Rol de la mujer

THE

QUARTERLY JOURNALOF ECONOMICS

Vol. 128 May 2013 Issue 2

ON THE ORIGINS OF GENDER ROLES: WOMEN AND THEPLOUGH*

Alberto Alesina

Paola Giuliano

Nathan Nunn

The study examines the historical origins of existing cross-cultural differ-ences in beliefs and values regarding the appropriate role of women in society.We test the hypothesis that traditional agricultural practices influenced thehistorical gender division of labor and the evolution of gender norms. We findthat, consistent with existing hypotheses, the descendants of societies that trad-itionally practiced plough agriculture today have less equal gender norms, mea-sured using reported gender-role attitudes and female participation in theworkplace, politics, and entrepreneurial activities. Our results hold lookingacross countries, across districts within countries, and across ethnicitieswithin districts. To test for the importance of cultural persistence, we examinethe children of immigrants living in Europe and the United States. We find thateven among these individuals, all born and raised in the same country, thosewith a heritage of traditional plough use exhibit less equal beliefs about genderroles today. JEL Codes: D03, J16, N30.

* We thank Larry Katz, Elhanan Helpman, and five anonymous referees forcomments that substantially improved the paper. We also thank Samuel Bowles,David Clingingsmith, Matthias Doepke, Esther Duflo, Raquel Fernandez, NicoleFortin, Oded Galor, Claudia Goldin, Pauline Grosjean, Judith Hellerstein, VivianHoffman, Edward Miguel, Rohini Pande, Louis Putterman, John Wallis, as wellas participants at various conferences and seminars. We also thank Eva Ng,Matthias Baeuml, Ellora Derenoncourt, Lauren Morris, and Benjamin Schoeferfor excellent research assistance. Paola Giuliano gratefully acknowledges supportfrom the UCLA Senate and thanks the Russell Sage Foundation for its wonderfulhospitality.

! The Author(s) 2013. Published by Oxford University Press, on behalf of President andFellows of Harvard College. All rights reserved. For Permissions, please email: [email protected] Quarterly Journal of Economics (2013), 469–530. doi:10.1093/qje/qjt005.Advance Access publication on February 19, 2013.

469

at Harvard U

niversity Library on A

pril 24, 2013http://qje.oxfordjournals.org/

Dow

nloaded from

http://qje.oxfordjournals.org/

I. INTRODUCTION

This study examines an important deeply held belief thatvaries widely across societies: the appropriate or natural role ofwomen in society. In some societies, the dominant belief is thatwomen should be allowed to participate freely, and equally tomales, in employment outside the home. In others, there is thevery different view that the appropriate place for women is withinthe home, and they are discouraged from participating in activ-ities outside the domestic sphere. These differences can be mostclearly seen in surveys that report attitudes about gender roles.For example, the proportion of respondents in the World ValuesSurvey that ‘‘agree’’ with the statement that ‘‘when jobs arescarce, men should have more right to a job than women’’varies widely across countries, ranging from 3.6% (in Iceland)to 99.6% (in Egypt).1

Our interest is in explaining the origins of these culturaldifferences. Specifically, we test the hypothesis, put forth byEster Boserup (1970), that differences in gender roles havetheir origins in the form of agriculture traditionally practiced inthe pre-industrial period. Boserup identifies important differ-ences between shifting cultivation and plough cultivation.Shifting cultivation is labor intensive and uses handheld toolslike the hoe and the digging stick. Plough cultivation, by contrast,is much more capital intensive, using the plough to prepare thesoil. Unlike the hoe or digging stick, the plough requires signifi-cant upper body strength, grip strength, and bursts of power,which are needed to either pull the plough or control theanimal that pulls it. Because of these requirements, whenplough agriculture is practiced, men have an advantage in farm-ing relative to women.

Given the important role of soil preparation in agriculture,which accounts for about one-third of the total time spent inagricultural tasks, societies that traditionally practiced ploughagriculture—rather than shifting cultivation—developed a spe-cialization of production along gender lines. Men tended to workoutside the home in the fields, while women specialized in

1. The figures are based on the 4th wave of the World Values Survey, whichincludes information from 62 countries surveyed between 1999 and 2004. Objectiveoutcomes, like female labor force participation (FLFP), also exhibit significant vari-ation (Antecol 2000; Fortin 2005; Fernandez 2007; Fernandez and Fogli 2009). In2000, the FLFP rate ranged from 16.1% (Pakistan) to 90.5% (Burundi).

QUARTERLY JOURNAL OF ECONOMICS470

at Harvard U



Dow

nloaded from


activities within the home. This division of labor then generatednorms about the appropriate role of women in society. Societiescharacterized by plough agriculture, and the resulting gender-based division of labor, developed the belief that the naturalplace for women is within the home. These cultural beliefs tendto persist even if the economy moves out of agriculture, affectingthe participation of women in activities performed outside thehome, such as market employment, entrepreneurship, or partici-pation in politics.

To test Boserup’s hypothesis, we combine pre-industrialethnographic data, reporting whether societies traditionallypracticed plough agriculture, with contemporary measures of in-dividuals’ views about gender roles, as well as measures of femaleparticipation in activities outside the home. Our analysis exam-ines variation across countries, ethnic groups, and individuals.Consistent with Boserup’s hypothesis, we find a strong androbust positive relationship between historical plough use andunequal gender roles today. Traditional plough use is positivelycorrelated with attitudes reflecting gender inequality and nega-tively correlated with female labor force participation, femalefirm ownership, and female participation in politics.2

Although these findings support Boserup’s hypothesis, theyare also consistent with other interpretations. For example, wewould observe the same relationships if societies with attitudesfavoring gender inequality were more likely to adopt the ploughhistorically, and these attitudes persist today. To better under-stand whether traditional plough use has a causal impact on sub-sequent cultural norms, we control for an exhaustive set ofobservable characteristics.

In our baseline set of covariates, we include controls for anumber of historical characteristics of ethnic groups, such asthe suitability of their environment for agriculture, whetherthey had domesticated animals, the extent to which they livedin tropical climates, their level of political development, andtheir level of economic development. We also flexibly control forcurrent country-level per capita GDP. We show that the results

2. Our analysis is concerned with a very specific aspect of gender equality:whether it is believed to be appropriate for women to work outside the home as itis for men. There are many other gender differences that contribute to gender in-equality more broadly defined. However, throughout the paper when we refer togender equality (or inequality), we are referring specifically to gender differences inemployment outside the home.

ORIGINS OF GENDER ROLES 471

at Harvard U



Dow

nloaded from


are robust to controlling for additional historical characteristicsof ethnic groups and contemporary characteristics of countriesthat are potentially correlated with traditional plough use andbeliefs about gender roles today.

Motivated by the possibility that our cross-national resultsmay be biased by other characteristics of countries, we move to amore micro-level analysis, examining cross-individual variationin female labor force participation and beliefs about gender equal-ity. With this strategy, we are able to restrict the analysis tovariation within countries and even variation within districtswithin countries. We find similar impacts when we examinethis finer variation.

We then turn to an analysis of the importance of ancestraldifferences in specific geo-climatic characteristics that impactedwhat types of crops could be cultivated in the historical locations.This analysis is motivated by Pryor (1985), who argues that thebenefit of using the plough differed depending on the types ofcrops cultivated. The plough is more beneficial for crops that re-quire large tracts of land to be prepared in a short period of time(e.g., due to multiple-cropping), and that can be grown only insoils that are not shallow, not sloped, and not rocky.3 Thesecrops, which Pryor refers to as ‘‘plough-positive’’, include teff,wheat, barley, rye, and wet rice. These can be contrasted to‘‘plough-negative’’ crops, such as maize, sorghum, millet, andvarious types of root and tree crops, which require less land tobe prepared over a longer period of time, and/or can be cultivatedon thin, sloped, or rocky soils, where using the plough is difficult.Unlike plough-positive crops, plough-negative crops benefit muchless from the adoption of the plough.

Using data from the Food and Agriculture Organization(FAO), we identify the geo-climatic suitability of finely definedlocations for growing plough-positive cereals (wheat, barley,and rye) and plough-negative cereals (sorghum, foxtail millet,and pearl millet). Except for their benefit from plough use, thetwo sets of cereals are otherwise similar. Both have been culti-vated in the Eastern Hemisphere since the Neolithic revolution;

3. For a recent study documenting the link between soil type and plough use inmodern India, see Carranza (2010). In particular, she shows that in contemporaryIndia, plough technology is more likely to be adopted with deep loamy soils ratherthan shallow clay soils, and that it is associated with less participation of women inagriculture.


at Harvard U



Dow

nloaded from


require similar preparations for consumption, all being used forflour, porridge, bread, or in beverages; and produce similar yieldsand therefore are able to support similar population densities.

We show that, consistent with Pryor (1985), the suitability ofa location for cultivating plough-positive vs. plough-negativecrops predicts whether the plough was adopted. We also showthat they predict gender norms today. Finally, we use ethnicgroups’ geo-climatic conditions for growing plough-positive andplough-negative cereals as instruments for historical plough use.Motivated by concerns that our instruments may be correlatedwith other geographic characteristics, we check that the esti-mates are robust to flexibly controlling for geographic covariates,including overall agricultural suitability, temperature, precipita-tion, soil depth, and slope. We find that the IV procedure gener-ates estimates that are consistent with the OLS estimates.

Our analysis then turns to underlying mechanisms. It is pos-sible that the long-term impact of the plough reflects persistentcultural beliefs. However, it is also possible that part of thelong-term impact arises because historical plough use promotedthe development of institutions, policies, and markets that areless conducive to the participation of women in activities outsidethe home. To distinguish these two channels, we exploit the factthat cultural norms and beliefs—unlike institutions, policies, andmarkets—are internal to the individual. Therefore, when individ-uals move, their beliefs and values move with them, but theirexternal environment remains behind. Exploiting this fact, weexamine variation in cultural heritage among children of immi-grants living in either the United States or Europe. All individ-uals born and raised abroad have been exposed to the sameinstitutions and markets. In effect, the analysis holds externalfactors constant, while examining variation in individuals’ in-ternal beliefs and values. We find that individuals from culturesthat historically used the plough have less equal gender normsand that women from cultures that used the plough participateless in the workforce. These results provide evidence that part ofthe importance of the plough arises through its impact on in-ternal beliefs and values.

Our findings contribute to a deeper understanding of the ori-gins of cultural norms and beliefs. Studies have documented thecontinuity of cultural norms over remarkably long periods of time(e.g., Voigtlander and Voth 2012). Others show that historicalfactors influence the evolution of culture over time by affecting


at Harvard U



Dow

nloaded from


the relative costs and benefits of different cultural traits. Guiso etal. (2008) provide evidence that the formation of medieval com-munes had a long-term impact on the level of social capital withinNorthern Italy. Similarly, Becker et al. (2010) and Grosjean(2011b) provide evidence of historical state boundaries havinglasting cultural impacts. Nunn and Wantchekon (2011) showthat Africa’s slave trade generated a culture of distrust that con-tinues to persist today. Nisbett and Cohen (1996) and Grosjean(2011a) show that the culture of honor in the US South has itsorigins in a tradition of herding among the Scots-Irish. The find-ings of our paper add to this line of enquiry by providing add-itional evidence that shows that historical factors—namely,differences in traditional farming practices—have shaped theevolution of norms and beliefs about the appropriate role ofwomen in society.

Our focus on a historical determinant of gender roles is notmeant to imply that other short-run factors are unimportant. Anumber of existing studies have shown the importance of deter-minants like economic development, medical improvements,technological change, and the production structure of the econ-omy (e.g., Goldin 2006; Ross 2008; Albanesi and Olivetti 2007;2009; Doepke and Tertilt 2009; Iversen and Rosenbluth 2010).As we show, accounting for these important factors, there re-mains a strong persistent impact of traditional plough use ongender norms today.

The paper is organized as follows. We begin, in Section II, bydescribing the conceptual framework underlying the hypothesistested in the paper. Section III then documents that in societiesthat traditionally used plough agriculture, women did in fact par-ticipate less in farmwork and other activities outside domesticsphere. In Section IV, we then explain the procedure used tolink traditional plough use, which is measured at the ethnicitylevel, to contemporary data on gender norms and female partici-pation outside the home. Sections V and VI report OLS estimatesof the relationship between traditional plough use and genderoutcomes today, examining variation across individuals andcountries. Section VII turns to the issue of causality, reportingOLS estimates that control for an extensive set of observablecharacteristics, as well as the IV estimates. In Section VIII, wethen turn to mechanisms, using the children of immigrants in theUnited States and Europe to test for cultural transmission as apotential channel. Section IX offers concluding thoughts.


at Harvard U



Dow

nloaded from


II. CONCEPTUAL FRAMEWORK

The possibility that modern gender roles have their origins inthe form of agriculture practiced traditionally has long beenrecognized. Baumann’s (1928) early study examined the relation-ship between matriarchy and the use of the hoe in Africa. EsterBoserup (1970) further developed the observation, identifying im-portant differences between shifting cultivation and plough cul-tivation. Unlike shifting cultivation, which is labor intensive anduses handheld tools like the hoe and digging stick, plough agri-culture requires significant upper body strength, needed to eitherpull the plough or control the animal that pulls it. Because ofthese physical requirements, when plough agriculture is prac-ticed, men have an advantage relative to women. This advantageis also reinforced by the fact that when the plough is used, there isless need for weeding, a task typically undertaken by women andchildren. In addition, child care, a task almost universally per-formed by women, is most compatible with activities that can bestopped and resumed easily and that do not put children indanger; characteristics that are satisfied for hoe agriculture,but not for plough agriculture, especially when animals areused to pull the plough.

Due to the importance of soil preparation, societies that trad-itionally practiced plough agriculture—rather than shifting hoecultivation—tended to develop a specialization of productionalong gender lines. Men tended to work outside the home in thefields, while women specialized in activities within the home.This division of labor then generated norms about the appropri-ate role of women in society. Societies characterized by ploughagriculture, and the resulting gender-based division of labor, de-veloped a cultural belief that the natural place for women iswithin the home.

We view cultural beliefs as decision-making heuristics or‘‘rules-of-thumb’’ that are employed in uncertain or complex en-vironments. Using theoretical models, Boyd and Richerson (1985)show that if information acquisition is either costly or imperfect,it can be optimal for individuals to develop heuristics orrules-of-thumb in decision making. By relying on general beliefsabout the right thing to do in different situations, individuals maynot behave in a manner that is precisely optimal in every in-stance, but they save on the costs of obtaining the informationnecessary to always behave optimally. It is increasingly


at Harvard U



Dow

nloaded from


understood that many important aspects of human behavior areguided by decision-making heuristics, which manifest themselvesas values, beliefs, or gut feelings about the appropriate action incertain situations (Gigerenzer 2007; Kahneman 2011). In prac-tice, these heuristics often take the form of deeply held traditionalvalues or religious beliefs.

In standard models of cultural evolution (e.g., Boyd andRicherson 1985), the distribution of cultural beliefs in the popu-lation evolves through a natural-selection-like process deter-mined by relative payoffs. Within this framework, Boserup’shypothesis suggests that in societies that engaged in plough agri-culture, cultural beliefs about gender inequality were relativelybeneficial. Therefore, these norms may have evolved inplough-agriculture societies but not hoe-agriculture societies.Because of the persistent nature of cultural beliefs, norms ofgender inequality may persist even after the economy movesout of agriculture or industrializes, affecting the participation ofwomen in activities performed outside the home, such as marketemployment, entrepreneurship, or participation in politics.

There are a number of reasons why we may observe persist-ence. First, the underlying cultural traits may be reinforced bypolicies, laws, and institutions, which affect the benefit of beliefsabout gender inequality. A society with traditional beliefs aboutgender inequality may perpetuate these beliefs by institutionaliz-ing unequal property rights, voting rights, parental leavepolicies, etc. Another source of persistence can arise from a com-plementarity between cultural beliefs and industrial structure.Beliefs of gender inequality may cause a society to specialize inthe production of capital- or brawn-intensive industries, which inturn decreases the relative costs of norms about gender inequal-ity, thereby perpetuating this trait.

A third explanation that does not rely on these forms of com-plementarity is that cultural beliefs, by definition, are inherentlysticky. The benefit of decision-making rules-of-thumb is that theycan be applied widely in a number of environments, saving on theneed to acquire and process information with each decision. Thefocus of our analysis is on testing whether there is any evidencefor the persistence of cultural beliefs independent of external fac-tors that generate complementarities that cause persistence. Oursubnational analysis, with either country fixed effects or subna-tional district fixed effects, speaks to this by accounting for exter-nal factors such as industrial structure and domestic policies and


at Harvard U



Dow

nloaded from


institutions. More directly, our analysis of the children of immi-grants living in the United States and Europe tests for a relation-ship between traditional plough agriculture and cultural beliefswhile holding constant the external environment.

At first blush, some aspects of Boserup’s hypothesis likelyhave intuitive appeal, while others may not. If one is seeking toexplain differences in female participation in activities outsidethe home, then it is natural to examine the activity that histor-ically has been the most important, namely, agriculture. Withinagriculture, the two traditional forms are slash-and-burn hoeagriculture and intensive plough agriculture. It is possible thatthis important distinction had long-term impacts.

One aspect of Boserup’s theory that may initially appear un-appealing is that Western societies, namely Europe and its off-shoots, have long practiced plough agriculture, and thesesocieties also have extremely equal gender roles today. This ob-servation seems to directly contradict Boserup’s hypothesis thatlinks a tradition of plough agriculture to unequal gender roles.However, the presumption that Europe and its offshoots are themost gender equal today is to a large extent a slightly misplacedWestern-centric view of gender norms. Although it is true thatprogress has been made due to changes in the second half of thetwentieth century, even today, Europe and its offshoots, althoughabove average, are not among the most gender-equal countries.For example, the share of national political positions held bywomen in the United States in 2000 was 13%, which places it50th out of the 156 countries for which data are available. Thefemale labor force participation rate in the United States in thesame year was 59.5, ranking it 47th of the 181 with availabledata. In addition, these statistics appear even less equal whenone takes into account the high per capita income of Westernnations.

III. THE HISTORICAL IMPACTS OF TRADITIONAL PLOUGH

AGRICULTURE

We begin by first confirming that societies that traditionallyused plough agriculture had lower female participation in agri-cultural activities. We also check whether plough use was asso-ciated with differences in other activities within and outside thedomestic sphere.


at Harvard U



Dow

nloaded from


Our analysis uses information on pre-industrial plough usefrom the Ethnographic Atlas, a worldwide ethnicity-level data-base constructed by George Peter Murdock that contains ethno-graphic information for 1,265 ethnic groups. Information forsocieties in the sample has been coded for the earliest period forwhich satisfactory ethnographic data are available or can be re-constructed. The earliest observation dates are for groups in theOld World where early written evidence is available. For theparts of the world without a written history, the information isfrom the earliest observers of these cultures. For some cultures,the first recorded information is from the 20th century, but evenfor these cultures, the data capture as much as possible the char-acteristics of the ethnic group prior to European contact. For allgroups in the data set, the variables are taken from the societiesprior to industrialization.4

The database provides information on whether societies trad-itionally used the plough. Ethnicities are classified into one of threemutually exclusive categories: (1) the plough was absent, (2) theplough existed at the time the group was observed, but it was notaboriginal, and (3) the plough was aboriginal and found in the so-ciety prior to contact.5 There is no evidence of groups switchingfrom one form of agriculture to another and then back again. Inother words, the use of the plough, once adopted, remains stableover time. Using this categorization, we construct an indicator vari-able for traditional plough agriculture that equals one if the ploughwas present (whether aboriginal or not) and zero otherwise.6

It is possible that the plough had a larger impact on gendernorms if it was adopted early. However, because of data

4. In total, 23 ethnicities are observed during the 17th century or earlier, 16during the 18th century, 310 during the 19th century, 876 between 1900 and 1950,and 31 after 1950. For nine ethnicities, an exact year is not provided. AppendixTable A2 (see the supplementary material online) reports the number of the ethni-cities present in the Ethnographic Atlas by continent, and Figure A1 maps theirlocations. The majority of the ethnicities sampled are in Africa, followed by NorthAmerica and then Asia. Fewer observations are present within the European con-tinent, in part, because researchers felt that analysis of European cultures washistory rather than ethnography (since written records existed for these cultures).

5. Of the 1,156 ethnicities for which information exists, for 997 the plough wasabsent, for 141 the plough was adopted and aboriginal, and for 18 it was adopted,but following European contact.

6. All of the results reported in the paper are robust to an alternative definitionof traditional plough use that does not include the 18 ethnic groups that adopted theplough after European contact. See Appendix Table A7.


at Harvard U



Dow

nloaded from

http://hwmaint.qje.oxfordjournals.org/cgi/content/full/doi:10.1093/qje/qjt005/-/DC1


limitations, we are unable to test for this. From the database, weknow the approximate date of adoption only if it occurred afterEuropean contact. For the others, we do not have any informationon the timing of adoption. Given this, our estimates should be in-terpreted as the average effect of having adopted the ploughamong all ethnic groups that did so prior to industrialization.There may be heterogeneity within the group of adopters (e.g.,based on date of adoption), but we are able to estimate only anaverage effect.

We measure traditional female participation in agricultureusing information on the gender-based division of labor in agri-culture reported in the Ethnographic Atlas. Ethnicities aregrouped into one of the following five categories measuringrelative participation in agriculture by gender: (1) males only,(2) males appreciably more, (3) equal participation, (4) femalesappreciably more, and (5) females only.7 Using this information,we construct a variable that takes on integer values ranging from1 to 5 and increases with female specialization in agriculture.

When examining the relationship between the gender-baseddivision of labor in agriculture and plough use, we control for anumber of characteristics of ethnic groups that may be correlatedwith plough use and gender roles. We control for the presence oflarge domesticated animals, a measure of economic development,and a measure of political complexity, all taken from the Ethno-graphic Atlas. The presence of domesticated animals is measuredwith an indicator variable that equals one if large domestic ani-mals were present in the society. Economic development is mea-sured using the density of ethnic groups’ settlements. Ethnicitiesare grouped into the following categories: (1) nomadic or fullymigratory, (2) semi-nomadic, (3) semi-sedentary, (4) compactbut not permanent settlements, (5) neighborhoods of dispersedfamily homesteads, (6) separate hamlets forming a single com-munity, (7) compact and relatively permanent settlements, and(8) complex settlements. With this information, we construct avariable that takes on integer values, ranging from 1 to 8 and

7. The original classification in the Ethnographic Atlas distinguishes ‘‘differ-entiated but equal participation’’ from ‘‘equal participation’’. Since this distinctionis not relevant for our purposes, we combine the two categories into a single cat-egory of equal participation. In addition, for 232 ethnic groups, agriculture was notpracticed and therefore there is no measure of female participation in agriculture.For an additional 315 ethnic groups, information for the variable is missing. Theseethnic groups (547 in total) are omitted from the analysis.


at Harvard U



Dow

nloaded from


increasing with settlement density. Political complexity is mea-sured by the levels of jurisdictional hierarchies in the society.8

We also control for two measures of the geographic conditionsof the traditional location of each ethnic group. For each ethni-city, we know the geographic coordinates of the centroid of thegroup historically. Using this information, we calculate the frac-tion of land within a 200-kilometer radius of the centroid that isdefined as suitable for the cultivation of crops. The crop suitabil-ity data are from the FAO’s Global Agro-Ecological Zones (GAEZ)2002 database (Fischer et al. 2002), which reports suitabilitymeasures for 5 arc-minute by 5 arc-minute (approximately56 km by 56 km) grid cells globally. We also use the same proced-ure to control for the proportion of land within a 200-kilometerradius that is defined as being tropical or subtropical.

OLS estimates examining the historical relationship be-tween traditional plough use and female participation in agricul-ture are reported in column 1 of Table I. The specificationreported includes the five control variables. The estimates iden-tify a negative relationship between plough use and participationof women in agriculture. The use of the plough is associated witha reduction in the female participation in agriculture variable of0.88, which is large given that its mean is 3.0 and its standarddeviation is 1.0.9 This is consistent with the analysis of Boserup(1970), as well as the observations of anthropologists like Bau-mann (1928) and Whyte (1978).10

A natural question that arises is the exact nature of this de-cline in female participation in agriculture; specifically, whetherthe decline is in all agricultural tasks or if it is focused in only afew (such as soil preparation). Unfortunately, the EthnographicAtlas does not provide information on specific tasks within agri-culture. We therefore complement our analysis by using Murdockand White’s (1969) Standard Cross-Cultural Sample (SCCS),which does contain this information. The SCCS contains ethno-graphic information on 186 societies, intentionally chosen to be

8. As we report in Appendix Table A13, traditional plough use is positivelyassociated with all three variables, and most robustly with political complexity.

9. Descriptive statistics are reported in Table A1 of the appendix.10. Ideally, we would also like to observe the absolute participation of women in

agriculture, not just their participation relative to men. Unfortunately, such ethno-graphic data are not widely available. In Section V, where we examine contempor-ary outcomes, the measures of female participation in the labor force, firmownership, and politics are absolute measures and are not relative to men.


at Harvard U



Dow

nloaded from


TA

BL

EI

TR

AD

ITIO

NA

LP

LO

UG

HU

SE

AN

DF

EM

AL

EP

AR

TIC

IPA

TIO

NIN

PR

E-I

ND

US

TR

IAL

AG

RIC

UL

TU

RE

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Dep

end

ent

vari

able

:T

rad

itio

nal

part

icip

ati

onof

fem

ale

sre

lati

ve

tom

ale

sin

the

foll

owin

gta

sks:

Over

all

agri

cult

ure

Lan

dcl

eara

nce

Soi

lp

rep

ara

tion

Pla

nti

ng

Cro

pte

nd

ing

Harv

esti

ng

Mea

nof

dep

.var.

3.0

42.8

31.4

52.1

52.8

63.1

63.2

3

Tra

dit

ion

al

plo

ugh

agri

cult

ure�

0.8

83**

*�

1.1

36**

*�

0.4

34**

�1.1

82**

*�

1.2

90**

*�

1.1

88**

*�

0.9

54**

*(0

.225)

(0.2

40)

(0.1

97)

(0.3

20)

(0.3

06)

(0.3

51)

(0.2

71)

Eth

nog

rap

hic

con

trol

syes

yes

yes

yes

yes

yes

yes

Obse

rvati

ons

660

124

129

124

131

122

131

Ad

just

edR

-squ

are

d0.1

30.1

90.1

40.1

00.0

90.1

30.1

6R

-squ

are

d0.1

40.2

30.1

80.1

40.1

30.1

80.2

0

Not

es.

Th

eu

nit

ofob

serv

ati

onis

an

eth

nic

gro

up

.In

colu

mn

1,

eth

nic

gro

up

sare

from

the

Eth

nog

rap

hic

Atl

as,

an

din

colu

mn

s2–7,

they

are

from

the

Sta

nd

ard

Cro

ss-C

ult

ura

lS

am

ple

.T

he

dep

end

ent

vari

able

mea

sure

str

ad

itio

nal

fem

ale

part

icip

ati

onin

ap

art

icu

lar

agri

cult

ura

lact

ivit

yin

the

pre

-in

du

stri

al

per

iod

.T

he

vari

able

sta

ke

onin

teger

valu

esbet

wee

n1

an

d5

an

dare

incr

easi

ng

infe

male

part

icip

ati

on.

‘‘Tra

dit

ion

al

plo

ugh

use

’’is

an

ind

icato

rvari

able

that

equ

als

one

ifth

ep

lou

gh

was

trad

itio

nall

yu

sed

inp

re-i

nd

ust

rial

agri

cult

ure

.F

orth

eE

thn

ogra

ph

icA

tla

s,th

em

ean

(an

dst

an

dard

dev

iati

on)

ofth

etr

ad

itio

nal

plo

ugh

agri

cult

ure

vari

able

is0.1

86

(0.3

90),

an

dfo

rth

eS

CC

Sit

is0.2

34

(0.4

25);

thes

eco

rres

pon

dto

the

sam

ple

sfr

omco

lum

ns

1an

d2,

resp

ecti

vel

y.

Th

esa

me

stati

stic

sfo

rth

eot

her

colu

mn

sare

slig

htl

yd

iffe

ren

t.‘‘E

thn

ogra

ph

icco

ntr

ols’

’in

clu

de:

the

suit

abil

ity

ofth

elo

cal

envir

onm

ent

for

agri

cult

ure

,th

ep

rese

nce

ofla

rge

dom

esti

cate

dan

imals

,th

ep

rop

orti

onof

the

loca

len

vir

onm

ent

that

istr

opic

al

orsu

btr

opic

al,

an

ind

exof

sett

lem

ent

den

sity

,an

dan

ind

exof

pol

itic

al

dev

elop

men

t.F

iner

det

ail

sabou

tvari

able

con

stru

ctio

nare

pro

vid

edin

the

text

an

dap

pen

dix

.C

oeffi

cien

tsare

rep

orte

dw

ith

robu

stst

an

dard

erro

rsin

bra

cket

s.C

olu

mn

1re

por

tsC

onle

yst

an

dard

erro

rsad

just

edfo

rsp

ati

al

corr

elati

on.

***,

**,

an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


representative of the full sample, and to be historically and cul-turally independent from the other ethnic groups in the sample.The database was constructed by first grouping the 1,265 socie-ties from the Ethnographic Atlas into 186 clusters of closelyrelated cultures. A particularly well-documented and representa-tive ethnic group was then chosen for each cluster, and theseconstitute the observations in the SCCS.11

Using the SCCS data, we first replicate the regression re-ported in column 1. As shown in column 2, we find a similarresult: plough use is associated with a decline in female partici-pation in agriculture of 1.14, which is roughly equal to aone-standard-deviation change in the dependent variable. In col-umns 3–7, we estimate the association between plough use andfemale participation in the following agricultural tasks: landclearance, soil preparation, planting, crop tending, and harvest-ing. The estimates show that plough use is associated with lessfemale participation in all agricultural tasks, with the largestdeclines in soil preparation, planting, and crop tending.

In columns 1–9 of Table II, we consider the relationship be-tween plough use and female participation in the followingnon-agricultural activities: caring for small animals, caring forlarge animals, milking, cooking, fuel gathering, water fetching,burden carrying, handicraft production, and trading.12 We findthat the plough tends not to be significantly correlated withfemale participation in other activities. The exception is thatplough use is associated with significantly less female participationin burden carrying. We find some evidence of more female partici-pation in caring for large animals, caring for small animals, andmilking, although none of the coefficients is statistically significant.

These results are consistent with women working less insocieties that traditionally used the plough. This interpretationof the correlations is fully consistent with Boserup’s hypothesis.What is important for her argument is that when plough

11. Table A2 in the appendix reports the distribution of ethnicities in theEthnographic Atlas and SCCS by continent, while Table A3 lists the names of allethnic groups in the SCCS.

12. If an activity is not present in a society, then the dependent variable is codedas missing. This is the reason for the different number of observations in eachregression.


at Harvard U



Dow

nloaded from


TA

BL

EII

TR

AD

ITIO

NA

LP

LO

UG

HU

SE

AN

DT

RA

DIT

ION

AL

FE

MA

LE

PA

RT

ICIP

AT

ION

OU

TS

IDE

OF

AG

RIC

UL

TU

RE

INT

HE

PR

E-I

ND

US

TR

IAL

PE

RIO

D

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

:T

rad

itio

nal

part

icip

ati

onof

fem

ale

sre

lati

ve

tom

ale

sin

the

foll

owin

gta

sks:

Cari

ng

for

small

an

imals

Cari

ng

for

larg

ean

imals

Mil

kin

gC

ook

ing

Fu

elgath

erin

gW

ate

rfe

tch

ing

Bu

rden

carr

yin

gH

an

dic

raft

sT

rad

ing

Mea

nof

dep

.var.

3.5

31.7

33.2

54.6

53.9

04.6

43.4

72.7

82.4

7

Tra

dit

ion

al

plo

ugh

use

0.1

40.0

64

0.6

3�

0.0

19

�0.6

38

�0.0

52

�0.9

62**

�0.1

57

�0.1

55

(0.5

17)

(0.2

54)

(0.6

97)

(0.1

08)

(0.4

03)

(0.2

05)

(0.3

78)

(0.2

74)

(0.5

42)

Eth

nog

rap

hic

con

trol

syes

yes

yes

yes

yes

yes

yes

yes

yes

Obse

rvati

ons

88

95

48

173

159

154

135

74

59

Ad

just

edR

-squ

are

d�

0.0

2�

0.0

20.0

30.0

1�

0.0

01

0.0

10.1

20.0

7�

0.0

1R

-squ

are

d0.0

50.0

40.1

40.0

40.0

40.0

40.1

60.1

50.1

0

Not

es.

Th

eu

nit

ofob

serv

ati

onis

an

eth

nic

gro

up

from

the

Sta

nd

ard

Cro

ss-C

ult

ura

lS

am

ple

.T

he

dep

end

ent

vari

able

mea

sure

str

ad

itio

nal

fem

ale

part

icip

ati

onin

ap

art

icu

lar

act

ivit

yin

the

pre

-in

du

stri

al

per

iod

.T

he

vari

able

sta

ke

onin

teger

valu

esbet

wee

n1

an

d5

an

dare

incr

easi

ng

infe

male

part

icip

ati

on.

‘‘Tra

dit

ion

al

plo

ugh

use

’’is

an

ind

icato

rvari

able

that

equ

als

one

ifth

ep

lou

gh

was

trad

itio

nall

yu

sed

inp

re-i

nd

ust

rial

agri

cult

ure

.T

he

mea

n(a

nd

stan

dard

dev

iati

on)

ofth

isvari

able

is0.2

39

(0.4

29);

this

corr

esp

ond

sto

the

sam

ple

from

colu

mn

1.

‘‘Eth

nog

rap

hic

con

trol

s’’i

ncl

ud

e:th

esu

itabil

ity

ofth

elo

cal

envir

onm

ent

for

agri

cult

ure

,th

ep

rese

nce

ofla

rge

dom

esti

cate

dan

imals

,th

ep

rop

orti

onof

the

loca

len

vir

onm

ent

that

istr

opic

al

orsu

btr

opic

al,

an

ind

exof

sett

lem

ent

den

sity

,an

dan

ind

exof

pol

itic

al

dev

elop

men

t.F

iner

det

ail

sabou

tvari

able

con

stru

ctio

nare

pro

vid

edin

the

text

an

dap

pen

dix

.C

oeffi

cien

tsare

rep

orte

dw

ith

robu

stst

an

dard

erro

rsin

bra

cket

s.**

*,**

,an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


agriculture is used, women participate less in work outside thehome than when hoe agriculture is practiced. Whether womenwork more or less hours under one form than under the otherhas no direct effect on her argument. In addition, her hypothesismakes no claims about whether women are better or worse offunder one agricultural system than under the other.

IV. LINKING THE PAST TO THE PRESENT: DATA AND

METHODOLOGY

We next turn to an examination of the long-term impact oftraditional plough use. To do this, we link historical plough use,measured at the ethnicity level, with current outcomes of inter-est, measured at the location level (either countries or districtswithin countries) today. This requires an estimate of the geo-graphic distribution of ethnicities across the globe today. We con-struct this information using the 15th edition of Ethnologue:Languages of the World (Gordon 2005), a data source that mapsthe current geographic distribution of 7,612 different languages,each of which we manually matched to one of the 1,265 ethnicgroups from the Ethnographic Atlas. The Ethnologue provides ashape file that divides the world’s land into polygons, with eachpolygon indicating the location of a specific language as of 2003.We also use the Landscan 2000 database, which reports esti-mates of the world’s population in 2000 for 30 arc-second by 30arc-second (roughly 1 km by 1 km) grid cells globally. We combinethe Ethnologue shape file with the Landscan raster data to obtainan estimate of the distribution of language groups across theglobe today. This information is used to link the historicalethnicity-level data to our current outcomes of interest, measuredat the location level.

We illustrate our procedure with the example of Ethiopia.Figure Ia shows a map of the land inhabited by different ethnicgroups, i.e., groups speaking different languages. Each polygonrepresents the approximate borders of a group (from Ethnologue).One should not think of the borders as precisely defined bound-aries, but rather as rough measures indicating the approximatelocations of different language groups. The map also shows theLandscan estimate of the population of each cell within the coun-try. A darker shade indicates greater population.


at Harvard U



Dow

nloaded from


(a)

(b)

Population density and language groups

Population density, language groups, and their traditional plough use

FIGURE I

Populations, Language Groups, and Historical Plough Use within Ethiopia


at Harvard U



Dow

nloaded from


From the Ethnographic Atlas, we know whether each ethnicgroup used the plough historically. We define Iplough

e to be a vari-able equal to one if ethnic group e engaged in plough agricultureand zero otherwise. By matching each of the 7,612 Ethnologuelanguage groups to one of the 1,265 Ethnographic Atlas ethnicgroups, we can determine whether the ancestors of each languagegroup engaged in plough agriculture. This information is shownin Figure Ib. We thus have an estimate of ancestral plough useamong individuals across the world, observed at a level of 1 km by1 km. We then combine this with information on modern bound-aries to construct district- and country-level averages of ancestralplough use. This provides an estimate of the fraction of the popu-lation currently living in a district (or country) with ancestorsthat traditionally engaged in plough agriculture.

To be more precise, let Ne, i, d, c denote the number of individ-uals of ethnicity e living in grid-cell i located in district d in coun-try c. We construct a population-weighted average of Iplough

e for allethnic groups living in a district d. The district-level measure ofthe fraction of the population with ancestors that traditionallyused the plough, Ploughd, c, is given by

Ploughd, c ¼X

e

X

i

Ne, i, d, c

Nd, c� Iplough

e ,ð1Þ

where Nd, c is the total number of people living in district d incountry c. The same procedure is also used to construct acountry-level measure Ploughc, except that the average is takenover all grid-cells in country c.

Figure II shows the global distribution of languages based onthe Ethnologue data, as well as the historical plough use for eachgroup (i.e., the global version of Figure Ib). Uninhabited land isshown as dark gray.

The map highlights a number of potential concerns that weaddress in our empirical analysis. First, there is little variationwithin Europe and within sub-Saharan Africa. Therefore, there isthe concern that our estimates are simply driven by broad differ-ences between regions like Europe and Africa. Motivated by thisconcern, we also estimate a number of different specificationsidentified from within-country variation. These are reported inSection VI. We also report cross-country estimates omittingEurope and sub-Saharan Africa. These are reported in Section V.


at Harvard U



Dow

nloaded from


FIG

UR

EII

Tra

dit

ion

al

Plo

ugh

Use

am

ong

the

Eth

nic

/Lan

gu

age

Gro

up

sG

loball

y


at Harvard U



Dow

nloaded from


As well, the lack of variation in traditional plough use withinEurope implies that Boserup’s hypothesis and our empirical ana-lysis cannot explain existing differences in gender role beliefswithin Western Europe (and its offshoots).

Second, most of the variation appears to be at the macro levelwith much less variation across smaller geographic units.Although this is true in part, there is significant micro-level vari-ation in many parts of the world, much of which cannot easily beseen at the global scale presented in Figure II. In Section VI, weshow that our results are robust to examining within-countryvariation across ethnic groups and districts.

A third concern with the Ethnologue data is that informationis missing for some parts of the world. This is due to uncertaintyor a lack of information about the boundaries of some languagegroups. As is apparent from Figure II, this occurs primarily inLatin America.

We undertake a number of strategies to assess whether themissing data are systematically biasing our estimates. The first isto check the robustness of the results to omitting countries thathave a significant proportion of their language data coded asmissing. Another strategy is to construct alternative measureswhere we make assumptions regarding the missing languagedata. The first assumption we employ is to assume that all inhab-itants in unclassified territories speak the official national lan-guage of the country. The second strategy is to impute themissing data using information on the spatial distribution ofethnic groups from the Geo-Referencing of Ethnic Groups(GREG) database (Weidmann et al. 2010). Like the Ethnologue,the GREG database provides a shape file that divides the world’sland into polygons, with each polygon indicating the location of aspecific ethnicity. The shortcoming of the GREG database is thatethnic groups are much less finely identified relative to theEthnologue database. The GREG database identifies 1,364ethnic groups, while the Ethnologue identifies 7,612 languagegroups. The spatial distribution of traditional plough agricultureglobally using these alternative procedures is reported inAppendix Figures A2a and A2b.

The traditional plough use variables, constructed usingeither imputation method, are highly correlated with the baselinemeasure. At the country level, the correlation between: (1) ourbaseline variable and the measure with missing languagesimputed using the country’s national language is 0.89, (2) our


at Harvard U



Dow

nloaded from


baseline measure and the measure imputed using ethnic groupsfrom the GREG database is 0.91, and (3) the two variables withimputed values is 0.98.13

V. COUNTRY-LEVEL OLS ESTIMATES

Having constructed country- and district-level measures oftraditional plough use, we are able to examine the relationshipbetween historical plough use and the role of women in societiestoday. We begin by examining variation at the country level. Ourcountry-level estimates examine three outcomes of interest, all ofwhich are intended to reflect cultural attitudes and beliefs aboutthe role of women in society. The first measure we consider is eachcountry’s female labor force participation rate (FLFP) in 2000. Wealso examine women’s participation in more narrowly specifiedactivities outside the domestic sphere: entrepreneurship (mea-sured by the share of firms with a woman among its principalowners) and national politics (measured by the proportion ofseats held by women in national parliament).14

We first examine the unconditional relationship betweeneach outcome and our measure of traditional plough agriculture.These are reported in Figure III. For two of the three outcomesof interest, we find that the relationship is in the hypothesizeddirection and statistically significant. Traditional plough use isassociated with less female labor force participation and lessfemale firm ownership. Both relationships are statistically sig-nificant. However, the bivariate relationship between female par-ticipation in politics and traditional plough use is insignificantand positive.

13. Appendix Figures A3a–A3c report maps showing the country-level aver-ages for each of the three plough variables.

14. Female labor force participation is taken from the World Bank’s WorldDevelopment Indicators. The variable is measured as the percentage of womenaged 15 to 64 that are in the labor force; it ranges from 0 to 100. The share of acountry’s firms with some female ownership is measured as the percentage of sur-veyed firms with a woman among the principal owners. The measures are from theWorld Bank Enterprise Surveys, which were undertaken between 2005 and 2011,depending on the country. The proportion of seats in national parliament is mea-sured as the percent of parliamentary seats, in a single or lower chamber, held bywomen. The variable, measured in 2000, is taken from the United Nations’ Women’sIndicators and Statistics Database. The sample of countries for each variable isreported in Table A4 of the appendix.


at Harvard U



Dow

nloaded from


FIGURE III

Bivariate Correlations with Traditional Plough Use


at Harvard U



Dow

nloaded from


In Appendix Figure A4, we show that the relationships aresimilar if we condition on continent fixed effects (North America,South America, Europe, Asia, Africa, and Oceania). The one dif-ference is that the relationship between traditional plough useand female participation in politics is now negative—consistentwith Boserup’s hypothesis—although it remains statisticallyinsignificant.

We test Boserup’s hypothesis by estimating the followingequation:

yc ¼ �þ �Ploughc þ XHc !þ XC

c &þ "c,ð2Þ

where yc is an outcome of interest, c denotes countries, andPloughc is our measure of the traditional use of the ploughamong the ancestors of the citizens in country c. XH

c and XCc are

vectors of historical ethnographic and contemporary control vari-ables, each measured at the country level.

The ethnographic control variables XHc are intended to cap-

ture historical differences between societies that had adopted

FIGURE III

Continued


at Harvard U



Dow

nloaded from


TA

BL

EII

I

CO

UN

TR

Y-L

EV

EL

OL

SE

ST

IMA

TE

SW

ITH

HIS

TO

RIC

AL

CO

NT

RO

LS

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep

end

ent

vari

able

:

Fem

ale

labor

forc

ep

art

icip

ati

onin

2000

Sh

are

offi

rms

wit

hfe

male

own

ersh

ip,

2003–2010

Sh

are

ofp

olit

ical

pos

itio

ns

hel

dby

wom

enin

2000

Aver

age

effe

ctsi

ze(A

ES

)

Mea

nof

dep

.var.

51.0

334.7

712.1

12.3

1

Tra

dit

ion

al

plo

ugh

use

�14.8

95**

*�

15.9

62**

*�

16.2

43**

*�

17.8

06**

*�

2.5

22

�2.3

03

�0.7

36**

*�

0.9

20**

*(3

.318)

(3.8

81)

(3.8

54)

(4.4

75)

(1.9

67)

(2.3

53)

(0.0

84)

(0.1

00)

His

tori

cal

con

trol

s:A

gri

cult

ura

lsu

itabil

ity

9.4

07**

9.0

17**

1.5

14

4.6

19

1.0

09

�0.6

87

0.3

12**

0.3

25**

(3.8

85)

(4.2

36)

(5.3

58)

(5.8

36)

(2.7

99)

(2.9

25)

(0.1

29)

(0.1

33)

Tro

pic

al

clim

ate

�8.6

44**

*�

12.3

89**

*�

11.0

91**

*�

3.9

74

�7.6

71**

*�

5.6

18**

�0.3

22**

*�

0.0

04

(2.6

98)

(3.3

02)

(3.6

08)

(5.5

42)

(2.3

70)

(2.2

65)

(0.0

83)

(0.1

02)

Pre

sen

ceof

larg

ean

imals

10.9

03**

2.3

5�

0.6

49

4.4

75

�9.1

52**

�7.3

38

0.1

74

0.2

96**

(5.0

32)

(5.9

56)

(9.1

30)

(10.0

34)

(4.0

52)

(4.7

74)

(0.1

11)

(0.1

45)

Pol

itic

al

hie

rarc

hie

s�

0.7

87

0.4

47

1.5

02

0.5

20.9

06

0.6

99

0.0

80**

0.0

62

(1.6

22)

(1.6

24)

(1.8

45)

(1.7

73)

(0.7

40)

(0.7

77)

(0.0

40)

(0.0

43)

Eco

nom

icco

mp

lexit

y0.1

70

1.1

57

1.8

10*

0.5

17

1.0

82**

0.7

27

0.0

48**

0.0

18

(0.8

49)

(0.8

59)

(1.0

23)

(1.3

51)

(0.4

91)

(0.5

10)

(0.0

21)

(0.0

26)

Con

tin

ent

fixed

effe

cts

no

yes

no

yes

no

yes

no

yes

Obse

rvati

ons

177

177

128

128

153

153

153

153

Ad

just

edR

-squ

are

d0.2

00.2

40.1

40.1

60.1

40.1

40.2

40.2

7R

-squ

are

d0.2

20.2

80.1

80.2

30.1

70.2

00.2

50.3

0

Not

es.

OL

Ses

tim

ate

sare

rep

orte

dw

ith

robu

stst

an

dard

erro

rsin

bra

cket

s.T

he

un

itof

obse

rvati

onis

aco

un

try.

‘‘Tra

dit

ion

al

plo

ugh

use

’’is

the

esti

mate

dp

rop

orti

onof

citi

zen

sw

ith

an

cest

ors

that

use

dth

ep

lou

gh

inp

re-i

nd

ust

rial

agri

cult

ure

.T

he

vari

able

ran

ges

from

0to

1.

Th

em

ean

(an

dst

an

dard

dev

iati

on)

for

this

vari

able

is0.5

22

(0.4

73);

this

corr

esp

ond

sto

the

sam

ple

from

colu

mn

s1

an

d2.

‘‘Fem

ale

labor

forc

ep

art

icip

ati

on’’

isth

ep

erce

nta

ge

ofw

omen

inth

ela

bor

forc

e,m

easu

red

in2000.

Th

evari

able

ran

ges

from

0to

100.

‘‘Sh

are

offi

rms

wit

hfe

male

own

ersh

ip’’

isth

ep

erce

nta

ge

offi

rms

inth

eW

orld

Ban

kE

nte

rpri

seS

urv

eys

wit

hso

me

fem

ale

own

ersh

ip.

Th

esu

rvey

sw

ere

con

du

cted

bet

wee

n2003

an

d2010,

dep

end

ing

onth

eco

un

try.

Th

evari

able

ran

ges

from

0to

100.

‘‘Sh

are

ofp

olit

ical

pos

itio

ns

hel

dby

wom

en’’

isth

ep

rop

orti

onof

seats

inp

arl

iam

ent

hel

dby

wom

en,

mea

sure

din

2000.

Th

evari

able

ran

ges

from

0to

100.

Th

en

um

ber

ofob

serv

ati

ons

rep

orte

dfo

rth

eA

ES

isth

eaver

age

nu

mber

ofob

serv

ati

ons

inth

ere

gre

ssio

ns

for

the

thre

eou

tcom

es.

***,

**,

an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


plough agriculture and those that had not. They include the pres-ence of large domesticated animals, economic development mea-sured by the density of settlement, levels of political authority inthe society, agricultural suitability, and the presence of a tropicalclimate. These are the same set of historical controls as in theregressions reported in Tables I and II. We construct country-level measures of each variable using the same procedure thatis used to construct the historical plough use variable. Thus, theethnographic controls capture the historical characteristics of acountry’s ancestors. OLS estimates of equation (2) controlling forhistorical control variables XH

c are reported in Table III. Theeven-numbered columns also include controls for continentfixed effects.15

The contemporary control variables XCc include the natural

log of a country’s real per capita GDP measured in 2000, as wellas the variable squared. Allowing for a non-linear relationship ismotivated by the observed U-shaped bivariate relationship be-tween economic development and female labor force participation(Goldin 1995). Estimates of equation (2) including the contempor-ary controls are reported in Table IV.

The estimates show that in countries with a tradition ofplough use, women are less likely to participate in the labormarket, are less likely to own firms, and are less likely to partici-pate in national politics.16 Although the estimated relationshipsbetween traditional plough use and either female labor force par-ticipation or female firm ownership are similar with or withoutthe contemporaneous income controls, this is not true for femaleparticipation in politics. Once we control for per capita income,the magnitude of the relationship between traditional plough useand female participation in politics roughly doubles and becomesstatistically significant. This can be explained by the fact thatthere is a positive relationship between per capita income andfemale participation in politics (� ¼ 0:44) and between plough

15. One concern with the reported estimates is that many of the control vari-ables are potentially endogenous to traditional plough use. Theresults are similar ifwe control only for the geographically determined historical covariates, namely,agricultural suitability and the presence of a tropical climate. Estimates with onlythese controls are reported in Appendix Table A5.

16. Many countries have introduced quotas to increase the participation ofwomen in politics. We have checked the robustness of our results excluding coun-tries with gender quotas. For the restricted sample of 104 countries, the estimatedcoefficient is –5.91 with a standard error of 1.97.


at Harvard U



Dow

nloaded from


TA

BL

EIV

CO

UN

TR

Y-L

EV

EL

OL

SE

ST

IMA

TE

SW

ITH

HIS

TO

RIC

AL

AN

DC

ON

TE

MP

OR

AR

YC

ON

TR

OL

S

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep

end

ent

vari

able

:

Fem

ale

labor

forc

ep

art

icip

ati

onin

2000

Sh

are

offi

rms

wit

hfe

male

own

ersh

ip,

2003–2010

Sh

are

ofp

olit

ical

pos

itio

ns

hel

dby

wom

enin

2000

Aver

age

effe

ctsi

ze(A

ES

)

Mea

nof

dep

.var.

51.3

535.1

711.8

32.3

1

Tra

dit

ion

al

plo

ugh

use

�12.4

01**

*�

12.9

30**

*�

15.2

41**

*�

16.5

87**

*�

4.8

21**

*�

5.1

29**

�0.7

43**

*�

0.8

45**

*(2

.964)

(3.5

37)

(4.0

60)

(4.9

60)

(1.7

82)

(2.0

61)

(0.0

80)

(0.0

91)

His

tori

cal

con

trol

s:A

gri

cult

ura

lsu

itabil

ity

6.0

73

7.1

81*

0.8

03

4.3

22

2.1

98

1.0

81

0.2

62*

0.3

42**

(3.6

96)

(4.1

75)

(5.4

47)

(6.0

71)

(2.6

05)

(2.5

48)

(0.1

39)

(0.1

39)

Tro

pic

al

clim

ate

�9.7

18**

*�

10.9

06**

*�

10.4

32**

*�

3.7

12

�6.0

86**

*�

4.1

69*

�0.3

62**

*�

0.0

6(2

.487)

(3.0

70)

(3.7

62)

(5.7

11)

(2.0

94)

(2.3

96)

(0.0

84)

(0.1

01)

Pre

sen

ceof

larg

ean

imals

�2.0

15

�2.1

66

2.7

07

5.6

10

�5.7

18

�4.6

88

0.0

05

0.2

01

(5.3

72)

(6.0

72)

(9.7

45)

(10.4

17)

(3.5

65)

(4.1

32)

(0.1

21)

(0.1

46)

Pol

itic

al

hie

rarc

hie

s0.7

79

1.1

81

1.1

28

0.2

07

0.7

44

0.6

56

0.1

02**

0.0

70*

(1.5

15)

(1.4

82)

(1.9

41)

(1.8

78)

(0.8

22)

(0.8

07)

(0.0

40)

(0.0

42)

Eco

nom

icco

mp

lexit

y1.1

57

1.4

11*

1.6

93

0.7

64

0.4

54

0.3

33

0.0

63**

*0.0

27

(0.7

93)

(0.8

15)

(1.1

29)

(1.3

82)

(0.4

87)

(0.5

02)

(0.0

23)

(0.0

26)


at Harvard U



Dow

nloaded from


TA

BL

EIV

(CO

NT

INU

ED)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep

end

ent

vari

able

:

Fem

ale

labor

forc

ep

art

icip

ati

onin

2000

Sh

are

offi

rms

wit

hfe

male

own

ersh

ip,

2003–2010

Sh

are

ofp

olit

ical

pos

itio

ns

hel

dby

wom

enin

2000

Aver

age

effe

ctsi

ze(A

ES

)

Con

tem

por

ary

con

trol

s:ln

inco

me

in2000

�34.6

12**

*�

32.6

85**

*10.7

66

6.3

85

�6.5

30

�6.6

16

�0.7

76**

*�

0.8

15**

*(6

.528)

(7.0

23)

(9.9

86)

(10.4

82)

(4.0

71)

(4.3

35)

(0.2

21)

(0.2

31)

lnin

com

ein

2000

squ

are

d2.0

38**

*1.9

36**

*�

0.7

07

�0.5

23

0.5

39**

0.5

35*

0.0

51**

*0.0

51**

*(0

.406)

(0.4

31)

(0.6

88)

(0.7

06)

(0.2

71)

(0.2

81)

(0.0

15)

(0.0

15)

Con

tin

ent

fixed

effe

cts

no

yes

no

yes

no

yes

no

yes

Obse

rvati

ons

165

165

123

123

144

144

144

144

Ad

just

edR

-squ

are

d0.3

70.3

60.1

10.1

30.2

70.2

70.2

60.3

0R

-squ

are

d0.4

00.4

10.1

60.2

20.3

10.3

40.2

80.3

3

Not

es.

OL

Ses

tim

ate

sare

rep

orte

dw

ith

robu

stst

an

dard

erro

rsin

bra

cket

s.T

he

un

itof

obse

rvati

onis

aco

un

try.

‘‘Tra

dit

ion

al

plo

ugh

use

’’is

the

esti

mate

dp

rop

orti

onof

citi

zen

sw

ith

an

cest

ors

that

use

dth

ep

lou

gh

inp

re-i

nd

ust

rial

agri

cult

ure

.T

he

vari

able

ran

ges

from

0to

1.

Th

em

ean

(an

dst

an

dard

dev

iati

on)

ofth

isvari

able

is0.5

25

(0.4

72);

this

corr

esp

ond

sto

the

sam

ple

from

colu

mn

s1

an

d2.

‘‘Fem

ale

labor

forc

ep

art

icip

ati

on’’

isth

ep

erce

nta

ge

ofw

omen

inth

ela

bor

forc

e,m

easu

red

in2000.

Th

evari

able

ran

ges

from

0to

100.

‘‘Sh

are

offi

rms

wit

hfe

male

own

ersh

ip’’

isth

ep

erce

nta

ge

offi

rms

inth

eW

orld

Ban

kE

nte

rpri

seS

urv

eys

wit

hso

me

fem

ale

own

ersh

ip.

Th

esu

rvey

sw

ere

con

du

cted

bet

wee

n2003

an

d2010,

dep

end

ing

onth

eco

un

try.

Th

evari

able

ran

ges

from

0to

100.

‘‘Sh

are

ofp

olit

ical

pos

itio

ns

hel

dby

wom

en’’

isth

ep

rop

orti

onof

seats

inp

arl

iam

ent

hel

dby

wom

en,

mea

sure

din

2000.

Th

evari

able

ran

ges

from

0to

100.

Th

en

um

ber

ofob

serv

ati

ons

rep

orte

dfo

rth

eA

ES

isth

eaver

age

nu

mber

ofob

serv

ati

ons

inth

ere

gre

ssio

ns

for

the

thre

eou

tcom

es.

***,

**,

an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


use and per capita income (� ¼ 0:35). In other words, even if trad-itional plough use has a negative impact on female participationin politics, countries with a tradition of plough use also tend to bericher, and therefore, all else equal, have more female participa-tion in politics. Therefore, the strength of the coefficient forplough use depends on whether we control for contemporaryincome.

The reason that the estimates for the two other outcomes ofinterest—female firm ownership and female labor force partici-pation—are not sensitive to conditioning on per capita income isdue to the fact that the relationship between per capita incomeand the other outcomes is either zero (i.e., female firm ownership)or U-shaped (FLFP), both of which result in a zero correlationbetween the dependent variable and per capita income.

The partial correlation plots for traditional plough use areshown in Figures IVa–IVc (for columns 1, 3, and 5 of Table IV).From the figures it is clear that the coefficients for traditionalplough use are not being influenced by a small number of coun-tries. The plots also show that the coefficient estimates are notidentified only from broad differences across regions, but fromfiner within-region variation. For example, we observe Africancountries in the Northwest corner (e.g., Rwanda, Madagascar)and in the Southeast corner (e.g., Eritrea, Mauritania, Ethiopia,etc.). This is confirmed by the fact that the point estimates con-trolling for continent fixed effects are essentially identical to theestimates without the fixed effects (comparing the odd-numberedcolumns to the even-numbered columns in Tables III and IV).

Not only are the coefficient estimates statistically significant,but they are also economically meaningful. Based on the esti-mates from column 1, a one-standard-deviation increase in trad-itional plough use (0.472) is associated with a reduction of femalelabor force participation (FLFP) of 5.85 percentage points(12:401� 0:472), which is equal to 11.4% of the sample meanfor FLFP and 38% of its standard deviation. The impact of thesame increase in traditional plough use on the share of firms withsome female ownership (based on the column 3 estimate) is areduction of 7.19 percentage points, which is 20% of the outcome’smean and 48% of its standard deviation. The reduction on theparticipation of women in politics (using the column 5 estimate)is 2.28 percentage points, which is 19% of the outcome’s mean and25% of its standard deviation.


at Harvard U



Dow

nloaded from


(a)

(b)

FIGURE IV

Partial Correlation Plots


at Harvard U



Dow

nloaded from


FIGURE IV

Continued


at Harvard U



Dow

nloaded from


An alternative way to assess the magnitude of the estimatesis to calculate the proportion of the total variation that they ex-plain. By this metric, traditional plough use also explains a siz-able proportion of differences in gender roles across countries.When female labor force participation is the dependent variable(column 1 of Table IV), the inclusion of the historical plough usevariable increases the R-squared by 0.06 (from 0.34 to 0.40).Therefore, a tradition of plough agriculture accounts for 6% ofthe total variation in FLFP and 9.1% of the residual variationleft unexplained by the control variables.17 For the share offirms with female ownership, traditional plough use accountsfor 11% of the total variation and 11.6% of the residual variation.For women’s participation in politics, historical plough use ex-plains 3% of the total variation and 4.2% of the residual variation.

Columns 9 and 10 of Tables III and IV report the estimatedaverage effect size (AES) for the three outcomes of interest(Kling et al. 2007).18 The AES estimates are very similar tothe OLS estimates. According to the estimate from column 7, aone-standard-deviation increase in traditional plough use is asso-ciated with an average decrease (across the three outcomes) equalto approximately 0:472� 0:743 ¼ 0:351 standard deviations.

V.A. The Persistence of Female Labor Force Participation

Up to this point, we have shown that historical plough use isassociated with less female participation in agriculture historic-ally and with less female participation in the labor force today.These two correlations suggest long-term persistence in femaleparticipation in activities outside the home. To confirm this, weregress female labor force participation today on the measure ofwomen’s historical participation in agriculture constructed fromthe Ethnographic Atlas. The regression also controls for our fullset of covariates from equation (2). The partial correlation plotillustrating the relationship between the two variables is shownin Figure IVd. As is apparent from the figure, there is strongpersistence over time. Female labor force participation today

17. This is calculated as ð0:40� 0:34Þ=ð1� 0:34Þ ¼ 0:091 or 9.1%.18. In implementation, we follow Kling et al. (2007) and Clingingsmith et al.

(2009) and calculate the AES as 1K

PKk¼1

�k

�k, where K is the total number of outcomevariables,�k is the estimated plough use coefficient for outcome variable k, and �k isthe standard deviation of outcome k. To properly calculate the sample variance ofthe AES, the coefficients �k are jointly estimated in a seemingly unrelated regres-sion framework.


at Harvard U



Dow

nloaded from


and female participation in agriculture in the past are verystrongly correlated.

Despite the evidence of persistence over time on average, it isimportant to recognize that there are well-documented excep-tions to this rule. For example, Goldin and Sokoloff (1984) docu-ment that within the northeastern United States, the low relativeproductivity of women and children in agriculture (and their lowparticipation in this sector) allowed them to actively participatein the manufacturing sector. In this setting, initial femalelabor force participation in agriculture is inversely related to sub-sequent participation in manufacturing, showing a lack ofcontinuity of female labor force participation over time as indus-trialization occurred.

V.B. Robustness Checks

We first consider the robustness of our results to the use ofalternative plough measures. Using either of the two methodsdescribed above for imputing missing language data—either offi-cial language or GREG ethnicity—yields estimates that are quali-tatively identical to the estimates using our baseline variable. Asreported in Appendix Table A6, the alternative measures yieldnearly identical point estimates that are highly significant. Wealso test the robustness of our results to defining traditionalplough use in a slightly different manner. Our baseline variableequals one if an ethnic group traditionally used the plough inpre-industrial agriculture, whether or not adoption occurredprior to European contact. We alternatively code the indicatorvariable to equal one only if adoption occurred prior toEuropean contact. The estimates using this alternative defin-ition, reported in Appendix Table A7, are virtually identical toour baseline estimates.

We also check the robustness of our results to the use of dif-ferent samples. First, to ensure that our findings are not beingdriven by measurement error, we omit 17 countries that have asignificant proportion of missing language data. The countriesinclude Australia, Argentina, Bolivia, Brazil, Chile, Colombia,Costa Rica, Ecuador, Guatemala, Honduras, Mexico, NewZealand, Nicaragua, Panama, Paraguay, Peru, and Venezuela.The estimates reported in panel A of Appendix Table A8 showthat the estimated impact of the plough remains robust.


at Harvard U



Dow

nloaded from


Motivated by the fact that there is little variation in ploughuse within Europe, we also test the robustness of our results tothe omission of European countries and neo-European countries(United States, Canada, Australia, and New Zealand) from thesample. As reported in panel B of Appendix Table A8, the resultsremain robust.

Examining Figure II, it is apparent that many of the countrieswith ancestors that traditionally did not use the plough are locatedin sub-Saharan Africa. We also check that our results do not reflectdifferences only between sub-Saharan Africa and the rest of theworld by re-estimating equation (2) after omitting all sub-SaharanAfrican countries (panel C of Appendix Table A8). We obtain verysimilar estimates when we omit these countries from the sample.19

An alternative strategy is to disaggregate the African continentindicator variable into an indicator variable for sub-SaharanAfrica and an indicator variable for Northern Africa. The estimatesare also robust to this procedure (panel D of Appendix Table A8).

Today’s industrial countries have witnessed dramatic chan-ges in gender roles in recent decades. For example, within theUnited States, social norms regarding women’s employment out-side the home began to change dramatically in the late 1960s and1970s. This is reflected by female labor force participation rates,which increased from 25 to 46 percent (among women age 35–44)from 1950 to 1970 (Goldin 2006, 6–8). In light of these recentchanges, it is important to check that we find a similar impactof the plough on gender norms prior to these changes. We there-fore also examine female labor force participation and female par-ticipation in politics in the 1950s and 1960s. We are unable toexamine female firm ownership due to a lack of data prior to the2000s. As detailed in Section A5 of the online appendix, we obtainqualitatively and quantitively similar estimates when we exam-ine this earlier time period.

VI. SUBNATIONAL ESTIMATES

VI.A. Evidence from the World Values Surveys

We now turn to specifications that examine variation acrossindividuals, linking them to a tradition of plough agriculture

19. We also obtain similar estimates if we omit all African countries, not justsub-Saharan African countries.


at Harvard U



Dow

nloaded from


using the district in which they live. The analysis relies on datafrom the World Value Survey (WVS), a compilation of nationalindividual-level surveys on a wide variety of topics, including at-titudes and preferences. They also include information on stand-ard demographic characteristics, such as gender, age, education,labor market status, income, and religion.20 Using the WVS, weconstruct an indicator variable that equals one if a woman is inthe labor force, which is defined as full-time, part-time, or self-employment. Women are not in the labor force if they report beingretired, a housewife, or a student.21

We also examine two measures of individual (male andfemale) attitudes about the appropriate role of women in society.The first measure is based on each respondent’s view of the fol-lowing statement: ‘‘When jobs are scarce, men should have moreright to a job than women.’’ The respondents are asked to chooseamong ‘‘agree,’’ ‘‘disagree,’’ ‘‘neither,’’ or ‘‘don’t know.’’ We omitobservations for which the respondents answered ‘‘neither’’ or ‘‘Idon’t know,’’ and code ‘‘disagree’’ as 0 and ‘‘agree’’ as 1.22

We also consider a second variable derived from survey re-sponses to the following statement: ‘‘On the whole, men makebetter political leaders than women do.’’ Respondents are askedto choose among ‘‘strongly disagree,’’ ‘‘disagree,’’ ‘‘agree, ‘‘agreestrongly,’’ or ‘‘don’t know.’’ We omit observations in which therespondent answered ‘‘don’t know’’ and create a variable thattakes the value of 1 for ‘‘strongly disagree,’’ 2 for ‘‘disagree,’’ 3for ‘‘agree,’’ and 4 for ‘‘agree strongly.’’ By construction, both at-titude variables are increasing in the extent to which a respond-ent’s view reflects a belief of gender inequality.

20. Five waves of the WVS were carried out between 1981 and 2007. In ouranalysis, we use the four most recent waves of the survey, since the first wavedoes not contain information on the districts in which respondents live. Becauseregional classifications often vary by wave, we use the wave with the most finelydefined location data. A list of countries for each question is reported in Table A9 ofthe appendix.

21. The results are qualitatively identical if we exclude retired women and stu-dents from the sample.

22. We omit observations for which the respondent answered ‘‘neither’’ since itis unclear whether this represents an intermediate view, whether they have chosennot to answer the question, or whether they do not know their answer. If we inter-pret this response as reflecting an intermediate position and code a variable thattakes the values 0, 1, or 2, we obtain qualitatively identical results to what we reporthere.


at Harvard U



Dow

nloaded from


An appealing characteristic of the two WVS variables is thatthey measure the values that underlie the objective outcomesexamined in the cross-country analysis. The first question reflectsdifferences in individual beliefs about whether women shouldhave equal access to jobs, which is likely an important factorunderlying observed differences in female labor force participa-tion across countries. The second question reflects values aboutthe ability of women to take on roles of leadership and responsi-bility, which likely affects observed differences in female partici-pation in politics and female firm ownership. Therefore, there is aclose link between the objective measures from the country-levelanalysis and the subjective measures in the individual-levelanalysis.

Examining the three outcomes—female participation in thelabor force, attitudes about female employment, and attitudesabout female leadership—we estimate the following individual-level equation:

yi, d, c ¼ �c þ �Ploughd þ XHd &þ Xi(þ "i, d, c,ð3Þ

where i denotes an individual, d a district, and c a country.Ploughd is our measure of traditional plough use among the an-cestors of individuals living in district d. �c denotes country fixedeffects. XH

d includes the same historical ethnographic variables asin equation (2), but measured at the district level rather than thecountry level. Xi denotes current individual-level controls: age,age squared, marital status fixed effects, educational attainmentfixed effects, and gender (for the attitude regressions only).Standard errors are clustered at the district level.

The odd numbered columns of Table V report OLS estimatesof equation (3), but with continent fixed effects rather than coun-try fixed effects. In this specification, we include XC

c , the samecontemporary country-level controls as in equation (2). We find anegative relationship between traditional plough use and currentfemale labor force participation, and a positive relationshipbetween plough use and attitudes reflecting gender inequality.All three relationships are statistically significant. The estimatedmagnitudes of the female labor force participation effects aresimilar to the cross-country estimates reported in columns 1and 2 of Table IV. A one-standard-deviation increase in trad-itional plough use (0.415) implies a reduction in female laborforce participation of 0.073 or 7.3 percentage points, roughly


at Harvard U



Dow

nloaded from


TA

BL

EV

IND

IVID

UA

L-L

EV

EL

OL

SE

ST

IMA

TE

SU

SIN

GW

VS

DA

TA

(1)

(2)

(3)

(4)

(5)

(6)

Dep

end

ent

vari

able

:

Fem

ale

labor

forc

ep

art

icip

ati

on,

1995–2007

Wh

enjo

bs

are

scarc

e,1995–2007

Men

bet

ter

pol

itic

al

lead

ers,

1995–2007

Mea

nof

dep

.var.

0.5

50.5

50.4

60.4

72.6

22.6

4

Tra

dit

ion

al

plo

ugh

use

�0.1

77**

*�

0.0

02

0.1

93**

*0.1

00*

0.2

24**

*0.3

04**

*(0

.035)

(0.0

31)

(0.0

33)

(0.0

59)

(0.0

69)

(0.1

17)

Ind

ivid

ual

&d

istr

ict

con

trol

syes

yes

yes

yes

yes

yes

Con

tem

por

ary

cou

ntr

yco

ntr

ols

yes

n/a

yes

n/a

yes

n/a

Fix

edef

fect

sco

nti

nen

tco

un

try

con

tin

ent

cou

ntr

yco

nti

nen

tco

un

try

Nu

mber

ofco

un

trie

s73

78

74

79

50

55

Nu

mber

ofd

istr

icts

672

698

674

700

453

479

Obse

rvati

ons

43,8

01

47,5

87

80,3

03

87,5

28

64,2

15

72,1

52

Ad

just

edR

-squ

are

d0.1

70.2

70.2

10.2

80.1

90.2

6R

-squ

are

d0.1

70.2

70.2

10.2

80.1

90.2

6

Not

es.

Th

eta

ble

rep

orts

OL

Ses

tim

ate

s,w

ith

stan

dard

erro

rscl

ust

ered

at

the

dis

tric

tle

vel

.T

he

un

itof

obse

rvati

onis

an

ind

ivid

ual.

Inco

lum

ns

1an

d2,

the

sam

ple

incl

ud

esw

omen

only

an

dth

ed

epen

den

tvari

able

isan

ind

icato

rvari

able

that

equ

als

one

ifsh

eis

inth

ela

bor

forc

e.T

he

esti

mate

sin

colu

mn

s3–6

incl

ud

em

enan

dw

omen

.T

he

dep

eden

tvari

able

sm

easu

rere

spon

den

ts’

self

-rep

ort

att

itu

des

regard

ing

gen

der

role

s.A

hig

her

valu

ein

dic

ate

sbel

iefs

abou

tgre

ate

rin

equ

ali

tybet

wee

nm

enan

dw

omen

.‘‘W

hen

jobs

are

scarc

e’’

tak

eson

the

valu

eof

zero

oron

e,w

hil

e‘‘m

enbet

ter

pol

itic

al

lead

ers’

’ta

kes

onin

teger

valu

esbet

wee

n1

an

d4.

‘‘In

div

idu

al

con

trol

s’’

are

:age,

age

squ

are

d,

du

mm

ies

for

pri

mary

an

dse

con

dary

edu

cati

on(t

he

excl

ud

edgro

up

iste

rtia

ryed

uca

tion

),gen

der

(for

gen

der

att

itu

de

dep

end

ent

vari

able

son

ly)

an

dan

ind

icato

rvari

able

for

mari

tal

statu

s.‘‘T

rad

itio

nal

plo

ugh

use

’’is

the

esti

mate

dp

rop

orti

onof

ind

ivid

uals

livin

gin

ad

istr

ict

wit

han

cest

ors

that

use

dth

ep

lou

gh

inp

re-i

nd

ust

rial

agri

cult

ure

.T

he

mea

n(a

nd

stan

dard

dev

iati

on)

ofth

isvari

able

is0.7

24

(0.4

25);

this

corr

esp

ond

sto

the

sam

ple

from

colu

mn

1.

‘‘Dis

tric

tco

ntr

ols’

’in

clu

de

dis

tric

t-le

vel

mea

sure

sof

:an

cest

ral

suit

abil

ity

for

agri

cult

ure

,fr

act

ion

ofan

cest

ral

lan

dth

at

was

trop

ical

orsu

btr

opic

al,

an

cest

ral

dom

esti

cati

onof

larg

ean

imals

,an

cest

ral

sett

lem

ent

patt

ern

s,an

dan

cest

ral

pol

itic

al

com

ple

xit

y.

‘‘Con

tem

por

ary

cou

ntr

yco

ntr

ols’

’in

clu

de:

the

natu

ral

log

ofre

al

per

cap

ita

GD

P,

an

dit

ssq

uare

,m

easu

red

inth

esa

me

yea

ras

the

dep

end

ent

vari

able

.**

*,**

,an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


equal to 13% of the sample average. The impact of aone-standard-deviation increase in the two attitude measures is0.08 and 0.09 (equal to 17 and 4% of the sample averages of thesevariables).

In the even-numbered columns of Table V, we control forcountry fixed effects. We continue to find a positive and signifi-cant relationship between traditional plough agriculture and at-titudes reflecting gender inequality. Although we estimate anegative relationship between traditional plough use andfemale labor force participation, the estimated magnitude issmall and the coefficient is statistically insignificant. For approxi-mately half of the countries in the sample, there is no subnationalvariation in the district-level plough measure. As we report inAppendix Table A10, we obtain similar estimates when we omitthese countries from the sample or when we restrict the samplefurther, additionally dropping countries with the least amount ofsubnational variation.

VI.B. Evidence from IPUMS-International

One explanation for the weaker WVS estimates—particu-larly the insignificant estimates for female labor force participa-tion—is attenuation bias arising from the imperfectly constructedplough use variable. Because we do not observe individual ethni-city, we are forced to use an individual’s district of residence as animperfect proxy. We now turn to another data source, IPUMS-International Census data, that records respondents’ ethnic iden-tity as well as their employment status. This allows us to examinefemale labor force participation using a more accurate measure ofancestral plough use. Because the IPUMS surveys do not recordinformation about gender-role beliefs, we are unable to examinethese outcomes.

Our estimating equation is

yi, d, e ¼ �d þ �Ploughe þ XHe &þ Xi(þ "i, d, e,ð4Þ

where i denotes an individual, d denotes a district within a coun-try, and e is ethnicity. Ploughe is our measure of traditional ploughuse among the ancestors of individuals of ethnicity e. �d denotesdistrict fixed effects.23 The vector XH

e includes the same historical

23. For three countries—Mongolia, Uganda, and Malaysia—there are multiplesurvey years. For these countries, we also include survey-year fixed effects and


at Harvard U



Dow

nloaded from


TA

BL

EV

I

IND

IVID

UA

L-L

EV

EL

OL

SE

ST

IMA

TE

SU

SIN

GIP

UM

S-I

NT

ER

NA

TIO

NA

LD

AT

A

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

:F

emale

labor

forc

ep

art

icip

ati

onin

dic

ato

r

Bol

ivia

Ch

ile,

Cam

bod

ia,

Mala

ysi

a,

Mon

gol

ia,

Nep

al,

Ph

ilip

pin

es,

Ugan

da,

All

2001

2002

2008

70,

80,

91,

00

1989,

2000

2001

1990

1991,

2002

cou

ntr

ies

Mea

nof

dep

.var.

0.4

40.4

00.7

80.4

00.3

80.5

40.3

90.5

60.4

9

Tra

dit

ion

al

plo

ugh

use�

0.0

35**

*�

0.0

73**

*�

0.0

64**

�0.0

80**

*�

0.0

06

�0.1

00**

0.0

35

�0.0

79**

*�

0.0

40**

(0.0

02)

(0.0

03)

(0.0

27)

(0.0

16)

(0.0

13)

(0.0

43)

(0.0

23)

(0.0

20)

(0.0

19)

Ind

ivid

ual

&et

hn

icit

yco

ntr

ols

yes

yes

yes

yes

yes

yes

yes

yes

yes

Dis

tric

tfi

xed

effe

cts

926

24

15

23

14

77

4192

Eth

nic

gro

up

s6

511

21

10

16

21

60

150

Obse

rvati

ons

173,8

04

505,1

14

432,4

81

319,5

80

125,3

49

710,6

62

1,2

66,3

63

1,0

03,3

21

4,5

36,6

74

Ad

just

edR

-squ

are

d0.0

70.1

70.1

90.1

00.5

00.1

90.1

30.0

90.1

5R

-squ

are

d0.0

70.1

70.1

90.1

00.5

00.1

90.1

30.0

90.1

5

Not

es.

OL

Ses

tim

ate

sare

rep

orte

d,

wit

hst

an

dard

erro

rscl

ust

ered

at

the

eth

nic

ity

level

inbra

cket

s.T

he

un

itof

obse

rvati

onis

afe

male

ind

ivid

ual.

Th

ed

epen

den

tvari

able

isan

ind

icato

rvari

able

that

equ

als

one

ifth

ein

div

idu

al

isre

por

ted

tobe

inth

ela

bor

forc

e.T

he

tim

ep

erio

dd

iffe

rsfo

rea

chco

un

try:

Bol

ivia

,2001;

Ch

ile,

2002;

Cam

bod

ia,

2008;

Mala

ysi

a,

1970,

1980,

1991,

an

d2000;

Mon

gol

ia,

1989

an

d2000;

Nep

al,

2001;

the

Ph

ilip

pin

es,

1990;

an

dU

gan

da,

1991,

2002.

‘‘Tra

dit

ion

al

plo

ugh

use

’’in

dic

ate

sw

het

her

the

eth

nic

gro

up

tow

hic

hea

chin

div

idu

al

bel

ongs

use

dth

ep

lou

gh

inp

re-i

nd

ust

rial

agri

cult

ure

.T

he

mea

n(a

nd

stan

dard

dev

iati

on)

ofth

isvari

able

is:

for

Bol

ivia

,0.5

17

(0.5

00);

for

Ch

ile,

0.9

59

(0.1

97);

for

Cam

bod

ia,

0.9

89

(0.1

03);

for

Mala

ysi

a,

0.9

29

(0.2

57);

for

Mon

gol

ia,

0.8

84

(0.3

20);

for

Nep

al,

0.9

71

(0.1

67);

for

the

Ph

ilip

pin

es,

0.3

27

(0.4

69);

an

dfo

rU

gan

da,

0.0

65

(0.2

47).

‘‘In

div

idu

al

con

trol

s’’

incl

ud

e:age,

age

squ

are

d,

fixed

effe

cts

for

edu

cati

onal

att

ain

men

t,an

ind

icato

rvari

able

for

mari

tal

statu

s,an

dan

urb

an

/ru

ral

ind

icato

rvari

able

.‘‘E

thn

icit

yco

ntr

ols’

’in

clu

de:

an

cest

ral

suit

abil

ity

for

agri

cult

ure

,fr

act

ion

ofan

cest

ral

lan

dth

at

was

trop

ical

orsu

btr

opic

al,

an

cest

ral

dom

esti

cati

onof

larg

ean

imals

,an

cest

ral

sett

lem

ent

patt

ern

s,an

dan

cest

ral

pol

itic

al

com

ple

xit

y.

For

cou

ntr

ies

wit

hd

ata

avail

able

for

mor

eth

an

one

wave,

we

als

oco

ntr

olfo

rsu

rvey

-wave

fixed

effe

cts

an

dd

istr

ict-

wave

fixed

effe

cts.

***,

**,

an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


ethnographic variables as in equation (2), but measured at theethnicity level rather than the country level. Xi denotes currentindividual-level controls: age, age squared, marital status, educa-tional attainment fixed effects, and an urban/rural indicator vari-able. We estimate equation (4) separately for each country,clustering standard errors at the ethnicity level. We also estimatea pooled regression that includes all countries. The sample in-cludes all countries in IPUMS-International that report informa-tion on respondents’ ethnicity at a sufficiently fine level and thathave within-country variation in traditional plough use.

Estimation results are reported in Table VI. In total there areeight countries: five from Asia, two from Latin America, and onefrom Africa. Columns 1–8 report separate estimates for each coun-try, while column 9 reports the average effect across all countries.The results show that overall there is a negative and statisticallysignificant relationship between female participation in the laborforce and a tradition of ancestral plough use. This is the case for sixof the eight countries in the country-by-country regressions.Although there is significant heterogeneity in the magnitude ofthe estimated impact of the plough on FLFP, the effect of aswitch from non-plough use to plough use generally ranges from4 to 10 percentage points (and the average effect from column 9 is 4percentage points). These magnitudes are approximately half thesize of the effect found in the cross-country analysis.24

For two countries, Mongolia and the Philippines, the coeffi-cient is not statistically different from zero. The lack of a relation-ship within Mongolia is explained by the fact that agriculturewas relatively unimportant within this region, with animalhusbandry being the primary form of subsistence. For theethnic groups that had adopted the plough, only 6–15% of

allow the district fixed effects to vary by survey year. The specific time periodscovered are different for each country. Full details are reported in the appendix.

24. It is also possible to generate estimates for Belarus, although only two ethnicgroups, accounting for 0.2% of the total population, did not use the plough tradition-ally. Given the imbalance in traditional plough use, we do not report these estimates.Consistent with Boserup’s hypothesis, the estimated coefficient is negative and stat-istically significant, although the estimated magnitude is extremely large. The coef-ficient is –0.522 with a standard error of 0.198. The estimated average effect across allcountries is identical if Belarus is also included (coeff: –0.40; s.e.: 0.019).


at Harvard U



Dow

nloaded from


subsistence consumption was from agriculture. Therefore, al-though the plough had been adopted, its impact appears tohave been muted because of the relative unimportance of agricul-ture. In this respect, Mongolia is very different from the rest ofthe countries in the sample. Among the other seven countries,agriculture accounted for at least 46% of each ethnic group’sfood consumption.

The reason for the different relationship within the Philip-pines remains unclear. However, it is interesting that in terms ofgender relations the Philippines is exceptional in many dimen-sions. For example, the Philippines (along with Mongolia) arethe only East Asian countries for which educational attainmentis higher for girls than for boys (Estudillo et al. 2001, WorldBank, 2012). In addition, the Philippines consistently ranksamong the top 5–10 most gender-equal countries globally,which is significantly higher than other Asian countries andother countries at a similar level of economic development(Hausman et al. 2012).

VII. ADDRESSING THE ISSUE OF CAUSALITY

VII.A. Controlling for Observable Characteristics

A potential concern with the OLS estimates reported up tothis point is that locations that historically had less equalgender-role attitudes may have had a higher likelihood of invent-ing or adopting the plough. This would bias the OLS estimatesaway from zero. It is also possible that locations that were eco-nomically more developed were more likely to have adopted theplough. Since these areas today tend to be richer and more proneto equal gender-role attitudes, this would tend to bias the OLSestimates toward zero. Our first strategy to address these, andrelated, concerns is to control for observable characteristics.

Our analysis makes a distinction between traditional ploughagriculture and all other forms of subsistence. In our coding,non-plough societies include both agricultural societies practicingshifting hoe agriculture as well as societies not engaged in agri-culture, such as hunter-and-gatherer and herding societies. It ispossible that the status of women is also affected by the extent towhich a society participates in these non-agricultural activities.We account for this by constructing two variables that measurethe proportion of ancestors’ subsistence that was provided by


at Harvard U



Dow

nloaded from


TA

BL

EV

II

RO

BU

ST

NE

SS

OF

OL

SE

ST

IMA

TE

ST

OA

DD

ITIO

NA

LC

OV

AR

IAT

ES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep

end

ent

vari

able

:F

emale

labor

forc

ep

art

icip

ati

onin

2000

Mea

nof

dep

.var.

51.3

551.5

551.3

551.4

851.2

652.0

951.4

852.1

3

Tra

dit

ion

al

plo

ugh

use

�10.8

92**

*�

12.7

14**

*�

12.3

56**

*�

12.3

36**

*�

12.7

21**

*�

14.6

18**

*�

9.9

13**

*�

9.2

34**

(3.8

48)

(3.2

55)

(2.9

93)

(3.0

19)

(3.3

64)

(3.4

82)

(3.1

60)

(4.3

01)

His

tori

cal

con

trol

s:P

ract

ices

inte

nsi

ve

agri

cult

ure

yes

yes

Pro

p.

ofsu

bsi

st.

from

her

din

gyes

yes

Pro

p.

ofsu

bsi

st.

from

hu

nti

ng

yes

yes

Abse

nce

ofp

rivate

pro

per

tyyes

yes

Patr

iloc

al

marr

iages

yes

yes

Matr

iloc

al

marr

iages

yes

yes

Nu

clea

rfa

mil

yst

ruct

ure

yes

yes

Exte

nd

edfa

mil

yst

ruct

ure

yes

yes

Yea

ret

hn

icit

ysa

mp

led

yes

yes

Con

tem

por

ary

con

trol

s:Y

ears

ofci

vil

con

flic

ts(1

816–2007)

yes

yes

Yea

rsof

inte

rsta

teco

nfl

icts

(1816–2007)

yes

yes

Ru

gged

nes

syes

yes

Com

mu

nis

min

dic

ato

ryes

yes


at Harvard U



Dow

nloaded from


TA

BL

EV

II

(CO

NT

INU

ED)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Dep

end

ent

vari

able

:F

emale

labor

forc

ep

art

icip

ati

onin

2000

Fra

ctio

nof

Eu

rop

ean

des

cen

tyes

yes

Oil

pro

du

ctio

np

erca

pit

ayes

yes

Agri

cult

ura

lsh

are

ofG

DP

yes

yes

Man

ufa

ctu

rin

gsh

are

ofG

DP

yes

yes

Ser

vic

essh

are

ofG

DP

yes

yes

Fra

ctio

nof

pop

.C

ath

olic

yes

yes

Fra

ctio

nof

pop

.P

rote

stan

tyes

yes

Fra

ctio

nof

pop

.C

hri

stia

n(o

ther

)yes

yes

Fra

ctio

nof

pop

.M

usl

imyes

yes

Fra

ctio

nof

pop

.H

ind

uyes

yes

Base

lin

eco

ntr

ols

yes

yes

yes

yes

yes

yes

yes

yes

Obse

rvati

ons

165

163

165

163

153

154

163

142

Ad

just

edR

-squ

are

d0.3

90.3

80.3

60.3

50.4

20.3

50.5

10.5

4R

-squ

are

d0.4

30.4

30.4

00.4

00.4

60.4

00.5

50.6

4

Not

es.

OL

Ses

tim

ate

sare

rep

orte

dw

ith

robu

stst

an

dard

erro

rsin

bra

cket

s.T

he

un

itof

obse

rvati

onis

aco

un

try.

‘‘Tra

dit

ion

al

plo

ugh

use

’’is

the

esti

mate

dp

rop

orti

onof

citi

zen

sw

ith

an

cest

ors

that

use

dth

ep

lou

gh

inp

re-i

nd

ust

rial

agri

cult

ure

.T

he

vari

able

ran

ges

from

0to

1.

Th

em

ean

(an

dst

an

dard

dev

iati

on)

ofth

isvari

able

is0.5

48

(0.4

68);

this

corr

esp

ond

sto

the

sam

ple

from

colu

mn

s1

an

d3.

‘‘Fem

ale

labor

forc

ep

art

icip

ati

on’’

isth

ep

erce

nta

ge

ofw

omen

inth

ela

bor

forc

e.T

he

vari

able

ran

ges

from

0to

100.

Each

regre

ssio

nin

clu

des

the

full

set

ofco

ntr

olvari

able

s(h

isto

rica

lan

dco

nte

mp

orary

).T

he

his

tori

cal

con

trol

sin

clu

de:

an

cest

ral

suit

abil

ity

for

agri

cult

ure

,fr

act

ion

ofan

cest

ral

lan

dth

at

was

trop

ical

orsu

btr

opic

al,

an

cest

ral

dom

esti

cati

onof

larg

ean

imals

,an

cest

ral

sett

lem

ent

patt

ern

s,an

dan

cest

ral

pol

itic

al

com

ple

xit

y.

Th

eco

nte

mp

orary

con

trol

sin

clu

de

the

natu

ral

log

ofre

al

per

cap

ita

GD

Pan

dit

ssq

uare

,m

easu

red

inth

esa

me

yea

ras

the

dep

end

ent

vari

able

.T

he

coef

fici

ents

an

dst

an

dard

erro

rsfo

rth

eco

ntr

olvari

able

sare

rep

orte

din

the

onli

ne

ap

pen

dix

.**

*,**

,an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


hunting and by the herding of large animals.25 In addition, wealso address the possibility that our plough measure is simplycapturing the intensity of agriculture by directly controlling foragricultural intensity. As shown in column 1 of Table VII, theinclusion of these three control variables has little impact onthe coefficient of interest.26

A prominent determinant of cultural differences in gender-role attitudes was proposed by Frederick Engels (1902). Heargued that gender inequality arose due to the intensification ofagriculture, which resulted in the emergence of private property,which was monopolized by men. The control of private propertyallowed men to subjugate women and to introduce exclusive pa-ternity over their children, replacing matriliny with patrilinealdescent, making wives even more dependent on husbands andtheir property. As a consequence, women were no longer activeand equal participants in community life. To account for this al-ternative determinant, we control for the proportion of a coun-try’s ancestors without land inheritance rules; with patrilocalpost-marital residence rules; and with matrilocal post-maritalrules. We also control for the proportion of a country’s ancestorswith a nuclear family structure and with an extended familystructure. As shown in column 2, the results remain robust tothe inclusion of these additional controls.

In column 3, we control for the fact that in the EthnographicAtlas ethnicities are observed at different points in time. In gen-eral, less-developed ethnic groups, without written records orwith less external contact, tend to have information from morerecent time periods. Although the characteristics of interesttend to be slow-moving and typically do not change significantlyin short periods of time, this could still introduce some measure-ment error. Motivated by this concern, we check that the resultsremain robust to controlling for the average date of observation ofancestors in the Ethnographic Atlas. The estimates are very simi-lar with this additional control.

We next turn to the possibility that warfare may have a sys-tematic impact on beliefs about gender roles. Theoretically, the

25. The sources and method of construction of all control variables fromTable VII are documented in the online appendix.

26. For brevity, we report only the estimates with female labor force participa-tion as the dependent variable. In addition, we do not report the coefficient esti-mates for the control variables. For the interested reader, these are reported inTable A11 of the appendix.


at Harvard U



Dow

nloaded from


expected direction of the effect is unclear (Whyte 1978). Involve-ment in warfare may cause societies to become more hierarchicaland male dominated, suggesting a negative relationship betweenconflict and female work outside the home. On the other hand,being involved in warfare can generate a greater need for femaleinvolvement outside the home, which may in turn affect the evo-lution of beliefs about gender. We control for the potential im-pacts of warfare, by calculating, for each country, the number ofyears since 1816 (the first year data are available) that the coun-try was involved in either internal or interstate warfare. To con-trol for the possibility that invasions are easier on the types of flatground that could also favor the diffusion of the plough, we alsocontrol for Nunn and Puga’s (2012) terrain ruggedness index. Theresults, reported in column 4, show that the impact of traditionalplough use is robust to controlling for a country’s exposure towarfare during the 19th and 20th centuries.

Communism and large historical migrations (most notablymass European migrations during the 17th to the 20th centuries)are other recent historical episodes that may have had a largeimpact on beliefs about gender. Communist regimes typically im-plemented policies to eliminate gender differences in the economyand promote female participation outside the home. For manyparts of the world, the adoption of the plough may be associatedwith the diffusion of other aspects of Western European societies.In column 5, we include an indicator variable that equals one ifthe country was under a communist regime in the post-WWIIperiod. We also control for the fraction of each country’s popula-tion in 2000 with ancestors from Western Europe. The estimatesshow that the results remain robust. This finding is consistentwith the more brute-force strategy, reported in Table A8 of theappendix, that omits all European and neo-European countries(Australia, New Zealand, Canada, and the United States) fromthe sample.

Ross (2008) argues that a country’s endowment of oil reservesis an important determinant of beliefs about the role of women insociety. According to his hypothesis, oil causes a country’s domes-tic currency to strengthen, making exports less competitive andcausing a decline in light manufacturing, a sector particularlywell suited for female employment. We account for the import-ance of oil by controlling for per capita oil production in 2000. Tocapture in a more general way the phenomenon Ross (2008)argues for—namely, that a country’s economic structure can


at Harvard U



Dow

nloaded from


affect female labor force participation—we also include the shareof a country’s GDP that is in agriculture, in manufacturing, andin services. As shown in column 6, the impact of the plough re-mains robust.

The last factor that we consider is religion. For all of ourcontrol variables there is the concern that they may be endogen-ous to traditional plough agriculture and thus we are in danger ofover-controlling. However, this is particularly true for religion,which is an important part of Boserup’s (1970) hypothesizedmechanism. She argues that the plough had an impact on reli-gious practices, in particular women’s wearing of the veil and theburqa. She explains that plough cultivation ‘‘shows a predomin-antly male labor force. The land is prepared for sowing by menusing draught animals, and this . . . leaves little need for weedingthe crop, which is usually the women’s task . . . Because villagewomen work less in agriculture, a considerable fraction of themare completely freed from farm work. Sometimes such womenperform purely domestic duties, living in seclusion within theirown homes only appearing in the street wearing a veil, a phenom-enon associated with plough culture and seemingly unknown inregions of shifting cultivation where women do most of the agri-cultural toil’’ (13–14).

Braudel (1998) also argues that, historically, religious beliefswere endogenous to the practice of plough agriculture. He de-scribes how the adoption of the plough in Mesopotamia aroundthe fourth millennium BCE was accompanied not only by a move-ment of women out of agriculture and a shift from matriarchy topatriarchy, but also by a change in religious beliefs. There was ashift away from ‘‘the reign of the all-powerful mother goddessesand immemorial fertility cults presided over by priestesses’’ andtoward ‘‘male gods and priests’’ (Braudel 1998, 71). More gener-ally, religious beliefs can be interpreted as one manifestation of asociety’s overall cultural views.

With this caveat in mind, we examine how the inclusion ofreligion affects the OLS estimates of the impact of traditionalplough agriculture. In column 7, we control for five variablesthat measure the proportion of a country’s population that isCatholic, Protestant, other Christian, Muslim, and Hindu. Con-trolling for religion, traditional plough agriculture continues tohave a sizable and statistically significant impact, although theestimated magnitude decreases to 9.9 (from 12.4), a decline of20%. This is consistent with religion and its transmission over


at Harvard U



Dow

nloaded from


time being one of the mechanisms underlying the persistentimpact of traditional plough use.

In column 8, we include all covariates in one specification.Although the plough use coefficient decreases slightly, it remainshighly significant. Overall, the estimated impact of the ploughremains highly robust across the various specifications reportedin Table VII. The coefficient is always negative and statisticallysignificant, and the point estimates remain fairly stable, rangingfrom –9.2 to –14.6.

VII.B. The Importance of Ancestral Geo-Climatic Conditions

It has been hypothesized that an important determinant ofwhether the plough was adopted was the characteristics of cropsthat could be grown in a particular location (Pryor 1985). Theprimary benefit of the plough is that it facilitates the cultivationof larger amounts of land over a shorter period of time. This cap-ability is more advantageous for crops that require specific plant-ing conditions that are met only during narrow windows of timeor for crops that require larger tracts of land to cultivate a givenamount of calories. The benefit of the plough is also reducedfor crops grown in swampy, sloped, rocky, or shallow soils, all ofwhich make the plough less efficient or impossible to use. Takingthese factors into consideration, Pryor (1985) has classifiedcrops into those whose cultivation benefits greatly from the adop-tion of the plough—he calls these plough-positive crops—andthose whose cultivation benefits less—called plough-negativecrops. Plough-positive crops, which include wheat, teff,barley, and rye—tend to have shorter growing seasons andtend to be cultivated on relatively large expanses of land (percalorie of output) that tends to be flat and with deep soilthat is not too rocky or swampy. Plough-negative crops, whichinclude sorghum, maize, millet, roots, tubers, and treecrops, tend to yield more calories per acre, have longer growingseasons, and can be cultivated on more marginal land (Pryor1985, 732).

We examine Pryor’s hypothesis empirically by first measur-ing the average suitability of the ancestral environment for cul-tivating plough-positive cereal crops—wheat, barley, and rye andplough-negative cereals—foxtail millet, pearl millet, and sor-ghum. We choose these two groups of crops because they are com-parable in many other dimensions except for the extent to which


at Harvard U



Dow

nloaded from


they benefit from the use of the plough. For example, we are notcomparing cereals to ground crops, which according to Scott(2009) are important for the practice of ‘‘escape agriculture’’ andstate formation. Both sets of crops are cereals that have beencultivated in the Eastern Hemisphere since the Neolithic revolu-tion (Mazoyer and Roudart, 2006, 71–99; Crawford, 2009; Luet al. 2009). Both sets of crops require similar preparations forconsumption, all being used for flour, porridge, bread, or in bev-erages (Recklein 1987). Both sets also produce similar yields, andtherefore neither clearly dominates the other in terms of thepopulation it can support (Pryor 1985, 732).

Information on the suitability of a location for cultivatingplough-positive and plough-negative cereal crops is taken fromthe FAO’s Global Agro-Ecological Zones (GAEZ) 2002 database(Fischer et al. 2002). The database reports, for 5 arc-minute by 5arc-minute grid-cells globally, the suitability of the location forcultivating a variety of different crops. The data are constructedusing information on a location’s precipitation, frequency of wetdays, mean temperature, daily temperature range, vapor pres-sure, cloud cover, sunshine, ground-frost frequency, wind speed,soil slope, and soil characteristics. These characteristics are com-bined with the specific growing requirements of crops to producea measure of whether each crop can be grown in each location,and if so, how productively.

It is important to note that the models of crop growth arebased solely on technical requirements and constraints for cropgrowth. The model’s parameters, and the final measures, are notaffected by which crops are actually grown in a particular loca-tion. The final estimates are not simple functions of the geo-graphic characteristics used, but are based on precise, highlynon-linear crop-specific models of evapotranspiration, water bal-ance, temperature profiles, temperature growing periods, lengthof growing period, and thermal regimes. This last point is par-ticularly important as it allows us to check the robustness of ourfindings to controlling directly for important geo-climatic charac-teristics which may differ systematically in plough-positive andplough-negative environments.

We construct the instruments by first identifying the landtraditionally inhabited by each ethnic group in the EthnographicAtlas. We use all land within 200 kilometers of an ethnic group’scentroid and measure the amount of land within this area thatcan grow each of the cereal crops that comprise the


at Harvard U



Dow

nloaded from


instruments.27 Let xwe , xb

e , xre, xs

e, xfme , and xpm

e denote the amountof land that can cultivate wheat, barley, rye, sorghum, foxtail mil-let, and pearl millet, respectively. Further, let xall be the amountof land that could grow any of the crops (i.e., the amount of arableland). The ethnicity-level measures of suitability for plough-positive crops is given by: Areapos

e ¼13 ðx

we þ xb

e þ xreÞ=x

alle ; while

the ethnicity-level measure of suitability for plough-negativecrops is Areaneg

e ¼ 13 ðx

se þ xfm

e þ xpme Þ=x

alle . Intuitively, the instru-

ments measure the average suitability for each type of crop, nor-malized by the overall suitability for cultivation in general.28

Using the procedure explained by equation (1), we then con-struct district- and country-level averages of our plough-positiveand plough-negative ancestral suitability. The variables measurethe proportion of the population with ancestors that lived in cli-mates that could grow plough-positive cereals (wheat, barley, andrye) and the proportion that lived in climates that could growplough-negative cereals (sorghum, foxtail millet, and pearlmillet).

We begin by testing Pryor’s (1985) hypothesis that crop typeaffected the adoption of the plough. Panel A of Table VIII reportsestimates of the specifications from Table III, but with traditionalplough use as the dependent variable and the two crop suitabilitymeasures as additional explanatory variables. The estimatesshow that the adoption of the plough is positively correlatedwith an environment suitable for the cultivation of plough-positive cereals, but not with plough-negative cereals. In allspecifications, the difference between the two coefficients is stat-istically significant. These estimates provide confirmation forPryor’s hypothesis.

Next, we consider the relationship between crop suitabilitiesand current gender roles. Panel B of Table VIII reports thesereduced-form estimates. Consistent with their impacts onplough use shown in Panel A, we find that having an ancestralenvironment that was better able to cultivate plough-positivecrops is always associated with less-equal gender roles today.

27. The GAEZ database measures suitability as a proportion of maximum at-tainable yield. We define locations that obtain at least 40% of the maximum yield assuitable. The suitability for each crop is shown in Appendix Figures A5 and A6.

28. Our procedure assumes that the GAEZ data provide an unbiased measure ofrelative historical suitability. In a different context, Nunn and Qian (2011) provideevidence of the validity of this assumption, showing that the GAEZ suitabilitymeasure for potatoes is highly correlated with historical potato production.


at Harvard U



Dow

nloaded from


TA

BL

EV

III

CO

UN

TR

Y-L

EV

EL

2S

LS

AN

DR

ED

UC

ED-F

OR

ME

ST

IMA

TE

S

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Pan

elA

.F

irst

stage

2S

LS

esti

mate

s.D

epen

den

tvari

able

:T

rad

itio

nal

plo

ugh

use

Mea

nof

dep

.var.

0.5

30.4

40.5

40.5

1

Plo

ugh

-pos

itiv

een

vir

onm

ent

0.7

44**

*0.6

29**

*0.8

61**

*0.6

73**

*0.8

20**

*0.6

85**

*0.8

74**

*0.7

17**

*(0

.084)

(0.0

89)

(0.0

78)

(0.1

03)

(0.0

82)

(0.1

04)

(0.0

89)

(0.1

18)

Plo

ugh

-neg

ati

ve

envir

onm

ent

0.1

19

0.1

85

0.1

00

0.1

15

0.1

32

0.1

87

0.1

29

0.1

42

(0.1

22)

(0.1

33)

(0.1

66)

(0.1

71)

(0.1

30)

(0.1

41)

(0.1

81)

(0.1

88)

Equ

ali

tyof

coef

fici

ents

(p-v

alu

e)0.0

00.0

00.0

00.0

00.0

00.0

00.0

00.0

0F

-sta

t(p

lou

gh

vari

able

s)40.2

125.0

666.8

021.8

851.9

621.8

849.5

418.5

2

Dep

end

ent

vari

able

(pan

els

B&

C):

Fem

ale

labor

forc

ep

art

icip

ati

onin

2000

Sh

are

offi

rms

wit

hfe

male

own

ersh

ip,

2005–2011

Sh

are

ofp

olit

ical

pos

itio

ns

hel

dby

wom

enin

2000

Aver

age

effe

ctsi

ze(A

ES

)

Mea

nof

dep

.var.

51.1

035.0

411.8

62.3

1

Pan

elB

.R

edu

ced

-for

mes

tim

ate

s

Plo

ugh

-pos

itiv

een

vir

onm

ent

�10.6

44**

*�

11.2

99**

*�

13.1

64**

�12.6

92**

�5.8

00**

�6.8

40**

�0.6

39**

*�

0.7

74**

*(3

.816)

(4.2

85)

(5.6

10)

(6.2

14)

(2.5

34)

(2.7

90)

(0.2

14)

(0.2

88)


at Harvard U



Dow

nloaded from


TA

BL

EV

III

(CO

NT

INU

ED)

Dep

end

ent

vari

able

(pan

els

B&

C):

Fem

ale

labor

forc

ep

art

icip

ati

onin

2000

Sh

are

offi

rms

wit

hfe

male

own

ersh

ip,

2005–2011

Sh

are

ofp

olit

ical

pos

itio

ns

hel

dby

wom

enin

2000

Aver

age

effe

ctsi

ze(A

ES

)

Plo

ugh

-neg

ati

ve

envir

onm

ent

18.9

28**

*19.5

71**

*6.0

72

9.1

34

�2.9

75

�2.8

68

0.6

07

0.6

53*

(6.5

06)

(6.3

29)

(9.9

26)

(10.4

01)

(6.0

93)

(6.2

58)

(0.3

91)

(0.3

93)

Equ

ali

tyof

coef

fici

ents

(p-v

alu

e)0.0

00.0

00.0

20.0

20.5

60.4

70.0

00.0

0F

-sta

t(p

lou

gh

vari

able

s)14.8

712.4

95.4

14.4

63.4

43.4

09.1

97.1

1

Pan

elC

.S

econ

d-s

tage

2S

LS

esti

mate

s

Tra

dit

ion

al

plo

ugh

use

�21.6

30**

*�

25.0

13**

*�

17.4

86**

*�

22.6

89**

*�

6.4

60**

*�

9.7

26**

*�

0.9

18**

*�

1.3

13**

*(5

.252)

(7.5

13)

(5.5

33)

(7.6

20)

(2.3

34)

(3.7

50)

(0.2

25)

(0.3

88)

Hau

sman

test

(p-v

alu

e)0.0

20.0

40.5

60.4

00.2

20.1

00.3

30.1

6H

an

sen

J0.0

00.0

00.4

10.3

10.7

20.8

60.0

50.0

6H

isto

rica

l&

con

tem

por

ary

con

trol

syes

yes

yes

yes

yes

yes

yes

yes

Con

tin

ent

FE

sn

oyes

no

yes

no

yes

no

yes

Obse

rvati

ons

160

160

122

122

140

140

104

104

Not

es.

Tw

o-st

age

least

squ

are

ses

tim

ate

sare

rep

orte

dw

ith

robu

stst

an

dard

erro

rsin

bra

cket

s.T

he

un

itof

obse

rvati

onis

aco

un

try.

‘‘Tra

dit

ion

al

plo

ugh

use

’’is

the

esti

mate

dp

rop

orti

onof

citi

zen

sw

ith

an

cest

ors

that

trad

itio

nall

yu

sed

the

plo

ugh

inp

re-i

nd

ust

rial

agri

cult

ure

.T

he

vari

able

ran

ges

from

0to

1.

Th

em

ean

(an

dst

an

dard

dev

iati

on)

ofth

isvari

able

is0.5

30

(0.4

71);

this

corr

esp

ond

sto

the

sam

ple

from

colu

mn

s1

an

d2.

‘‘Plo

ugh

-pos

itiv

een

vir

onm

ent’

’is

the

aver

age

fract

ion

ofan

cest

ral

lan

dth

at

was

suit

able

for

gro

win

gbarl

ey,

rye,

an

dw

hea

td

ivid

edby

the

fract

ion

that

was

suit

able

for

an

ycr

ops.

‘‘Plo

ugh

-neg

ati

ve

envir

onm

ent’

’is

the

aver

age

fract

ion

ofan

cest

ral

lan

dth

at

was

suit

able

for

gro

win

gfo

xta

ilm

ille

t,p

earl

mil

let,

an

dso

rgh

um

div

ided

by

the

fract

ion

that

was

suit

able

for

an

ycr

ops.

‘‘Fem

ale

labor

forc

ep

art

icip

ati

on’’

isth

ep

erce

nta

ge

ofw

omen

inth

ela

bor

forc

e.T

he

vari

able

ran

ges

from

0to

100.

‘‘Sh

are

offi

rms

wit

hfe

male

own

ersh

ip’’

isth

ep

erce

nta

ge

offi

rms

inth

eW

orld

Ban

kE

nte

rpri

seS

urv

eys

wit

hso

me

fem

ale

own

ersh

ip.

Th

esu

rvey

sw

ere

con

du

cted

bet

wee

n2005

an

d2011,

dep

end

ing

onth

eco

un

try.

Th

evari

able

ran

ges

from

0to

100.

‘‘Sh

are

ofp

olit

ical

pos

itio

ns

hel

dby

wom

en’’

isth

ep

rop

orti

onof

seats

inp

arl

iam

ent

hel

dby

wom

en.

Th

em

easu

reals

ora

nges

from

0to

100.

‘‘His

tori

cal

con

trol

s’’in

clu

de:

an

cest

ral

suit

abil

ity

for

agri

cult

ure

,fr

act

ion

ofan

cest

ral

lan

dth

at

was

trop

ical

orsu

btr

opic

al,

an

cest

ral

dom

esti

cati

onof

larg

ean

imals

,an

cest

ral

sett

lem

ent

patt

ern

s,an

dan

cest

ral

pol

itic

al

com

ple

xit

y.

‘‘Con

tem

por

an

eou

sco

ntr

ols’

’in

clu

de

the

natu

ral

log

ofre

al

per

cap

ita

GD

Pan

dit

ssq

uare

,m

easu

red

inth

esa

me

yea

ras

the

dep

end

ent

vari

able

.T

he

nu

mber

ofob

serv

ati

ons

rep

orte

dfo

rth

eA

ES

isth

eaver

age

nu

mber

ofob

serv

ati

ons

inth

ere

gre

ssio

ns

for

the

thre

eou

tcom

es.

***,

**,

an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


A better environment for growing plough-negative crops is gen-erally associated with more-equal gender roles today.

If one is willing to assume that the difference between ances-tral suitability for plough-positive and plough-negative crops iscorrelated with gender role attitudes today (conditional on cov-ariates) only through their impact on the plough, then we can usethe two measures to provide IV estimates of the impact of trad-itional plough use on current gender attitudes. Although this as-sumption (i.e., the validity of the exclusion restriction) cannot betaken for granted, we feel that with appropriate caveats, the IVestimates are informative. Recall that the controls already in-clude the proportion of land historically inhabited by an ethnicgroup that was tropical or subtropical, as well as its overall suit-ability. In addition, further analysis discussed below flexibly con-trols for additional geographic characteristics that may becorrelated with the suitability instruments and have an inde-pendent effect on gender norms today.

One important difference between plough-positive andplough-negative crops is that in many countries today, sorghumand millet are often less preferred to rice and wheat (Nadkarni1986; Kennedy and Reardon 1994). Although the reason for this isnot yet well understood, the inferior status is not due to lowernutritional value or greater output volatility (Nadkarni 1986,A113).

Panel C of Table VIII reports the IV estimates, which confirmthe OLS estimates. Historical plough use is associated with lessfemale labor force participation, less female firm ownership, andless female participation in politics.29 The magnitudes of the IVcoefficients are consistently greater than the OLS estimates. Apotential explanation for this is selection arising from the endo-geneity of plough adoption. All else equal, historically advancedsocieties were more likely to adopt the plough.30 Furthermore,they are more likely to be advanced today, with higher percapita income and more female participation in the labormarket. Therefore, selection introduces a positive relationship

29. The results are similar if we do not control for current per capita GDP. SeeAppendix Table A12.

30. This fact can be seen in Appendix Table A13, which reports the relationshipbetween traditional plough use and various ethnicity characteristics. Political de-velopment, and to some extent economic development, as measured by settlementcomplexity, are both positively correlated with a tradition of plough use.


at Harvard U



Dow

nloaded from


between historical plough use and female labor force participa-tion today, biasing the negative OLS estimates toward zero.

The validity of the IV estimates rests on the assumption thatholding overall agricultural suitability constant, the specific typeof cereal crop that a location could grow only impacts long-termgender attitudes only through the past adoption of the plough.The primary concern with this strategy is that the difference be-tween plough-positive and plough-negative environments may becorrelated with geographic features that affect gender attitudestoday through channels other than the plough. We check therobustness of our results to this concern by controlling for geo-graphic characteristics that are potentially correlated with thesuitability of the environment for plough-positive and plough-negative crops. Our controls include the terrain slope, soildepth, average temperature, and average precipitation of loca-tions inhabited by each country’s ancestors. The estimates, re-ported in Appendix Table A14, show that the IV estimatesremain robust to the inclusion of these geo-climatic characteris-tics, controlling for the measures linearly, non-linearly, or allow-ing for interactions between the characteristics.

VIII. CULTURAL TRANSMISSION AS PART OF THE

MECHANISM: EVIDENCE FROM THE CHILDREN OF

IMMIGRANTS

We now turn to a closer analysis of the causal mechanismsunderlying our results. It is likely that part of the long-term im-pacts of traditional plough use may not be solely due to the evo-lution and persistence of cultural norms, but also to thedevelopment of institutions, policies, laws, and markets thatare less conducive to the participation of women in activities out-side the domestic sphere. Through this channel, the ploughcauses less female participation in market activities because itaffects the costs and benefits of these activities, not because itaffects individuals’ beliefs about whether these are appropriateactivities for women. Our individual-level estimates, by control-ling for country fixed effects (WVS estimates) and even subna-tional district fixed effects (IPUMS estimates), provide evidenceconsistent with a tradition of plough use having some impact oninternal values and beliefs.


at Harvard U



Dow

nloaded from


We undertake an alternative strategy to better isolate thecausal impact of ancestral plough agriculture on cultural beliefsand values, and examine variation among the children of immi-grants, a group of individuals from diverse cultural backgroundsand different histories of ancestral plough use, but facing thesame external environment, including markets, institutions,laws, and policies.

The benefit of this exercise is that by-and-large we are hold-ing constant the external environment, while examining individ-uals from diverse cultural backgrounds. One shortcoming is thatthe children of immigrants are not a random sample of the fullpopulation in the home country. Therefore, the results should beunderstood with this in mind: they are an average effect of thesample we consider. Second, if immigrants and their childrentend to live in locations with many co-ethnics, then it is possiblethat informal institutions may be recreated in these areas, ex-plaining some of the persistence of traditional plough use.Again, the results should be understood with this possibility inmind.

The analysis examines the children of immigrants living inthe United States and Europe. That is, we examine individualsborn in the country, but whose parents were born abroad. Anindividual’s ancestry is defined by the country of birth of theirparents. The US data are from the March Supplement of theCurrent Population Survey (CPS).31 The European data arefrom the European Social Survey (ESS).

VIII.A. Evidence from the United States

The sample of children of US immigrants includes womenaged 15 to 64. The outcome of interest is the labor force partici-pation of women in the sample. Because this decision is likelyvery different for married women, we also examine this groupseparately. Our estimating equation is given by

yi, s, c ¼ �s þ �Ploughc þ XCc !þ XH

c &þ Xi(þ "i, s, c,ð5Þ

where i denotes the daughter of an immigrant parent (or immi-grant parents) who currently lives in state s, with country of

31. Because the Census stopped asking questions about parents’ country oforigin in 1970, this source cannot be used for the analysis. Starting in 1994, theCPS asks individuals about their country of origin and their parents’ country oforigin. We use all the years available since 1994.


at Harvard U



Dow

nloaded from


ancestry c. Country of ancestry is defined using either themother’s country of birth or the father’s country of birth. A listof the countries of ancestry and the number of children of immi-grants for each country is provided in Appendix Table A16.

The dependent variable, yi, s, c, is an indicator variable thatequals one if woman i is in the labor market. As in equation (2),Ploughc denotes ancestral plough use of those from country c. �s

denotes state fixed effects, which control for state-varying differ-ences in labor markets, laws, regulations, institutions, etc.32 XC

c

and XHc denote the same vectors of current and historical controls

as in equation (2). Xi indicates a vector of individual-level con-trols: fixed effects for educational attainment, a quadratic for age,marital status, fixed effects for whether the person lives in ametropolitan or rural area, and fixed effects for the year of thesurvey. When we examine the sample of married women, we alsoinclude controls for characteristics of the husband: a quadratic ofthe husband’s age, the husband’s education, and the natural logof his wage income. Because our variable of interest, Ploughc,varies only at the country-of-origin level, standard errors areclustered at this level.

OLS estimates of equation (5) are reported in Table IX.Columns 1–3 report estimates using the full sample of women.Column 1 identifies an individual’s ancestry by her father’s coun-try of birth, while column 2 uses the mother’s country of birth. Incolumn 3, we restrict the sample to women whose parents wereboth born in the same country. For all three specifications, we es-timate a negative relationship between a tradition of plough use inthe home country and participation in the labor force. Columns 4–9report estimates for the sample of married women. Columns 4–6reproduce the estimates of columns 1–3 with the married sample.We continue to find a negative impact of traditional plough use onfemale labor force participation. For both specifications, we findthat the estimated impact of traditional plough use is strongerwhen both parents have the same ancestry.

We also consider the possibility that a wife’s participation inthe labor market may be influenced by her husband’s beliefs and

32. We include state fixed effects to account for the potential selection of immi-grants (and their children) to different states. As shown in Appendix Table A18, weobtain similar estimates when we exclude these fixed effects. Our European resultsare also similar with or without destination-country fixed effects. See AppendixTable A20.


at Harvard U



Dow

nloaded from


TA

BL

EIX

DE

TE

RM

INA

NT

SO

FF

EM

AL

EL

AB

OR

FO

RC

EP

AR

TIC

IPA

TIO

NF

OR

US

CH

ILD

RE

NO

FIM

MIG

RA

NT

S

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Dep

end

ent

vari

able

:L

abor

forc

ep

art

icip

ati

onin

dic

ato

r,1994–2011

All

wom

enM

arr

ied

wom

en

Wom

an

’san

cest

ryW

oman

’san

cest

ryH

usb

an

d’s

an

cest

ry

Fath

er’s

cou

ntr

yM

oth

er’s

cou

ntr

y

Pare

nts

sam

eco

un

try

Fath

er’s

cou

ntr

yM

oth

er’s

cou

ntr

y

Pare

nts

sam

eco

un

try

Fath

er’s

cou

ntr

yM

oth

er’s

cou

ntr

y

Pare

nts

sam

eco

un

try

Mea

nof

dep

.var.

0.6

30.6

30.6

00.6

80.6

90.6

90.7

00.7

10.7

0

Tra

dit

ion

al

plo

ugh

use

�0.0

44**

*�

0.0

43**

�0.0

62**

*�

0.0

94**

�0.1

18**

*�

0.1

36**

�0.0

65**

*�

0.0

45**

�0.0

58**

(0.0

15)

(0.0

18)

(0.0

20)

(0.0

46)

(0.0

43)

(0.0

54)

(0.0

24)

(0.0

22)

(0.0

24)

Obse

rvati

ons

57,1

38

55,3

41

32,7

76

10,2

06

9,5

08

6,8

35

35,3

93

35,1

58

23,1

24

Ad

just

edR

-squ

are

d0.2

30.2

30.2

50.1

00.1

00.1

10.0

80.0

80.0

8R

-squ

are

d0.2

30.2

30.2

60.1

10.1

10.1

20.0

90.0

80.0

9

Not

es.

OL

Ses

tim

ate

sare

rep

orte

dw

ith

stan

dard

erro

rscl

ust

ered

at

the

cou

ntr

yle

vel

.A

nob

serv

ati

onis

ad

au

gh

ter

ofan

imm

igra

nt

toth

eU

nit

edS

tate

s,su

rvey

edbet

wee

n1994

an

d2011.

‘‘Tra

dit

ion

al

plo

ugh

use

’’is

the

fract

ion

ofci

tize

ns

wit

han

cest

ors

that

use

dth

ep

lou

gh

inp

re-i

nd

ust

rial

agri

cult

ure

inth

efa

ther

’sco

un

try

ofor

igin

ofth

ech

ild

ren

ofim

mig

ran

ts.

Th

em

ean

(an

dst

an

dard

dev

iati

on)

for

this

vari

able

is0.5

70

(0.4

54);

this

corr

esp

ond

sto

the

sam

ple

from

colu

mn

1.

All

regre

ssio

ns

incl

ud

e:st

ate

-of-

resi

den

cefi

xed

effe

cts,

ind

ivid

ual

con

trol

s(a

ge,

age

squ

are

d,

edu

cati

onal

att

ain

men

tfi

xed

effe

cts

for

less

than

hig

hsc

hoo

l,h

igh

sch

ool,

mor

eth

an

hig

hsc

hoo

l,an

ind

icato

rvari

able

for

bei

ng

sin

gle

,yea

rof

surv

eyfi

xed

effe

cts,

an

dm

etro

pol

itan

fixed

effe

cts

for

wit

hin

met

rop

olit

an

cen

tral

city

,ou

tsid

eof

met

rop

olit

an

cen

tral

city

,an

dn

otli

vin

gin

am

etro

pol

itan

are

a),

his

tori

cal

cou

ntr

yco

ntr

ols

(an

cest

ral

suit

abil

ity

for

agri

cult

ure

,fr

act

ion

ofan

cest

ral

lan

dth

at

was

trop

ical

orsu

btr

opic

al,

an

cest

ral

dom

esti

cati

onof

larg

ean

imals

,an

cest

ral

sett

lem

ent

patt

ern

s,an

dan

cest

ral

pol

itic

al

com

ple

xit

y),

an

dco

nte

mp

oran

eou

sco

un

try

con

trol

s(t

he

natu

ral

log

ofre

al

per

cap

ita

GD

Pan

dit

ssq

uare

,m

easu

red

inth

esa

me

yea

ras

the

dep

end

ent

vari

able

).C

olu

mn

s4–9

als

oin

clu

de

hu

sban

dco

ntr

ols

(hu

sban

d’s

age,

age

squ

are

d,

hu

sban

d’s

edu

cati

onal

att

ain

men

tfi

xed

effe

cts

for

less

than

hig

hsc

hoo

l,h

igh

sch

ool,

an

dm

ore

than

hig

hsc

hoo

l,an

dh

usb

an

d’s

natu

ral

log

ofre

al

wage

inco

me)

.**

*,**

,an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


values, which were transmitted from his parents. Columns 7–9reproduce the estimates of columns 4–6, but identify ancestryusing the husband’s parents rather than the wife’s parents. Theestimates provide evidence that a tradition of plough use amongthe husband’s ancestors also affects the wife’s participation in thelabor market.33

By examining the children of immigrants, we are able to es-timate the impact of traditional plough use on FLFP, while hold-ing constant external factors that vary at the state level. Thisprovides increased confidence that the estimated impact moreclosely approximates the true impact of the plough workingthrough cultural transmission. Therefore, a comparison of themagnitudes of the children-of-immigrant estimates to thecountry-level estimates allows us to glean evidence of the import-ance of cultural transmission, relative to other channels.34 Theimpact of the plough on the daughters of immigrants to theUnited States is much smaller than the estimated impact fromthe country-level regressions. Consider the estimated impact onfemale labor force participation from a one-unit increase in his-torical plough use. Using the country-level OLS estimates(column 1 of Table IV), this is associated with an increase inFLFP by 12.4 percentage points. The magnitudes from thechildren-of-immigrant estimates are much smaller, generallyranging from 4.3 to 6.2 (for the sample of all women). Althoughone must interpret these findings with appropriate caution, theysuggest that the transmission of internal norms (as identified inthe children-of-immigrant regressions) may account for roughly35–50 percent of the total effect.

VIII.B. Evidence from Europe

We now turn to the European sample of children of immi-grants from the ESS. One advantage of the sample is that theESS, like the WVS, asks respondents their view about the state-ment ‘‘When jobs are scarce, men should have more right to a jobthan women.’’ The potential responses, however, are slightly

33. As we show in Appendix Table A17, we obtain similar results if we do notcontrol for origin country contemporaneous per capita income.

34. To facilitate such a comparison, we intentionally construct the country-level, individual-level, and children-of-immigrants estimating equations to be assimilar as possible. For example, they all include exactly the same country-levelhistorical and contemporary control variables.


at Harvard U



Dow

nloaded from


different than in the WVS. The respondents are asked to chooseamong ‘‘agree strongly,’’ ‘‘agree,’’ ‘‘neither agree nor disagree,’’ ‘‘dis-agree,’’ and ‘‘disagree strongly.’’35 From the survey question, weconstruct two measures. The first variable takes integer valuesfrom 1 to 5 and is increasing in the strength of agreement withthe statement. The second variable is an indicator that equals oneif the individual agrees or strongly agrees with the statement andzero if he or she disagrees or strongly disagrees; people who neitheragree nor disagree are excluded from the sample.36 The secondvariable, although less precise than the first, is constructed inthe same manner as the dependent variable used in theindividual-level analysis, providing the best comparison of thechildren-of-immigrant regressions with these estimates.

We estimate the impact of the plough on the children of im-migrants using the following equation:

yi, d, c ¼ �d þ �Ploughc þ XCc !þ XH

c &þ Xi(þ "i, d, c,ð6Þ

where i denotes the child of an immigrant who is currently living indestination-country d with country of ancestry c. The dependentvariable, yi, d, c is the gender role attitude measure described above.Ploughc denotes traditional plough use in country c;37 �d denotesdestination-country fixed effects; XC

c and XHc denote the same vec-

tors of current and historical controls as in equation (5); and Xi

denotes a vector of individual-level control variables that includesyears of education, age, age squared, gender, marital status, fixedeffects for city size, and fixed effects for the year of the survey.

The OLS estimates are reported in Table X. The odd-numbered columns report estimates using the dependent vari-able that ranges from one to five, and the even-numbered col-umns report estimates using the indicator variable. Columns 1–2 identify a person’s ancestry by their father’s country of birth,and columns 3–4 use the mother’s country of birth. In columns5–6, we restrict the sample to individuals for whom both parentswere born in the same country. In all six specifications, we esti-mate a positive relationship between traditional plough use and

35. The responses in the WVS are ‘‘agree,’’ ‘‘disagree,’’ ‘‘neither,’’ or ‘‘don’tknow.’’

36. The estimates for the 1-to-5 integer variable are similar if we omit peoplewho ‘‘neither agree or disagree’’.

37. The origin countries and the number of observations from each are reportedin Appendix Table A16.


at Harvard U



Dow

nloaded from


TA

BL

EX

DE

TE

RM

INA

NT

SO

FG

EN

DE

RA

TT

ITU

DE

SO

FE

UR

OP

EA

NC

HIL

DR

EN

OF

IMM

IGR

AN

TS

(1)

(2)

(3)

(4)

(5)

(6)

Dep

end

ent

vari

able

s:‘‘W

hen

jobs

are

scarc

e..

.’’su

rvey

resp

onse

,2004–2011

Fath

er’s

cou

ntr

yM

oth

er’s

cou

ntr

yS

am

eco

un

try

1–5

scale

Ind

icato

r1–5

scale

Ind

icato

r1–5

scale

Ind

icato

r

Mea

nof

dep

.var.

2.5

40.3

22.5

30.3

22.6

20.3

5

Tra

dit

ion

al

plo

ugh

use

0.2

19**

0.0

73**

0.2

14**

0.0

70**

0.2

98**

*0.0

94**

(0.0

91)

(0.0

34)

(0.0

86)

(0.0

33)

(0.0

96)

(0.0

38)

Obse

rvati

ons

15,5

45

13,0

24

15,2

60

12,7

88

10,5

35

8,7

80

Ad

just

edR

-squ

are

d0.1

80.1

60.1

70.1

60.1

70.1

6R

-squ

are

d0.1

80.1

70.1

70.1

60.1

70.1

7

Not

es.

Th

eta

ble

rep

orts

OL

Ses

tim

ate

s,w

ith

stan

dard

erro

rscl

ust

ered

at

the

cou

ntr

yle

vel

.A

nob

serv

ati

onis

the

chil

dof

an

imm

igra

nt,

rep

orte

din

thre

ew

aves

ofth

eE

uro

pea

nS

ocia

lS

urv

ey(E

SS

).T

he

thre

ew

aves

incl

ud

eth

ese

con

d(2

004–2005),

the

fou

rth

(2008–2009),

an

dth

efi

fth

(2010–2011).

‘‘Tra

dit

ion

al

plo

ugh

use

’’is

the

fract

ion

ofci

tize

ns

wit

han

cest

ors

that

use

dth

ep

lou

gh

inp

re-i

nd

ust

rial

agri

cult

ure

inth

efa

ther

’sco

un

try

ofor

igin

ofth

ech

ild

ren

ofim

mig

ran

ts.

Th

em

ean

(an

dst

an

dard

dev

iati

on)

for

this

vari

able

is0.8

91

(0.2

73);

this

corr

esp

ond

sto

the

sam

ple

from

colu

mn

1.

All

regre

ssio

ns

con

trol

for

33

cou

ntr

y-o

f-d

esti

nati

onfi

xed

effe

cts,

two

surv

ey-y

ear

fixed

effe

cts

for

3d

iffe

ren

tsu

rvey

waves

,in

div

idu

al

con

trol

s(a

ge,

age

squ

are

d,

the

nu

mber

ofyea

rsof

edu

cati

on,

agen

der

ind

icato

rvari

able

,an

ind

icato

rvari

able

for

bei

ng

sin

gle

,an

dtw

oci

tysi

zein

dic

ato

rvari

able

s),

his

tori

cal

orig

in-c

oun

try

con

trol

s(a

nce

stra

lsu

itabil

ity

for

agri

cult

ure

,fr

act

ion

ofan

cest

ral

lan

dth

at

was

trop

ical

orsu

btr

opic

al,

an

cest

ral

dom

esti

cati

onof

larg

ean

imals

,an

cest

ral

sett

lem

ent

patt

ern

s,an

dan

cest

ral

pol

itic

al

com

ple

xit

y),

an

dco

nte

mp

oran

eou

sor

igin

-cou

ntr

yco

ntr

ols

(th

en

atu

ral

log

ofre

al

per

cap

ita

GD

Pan

dit

ssq

uare

,m

easu

red

inth

esa

me

yea

ras

the

dep

end

ent

vari

able

).**

*,**

,an

d*

ind

icate

sign

ifica

nce

at

the

1%

,5%

,an

d10%

level

s.


at Harvard U



Dow

nloaded from


beliefs about gender inequality. The estimated impact of trad-itional plough use is stronger when both parents come from thesame country.38

As with the US sample, the estimated impact of the plough issmaller when examining variation across the children of immi-grants. The immigrant estimates that are most directly compar-able to the individual-level estimate (i.e., column 3 of Table V)are the even-numbered columns of Table X. The estimated impactfrom the individual-level regression is 0.193, while the impactsfrom the children-of-immigrants regressions range from 0.070 to0.094. This suggests that between 36 and 49 percent of the totalimpact identified in Table V may be explained by cultural persist-ence. This is similar to the figure for female labor force participa-tion using the US children-of-immigrants sample (35 to 50percent).

IX. CONCLUSIONS

It has long been hypothesized that the traditional use ofshifting hoe cultivation vs. plough agriculture was an importantdeterminant of the evolution and persistence of traditionalgender norms. We formally test this hypothesis by combiningethnographic data on traditional plough use with contemporarydata, measuring gender norms and female participation outsidethe domestic sphere.

Our findings provide evidence that current differences ingender attitudes and female behavior are indeed shaped by dif-ferences in traditional agricultural practices. Specifically, wehave shown that individuals, ethnicities, and countries whoseancestors engaged in plough agriculture have beliefs that exhibitgreater gender inequality today and have less female participa-tion in non-domestic activities, such as market employment, firmownership, and politics. In an effort to better identify a channel ofcultural persistence, we examined the children of immigrants.We find that among these individuals who face the same labormarkets, institutions, and policies, a heritage of traditional

38. Unlike the US estimates, the European estimates are sensitive to condi-tioning on origin-country contemporaneous per capita income. As we show inAppendix Table A19, if we do not control for this variable, the magnitude of therelationship between traditional plough use and gender attitudes is reduced byapproximately one-half.


at Harvard U



Dow

nloaded from


plough agriculture is still associated with more unequal genderattitudes and less female labor force participation.

Harvard University, IGIER Bocconi, NBER, and CEPR

University of California–Los Angeles, NBER, CEPR,

and IZA

Harvard University, NBER, and BREAD

Supplementary Material

An Online Appendix for this article can be found at QJEonline (qje.oxfordjournals.org).

REFERENCES

Albanesi, Stefania, and Claudia Olivetti, ‘‘Gender Roles and TechnologicalProgress,’’ (2007) Working Paper 13179, NBER.

———, ‘‘Home Production, Market Production, and the Gender Wage Gap:Incentives and Expectations,’’ Review of Economic Dynamics, 12 (2009),80–107.

Antecol, Heather, ‘‘An Examination of Cross-Country Differences in the GenderGap in Labor Force Participation Rates,’’ Labour Economics, 7, no. 4 (2000),409–426.

Baumann, Hermann, ‘‘The Division of Work According to Sex in African HoeCulture,’’ Africa: Journal of the International African Institute, 1, no. 3(1928), 289–319.

Becker, Sasha O., Katrin Boeckh, Christa Hainz, and Ludger Woessman, ‘‘TheEmpire Is Dead! Long Live the Empire,’’ (Mimeo, University of Stirling, 2010).

Boserup, Ester, Woman’s Role in Economic Development (London: George Allenand Unwin Ltd, 1970).

Boyd, Robert, and Peter J. Richerson Culture and the Evolutionary Process(Chicago: University of Chicago Press, 1985).

Braudel, Fernand, Mediterranean in the Ancient World (London: Allen Lane,1998).

Carranza, Eliana, ‘‘Soil Endowments, Production Technologies and MissingWomen in India,’’ (Mimeo, Harvard University, 2010).

Clingingsmith, David, Asim Ijaz Khwaja, and Michael Kremer, ‘‘Estimating theImpact of the Hajj: Religion and Tolerance in Islam’s Global Gathering,’’Quarterly Journal of Economics (2009), 1133–1170.

Crawford, Gary W., ‘‘Agricultural Origins in North China Pushed Back to thePleistocene-Holocene Boundary,’’ Proceedings of the National Academy ofSciences, 106, no. 18 (2009), 7271–7272.

Doepke, Matthias, and Michele Tertilt, ‘‘Women’s Liberation: What’s in It forMen?,’’ Quarterly Journal of Economics, 124, no. 4 (2009), 1541–1591.

Engels, Frederick, The Origin of the Family, Private Property, and the State(Chicago: Charles H. Kerr & Company Cooperative, 1902).

Estudillo, Jonna P., Agnes R. Quisumbing, and Keijiro Otsuka, ‘‘GenderDifferences in Land Inheritance, Schooling, and Lifetime Income: Evidencefrom the Rural Philippines,’’ Journal of Development Studies, 37, no. 4 (2001),23–48.

Fernandez, Raquel, ‘‘Women, Work, and Culture,’’ Journal of the EuropeanEconomic Association, 5 (2007), 305–332.

Fernandez, Raquel, and Alessandra Fogli, ‘‘Culture: An Empirical Investigation ofBeliefs, Work, and Fertility,’’ American Economic Journal: Macroeconomics,1, no. 1 (2009), 146–177.


at Harvard U



Dow

nloaded from

http://hwmaint.qje.oxfordjournals.org/cgi/content/full/doi:10.1093/qje/qjt005/-/DC1


Fischer, Gunther, Harrij van Nelthuizen, Mahendra Shah, andFreddy Nachtergaele, Global Agro-Ecological Assessment for Agriculture inthe 21st Century: Methodology and Results (Rome: Food and AgricultureOrganization of the United Nations, 2002).

Fortin, Nicole, ‘‘Gender Role Attitudes and the Labour Market Outcomes ofWomen Across OECD Countries,’’ Oxford Review of Economic Policy, 21(2005), 416–438.

Gigerenzer, Gerd, Gut Feelings: The Intelligence of the Unconscious (New York:Penguin Group, 2007).

Goldin, Claudia, ‘‘The U-Shaped Female Labor Force Function in EconomicDevelopment,’’ In Investment in Women Human Capital, Paul Schultz, T.,ed. (Chicago: University of Chicago Press, 1995), 61–90.

———, ‘‘The Quiet Revolution That Transformed Women’s EmploymentEducation and Family,’’ American Economic Review Papers andProceedings, 96, no. 2 (2006), 1–23.

Goldin, Claudia, and Kenneth Sokoloff, ‘‘The Relative Productivity Hypothesis ofIndustrialization: The American Case, 1820 to 1850,’’ Quarterly Journal ofEconomics, 99, no. 3 (1984), 461–487.

Gordon, Raymond G. Ethnologue: Languages of the World. 15th ed. (Dallas, TX:SIL International, 2005).

Grosjean, Pauline, ‘‘A History of Violence: The Culture of Honor as a Determinantof Homicide in the US South,’’ (Mimeo, University of New South Wales,2011a).

———, ‘‘The Weight of History on European Cultural Integration: A GravityApproach,’’ American Economic Review Papers and Proceedings, 101, no. 3(2011b), 504–508.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales, ‘‘Long-Term Persistence,’’(Mimeo, 2008).

Hausman, Ricardo, Laura D. Tyson, and Saadia Zahidi, The Gender Gap Report,2012 (Geneva: World Economic Forum, 2012).

Iversen, Torben, and Francis Rosenbluth, Women, Work, and Politics: ThePolitical Economy of Gender Inequality (New Haven, CT: Yale UniversityPress, 2010).

Kahneman, Daniel, Thinking, Fast and Slow (New York: Farar, Straus & Giroux,2011).

Kennedy, Eileen, and Thomas Reardon, ‘‘Shift to Non-Traditional Grains in theDiets of East and West Africa: Role of Women’s Opportunity Cost of Time,’’Food Policy, 19, no. 1 (1994), 45–56.

Kling, Jeffrey R., Jeffrey B. Liebman, and Lawrence F. Katz, ‘‘ExperimentalAnalysis of Neighborhood Effects,’’ Econometrica, 75, no. 1 (2007), 83–119.

Lu, Houyuan, Jianping Zhang, Kam biu Liu, Naiqin Wu, Yumel Li, Kunshu Zhou,Maolin Ye, Tianyu Zhang, Haijang Zhang, Xiaoyan Yang, Licheng Shen,Deke Xu, and Quan Li, ‘‘Earliest Domestication of Common Millet(Panicum Miliaceum) in East Asia Extended to 10,000 Years Ago,’’Proceedings of the National Academy of Sciences, 106, no. 18 (2009),7367–7372.

Mazoyer, Marcel, and Laurence Roudart, A History of World Agriculture: Fromthe Neolithic Age to the Current Crisis (New York: Monthly Review Press,2006).

Murdock, George Peter, and Douglas R. White, ‘‘Standard Cross-CulturalSample,’’ Ethnology, 8, no. 4 (1969), 329–369.

Nadkarni, M. V., ‘‘ ‘Backward’ Crops in Indian Agriculture: Economy of CoarseCereals and Pulses,’’ Economic and Political Weekly, 21, no. 38/39 (1986),A113–A118.

Nisbett, Richard E., and Dov Cohen, Culture of Honor: The Psychology of Violencein the South (Boulder: Westview Press, 1996).

Nunn, Nathan, and Diego Puga, ‘‘Ruggedness: The Blessing of Bad Geography inAfrica,’’ Review of Economics and Statistics, 94, no. 1 (2012), 20–36.

Nunn, Nathan, and Leonard Wantchekon, ‘‘The Slave Trade and Origins ofMistrust in Africa,’’ American Economic Review, 101, no. 7 (2011),3221–3252.


at Harvard U



Dow

nloaded from


Nunn, Nathan, and Nancy Qian, ‘‘The Potato’s Contribution to Population andUrbanization: Evidence from a Historical Experiment,’’ Quarterly Journal ofEconomics, 126, no. 2 (2011), 593–650.

Pryor, Frederic, ‘‘The Invention of the Plow,’’ Comparative Studies in Society andHistory, 27, no. 4 (1985), 727–743.

Recklein, John C., A Profile of Economic Plants (Oxford: Transaction Books,1987).

Ross, Michael L., ‘‘Oil, Islam, and Women,’’ American Political Science Review,102, no. 1 (2008), 107–123.

Scott, James C., The Art of Not Being Governed (New Haven, CT: Yale UniversityPress, 2009).

Voigtlander, Nico, and Hans-Joachim Voth, ‘‘Persecution Perpetuated: TheMedieval Origins of Anti-Semitic Violence in Nazi Germany,’’ QuarterlyJournal of Economics, 127, no. 3 (2012), 1339–1392.

Weidmann, Nils B., Jan Ketil Rod, and Lars-Erik Cederman, ‘‘RepresentingEthnic Groups in Space: A New Dataset,’’ Journal of Peace Research, 47,no. 4 (2010), 491–499.

Whyte, Martin King, The Status of Women in Pre-Industrial Societies (Princeton,NJ: Princeton University Press, 1978).

World Bank, Toward Gender Equality in East Asia and the Pacific (Washington,DC: World Bank, 2012).


at Harvard U



Dow

nloaded from


Rol de la mujer

Documents

Transcript of Rol de la mujer