century. As it can be noticed by looking at Figure 1.6, there appears to be some

spatial correlation in fertility changes between 1900 and 1920, with larger declines taking place

along the east coast and in some north-western provinces. However, when looking at the spatial

distribution of changes in children’s literacy, a similar pattern stands out (Figure 1.7). Provinces

that improved most in children’s literacy rates are located in the eastern and north-western

regions of Spain. This might suggest that also the spread of new attitudes towards child quality

was taking place.

A first assessment of the degree of spatial autocorrelation in the decline of fertility across Spanish

provinces can be obtained by looking at the Moran’s I.30

28Among the additional control variables, the share of women is defined as women aged 16-45 over populationwhile the share of married women is computed as married women aged 16-45 over women of the same age.

29Other studies dealing with the role of spatial diffusion are Tolnay (1995), Goldstein (2010) et al. and Murphy(2010).

30Moran’s I is a measure of spatial autocorrelation characterizing the relationship of the values of a variablewith the geographical location where they were measured.

26


Tab

le1.6:

Determinan

tsof

fertility

declines:long

-tim

ediffe

rences.2S

LSDependent

∆(L

og)Age-adjusted

∆(L

og)Marital

∆(L

og)

variable

Children’sed

ucation

fertility

(Ig)

TFR

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

First

stage

Second

stag

e

∆(L

og)C

hildren’se

ducation

-0.345

8***

-0.373

2***

-0.3123*

**-0.331

1***

(age-adjusted)

[0.103

6][0.097

6][0.091

6][0.105

7]Sh

areo

fmed

ium-la

rge

-1.822

6***

-1.509

2***

-1.695

9***

-1.642

5***

livestock

owne

rs[0.418

7][0.417

2][0.474

9][0.473

1]∆Localsupp

orttoed

ucation

0.26

85**

*0.27

36**

*0.28

37**

*[0.094

1][0.087

3][0.091

0]∆Sh

areinag

riculture(

men

)-0.503

9-0.582

5-0.620

4-0.697

2-0.057

5-0.060

4-0.050

3-0.066

2[0.503

3][0.508

5][0.509

7][0.512

3][0.116

4][0.124

0][0.110

3][0.129

8]∆Sh

areinindu

stry

-2.929

4**

-2.928

7**

-3.181

4**

-3.846

7***

-0.174

4-0.230

2-0.259

9-0.863

8[1.197

1][1.085

3][1.176

2][1.202

0][0.432

3][0.458

4][0.360

8][0.605

2]∆Sh

areu

rban

-0.456

0-0.441

4-0.599

3-0.810

10.01

370.00

71-0.078

9-0.318

3[0.775

8][0.754

4][0.744

0][0.718

4][0.345

0][0.356

4][0.308

0][0.293

6]∆(L

og)C

hild

mortalityrate

-0.116

9-0.042

50.02

120.01

51[0.123

6][0.132

3][0.059

3][0.060

1]∆(L

og)Infan

tmortalityrate

0.14

520.20

590.14

48**

0.14

32*

[0.174

9][0.167

6][0.070

2][0.073

5]∆Life

expe

ctan

cyat

15-0.0001

0.00

140.00

630.01

13-0.006

4-0.006

5-0.004

1-0.007

5[0.013

5][0.013

4][0.012

4][0.012

1][0.007

5][0.007

6][0.006

2][0.006

3]∆(L

og)P

opulationde

nsity

1.14

79**

*1.32

27**

*1.51

22**

*1.81

43**

*0.00

170.01

030.06

900.29

17[0.364

8][0.355

7][0.411

7][0.393

9][0.166

3][0.169

5][0.160

8][0.201

8]∆(L

og)In-m

igration

0.15

290.10

110.09

290.13

890.0569

0.0641

0.04

940.08

66[0.125

9][0.124

7][0.127

8][0.130

3][0.066

8][0.065

5][0.058

8][0.068

4]∆(L

og)Tem

p.in-m

igration

-0.003

40.01

030.00

440.00

50-0.0051

-0.004

5-0.010

1-0.014

4*[0.020

9][0.022

9][0.023

1][0.023

0][0.009

1][0.009

2][0.008

3][0.008

7]∆Nup

tiality

-1.640

9**

0.52

71[0.747

0][0.538

4]Con

stan

t0.40

41**

*0.33

52**

*0.38

05***

0.25

33**

0.02

650.03

260.03

67-0.024

7[0.079

4][0.076

2][0.084

4][0.106

0][0.037

5][0.038

0][0.032

7][0.037

9]F-statistic

1ststag

e18

.95

13.64

16.00

15.06

Han

senJtest

(p-value

)0.51

0.2

0.51

Observation

s48

4848

4848

4848

48

***,

**,*

deno

testatisticals

ignifican

ceat

1%,5

%an

d10

%levels,r

espe

ctively.

Allvariab

lesareexpressedas

chan

gesbe

tween19

00an

d19

20.Children’s

educationis

theshareof

child

renag

ed5-15

that

canread

andwrite.Rob

uststan

dard

errors

repo

rted

inpa

renthe

ses.

27


Table 1.7: Determinants of fertility declines: long-time differences. 3SLS and robustness checksDependent variable ∆ (Log)Ig ∆ (Log)TFR ∆ (Log)

CBR1900−17

(1) (2) (3) (4) (5) (6)No No No No

∆1910−20 ∆1910−20 ∆1910−20 ∆1910−20

cmr>0 cmr>0 cmr>0 cmr>0Panel A: Three-stage least squares

∆(Log)Children’s education -0.3458*** -0.3860*** -0.3812*** -0.4298*** -0.3946*** -0.6153***(age-adjusted) [0.1135] [0.1212] [0.1056] [0.1238] [0.0985] [0.1821]∆Share in agriculture (men) -0.0575 -0.1033 -0.1554 -0.1436 -0.1899 -0.3047

[0.1497] [0.1443] [0.1445] [0.1451] [0.1384] [0.2139]∆Share in industry -0.1744 0.0808 -0.1713 -0.3393 -0.6811 -1.2414

[0.5149] [0.4724] [0.4987] [0.5055] [0.5141] [0.7727]∆Share urban 0.0137 -0.6308 -0.6015 -1.0952** -1.1306** -0.0870

[0.3228] [0.5195] [0.5160] [0.5294] [0.5043] [0.4499]∆(Log)Childmortality rate 0.0212 0.1566** 0.1149 -0.0019

[0.0645] [0.0769] [0.0796] [0.0912]∆(Log) Infantmortality rate 0.1863** 0.1822**

[0.0824] [0.0797]∆Life expectancy at 15 -0.0064 -0.0076 -0.0029 -0.0092 -0.0015 -0.0063

[0.0090] [0.0119] [0.0123] [0.0124] [0.0124] [0.0131]∆(Log)Population density 0.0017 0.3505 0.4543* 0.5967** 0.7602*** 0.4870*

[0.1698] [0.2551] [0.2681] [0.2738] [0.2789] [0.2503]∆(Log) In-migration 0.0569 0.1006 0.0625 0.1061 0.0747 0.0294

[0.0651] [0.0733] [0.0724] [0.0783] [0.0721] [0.0911]∆(Log) Temp. in-migration -0.0051 0.0282** 0.0074 0.0165 -0.0035 0.0014

[0.0102] [0.0118] [0.0132] [0.0119] [0.0126] [0.0145]∆Nuptiality 1.2610** 0.9733**

[0.5303] [0.4891]∆Sharewomen 4.9653

[3.2793]∆Sharemarriedwomen 0.4908

[0.8049]

Panel B: Second stage for ∆(Log)Age-adjusted children’s education

∆Local support 1.5610** 1.4072** 1.0604*** 1.3856** 1.1173*** 1.4704**to education [0.6346] [0.6004] [0.3487] [0.5883] [0.3618] [0.6341]

Panel C: First stage for ∆Local support to education

Share ofmedium-large -1.1676** -1.6188** -2.4533*** -1.6220** -2.4086*** -1.1023**livestock owners [0.4974] [0.7791] [0.8534] [0.7806] [0.8568] [0.5033]Observations 48 27 27 27 27 48

***, **,* denote statistical significance at 1%, 5% and 10% levels, respectively. All variables are expressed aschanges between 1900 and 1920 except the dependent variable in column 5.

28


(−.023,.048](−.061,−.023](−.08,−.061](−.115,−.08](−.181,−.115][−.427,−.181]

Change in (log) marital fertility (1900−1920)

Figure 1.6: Change in (log) marital fertility (1910-1930)

(.48,.76](.38,.48](.27,.38](.22,.27](.17,.22][−.01,.17]

Change in (log) age−adjusted children’s education (1900−1920)

Figure 1.7: Change in (log) age-adjusted children’s literacy, aged 5-15 (1900-1920)

29


Moran scatterplot (Moran’s I = 0.069)dlnifm

Wz

z−4 −3 −2 −1 0 1 2

−1

0

1

Figure 1.8: Moran’s I: marital fertility

Moran scatterplot (Moran’s I = 0.078)dlnisf

Wz

z−4 −3 −2 −1 0 1 2

−1

0

1

Figure 1.9: Moran’s I: TFR

Figures 1.8 and 1.9 show the Moran scatterplot of the relationship between the change in (log)

marital fertility and (log) TFR between 1900 and 1920 respectively and their corresponding spa-

tially lagged component. As it can be noticed the majority of observations are placed in the first

and third quadrants, suggesting the existence of (positive) spatial autocorrelation (i.e. provinces

characterized by larger declines in fertility surrounded by provinces with a similar pattern, and

similarly for provinces with smaller declines). We account for the role of spatial dependence in

explaining fertility declines by estimating spatial lag and error models via maximum likelihood

estimator (MLE) following Anselin (1988). The spatial lag model is defined as follows:

∆ferti = ρW∆fertj + α1∆educi + α2∆Xi + const+ εi (1.6)

where j 6= i, W is the spatial weight matrix and W∆ferti is the spatially lagged dependent

variable.31

Instead the spatial error model includes a spatial component in the error term:

31The inverse distance spatial weights matrix is computed using latitude and longitude of the capital city ofeach province. This choice is consistent with the fact that new behavioural and cultural norms tend to spreadfirst in the urban environment and then to diffuse also in rural areas.

30


∆ferti = α1∆educi + α2∆Xi + const+ εi where εi = λWεi + ψi (1.7)

where Wεi is the spatially lagged error term.

Table 1.8 shows the results of estimating Equations 1.6 and 1.7 by MLE.32 Even if there is some

evidence suggesting that a diffusion process might have been in place (positive and significant

rho), we still observe a negative relationship between changes in children’s education and changes

in fertility.

1.6 Conclusion

This paper finds evidence supporting a negative association between quality and quantity of chil-

dren using provincial level data for early 20th century Spain. In the period under consideration,

Spain witnessed the beginning of a process of transformation of the educational system that took

place with large heterogeneity across the country. It is then reasonable to think that part of the

changes in schooling across provinces were due to the different local support that citizens gave to

this process. Taking advantage of this fact this study establishes a causal effect linking children’s

education and parents’ fertility by exploiting instrumental variable strategies that use direct and

indirect measures of local support to education expansion. Specifically, evidence points out that

increases in children’s literacy, which proxies for higher primary school attendance and comple-

tion, are related to declines in fertility. Our result indicates that, by focusing on changes rather

than on levels as the concept of fertility transition suggests and by tackling the endogeneity is-

sues related to the quantity-quality trade-off, there is evidence of a negative association between

children’s education and parents’ fertility across Spanish provinces at the beginning of the 20th

century. This adds evidence in favour of theories arguing that increases in children’s quality are

important in explaining fertility declines.

32Estimation results using infant mortality instead of child mortality (not reported) are qualitatively similar.

31


Table

1.8:Determ

inantsof

fertilitydeclines:

long-timedifferences.

Spatiallagand

errormodels

(MLE

)Dependent

variable∆(L

og)∆(L

og)∆(L

og)∆(L

og)∆(L

og)∆(L

og)∆(L

og)∆(L

og)Ig

TFR

Ig

TFR

Ig

TFR

Ig

TFR

(1)(2)

(3)(4)

(5)(6)

(7)(8)

Model

SpatiallagSpatialerror

SpatiallagSpatialerror

∆(L

og)Children’seducation

-0.1573**-0.1965***

-0.1422*-0.1859**

-0.1519**-0.1761***

-0.1687**-0.1957***

(age-adjusted)[0.0697]

[0.0736][0.0832]

[0.0861][0.0653]

[0.0672][0.0773]

[0.0744]

∆Sharein

agriculture(men)

0.16850.1393

0.16660.1409

-0.0279-0.0195

-0.0381-0.0315

[0.1875][0.1736]

[0.1609][0.1662]

[0.1079][0.1165]

[0.1385][0.1420]

∆Sharein

industry0.5741

-0.06910.6226

-0.00180.2599

-0.17130.2076

-0.2387[0.4751]

[0.4190][0.5261]

[0.6044][0.3623]

[0.4058][0.4802]

[0.5483]∆Shareurban

-0.1625-0.3501

-0.1888-0.3453

-0.0065-0.1634

0.0269-0.1382

[0.3692][0.3591]

[0.3751][0.3730]

[0.3013][0.2824]

[0.3255][0.3067]

∆(L

og)Population

density-0.0540

0.1465-0.0602

0.1309-0.0378

0.0846-0.0459

0.0893[0.1698]

[0.1898][0.1947]

[0.2220][0.1569]

[0.1825][0.1581]

[0.1804]∆(L

og)Child

mortality

rate0.0233

0.01270.0196

0.01300.0540

0.03850.0534

0.0406[0.0677]

[0.0658][0.0680]

[0.0681][0.0468]

[0.0446][0.0602]

[0.0586]∆Lifeexpectancy

at15-0.0060

-0.0096-0.0051

-0.0086-0.0063

-0.0112-0.0062

-0.0111[0.0088]

[0.0094][0.0103]

[0.0104][0.0076]

[0.0073][0.0084]

[0.0086]∆(L

og)In-m

igration0.1074

0.14640.1068*

0.1408**0.0190

0.05000.0142

0.0397[0.0876]

[0.0915][0.0629]

[0.0652][0.0665]

[0.0652][0.0556]

[0.0584]∆Nuptiality

0.58380.6037

0.9971**0.9826**

[0.5356][0.5626]

[0.4472][0.4677]

∆(L

og)Tem

poraryin-m

igration-0.0089

-0.0130-0.0092

-0.0135[0.0089]

[0.0084][0.0098]

[0.0095]Constant

0.03470.0452

-0.0362-0.0839

0.02930.0622

-0.0165-0.0411

[0.0447][0.0728]

[0.0578][0.0660]

[0.0430][0.0721]

[0.0447][0.0518]

ρ0.6090*

0.6367*0.5645

0.5856λ

0.56150.5168

0.20380.1479

Observations

4949

4949

4848

4848

***,**,*denote

statisticalsignificanceat

1%,5%

and10%

levels,respectively.Allvariables

areexpressed

aschanges

between

1900and

1920.Children’s

educationis

theshare

ofchildren

aged5-15

thatcan

readand

write.

Robust

standarderrors

reportedin

parentheses.

32

Chapter 2

Quantity affects quality: fertility,

education, and gender in 1887 Spain

2.1 Introduction

This paper aims at testing the effect of quantity of (surviving) children on quality of children

in historical context. Using district level data for Spain in the year 1887, it provides evidence

suggesting that parents’ fertility had a significant negative effect on boys’ literacy rates while

evidence is weaker and less robust for girls. On the one hand the significant impact of quantity of

children on boys’ education confirms the existence of a quantity-quality trade-off in the historical

context of late 19th century Spain. Hence, this suggests that in larger families male children

were more likely to be unschooled. On the other hand, this result is consistent with the existence

of cultural heterogeneity across Spain regarding the role of women in society, suggesting that

parents’ fertility - through a budgetary and economic mechanism - was not among the main

determinants of girls’ schooling attainments. For example, if women were not expected to work

outside the house or social norms were such that they were not supposed to attend school, the

number of their siblings would not have affected significantly their school attendance.

The formalization of the theory of the demand for children and of the quantity-quality trade-

2 Quantity affects quality: fertility, education, and gender in 1887 Spain

off as an economic mechanism date back to Becker’s theory of the family (e.g. Becker and

Lewis 1973; Becker 1981). More recently the interplay of quality and quantity of children has

gained attention from the literature analysing the determinants of fertility choice and fertility

transitions (see Guinnane 2011; Galor 2012). The development of unified growth theories suggest

that increasing returns to education - due to technological progress - would lead parents’ to invest

more in the quality of their offspring, thus reducing their quantity to keep their budget balanced

(Galor and Weil 2000). The central mechanism of unified growth theories entails a change in the

relative price of quality with respect to quantity of children. A change in the price of quality

- as due to higher returns to schooling - would lead parents to react by changing their optimal

number of children. In this paper we do not measure (or exploit) changes in returns to schooling

or in relative prices.1 Our paper is a simple test of the existence of a child quantity-quality

trade-off in late 19th century Spain.

Most of the studies investigating the effect of family size (i.e. quantity) on quality of children

uses modern data (e.g. Angrist et al. 2005; Black et al. 2005). This paper closely relates to

the literature that recently focused on analysing the quantity-quality trade-off in a historical

context. Within this literature, Bleakley and Lange (2009) explore the causal effect of education

on fertility by exploiting the eradication policy of the hookworm disease in Southern states

of North America in 1910. Their study argues that this eradication increased the return to

schooling and hence reduced the price of child quality, thus increasing school attendance and

reducing fertility. Becker et al. (2010, 2012) use data on Prussian counties in 1849 and find

evidence of the quantity-quality trade-off in a context in which the demographic transition has

not yet taken place.2 Becker et al. (2010) show that the negative correlation between quantity

and quality of children can be given a causal interpretation. One of their instrumental variable

strategies employs sex ratios in the adult population to instrument fertility levels. Klemp and

Weisdorf (2011) find a significant negative impact of family size on children’s literacy using data

1Bleakley and Lange (2009) claim their are able to measure a change in relative prices of children’s qualityand quantity.

2Our paper is closely related to Becker et al. (2010) as we study the existence of a quantity-quality trade-off in1887, that is before the onset of the demographic transition in Spain. The onset of the transition at the countrylevel is dated 1910 (Reher 2004).

34


from Anglican parish registers in England over the period 1700-1830. One of the novelties of

their approach is their instrumental variable strategy. Fertility is instrumented using marital

fecundability, measured by the time interval elapsed from marriage to the first birth. Finally,

Fernihough (2011) finds evidence of the quantity-quality trade-off using census data for Ireland

in 1911: using data for Belfast and Dublin, he finds that higher fertility (measured by the number

of siblings) reduced the probability of school attendance.

In 1857 the Ley Moyano established in Spain compulsory schooling attendance for children aged

between 6 and 9 years.3 As it can be noticed from Table 2.1 children’s literacy - defined as the

share of children aged 5-15 able to read and write - was around 24% in 1887. Gender differences

were also evident as boys’ literacy (29%) was higher than girls’ (19%). Hence, even if school

attendance became compulsory for children aged 6-9 in 1857, in practice children’s literacy was

still relatively low in 1887. Quality of children is usually measured by educational achievements.

Population census data used in this analysis provide information on literacy rates at the judicial

district level. Consequently the measure of quality of children used throughout the paper is

children’s literacy, measured for children aged 5-15.4 This appears to be a good proxy both for

primary school attendance and completion, since literacy is one of the main outputs of primary

school and compulsory schooling in 1887 was restricted to children aged 6-9. A quantity-quality

trade-off exists if, on average, when the number of children is high, quality of children is low

since fewer resources are available for each of them. While there are no specific cultural reasons

why the quantity-quality trade-off should not apply for boys, there are good reasons to believe

this might be the case for girls. In particular, girls might be unschooled not because they have

many siblings but because cultural and social norms are not favourable for them to get educated.

One can expect that in a period in which female education is driven by several factors that go

beyond a pure budgetary mechanism, evidence of the quantity-quality trade-off will be stronger

if considering only boys’ education as indicator of children’s quality.5

3The age limit for compulsory schooling was raised to 12 years in 1909.4Using literacy for children aged 9-15 does not change the qualitative results of the analysis.5This is also in light of the fact that when the analytical framework is a simple cross-section, controlling for

heterogeneity in cultural and social norms is a difficult task.

35


Table 2.1: Descriptive statistics(1) (2) (3) (4)

Mean Std. dev. Min Max

District level data

Child-woman ratio 1 0.67 0.09 0.34 0.89

Child-woman ratio 2 1.02 0.11 0.46 1.34

Children’s literacy (aged 5-15) 0.24 0.13 0.02 0.59

Girls’ literacy, (aged 5-15) 0.19 0.11 0.01 0.55

Boys’ literacy (aged 5-15) 0.29 0.15 0.04 0.67

Share in agriculture, men (aged 21-40) 0.72 0.15 0.07 0.93

Share urban 0.1 0.24 0.00 1

Temporary male migration 0.00 0.21 -0.3 4.37

Share in industry, all (aged 21-40) 0.01 0.04 0.00 0.4

Adult literacy (aged 21-50) 0.35 0.13 0.11 0.7

Women-to-men ratio (aged 21-50) 1.09 0.23 0.61 2.42

Population 37085.34 33322.55 7410 470283

Data for 473 judicial districts collected from Spanish population census in 1887.

36


The data used in the empirical analysis come from the population census of 1887 which pro-

vides data at the judicial district (partido judicial) level. These units, much smaller in size than

provinces, allow for a relative large cross-sectional sample of more than 400 observations.6 Using

these district-level data, this paper studies the effect of quantity on quality of children allowing

for gender-specific measures of quality. To capture the number of children, we use two alter-

native measures of fertility. One is the child-woman ratio computed as the number of children

aged 0-5 over the number of women aged 16-45 (labelled Child-woman ratio 1 ). The other is

the child-woman ratio computed as the number of children aged 6-15 over the number of women

aged 21-50 (labelled Child-woman ratio 2 ). As mentioned above, quality of children is proxied

by the share of literate (i.e. able to read and write) children aged 5-15. Three measures are

considered: one that includes both boys and girls, one that includes only girls, and another that

considers only male children.

Potential bias of OLS estimates is tackled using an instrumental variable strategy that employs

women-to-men ratios (WMRs hereafter) in the adult population to instrument fertility levels.

WMRs in the adult population identify exogenous variation in parental fertility: it seems reason-

able to assume that they affect children’s education only through their effect on parents’ fertility

behaviour. This strategy is employed in Becker et al. (2010) when studying the effect of parents’

fertility on children’s education across Prussian counties in the 19th century. We cannot use

other - probably more plausible - instruments that have been used in the literature (e.g. twins,

length of time to first birth) as they are not available.

Evidence of a negative effect of parents’ fertility on boys’ education is strong, while the rela-

tionship is weaker when considering girls’ education. The result is robust to various robustness

checks, including the presence of spatial dependence. One the one hand, this study contributes to

the literature by widening the historical empirical evidence on the quantity-quality trade-off to a

Southern European country. On the other hand it suggests that when studying the existence of

a quantity-quality trade-off, especially in a historical and cultural context in which heterogeneity

regarding gender roles is present, distinguishing children’s quality by gender might be important.

6This allows to tackle one of the main criticisms of previous studies analysing, for example, fertility behaviour:the use of statistical units of analysis that are too aggregated (Brown and Guinnane 2007).

37


Therefore, conclusions drawn on the base of a measure of quality that does not distinguish female

and male children’s education could be mistaken.

The paper is structured as follows. Section 2.2 describes the data and the empirical strategy.

Section 2.3 shows the results of the analysis. Finally, Section 2.4 concludes.

2.2 Data and empirical strategy

2.2.1 Data

This paper studies the relationship between quantity and quality of children in historical Spain

using district level data. Data is taken from a population census carried out in 1887.7 One of the

advantages that the 1887 census provides is that several data is available not only at the provin-

cial level (49 observations) but also at a much more disaggregated level, that is district-level.8

To capture fertility we use two measures: one is the child-woman ratio computed as the number

of children aged 0-5 over the number of women aged 16-45 (labelled Child-woman ratio 1 ). The

other is the child-woman ratio computed as the number of children aged 6-15 over the number

of women aged 21-50 (labelled Child-woman ratio 2 ). The latter measure includes children aged

6-15 in order to remove the impact that child mortality rates might have on the number of

surviving children. Assuming that women did not have normally children before being 15 years

old, we consider women aged 21-50 to match the age lower bound for children (6).9 The child-

woman ratio is to our knowledge the best indicator of fertility we can obtain using district-level

data. The use of disaggregated data (such as district-level data in this paper) entails a trade-off

between sample size and the quality of some indicators. We assume this is a good indicator since

other studies dealing with the same issue have recently used it to proxy fertility (Becker et al.

2010; Becker et al. 2012). To capture quality of children we use the share of children (aged 5-15)

7The population census is available at www.ine.es.8To give an idea about their size, the districts were on average populated by 37085 individuals with a min-max

of 7410-470283 inhabitants. The maximum value corresponds to the Madrid district.9Using the age range 16-50 for women to compute the child-woman ratio does not affect the results of our

analysis.

38


that can read and write, which is also disaggregated by gender. Literacy is used as proxy for

basic schooling attainment as it is the main output of primary school: a literate child is expected

to attend or have attended primary school.

The analysis controls for several possible determinants of children’s educational levels. We mea-

sure the stage of development of the district by using a measure of the dependence on agriculture

(measured by the share of men aged 21-40 working in the primary sector). The development of the

industrial sector is proxied by the share of adult men and women aged 21-40 that work in industry

where industry includes manufacturing, mines and related industries ("industrias derivadas").10

Table 2.1 displays some descriptive statistics that characterize the sample in the year 1887. We

notice that due to the definition of industry some districts (precisely 8) are characterized by

0 shares. Very low shares of industrial employment identify extremely rural environments: of

course this depends to some extent on the definition of the industry sector, but it is in line with

the low industrial development that characterized Spain in 1887: the average share is 0.01 and

also the maximum figure (0.4) is not particularly high. As a proxy of urban environment we use

the share of population living in urban areas defined as the fraction of individuals living in towns

with more than 20000 inhabitants and in the capital city of each province. In addition we use a

dummy variable that takes value one for districts where the capital of each province is located.

This would control for the role of administrative and public jobs opportunities on stimulating

the demand for human capital (at least in terms of literacy and numeracy) and consequently

school attendance. To capture cultural and social norms that affected both parents’ fertility and

children’s education, we use province dummies aiming at capturing within-provinces cultural and

historical similarities: the rationale for this is that individuals living in districts belonging to the

same province are more likely to share common cultural, social and historical characteristics. In

our instrumental variable strategy, we include a measure of temporary men’s migration defined

as the difference between married males and married females, divided by married females. This

measure serves to control for possible migratory phenomena that might happen between districts

because of sex imbalances: Becker et al. (2010) use it for this purpose.

10The disaggregation available regarding occupations does not allow for other categorizations. Transportationis not included in the industry.

39


2.2.2 Empirical strategy

The relationship between children’s education and quantity of children is characterized as follows:

educi = γ1 ferti + γ2Xi + ψi (2.1)

where educi is the share of literate children (boys and/or girls) aged 5-15 in judicial district i,

ferti is the child-woman ratio in judicial district i and Xi includes district-level control variables

In order to tackle endogeneity problems due to omitted variables, measurement error and si-

multaneity, an instrumental variable approach is employed. As stated before, fertility levels are

instrumented with the WMR in the adult (i.e. aged 21-50) population.11 The latter should cap-

ture the tightness of the marriage market and consequently the likelihood of couple formation

which is one of the most important factors in determining fertility behaviour, as suggested by

Becker et al. (2010) arguing that (p. 187): "A lower sex ratio (less men than women) establishes

a constraint on the number of children that a typical household may have, e.g., because it leads to

later marriage or decreases the marriage rate of women, pushing fertility rates down... Econo-

metrically, the identifying assumption for this instrument is that the sex ratio is exogenously

determined by differential birth and death rates and that it affects fertility behavior only through

its influence on the probability of finding a mate, but is otherwise unrelated to education." As the

authors point out, migration flows could affect the WMR and being also related to educational

outcomes. We use the same indicator of migration they propose to control for this fact. In

addition the inclusion of province dummies - to control for common cultural characteristics -

should partially account for cultural factors that affect both the WMR and children’s education.

11We use as instrument the log of the WMR to reduce the effect of potential outliers. As we notice from thedescriptive statistics, the WMR varies from 0.61 to 2.42. Five districts take on a value above 2, but these outliersdo not affect significantly the average picture (1.09) that is similar to the one observed, for example, in 1849Prussia where the women-to-men ratio (age 15-45) was 1.01 (Becker et al. 2010).

40


Figure 2.1: Boys’ literacy in 1887: larger dots stand for higher literacy rates.

2.3 Quantity and quality of children: results

We estimate the association between fertility and children’s literacy controlling for the set of

controls mentioned in the previous section. In addition we include also latitude and longitude

of each district as there appears to be a North-South divide in children’s literacy (see Figure

2.1).12 Also, we include a measure of adult literacy to account for persistence in educational

choice. We limit its introduction to OLS regressions as it might have a counter-intuitive effect

leading to a difficult interpretation of our estimates. This because after the introduction in 1857

of compulsory schooling for children aged 6-9, there were no significant educational or structural

reforms before 1900. Hence, we can assume that the incentives to get educate were in practice

constant in the second half of the 19th century. Introducing adult literacy - that is the result

of past children’s literacy - might result in the impossibility of identifying the factors explaining

educational levels: as we notice from Table 2.2, the high coefficient associated to adult literacy

12In addition, latitude and longitude should partially capture differences in mortality risk, including childhoodmortality.

41


suggests this might be the case.13 Table 2.2 displays OLS estimates of Equation 2.1 using the

first measure of fertility (labelled child-woman ratio 1 ). Overall the association of quantity of

children and literacy is negative but not always significant; also it is stronger when the boys-

specific indicator of quality is used. The inclusion of province dummies seems appropriate as their

coefficients are jointly significantly different from zero in all specifications. We obtain similar

results, but better in terms of significance, using the child-woman ratio 2 as dependent variable

(Table 2.3). The coefficients are always significantly different from zero: a possible explanation

is that the child-woman ratio 2 partially accounts for child mortality. In fact, by reducing

the number of surviving children, higher child mortality can be positively related to children’s

literacy. However causal interpretation cannot be drawn using OLS estimates, which might be

biased for several reasons mentioned above. To address this issue, fertility is instrumented with

the (log of the) women-to-men ratio (WMR), a measure of the tightness of the marriage market.

We start our analysis by estimating the reduced form of this empirical model, that is we look at

the association between the WMR and children’s literacy. Table 2.4 displays the results: these

suggest that higher WMRs - i.e. few men relative to women - are significantly associated with

higher literacy levels but only for boys. We then move to the IV regressions: Table 2.5 reports

the first-stage and second-stage estimates obtained using child-woman ratio 1 to proxy fertility.

As it can be noticed the instrument is significantly correlated with the endogenous variable (see

also Figure 2.2): tighter marriage markets are associated with lower fertility levels.14 Second

stage estimates present a twofold picture similar to the one from the reduced form model: causal

evidence is absent when considering a general measure of children’s education, while it appears

significant when using boys’ literacy as indicator of quality. Similar results are obtained using

child-woman ratio 2 as dependent variable (see Table 2.6). This suggests that across Spain in

the late 19th century parents’ fertility behaviour had an impact mainly on their sons’ education

rather than on their daughters’.

13When regressing children’s literacy only on adult literacy we get an R-squared above 0.86 and the associatedcoefficient is above 1. Also, by looking at Table 2.2 we notice that including adult literacy does not add much interms of model explanation (R-squared increases but slightly), while it seems to be capturing the effect due tothe dependence on agriculture, urbanization (capital city) and (partially) latitude.

14The set of controls X used to obtain the partial correlation plot in Figure 2.2 is the same as the one displayedin Table 2.5.

42


Tab

le2.2:

Qua

ntityan

dqu

alityof

child

ren:

cross-sectionOLS

Dependent

Boys’

Girls’

Children’s

variable

education

education

education

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Child-w

oman

ratio1

-0.088

*-0.078

-0.112

***

-0.033

-0.031

-0.058

*-0.054

-0.049

-0.079

***

[0.051

][0.052

][0.033

][0.050]

[0.050

][0.034

][0.047

][0.049

][0.030

]Sh

areinagriculture(

men

)-0.221

***

-0.220

***

0.01

4-0.285

***

-0.283

***

-0.093

***

-0.254

***

-0.253

***

-0.041

[0.042

][0.042

][0.032

][0.040]

[0.040

][0.033

][0.038

][0.038

][0.029

]Sh

areu

rban

-0.000

0.00

2-0.013

-0.019

-0.018

-0.030

***

-0.009

-0.008

-0.021

**[0.018

][0.018

][0.011

][0.016]

[0.016

][0.010

][0.016

][0.016

][0.009

]Province’sc

apital(dum

my)

0.03

0**

0.03

0**

-0.002

0.02

7**

0.028*

*0.00

10.02

8**

0.02

8**

>-0.001

[0.013

][0.013

][0.007

][0.011]

[0.011

][0.007

][0.011

][0.011

][0.006

]Sh

areinindu

stry

-0.199

-0.192

0.04

5-0.396

***

-0.389

***

-0.196

***

-0.298

**-0.291

**-0.077

[0.142

][0.136

][0.068

][0.104]

[0.104

][0.074

][0.118

][0.116

][0.062

]Latitud

e0.03

7***

0.01

2*0.01

6*-0.004

0.02

6***

0.00

3[0.010

][0.007

][0.009

][0.007

][0.009

][0.006

]Lon

gitude

-0.006

-0.004

0.00

20.00

3-0.002

-0.000

[0.007

][0.004

][0.006

][0.003

][0.006

][0.003

]Adu

ltliteracy

0.80

5***

0.65

4***

0.73

0***

[0.042

][0.041

][0.037

]Con

stan

t0.55

5***

-1.107

***

-0.405

0.51

8***

-0.338

0.23

20.53

3***

-0.690

*-0.054

[0.047

][0.388

][0.249

][0.042]

[0.358

][0.276

][0.042

][0.358

][0.245]

Provincedu

mmies

yes

yes

yes

yes

yes

yes

yes

yes

yes

F-testof

jointsign

ificance

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

0.00

(p-value

)

R2

0.87

60.88

00.95

10.83

10.83

30.91

50.86

50.86

80.94

8

***,

**,*

deno

tesstatisticalsign

ificanceat

1%,5%

and10%

levels,respectively.Allestimations

includ

e47

3ob

servations.Child-w

oman

ratio1is

thenu

mbe

rof

child

renag

ed0-5over

thenu

mbe

rof

wom

enaged

16-45.

Children’s/bo

ys’/girls’educationistheshareof

child

ren/

boys/g

irlsaged

5-15

that

canread

andwrite.Rob

ust

stan

dard

errors

repo

rted

inpa

rentheses.

43


Table

2.3:Quantity

andquality

ofchildren:

cross-sectionOLS.A

lternativemeasure

offertility

Dependent

Boys’

Girls’

Children’s

variableeducation

educationeducation

(1)(2)

(3)(4)

(5)(6)

(7)(8)

(9)

Child-w

oman

ratio2

-0.128***-0.110***

-0.080***-0.086***

-0.079**-0.054**

-0.105***-0.092***

-0.065***[0.035]

[0.036][0.025]

[0.033][0.033]

[0.023][0.032]

[0.033][0.021]

Shareinagriculture(m

en)-0.192***

-0.197***0.016

-0.260***-0.261***

-0.087**-0.227***

-0.230***-0.036

[0.042][0.042]

[0.032][0.041]

[0.041][0.034]

[0.039][0.039]

[0.029]Shareurban

-0.0020.000

-0.009-0.023

-0.021-0.029***

-0.012-0.010

-0.019**[0.017]

[0.017][0.011]

[0.015][0.015]

[0.009][0.015]

[0.015][0.008]

Province’scapital(dum

my)

0.030**0.030**

-0.0010.027**

0.027**0.002

0.028**0.028**

0.000[0.013]

[0.013][0.007]

[0.011][0.011]

[0.007][0.011]

[0.011][0.006]

Shareinindustry

-0.183-0.182

0.028-0.375***

-0.372***-0.201***

-0.279**-0.277**

-0.086[0.140]

[0.134][0.067]

[0.102][0.102]

[0.073][0.116]

[0.114][0.061]

Latitude

0.033***0.010

0.013-0.006

0.022**0.001

[0.010][0.006]

[0.009][0.007]

[0.009][0.006]

Longitude

-0.006-0.005

0.0010.003

-0.002-0.001

[0.007][0.004]

[0.006][0.003]

[0.006][0.003]

Adult

literacy0.794***

0.648***0.722***

[0.043][0.041]

[0.038]Constant

0.622***-0.911**

-0.3190.578***

-0.1770.307

0.597***-0.513

0.026[0.051]

[0.396][0.248]

[0.043][0.371]

[0.287][0.044]

[0.367][0.248]

Province

dummies

yesyes

yesyes

yesyes

yesyes

yesF-test

ofjoint

significance0.00

0.000.00

0.000.00

0.000.00

0.000.00

(p-value)

R2

0.8780.882

0.9510.834

0.8350.916

0.8680.870

0.948

***,**,*

denotesstatistical

significanceat

1%,5%

and10%

levels,respectively.

Allestim

ationsinclude

473observations.

Child-w

oman

ratio2is

thenum

berof

childrenaged

6-15over

thenum

berof

wom

enaged

21-50.Children’s/boys’/girls’

educationis

theshare

ofchildren/boys/girls

aged5-15

thatcan

readand

write.

Robust

standarderrors

reportedin

parentheses.

44


−.2

−.1

0.1

.2e(C

hild

wom

an r

atio 1

| X

)

−.4 −.2 0 .2 .4e(Log WMR | X)

Figure 2.2: WMR and fertility

Table 2.4: WMR and children’s education: reduced form. OLSDependent Boys’ Girls’ Children’svariable education education education

(1) (2) (3)

(Log)Women-to-men ratio (WMR) 0.061** 0.009 0.030[0.029] [0.030] [0.027]

Share in agriculture (men) -0.229*** -0.288*** -0.259***[0.041] [0.039] [0.037]

Share urban 0.007 -0.016 -0.005[0.018] [0.016] [0.016]

Province’s capital (dummy) 0.035*** 0.029*** 0.031***[0.013] [0.011] [0.011]

Share in industry -0.182 -0.395*** -0.291**[0.142] [0.105] [0.119]

Latitude 0.036*** 0.016* 0.025***[0.010] [0.009] [0.009]

Longitude -0.006 0.001 -0.002[0.007] [0.006] [0.006]

Constant -1.120*** -0.356 -0.705**[0.381] [0.355] [0.355]

Province dummies yes yes yesF-test of joint significance 0.00 0.00 0.00(p-value)

R2 0.880 0.833 0.868

***, **,* denotes statistical significance at 1% , 5% and 10% levels, respectively. All estimations include 473 observations.Children’s/boys’/girls’ education is the share of children/boys/girls aged 5-15 that can read and write. Robust standarderrors reported in parentheses.

45


Table 2.5: Quantity and quality of children: cross-section IV. First and second stage estimatesDependent Child-woman Boys’ Girls’ Children’svariable ratio 1 education education education

(1) (2) (3) (4)

First stage Second stage

Child-woman ratio 1 -0.196** -0.014 -0.087[0.095] [0.100] [0.090]

(Log)Women-to-men ratio (WMR) -0.280***[0.028]

Share in agriculture (men) 0.172*** -0.193*** -0.284*** -0.242***[0.036] [0.044] [0.041] [0.040]

Share urban -0.077*** -0.010 -0.019 -0.013[0.016] [0.018] [0.017] [0.017]

Province’s capital (dummy) -0.033*** 0.030** 0.030*** 0.030***[0.011] [0.012] [0.010] [0.010]

Temporarymalemigration 0.013** -0.022*** -0.020*** -0.021***[0.006] [0.005] [0.005] [0.005]

Share in industry 0.203** -0.133 -0.384*** -0.264**[0.102] [0.137] [0.105] [0.117]

Latitude -0.000 0.036*** 0.016* 0.025***[0.010] [0.010] [0.009] [0.009]

Longitude 0.005 -0.005 0.001 -0.002[0.006] [0.007] [0.005] [0.006]

Constant 0.560 -0.888** -0.231 -0.536[0.403] [0.407] [0.354] [0.365]

Province dummies yes yes yes yes

First-stage F-statistics 100.11

***, **,* denotes statistical significance at 1% , 5% and 10% levels, respectively. All estimations include 473 observations.Child-woman ratio 1 is the number of children aged 0-5 over the number of women aged 16-45. Children’s/boys’/girls’education is the share of children/boys/girls aged 5-15 that can read and write. Robust standard errors reported inparentheses. The instrument for the child-woman ratio is the log of the women-to-men ratio (WMR) in the adultpopulation (aged 21-50).

46


Table 2.6: Quantity and quality of children: cross-section IV. Alternative dependent variableDependent Child-woman Boys’ Girls’ Children’svariable ratio 2 education education education

(1) (2) (3) (4)

First stage Second stage

Child-woman ratio 2 -0.187** -0.013 -0.083[0.092] [0.095] [0.085]

(Log)Women-to-men ratio (WMR) -0.294***[0.051]

Share in agriculture (men) 0.321*** -0.166*** -0.282*** -0.230***[0.060] [0.052] [0.049] [0.048]

Share urban -0.063** -0.007 -0.018 -0.012[0.028] [0.017] [0.017] [0.016]

Province’s capital (dummy) -0.027 0.031*** 0.030*** 0.030***[0.018] [0.012] [0.010] [0.010]

Temporarymalemigration 0.016 -0.021*** -0.020*** -0.021***[0.014] [0.005] [0.005] [0.005]

Share in industry 0.168 -0.142 -0.385*** -0.268**[0.157] [0.137] [0.104] [0.115]

Latitude -0.031*** 0.030*** 0.015* 0.022**[0.011] [0.010] [0.009] [0.009]

Longitude -0.002 -0.007 0.001 -0.003[0.009] [0.006] [0.005] [0.005]

Province dummies yes yes yes yes

First-stage F-statistics 33.73

***, **,* denotes statistical significance at 1% , 5% and 10% levels, respectively. All estimations include 473 observations.Child-woman ratio 2 is the number of children aged 6-15 over the number of women aged 21-50. Children’s/boys’/girls’education is the share of children/boys/girls aged 5-15 that can read and write. Robust standard errors reported inparentheses. The instrument for the child-woman ratio is the log of the women-to-men ratio (WMR) in the adultpopulation (aged 21-50).

47


2.3.1 Allowing for spatial dependence

Children’s education might be also driven by a diffusion process, where the geographical spread

of attitudes towards education might play a relevant role. As mentioned above, Figure 2.1 shows

the geographical distribution of boys’ literacy where larger dots stand for higher literacy rates.

We notice that there are geographical patterns, with Northern areas on average characterized by

higher children’s education.

We check the degree of spatial autocorrelation in boys’ education across Spanish districts by

looking at the Moran’s I. Moran’s I is a measure of spatial autocorrelation characterizing the

relationship of the values of a variable with the geographical location where they were measured.

Figure 2.3 shows the Moran scatterplot of the relationship between the share of literate boys (aged

5-15) and its corresponding spatially lagged component. As it can be noticed, the majority of

observations are placed in the first and third quadrants, suggesting the existence of positive

spatial autocorrelation (i.e. districts characterized by higher boys’ education surrounded by

districts with a similar pattern, and similarly for districts with lower literacy rates).

To assess whether accounting for spatial dependence affects the negative association between

quantity of children and boys’ education, spatial lag and error models are estimated using OLS

and IV (Anselin 1988).15

The spatial lag model is defined as follows:

educi = ρWeduci + γ1 ferti + γ2Xi + ψi (2.2)

where W is the spatial weight matrix and Weduci is the spatially lagged dependent variable.

15The inverse distance spatial weights matrix is computed using latitude and longitude of the seat of each dis-trict. Latitude and longitude are available at http://www.businessintelligence.info/docs/listado-longitud-latitud-municipios-espana.xls

48


Moran scatterplot (Moran’s I = 0.292)educm5_15_h

Wz

z−2 −1 0 1 2 3

−1

0

1

Figure 2.3: Moran scatter plot of boys’ literacy in 1887

Instead the spatial error model includes a spatial component in the error term:

educi = γ1 ferti + γ2Xi + µi where µi = λWµi + εi (2.3)

where Wµi is the spatially lagged error term.

Table 2.7 displays the estimations of the empirical models: OLS spatial lag model (columns 1

and 2), OLS spatial error model (columns 3 and 4), and 2SLS spatial lag model (columns 5

and 6).16 As estimation results suggest, a process of diffusion might be in place (positive and

significant rho) but it does not affect the negative association between fertility and boys’ literacy

across Spanish districts in 1887.

16Using a maximum likelihood estimator to estimate the spatial lag and error models yields similar results.

49


Table 2.7: Quantity and quality of children: spatial lag and error modelsDependent variable Boys’ education

(1) (2) (3) (4) (5) (6)

Model Spatial Spatial Spatial Spatial Spatial Spatiallag lag error error lag lag

OLS OLS OLS OLS 2SLS 2SLS

Child-woman ratio 1 -0.095** -0.069 -0.293***[0.047] [0.049] [0.094]

Child-woman ratio 2 -0.110*** -0.102*** -0.269***[0.033] [0.035] [0.094]

Share in agriculture (men) -0.204*** -0.185*** -0.220*** -0.197*** -0.152*** -0.119**[0.038] [0.038] [0.039] [0.040] [0.044] [0.052]

Temporarymalemigration -0.031*** -0.030*** -0.027*** -0.025*** -0.034*** -0.032***[0.006] [0.006] [0.006] [0.006] [0.006] [0.006]

Share urban 0.003 0.003 0.001 -0.001 -0.008 -0.003[0.017] [0.016] [0.017] [0.016] [0.018] [0.017]

Province’s capital (dummy) 0.033*** 0.033*** 0.031*** 0.032*** 0.030** 0.032**[0.012] [0.012] [0.012] [0.012] [0.012] [0.012]

Share in industry -0.179 -0.176 -0.188 -0.176 -0.116 -0.130[0.129] [0.128] [0.130] [0.128] [0.142] [0.143]

Latitude 0.020** 0.017* 0.037*** 0.033*** -0.006 -0.010[0.009] [0.010] [0.009] [0.009] [0.013] [0.014]

Longitude -0.006 -0.007 -0.005 -0.006 -0.005 -0.007[0.006] [0.006] [0.007] [0.007] [0.006] [0.006]

Constant -0.454 -0.280 -0.826** -0.639 0.182 0.458[0.398] [0.409] [0.406] [0.413] [0.473] [0.532]

Province dummies yes yes yes yes yes yes

ρ 0.900*** 0.894*** 2.097*** 1.862***λ 0.372 0.233

***, **,* denotes statistical significance at 1% , 5% and 10% levels, respectively. All estimations include 473 observations.Child-woman ratio 1 is the number of children aged 0-5 over the number of women aged 16-45 while child-woman ratio 2is the number of children aged 6-15 over the number of women aged 21-50. Boys’ education is the share of boys aged 5-15that can read and write.

50


2.4 Conclusion

This paper studies the association between quantity and quality of children in historical Spain

using macroeconomic (i.e. district level) data. While evidence of a negative effect of parents’

fertility on a general measure of child literacy is weak, the relationship is significant when consid-

ering only boys’ literacy. This result adds to the within-country empirical evidence that supports

the existence of a quantity-quality trade-off in a historical context (e.g. Becker et al. 2010, Klemp

and Weisdorf 2011) for a Southern European country, thus widening the literature on this issue.

As one can expect that in a historical period in which girls’ education is driven by several factors

that go beyond a pure budgetary mechanism, it also highlights that distinguishing quality of

children by gender might be important.

51

Chapter 3

Human capital, culture and the onset

of the fertility transition

3.1 Introduction

The transformation of an economy from a regime of Malthusian stagnation to one of sustained

growth is fundamentally linked to the process of the demographic transition. By turning to

negative the relation between income and fertility, this transition plays a key role in fostering

economic development and income growth (e.g. Galor and Weil 1999, 2000). As a consequence,

one would expect that countries that first experienced the onset of the demographic transition

would be relatively richer than those that experienced it later on or that have not yet experienced

it. Figure 3.1 shows a scatterplot of per-capita income in the year 2000 and the year at which each

country experienced its fertility transition. These dates have been estimated by Reher (2004)

and identify permanent declines in birth rates, assigning the year 2000 as the transition date

for countries that had not yet experienced the onset.1 The relation between these two variables

is strongly negative, which is consistent with the importance of experiencing a demographic

1These data has been recently used and cited in several papers like for instance Galor (2012) and Andersen etal. (2010). We drop 12 countries that were assigned the year 2000 to avoid arbitrariness.

3 Human capital, culture and the onset of the fertility transition

DZA

ARG

AUT

BHR

BGD

BRBBEL

BEN

BOL

BWABRA

CMR

CAN

CAF

CHL

CHN

COL

CRI

DNK

DOM

ECUEGYSLV

GBRFINFRA

GMB

DEU

GHA

GTM

GUY

HTI

HND

HUN

IND

IDNIRQ

ISRITA

JAM

JPN

JOR

KEN

KOR

KWT

LSO

LBR

MWI

MYS

MLI

MUS

MEX

NPL

NLD

NIC

NER

NOR

PAN

PRYPERPHL

PRT

PRI

RWA

SEN

SGP

ZAF

ESP

LKA

SDN

SWZ

SWECHE

SYR

TZA

THA

TGO

TTO

TUN

USA

URY VEN

ZMB

ZWE

68

10

12

(Lo

g)

GD

P p

er

ca

pita

in

20

00

1850 1900 1950 2000Onset of the DT

Figure 3.1: Onset of the fertility transition and GDP per capita in 2000

transition to enter the sustained growth regime.2

A recent strand of the literature highlights the role of culture in explaining economic development

across countries (Guiso et al. 2006; Spolaore and Wacziarg 2009). Spolaore and Wacziarg (2009)

explain a significant fraction of income differences across countries using their genetic distance

(relative to the technological frontier), which, according to their view, should measure barriers

to the adoption and diffusion of new technology from this frontier. Their measure of genetic

distance captures the general relatedness between populations: the closer two populations are in

terms of genetic distance, the smaller their differences in traits and social norms (e.g., beliefs,

habits, biases, etc.). On the other hand, the literature emphasizes also the role of historical

human capital in promoting a country’s development. Expansion of education is often regarded

as one of the fundamental factors in economic development. Comparative analysis suggests that,

2Another channel through which the demographic transition may spur a country’s per-capita income is theso-called demographic gift, by which a lower population growth rate decreases the dependency rate through itseffect on the population age structure (Bloom and Williamson 1998).

54


among several factors, historical differences in human capital might be responsible for different

paths of development observed during and after the colonization period. For example, Glaeser et

al. (2004) argue that European settlers brought their human capital where they settled in large

numbers, thus fostering technological progress, growth and better institutions.

Following Spolaore and Wacziarg (2009), that use genetic distance to the United Kingdom (UK)

and the United States (US) as a proxy for cultural relatedness to the technological frontier, we

show that genetic distance to the UK (US) has been important in shaping the timing of the fertil-

ity transition across countries. This result is consistent with an indirect channel working through

technology diffusion as in Spolaore and Wacziarg (2009, 2011). Larger genetic distance to the

technological frontier would delay technology adoption and lower productivity and the demand

for human capital, consequently leading to a late onset of the fertility transition. The mechanism

we highlight here follows Galor and Weil (2000) who argue that increasing technological progress

boosts the demand for human capital and, because of the higher return to education, households

eventually trade quantity for quality of children. When a significant fraction of families de-

cides to have fewer and more educated children, the onset of demographic transition takes place.

Therefore, culture and informal institutions, by affecting incentives to innovate and accumulate

human capital, might have shaped the timing of fertility transitions and, consequently, the cur-

rent distribution of income across countries throughout the world. Our reasoning is that genetic

distance to the UK, through its effect on technology adoption and human capital accumulation,

facilitate the onset of the transition, but this does not necessarily mean that the technological

frontier has to be the first country to experience such transition.3

In our analysis we use the UK as the main reference country since it was the technological

leader until the early twentieth century. However, given that most of the fertility transitions in

our sample took place after 1950, we also consider using the US as the reference country.4 The

timing of the demographic transition differs widely across countries, as shown in Table 3.1, which

3There are other factors that are important in explaining the onset of fertility transitions across countries.In fact, the UK, which belongs to the group of "early" transitions (i.e. before 1950) is not the first countryexperiencing the onset - Sweden had its transition in 1865, according to Reher.

4In our largest sample 23 out of 124 countries experienced the onset of the transition before 1950, excludingthe countries assigned a transition in the year 2000.

55


lists the years at which the different countries reached their fertility transition as estimated in

Reher (2004). Figure 3.2 displays a histogram of these dates.5

Table 3.1: Reher’s (2004) estimates of the onset of the demographic transition

Albania 1965 Denmark 1910 Korea, Rep. 1960 Portugal 1925Algeria 1975 Djibouti 1985 Korea, Dem. Rep. 1970 Qatar 1955Angola 1995 Dominican Rep. 1965 Kuwait 1975 Romania 1935Antigua 1960 Ecuador 1970 Kyrgyzstan 1965 Rwanda 1995Argentina 1910 Egypt 1965 Laos 1995 Saudi Arabia 1980Armenia 1965 El Salvador 1965 Lebanon 1965 Senegal 1980Austria 1915 Eritrea 1990 Lesotho 1985 Seychelles 1955Azerbaijan 1965 Ethiopia 1990 Liberia 1995 Singapore 1955Bahamas 1965 Finland 1915 Libya 1980 South Africa 1975Bahrain 1970 France 1900 Madagascar 1990 Spain 1910Bangladesh 1980 Gambia 1985 Malawi 1980 Sri Lanka 1960Barbados 1955 Georgia 1965 Malaysia 1965 Sudan 1980Belgium 1905 Germany 1900 Mali 1995 Suriname 1965Belize 1965 Ghana 1985 Mauritania 1980 Swaziland 1975Benin 1985 Guatemala 1985 Mauritius 1960 Sweden 1865Bhutan 1995 Guinea 1995 Mexico 1970 Switzerland 1910Bolivia 1975 Guyana 1965 Mongolia 1975 Syria 1985Botswana 1975 Haiti 1985 Morocco 1965 Taiwan 1955Brazil 1965 Honduras 1985 Myanmar (Burma) 1975 Tanzania 1975Brunei 1960 Hungary 1890 Namibia 1990 Thailand 1965Bulgaria 1925 India 1960 Nepal 1995 Togo 1985Burundi 1995 Indonesia 1970 Netherlands 1910 Trinidad and Tobago 1965Cameroon 1980 Iran 1985 Nicaragua 1985 Tunisia 1965Canada 1905 Iraq 1975 Niger 1985 Turkmenistan 1965Central Afr. R. 1990 Israel 1955 Nigeria 1995 United Kingdom 1910Chile 1960 Italy 1925 Norway 1905 United States 1925China 1970 Ivory Coast 1985 Oman 1995 Uruguay 1890Colombia 1965 Jamaica 1925 Panama 1970 Uzbekistan 1965Comoros 1990 Japan 1950 Paraguay 1985 Venezuela 1965Costa Rica 1965 Jordan 1975 Peru 1975 Vietnam 1980Cuba 1920 Kenya 1980 Philippines 1955 Zambia 1980 Zimbabwe 1970

Excluding countries that were assigned the onset in the year 2000. These are: Afghanistan, BurkinaFaso, Chad, Congo, Democratic Republic of Congo, Gabon, Guinea Bissau, Mozambique, Sierra Leone,Somalia, Uganda, Yemen.

5Some of Reher’s onset dates differ from other sources (Coale and Watkins 1986, Bailey 2009). In Section 3.3.3we check the robustness of our results using alternative dates.

56


05

10

15

20

25

Nu

mb

er

of

ep

iso

de

s

1850 1900 1950 2000Year of the onset of the demographic transition (Reher, 2004)

Figure 3.2: Year of the onset of the demographic transition (Reher 2004)

As the data show, most of the countries that experienced the transition in the late 19th and

early 20th centuries, were located in Western Europe. In contrast, most countries belonging to

Asia, Africa, and Latin America experienced a late transition (that is, after 1950).

In this paper we exploit cross-country variation to shed light on the determinants of the fertility

transition around the world. Several mechanisms have been proposed to explain the fertility

transition: a rise in the demand for human capital (Galor and Weil 2000), a rise in income

during industrialization (Becker and Lewis 1973, Becker, 1981), a reduction in child and infant

mortality rates (Coale 1973, van de Walle 1986, Sah 1991, Galloway et al. 1998, Eckstein et al.

1999, Kalemli-Ozcan 2002, Angeles 2010), and a reduction in gender gaps (Galor and Weil 1996,

Goldin 1990, and Lagerlöf 2003).6 Data limitations prevent us to run a formal horserace between

these competing explanations of the triggers of the demographic transition. Instead, our goal is

to explore the contribution of a specific variable to this process and rationalize the mechanism

6Guinnane (2011) and Galor (2012) provide reviews of the factors behind fertility transitions.

57


through which it operates. In particular, here we focus on a country’s cultural relatedness to

the technological frontier and show that its impact can be mainly attributed to its effect on

human capital accumulation. We also test whether other measures of historical institutions -

executive constraints and polity2 scores - are related to the onset of the demographic transition

across countries. Contrary to the proxy of informal institutions (i.e. genetic distance), these

alternative measures of historical formal institutions do not show a robust relationship with the

year of the onset. This result is consistent with cultural relatedness to technological frontier

favouring technology adoption (Spolaore and Wacziarg 2009, 2011) which foster human capital

accumulation and the onset of the fertility transition (Galor 2012). Figure 3.3 shows a strong

positive correlation between the demographic transition years and genetic distance to the UK.

The main findings of our paper can be summarized as follows. First, a large genetic distance

with respect to the UK (US) delays a country’s fertility transition. Second, when we instrument

a country’s schooling levels in 1870 with genetic distance to the UK and the percentage of

Protestants in the population - or alternatively with a country’s physical distance from Germany

- we find a strong causal effect of human capital on the onset of the fertility transition, as predicted

by Galor andWeil (2000).The mechanism behind this relationship is as follows. Genetic proximity

to the UK enhances a country’s demand for human capital. Protestantism, on the other hand,

is associated with a boost in the supply of human capital. These two effects enhance human

capital accumulation which in turn induces families to reduce their offspring, triggering the

fertility transition.

The paper is organized as follows. Section 3.2 summarizes the sparse empirical literature that

has attempted to isolate different triggers of fertility transitions across countries. Section 3.3

describes the data and methodology used in the analysis. Section 3.4 presents the finding that

genetic distance from the UK is a robust determinant of the onset of the fertility transition across

countries. Section 3.5 illustrates the mechanism at work. Finally, Section 3.6 concludes.

58


ALB

DZA

AGO

ATG

ARG

ARM

AUT

AZE BHSBHR

BGD

BRB

BEL

BLZ

BEN

BTN

BOL BWA

BRABRN

BGR

BDI

CMR

CAN

CAF

CHL

CHNCOL

COM

CRI

CUB

DNK

DJI

DOMECU

EGY SLV

ERIETH

FIN

FRA

GMB

GEO

DEU

GHAGTM

GIN

GUY

HTIHND

HUN

IND

IDN

IRN

IRQ

ISR

ITA

CIV

JAM

JPN

JORKEN

KOR

PRKKWT

KGZ

LAO

LBN

LSO

LBR

LBY

MDG

MWI

MYS

MLI

MRT

MUS

MEXMNG

MAR

MMR

NAMNPL

NLD

NIC NER

NGA

NOR

OMN

PAN

PRY

PER

PHL

PRT

QAT

ROU

RWA

SAU SEN

SYCSGP

ZAF

ESP

LKA

SDN

SUR

SWZ

SWE

CHE

SYR

TWN

TZA

THA

TGO

TTOTUN TKM

USA

URY

UZBVEN

VNM ZMB

ZWE7

.52

7.5

47

.56

7.5

87

.6(L

og

) O

nse

t o

f th

e D

T

3 4 5 6 7 8(Log) Genetic distance to the UK (weighted)

Correlation=0.70***

Figure 3.3: Genetic distance to the UK and the onset of the fertility transition

3.2 Literature review

Since our contribution is purely empirical, in this section we limit ourselves to discussing the

empirical papers that analyze possible triggers of fertility transitions.7 The Princeton European

Fertility Project (e.g. see Coale andWatkins 1986) was one of the first comprehensive studies that

used data from the 19th century to document different demographic transitions in Europe and

analyze their possible triggers. The emphasis in this project, however, was mainly on cultural and

sociological explanations, ignoring economic factors. More recently, the development of unified

growth theories that seek to explain economic growth in the very long run has spurred interest

in identifying the role of different socio-economic factors in explaining demographic transitions.

The first - and most common - methodological approach has been to study the correlation between

fertility and income at different time periods. For instance, using a sample of countries in the

7The literature review in Galor (2012) also includes theoretical papers.

59


1960-1999 period, Lehr (2009) examines the existence of different regimes in terms of fertility

dynamics. She finds that, at early stages of development, increases in productivity and primary

schooling-enrolment are typically associated with increases in fertility. In contrast, at higher

levels of development, productivity and education are shown to be negatively associated with

fertility, whereas the level of parents’ human capital has a somewhat positive effect. In all periods,

increases in secondary-schooling enrolment are correlated with drops in fertility rates. Herzer et

al. (2012) find evidence that increasing income and falling mortality are the main explanatory

factors of fertility declines over the 20th century across a selected sample of countries. Murtin

(forthcoming) uses data for a large panel of countries since 1870 and concludes that education

is the main trigger of changes in the birth rate and that the effect of health improvements is of

second order. Becker et al. (2010) use data on Prussian counties in 1849 and identify a negative

relation between child quantity and education in a context in which the demographic transition

has not yet taken place. Another finding of their study is that the initial level of education is

a good predictor of the demographic transition that occurred in Prussia during the 1880-1905

period. Finally, Murphy (2010) analyzes historical French département data for the late 19th

century and finds that both economic and cultural factors had an effect on different fertility

patterns across these geographical units. In particular, education, measured as female literacy

and child enrolment in primary schools, has a negative impact on fertility, whereas wealth is

correlated with larger family sizes.8

A different approach is to use information on the years of the onset of fertility transitions in

different countries to directly identify their main historical determinants. Andersen et al. (2010)

use this strategy to analyze how cataract incidence explain cross-country variation in labour

productivity. They argue that an earlier onset of vision loss reduces the return to human capital,

8There are several studies that focus on the closely related children’s quantity-quality trade-off. For instance,Rosenzweig and Wolpin (1980) were the first to use exogenous variations in fertility to identify the effect of childquantity on child quality. They instrumented child quantity with increases in family size resulting from multiplebirths and show that child quantity significantly reduces children’s education. Bleakley and Lange (2009) explorethe causal effect of education on fertility by exploiting the eradication policy of the hookworm disease in southernstates in North America. Their paper argues that this eradication increased the return to schooling and hencereduced the price of child quality. This exogenous change, in turn, increased school attendance and reducedfertility. Other relevant papers are Angrist et al. (2005), Black et al. (2005), and Qian (2009). See Schultz (2008)for a summary of this literature.

60


and hence delays the demographic transition.

3.3 Data and methodology

3.3.1 Baseline analysis

In our baseline analysis our main variable of interest is a proxy of informal institutions. Specifi-

cally we consider a measure of cultural relatedness, genetic distance to the UK (or the US) taken

from Spolaore and Wacziarg (2009) - SW henceforth - aiming to capture cultural proximity to

the technological frontier.9

We first investigate the effect of genetic distance to the UK on the timing of the fertility transition

across countries by estimating the following model using ordinary least squares (OLS):10

log onseti = β1 + β2 ∗ log gendisti,UK + β′3Xi + εi (3.1)

where log onseti represents (the log of) the year of the onset of the fertility transition in country

i, log gendisti,UK represents (the log of) genetic distance to the UK in country i, Xi is a set of

country i control variables, and ε is a standard error term. Xi includes different sets of standard

determinants of long-run development and productivity used in the literature. To account for

the potential effect of geography and climate, we control for the absolute latitude of a country’s

centroid, the average distance to the nearest ice-free coast, the malaria ecology index, and a set of

continental dummies (Africa, Asia, Europe, North America, and South America). The historical

variables included in the regressions are population density in 1400 and the years passed since

the Neolithic revolution (i.e. the agricultural transition).11 We also control for the type of legal

origins (British common law, French civil law, socialist law, German civil law, and Scandinavian

9Throughout our analysis we use the measure weighted genetic distance that accounts for sub-populations’genetic groups. The other measure provided by Spolaore and Wacziarg (2009), named dominant genetic distance,considers only the largest groups of each country’s population.

10We take logs of the two key variables to reduce the impact of outliers.11Data on the agricultural transition are from Louis Putterman’s Agricultural Transition Year Country Data

Set.

61


law) and the 1900 shares of protestants, catholics and muslims.12

Table 3.2 contains the definitions and sources of all the variables used in the cross-sectional

exercise.

3.3.2 Bilateral analysis

In this section we follow an approach similar to SW. We assess whether the role of cultural

relatedness to the technological frontier as a determinant of the fertility transition is still present

using a bilateral approach considering countries pair by pair. One advantage of this approach

is that it makes use of a much larger dataset and so it helps increasing the precision of our

estimates. To do so, we regress the distance in the onset of the fertility transition between each

pair of countries on their genetic distance relative to the UK (US) and on a set of controls very

similar to those of SW aimed at capturing geographical, climatic, and historical differences which

can be interpreted as distances. We account for the effect of geographical distances by including

the absolute difference in latitudes and longitudes, the geodesic distance between countries, a

dummy that takes a value of one if both countries in the pair are contiguous, a dummy that

takes a value of one if at least one country is landlocked, a dummy that takes a value of one if at

least one country is an island, and a measure of climatic similarity based on 12 Koeppen-Geiger

climate zones.13 We also add as covariates a set of dummies that take a value of one if two

countries in a pair are located in the same continent. We include a measure of transportation

costs based on freight rates for surface transport (sea or land).14 To control for common historical

and cultural characteristics we use a dummy taking a value of one if both countries in a pair share

the same legal origins, and zero otherwise; a dummy taking a value of one if both countries in a

pair share the same colonial origins, and zero otherwise; and a dummy taking a value of one if

both countries share a common official language. As for climate, religious similarity is measures

12Religion adherence is particularly important in our context as some religions differed substantially in thepromotion of literacy and education (Ferguson, 2011).

13This is measured as the average absolute value difference in the percentage of land area in each of the 12climate zones between two countries.

14Transportation-cost data is from http://www.importexportwizard.com/. The measure refers to 1000kg ofunspecified freight transported over sea or land, with no special handling.

62


with the average absolute value difference, between two countries, in the percentages of religions

followers in 1900 in each of 10 religious categories. All variables used in this section, along with

their sources, are listed in Table 3.3. Our estimation model in this case is the following:

|log onseti − log onsetj | = α+ β|log gendisti,UK − log gendistj,UK |+ γ′Qi,j + εi,j (3.2)

where |log onseti − log onsetj | represents the absolute value of the log difference in the year of

the onset of the fertility transition between country i and j, |log gdi,UK − log gdj,UK | represents

the absolute value of the genetic (log) distance relative to the UK between country i and j and

Qi,j includes the mentioned measures of geographical, climatic and historical distances between

country i and j. Finally, εi,j is the error term associated with the country pair ij.15 This approach

allows us to investigate whether differences (and similarities) in culture (relative to the UK and

the US) explain the distance in the timing of the onset of fertility transitions between pairs

of countries. Specifically, we ask whether similar (different) timing in the onset is explained

by similar (different) culture (relative to the UK and US), controlling for the effect of similar

(different) geographical, climatic, and historical contexts.

3.3.3 Robustness checks

Next we perform two robustness checks. First, we use alternative dates for the onset of fertility

transitions for those who experienced an "early transition". Reher’s dates might be criticized

especially for some countries as France which are assigned a relatively late onset. To account for

this, we use dates from Coale and Watkins (1986) and Bailey (2009) which are directly related

to the European Fertility Project. Using alternative onset dates is a sensible thing to do, since

some of Reher dates have been criticized on two grounds.

15As Spolaore and Wacziarg (2009) point out, spatial correlation results from the construction of the dependentvariable. We follow their strategy to address this issue by using two-way clustered standard errors.

63


Table

3.2:Variables

anddata

sources:cross-section

analysis

Variable

nameand

descriptionSource

Onset

ofthe

fertilitytransition

Reher

(2004);Bailey

(2009)Genetic

distanceto

theUK

(USA

),weighted

Spolaoreand

Wacziarg

(2009)Executive

constraintsin

1850and

1900Polity4,version

3(2008)

Polity2

scorein

1850and

1900Polity4,version

3(2008)

Absolute

valueof

latitudeof

countrycentroid

Nunn

andPuga

(2012);andGallup

etal.(2001)

Average

distanceto

nearestice-free

coast(1000

km)

Nunn

andPuga

(2012)Continentaldum

mies

Nunn

andPuga

(2012)Malaria

ecologyindex

Sachset

al.(2004)

Population

densityin

1400Nunn

andPuga

(2012)Years

passedsince

theNeolithic

revolutionPutterm

an(2006)

LegaloriginsNunn

andPuga

(2012)Shares

ofreligion

followers

in1900

Robert

Barro’s

website

Average

yearsof

education(age

15-64)Morrisson

andMurtin

(2009)Geodesic

distanceto

Germ

anyhttp://w

ww.cepii.fr/anglaisgraph/bdd/distances.htm

64


Tab

le3.3:

Variables

andda

tasources:

bilaterala

nalysis

Variablena

mean

ddescription

Source

Onset

ofthefertility

tran

sition

Reher

(2004)

Genetic

distan

cerelative

toUK,w

eigh

ted

Spolaore

andWacziarg(2009)

Absolutevalueof

latitude

ofcoun

trycentroid

Nun

nan

dPug

a(2012);a

ndGallupet

al.(2001)

Con

tinental

dummies

Nun

nan

dPug

a(2012)

Dum

myforland

locked

Nun

nan

dPug

a(2012)

Dum

myforisland

CIA

Factbo

okDum

myforcoun

tries’

contiguity

http://w

ww.cepii.fr/ang

laisgrap

h/bd

d/distan

ces.htm

Legalo

rigins

Nun

nan

dPug

a(2012)

Colon

ialh

istory

Nun

nan

dPug

a(2012)

Areain

each

Kop

perclim

atic

zone

Gallupet

al.(2001)

Absolutevalueof

long

itud

eof

coun

trycentroid

Nun

nan

dPug

a(2012);a

ndGallupet

al.(2001)

Geodesicdistan

cebe

tweencoun

tries

http://w

ww.cepii.fr/ang

laisgrap

h/bd

d/distan

ces.htm

Com

mon

official

lang

uagesbe

tweenpa

irof

coun

tries

http://w

ww.cepii.fr/ang

laisgrap

h/bd

d/distan

ces.htm

Shares

ofrelig

ionfollo

wersin

1900

Rob

ertBarro’s

website

Transpo

rtationcosts

http://w

ww.im

portexpo

rtwizard.com/

65


First, some of the "early" transitions in Reher seem to take place too late. This seems to be the

case for instance in France, where other sources suggest that the fertility transition took place

around 1827 rather than 1900. Second, there seems to be too much bunching across fertility

transition years in the Reher estimates, as all the dates occur precisely at the beginning of a

decade or exactly in the middle of it. Table 3.4 shows the discrepancies in the dates calculated

by Reher (2004), Coale and Watkins (1986) and Bailey (2009). The first thing to notice is that

the discrepancies only occur in Western countries, the ones that were the focus of Coale and

Watkins (1986) and Bailey (2009). Second, the Coale-Watkins dates and the Bailey’s ones are

very similar in most cases. One exception is France, for which Coale-Watkins estimate that

the fertility transition took place in 1827, while Bailey’s date is 1814. The second robustness

check we perform is to control for alternative measures of formal historical institutions, namely

executive constraints and an index of democracy scores in 1850 and 1900.16 Although there is no

formal theory that directly links the fertility transition to the quality of institutions it can be the

case that human capital promotion is enhanced by a well-functioning institutional framework.

Table 3.4: Alternative fertility transitions datesCountry Reher Coale-Watkins Bailey

Austria 1915 1907 1908Belgium 1905 1881 1882Denmark 1910 1898 1899England 1905 1892 1892Finland 1915 1912 1911France 1900 1827 1814Hungary 1890 1910 1900Italy 1925 1913 1912Netherlands 1910 1897 1897Norway 1905 1903 1904Portugal 1925 1916 1916Spain 1910 1920 1919Sweden 1865 1902 1897Switzerland 1910 1887 1886

Coale and Watkins (1986) also provide transition dates for Germany, Greece, Ireland, Russia, and Scotland. Weomit those here since Reher does not include these countries in his sample. Bailey (2009) adds Bulgaria and

Wales to this list.

16The variables we use are xconst and polity2 from the data set Polity IV and measure a country’s institutionalframework. See Marshall and Jaggers (2008) for a detailed description.

66


3.4 Results: genetic distance to the technological frontier and the

onset of fertility transition

3.4.1 Baseline analysis

Following SW, genetic distance might indirectly affect the timing of the fertility decline as it

proxies for a cultural environment favourable to technological progress and adoption of innova-

tions. This would favour education and human capital accumulation, then triggering an earlier

onset of the fertility transition (Galor 2012). Here we test for the existence of this indirect

channel. In Section 3.5 we will provide evidence suggesting that this mechanism is plausible

and that the effect of genetic distance from the technological frontier on the timing of the on-

set of the fertility transition is accounted by historical levels of educational attainments. Table

3.5 shows the estimation results obtained by regressing the timing of the fertility transition on

genetic distance to the UK. Specification 1 simply uses the log of genetic distance to the UK

as regressor. Its impact is positive and statistically significant, suggesting that a larger genetic

distance from the UK (i.e. a larger difference in the cultural environment with respect to the

technological frontier) delays the onset of the fertility transition. This estimate is qualitatively

similar if one adds geography and climate, history, legal origins, and religion as controls. In-

cluding all these regressors simultaneously in the same specification does not significantly alter

the results (column 6); the same applies when considering genetic distance to the US (column

7) which, as mentioned above, it may be considered the technological frontier after 1950. The

size of these estimates is quantitatively important. The coefficients from specification 6 suggest

that, for instance, if Lesotho had been culturally as similar - in terms of genetic distance - to the

British population as Spain, then it would have experienced a fertility transition twenty-eight

years earlier than what the model predicts (in 1944 rather than 1972).

67


Table

3.5:Cross-section

OLS:determ

inantsof

theonset

offertility

transitions(1)

(2)(3)

(4)(5)

(6)(7)

(Log)Genetic

distance0.0093***

0.0024**0.0106***

0.0085***0.004**

0.0034**to

theUK

[0.0008][0.0012]

[0.0011][0.0008]

[0.0017][0.0016]

(Log)Genetic

distance0.0055*

tothe

USA

[0.0028]Geography

andclim

ateno

yesno

noyes

yesyes

History

nono

yesno

yesyes

yesLegalorigins

nono

noyes

yesyes

yesReligion

nono

nono

noyes

yesR-squared

0.490.73

0.530.61

0.780.81

0.8Observations

124116

114124

109108

108**,**,*

denotesstatisticalsignificance

at1%

,5%and

10%levels,respectively.

Estim

ationwith

robuststandard

errors(reported

insquared

brackets).Allregressions

includeaconstant.

Dependent

variable:(Log)

Onset

ofthefertility

transitionas

inReher

(2004).

68


3.4.2 Bilateral analysis

Table 3.6 shows the OLS estimates obtained from the regression in Equation 3.2. In column 1,

where we do not add any control variable, larger differences in genetic distance (relative to the

UK) are associated with larger time distances in the onset of fertility transition. In columns 2 to

5, we add different controls. In particular, column 2 adds measures of geographical differences,

column 3 includes a measure of climatic similarity, column 4 includes a set of continental dum-

mies, whereas column 5 adds the measure of transportation costs described above. Throughout

all specifications, including column 6 where we add all the controls, larger genetic distances (rel-

ative to the UK) are associated with wider differences in the timing of the fertility transition.

In columns 7-9 we add controls for similar legal origins, colonial history, language and religion,

respectively. The inclusion of these variables affects neither the significance nor the size of the

coefficient associated with the difference in genetic distance relative to the UK. These results

again provide strong evidence of the importance of cultural differences - specifically relative to

the technological frontier - in determining international differences in the onset of the fertility

transition.

3.4.3 Robustness checks

Using alternative dates of the onset of fertility decline for some of the transitions (mainly "early"

ones, which correspond to Western countries) does not alter our main result as it can be noticed

by looking at Table 3.7.17 If anything, the association is strengthened as, in all cases, the

coefficients of interest are larger in absolute value using these onset dates. We also test the role

of formal institutions on the timing of fertility transition using different proxies as executive

constraints and a democracy index scores measured in 1850 and 1900 (from the data set Polity

IV, see Marshall and Jaggers 2008). The question we ask here is whether the effect of genetic

distance to the technological frontier on the timing of the onset of the fertility transition is robust

17For the sake of brevity we only report the results using the alternative dates from Coale and Watkins (1986).Considering the Bailey (2009) dates gives us almost identical estimates.

69


Table

3.6:Bilateralanalysis:

OLS

(1)(2)

(3)(4)

(5)(6)

(7)(8)

(9)

Genetic

logdistance

0.0081***0.0071***

0.0078***0.0075***

0.0081***0.006***

0.006***0.0058***

0.0057***relative

tothe

UK

[0.0008][0.0008]

[0.0008][0.0008]

[0.0008][0.0008]

[0.0008][0.0009]

[0.0008]Geography

noyes

nono

noyes

yesyes

yesClim

ateno

noyes

nono

yesyes

yesyes

Continentaldum

mies

nono

noyes

noyes

yesyes

yesTransportation

costsno

nono

noyes

yesyes

yesyes

Legalorigins,colonialno

nono

nono

noyes

noyes

historyand

languageReligion

nono

nono

nono

noyes

yesObservations

72606328

63287260

72606328

63285995

5995**,

**,*denotes

statisticalsignificance

at1%

,5%

and10%


Standarderrors

areclustered

(two-w

ay)and

reportedin

squaredbrackets.

Allregressions

includeaconstant.

Dependent

variable:Absolute

logdifference

inthe

onsetof

thefertility

transition.

70


Tab

le3.7:

Cross-section

OLS

:alterna

tive

onsetda

tes

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(Log)Genetic

distan

ce0.0106***

0.004***

0.0121***

0.01***

0.0051***

0.0044***

totheUK

[0.0011]

[0.0014]

[0.0013]

[0.0014]

[0.0014]

[0.0013]

(Log)Genetic

distan

ce0.0064***

totheUSA

[0.0022]

Geograp

hyan

dclim

ate

noyes

nono

yes

yes

yes

History

nono

yes

noyes

yes

yes

Legalo

rigins

nono

noyes

yes

yes

yes

Religion

nono

nono

noyes

yes

R-squ

ared

0.53

0.73

0.58

0.6

0.78

0.8

0.79

Observation

s124

116

114

124

109

108

108

**,*

*,*deno

tesstatisticalsignific

ance

at1%

,5%

and10%

levels,respe

ctively.

Estim

ationwithrobu

ststan

dard

errors

(rep

ortedin

squa

redbrackets).

Allregression

sinclud

eaconstant.Dep

endent

variab

le:(L

og)Onset

ofthefertility

tran

sition

asin

Reher

(2004)

andCoale

andWatkins

(1986).

71


Table 3.8: Cross-section OLS: alternative measures of historical institutions(1) (2) (3) (4)

(Log) Genetic distance to the UK 0.0077*** 0.0079*** 0.0071** 0.0073***[0.0025] [0.0025] [0.0025] [0.0024]

(Log of) Executive constraints in 1850 -0.0007[0.0014]

Polity2 in 1850 0.0001[0.0003]

(Log of) Executive constraints in 1900 -0.0019[0.0017]

Polity2 in 1900 -0.0001[0.0002]

Geography and climate yes yes yes yesHistory yes yes yes yesLegal origins yes yes yes yesR-squared 0.9 0.9 0.91 0.91Observations 36 36 39 39

**, **,* denotes statistical significance at 1% , 5% and 10% levels, respectively. Estimationwith robust standard errors (reported in squared brackets). All regressions include a constant.Dependent variable: (Log) Onset of the fertility transition as in Reher (2004). In columns (3,4)onsets taking place before 1900 are dropped.

to controlling for early formal institutions.18 Consistent with our hypothesis, Table 3.8 shows

that the effect of genetic distance to the UK is still positive and significant in spite of the

considerable drop in the number of observations while these proxies of formal institutions do not

have a significant effect.

3.5 Verification of the mechanism: genetic distance to technolog-

ical frontier, education and fertility transition

In this section we provide evidence supporting the channel of causation we think might be driving

our results. Specifically we show that the impact of cultural relatedness to the technological

frontier on the onset of fertility transition can be attributed mainly to its effect on human capital

18When using measures of formal institutional quality in 1900 we exclude four countries that experienced theonset of the transition before (or in 1900) to avoid reverse causality issues. These are: Hungary, Germany, Swedenand Uruguay.

72


accumulation. This is consistent with the idea that cultural relatedness to the technological

frontier favoured technology adoption (Spolaore and Wacziarg 2009, 2011) which in turn fostered

human capital accumulation and the onset of the fertility transition (Galor 2012). As a first

step we show that genetic distance to the technological frontier is an important determinant of

historical educational attainments. We use average years of education in the population aged

15-64 from Morrisson and Murtin (2009) in the year 1870 to capture historical schooling levels.

As we can notice in Table 3.9, genetic distance to the UK has a negative and significant effect on

schooling in 1870, in line with our reasoning, even after controlling for geography and climate,

legal origins, history and religion. Our strategy to disentangle the mechanism at work goes as

follows. We use genetic distance to the UK as an instrument for schooling levels in 1870 to assess

the impact of historical education levels on the timing of the onset of fertility transitions across

countries. In order to show that we can plausibly argue that genetic distance to the UK affected

the onset of fertility transitions mainly through education - by favouring technology adoption -

we use additional instruments so that we can use an overidentification test to check whether the

instruments are valid. The additional instruments we use are the share of Protestants in 1900

and a country’s (log) distance from Germany. The former is likely to capture heterogeneity in

the incentives to get educated. As Becker and Woessmann (2009) and Ferguson (2011) point

out, adherents to Protestantism had a big incentive to become educated, in order to be able to

read and interpret the Bible by themselves, a crucial element of Protestantism:

"Because of the central importance in Luther’s thought of individual reading of the Bible,

Protestantism encouraged literacy, not to mention printing, and these two things unquestion-

ably encouraged economic development (the accumulation of ’human capital’) as well as sci-

entific study. This proposition holds good not just for countries like Scotland, where spending

on education, school enrolment and literacy rates were exceptionally high, but for the Protes-

tant world as a whole. Wherever Protestant missionaries went, they promoted literacy, with

measurable long-term benefits to the societies they sought to educate; the same cannot be said

of Catholic missionaries throughout the period of the Counter-Reformation to the reforms of

the Second Vatican Council (1962-5)..." (Ferguson 2011, pp. 259)

73


Following this reasoning, we also use as an additional instrument the (log) distance to Germany.

This would capture the heterogeneity in the spread of the Protestant reform which started in

Germany in the early 16th century.19 Table 3.10 displays the results from the first and second

stage of the IV regressions. We run regressions without additional controls in columns (1,2,5,6)

and including the controls used previously, that is geography and climate, history and legal origins

(columns 3,4,7,8). As it can be noticed, the instruments are well correlated with the endogenous

variable and in all cases we cannot reject the null hypothesis that the instruments are valid.

The Hansen-J test checks if genetic distance to the UK and the additional instrument have an

effect on the onset of fertility transitions that goes beyond their effect on initial (historical)

educational attainments. As we cannot reject instruments validity throughout all specifications,

it’s likely that genetic distance (which should capture cultural differences) with respect to the

technological frontier affected the onset of fertility transitions mainly through education and

human capital accumulation. This result is in line with the reasoning that cultural distance to

the technological frontier affected the diffusion of technology and this, consequently, affected the

incentives for human capital accumulation, the quantity-quality trade-off and the timing of the

onset of fertility decline. As we saw above, the estimated coefficients have a strong economic

significance. Using the estimates from the last column of Table 3.10 we can make the following

calculation. Suppose it would have been possible to rise the schooling level of a country like India

in 1870 to the level of Switzerland, keeping everything else constant. In that case, our IV model

predicts that India would have then experienced its fertility transition in 1925, rather than in

1966, forty-one years earlier.

3.6 Conclusion

This paper contributes to our understanding of the main determinants of the fertility transitions

across a large sample of rich and developing countries. We provide evidence suggesting that a

specific type of informal institutions or culture, genetic distance to the technological frontier (the

19The logic is similar to Becker and Woessmann (2009) and Becker et al. (2010) who use distance to Wittenbergas an instrument for education in a cross-county framework in 19th century Prussia.

74


Tab

le3.9:

Cross-section

OLS

:determinan

tsof

historical

scho

olinglevels

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(Log)Genetic

distan

ce-1.009***

-0.4709*

-1.1583**

-0.9417***

-0.7135**

-0.9101***

totheUK

[0.1355]

[0.2393]

[0.1893]

[0.1522]

[0.2741]

[0.2564]

(Log)Genetic

distan

ce-1.592***

totheUSA

[0.5778]

Geograp

hyan

dclim

ate

noyes

nono

yes

yes

yes

History

nono

yes

noyes

yes

yes

Legalo

rigins

nono

noyes

yes

yes

yes

Religion

nono

nono

noyes

yes

R-squ

ared

0.52

0.66

0.59

0.68

0.77

0.85

0.83

Observation

s68

6766

6865

6565

**,*

*,*deno

tesstatisticalsignific

ance

at1%

,5%

and10%

levels,respe

ctively.

Estim

ationwithrobu

ststan

dard

errors

(rep

ortedin

squa

redbrackets).

Allregression

sinclud

eaconstant.Dep

endent

variab

le:averageyearsof

scho

olingin

thepo

pulation

aged

15-64

in1870.

75


Table

3.10:Genetic

distanceto

UK,education

andonset

offertility

transitions.Cross-section

IVDependent

variableSchooling

in1870

(Log)onset

offertility

transition(1)

(2)(3)

(4)(5)

(6)(7)

(8)First

stageSecond

stage

Schoolingin

1870-0.0089***

-0.0096***-0.004***

-0.0034**[0.001]

[0.0012][0.0015]

[0.0014](Log)

Genetic

distance-0.8085***

-0.7186***-0.7616***

-0.4992*to

theUK

[0.1237][0.1914]

[0.2272][0.2885]

Shareprotestants

3.0883***3.4475**

in1900

[0.8173][1.3713]

(Log)Distance

to-0.3973*

-0.8239**Germ

any[0.2039]

[0.3451]Geography

andclim

ateno

noyes

yesno

noyes

yesHistory

nono

yesyes

nono

yesyes

Legaloriginsno

noyes

yesno

noyes

yesFirst

stageF-statistic

60.3827.74

9.317.04

Hansen

p-value0.44

0.660.83

0.44Observations

6463

6160

6463

6160

**,**,*

denotesstatistical

significanceat

1%,5%

and10%


Estim

ationwith

robuststandard

errors(reported

insquared

brackets).Allregressions

includeaconstant.

Schoolingis

averageyears

ofeducation

inthe

populationaged

15-64.The

onsetof

thefertility

transitionis

takenfrom

Reher

(2004):countries

assignedthe

onsetof

thedem

ographictransition

before1870

arenot

includedin

thesam

ple.

76


UK or the US), has been a crucial factor of the timing of the fertility transition across these

economies. We provide evidence that genetic distance to the technological frontier affected the

timing of the onset of the fertility transitions through an indirect channel working through tech-

nology diffusion as suggested in Spolaore and Wacziarg (2009, 2011). A larger genetic distance

to the technological frontier would delay technology adoption and lower the demand for human

capital, consequently leading to a late onset of the fertility transition (Galor 2012). We first

estimate a reduced form regression in which genetic distance to the UK is a strongly significant

variable in explaining the timing of the fertility transitions across countries, even after controlling

for a large set of geographical and historical variables. We show that these estimates are robust

to considering a bilateral analysis that compares pairs of countries in terms of their onsets of the

fertility transitions and their differences in terms of their distance to the UK. Also the result is

robust to using different estimates of the fertility transition dates. Finally, we unveil a plausible

mechanism that may be behind this reduced form. A large cultural difference with respect to the

UK may be proxying a lower technological adoption and less incentives to accumulate human

capital, which in turn may delay the onset of the fertility transition. We test this possible channel

by instrumenting a country’s initial levels of human capital with genetic distance from the UK,

and two measure of the degree of spread of Protestantism, a religion known for its emphasis on

the promotion of human capital among its followers. Our finding that cultural characteristics

matter as triggers of fertility transitions may be seen as a bridge between the literature that

emphasizes the importance of economic determinants of these transitions (e.g. Galor 2012) and

the one that points to purely cultural factors (e.g. Coale and Watkins 1986).

77

Bibliography

[1] Andersen, T. B., Dalgaard, C.-J. and Selaya, P. (2010), "Eye Disease and Development,"

LEPAS Conference Papers, vol. 1: The Mechanics of Aging, Vienna.

[2] Angeles, L. (2010), "Demographic transitions: analyzing the effects of mortality on fertility,"

Journal of Population Economics, 23 (1), pp. 99-120.

[3] Angrist, J. D., Lavy, V. and Schlosser, A. (2005), "New Evidence on the Causal Link between

the Quantity and Quality of Children," NBER working paper 11835. Cambridge, MA.

[4] Anselin, L. (1988), Spatial Econometrics: Methods and Models. Kluwer Acedemic Publishers.

[5] Bailey, A. K. (2009), "How Personal Is the Political? Democratic Revolution and Fertility

Decline," Journal of Family History, 34 (4), pp. 407-425.

[6] Becker, G. S., and Lewis, H. G. (1973), "On the Interaction between the Quantity and

Quality of Children," Journal of Political Economy, 81, pp. 279-288.

[7] Becker, G. S. (1981), A Treatise on the Family. Cambridge and London: Cambridge Uni-

versity Press.

[8] Becker, S. O. and Woessmann, L. (2009), "Was Weber Wrong? A Human Capital Theory of

Protestant Economic History," The Quarterly Journal of Economics, 124 (2), pp. 531-596.

[9] Becker, S. O., Cinnirella, F. and Woessmann, L. (2010), "The Trade-off between Fertility

and Education: Evidence from Before the Demographic Transition," Journal of Economic

Growth, 15(3), pp. 177-204.

3 Bibliography

[10] Becker, S. O., Cinnirella, F. and Woessmann, L. (2012), "The effect of investment in chil-

dren’s education on fertility in 1816 Prussia," Cliometrica, 6, pp. 29-44.

[11] Black, S. E., Devereux, P. J. and Salvanes, K. G. (2005), "The More the Merrier? The Effect

of Family Size and Birth Order on Children’s Education," Quarterly Journal of Economics

120 (2), pp. 669-700.

[12] Bleakly, H., and Lange, F. (2009), "Chronic Disease Burden and the Interaction of Educa-

tion, Fertility, and Growth," Review of Economics and Statistics, 91 (1), pp. 52-65.

[13] Bloom, David E., and Jeffrey G. Williamson (1998), "The Demographic Transition and

Economic Miracles in Emerging Asia", The World Bank Economic Review, 12 (3), pp. 419-

455.

[14] Brown, J. C. and Guinnane, T. W. (2002), "Fertility Transition in a Rural, Catholic Popu-

lation: Bavaria 1880-1910," Population Studies, 56 (1), pp. 35-49.

[15] Brown, J. C. and Guinnane, T. W. (2007), "Regions and Time in the European Fertility

Transition: Problems in the Princeton Project’s Statistical Methodology," Economic History

Review, 60 (3), pp. 574-95.

[16] Coale, A. J. (1973), "The Demographic Transition." in International Population Conference.

vol. 1, International Union for the Scientific Study of Population: Liege, pp. 53-72.

[17] Coale, A. J., and Watkins, S. C. (1986), The Decline of Fertility in Europe. Princeton, NJ:

Princeton University Press.

[18] Delgado, M. (2009), "La fecundidad de las provincias espan̋olas en perspectiva historica,"

Estudios geográficos, 267, pp. 387-442.

[19] Dopico F. and Reher D. S. (1999), "El declive de la mortalidad en Espan̋a, 1860-1930",

Monografias ADEH, 1, Zaragoza, Asociacion de Demografia Historica, 1999.

[20] Eckstein, Z., Mira, P. and Wolpin, K. (1999), "A Quantitative Analysis of Swedish Fertility

Dynamics: 1751-1990," Review of Economic Dynamics, 2, pp. 137-65.

80

3 Bibliography

[21] Ferguson, N. (2011), Civilization, Penguin Books.

[22] Fernihough, A. (2011), "Human Capital and the Quantity-Quality Trade-Off during the

Demographic Transition: New Evidence from Ireland," Working Papers 201113, School Of

Economics, University College Dublin.

[23] Galloway, P.R., Lee, R. D., and Hammel, E. A. (1994), "Fertility decline in Prussia 1875 to

1910: a pooled cross-section time series analysis", Population Studies, 48 (1), pp. 135-158.

[24] Galloway, P.R., Lee, R. D. and Hammel, E. A. (1998), "Infant Mortality and the Fertil-

ity Transition: Macro Evidence from Europe and New Findings from Prussia." Chapter

6 in From Death to Birth: Mortality Decline and Reproductive Change, edited by M. R.

Montgomery, B. Cohen (ed). Washington, D.C: National Academy Press

[25] Gallup, J. L., Sachs, J. D. and Mellinger, A. D. (1999), "Geography and Economic Devel-

opment," CIDHarvard University Working Paper No. 1, March.

[26] Galor, O. (2012), "The Demographic Transition: Causes and Consequences," Cliometrica,

6 (1), pp. 1-28.

[27] Galor, O. and Weil, D. N. (1996), "The Gender Gap, Fertility, and Growth," American

Economic Review, 86 (3), pp. 374-387.

[28] Galor, O. and Weil, D. N. (1999), "From Malthusian Stagnation to Modern Growth," Amer-

ican Economic Review, 89 (2), pp. 150-154.

[29] Galor, O. and Weil, D. N. (2000), "Population, Technology, and Growth: From Malthusian

Stagnation to the Demographic Transition and Beyond," American Economic Review, 90

(4), pp. 806-828.

[30] Galor, O., Moav, O. and Vollrath, D. (2009), "Inequality in Landownership, the Emergence

of Human-Capital Promoting Institutions, and the Great Divergence," Review of Economic

Studies, 76 (1), pp. 143-179.

81

3 Bibliography

[31] Glaeser, E. L., La Porta, R., Lopez-de-Silanes,F. and Shleifer, A. (2004), "Do Institutions

Cause Growth?," Journal of Economic Growth, 9 (3), pp. 271-303.

[32] Goldin, C. (1990), Understanding the Gender Gap: An Economic History of American

Women. Oxford University Press, New York.

[33] Goldstein, J. R. and Klüsener, S. (2010), "Culture revisited: a geographic analysis of fertility

decline in Prussia," MPIDR Working Paper, WP-2010-012.

[34] Guinnane, T. W. (2011), "The Historical Fertility Transition and Theories of Long-Run

Growth: A Guide for Economists," Journal of Economic Literature, 49 (3), pp. 589-614.

[35] Guiso L., Sapienza, P. and Zingales, L. (2006), "Does Culture Affect Economic Outcomes?"

Journal of Economic Perspectives 20 (2), pp. 23-48.

[36] Guzman, R. and Weisdorf, J. L. (2010), "Product Variety and the Demographic Transition",

Economics Letters, 107(1), pp. 74-76.

[37] Herzer, D., Strulik, H. and Vollmer, S. (2012), "The long-run determinants of fertility:

one century of demographic change 1900-1999," Journal of Economic Growth, 17 (4), pp.

357-385.

[38] INE, Instituto Nacional de Estadística (www.ine.es).

[39] Kalemli-Ozcan, S. (2002), "Does the Mortality Decline Promote Economic Growth?", Jour-

nal of Economic Growth 7, pp. 411-439.

[40] Kalemli-Ozcan, S. (2003), "A stochastic model of mortality, fertility, and human capital

investment," Journal of Development Economics, 70, pp. 103-118.

[41] Klemp, M. and Weisdorf, J. L. (2011), "The Child Quantity-Quality Trade-Off during the

Industrial Revolution in England", Discussion Papers 11-16, University of Copenhagen,

Department of Economics.

[42] Lagerlöf, N. (2003), "Gender Equality and Long-Run Growth", Journal of Economic Growth

8 (4), pp. 403-426.

82

3 Bibliography

[43] Leasure, J. W. (1963), "Factors involved in the decline of fertility in Spain, 1900-1950",

Population Studies, 16 (3), pp. 271-285.

[44] Lehr, C. S. (2009), "Evidence on the Demographic Transition." Review of Economics and

Statistics 91 (4), pp. 871-887.

[45] Marshall, M. G. and Jaggers, K. (2008), "Polity IV project. Political Regime Characteristics

and Transitions, 1800-2002," Integrated Network for Societal Conflict Research (INSCR),

Program Center for International Development and Conflict Management (CIDCM), Uni-

versity of Maryland.

[46] Mas Ivars, M. and Cucarella Tormo, V. (2009), "Series históricas de capital público en

Espan̋a y su distribución territorial (1900-2005)", Fundación BBVA.

[47] Mitchell, B. R. (2007), "International historical statistics Europe 1750-2005", Palgrave

Macmillan.

[48] Morrisson, C. and Murtin, F. (2007), "Education Inequalities: a Global Perspective Since

1870", PSE Working Paper.

[49] Morrisson, C. and Murtin, F. ( 2009), "The Century of Education." Journal of Human

Capital 3(1): pp. 1-42.

[50] Murphy, T. E. (2010), "Old Habits Die Hard (Sometimes): What can Département hetero-

geneity tell us about the French fertility decline?", IGIER working paper No. 364.

[51] Murtin, F. (forthcoming), "Long-term Determinants of the Demographic Transition, 1870-

2000," Review of Economics and Statistics.

[52] Nunn, N. and Puga, D. (2012), "Ruggedness: the Blessing of Bad Geography in Africa,"

Review of Economics and Statistics, 94 (1), pp. 20-36.

[53] Núñez, C.E. (2005a), "Educación". In Carreras, A. y Tafunell, X. (coords.), "Estadísticas

históricas de Espan̋a, siglos XIX y XX". Fundación BBVA, Bilbao, Tomo 1, pp. 155-245.

83

3 Bibliography

[54] Núñez, C.E. (2005b), "A Modern Human Capital Stock. Spain in the Nineteenth and Twen-

tieth Centuries," in Magnus Jerneck, Magnus MÃűrner, Gabriel Tortella y Sune Akerman

editors, "Different Paths to Modernity. A Nordic and Spanish Perspective". Lund: Nordic

Academic Press. pp. 122-142.

[55] Putterman, L. (2006), Agricultural Transition Dataset, Brown University.

[56] Qian, N. (2009), "Quantity-Quality and the One Child Policy: The Only-Child Disadvantage

in School Enrolment in Rural China." NBER Working Paper 14973. Cambridge, MA.

[57] Reher, D. S. (2004), "The Demographic Transition Revisited as a Global Process," Popula-

tion Space and Place, 10, pp. 19-42.

[58] Reher, D. S. and Iriso-Napal, P.L. (1989), "Marital Fertility and its Determinants in Rural

and in Urban Spain, 1887-1930," Population Studies, 43 (3), pp. 405-427.

[59] Rosenzweig, M. R., and Wolpin, K. I. (1980), "Testing the Quantity-Quality Fertility Model:

The Use of Twins as a Natural Experiment," Econometrica, 48 (1), pp. 227-240.

[60] Sachs, Jeffrey, Kiszewski, A., Mellinger, A., Spielman, A., Malaney, P., and Ehrlich, S.

(2004), "A Global Index of the Stability of Malaria Transmission," American Journal of

Tropical Medicine and Hygiene 70 (5), pp. 486-498.

[61] Sah, R. K. (1991), "The Effects of Child Mortality Changes on Fertility Choice and Parental

Welfare," Journal of Political Economy, 99 (3), pp. 582-606.

[62] Schultz, T. P. (2008), "Population Policies, Fertility, Women’s Human Capital, and Child

Quality." In T. Paul Schultz, John Strauss (eds.), Handbook of Development Economics, vol.

4: 3249-3303. Amsterdam: Elsevier.

[63] Silvestre, J. (2007), "Temporary Internal Migrations in Spain, 1860-1930," Social Science

History, 31 (4), pp. 539-574.

[64] Spolaore, E. and Wacziarg, R. (2009), "The Diffusion of Development," Quarterly Journal

of Economics 124 (2), pp. 469-529.

84

3 Bibliography

[65] Spolaore, E., and R. Wacziarg (2011), "Long Term Barriers to the International Diffusion

of Innovations." in Jeffrey Frankel and Christopher Pissarides, eds., NBER International

Seminar on Macroeconomics 2011, Chapter 1, pp. 11-46. Chicago: University of Chicago

Press, May.

[66] Tolnay, S. E. (1995), "The Spatial Diffusion of Fertility: A Cross-Sectional Analysis of

Counties in the American South, 1940," American Sociological Review, 60, pp. 299-308.

[67] van de Walle, F. (1980), "Education and the Demographic Transition in Switzerland," Pop-

ulation and Development Review, 6 (3), pp. 463-472.

[68] van de Walle, F. (1986), "Infant Mortality and the European Demographic Transition."

in The Decline of Fertility in Europe, Coale, A. J., and S. C. Watkins (eds). Princeton

University Press: Princeton: pp. 201-233.

85

essays in comparative economic development alberto...

Documents