On the Origins of Gender Roles: Women and the Plough

Alberto Alesina Paola Giuliano April 2011 Nathan Nunn

Abstract: This paper seeks to better understand the historic origins of current differences in norms and beliefs about the appropriate role of women in society. We test the hypothesis that traditional agricultural practices inuenced the historic gender division of labor and the evolution and persistence of gender norms. We nd that, consistent with existing hypotheses, the descendants of pre-industrial societies that practiced plough agriculture, today have lower rates of female participation in the work place, in politics, and in entrepreneurial activities, as well as attitudes reecting gender inequality. We identify the causal impact of traditional plough use on gender norms today by exploiting variation in the historic geo-climatic suitability of the environment for growing crops that differentially beneted from the adoption of the plough. Our IV estimates, based on this variation, support the ndings from OLS. To isolate the importance of cultural transmission as a mechanism, we examine female labor force participation of second generation immigrants living within the US.

thank Samuel Bowles, David Clingingsmith, Pauline Grosjean, Judith Hellerstein, Edward Miguel, as well as seminar participants at the Bank of Italy, Brown University, Harvard University, Hong Kong University of Science and Technology, MIT, New York University, Sciences Po, UCLA Kaler Meeting, University of Oklahoma, Washington University St. Louis, World Bank, Stanfords SITE Conference, Coevolution of Behaviors and Institutions Conference, AEA Annual Meetings, Brooking Africa Growth Forum, and the IZA/Science Po Workshop on Trust, Civic Spirit and Economic Performance for valuable comments. We also thank Eva Ng for excellent research assistance. Giuliano gratefully acknowledges support from the UCLA Senate. Harvard University, IGIER Bocconi, NBER and CEPR. (email: aalesina@harvard.edu) University of California Los Angeles, NBER, CEPR and IZA. (email: paola.giuliano@anderson.ucla.edu) Harvard University, NBER and BREAD. (email: nnunn@fas.harvard.edu)

We

1. Introduction
The role of women in the family, in the work force, and in society varies across nations. In some cultures the social norm is for women to work outside the house, while in others the norm is for women to remain within the home, not actively participating in activities outside of the domestic sphere. This study seeks to better understand the reasons underlying these differences. We test the hypothesis, originally put forth by Boserup (1970), that cross-cultural differences in gender role norms and attitudes arose from differences in agricultural technologies used traditionally. In particular, she identies important differences between shifting cultivation and plough cultivation. Shifting cultivation, which uses hand-held tools like the hoe and the digging stick, is labor intensive and women actively participate in farm work. Plough cultivation, by contrast, is much more capital intensive, using the plough to prepare the soil. Unlike the hoe or digging stick, the plough requires signicant upper body strength, grip strength, and burst of power, which are needed to either pull the plough or control the animal that pulls it.1 Because of these requirements, when plough agriculture is practiced, men have an advantage in farming relative to women (Murdock and Provost, 1973a). Also reinforcing this gender-bias in ability is the fact that when the plough is used, there is less need for weeding, a task typically undertaken by women and children (Foster and Rosenzweig, 1996). In addition, child care, a task almost universally performed by women, is most compatible with activities that can be stopped and resumed easily and do not put children in danger. These are characteristics that hold for hoe agriculture, but not for plough agriculture, especially if animals are used to pull the plough. The result is that societies that traditionally practiced plough agriculture rather than shifting cultivation developed a specialization of production along gender lines. Men tended to work outside of the home in the elds, while women specialized in activities within the home.2 This
1 See Pitt, Rosenzweig and Hassan (2010) for evidence from Bangladesh and the USA on the distribution of strength by gender. 2 Boserup (1970) in her analysis clearly describes the relationship between traditional plough use and gender norms, even hypothesizing that the use of the veil may be associated with traditional plough use. She writes that plough cultivation shows a predominantly male labor force. The land is prepared for sowing by men using draught animals, and this. . . leaves little need for weeding the crop, which is usually the womens task. . . Because village women work less in agriculture, a considerable fraction of them are completely freed from farm work. Sometimes such women perform purely domestic duties, living in seclusion within their own homes only appearing in the street wearing a veil, a phenomenon associated with plough culture and seemingly unknown in regions of shifting cultivation where women do most of the agricultural toil. (pp. 1314)

division of labor then generated norms about the appropriate role of women in society.3 Societies characterized by plough agriculture, and the resulting gender division of labor, developed the belief that the natural place for women is within the home. These cultural beliefs tend to persist even even if the economy moves out of agriculture, affecting the participation of women on activities performed outside of the home, such as market employment, entrepreneurship, or participation in politics. To test Boserups hypothesis, we combine pre-industrial ethnographic data, reporting whether societies traditionally used plough agriculture with contemporary measures of individuals views about gender roles, as well as measures of female participation in activities outside of the home. Our analysis examines variation across countries, ethnic groups, and individuals. Consistent with Boserups hypothesis, we nd a strong and robust negative relationship between historic plough-use and attitudes of gender equality today. Traditional plough-use is positivly correlated with attitudes reecting gender inequality and negatively correlated with female labor force participation, female rm ownership, and female participation in politics. Although these ndings support Boserups hypothesis, they are also consistent with other interpretations. For example, we would observe the same relationships if societies with attitudes favoring gender inequality were more likely to adopt the plough historically and if these attitudes continue to persist today. To better understand whether past plough use did have a causal impact on subsequent cultural norms, we instrument historic plough-use using specic geo-climatic conditions of a societys historic location which affected the relative benets of adopting the plough. As Pryor (1985) shows, the benet of the plough depends on the crop being cultivated. The plough is more benecial for crops that require large tracts of land to be prepared in a short period of time (e.g., due to multiple-cropping), and can only be grown in soils that are not shallow, sloped, or rocky.4 These crops, 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 and various types of root and tree crops, which require less land to be prepared over a longer period of time, and/or can be cultivated on thin, sloped or rocky soils,
3 Prior to Boserup anthropologists and ethnographers had recognized a relationship between traditional gender roles and the use of the hoe (Baumann, 1928). However, Boserup was the rst to argue for the impact of hoe and plough use on the subsequent evolution of norms and values, and their importance for the development process. 4 For a recent study documenting the link between soil type and plough-use in modern India see Carranza (2010). In particular, she shows that plough technology is more likely to be adopted with deep loamy soils rather than shallow clay soils.

where using the plough is difcult. Unlike plough-positive crops, plough-negative crops benet much less from the adoption of the plough. Using data from the FAO, we identify the geo-climatic suitability of nely dened locations for growing plough-positive cereals (wheat, barley and rye) and plough-negative cereals (sorghum and millet). We then use the relative differences in ethnic groups geo-climatic conditions for growing plough-positive and plough-negative cereals as instruments for historic plough use. We nd that the IV estimates provide results consistent with the OLS estimates. Traditional plough use is associated with attitude of gender inequality, as well as less female labor force participation, female rm-ownership, and female participation in politics. Our analysis then considers potential underlying mechanisms. It is possible that the long-term effect of the plough reects persistent cultural beliefs. However, it is also possible that part of the long-term impact arises because historic plough-use promoted the development of institutions, policies and markets that are less conducive to the participation of women in activities outside of the home.5 To distinguish these two channels we exploit the fact that cultural norms and beliefs unlike institutions, policies and markets are internal to the individual. Therefore, when individuals move, their beliefs and values move with them, but their external environment remains behind. Exploiting this fact, we examine variation in cultural heritage among second generation immigrants living in the US. All individuals born and raised in the US have been exposed to the same institutions and markets. In effect, the analysis holds all external factors constant, while examining variation in individuals internal beliefs and values. We nd that women from cultures that historically used the plough have lower rates of labor force participation in the US. This provides evidence that part of the importance of the plough arises through its impact on internal beliefs and values. The relationship between traditional plough use and gender roles has been well-studied in the elds of history, anthropology and sociology. Since Boserups initial hypothesis, various scholars have, through qualitative analysis, examined the relationship between the plough and attitudes toward gender roles (Goody, 1976, Whyte, 1978 and Braudel, 1998). A particularly interesting case is Braudels (1998) description of how gender relations, culture, and society were impacted by the adoption of the plough in Mesopotamia between 4,000 and 6,000 BC. He writes: Until
5 See the recent studies by Alesina, Algan, Cahuc and Giuliano (2010), Guiso, Sapienza and Zingales (2008b) and Tabellini (2008) that investigate feedback effects between culture and institutions.

now, women had been in charge of the elds and gardens where cereals were grown: everything had depended on their tilling the soil and tending the crop. Men had been rst hunters, then herdsmen. But now men took over the plough, which they alone were allowed to use. At a stroke, it might seem that the society would move from being matriarchal to patriarchal: that there would be a shift away from the reign of the all-powerful mother goddesses. . . and towards the male gods and priests who were predominant in Sumer and Babylon. . . and was accompanied with a move towards male domination of society and its beliefs (p. 71). Our focus on a historical determinant of gender roles is not meant to imply that other factors, particularly factors that can change signicantly over time, are unimportant. A number of existing studies have examined other important determinants, including economic development, medical progress, and the production structure of the economy (e.g., Iversen and Rosenbluth, 2010, Goldin, 2006, Ross, 2008, and Albanesi and Olivetti, 2007, 2009).6 As we show in section 4, even accounting for these important factors, there remains a strong persistent impact of the plough on gender norms today. A logical implication of our nding of a deep historical determinant of gender norms is that there must be some persistence in gender norms and female activity outside of the home over time. We provide evidence for this in section 4 where we show a very strong positive relationship between pre-industrial female participation in agriculture and female labor force participation today. Our ndings add to a recent line of research that has emphasized the importance of cultural norms and beliefs as important factors underlying the persistent differences in gender roles across societies (Alesina and Giuliano, 2010, Fernandez, 2007, Fernandez and Fogli, 2009, and Fortin, 2005, 2009). Although the link between gender norms and female labor force participation is wellestablished, little is known about the origin of these cultural differences. Our ndings suggest that an important determinant of these differences is the nature of traditional farming practices. Our ndings also provide an example of how historic factors shape the evolution and persistent of norms and beliefs. Thus, they contribute a number of recent studies that also seek to explain the historic determinants of various cultural characteristics today. For instance, Guiso, Sapienza and
of the more novel hypotheses is provided by Ross who argues that countries that specialize in oil production crowd out the production of low-end export-oriented manufacturing activities, like textiles and footwear, which are particularly well-suited for female employment. Therefore, specialization in oil results in less female labor force participation and attitudes of gender inequality. We test explicitly for Ross hypothesis in our analysis.
6 One

Zingales (2008a) test Robert Putnams hypothesis of the historic origins of regional differences of social capital and trust within Italy.7 Nunn and Wantchekon (2011) examine the historic roots of mistrust within sub-Saharan Africa. Grosjean (2010b) examines the historical origins of a culture of honor in the US South, and Grosjean (2010a) and Becker, Boeckh, Hainz and Woessman (2010) examine the lasting impact that historic empires had on cultural outcomes. We begin our analysis by rst documenting that in societies that traditionally used plough agriculture women did in fact participate less in farm-work and other activities outside of the domestic sphere. In section 3, we then explain the procedure used to link the historical use of the plough, which is measured at the ethnicity level, to current data on gender norms or female labor force participation, measured either at the country or individual level. Section 4 and 5 report OLS and IV estimates of the relationship between traditional plough use and gender outcomes today, examining variation across individuals and countries. In section 6, we then turn to mechanisms, using second generation US immigrants to test for persistent impacts of the plough arising through cultural transmission. Section 7 offers concluding thoughts.

2. The historic impacts of traditional plough use

We begin our analysis by rst conrming that societies that traditionally used plough agriculture had lower female participation in agricultural activities. We also check whether plough use was associated with differences in other activities within and outside of the domestic sphere. Our analysis relies on information on traditional plough use taken from the Ethnographic Atlas, a world wide ethnicity-level database constructed by George Peter Murdock that contains ethnographic information for 1,267 ethnic groups around the world. Information for societies in the sample have been coded for the earliest period for which satisfactory ethnographic data are available or can be reconstructed. The earliest observation dates are for groups in the Old World where early written evidence is available. For the parts of the world without a written history the information is from the earliest observers of these cultures. For some cultures the rst recorded information is from the early 20th century. However, even for these observations, the data should capture, to the maximum extent possible, the characteristics of the ethnic group prior to European contact. For all groups in the dataset, the variables are taken from the societies prior to industrialization.
7 See

also Guiso, Sapienza and Zingales (2004) on social capital and nancial development.

The database contains a measure of the historic use of plough agriculture. Groups are classied into one of three mutually exclusive categories: (i) the plough was absent, (ii) the plough existed at the time the group was observed but it was not aboriginal, and (iii) the plough was aboriginal and found in the society prior to contact. There are data on plough use for 1,158 of the 1,267 societies in the database. There is hardly any evidence of groups repeatedly switching from one form of agriculture to another. In other words, the use (or non-use) of the plough remains stable over time. The database does record adoption if it occurred after European contact. However, we do not have the exact date of adoption for the other cases of adoption. It is possible that the plough has a bigger effect on gender norms amongst groups that adopted early, and therefore have used the technology for a longer period of time. However, because of data limitation, we are unable to test for this. Therefore our estimates should be interpreted as the average effect of having adopted the plough among all ethnic groups that did so prior to industrialization. There may be heterogeneity within the group of adopters, but we are only able to estimate an average effect. The number of societies that did not use the plough is greater than the number that did. Descriptive statistics for all the data used in the paper are reported in Appendix Table A1. In the sample, 86% of the ethnicities did not use the plough, 12.18% of the societies used the plough, and in 1.5% of the societies the plough was not initially used, but it was adopted after European contact. However, this actually provides an inaccurate description of the extent of plough use historically. First of all, the database under-samples European ethnic groups. Second, ethnic groups that adopted the plough were larger historically, and are larger today. For example, many of the ethnic groups that did not adopt the plough are indigenous groups located in the Americas, with small populations historically and even smaller populations today. More generally, the ethnic groups are not of equal size or importance today as compared to the historical period to which the Ethnographic Atlas refers to. For our analysis (as we describe below) we rst link the historical data to information about the current population distributions of ethnic groups, as a second step we link the information about the population weighted distribution on the use of the plough to contemporary datasets on female labor force participation and gender role attitudes. Our analysis is therefore not biased by the fact that the Ethnographic Atlas over-samples small groups or groups that are less populous today. In addition, in past centuries there have been signicant migrations of groups, particularly Europeans and Africans across the Atlantic. Our analysis also takes this

into account, since we match the ethnographic data to current outcome data based on ethnic groups and not geographic locations and we are therefore able to follow ethnic groups that have moved. We construct a plough indicator variable that takes on the value of one if the plough was present (whether aboriginal or not) among the ethnic groups and zero otherwise. Nineteen percent of societies in the sample used the plough. Female labor force participation is a categorical variable which is increasing in the degree of participation of women in agriculture. In particular, the variable indicates whether agriculture is a male or female dominated activity and can take the following values: (1) males only, (2) males appreciably more, (3) equal participation, (4) female appreciably more, and (5) females only.8 Thirty two percent of ethnic groups historically had either mostly men or only men working in agriculture, 32 percent had equal participation and 36 percent had either mostly female or only female participation.9 We estimate an OLS regression of female participation in agriculture on the presence of the plough. In all specications, we control for the presence of domesticated bovine or equine animals since low participation of women in agriculture could be due to the female monopoly over the care of domesticated animals. This variable equals one if the ethnic group has bovine or equine animals as predominant type of animal husbandry. We also include measures of economic and political complexity of the ethnic groups. Economic complexity is measured by a variable (increasing in the level of economic complexity) indicating the settlement pattern of the ethnic group. This variable can take the following values: (1) nomadic or fully migratory, (2) semi-nomadic, (3) semi-sedentary, (4) compact but not permanent settlements, neighborhoods of dispersed family homesteads, (5) separate hamlets, (6) forming a single community, (7) compact and relatively permanent settlements and (8) complex settlements. We proxy for political complexity with a variable that measures the number of levels of jurisdictional hierarchy beyond the local community. The two variables have been shown to be correlated with economic development and societal complexity (Murdock and Provost, 1973b). OLS results are reported in Table 1. Column 1 shows a negative relationship between historic plough use and historic participation of women in agriculture. A one standard deviation increase
original categorization from the Ethnographic Atlas distinguished between differentiated but equal participation and equal participation. Since this distinction is not relevant for our purposes, we combine the two categories into one. 9 Information on female participation in agriculture is missing for 547 observations in the sample. For 232 ethnic groups agriculture was not practiced and for 315 groups the data are missing.
8 The

in the use of the plough, implies a reduction in female participation in agriculture of 0.30 (10% of the sample average of the left hand side variable). The Ethnographic Atlas unfortunately does not provide any detail on the type of tasks women do when they work in agriculture. We therefore complement our analysis by using Murdock and Whites (1969) Standard Cross-Cultural Sample (SCCS), a dataset containing ethnographic information on 186 societies, intentionally chosen to be representative and historically and culturally independent from one another. The authors group the 1267 societies from the Ethnographic Atlas into 186 clusters of closely related cultures. They then chose, for each, a particularly well-documented and representative culture to be an observation in the SCCS. Using the SCCS data, we rst replicate the regressions using the Ethnographic Atlas. As shown in column 2, we nd similar results.10 In columns 314, we then look specically at gender role specialization in the following specic tasks: land clearance, soil preparation, planting, crop tending, harvesting, care of small and large animals, milking, cooking, fuel gathering, water fetching and burden carrying. Each specialization variable is coded on a 1 to 5 integer scale, increasing in the participation of women: (1) male exclusively, (2) males predominantly, (3) equal division, (4) females predominantly, and (5) females exclusively. We nd that if the plough was used, women tended to participate signicantly less in the primary agricultural activities, including land clearance soil preparation, planting, crop tending and harvesting (although for land clearance the coefcient is smaller in magnitude and statistically insignicant). This is consistent with Boserups assertion that traditional plough use was associated with less female participation in agriculture outside of the home. We also nd evidence that women in plough societies tended to participate less in other activities performed outside the house, including fuel gathering, water fetching, and burden carrying (although the coefcient for water fetching is smaller and not statistically different from zero). We dont nd statistically signicant evidence that women in plough societies increased their participation in other activities, including caring for large or small animals, milking or cooking. For these activities the estimated coefcients are positive but always statistically insignicant. Overall, the ethnographic evidence conrms that women participated less in farm activities in societies that historically practiced plough agriculture. This is consistent with the analysis of
10 The magnitude of the coefcient is slightly higher. An increase in one standard deviation in the use of the plough implies a decline in female labor force participation of 0.41, which is roughly equal to 14% of the sample average of this variable in the SCCS.

Table 1: Historic plough use and historic female participation in agriculture.

Panel A. Dependent variables: Female par.cipa.on in the following (agriculture-related) tasks: Par.cipa.on in agriculture (1) (2) -0.780*** -1.095*** (0.108) (0.250) 698 132 0.098 0.171 Land clearance (3) -0.280 (0.204) 137 0.040 Soil prepara.on (4) -1.055*** (0.353) 132 0.092 Plan.ng (5) -1.150*** (0.342) 139 0.097 Crop tending (6) -0.895** (0.367) 129 0.148 Harves.ng (7) -0.704** (0.307) 139 0.156

Historic plough use Observa.ons R-squared

Panel B. Dependent variables: Female par.cipa.on in the following (addi.onal) tasks: Caring for Caring for large small animals animals (8) (9) 0.349 0.220 (0.560) (0.276) 95 96 0.034 0.049 Milking (10) 0.738 (0.711) 48 0.030 Cooking (11) 0.085 (0.149) 182 0.019 Fuel gathering Water fetching (12) (13) -0.940** -0.219 (0.410) (0.240) 166 159 0.041 0.046 Burden carrying (14) -1.160*** (0.374) 144 0.151

Historic plough use Observa.ons R-squared

Notes: The un unit of observa.on is ethnicity. Coecients are reported with robust standard errors in brackets. ***, ** and * indicate signicance at the 1, 5 and 10% levels. Column 1 reports evidence from the Ethnographic Atlas; Columns 2-14 report evidence from the Standard Cross Cultural Sample. Par.cipa.on in agriculture is a variable taking on integer values between 1 and 5 that quan.es sex dierences in agriculture and is increasing in female par.cipa.on in agriculture. Columns 3-14 are variables taking on interger values between 1 and 5 and are increasing in female par.cipa.on of the ac.vity.

Boserup (1970), as well as the observations of anthropologists like Baumann (1928) and Whyte (1978).

3. Linking the past to the present: Data and methodology

We now turn to an examination of the long-term impact of historical plough use. We link the historic ethnographic data, measured at the ethnicity level, with our outcomes of interest, measured at the location-level, either countries or districts within countries today. To do this, we need an estimate of the location and distribution of ethnicities across the globe today. We construct this information using two datasets: the 15th edition of the Ethnologue: Languages of the World (Gordon, 2005) and the Landscan 2000 database. The former reports the current geographic distribution of 7,612 different languages, each of which we manually matched to the appropriate ethnic group from the Ethnographic Atlas. The database provides a shape le that divides the worlds land into polygons, with each polygon indicating the location of a specic language. We also use the Landscan 2000 database, which reports estimates of the worlds population for 30 arc-second by 30 arc-second (roughly 1km by 1km) grid-cells globally.11 We combine the
11 The Landscan 2000 database was produced by Oakridge Laboratories in cooperation with the US Government and NASA.

Ethnologue shape le with the Landscan raster le to obtain an estimate of the global distribution of language groups across the globe today. This information is then used to link the historic ethnicity-level data to our current outcomes of interest, measured at the location-level. We illustrate our procedure with the example of Ethiopia. Figure 1a shows a map of the land inhabited by different ethnic groups, i.e. groups speaking different languages. Each polygon represents the approximate borders of a group (from Ethnologue). One should not think of the borders as precisely dened boundaries, but rather as rough measures indicating the approximate locations of different language groups. The map also shows the Landscan estimate of the population of each cell within the country. A darker shade indicates greater population. From the Ethnographic Atlas we know whether each ethnic group used the plough. Dene
plough Ie to be a variable equal to one if ethnic group e used plough agriculture and zero otherwise.

We rst match to each of the 7,612 language groups globally, one of the 1,267 ethnic groups for which we have plough-use data. After the matching procedure, we know for each language group whether their ancestors engaged in plough agriculture. This information is shown in gure 1b. We then use information on the location of modern district and country boundaries to construct district-level and country-level averages of the historic plough measure. The procedure is shown visually for the district-level averages in gures 2a and 2b. Intuitively, the procedure creates a population-weighted average plough measure for all grid-cells within a district. This provides an estimate of the fraction of the population currently living in a district (or country) with ancestors that traditionally engaged in plough agriculture. To be more precise, let Ne,i,d,c denote the number of individuals of ethnicity e living in grid-cell
plough i located in district d in country c. We then construct a population-weighted average of Ie for

all ethnic groups living in a district d. The district-level measure of the fraction of the population with ancestors that traditionally used the plough, Ploughd,c , is given by: Ploughd,c =

e i

Ne,i,d,c plough Ie Nd,c

(1)

where Nd,c is the total number of people living in district d in country c. The same procedure is used to construct a country-level measure Ploughc as well, except that an average is taken over all grid-cells in country c. Figure 3a shows the global distribution of languages based on the Ethnologue data, as well as historic plough use for each group. The gure also shows inhabited land in dark grey. One

10

Ethnologue languages

Legend

4
(a) Population density and language groups
Ethnologue languages Plough not used Plough used

Legend

4
(b) Population density, language groups and their traditional plough use

Figure 1: Populations, language groups, and historic plough-use within Ethiopia. 11

District boundaries District boundaries Ethnologue languages Plough not used Plough used

Legend

4
(a) Population density, language groups their traditional plough use, and districts today
Historic plough use 0.00-0.01 0.01-0.39 0.39-0.76 0.76-0.94 0.94-1.00

Legend

4
(b) District averages of plough use among inhabitants ancestors

Figure 2: Traditional plough-use across districts within Ethiopia 12

problem with the Ethnologue data is that the information is missing for some parts of the world. This is due to uncertainty or a lack of information about the boundaries of language groups in that location. As it is apparent from the map, this primarily occurs in South America. We undertake three strategies in order to address this issue. The rst is to ignore the missing languages and calculate country and district measures using the data that exist (i.e., shown in Figure 3a.) This is the strategy that has been undertaken by other studies using the Ethnologue language data (e.g., Michalopoulos (2008)). Our second strategy is to assume that all inhabitants in the unclassied territories speak the national language of the country. The spatial distribution of historic plough use using this imputation procedures is reported in gure 3b. Our third strategy is to impute the language of the inhabitants using information on the spatial distribution of ethnic groups from the Geo-Referencing of Ethnic Groups (GREG) database (Weidmann, Rod and Cederman, 2010). Like the Ethnologue, the GREG database provides a shape le that divides the worlds land into polygons, with each polygon indicating the location of a specic ethnicity. The shortcoming of the GREG database is that ethnic groups are much less nely identied relative to the Ethnologue database. The GREG database identies 1,364 ethnic groups, while the Ethnologue identies 7,612 ethnic groups.12 The spatial distribution of historic plough use using this procedures is shown in gure 3c. In gures 4a4c, we report population weighted country-level averages of historic plough use for each of the three strategies used to address the missing language data. Some general patterns appear no matter which methodology we choose. Groups within sub-Saharan Africa generally did not use the plough. The majority of the European countries used the plough historically, together with some African countries like Eritrea, Ethiopia and the countries of Northern and Southern Africa, as well as a number of Asian countries. In our analysis we use the plough variable without missing values imputed as our baseline measure. We also show that our results are robust to the use of either variable that imputes the missing language data. This robustness is explained by the high correlation among the three plough measures. At the country-level, the correlation between: (i) our baseline variable and the measure with missing languages imputed using the countrys national language is 0.89; (ii) our baseline measure and the measure imputed using ethnic groups from the GREG database is 0.91;
12 An alternative strategy is to rely only on the coarser GREG classication and map. Our results are robust to this procedure as well.

13

Legend

Historic plough use

No plough use

Missing plough data Plough use, not indigenous

Missing language data

Unpopulated land Populated but no Ethnologue data

Indigenous plough use

(a) Missing language information not imputed

Legend

Historic plough use

No plough use

Missing plough data Plough use, but not indigenous

Missing language data

Unpopulated Populated but no language data

Indigenous plough use

(b) Missing language information imputed using the countrys ofcial language

Legend

Historic plough use

Missing plough data No plough use Plough use, but not indigenous

Missing language data

Unpopulated Populated but no language data

Indigenous plough use

(c) Missing language information imputed using GREG ethnic groups

Figure 3: Historic plough use among the ethnic/language groups in the Ethnologue 14

Legend

Historic plough use

0.000000 - 0.024938 0.024939 - 0.133554 0.133555 - 0.233165 0.233166 - 0.323672 0.323673 - 0.569498 0.569499 - 0.821691 0.821692 - 0.889259 0.889260 - 0.943730 0.943731 - 0.985092 0.985093 - 1.000000

(a) Missing language information not imputed

Legend

Historic plough use

0.000000 - 0.061609 0.061610 - 0.251907 0.251908 - 0.403964 0.403965 - 0.573113 0.573114 - 0.676111

0.676112 - 0.783906

0.783907 - 0.874715 0.874716 - 0.928645 0.928646 - 0.981442 0.981443 - 1.000000

(b) Missing language information imputed using the countrys ofcial language

Legend

Historic plough use

0.000000 - 0.043163 0.043164 - 0.133554 0.133555 - 0.323672 0.323673 - 0.629320 0.629321 - 0.803035 0.803036 - 0.870872 0.870873 - 0.908499 0.908500 - 0.953173 0.953174 - 0.985101 0.985102 - 1.000000

(c) Missing language information imputed using GREG ethnic groups

Figure 4: Average historic plough use among the ancestors of each country 15

and (iii) the two variables with imputed values is 0.99.13

4. OLS estimates
Having constructed country and district-level measures of traditional plough use, we are now able to examine the relationship between historic plough use and measures of the role of women in society today. We begin by examining variation at the country level.

A. Country-level estimates We test our hypothesis by estimating the following equation:

C H yc = + Ploughc + Xc + Xc + c

(2)

where y is the outcome of interest, c denotes countries, Ploughc is our measure of the historic
C H use of the plough among the ancestors of the citizens in country c, and Xc and Xc are vectors of C current controls and historic ethnographic controls, all measured at the country level. Xc includes

the natural log of a countrys real per capita GDP measured in 2000, as well as the variable squared. This is important since economic development is known to be non-linearly associated with female labor force participation (Goldin, 1995). We also include an indicator variable that equals one if the country was formerly communist, since these regimes implemented policies to
H eliminate gender differences in the economy.14 The historic ethnographic controls Xc include

agricultural suitability, the presence of domesticated bovine or equine animals, the presence of a tropical climate (either tropical or subtropical), the levels of jurisdictional hierarchy beyond the local community, and the economic development of the ethnic groups currently living within the country (dened above). We construct these variables using the same manner used to construct the historic plough use variable. The measures capture the historic characteristics of the ancestors of those in our sample. A more detailed description of each control variable is provided in the papers appendix. Table 2 reports the country-level OLS estimates. In columns 1 and 2, the dependent variable is a countrys female labor force participation rate in 2000. In columns 36, we examine womens
13 Descriptive 14 Alesina

statistics for the three measures are shown in Appendix Table A1. and Fuchs-Schundeln (2007) show how the impact of communist regimes on individual beliefs can be long

lasting.

16

participation in more narrowly specied activities outside of the domestic sphere: entrepreneurship (measured by the share of rms with owners or managers that are female) and national politics (measured by the proportion of seats held by women in national parliament)15 . The even numbered columns include controls for continent xed effects, while the odd numbered columns do not. The estimates show that in countries with a tradition of plough-use, women are less likely to participate in the labor market, are less likely to own or manage rms, and are less likely to participate in politics.16 The point estimates (using the odd numbered columns) suggest that an increase in one standard deviation in plough-use (0.474 for the full sample) is associated with a reduction of female labor force participation of (15.506 0.474 =) 7.35 (equivalent to 14.2% of the sample average for FLFP and 47% of the standard deviation); of 5.23 of the share of rms with some female ownership (16% of the sample average and 38% of the standard deviation); and a reduction of the participation of women in politics by 2.66 (22% of the sample average and 30% of the standard deviation). Columns 7 and 8 report the estimated average effect size (AES) for the three dependent variables examined in columns 16. We computed the AES following Kling, Liebman, Katz and Sanbonmatsu (2004). Let k indicate the estimated plough coefcient for outcome variable k and k the standard deviation of outcome k, the average effect size is equal to
1 K

K 1 k=

k , k

where K

is the total number of outcome variables. To properly calculate the sample variance of the AES, the coefcients k are jointly estimated in a seemingly unrelated regression framework.17 AES estimates reduce the possibility of Type I (that results on any of our outcomes is due to chance) and Type II error (the risk of low statistical power). The AES estimates conrm the ndings when examining the outcomes individually: historic plough use is associated with less female participation in activities outside of the home. As well, the implied magnitudes are similar. According to the AES estimate, a one standard deviation increase in plough use is associated with an average decrease (for the three outcomes) of 0.40 standard deviations. An alternative way to assess the magnitude of the estimates is to calculate the proportion of
about the sources of the left hand side variables are reported in the Appendix. female participation in national politics may be affected by the type of government, we also control for each countrys level of democracy in 2000 when this outcome is examined. The extent of democracy is measured using the polity2 measure from the Polity IV database, which is a variable that takes on integer values ranging from 10 (high autocratic) to +10 (highly democratic). 17 See Clingingsmith, Khwaja and Kremer (2009) for an alternative application and further details.
16 Because 15 Details

17

Table 2: Country level OLS estimates.

Female labor force participation (1) (2) Historic plough use Historic controls: Agricultural suitability Domesticated animals Tropics Political hierarchies Economic complexity Contemporary controls: ln income, ln income2 Communism indicator Polity2 Continent fixed effects Observations R-squared -16.506*** (3.547) yes yes yes yes yes yes yes no no 159 0.412 -15.417*** (3.561) yes yes yes yes yes yes yes no yes 159 0.429 Dependent variable: Share of firms with some female ownership Females in politics (3) (5) (4) (6) -11.052** (4.287) yes yes yes yes yes yes yes no no 105 0.154 -11.540** (5.152) yes yes yes yes yes yes yes no yes 105 0.205 -5.606*** (2.128) yes yes yes yes yes yes yes yes no 125 0.279 -4.245* (2.218) yes yes yes yes yes yes yes yes yes 125 0.315

Average effect size (AES) (7) (8) -0.849*** (0.140) yes yes yes yes yes yes yes no no 135
a

-0.796*** (0.137) yes yes yes yes yes yes yes no yes 135a

Notes : OLS estimates are reported with robust standard errors in brackets. The unit of observation is a country. ***, ** and * indicate significance at the 1, 5 and 10% levels. aThis is the average number of observations in the regressions for the three outcomes.

the total variation they explain. By this metric as well, historic plough use explains a sizable proportion of differences in gender roles across countries. When female labor force participation is the dependent variable (column 1 of Table 2), the inclusion of the historic plough use variable increases the R-squared by 0.086 (from 0.326 to 0.412). Therefore, traditional plough use accounts for 8.6% of the total variation in FLFP and 12.8% of the residual variation in FLFP that is unaccounted for by our control variables.18 For the share of rms with female ownership,

traditional plough use accounts for 5% of the total variation and 6% of the residual variation. For the participation of women in politics, historic plough use explains 3% of the total variation and 4% of the residual variation. The estimated coefcients for our control variables are generally as expected. For example we nd evidence of a U-shaped relationship between per capita income and female labor force participation, as well as the other outcomes. This is consistent with previous studies that also nd this same non-monotonic relationship (Goldin, 1995). We also nd that countries that experience a period of communism have higher rates of female labor force participation.
18 This

is calculated as: (0.412 0.326)/(1 0.326) = 0.128 or 12.8%.

18

Robustness to alternative plough measures Estimates using either of the two methods for imputing missing language data are qualitatively identical to the estimates using our baseline variable. Estimates of equation (2) using the two alternative plough variables are reported in Appendix Table A2. We nd the alternative measures yield nearly identical point estimates that are highly signicant. As a nal strategy to ensure that our ndings are not being driven by measurement error, we omit 17 countries that have a signicant proportion of missing data.19 The estimates, reported in the appendix in Table A3, show that the estimated impact of the plough remains robust to the smaller sample.

The persistence of female labor force participation To this point, we have shown that historic plough use is associated with less female participation in agriculture historically, and with less female labor force participation today. These two

correlations imply long-term persistence in female participation in activities outside of the home. As a check for the persistence of cross-country differences in the participation of women in the labor market, we regress female labor force participation today on the measure of womens participation in agriculture constructed from the Ethnographic Atlas. The regression also controls for our full set of covariates from equation (2) above. The partial correlation plot, showing the relationship between historic female participation in agriculture and FLFP today, is shown in gure 5. As is apparent from the gure, there is strong persistence over time. Female labor market participation today and female participation in agriculture in historical societies are very strongly correlated, even after controlling for our set of covariates. In light of existing studies, this persistence is perhaps surprising. In fact, Goldin and Sokoloff (1984) document that within the Northeastern United States, the low relative productivity of women and children in agriculture (and hence their low participation) spurred industrialization and their active participation in the manufacturing sector. In their setting, female labor force participation in agriculture was inversely related to participation in manufacturing, suggesting a lack of continuity of female labor market participation overtime as industrialization occurs. Our results, however, show that this example does not appear to be general. Instead, areas with low female participation in agriculture and plough use because of the persistence of norms and
countries include Australia, Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Ecuador, Guatemala, Honduras, Mexico, New Zealand, Nicaragua, Panama, Paraguay, Peru and Venezuela.
19 The

19

40

VUT THA

TZA

BWA

RWA LSO

40

MDG ISL KEN GIN BHS AGO BDI LAO JAM CHNKAZ GAB UGA PNG BFA BOL COM MOZ AZE ZWE PER MNG PRT KHM BRN BRA GMB VNM MRT SEN NZL URY VCT VEN FIN DNK CAN DOM GHA CHE NOR TTO SWE LTU RUS ARG SGP CAF ZAF LCA PHL HTI MYS CZE ZMB MWI ISR SVK COG USA PAN ESTBLRGBR IDN KOR NLD CYP PRY BIH TKM CPV TCD CMR FRA POL LVABGD UZB ALB SVN ARM ETH BLZ ESP AUT DZA KGZ GRC ECUBEN GEO SLV NPL GUY LBR TGO IRL GTM MEX JPN FJI UKR CIV COL BGRHRV CRI BEL MDA DEU SLE HUN ITA PRI HND LBY SUR ARE BTN IRN LKA MKD CHL GNQ GNB TUR OMN MAR SAU ERI LUX EGY NGA LBN IND JOR SDN SYR TJK NER SLB MLI PAK

e( FLFP_2000 | X ) 0 20

SWZ NCL

NAM

20

NIC

0 1 e( Historical_FLFP_agriculture | X )

coef = 4.6895409, (robust) se = 1.3788193, t = 3.4

Figure 5: Historic female participation in agriculture and current FLFP. beliefs, today continue to have low levels of female participation in activities outside of the home, even after the economy moves out of agriculture and into industry.

Controlling for alternative hypotheses We now test the robustness of our estimates to alternative determinants of gender roles that have been suggested by various scholars. One of the most prominent alternative explanations for the evolution of gender role differences across societies is from Engels (1902). He argues that gender inequality arose as a result of the intensication of agriculture, which resulted in the emergence of private property, which was monopolized by men. Their control of private property allowed men to subjugate women and to introduce exclusive paternity over their children, replacing matriliny with patrilineal descent, making wives even more dependent on husbands and their property. Another consequence of this is that women were no longer active and equal participants in the community life. We account for this potential mechanism by including a number of observable characteristics of pre-industrial societies which capture the different determinants from Engelss theory. Specically, we control for a variable that equals the proportion of a countrys ancestors that

20

practiced intensive agriculture.20 We also control for the proportion of a countrys population with ancestors without inheritance rules for land, which we take as an indicator for the absence of property rights in land.21 The last control variables are two measures that capture the proportion of a countrys ancestors with patrilocal post-marital residence rules and with matrilocal rules. These capture the extent to which societys were matrilineal versus patrilineal.22 Columns 13 of Table 3 report estimates of equation (2) with the additional controls included. We report the estimates with female labor force participation as the dependent variable.23 The estimated impact of the traditional plough use remains robust to the inclusion of the additional controls: the coefcient remains negative and statistically signicant, and its magnitude changes little from the baseline value of 16.5. It has been hypothesized that the status of women may be affected by the extent to which families are nuclear as opposed to including extended relatives. It has been argued that cultures with large-extended families typically have more hierarchical and less egalitarian structures. Since hierarchies tend to be dominated by men, this results in a subordinate status of women (Engels, 1902, Boserup, 1970, Barry, Bacon and Child, 1957).24 We control for the potential impact of family structures by controlling for the proportion of a countrys ancestors that lived in nuclear or extended families.25 Estimates with these additional controls are reported in column 4. The estimate for historic plough use remains robust. It is possible that the status of women is also affected by the extent to which a society
20 The measure is constructed from variable v28 in the Ethnographic Atlas, which classies societies based on their agricultural intensity into one of the following categories: no agriculture, casual agriculture, extensive or shifting agriculture, horticulture, intensive agriculture and intensive irrigated agriculture. Our control captures societies belonging to the last two categories. 21 The original Ethnographic Atlas question (v75) measures the inheritance distribution for land and identies the following categories: equal or relatively equal, exclusively or predominantly to the one adjudged best qualied, ultimogeniture, primogeniture and absence of inheritance of real property. Our measure identies ethnic groups belonging to the last categories. 22 The measures are constructed from variable v12 of the Ethnographic Atlas, which groups societies into the following categories: avunculal, ambilocal, optionally uxorilocal or avunculocal, optionally patrilocal, matrilocal, neolocal, no common residence, patrilocal, uxorilocal or virilocal. 23 The estimates for the other outcomes of interest are also robust to the inclusion of the additional controls. 24 A similar but different theory stresses competence rather than authority. In large families with many adults, a gender division of labor can more easily develop. In nuclear families with only a husband and a wife, it is more likely that either adult will need to substitute for the other. Therefore, the wife will be involved in activities ordinarily done by men. 25 The information is taken from variable v8, which classies ethnic groups family structures into the following categories: (i) independent (monogamous) nuclear family, (ii) independent (polygynous) nuclear family, (iii) independent polyandrous families, (iv) polygynous (with co-wives), (v) polygynous (without co-wives), (vi) minimal extended families, (vii) small extended families, (viii) large extended families. We construct a variable for the proportion of nuclear families (including indipendent monogamous and polygynous nuclear families) and a variable for the proportions of extended families (including minimal, small and large extended families).

21

participates in certain non-agricultural activities, which are not particularly suitable to female involvement. To our minds, the most likely candidates are hunting and the herding of large animals, which may also require signicant strength. We construct a measure of the average historical dependence (measured as a percentage) of a countrys ancestors on hunting and animal husbandry. Both variables are measured as the proportion of all subsistence activities that are comprised of each activity.26 As shown in column 5, the inclusion of these controls has little impact on the estimated effect of past plough use. A potentially important determinant of differences in gender roles is religion. We check the robustness of our estimates by controlling for the proportion of a countrys population that is: Catholic, Protestant, other Christian, Muslim, and Hindus.27 As shown in column 6, the estimated impact of the plough remains robust. In recent work, Ross (2008) challenges the idea that religion has an important impact on gender attitudes and female labor force participation. Instead, he argues for the importance of a country endowment of oil reserves. According to his hypothesis, oil causes a countrys domestic currency to strengthen, making exports less competitive and causing a decline in light manufacturing, a sector particularly well-suited for female employment. We account for this possibility by controlling for per capita oil production and the trade-to-GDP ratio, both measured in 2000.28 We also control for the economic structure of each country, by including measures of the share of GDP accounted for by agriculture, manufacturing and services. Estimates with these additional controls are reported in columns 79. The impact of the plough remains robust to the set of controls meant to capture the hypothesized mechanism from Ross (2008). The last factor that we consider stems from the fact that part of the variation in ancestral plough use arises because of European migration to different parts of the world. European ancestry may have impacts on economic development, gender attitudes and female labor force participation independent of historic plough use. We account for this possibility by controlling for the fraction of each countrys population in 2000 whose ancestors came from Europe. The measure is taken from Nunn and Puga (2011), who calculate the variable using Weil and Puttermans (2010) World Migration Matrix. As reported in column 10, the estimates remain robust to controlling for
variables are constructed from variables v2 and v4 of the Ethnographic Atlas. data are taken from McCleary and Barros (2006) Religion Adherence dataset. 28 The oil production data are from BP Oil and the trade-to-GDP ratio is taken from the World Banks World Development Indicators.
27 The 26 The

22

Table 3: Robustness of OLS estimates to alternative controls.

(1) Historic plough use Prac@ces intensive agriculture Absence of private property Patrilocal society, matrilocal society Nuclear family, extended family Propor@on of subsistence from: hun@ng, herding Baseline historic and contemporary controls Observa@ons R-squared Historic plough use Prop of pop belonging to ve major religions Oil produc@on per capita Trade/GDP Agric., manuf. and services share of GDP Frac@on of European descent Observa@ons R-squared -14.734*** (4.856) yes (2) Dependent variable: FLFP (3) (4) -17.073*** (3.568) -14.570*** (3.568) (5) -15.417*** (3.692)

-16.814*** (3.769) yes

yes yes yes 159 0.41 (6) -12.041*** (3.967) yes yes 156 0.41 (7) -16.435*** (3.536) yes yes yes 157 0.57 157 0.43 159 0.42 157 0.42 yes 151 0.42 yes 159 0.42 (8) -15.767*** (3.538) yes 159 0.46 (9) -18.474*** (4.112) yes yes 159 0.43 (10) -15.375*** (3.844)

Notes: OLS es@mates are reported with robust standard errors in brackets. The unit of observa@on is a country. ***, ** and * indicate signicance at the 1, 5 and 10% levels.

European ancestry.29 Overall, the estimated impact of the plough remains highly robust across the ten specications in Table 3. The coefcient is always negative and statistically signicant, and the point estimate remains reasonably stable, ranging from 0.12 to 0.18. Although we do not report the estimates here, we also nd that the other two outcomes variables the share of rms with female ownership and the participation of women in politics also remain robust to the inclusion of all these controls. As well, the results also remain robust if the controls are included with continent xed effects.

B. Individual-level estimates We now turn to our specication that examines variation across individuals, linking them to a tradition of plough agriculture using the district they live in. This analysis relies on data from the World Value Survey (WVS), a compilation of national individual-level surveys on a wide variety of topics, including attitudes and preferences, as well as information on standard demographic
29 An alternative, more blunt strategy, is to omit all European and the neo-Europe countries (Australia, New Zealand, Canada, the US) from the sample. As reported in Appendix Table A4, the results are also robust to this strategy.

23

characteristics, such as gender, age, education, labor market status, income, religion, etc.30 Using the WVS we construct, for females, an indicator variable that equals one if she is in the labor force, which includes full-time, part-time or self-employment. Women are not in the labor force if they report being retired, a housewife or a student.31 We also examine two measures of individuals (male and female) attitudes about the appropriate role of women in society. The rst measure is based on each respondents view of the following statement: When jobs are scarce, men should have more right to a job than women. The respondents are then asked to choose between agree, disagree, neither or dont know. We omit observations for which the respondents answered neither or dont know, and code disagree as 0 and agree as 1.32 Therefore, the variable is increasing in the extent to which a respondents view is characterized by gender inequality. We also consider a second variable derived from a survey question where the respondents are given the following statement On the whole, men make better political leaders than women do. Respondents are then asked to choose between strongly disagree, disagree, agree, agree strongly, or dont know. We omit observations in which the respondent answered dont know and create a variable that takes on the value of 1 for strongly agree, 2 for disagree, 3 for agree and 4 for agree strongly. This variable, like the rst is increasing in the respondents view of gender inequality. We nd the two subjective value measures particularly appealing because they could reect the values potentially underlying cross country differences in female outcomes. The rst question reects values of female access to jobs, which may underlie part of the observed differences in female labor force participation rates across countries. The second question reects values about the ability of women to take on roles of leadership and responsibility, which may determine observed differences in female participation in politics and female rm ownership and management. Therefore, there is a close link between the objective measures from the country-level analysis and
30 The ve waves of the WVS were carried out in the following years: 1981-1984, 1990-1993, 1995-1997, 1999-2004 and 20052007. The coverage varies depending on the wave (starting with 22 countries in 1980, reaching 81 countries in the fourth wave and 57 in the fth). In our analysis, we use the four most recent waves of the World Values Survey, since the rst wave does not contain information on the district in which the respondent lives. Because regional classications often vary by wave, we choose the wave with the most nely dened regions. 31 The results are qualitatively identical if we exclude retired women and students from the sample. 32 We omit observations that respond neither because it is ambiguous whether this represents an intermediate view or whether they have chosen not to answer the question or whether they do not know their answer. If we interpret this response as reecting an intermediate position and code a variable that takes on the values 0, 1, and 2, then we obtain qualitatively identical results to what we report here.

24

Table 4: Individual-level OLS estimates.

Dependent variables: FLFP (1) -0.214*** (0.034) yes yes yes yes 38,832 0.186 When jobs are scarse (2) 0.245*** (0.029) yes yes yes yes 71,656 0.209 Men be`er Average eect size poliIcal leaders (AES) (3) (4) 0.397*** 0.451*** (0.075) (0.063) yes yes yes yes 59,288 0.178 yes yes yes yes 65,472

Historic plough use Individual controls Current country controls Historic district controls ConInent FE ObservaIons R-squared

Notes: The table reports OLS esImates, with standard errors clustered at the country level. Individual controls are age, age squared, educaIon, gender (for gender aXtudes only), marital status, and income. Current country controls include ln income, ln income squared and a communism indicator variable. Historic district controls include agricultural suitability, domesIcated animals, tropical areas, poliIcal hierarchies, and economic complexity. The AES reported in column 4 is for the two subjecIve belief measures from columns 2 and 3. ***, ** and * indicate signicance at the 1, 5 and 10% levels.

the subjective measures in the individual-level analysis. Examining these three outcomes female participation in the labor force, attitudes about female employment, and attitudes about female leadership skills we estimate the following individual-level equation:
C H yi,d,c = r (c) + Ploughd + Xc + Xd + Xi + i,d,c

(3)

where i denotes an individual, d denotes a district within a country c, and r (c) denote the continent of country c. Ploughd is our measure of the historic use of the plough among the
C ancestors of individuals living in district d. Xc are the same current country-level controls as in H equation (2), and Xd includes the same historic ethnographic variables as in equation (2), but

measured at the district level. Xi denotes current individual-level controls: age, age squared, as well as xed effects for marital status, educational attainment, and income levels. The equation also includes continent xed effects, denoted r (c) . To be as conservative as possible, we cluster the standard errors at the country level. Tables 4 reports OLS estimates of equation (3). Consistent with the country-level estimates, we nd a negative relationship between historic plough use and female labor force participation today, and a positive relationship between historic plough use and current attitudes reecting gender inequality. In terms of the magnitude of the effects, they are similar to the cross-country 25

estimates. An one-standard-deviation increase in the plough variable implies a reduction in female labor force participation of 0.09 (which is roughly equal to 16% of the sample average) and an increase in the gender attitudes variables of 0.10 and 0.21 (which are 21% and 8% of the sample averages, respectively).

5. IV estimates
The core concerns with the OLS correlations reported to this point are selection and omitted variables bias. For example, locations that historically had attitudes prone to less equal gender roles may have been more likely to invent or adopt the plough (in this case our OLS estimates would be biased away from zero). On the other hand, locations that were economically more developed were more likely to adopt the plough. Because today these areas are richer and more prone to attitudes about gender role equality, this will tend to bias our OLS towards zero. Our rst strategy to address these concerns was to control for observable characteristics. This was done in the previous section. Here we pursue an alternative strategy using instrumental variables. As instruments we rely on a determinant of historic plough use that has been emphasized in the anthropological literature: specic characteristics of the geo-climatic environment which determined the type of crops that could grow in a particular location (Pryor, 1985). Because the use of the plough involves a trade-off between larger up-front xed costs, but an ability to cultivate large amounts of land over a short period of time, the benet of the plough is greater for crops that have short cultivation periods (even multiple cropping per year) and require large amounts of land for a xed yield of calories. Further the benet of the plough is reduced when the crops are grown in swampy, sloped, rocky, or shallow soils, all of which make the plough much less efcient or impossible to use. Taking these factors into consideration, Pryor (1985) has classied crops into those whose cultivation benets greatly from the adoption of the plough i.e., plough positive crops and those whose cultivation benets less plough-negative crops. Plough-positive crops, which tend to be cultivated on larger expanses of land (per calorie of output), have shorter growing seasons, and grow on at, deep, soil that is not too rocky or swampy, include wheat, teff, barley and rye. On the other hand plough negative crops, which tend to yield more calories per acre, have longer growing seasons, and can be cultivated on more marginal land, include sorghum, maize, millet, roots and tubers, and tree crops (Pryor, 1985, p.

26

732).33 Because the cultivation of plough-positive and plough-negative crops is an endogenous outcome, we do not use this as our instrument. Instead, we measure geo-climatic conditions that are unaffected by human actions, but which impact the suitability of a location for growing both types of crops. Our strategy uses two instruments. The rst is a measure of the average suitability of the geo-climatic conditions of the location of each observations ancestors for cultivating wheat, barley and rye, which are plough-positive cereal crops. The second is the same measure of ancestral suitability, but for cultivating millet and sorghum, which are plough-negative cereal crops. We choose to identify off of conditions related to cereal crops that are plough-positive and plough-negative because we feel that these are the most comparable. The crops being compared are all cereals that have been grown in the Old World for millennia. Given our instruments, identication relies on the assumption that holding overall crop productivity constant (which we condition on), the type of cereal crop that a location can grow only impacts long-term gender attitudes through the adoption of the plough. In other words, the plough-positive and plough-negative cereal crop distinction is only important for long-term gender roles because it impacted the historic adoption of the plough. We obtain information on the suitability of a location for cultivating plough-positive and plough-negative cereal crops from the FAOs Global Agro-Ecological Zones (GAEZ) 2002 database (Fischer, van Nelthuizen, Shah and Nachtergaele, 2002). The database reports suitability for the cultivation of numerous crops for 5 arc minutes by 5 arc-minute grid-cell globally. The measures are constructed from a host of data on the geo-climatic conditions of a location: precipitation, frequency of wet days, mean temperature, daily temperature range, vapor pressure, cloud cover, sunshine, ground-frost frequency, wind speed, soil slope, and soil characteristics. These data are then combined with the specic growing requirements of crops to produce a measure of whether the crop could be grown in that location, and if so, how productively. It is important to note that the climate models are very sophisticated and therefore the crop suitability measures are not simple functions of the geographic characteristics the models use. In addition, the measures are objectively calculated, and not affected by where crops are actually cultivated.34 We construct the instruments by rst identifying the land traditionally inhabited by each ethnic
33 In his study, Pryor shows empirically that the existence of plough positive crops is positively correlated with the adoption of the plough. 34 For a detailed discussion of the data and a different application see Nunn and Qian (2011).

27

group in the Ethnographic Atlas, which reports the historical latitude and longitude of the centroid of each ethnic group. We then identify all land within 200 kilometers of the centroid, and assume that this is a reasonable representation of the land traditionally inhabited by the group. We then measure the amount of land within this area that can grow each of the cereal crops that comprise the instruments. Let xw , xb , xr , xs , and xm be the amount of land that can cultivate e e e e e wheat, barley, rye, sorghum and millet, respectively. Further, let xall be the amount of land that could grow any crop (i.e., the amount of arable land). The ethnicity-level measures of suitability for plough-positive crops is given by: Areapos = e
1 w 3 ( xe

+ xb + xr )/xall . While ethnicity-level e e e

measure of suitability for plough-negative crops is: Areaneg = 1 (xs + xm )/xall . Intuitively, the e e e e 2 instruments measure the average suitability for type of crop, normalized by the overall suitability for cultivation in general. Using the procedure explained in equation (1), we then construct district and country-level averages of our plough-positive and plough-negative instruments. Intuitively, the instruments measure the proportion of a country or districts population whose ancestor had a climate that could grow plough-positive cereals (wheat, barley and rye) and plough negative cereals (sorghum and millet). To provide the reader with a better sense of the instruments, Figure 6 shows the locations in the world that are classied as being suitable for the cultivation of the plough positive cereals wheat, barley and rye while Figure 7 shows suitability for the plough-negative cereals millet and sorghum. A number of facts are apparent from the maps. The rst is that there are many parts of the world that can grow plough-positive crops, but not plough-negative crops and vice versa. This provides an indication that the instruments may have variation independent from each other and therefore some predictive power. Second, relative to plough-positive crops, plough-negative crops appear to be relatively better suited for tropical and subtropical climates and plough-positive crops better suited for temperate climates. If these differences in climate caused other important differences between societies which affect gender attitudes today, then the exclusion restriction will not be satised. This is part of the motivation behind the set of covariates that we control for in both the OLS and IV estimates. Recall, that the controls include the proportion of land, historically inhabited by an ethnic group, that was tropical or subtropical. We also control for a number of historic measures of political/economic development, which may have been correlated with the presence of a tropical climate. Finally, our regressions also include

28

Wheat

Not suitable Suitable

(a) Wheat suitability

Not suitable Suitable

(b) Barley suitability

Rye

Not suitable Suitable

(c) Rye suitability

Figure 6: Maps displaying the global suitability of plough-positive crops, wheat, barley and rye.

29

Millet

Not suitable Suitable

(a) Millet suitability

Sorghum

Not suitable Suitable

(b) Sorghum suitability

Figure 7: Maps displaying the global suitability of plough-negative crops, millet and sorghum.

30

other potential determinants of plough use that may also affect gender attitudes today, including historic population density, as measured by settlement patterns, as well as a measure of overall agricultural suitability.35 An important nal point arises from the fact that the plough-positive and plough-negative cereals used in the construction of our instruments were all originally grown in the Eastern hemisphere and were not cultivated in the Americas until after 1500. This is not a concern to identication, but it is a fact that makes the rst stage relationship weaker than it would be otherwise. For the large proportion of the population in the Americas whose ancestors are from the Eastern hemisphere, the instrument will provide predictive power. It is only for the indigenous populations of the Americas that the instrument will not vary plough adoption. This should be kept in mind when interpreting the IV estimates as a local average treatment effect (LATE). In other words, the estimates are an average effect among the ethnic groups whose plough adoption was affected by the geo-climatic suitability for growing the cereal crops. Because the crops were not indigenous to the Americas, we would not necessarily expect the indigenous groups from the Americas to be within the group.

A. Country-level estimates Table 5 reports IV estimates of the specications from Table 2. The rst stage estimates are reported in the lower panel and the second stage estimates are in the top panel. The rst stage estimates show that the historic suitability for the cultivation of plough-positive cereals is always positively correlated with the adoption of the plough, while suitability for the cultivation of plough-negative cereals is generally negatively correlated with plough use. In all specications, the difference between the two coefcients is statistically signicant. The F -test for joint signicance of the two instruments is also reported in the table. The F -statistics range from about 511, suggesting that for some specications there is a potential concern about weak instruments. For this reason, in Panel A, we also report conditional likelihood ratio (CLR) condence intervals and LIML estimates in addition to the regular 2SLS estimates. The IV estimates, reported in Panel A of the table, conrm the OLS estimates. Historic plough use is associated with less female labor force participation, less female ownership of rms, and less female participation in politics. The magnitude of the IV coefcients are consistently greater
35 See

Bobek (1962) and Boserup (1970) for more on these determinants.

31

Table 5: Country level IV estimates.

Panel A. Second stage. Dependent variable: Femal labor force parLcipaLon (1) (2) -25.853** -26.423** (10.051) (12.465) Share of rms with some female ownership (3) (4) -19.939* -26.274* (11.932) (14.439) Females in poliLcs (5) (6) -16.820* -23.089* (9.243) (12.331) Average eect size (AES) (7) (8) -1.451*** -1.946*** (0.467) (0.635)

Historic plough use (2SLS) Historic plough use (LIML) p-value CLR intervals Historic controls Contemporary controls ConLnent FEs ObservaLons R-squared Plough-posiLve environment Plough-negaLve environment Equal coe (p-value)
F-stat (excl instr)

-27.099 -28.25 -28.761 -43.12 -16.826 -23.16 0.01 0.03 0.03 0.02 0.05 0.03 [-54.88, -7.02] [-66.24, -3.47] [-71.65, -2.70] [-142.60, -6.77] [-47.06, 0.29] [-75.07, -2.55] yes yes no 157 0.393 0.412*** (0.119) -0.120 (0.091) 0.00 10.76 0.15 yes yes yes 157 0.427 0.377*** (0.101) -0.079 (0.075) 0.00 7.90 0.19 yes yes no 104 0.014 yes yes yes 104 0.128 yes yes no 124 0.167 yes yes yes 124 0.045 yes yes no 133 yes yes yes 133

Panel B. First stage. Dep var: Historic plough use 0.656*** 0.561*** 0.401*** 0.340*** (0.150) (0.143) (0.140) (0.117) -0.017 0.001 -0.032 -0.026 (0.101) (0.075) (0.103) (0.087) 0.00 11.68 0.44 0.00 7.71 0.29 0.00 5.63 0.18 0.01 4.54 0.09

Hausman test (p-value)

Notes: IV esLmates are reported with robust standard errors in brackets. The unit of observaLon is a country. Historic controls include agricultural suitability, domesLcated animals, tropics, poliLcal hierarchies and economic complexity. Contemporary controls include ln income, ln income squared, a communism indicator, and polity (in columns 5 and 6 only). The number of observaLons for the AES esLmates is the average number observaLons in the regressions for each outcome. ***, ** and * indicate signicance at the 1, 5 and 10% levels.

than the OLS estimates. This is potentially explained by selection arising from the endogeneity of plough adoption. All else equal, historically advanced societies were more likely to adopt the plough. Further, they are more likely to also be advanced today with higher per capita incomes and more female participation in the labor market. Therefore, selection introduces a positive relationship between historic plough use and female labor force participation today, biasing the negative OLS estimates towards zero.

Robustness checks There are a number of potential concerns associated with our IV strategy. The primary concern is that the difference between plough-positive and plough-negative environments may be correlated with geographic features that affect gender attitudes today through channels other than the plough. We check the likelihood of this concern by controlling for a host of geographic characteristics that are potentially correlated with the suitability of the environment for plough-positive and plough-negative crops. Our controls36 include terrain slope, soil depth, average temperature and average precipitation of locations inhabited by each countrys ancestors.
36 The

details of these controls, including their sources are provided in the appendix.

32

Table 6: Robustness of IV estimates to additional geographic controls.

Dependent variable: FLFP

(1) Historic plough use Terrain slope Soil depth Average temperature Average precipitaGon Baseline historic and contemporary controls ObservaGons R-squared -33.573*** (11.777) yes

(2) -25.539** (10.371) yes

(3) -27.842*** (8.841)

(4) -18.427** (9.068)

(5) -25.089*** (9.300) yes yes yes yes yes 157 0.44

yes yes 157 0.29 yes 154 0.38 yes 157 0.38 yes yes 157 0.38

Notes: OLS esGmates are reported with robust standard errors in brackets. The unit of observaGon is a country. ***, ** and * indicate signicance at the 1, 5 and 10% levels.

The IV estimates (see Table 6) remain robust to the inclusion of these additional factors. In all specications, the estimated impact of past plough use on current female labor force participation is negative and statistically signicant. Further, the point estimates remain very similar to the baseline estimate of 25.9. We also check the robustness of our IV estimates to the set of additional controls from Table 3. These estimates are reported in Table 7. The IV estimates, like the OLS estimates of the impact of the plough on female labor force participation, also remain robust to the additional controls.37

B. Individual-level estimates Table 8 reports IV estimates of equation (3), which examines variation across individuals in the WVS. Consistent with the country-level IV estimates, at the individual-level we continue to nd persistent impacts of historic plough use. We estimate a negative effect of past plough use on the participation of women in the labor force, and a positive effect on the prevalence of attitudes of gender inequality. Like the country-level estimates, at the individual level we also nd that the IV estimates are larger than the OLS estimates.

6. Cultural transmission as a mechanism: Evidence from US immigrants

We now turn to an examination of the causal mechanisms underlying our results. Although our focus is on the evolution and persistence of cultural norms, it is possible that part of the long term effect of historic use plough may arise because it facilitated the development of
37 The

results are also robust for the two other outcome variables.

33

Table 7: Robustness of IV estimates to alternative controls.

(1) Historic plough use Intensity of agriculture Absence of private property Patrilocal and matrilocal socieIes HunIng and herding of large animals Nuclear and extended families Baseline historic and contemporary controls ObservaIons R-squared Historic plough use FracIon of major religious denomin. FracIon of European descent Oil producIon Trade/GDP Agric., manuf. and services share of GDP Baseline historic and contemporary controls ObservaIons R-squared Second stage IV esImates: Dependent variable: FLFP (2) (3) (4) -25.563** (10.317) yes yes yes yes 157 0.27 (6) -18.269** (9.003) yes yes 154 0.41 (7) -21.806* (11.188) yes yes yes yes 155 0.55 yes 148 0.44 yes 157 0.40 yes 156 0.40 yes yes 149 0.43 yes 157 0.38 (8) -25.974*** (10.014) yes 157 0.40 (9) -25.606** (9.303) yes yes 157 0.38 (10) -25.362** (9.930) -28.014*** (10.584) -25.332*** (9.250) (5)

-42.454* (22.737) yes

-28.969*** (10.861)

Notes: IV esImates are reported with robust standard errors in brackets. The unit of observaIon is a country. ***, ** and * indicate signicance at the 1, 5 and 10% levels.

34

Table 8: Individual-level IV estimates.

Panel A. Second stage. Dependent variable: FLFP (1) -0.585** (0.242) -0.611 0.00 [-0.700, -0.524] yes yes yes yes 36,370 0.154 0.273*** (0.073) -0.075* (0.044) 0.00 7.13 0.00 When jobs are scarce (2) 0.576** (0.229) 0.576 0.00 [0.501, 0.652] yes yes yes yes 67,347 0.170 0.242*** (0.079) -0.041 (0.040) 0.00 4.74 0.03 Men be.er Average eect size poli2cal leaders (AES) (3) (4) 1.383*** 1.328*** (0.362) (0.386) 1.418 0.00 [1.308, 1.531] yes yes yes yes 55,454 0.105 0.338*** (0.088) -0.094* (0.055) 0.00 7.29 0.00 yes yes yes yes 61,400

Historic plough use Historic plough use (LIML) p-value CLR intervals Individual controls Current country controls Historic district controls Con2nent FE Observa2ons R-squared Plough-posi2ve environment Plough-nega2ve environment Equal coe (p-value) F-stat (excl instr) Hausman test (p-value)

Panel B. First stage. Dependent variable: Historic plough use.

Notes: The table reports IV es2mates, with standard errors clustered at the country level. The instruments are plough- posi2ve climate and plough-nega2ve climate. Individual controls are age, age squared, educa2on, gender (for gender actudes only), marital status, and income. Current country controls include ln income, ln income squared and a communism indicator variable. Historic district controls include agricultural suitability, domes2cated animals, tropical areas, poli2cal hierarchies, and economic complexity. The AES reported in column 4 is for the two subjec2ve belief measures from columns 2 and 3. ***, ** and * indicate signicance at the 1, 5 and 10% levels.

35

institutions and markets which are less conducive to the participation of women in activities outside of the domestic sphere. Through this channel, the plough causes less female participation in market activities because it affects the costs and benets of these activities, not because it affects individuals beliefs about whether these are appropriate activities for women. Our individual-level estimates showing an impact of the plough on gender-role attitudes provide evidence that the plough has ultimately impacted beliefs and values. However, these may only differ because of differences in external factors (e.g., institutions, policies, markets, etc), which in turn shape individual beliefs. To isolate the causal impact of the plough on individual beliefs and values, we examine variation among second generation immigrants a group of individuals from diverse cultural backgrounds, with different histories of ancestral plough use, but all facing the same external environment, including markets, institutions and policies. Using data from the March Supplement of the Current Population Survey (CPS), we examine whether women with an ancestry of plough agriculture, are less likely to be in the labor force.38 We identify the ancestry of women in the sample using their parents country-of-birth. This yields three possible measures using either the mothers country of birth, the fathers country of birth, or restricting the sample to women for which both parents country of birth is the same. The different denitions also provide evidence for whether cultural transmission is stronger from the father to the daughter, the mother to the daughter, or when both occur. To examine the impact of the plough on immigrant populations we estimate the following equation:
C H yi,s,c = s + Ploughc + Xc + Xc + Xi + i,s,c

(4)

where i denotes second generation women currently living in state s, whose country of origin is country c. As in equation (2), Ploughc denotes the historic plough-use of those in country c. The dependent variable, yi,s,c , is an indicator variable that equals one if individual i is in the
C H labor market. Xc and Xc denote the same vectors of covariates as in equation (2), which include

current and historic ethnographic controls. Xi indicates a vector of individual level controls, which includes dummies for education, a quadratic for age, real personal income, marital status and xed effects for whether the person lives in a metropolitan or rural area, and for the year the individual was surveyed. The regression also includes state xed effects (denote by s ). Because
in 1994, the CPS asks individuals about their country of origin and their parents country of origin. We use all the years available since 1994.
38 Starting

36

our variable of interest Ploughc only varies at the country of origin-level, we cluster all standard errors at this level. We estimate equation (4) using two different samples: all women and married women. When we estimate our regressions only for married women, we also control for characteristics of the womens husband, since they may play an important role in the wifes decision of whether to enter the labor market or not. Specically, we control for a quadratic of the husbands age, the husbands education, and his income. The OLS estimates are reported in Table 9. Columns 13 report estimates for the sample of all women. Column 1 identifys the womens ancestry by her fathers country of birth, while column 2 uses the mothers country of birth. In column 3 we restrict the sample to women whose parents were both born in the same country. For all three specications, we estimate a negative relationship between a tradition of plough-use in the home country and the womens participation in the labor force. IV estimates of the same specications, reported in columns 13 of Table 10 yield nearly identical results. Columns 49 of Tables 9 and 10 report OLS and IV estimates when the sample is restricted to married women. Columns 46 reproduce the estimates of columns 13 for the more restricted sample. We continue to nd a negative impact of traditional plough use on female labor force participation, but now we nd a smaller estimated impact when we trace ancestry using the fathers country of birth. The impact is roughly 50% lower for both the OLS and IV estimates. This provides evidence, that at least for the subsample of married women, the transmission of values and beliefs, (originating from the plough) is stronger from mothers to daughters than from fathers to daughters. We next consider the possibility that a married womens participation in the labor market may be inuenced by her husbands beliefs and values, which were transmitted from his parents and their country of origin. In columns 79 of Tables 9 and 10, reproduce the estimates of columns 46 of the tables, but measure ancestry using the husbands parents rather than the mothers parents. These estimates provide information on whether the husbands beliefs and values play a role in his wifes decisions. We nd an impact of a history of plough use among the husbands ancestors on the wifes participation in the labor market. Interestingly, we continue to nd a weaker estimate when we use the husbands fathers country of birth. In other words, even when looking at the husband, the estimates suggest weaker transmission from the father than from the

37

Table 9: Immigrant OLS regressions.

Dependent variable: Labor force parQcipaQon indicator All women Women's ancestry Women's ancestry Married women Husband's ancestry

Historic plough use

Father's country (1) -0.042*** (0.011) yes n/a yes yes yes 48,910 0.372 47,219 0.376 27,550 0.401 8,864 0.371 yes yes yes yes 8,261 0.37 yes n/a yes yes yes n/a yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes 5,832 0.386

Mother's country (2) -0.047*** (0.012)

Parents same country (3) -0.065*** (0.013)

Father's country (4) -0.015 (0.021)

Mother's country (5) -0.043* (0.022)

Parents same country (6) -0.044* (0.024)

Father's country (7) -0.027 (0.018) yes yes yes yes yes 9,505 0.365

Mother's country (8) -0.049** (0.024) yes yes yes yes yes 8,886 0.369

Parents same country (9) -0.047** (0.022) yes yes yes yes yes 7,211 0.381

38

Individual controls Husband controls Historic country controls Current country controls

State xed eects

ObservaQons R-squared

Notes: OLS esQmates are reported with standard errors clustered at the country level. An observaQon is a US immigrant. Individual controls include age, age squared, educaQon, marital status and income. Husband controls include husband's age, age squared, educaQon and income. Historic country controls include the origin country's historic agricultural suitability, domesQcaQon of animals, tropics, poliQcal hierarchies and economic complexity. Current country controls include ln income, ln income squared, and a communism indicator variable. ***, ** and * indicate signicance at the 1, 5 and 10% levels.

Table 10: Immigrant IV regressions.

Panel A. Second Stage. Dep var: Labor force participation indicator All women Women's ancestry Women's ancestry Married women

Historic plough use yes n/a yes yes yes 48,908 0.372 47,075 0.375 27,550 0.401 8,863 0.371 8,245 0.37 yes yes yes yes yes 5,832 0.386 yes n/a yes yes yes n/a yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes 9,503 0.365

Father's country (1) -0.045*** (0.013)

Mother's country (2) -0.040** (0.016)

Parents same country (3) -0.058*** (0.019) Father's country (4) -0.026 (0.025) Mother's country (5) -0.064** (0.029) yes yes yes yes yes 8,870 0.369

Parents same country (6) -0.060* (0.031)

Husband's ancestry Parents Father's Mother's same country country country (7) (8) (9) -0.050** -0.079** -0.068** (0.025) (0.034) (0.030) yes yes yes yes yes 7,209 0.381

Individual controls Husband controls Historic country controls Current country controls

39
1.195*** (0.168) -0.370* (0.212) 0.00 60.02 0.73 1.160*** (0.176) -0.360 (0.222) 0.00 48.16 0.50

State fixed effects

Observations R-squared

Plough-positive environ.

Plough-negative environ.

Equal coeff (p-value) F-stat (excl instr) Hausman test (p-value)

Panel B. First stage. Dep var: Historic plough use 1.297*** 1.211*** 1.223*** 1.27*** 1.194*** (0.176) (0.164) (0.170) (0.173) (0.152) -0.372** -0.361** -0.421* -0.402** -0.444** (0.187) (0.186) (0.227) (0.188) (0.186) 0.00 0.00 0.00 0.00 0.00 54.43 66.47 65.80 64.45 61.82 0.21 0.54 0.64 0.34 0.46

1.205*** (0.150) -0.348** (0.180) 0.00 71.19 0.19

1.241*** (0.149) -0.394** (0.187) 0.00 85.44 0.22

Notes: IV estimates are reported with standard errors clustered at the country level. An observation is a second generation US immigrant woman. Individual controls include age, age squared, education, marital status and income. Husband controls include husband's age, age squared, education and income. Historic country controls include the origin country's historic agricultural suitability, domestication of animals, tropics, political hierarchies and economic complexity. Current country controls include ln income, ln income squared, and a communism indicator variable. All regressions also control for state, year and metropolitan area fixed effects. ***, ** and * indicate significance at the 1, 5 and 10% levels.

mother. One may have expected stronger transmission from father to son, than from mother to son, but the data do not support this. Transmission of beliefs about gender appears strongest from the mother, whether it is to the wife or the husband. It is also interesting that the estimates of the importance of the plough through the husbands parents is in every specication stronger than the importance through the wifes parents (i.e., comparing columns 46 with 78). This suggests that the beliefs of the husband are at least as important as the beliefs of the wife in determining whether she enters the labor force. This is consistent with the previous ndings of Fernandez and Fogli (2009). Comparing the magnitude of the immigrant estimates with the country-level estimates allows us to glean some evidence of the relative importance of cultural transmission, relative to other channels, in explaining the relationship between historic plough use and female labor force participation. For the whole sample, based on the estimates from column 1 of Table 9, a one-standard-deviation increase in the historical plough use variable implies a decline in female labor force participation of the order of 0.02, which is approximately 3% of the sample average (the effect is double when parents come from the same country). For married women a one-standard-deviation increase in the historical plough use variable implies a decline in female labor force participation of 0.02, approximately 3% of the sample average (this effect is the same when both parents have the same heritage). We obtain similar results when looking at the husbands cultural origin. Overall the impact of the plough on US second generation immigrant women is much smaller than the estimated impact from the country-level regressions. This suggest that other causal mechanisms are also important, the most likely candidates being the nature of a countrys institutions, its policies, and the structure of economy. These all may have been affected by historic plough use and early gender attitudes and may shape behavior and attitudes today. Therefore, although we have identied a statistically important role for cultural transmission and cultural persistence in explaining the persistent impact of traditional plough use, this is far from the whole story.

7. Conclusions
Social anthropologists have long-considered the use of shifting hoe cultivation vs. plough cultivation as an important determinant of the evolution and persistence of traditional gender roles and 40

norms. We formally test this hypothesis by combining ethnographic data on traditional plough use, measured at the ethnicity level, with contemporary data measuring gender attitudes and female participation outside of the domestic sphere. Our ndings provide evidence that current differences in gender attitudes and female behavior have indeed been shaped by historical differences in the agricultural system. Specically, we have shown that individuals, ethnicities and countries whose ancestors used the plough today have beliefs that exhibit greater gender inequality today and women participate less in non-domestic activities, like market employment, entrepreneurship, and politics. In an effort to identify a channel of cultural persistence, we examined variation across second generation female immigrants born (and living) in the US, but from different cultural backgrounds. We nd that even examining this group of individuals who face the same labor market, institutions, and policies, a history of plough use is associated with less female labor force participation. Overall, our ndings provide evidence for the historical evolution of gender norms and beliefs, and their persistence and importance today.

References
Albanesi, Stefania and Claudia Olivetti, Gender Roles and Technological Progress, Working Paper 13179, NBER June 2007. and , Home Production, Market Production and the Gender Wage Gap: Incentives and Expectations, Review of Economic Dynamics, 2009, 12, 80107. Alesina, Alberto and Nicola Fuchs-Schundeln, Good-Bye Lenin (or Not?): The Effect of Communism on Peoples Preferences, American Economic Review, 2007, 97 (4), 15071528. and Paola Giuliano, The Power of the Family, Journal of Economic Growth, June 2010, 15 (2), 93125. , Yann Algan, Pierre Cahuc, and Paola Giuliano, Family Values and the Regulation of Labor, 2010. NBER Working Paper 15747. Barry, H., M. Bacon, and I. Child, A Cross-Cultural Survey of Some Sex Differences in Socialization, Journal of Abnormal and Social Psychology, 1957, 55, 327332. Baumann, Hermann, The Division of Work According to Sex in African Hoe Culture, Africa: Journal fo the International African Institute, 1928, 1 (3), 289319. Becker, Sasha O., Katrin Boeckh, Christa Hainz, and Ludger Woessman, The Empire is Dead! Long Live the Empire, 2010. Mimeo, University of Stirling. Bobek, Hans, The Main Stages in Socio-Economic Evolution from a Geographical Point of View, in Philip L. Wagner and Marvin W. Mikesell, eds., Readings in Cultural Geography, Chicago: University of Chicago Press, 1962, pp. 218247. 41

Boserup, Ester, Womans Role in Economic Development, London: George Allen and Unwin Ltd., 1970. Braudel, Fernand, Mediterranean in the Ancient World, London: Allen Lane, 1998. Carranza, Eliana, Soil Endowments, Production Technologies and Missing Women in India, 2010. Mimeo, Harvard University. Clingingsmith, David, Asim Ijaz Khwaja, and Michael Kremer, Estimating the Impact of the Hajj: Religion and Tolerance in Islams Global Gathering, Quarterly Journal of Economics, 2009, 124 (3), 11331170. Engels, Frederick, The Origin of the Family, Private Property and the State, Chicago: Charles H. Kerr & Company Cooperative, 1902. Fernandez, Raquel, Women, Work and Culture, Journal of the European Economic Association, 2007, 5 (23), 305332. and Alessandra Fogli, Culture: An Empirical Investigation of Beliefs, Work, and Fertility, American Economic Journal: Macroeconomics, 2009, 1 (1), 146177. Fischer, Gunther, Harrij van Nelthuizen, Mahendra Shah, and Freddy Nachtergaele, Global Agro-Ecological Assessment for Agriculture in the 21st Century: Methodology and Results, Rome: Food and Agriculture Organization of the United Nations, 2002. Fortin, Nicole, Gender Role Attitudes and the Labour Market Outcomes of Women Across OECD Countries, Oxford Review of Economic Policy, 2005, 21, 416438. , Gender Role Attitudes and Womens Labor Market Participation: Opting-Out, AIDS, and the Persistent Appeal of Housewifery, 2009. Mimeo, University of British Columbia. Foster, Andrew D. and Mark R. Rosenzweig, Comparative Advantage, Information and the Allocation of Workers Tasks: Evidence from an Agricultural Labour Market, Review of Economic Studies, 1996, 63 (3), 347374. Goldin, Claudia, The U Shaped Female Labor Force Function in Economic Development, in T. Paul Schultz, ed., Investment in Women Human Capital, Chicago: University of Chicago Press, 1995, pp. 6190. , The Quiet Revolution that Transformed Womens Employment Education and Family, American Economic Review Papers and Proceedings, 2006, 96 (2), 123. and Kenneth Sokoloff, The Relative Productivity Hypothesis of Industrialization: The American Case, 1820 to 1850, Quarterly Journal of Economics, 1984, 99 (3), 461487. Goody, Jack, Production and Reproduction, Cambridge, UK: Cambridge University Press, 1976. Gordon, Raymond G., Ethnologue: Languages of the World, 15th Edition, SIL International, 2005. Grosjean, Pauline, The Determinants of Cultural Integration: A Gravity Approach, 2010. Mimeo, University of San Francisco. , A History of Violence: Testing the Culture of Honor in the US South, 2010. Mimeo, University of San Francisco. Guiso, Luigi, Paola Sapienza, and Luigi Zingales, The Role of Social Capital in Financial Development, American Economic Review, 2004, 94 (3), 526556. 42

, and

, Long-Term Persistence, 2008. Mimeo.

, , and , Social Capital as Good Culture, Journal of the European Economic Association, 2008, 6 (2-3), 295320. Iversen, Torben and Francis Rosenbluth, Women, Work, and Politics: The Political Economy of Gender Inequality, New Haven: Yale University Press, 2010. Kling, Jeffrey, Jeffrey Liebman, Lawrence Katz, and Lisa Sanbonmatsu, Moving to Opportunity and Tranquility: Neighborhood Effects on Adult Economic Self-Sufciency and Health from a Randomized Housing Voucher Experiment, 2004. Mimeo, Harvard University. McCleary, Rachel M. and Robert J. Barro, Religion and Economy, Journal of Economic Perspectives, 2006, 20 (2), 4972. Michalopoulos, Stelios, The Origins of Ethnolinguistic Diversity: Theory and Evidence, 2008. Mimeo, Tufts University. Murdock, George Peter and Caterina Provost, Factors in the Division of Labor by Sex, Ethnology, 1973, 12 (4), 203225. and , Measurement of Cultural Complexity, Ethnology, 1973, 12 (4), 379392.

and Douglas R. White, Standard Cross-Cultural Sample, Ethnology, October 1969, 8 (4), 329369. Nunn, Nathan and Diego Puga, Ruggedness: The Blessing of Bad Geography in Africa, Review of Economics and Statistics, 2011, p. forthcoming. and Leonard Wantchekon, The Slave Trade and Origins of Mistrust in Africa, American Economic Review, 2011, p. forthcoming. and Nancy Qian, The Potatos Contribution to Population and Urbanization: Evidence from an Historical Experiment, Quarterly Journal of Economics, 2011, p. forthcoming. Pitt, Mark M., Mark R. Rosenzweig, and Nazmul Hassan, Human Capital Investment and the Gender Division of Labor in a Brawn-Based Economy, 2010. Mimeo. Pryor, Frederic, The Invention of the Plow, Comparative Studies in Society and History, 1985, 27 (4), 727743. Ross, Michael L., Oil, Islam and Women, American Political Science Review, 2008, 102 (1), 107 123. Tabellini, Guido, The Scope of Cooperation: Values and Incentives, Quarterly Journal of Economics, 2008, 123 (3), 905950. Weidmann, Nils B., Jan Ketil Rod, and Lars-Erik Cederman, Representing Ethnic Groups in Space: A New Dataset, Journal of Peace Research, 2010, 47 (4), 491499. Weil, David N. and Louis Putterman, Post-1500 Population Flows and the Long-Run Determinants of Economic Growth and Inequality, Quarterly Journal of Economics, November 2010, 125 (4), 16271682. Whyte, Martin King, The Status of Women in Pre-Industrial Societies, Princeton, N.J.: Princetone University Press, 1978. 43