## Abstract

This paper examines the long-term determinants of intimate partner violence (IPV) by analyzing its relationship with traditional family structures: stem families in which one child stays in the parental household and nuclear families in which all children leave the household upon marriage. My hypothesis is that coresidence with a mother-in-law increases a wife's contribution to nondomestic work, which may decrease the level of violence. I find that areas where stem families were socially predominant in the past currently have a lower IPV rate, and use differences in inheritance laws in medieval times as an instrument for the different family types.

## I. Introduction

WORLDWIDE, 30% of all women who have been in a relationship have experienced physical or sexual violence, or both, by their intimate partner (WHO, 2013). Exposure to this type of abuse has serious consequences for women's health, fatal injuries being the most extreme outcome: as many as 38% of all female murder victims are killed by their intimate partners, in contrast to 6% of all murdered men (WHO, 2013). Violence against women is both a public health priority and a matter of social justice, yet our understanding of the causes of this kind of abuse is limited.

So far, the economics literature on intimate partner violence (IPV) has analyzed its contemporary determinants. While acknowledging their importance, if we want to fully understand what causes violence against women, we need to complement this view with a longer-term perspective. Recent findings about how traditional agricultural practices determine current gender attitudes (Alesina, Giuliano, & Nunn, 2013) and about how gender identity norms help explain current and economic social outcomes (Bertrand, Kamenica, & Pan, 2015) highlight the interest in understanding the long-term determinants and cultural norms that sustain gender violence. Among these cultural factors, I focus on traditional family structure as the family has great power in shaping gender-related attitudes and is a fundamental institution in all societies. Hence, this paper contributes to the analysis of IPV causes by studying the relationship between IPV and traditional family type.

I focus on the effects of two family types, stem and nuclear. Each has a distinct residence and inheritance pattern. In patrilineal stem families, one son inherits all the land and remains in the parental home with his wife to continue the family line. Therefore, in stem families, at least two different generations are living together. In contrast, in nuclear families, all children receive a share of the inheritance when leaving the parental home to start their own independent households. There is thus no intergenerational cohabitation in nuclear families.

I find that territories where the stem family was prevalent in the past currently exhibit lower rates of IPV and more gender-equal attitudes. My hypothesis is that coresidence of the wife with other women (usually the mother-in-law) reduced the burden of household work, freeing up her time for nondomestic work. This allowed for a more productive role of the younger woman, who had an increased contribution to family subsistence in a stem family. In section IIC, I discuss different mechanisms by which higher female contribution to nondomestic work can trigger less domestic violence. I present evidence for the positive relationship between female labor force participation and stem families (see section A in the online appendix), using data from preindustrial societies around the world and for historical and contemporary Spain.

In my main empirical analysis, I use Spanish data for two reasons. First, Spain provides gold-standard IPV data as defined by the World Health Organization (WHO, 2013): a comprehensive survey data set for 1999 to 2006 on violence against women in Spain ($N=69,627$) in which IPV is measured through a set of specific questions. Second, the regional distribution of stem and nuclear family types in Spain is stable and remarkably persistent and has been traced back to the Middle Ages (Todd, 1990). To measure historical distribution of family types, I use 1860 census data and compute the average number of married and widowed women per household at the province level. I control for individual characteristics, and I include an extensive set of contemporary, historical, and geographic control variables. My linear probability model (LPM) results are robust to the inclusion of these covariates and show a negative and significant relationship between areas with a stem family tradition and IPV.

To better understand the causality of this relationship, I exploit a unique source of exogenous variation by using the Christian conquest of the Iberian Peninsula as an instrument for family types. The so-called Reconquista is a seven-century-long period (722–1492) in which several Christian kingdoms took control from Islamic rulers and repopulated significant parts of the Iberian Peninsula. The political structure at the beginning of the Christian conquest helps to explain the establishment of the different family types. In the west of what eventually became Spain, the stronger and more centralized monarchies had an interest in restricting the development of powerful landholding families. This interest was best served by the introduction of compulsory sharing of inheritance among all children, which led to nuclear families. Meanwhile, the eastern kingdoms had a more powerful feudal nobility that sought to maintain its landholdings intact through indivisible inheritance (i.e., the appointment of a single heir), which led to stem families.

The instrumental variable (IV) estimates are consistent with the LPM estimates. In section IVB, I discuss potential concerns about the validity of the exclusion restriction. Among other tests, I perform an analysis of plausible exogeneity advanced by Conley, Hansen, and Rossi (2012) and show that with appropriate caveats, the IV estimates are informing of a negative and significant impact of the stem family structure on contemporary IPV.

The prevalence of the stem family has decreased in Spain over the past century as the country has become fully industrialized. Currently, stem families are residual at best and are no longer prevalent in any province (see figure A1 in the online appendix). However, this family pattern persisted long enough to potentially explain behavior at a later time and in different circumstances. I argue that the internalization of the resulting cultural norms and their intergenerational transmission play a role in explaining why lower rates of IPV are currently found in territories where the stem family was predominant in the past.1 To further explore the cultural transmission channel, I use data from the World Values Survey for Spain and find that territories that had a stem family tradition in the past currently exhibit more gender-equal attitudes than nuclear family territories do.

To my knowledge, this is the first paper to look at the relationship between historical family types and IPV. This paper fits into three main strands in the literature. First, it contributes to the analysis of the causes of domestic violence. The bulk of this literature looks at how the distribution of bargaining power within the couple affects domestic violence. For instance, various authors have analyzed the effect of income (Tauchen, Witte, & Long, 1991), services for battered women (Farmer & Tiefenthaler, 1996), divorce laws (Stevenson & Wolfers, 2006), the gender wage gap (Aizer, 2010), cash transfers (Angelucci, 2008; Bobonis, González-Brenes, & Castro, 2013), and unemployment (Anderberg et al., 2016; Tur-Prats, 2017).2 Other papers treat IPV as a signal conveying dissatisfaction with the marriage (Bolch & Rao, 2002) or as an expressive mechanism triggered by an emotional cue (Card & Dahl, 2011). All of these are immediate determinants of domestic violence. Only Alesina, Brioschi, and La Ferrara (2016) carry out an empirical investigation of the long-term determinants of domestic violence in Africa.3 They show that norms about marriage patterns, living arrangements, and the productive role of women in ancient times are correlated with contemporary violence. When looking at family types and in line with my results, they find that where the stem family was socially predominant in the past, both men and women currently tend to be less favorable to gender violence. To explain their results, the authors take as a reference the hypothesis first developed in this paper: that wives in stem families (where the mother-in-law takes care of the children) have more time to work outside the house, and hence become more “valuable.”

Second, this paper also contributes to the literature that examines the historical origins of gender roles. Alesina et al. (2013) test Boserup's (1970) hypothesis that societies that traditionally practiced plow cultivation (as opposed to shifting hoe cultivation), and where men therefore had an advantage in farmwork, exhibit less equal gender norms today. Hansen, Jensen, and Skovsgaard (2015) examine the hypothesis that societies with a longer history of agriculture have less equal gender roles, even without the plow. They find that societies that experienced an earlier Neolithic revolution or an earlier transition to cereal agriculture currently have lower female labor force participation. Both studies associate less equality of gender roles with a historical division of labor in which “men tended to work outside the home in the fields, while women specialized in activities within the home.” This paper also looks to the traditional division of labor to explain gender inequalities by focusing on historical family types and argues that a stem family structure tends to increase female contribution to nondomestic work, regardless of the agricultural setting.

Third, this paper adds to the literature on family types, an important part of which has focused on large kinship groups versus nuclear families and the interaction of this dimension with cooperation and provision of goods and safety. In this respect, Greif (2006) highlights the relevance of family structure to the emergence of economic and political corporations in late medieval Europe, and Greif and Tabellini (2017) study two different ways of sustaining cooperation in China and Europe: the clan and the corporation. Bertocchi and Bozzano (2015) investigate the relationship between the education gender gap and the prevalent family structures over the late nineteenth century in Italy.

Finally, my work is also related to that of Grosjean (2014), who examines the historical origins of the culture of honor and violence in the U.S. South. She shows that historical settlements populated by Scottish and Scottish-Irish herders 200 years ago are still associated with homicide today. My paper also shows a long-lasting effect of historical events on interpersonal violence. More broadly, my findings add to a growing literature on the role of culture as a channel of long-term impacts of historical events on current outcomes.4

In section II, I present the conceptual and historical framework; I review family types, the historical background, and the relationship between family structure and domestic violence. Section III documents the data used and explains the main empirical strategy. Section IV reports the LPM and the IV results and discusses the validity of the IV estimates. Section V discusses potential transmission mechanisms and shows evidence in favor of the cultural transmission channel. Section VI concludes.

## II. Conceptual and Historical Framework

### A. Family Types

According to the work started by Le Play (1884), there are three basic types of families in all parts of the world and all ages of history. I describe them below, assuming patrilocality for stem and joint families as the prevailing pattern. First, in the joint or communitarian family, all sons remain with their parents and bring their wives to the family home upon reaching adulthood. When the family gets too large, the household is split. Second, in the stem family, only one son stays at the parental homestead, together with his wife and children. He will be the one who inherits the land and the family home, thus continuing the family line. All other children who wish to marry and start their own households leave the household.5 Third, in the nuclear family, all children leave the parental home to establish their own households. This classification is also used, with some variations, by Todd (1990). He suggests that family types in Western Europe have a stable and long-lasting pattern and traces back their origins to medieval times, and even earlier for some regions.

Only two of these family structures are found in Spain: stem and nuclear. This is consistent with the anthropological work done in Spain by Lisón Tolosana (1975, 1977). Stem and nuclear families differ in two dimensions: coresidence and inheritance patterns. Stem families are generally patrilocal with patrilineal impartible inheritance (see note 5). There is a higher degree of intergenerational cohabitation, and the impartible inheritance principle (which requires a single heir) serves the main purpose of preserving the family estate. Nuclear families are neolocal and present bilateral inheritance: children leave the home to form their own households, so there is no cohabitation of couples and, at least in Spain, the estate is allocated equally among children.

To measure the predominance of family types in Spain, I use the 1860 census, the first data set that allows us to reliably measure household types for the whole country (Mikelarena Peña, 1992). The indicator chosen to best capture family structure is the average number of married and widowed women per household at the province level.6 This indicator is preferred to measures of household size (number of people or number of adults per household), as well as indicators that do not correct for immigration (total number of married and widowed people in the house).

Figure 1 shows family types in Spain in 1860. Although this represents only one specific point in time, some authors (Reher, 1996; García González, 2011) show that these patterns have remained stable at least from the seventeenth century through the beginning of the 1970s. The social and economic changes operated in Spain during the twentieth century (full industrialization, demographic transition, and mass migration to cities) have weakened the traditional peasant stem family pattern.7

Figure 1.

Family Types in Spain, 1860

For each province, I compute the average number of married and widowed women in the household as a measure of the family structure. Values range from 0.87 to 1.19, with an average of 1. In the provinces shown in darker gray, the average number of widowed and married women in the household is higher and, consequently, the stem family is more prevalent. Own calculations using 1860 census data.

Figure 1.

Family Types in Spain, 1860

For each province, I compute the average number of married and widowed women in the household as a measure of the family structure. Values range from 0.87 to 1.19, with an average of 1. In the provinces shown in darker gray, the average number of widowed and married women in the household is higher and, consequently, the stem family is more prevalent. Own calculations using 1860 census data.

### B. Historical Background

In 711 A.D., Muslims from North Africa crossed into the Iberian Peninsula. After seven years of battling the Visigoths, they came to dominate most of the territory and established their authority over Al-Andalus, or Islamic Iberia. Muslim expansion into the rest of Europe was halted by the Franks in 732 at the Battle of Tours. As a result, Charlemagne established the Spanish March, a buffer zone in northeastern Spain (broadly between the Pyrenees and the Ebro River), to protect his empire against attacks from Al-Andalus.

At the same time, in northwestern Spain where many of the ousted Visigothic nobles had taken refuge, the Christian Kingdom of Asturias was consolidating. Their first significant victory against the Muslims was in the stronghold of Covadonga in 722. This event marked the beginning of the so-called Christian reconquest (la Reconquista). The takeover and repopulation of Iberia by Christian kingdoms lasted more than seven centuries and was completed in 1492 with the fall of Granada. The circumstances that gave rise to kingdoms with different political structures in the west and the east are key in understanding the emergence of different family patterns.

In the east of the Peninsula, distant imperial power allowed the counts of the Spanish March to gain independence from the Frankish empire, and they began to drive south and conquer territories under Muslim control. Still, the feudal system brought in by Charlemagne persisted for some time and gave rise to a tradition of pactism, at least in Catalonia (Sobrequés i Callicó, 1982). With this term, historians refer to the principle that limited royal power by requiring agreements between the king and parliament (the latter first representing only the noblemen and clergy, and later incorporating commoners). From 1137 until 1707, the eastern territories formed the Crown of Aragon, a loose confederation of realms, each of which kept its own institutions, laws, and privileges.

Meanwhile, the Christian kingdoms in the west were also expanding southward. In 1230, several earlier kingdoms were united into the Crown of Castile.8 Unlike the Crown of Aragon, Castilian monarchs fought to maintain and centralize power and establish homogeneous institutions and laws. The Crown of Castile had its own unified civil law, which applied as state law or ordinary law throughout its jurisdiction. Under Castilian inheritance laws, it was mandatory to leave four-fifths of the estate to descendants, two-thirds of that to be equally allocated among them and one-third to be allocated freely to the preferred descendant.

In the Crown of Aragon, however, each individual realm (Aragon, Catalonia, the Balearic Islands, and Valencia) had its own distinct civil law, known as foral laws after the fueros, or charters, they were based on. Most foral law is devoted to family institutions. They all have in common the age-old institution of the homestead: a stable peasant family together with the farmland that provides the family living.9 To ensure the enduring survival of the family through future generations, the homestead needed to remain undivided. Thus, family and inheritance laws were geared toward guaranteeing the conservation and continuity of the family estate. In this sense, one of the most paradigmatic features of family law in foral regions is the ability to appoint a single heir or heiress, as opposed to the more equal division of bequest among offspring that was required under Castilian law.

As shown in figure 2, freedom of testation was instituted by the thirteenth century in all the Crown of Aragon territories, as well as in some Basque regions and in Navarre. Therefore, indivisible inheritance (also known as impartible inheritance) was allowed, whereby a single heir or heiress could inherit all. This right originally began at the demand of the nobility, but later, it was extended to all citizens since it was found to be beneficial for the feudal system and for conquest purposes. Indivisible inheritance, and thus a stem family structure, would link a family to a piece of land, ensuring regular collection of rents by the feudal lord. At the same time, it would release the people needed to resettle newly conquered land.

Figure 2.

Spanish Territories with Freedom of Testation, Thirteenth Century

In the provinces shown in dark gray, freedom of testation was instituted by the thirteenth century. In all these territories (comprising the Crown of Aragon, some Basque regions, and Navarre), indivisible inheritance was allowed, whereby a single heir or heiress could inherit everything. Based on Ferrer-Alòs (2011).

Figure 2.

Spanish Territories with Freedom of Testation, Thirteenth Century

In the provinces shown in dark gray, freedom of testation was instituted by the thirteenth century. In all these territories (comprising the Crown of Aragon, some Basque regions, and Navarre), indivisible inheritance was allowed, whereby a single heir or heiress could inherit everything. Based on Ferrer-Alòs (2011).

Terradas (1984) links the origins of indivisible inheritance to the feudal system established in the Spanish March by the Franks. More generally, Goldschmidt and Kunkel (1971) underscore the historical relationship between indivisible inheritance and a strong, independent feudal nobility, whereas highly centralized authorities would institute partible (i.e. divisible) inheritance in order to restrict the development of powerful landholding families.10

### C. Conceptual Framework

There are several mechanisms through which stem families, as opposed to nuclear families, might lead to lower levels of domestic violence. One potential mechanism is that a wife's coresidence with her mother-in-law, a feature of stem families, allows the younger woman to do more work outside the home. The main idea is that due to the mother-in-law's comparative advantage in the domestic activity and child rearing, coresidence with her will increase the time that the younger wife has available to work outside the home.

This higher female contribution to work in stem families could decrease the level of domestic violence through different channels. For instance, if we assume that (a) violence decreases the wife's productivity and that (b) this productivity loss is higher in the nondomestic activity (e.g., farming) than in the domestic activity, then we could show that the presence of the mother-in-law in the household could decrease the husband's preferred level of violence.11 According to this channel, the higher wife's contribution to work will lead the husband to reduce violence against her since it would reduce her health and productivity.

Alternatively, an increased wife's contribution to the productive activity may increase her bargaining power. Actually, wives enjoyed more rights historically in some of the foral territories than under Castilian law: for instance, they had greater power to manage matrimonial assets and could appoint the heir or heiress. In the same line, widows in foral territories received a life interest in the property so that the farm could smoothly continue its activity after the death of the head of the household.12 Intrahousehold bargaining models predict that higher bargaining power of women is associated with less violence against them (Aizer, 2010).

As yet, we have considered that violence enters the husband's utility function directly and positively. However, if we regard violence as a controlling behavior rather than a source of direct gratification for the husband, we would also predict that in stem families, there would be less need for violence. If we assume that the husband gets more utility from the house services of the wife (relative to his utility if the wife works) than the wife herself, then in nuclear families, husbands could use violence against their wives to reduce their value in the labor market and induce them to stay at home. Since in stem families, domestic services are already provided by the mother-in-law, there would be less need to behave violently.

All the mechanisms reviewed so far were based on the assumption that in stem families, there is a higher female contribution to work. Nonetheless, nonwork channels could be at play: if we considered a negative effect of witnesses on domestic violence, we would again expect less violence in stem families, as there are more people living in the same house. The data available allow me to account for some of these alternative mechanisms (e.g., I include in my regressions the number of people living in the household at the time of the survey, which controls, at least partially, for an explanation that takes into account only the witness and surveillance channels).

It is worth emphasizing at this point that this study is focused on stem and nuclear families and does not apply to joint families, in which intrahousehold dynamics might differ considerably. A well-known example of the joint family that still prevails today is the Hindu joint family. Conventional wisdom and popular culture often place women living in joint families in a more subservient position, especially due to the patrilocal residence and the mother-in-law's control, and therefore more prone to domestic abuse. However, some studies have challenged these common beliefs and found higher prevalence of domestic violence among women living in nuclear families than in joint families (Mishra et al., 2014; Yugantar Education Society, 2002; Visaria, 1999). Proposed explanations point to the absence of elderly people in nuclear families to solve potential conflicts and control erratic behavior of men (Jawarkar et al., 2016), and to the higher prevalence of alcoholism, extramarital affairs, and desertion by husbands in nuclear families compared to joint families (Deshpande & Suleiman, 2014). From these studies, we learn that differences in IPV prevalence between joint and nuclear families are most likely related to residential patterns and intrahousehold dynamics rather than to inheritance practices. Noticing this might help alleviate concerns about the validity of my instrument, since both nuclear and joint families practice divisible inheritance, in contrast to stem families that traditionally appointed a single heir.

There could be other explanations by which the family structure can affect domestic violence. Although I cannot decisively rule out any particular channel, I provide evidence in favor of the female productivity mechanism (see section A in the online appendix for details). First, using information about preindustrial societies from the Ethnographic Atlas, I find that women who lived in societies with indivisible inheritance (as a proxy for stem family prevalence) contributed more to agriculture. Second, I show evidence from the Spanish 1887 census that links the prevalence of stem families to higher female labor activity rates. More evidence is shown by Sasaki (2002) for contemporary Japan and Borderías and Ferrer-Alòs (2017) for the first third of the twentieth century in Catalonia (Spain). Both studies document that the presence of an older woman in the household reduces the burden of household work, leading to increased female labor force participation. Third, using evidence from female labor force participation in Spain during the period 1999 to 2006, I find a positive correlation between contemporaneous women's employment rates and historical stem family prevalence. Interestingly, the effect of the historical family type on IPV persists after ruling out the effect of female labor force participation.

The main theory of this paper goes as follows. I assume that before the Middle Ages, there were no differences across regions in the Iberian Peninsula in terms of their family structure and gender attitudes. Exogenous changes in inheritance laws before the thirteenth century contributed to the establishment of two different family structures: stem families in the northeast and nuclear families in the west and southwest. Each family type allowed for a different productive role of women, which in turn fostered a dissimilar use of domestic violence. In stem family territories, women's work outside the home is higher, and this is associated with less violence against them. Evidence in favor of the link between stem families and higher female labor force participation is shown in the online appendix (see section A). Unfortunately, there are no data available on IPV in the past to prove that the traditional stem family (currently no longer socially predominant) was associated with lower domestic violence.13 I argue that these attitudes toward IPV persist to the present day. To test this, I look at the impact of traditional family structure on present IPV, instrumenting the former by the differences in inheritance laws that emerged in medieval times. Further, in sections V in the main text and A.3 in the online appendix, I discuss different transmission channels and find evidence in favor of the cultural channel.

## III. Data and Empirical Strategy

Intimate partner violence (IPV) data in this study come from three cross-sectional surveys on violence against women in Spain (Instituto de la Mujer, 1999, 2002, 2006). These surveys were conducted by telephone in 1999, 2002, and 2006 (sample sizes 20,552, 20,652, and 28,423, respectively). They contain a broad and representative sample of adult women (18 years and older) living in Spain and include both self-assessment of IPV and more nonsubjective criteria. I use direct questions about specific acts of violence experienced over a defined period of time since they tend to reveal more information than generic questions about experiencing “domestic violence” or “abuse.” This type of survey data approximates the gold standard for estimating the prevalence of any form of interpersonal violence (WHO, 2013).

In the introduction to the interview, respondents were told that they were participating in a survey about the situation of women in the household (regarding their health, housework, children, and so on). Later in the interview, they were asked whether they had encountered any of 26 situations related to domestic violence. The questions about these 26 situations are specifically designed to detect violence against women, and 13 are considered stand-alone indicators of domestic violence because they describe more serious situations. They encompass six types of violence against women: physical, sexual, psychological, economic, structural, and spiritual. Table 1 lists the 13 indicator questions.

Table 1.
Definition of Intimate Partner Violence in the Survey
 At the moment, how often has someone from your home or your intimate partner done any of the following? Doesn't allow you to see your family, friends or neighbors Takes the money you make or doesn't give you enough money to live on Calls you names or threatens you Decides the things you can or cannot do Insists on having sex even though he or she knows you don't want to Doesn't take your needs into account (leaves you the worst share of the food, the house, etc.) Makes you feel afraid Shoves or beats you when he or she is feeling angry Says you're incapable of doing anything on your own or without him or her Says everything you do is wrong, calls you clumsy Belittles your beliefs (going to church, voting for a political party, joining an organization, etc.) or doesn't value them Doesn't appreciate your work Says things to make you look bad in front of the children
 At the moment, how often has someone from your home or your intimate partner done any of the following? Doesn't allow you to see your family, friends or neighbors Takes the money you make or doesn't give you enough money to live on Calls you names or threatens you Decides the things you can or cannot do Insists on having sex even though he or she knows you don't want to Doesn't take your needs into account (leaves you the worst share of the food, the house, etc.) Makes you feel afraid Shoves or beats you when he or she is feeling angry Says you're incapable of doing anything on your own or without him or her Says everything you do is wrong, calls you clumsy Belittles your beliefs (going to church, voting for a political party, joining an organization, etc.) or doesn't value them Doesn't appreciate your work Says things to make you look bad in front of the children

Having aggregated the data for the three surveys, I construct an IPV indicator variable that takes the value 1 if the woman answers “often” or “sometimes” for at least one of these 13 questions, and 0 otherwise. Figure 3 shows the resulting map of IPV in Spanish provinces for the period 1999 to 2006.

Figure 3.

Intimate Partner Violence in Spain, 1999–2006

I construct an IPV indicator variable that takes the value 1 if the woman answers “often” or “sometimes” for at least one of the thirteen questions detailed in table 1, and 0 otherwise. This figure shows the resulting map of IPV aggregated at the province level for the period 1999 to 2006. Values range from 5.73 to 12.67 percentage points, and the average is 8.5. In the provinces shown in a darker gray, IPV prevalence is higher. Own calculations from Spanish surveys on violence against women (1999, 2002 and 2006).

Figure 3.

Intimate Partner Violence in Spain, 1999–2006

I construct an IPV indicator variable that takes the value 1 if the woman answers “often” or “sometimes” for at least one of the thirteen questions detailed in table 1, and 0 otherwise. This figure shows the resulting map of IPV aggregated at the province level for the period 1999 to 2006. Values range from 5.73 to 12.67 percentage points, and the average is 8.5. In the provinces shown in a darker gray, IPV prevalence is higher. Own calculations from Spanish surveys on violence against women (1999, 2002 and 2006).

These surveys also include information at the individual level on the respondent's level of education, occupational status, marital status and religious beliefs, the head of the household, the presence of children in the household, the number of people in the household, and the partner's level of education. Summary statistics for the sample used in the estimation are shown in table A5 in the online appendix.14

Figure A2 in the appendix graphically explores the relationship between contemporary IPV levels and 1860 province-level family types. In particular, the dots are the residuals of a regression of the IPV rate on the baseline covariates plotted against the residuals of a regression of the traditional family types on the same covariates, both residuals averaged at the province level.15 The linear regression fit shows a negative relationship between both variables.

To further3 study the relationship between contemporary IPV levels and 1860 province-level family types, I also control for province characteristics that might be correlated with violence against women and with family types. First, I control for a set of variables that capture the current level of economic development in each province, including both formal measures (GDP per capita and unemployment) and informal measures (a social capital indicator).16 I also add the respondent's religion (Catholic versus any other option) and the number of people in the household. To control for the level of economic development in the past, I include population density and urbanization rates for each province in 1787 and 1860.17 Finally, to control for the variable productivity of labor, land, and climate, I add geographical variables (ruggedness, average temperature, temperature range, rainfall, and frost days). Data sources are listed in table A6 in the online appendix.

Using all of these data, I run the following regression to study the relationship between IPV and the different family types:
$IPVi,p,y=α+βStemp+γXi,p,y+δZp,y+θyYeary+εi,p,y,$
(1)

where $IPVi,p,y$ is a binary variable that indicates if the woman $i$ from province $p$ on survey year $y$ is a target of violence from her intimate partner, $Stemp$ is the average number of married and widowed women per household in province $p$ based on the 1860 census, $Xi,p,y$ is a vector of control variables at the individual level, $Zp,y$ comprises regional controls at the province level, $Yeary$ are survey-year fixed effects, and $εi,p,y$ is the error term.

These linear probability model (LPM) estimates, however, might be biased away from 0 if societies that were initially more pro-women were also more likely to establish a stem family structure. Conversely, if more advanced societies were more prone to adopt nuclear family structure and to have more gender-equal role attitudes at the same time, then the LPM estimates might be biased toward 0. To address this important concern, I not only control for observable characteristics (past and present economic development and determinants of farm labor productivity) but also use an instrumental-variable strategy. I exploit a historical source of exogenous variation that is unique in the history of Europe and resort to the Christian conquest of the Iberian Peninsula to instrument the family types.

In particular, the political structure at the beginning of the Christian conquest is key to explaining the emergence of the different family types. On the western side of the Iberian Peninsula, the Christian kingdoms had an interest in restricting the development of powerful landholding families. This interest was best served by instituting equal allocation of bequest, which led to nuclear family patterns. Meantime, in the east, power was more decentralized, and the feudal nobility sought to maintain its holdings intact through indivisible inheritance, which led to stem family patterns.

To quantify this instrument, I use the map of provinces in which freedom of testation was in place by the thirteenth century, thus allowing indivisible inheritance. I construct an indicator variable that takes the value 1 if the province had freedom of testation by the thirteenth century and 0 otherwise.18 I use a two-stage least-square (2SLS) procedure to estimate equation (1). In the first stage, I estimate the effect of the medieval political process on becoming a stem family province,
$Stemi,p,y=α+σFreedomTestationp+γXi,p,y+δZp,y+θyYeary+ui,p,y,$
(2)

where $FreedomTestationp$ is an indicator variable that takes the value 1 if the province had freedom of testation by the thirteenth century.

## IV. Results

### A. Linear Probability Model Results

Table 2 reports the LPM estimates of regression (1). The results show that living in provinces where the stem family was more socially predominant in 1860 is associated with less contemporary IPV. In particular, an increase of one in the average number of married and widowed women in the household per province in 1860 is associated with a decrease of around 6 percentage points in the current IPV prevalence.19 This effect persists after controlling for contemporary, historical and geographic variables.

Table 2.
Linear Probability Model Results
Mean of Dependent Variable: 0.085
Intimate Partner Violence
(1)(2)(3)
Stem family −0.067*** −0.053* −0.062*
(0.0236) (0.0311) (0.0358)
Contemporary controls  Yes Yes
Historical and geographic controls   Yes
Observations 62,740 62,740 62,740
$R2$ 0.041 0.041 0.042
Mean of Dependent Variable: 0.085
Intimate Partner Violence
(1)(2)(3)
Stem family −0.067*** −0.053* −0.062*
(0.0236) (0.0311) (0.0358)
Contemporary controls  Yes Yes
Historical and geographic controls   Yes
Observations 62,740 62,740 62,740
$R2$ 0.041 0.041 0.042

Stem family is defined as the average number of married and widowed women in the household at the province level in 1860. Model 1 includes age, respondent's and partner's level of education, respondent's occupational status, marital status, the head of the household, the presence of children in the household, settlement population size, and year when survey was conducted. Model 2 adds contemporary controls (religion; number of people in the household; and GDP per capita, unemployment rate, and social capital at the province level). Model 3 adds historical controls (population density in 1787, 1860, and the survey year; urbanization rates in 1787 and 1860), and geographic controls at the province level (ruggedness index and climate variables including temperature, temperature range, rainfall, and frost days). Standard errors clustered by province in parentheses. *$p<0.10$, **$p<0.05$, and ***$p<0.01$.

Table 3 shows the results obtained using different definitions of IPV: physical and sexual violence, on the one hand, and psychological, economic, spiritual, and structural violence, on the other. Both sets of results are in general consistent with results for the overall measure of IPV.20

Table 3.
LPM Results with Different IPV Measures
(1)(2)(3)
A. Physical and sexual violence—Mean of dependent variable: 0.032
Stem family −0.04** −0.03** −0.04*
(0.014) (0.016) (0.023)
$R2$ 0.018 0.019 0.019
B. Psychological, economic, spiritual, and structural violence—Mean of dependent variable: 0.069
Stem family −0.05** −0.04 −0.05*
(0.021) (0.028) (0.026)
$R2$ 0.032 0.032 0.033
(1)(2)(3)
A. Physical and sexual violence—Mean of dependent variable: 0.032
Stem family −0.04** −0.03** −0.04*
(0.014) (0.016) (0.023)
$R2$ 0.018 0.019 0.019
B. Psychological, economic, spiritual, and structural violence—Mean of dependent variable: 0.069
Stem family −0.05** −0.04 −0.05*
(0.021) (0.028) (0.026)
$R2$ 0.032 0.032 0.033

Stem family is defined as the average number of married and widowed women in the household at the province level in 1860. Model 1 includes age, respondent's and partner's level of education, respondent's occupational status, marital status, the head of the household, the presence of children in the household, settlement population size, and year when the survey was conducted. Model 2 adds contemporary controls (religion; number of people in the household; and GDP per capita, unemployment rate, and social capital at the province level). Model 3 adds historical controls (population density in 1787, 1860, and the survey year; urbanization rates in 1787 and 1860) and geographic controls at the province level (ruggedness index and climate variables, including temperature, temperature range, rainfall, and frost days). Standard errors clustered by province in parentheses. *$p<0.10$, **$p<0.05$, and ***$p<0.01$.

### B. Results from Instrumental Variables

Table 4 shows the instrumental-variable estimates, which confirm the LPM estimates. Panel A reports the first-stage results of equation (2) showing how greater political decentralization (measured in terms of freedom of testation) had a positive effect on becoming a stem family province. The instrument is a powerful predictor of family types, as reflected by the F-statistic for all specifications.21

Table 4.
IV Results
(1)(2)(3)
A. First-stage results—Mean of dependent variable: 1.01 (stem family)
Freedom of testation 0.10*** 0.08*** 0.06***
(0.019) (0.020) (0.016)
$F$-statistic 26.34 14.24 12.76
B. Second-stage results—Mean of dependent variable: 0.085 (IPV)
Stem family −0.13*** −0.15** −0.23***
(0.044) (0.059) (0.085)
Contemporary controls  Yes Yes
Historical and geographic   Yes
$R2$ 0.041 0.041 0.041
Observations 62,740 62,740 62,740
(1)(2)(3)
A. First-stage results—Mean of dependent variable: 1.01 (stem family)
Freedom of testation 0.10*** 0.08*** 0.06***
(0.019) (0.020) (0.016)
$F$-statistic 26.34 14.24 12.76
B. Second-stage results—Mean of dependent variable: 0.085 (IPV)
Stem family −0.13*** −0.15** −0.23***
(0.044) (0.059) (0.085)
Contemporary controls  Yes Yes
Historical and geographic   Yes
$R2$ 0.041 0.041 0.041
Observations 62,740 62,740 62,740

As an instrument for prevailing family structure, this analysis uses a dummy variable indicating whether the province had freedom of testation in the thirteenth century. All models include age, respondent's and partner's level of education, respondent's occupational status, marital status, the head of the household, the presence of children in the household, settlement population size, and year when the survey was conducted. Model 2 adds contemporary controls. Model 3 adds historical and geographic controls. Standard errors clustered by province in parentheses. *$p<0.10$, **$p<0.05$, and ***$p<0.01$.

In my second-stage results shown in panel B of table 4 and consistent with my LPM estimates, I find a negative and statistically significant effect of the historical stem family on IPV: increasing by 1 the average number of married and widowed women in the household in 1860 would decrease the prevalence of IPV in the last decade in Spain by about 13 to 23 percentage points.22 The magnitudes of the coefficients are higher than in the LPM estimates, and again robust to the inclusion of different sets of covariates.

#### Validity of the IV estimates.

The validity of the IV results rests on the assumption that the political differences (and relate inheritance laws) of the Christian conquest affect IPV today, conditional on the controls included in the regression, only through their impact on historical family types. The first concern with this strategy is that the different political institutions could be correlated with different levels of development that might also affect violence against women.

To address this concern, I control for historical and contemporary measures of economic development. First, to control for the level of economic development in the preindustrial era, I include both urbanization rates and population density at the province level in 1787 and 1860. Second, to control for current economic development, I include GDP per capita and unemployment at the province level measured at the time of the survey. Third, as a measure of informal development, I also include an index of social capital at the province level at the time of the survey. Finally, to further control for differences in labor force participation and schooling, all the specifications include household-level information regarding the respondent's occupational status and the respondent's and partner's level of education. As reported in table 4, the negative effect of the stem family is robust to the inclusion of these controls.23

A related issue is the potential long-term impact of the expulsion of converted Muslims (Moriscos) after the Christian conquest had been completed.24 The expulsion of the Moriscos affected mainly the Kingdom of Valencia,25 and some studies suggest that economic effects were concentrated in this specific kingdom (Álvarez-Nogal and Prados de la Escosura, 2007). To address this concern, I run my regressions without including the Valencia region and find similar results.

As additional validity checks of the instrument, I first estimate the reduced form and report these results in table A10 in the online appendix. The results show a (statistically significant) lower prevalence of current IPV in provinces with greater political decentralization in the Middle Ages (measured in terms of freedom of testation in the thirteenth century). Second, I categorize the Spanish provinces into four regions—northeast, northwest, southeast, and southwest—and re-run the IV estimation including region fixed effects. Table A11 in the online appendix reports these results. Model 1 includes a north-south fixed effect to account for the north-south divide. The fact that eastern provinces may be systematically different to western provinces is taken into account in model 2. The specification in model 3 includes a fixed effect for each of the four Spanish regions. Reassuringly, all of these estimates are robust to the inclusion of different sets of fixed effects and similar in magnitude to the baseline IV results.

Another relevant concern with this IV is that freedom of testation might have a direct impact on gender equality and therefore on IPV. For instance, inheritance laws could affect the average bequest women received. Evidence on gender inequalities within the two inheritance systems is scarce and inconclusive.26 To address this and as a general validity test, I follow Conley et al. (2012), and perform the analysis of “plausible exogeneity” by allowing a direct impact of my instrument, freedom of testation, on IPV. I specify a range of plausible values of the direct impact, $b$, and for each of them, I obtain a different confidence interval estimate of the coefficient of interest $β$ and take the union of all these confidence interval estimates. This approach is more conservative and will generally lead to wider bounds than others that require making distributional assumptions about $b$, but for the same reason it requires only that the researcher specifies the maximum and minimum values that $b$ can take.

Table A12 in the online appendix reports the lower and upper bound 90% confidence intervals of the coefficient of interest $β$ for different ranges of $b$, from ($-$0.0025, 0.0025) to ($-$0.0085, 0.0085). The confidence interval of the estimated $β$ always contains negative values and does not overlap with 0 for values smaller than $b∈$ ($-$0.0085, 0.0085). If we assume that the coefficient from the reduced-form regression captures both $β$ and the direct effect of the instrument on the outcome,27 then the direct effect of freedom of testation on IPV should be around 60% of the overall effect for my IV results to become insignificant, which seems implausible.

As a last validity check, I follow Nevo and Rosen (2012) and estimate the bounds of $β$ with “imperfect instrumental variables.” The only assumption that the researcher needs to make is regarding the direction of the covariance between the instrument and the unobserved error term. Instead of assuming that it is 0 as in the traditional IV assumption, I assume that it is positive to reflect that inheritance might have been more beneficial to women in territories with freedom of testation. Since the first-stage coefficient is positive, only upper bounds are returned: the upper bound estimator is $-$0.23, and the upper bound confidence interval is $-$0.04.

Taken together, these results are reassuring that the impact of the stem family structure on IPV is negative and statistically significant, even allowing for plausible amounts of imperfect exogeneity.

## V. Transmission Channels

Different reasons may explain the persistence of this distinct culture of violence against women within Spain. In this section, I explore the potential transmission channels. On the one hand, the institutional environment could have either reinforced or countered internal beliefs about gender roles. For instance, stem and nuclear family regions could have established different labor market institutions, laws, or policies that interacted with culture. On the other hand, it might just be purely cultural transmission. Cultural traits are sticky and slow moving, and there is evidence of a high degree of intergenerational correlation of domestic violence (Pollak, 2004) and of the important role of intrafamily transmission of gender role attitudes (Thornton et al., 1983; Fernández et al., 2004).

Although I cannot completely rule out the institutional channel, the evidence I present is consistent with the cultural transmission channel. First, I am looking at within-country variation, which means that all regions are dealing with the same external environment in terms of the laws, policies, and markets determined by the central authority. Since the beginning of the modern era until the 1980s, the tendency in Spain was to unify regional institutions and policies and centralize power, with few and brief exceptions. Only a very few regions have managed to preserve their own institutions through the centuries. Still, family structure and internal beliefs persisted in territories with different degrees of institutional persistence. This allows us to apply a natural experiment approach: the Basque Country and Navarre kept their own institutions almost throughout this period; Aragon, Catalonia, and the Balearic Islands lost their legislative bodies in the eighteenth century but kept some of their own laws; and Valencia lost both its legislative body and its laws in the eighteenth century.28 Despite the different persistence of local institutions across these regions, all of these territories maintained a stem family structure and today exhibit more equal gender roles.

Regarding internal migration, it was not relevant in Spain until the second half of the nineteenth century, when some short-distance movements began to occur within the regions (Carreras & Tafunell, 2005). During the twentieth century, there were two waves of cross-province migration, both involving agricultural workers moving to industrial towns. The first began in the 1920s. The process was interrupted in the 1930s and 1940s due to the international economic crisis, the Spanish Civil War, and the war's aftermath. Migration resumed in greater numbers from the 1950s through the 1970s. Migrants typically left the southern agricultural regions (Andalucia, Extremadura, and Castilla-La Mancha) to settle in industrial conurbations in Madrid, the Basque Country, Catalonia, and Valencia. This means that a significant number of people from nuclear family regions moved to regions where the stem family had been traditionally prevalent. For this reason, my estimates of the effect of historical family type on IPV should be interpreted as a lower bound.

### A. Evidence from the World Values Survey

Table 3 already shows that the traditional structure of the family not only explains physical and sexual violence but also other kinds of less extreme violence, such as psychological, economic, spiritual, and structural abuse. Furthermore, in this section, I explore the links between historical family types and other measures of gender inequality, using the Spanish sample of the World Values Survey for 1990 to 2007. Apart from demographic data, this survey contains information about values and attitudes toward women. The degree of gender equality is measured through agreement or disagreement with four statements: (a) “When jobs are scarce, men should have more right to a job than women”; (b) “On the whole, men make better political leaders than women do”; (c) “Both the husband and wife should contribute to household income”; and (d) “Having a job is the best way for a woman to be an independent person.”29

For each of these statements, I generate a binary variable that takes the value 1 when the answers indicate beliefs more consistent with gender equality and 0 otherwise.30 To examine the effect of a traditional stem family structure on contemporary attitudes toward gender, I estimate the following equation:
$yi,r=α+βStemr+γXi,r+δzr+ei,r,$
(3)
where $yi,r$ takes the value 1 if individual $i$ living in region $r$ has more gender-equal beliefs. $Stemr$ measures the average number of widowed and married women in the household in the 1860 census, aggregated at the autonomous community level. $Xi,r$ includes control variables at the individual level: sex, age, marital status fixed effects, and educational-level fixed effects. $zr$ measures regional GDP per capita measured in the same year as the dependent variable (1990, 1995, 2000, and 2007). The Spanish sample of the World Values Survey contains only information about the respondent's autonomous community (NUTS 2–level region in the Eurostat classification), which is a higher level than the province (NUTS 3). Since I cluster standard errors by region and my historical data cover only sixteen regions, I report wild bootstrap standard errors with weights assigned at the region level.31

Table 5 reports the results. Controlling for other individual and regional variables, individuals currently living in a region where stem family was socially predominant in the past tend to have beliefs more consistent with gender equality (with the exception of the first measure, for which results do not find any statistically significant effect).

Table 5.
LPM Results from the Spanish World Values Survey: Attitudes toward Gender
Mean of Dependent VariableJob Scarcity 0.76Political Leadership 0.78Household Contribution 0.91Independence through Work 0.80
(1)(2)(3)(4)
Stem family −0.18 0.28* 0.31* 0.87***
(0.303) (0.147) (0.187) (0.162)
Observations 2,853 3,082 2,118 1,299
$R2$ 0.10 0.05 0.03 0.04
Mean of Dependent VariableJob Scarcity 0.76Political Leadership 0.78Household Contribution 0.91Independence through Work 0.80
(1)(2)(3)(4)
Stem family −0.18 0.28* 0.31* 0.87***
(0.303) (0.147) (0.187) (0.162)
Observations 2,853 3,082 2,118 1,299
$R2$ 0.10 0.05 0.03 0.04

The unit of observation is the adult individual (18 years or older) living in Spain between 1990 and 2007. The dependent variables are indicator variables, and value 1 refers to beliefs consistent with greater gender equality. (1) “When jobs are scarce, men should have more right to a job than women”; (2) “On the whole, men make better political leaders than women do”; (3) “Both the husband and wife should contribute to household income”; and (4) “Having a job is the best way for a woman to be an independent person.” Stem family measures the average number of widowed and married women per household based on the 1860 census and aggregated at the autonomous community level. Control variables include sex, age, marital status fixed effects, occupational status fixed effects, educational-level fixed effects, and GDP per capita at the autonomous community level measured in the same year as the dependent variable. Model 4 does not include educational level fixed effects since the dependent variable is defined only for 1990 and education information is missing for that year. Wild bootstrapped standard errors with weights assigned to the autonomous community level (sixteen clusters) in brackets. *$p<0.10$, **$p<0.05$, and ***$p<0.01$.

Potentially, one could argue that stem family territories might be more open-minded and that higher gender equality in these regions is simply an expression of greater tolerance there. To address this concern I look at non-gender-related attitudes. Using the same data set from the World Values Survey, I run equation (3) using as dependent variables life satisfaction, trust, and attitudes toward homosexuality and euthanasia by constructing indicator variables for the following questions: (a) “All things considered, how satisfied are you with your life as a whole these days?” (1 indicates satisfied, 0 dissatisfied); (b) “Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people?” (1 indicates most people can be trusted, 0 otherwise); (c) “Do you think homosexuality can always be justified, never be justified, or something in between?” (1 indicates justifiable, 0 otherwise); and (d) “Do you think euthanasia can always be justified, never be justified, or something in between?” (1 indicates justifiable, 0 otherwise).32 Table A13 in the appendix reports the results. In general, I find no statistically significant differences for these attitudes in stem family territories compared to nuclear family territories. An exception to this rule is made for beliefs about homosexuality, for which results show that in territories in which stem family was more prevalent in the past, individuals tend to find homosexuality less justifiable today.

## VI. Conclusion

Family is a primal and fundamental institution that affects all spheres in the society. Its importance in shaping values and attitudes toward gender is more than evident. In this paper, I analyze the effect of the family structure on the culture of violence against women. I look at the relationship between IPV in Spain and traditional family types (stem and nuclear). My hypothesis is that different family types shaped distinct gender roles and that this has had a long-term impact that explains differences in violence against women today even when these initial conditions no longer persist.

The results show that territories where the stem family was socially predominant in the past exhibit a lower prevalence of IPV today. To address potential endogeneity concerns, I control for an exhaustive set of observable contemporary, historical, and geographic characteristics. I also resort to the political structure at the beginning of the Christian conquest of the Iberian Peninsula (722–1492), as an instrument for family types. Both LPM and IV estimates show a negative relationship between stem family predominance and violence against women.

After discussing different mechanisms that could be consistent with this finding, I show evidence consistent with one of them: that coresidence with the mother-in-law reduced the wife's burden of household work and accentuated her productive role (see section A in the online appendix for details). First, ethnographic data show that indivisible inheritance, a feature of stem families, is positively associated with greater female contribution to agriculture in preindustrial societies. Second, I find a positive correlation between stem family and female employment using both historical and contemporary data for Spain.

Although the importance of the stem family has decreased over the past century, it persisted remarkably (from the beginning of the Middle Ages until the 1970s, evidence suggests)—long enough to potentially explain current behavior. In section V, I show evidence consistent with the idea that attitudes that arose from the traditional family structure and their intergenerational transmission have a role in explaining violence against women today. In this respect, World Values Survey data for Spain show that historical stem family territories have not only less IPV today but also more equal gender roles.

This study contributes to the understanding of the deep historical factors that underlie gender inequality. As reviewed by Giuliano (2014), several studies have found that agricultural technology, language, and geography can affect the role of women in society and have a long-lasting impact up to the present. This paper introduces historical family patterns as yet another element that may also underlie gender relations today. Additionally, it provides an example of the significance of historical events in explaining attitudes today.

## Notes

1

Thornton, Alwin, and Camburn (1983) and Fernández, Fogli, and Olivetti (2004) emphasize the important role of family attitudes toward women and intrafamily transmission of such attitudes in shaping women's role in society.

2

In this separate paper (Tur-Prats, 2017), I find that individuals have a different reaction to relative changes in male versus female unemployment depending on their underlying gender-identity norms. Using Spanish data, I find that in territories with a nuclear family tradition, a decrease in female unemployment relative to male unemployment increases IPV, potentially because men feel their traditional breadwinner role threatened. These effects are more than offset in territories where the stem family was socially predominant in the past.

3

Before that, Pollack (2004) recognized this significant gap in the literature and developed a theoretical model of the intergenerational transmission of domestic violence.

4

See, for instance, Voigtländer and Voth (2012), Guiso, Sapienza, and Zingales (2016), and Becker et al. (2016).

5

Typically, the firstborn son is the one to stay, but in some regions, parents can choose among their sons, and in a very few others, they can choose among both daughters and sons. If there are no sons, a daughter will typically remain in the household, bring her husband with her, and eventually inherit the house.

6

There are fifty provinces in Spain; this division was first introduced in 1833.

7

Figure A1 in the online appendix shows family structure according to the 2001 census. When computing the average number of married and widowed women per household at the province level, I find remarkably lower figures overall and a complete change in geographic pattern.

8

In the westernmost part of the Iberian Peninsula, the Kingdom of Portugal became independent in 1139.

9

The homestead bore a different name in each region, although it carried the same meaning everywhere. It was called baserria (or etxea) in what is now the Basque Country; torre in Aragon; mas or masia in Catalonia; barraca in Valencia; and so forth (Lisón Tolosana, 1972).

10

As illustrative examples, they cite feudal Japan and western Europe on the one hand and imperial and centralized China and Russia on the other.

11

Assumption b is likely to hold for any level of domestic violence. On the one hand, for extreme levels of physical violence (e.g., broken limbs), the wife will not be able to leave the house to perform any farming task. On the other hand, lower levels of violence and emotional abuse could impede cooperation between a husband and wife working together in the fields and lower farm productivity.

12

Moret y Prendesgast and Silvela (1863) compared family law in Castile and in the foral territories (Aragon, the Balearic Islands, Catalonia, Navarre, and some parts of the Basque Country). They found that widows had held a life interest in the estate in Aragon, Navarre, and Catalonia, although Catalonia stipulated this only until 1351. The life interest remained a common practice for widows in some regions of Catalonia and the Balearic Islands. In Navarre and the Basque Country, wives had greater power to manage matrimonial assets jointly owned as community property.

13

Beltrán Tapia and Gallego-Martínez (2018) show that the prevalence of stem families and the existence of waged labor opportunities for women had a beneficial effect on girls' survival in nineteenth-century Spain.

14

The number of observations is smaller because some information is missing for two provinces. Therefore, I run all my regressions with 48 provinces ($N=62,740$). Since I do not use the survey weights in the estimations, I present the summary statistics unweighted.

15

The baseline covariates include respondent's age, level of education, occupational status, marital status, the head of the household, the presence of children in the household, the partner's level of education, settlement population size, and year fixed effects.

16

GDP per capita is entered for the same year as the survey (1999, 2002, and 2006, respectively). The unemployment rate is entered for the same quarter as the survey (second quarter of 1999, first quarter of 2002, and first quarter of 2006). Social capital is entered for 1999, 2002, and 2005, taking 1983 as the baseline (=100 for all provinces). This social capital index combines several variables (ratio of loans to GDP as a measure of network connectivity, proportion of population with primary education or more, unemployment rate, employed individuals as a measure of network dimension, Gini index, life expectancy, population) and parameters (discount rate and reciprocity degree) in a theoretical framework advanced by Pérez García et al. (2006).

17

Urbanization rates in 1787 are missing for four provinces (Soria, Lugo, Orense, and Pontevedra), so I assign to these provinces the average value of the whole region (Soria belongs to Castilla-y-Leon and the remaining three to Galicia) for the same year.

18

There are thirteen provinces where freedom of testation had been established by the thirteenth century: Alicante, the Balearic Islands, Barcelona, Castellon, Gerona, Huesca, Lerida, Navarre, Tarragona, Teruel, Valencia, Vizcaya, and Zaragoza.

19

To put it differently, a change from the most nuclear to the most stem family province would be associated with a decrease in IPV today of 2 percentage points, which is equal to 24% of the sample mean for IPV.

20

I also consider a continuous measure of IPV, constructed as the sum of the thirteen indicator questions listed in table 1, that can take values between 0 and 13. I then estimate regression (1) using a Poisson model and, consistent with previous measures, find negative and significant results. These results are shown in table A7 in the online appendix.

21

In table A8 in the online appendix, I present some additional IV regressions. Panel A shows the first-stage results at the level of variation of the endogenous variable (i.e., at the province level), and the coefficients are very similar. In panels B and C, I present the first- and second-stage results, respectively, of an IV in which the endogenous variable (traditional stem type) is a binary variable that takes the value 1 if the average number of widowed and married women in the household in the 1860 census at the province level is higher than 1 and 0 otherwise. Again, the results are consistent with the baseline results using the continuous measure. If I construct the binary stem family at a value higher than 1.03 (the 75th percentile), I obtain similar results.

22

Taking estimates from column 2, a change from the most nuclear to the most stem family province would be associated with a decrease in IPV today of 5 percentage points, which is equal to 57% of the sample mean for IPV.

23

One might argue that these contemporary and historical controls, mostly aggregated at the province level, are “bad controls” in the sense that they could themselves be an outcome of the Christian conquest. I deal with this concern by running the IV estimation including only the fully exogenous set of control variables. If I control only by the geographic variables and the household-level information arguably unrelated to income (i.e., age, marital status, presence of children in the household, settlement population size, and year the survey was conducted), I find a similar effect of traditional family types on current IPV. Table A9 in the online appendix shows these results.

24

Chaney and Hornbeck (2016) investigate the economic dynamics of the 1609 expulsion of Moriscos from the Kingdom of Valencia. They suggest that Malthusian convergence was delayed due to the persistence of extractive institutions. By limiting labor income, these institutions discouraged migration to former Morisco areas and slowed the demographic response to labor scarcity.

25

Spain expelled approximately 300,000 Moriscos, 110,000 of them living in the Kingdom of Valencia and the rest scattered over the rest of Spain (LaPeyre, 1959).

26

Ferrer-Alòs (2014) found that in the single-heir system—linked to stem families—sons tended to be preferred to daughters, although this varies across stem family regions. In Guipuzcoa (Basque Country), 62% were heirs and 38% heiresses (Oliveri Korta, 2001). In Aragon, Navarre and Catalonia sons were also more often appointed heirs than daughters were. An exception to this rule is found in the Pyrenees, where they followed a strict rule of primogeniture, which implied an equal share of heirs and heiress. Under the equal allocation of bequest system, linked to nuclear families, in principle there was no distinction among siblings according to gender or birth order. However, in practice, families tended to favor some siblings at the expense of others to avoid excessive estate fragmentation. García González (2000) and Gómez Carrasco (2009), using a sample of wills from the eighteenth century in Albacete (Castilla-La Mancha), found evidence of discrimination against women. Even when bequests had similar values, the amount of land was substantially smaller in the inheritances that women received (Bartolomé Bartolomé & García Fernández, 2011).

27

This coefficient is between $-$0.012 and $-$0.013, from table A10 in the online appendix.

28

The Nueva Planta (New Foundation) decrees, signed by Philip V between 1707 and 1716 after winning the War of Spanish Succession, suppressed the political and administrative institutions of the regions that were part of the Crown of Aragon. Eventually, Aragon, Catalonia, and the Balearic Islands were allowed to keep their civil law. The Basque Country and Navarre were not affected, since they had supported Philip V.

29

The first two questions follow Alesina et al. (2013).

30

For statement a, I omit “neither” answer. For statements b to d, I aggregate “agree strongly” with “agree” and “strongly disagree” with “disagree.”

31

Cameron, Gelbach, and Miller (2008) suggest thirty as a rule-of-thumb cutoff for when the number of clusters can be considered small, but they indicate that in general, it will depend on the level of intracluster correlation and the number of observations per cluster.

32

For the trust statement, I assign the value 1 if the individual responds that “most people can be trusted” and 0 if the response is “can't be too careful.” Responses to the other three questions vary on a scale of 1 to 10. Following what I did when looking at attitudes toward gender, I aggregate answers 1 to 5 and 6 to 10.

## REFERENCES

Aizer
,
Anna
, “
The Gender Wage Gap and Domestic Violence
,”
American Economic Review
100
:
4
(
2010
),
1847
1859
.
Alesina
,
Alberto
,
Benedetta
Brioschi
, and
Eliana La
Ferrara
, “
Violence against Women: A Cross-Cultural Analysis for Africa
,”
NBER working paper
21901
(
2016
).
Alesina
,
Alberto
,
Paola
Giuliano
, and
Nathan
Nunn
, “
On the Origins of Gender Roles: Women and the Plough
,”
Quarterly Journal of Economics
128
:
2
(
2013
),
469
530
.
Álvarez-Nogal
,
Carlos
, and
de la Escosura
, “
The Decline of Spain (1500–1850): Conjectural Estimates,
European Review of Economic History
11
(
2007
),
319
366
.
Anderberg
,
Dan
,
Helmut
Rainer
,
Jonathan
, and
Tanya
Wilson
, “
Unemployment and Domestic Violence: Theory and Evidence
,”
Economic Journal
126
:
597
(
2016
),
1947
1979
.
Angelucci
,
Manuela
, “
Love on the Rocks: Domestic Violence and Alcohol Abuse in Rural Mexico
,”
B.E. Journal of Economics Analysis & Policy
8
:
1
(
2008
),
1
43
.
Bartolomé
Bartolomé
,
Juan
Manuel
, and
Máximo García
Fernández
, “
Patrimonios urbanos, patrimonios burgueses. Herencias tangibles y transmisiones inmateriales en la Castilla interior,
Stud. his., H.a mod.
33
(
2011
),
29
60
.
Becker
,
Sascha O.
,
Katrin
Boeckh
,
Christa
Hainz
, and
Ludger
Woessmann
, “
The Empire Is Dead, Long Live the Empire! Long-Run Persistence of Trust and Corruption in the Bureaucracy
,”
Economic Journal
126
:
590
(
2016
),
40
74
.
Beltrán Tapia
,
Francisco J.
, and
Domingo
Gallego-Martínez
, “
What Explains the Missing Girls in the Nineteenth-Century Spain,
Economic History Review
(
2018
), https://onlinelibrary.wiley.com/doi/abs/10.1111/ehr.12772.
Bertocchi
,
Graziella
, and
Monica
Bozzano
, “
Family Structure and Education Gender Gap: Evidence from Italian Provinces
,”
CESifo Economic Studies
61
:
1
(
2015
),
263
300
.
Bertrand
,
Marianne
,
Emir
Kamenica
, and
Jessica
Pan
, “
Gender Identity and Relative Income within Households
,”
Quarterly Journal of Economics
130
:
2
(
2015
),
571
614
.
Bobonis
,
Gustavo
,
Melissa
González-Brenes
, and
Roberto
Castro
, “
Public Transfers and Domestic Violence: The Roles of Private Information and Spousal Control
,”
American Economic Journal: Economic Policy
5
:
1
(
2013
),
179
205
.
Bolch
,
Francis
, and
Vijayendra
Rao
, “
Terror as a Bargaining Instrument: A Case Study of Dowry Violence in Rural India
,”
American Economic Review
92
:
4
(
2002
),
1029
1043
.
Borderías
,
Cristina
, and
Llorenç
Ferrer-Alòs
, “
The Stem Family and Industrialization in Catalonia (1900–1936)
,”
History of the Family
22
:
1
(
2017
),
34
56
.
Boserup
,
Ester
,
Woman's Role in Economic Development
(
London
:
George Allen and Unwin
,
1970
).
Cameron
,
A. Colin
,
Jonah B.
Gelbach
, and
Douglas L.
Miller
, “
Bootstrap-Based Improvements for Inference with Clustered Errors,
this review
90
:
3
(
2008
),
414
427
.
Card
,
David
, and
Gordon B.
Dahl
, “
Family Violence and Football: The Effect of Unexpected Emotional Cues on Violent Behavior
,”
Quarterly Journal of Economics
126
:
1
(
2011
),
103
143
.
Carreras
,
Albert
, and
Xavier
Tafunell
(coord.),
Estadísticas Históricas de España, siglos XIX y XX
(
Bilbao
:
Fundación BBVA
,
2005
).
Chaney
,
Eric
, and
Richard
Hornbeck
, “
Economic Dynamics in the Malthusian Era: Evidence from the 1609 Spanish Expulsion of the Moriscos
,”
Economic Journal
126
:
594
(
2016
),
1404
1440
.
Conley
,
Timothy G.
,
Christian B.
Hansen
, and
Peter E.
Rossi
, “
Plausibly Exogenous,
this review
94
:
1
(
2012
),
260
272
.
Deshpande
,
Hema
, and
Gee Ameena
Suleiman
,
Age, Family Type and Domestic Conflict in Bangalore
(
Bangalore
:
National Institute of Advanced Studies
,
2014
).
Farmer
,
Amy
, and
Jill
Tiefenthaler
, “
Domestic Violence: The Value of Services as Signals
,”
American Economic Review, Papers and Proceedings
86
:
2
(
1996
),
274
279
.
Fernández
,
Raquel
,
Alessandra
Fogli
, and
Claudia
Olivetti
, “
Mothers and Sons: Preference Formation and Female Labor Force Dynamics
,”
Quarterly Journal of Economics
119
:
4
(
2004
),
1249
1299
.
Ferrer-Alòs
,
Llorenç
, “Acceso y distribución de los medios de producción. Herencia y reproducción social,” in
Francisco
Chacón
and
Joan
Bestard
, eds.,
Familias: Historia de la sociedad española (del final de la Edad Media hasta nuestros días)
(
:
Ediciones Cátedra
,
2011
).
Ferrer-Alòs
,
Llorenç
Quién hereda?Desigualdades de género en el acceso a los derechos de propiedad y sistemas hereditarios en España,
AREAS Revista Internacional de Ciencias Sociales
33
(
2014
),
35
47
.
García González
,
Francisco
,
Las estrategias de la diferencia: Familia y reproducción social en la Sierra. (Alcaraz, siglo XVIII)
(
:
Ministerio de Agricultura, Pesca y Alimentación
,
2000
).
García González
,
Francisco
“Las estructuras familiares y su relación con los recursos humanos y económicos,” in
Francisco
Chacón
and
Joan
Bestard
, eds.,
Familias: Historia de la sociedad española (del final de la Edad Media hasta nuestros días)
(
:
Ediciones Cátedra
,
2011
).
Giuliano
,
Paola
, “Female Labor Force Participation: Persistence and Evolution,” in
Steven N.
Durlauf
and
Lawrence E.
Blume
, eds.,
The New Palgrave Dictionary of Economics
(
London
:
Palgrave Macmillan
,
2014
).
Goldschmidt
,
Walter
, and
Evalyn Jacobson
Kunkel
, “
The Structure of the Peasant Family
,”
American Anthropologist
73
:
5
(
1971
),
1058
1076
.
Gómez Carrasco
,
Cosme Jesús
, “
Herencia y transmisión del patrimonio a finales del Antiguo Régimen: Diferentes estrategias en la comunidad mercantil y en la élite local (Albacete, 1750–1830)
,”
Investigaciones Históricas
29
(
2009
)
97
128
.
Greif
,
Avner
, “
Family Structure, Institutions, and Growth: The Origins and Implications of Western Corporations
,”
American Economic Review Papers and Proceedings
96
:
2
(
2006
),
308
312
.
Greif
,
Avner
, and
Guido
Tabellini
, “
The Clan and the Corporation: Sustaining Cooperation in China and Europe
,”
Journal of Comparative Economics
16
:
9
(
2017
),
1
35
.
Grosjean
,
Pauline
, “
A History of Violence: The Culture of Honor and Homicide in the US South
,”
Journal of the European Economic Association
12
:
5
(
2014
),
1285
1316
.
Guiso
,
Luigi
,
Paola
Sapienza
, and
Luigi
Zingales
, “
Long-Term Persistence
,”
Journal of the European Economic Association
14
:
6
(
2016
),
1401
1436
.
Hansen
,
Casper Worm
,
Peter Sandholt
Jensen
, and
Christian
Skovsgaard
, “
Modern Gender Roles and Agricultural History: The Neolithic Inheritance
,”
Journal of Economic Growth
20
:
4
(
2015
),
365
404
.
Instituto de la Mujer
,
Macroencuesta sobre “Violencia contra las mujeres” en España
(
1999
,
2002, 2006
).
Jawarkar
,
Ajay K.
,
Himani
Shemar
,
Vinod R.
Wasnik
, and
Manisha S.
Chavan
, “
Domestic Violence against Women: A Cross-Sectional Study in Rural Area of Amravati District of Maharashtra, India
,”
International Journal of Research in Medical Sciences
4
:
7
(
2016
),
2713
2718
.
LaPeyre
,
Henri
,
Géographie de l'Espagne Morisque
(
Paris
:
S.E.V.P.E.N.
,
1959
).
Le Play
,
Frédéric
,
L'organisation de la famille selon le vrai modèle signalé par l'histoire de toutes les races et de tous les temps
, 3rd ed. (
Tours
:
A. Mame.
,
1884
).
Lisón Tolosana
,
Carmelo
, “Estructura antropológica de la familia en España,” in
Juan Rof
Carballo
,
La familia, diálogo recuperable
(
:
Editorial Karpos
,
1975
).
Lisón Tolosana
,
Carmelo
Invitación a la antropología cultural de España
(
La Coruña
:
Akal
,
1977
).
Mikelarena Peña
,
Fernando
, “
Las estructuras familiares en la España tradicional: geografía y análisis a partir del censo de 1860
,”
Boletín de la Asociación de Demografía Histórica
10
:
3
(
1992
),
15
61
.
Mishra
,
Ashok
,
S. K.
Patne
,
Ranjana
Tiwari
,
Dhiraj
Kumar Srivastava
,
Neeraj
Gour
, and
Manoj
Bansal
, “
A Cross-Sectional Study to Find Out the Prevalence of Different Types of Domestic Violence in Gwalior City and to Identify the Various Risk and Protective Factors to Domestic Violence
,”
Indian Journal of Community Medicine
39
:
1
(
2014
),
21
25
.
Moret y Prendesgast
,
Segismundo
and
Luis
Sevilla
,
La familia foral y la familia castellana
(
:
Academia Matritense de Jurisprudencia y Legislación
,
1863
).
Nevo
,
Aviv
, and
Rosen
, “
Identification with Imperfect Instruments,
this review
94
:
3
(
2012
),
659
671
.
Oliveri Korta
,
Oihane
,
Mujer y herencia en el estamento hidalgo guipuzcoano durante el Antiguo Régimen (S.XVI–XVIII)
(
Donostia
:
Diputación Foral de Guipuzcoa
,
2001
).
Pérez
García
,
Francisco, Lorenzo Serrano
Martínez
,
Vicente Montesinos
Santalucía
, and
Juan Fernández de Guevara
, “
Measurement of Social Capital and Growth: An Economic Methodology
,”
Fundación BBVA working paper
4
(
2006
).
Pollak
,
Robert A.
, “
An Intergenerational Model of Domestic Violence,
Journal of Population Economics
17
(
2004
),
311
329
.
Reher
,
David
,
La familia en España. Pasado y Presente
(
:
Alianza Universal
,
1996
).
Sasaki
,
Masaru
, “
The Causal Effect of Family Structure on Labor Force Participation among Japanese Married Women
,”
Journal of Human Resources
37
:
2
(
2002
),
429
440
.
Sobrequés i Callicó
,
Jaume
,
El pactisme a Catalunya: una praxi política en la historia del país
(
Barcelona
:
Edicions
62
,
1982
).
Stevenson
,
Betsy
, and
Justin
Wolfers
, “
Bargaining in the Shadow of the Law: Divorce Laws and Family Distress
,”
Quarterly Journal of Economics
121
:
1
(
2006
),
267
288
.
Tauchen
,
Helen V.
,
Ann Dryden
Witte
, and
Sharon K.
Long
, “
Domestic Violence: A Non-Random Affair
,”
International Economic Review
32
:
2
(
1991
),
491
511
.
,
Ignasi
,
El món històric de les masies
(
Barcelona
:
Curial
,
1984
).
Thornton
,
Arland
,
Duane F.
Alwin
, and
Donald
Camburn
, “
Causes and Consequences of Sex-Role Attitudes and Attitude Change,
American Sociological Review
48
(
1983
),
211
227
.
Todd
,
Emmanuel
,
La Invención de Europa
(
Barcelona
:
Tusquets Editores
,
1990
).
Tur-Prats
,
Ana
, “
Unemployment and Intimate-Partner Violence: A Gender-Identity Approach
,”
Barcelona GSE working paper
963
(
2017
).
Visaria
,
Leela
, “Violence against Women in India: Evidence from Rural Gujarat,” in
Domestic Violence in India: A Summary Report of Three Studies
(
Washington DC
:
International Center for Research on Women
,
1999
).
Voigtländer
,
Nico
, and
Hans-Joachim
Voth
, “
Persecution Perpetuated: Medieval Origins of Antisemitic Violence in Nazi Germany
,”
Quarterly Journal of Economics
127
:
3
(
2012
),
1339
1392
.
World Health Organization
,
Global and Regional Estimates of Violence against Women: Prevalence and Health Effects of Intimate Partner Violence and Non-Partner Sexual Violence
(
Geneva
:
World Health Organization
,
2013
)
Yugantar Education Society
.
A Study of Nature, Extent, Incidence and Impact of Domestic Violence among Women in the States of Andhra Pradesh, Chhattisgarh, Gujarat, Madhya Pradesh and Maharashtra
(
New Delhi
:
Ministry of Human Resource Development
,
2002
).

## Author notes

I thank Rohini Pande and four anonymous referees for comments that substantially improved the paper. I also thank Larbi Alaoui, Antonio Cabrales, Pierre-André Chiappori, Gordon Dahl, Mercedes Fernández-Martorell, Rosa Ferrer, Christian Fons-Rosen, Enrique García-Bernal, Paola Giuliano, Nathan Nunn, Javier Ortega, Luigi Pascali, Imran Rasul, Pablo Salvador Coderch, Marcos Vera-Hernández, Jordi Vidal-Robert, Hans-Joachim Voth, and seminar participants at the NBER Summer Institute, IE Business School, University College London, Pompeu Fabra University, City University (London), Warwick PhD Conference, Universitat de València, University of Mainz, Barcelona GSE Summer Forum, COSME-FEDEA, University of Adelaide, CERGE-EI, University of Balearic Islands, Utrecht University, Universitat de Barcelona, and University of Southern Denmark for insightful comments. Finally, I am grateful to Juan Torrecillas for sharing the 1887 census data. This project has been cofinanced by the Marie Curie FP7-PEOPLE-2012-COFUND Action, grant agreement 600387. I gratefully acknowledge financial support from Fundación Ramón Areces. All errors remain my own.

A supplemental appendix is available online at http://www.mitpressjournals.org/doi/suppl/10.1162/rest_a_00784.