Policy Research Working Paper 11135 From Patriarchy to Policy Norms, Votes, and Gender Equal Laws Maurizio Bussolo Jonah M. Rexer Lynn Hu South Asia Region Office of the Chief Economist May 2025 Policy Research Working Paper 11135 Abstract Legal institutions play an important role in shaping gender show that ancestral patriarchal culture is a strong predictor equality in economic domains, from inheritance to labor of contemporary norms, and conservative social norms are markets. But where do gender equal laws come from? Using associated with more gender inequality in the de jure legal cross-country data on social norms and legal equality, this framework, the de facto implementation of laws, and the paper investigates the socio- cultural roots of gender ineq- labor market. The paper presents evidence for a political uity in the legal system and its implications for female selection mechanism linking norms to laws: countries with labor force participation. To identify the impact of social more conservative norms elect political leaders who are norms, the analysis uses an empirical strategy that exploits more hostile to gender equality, who then pass less progres- pre-modern differences in ancestral patriarchal culture as sive legislation. The results highlight the cultural roots and an instrument for present-day gender norms. The findings political drivers of legalized gender inequality. This paper is a product of the Office of the Chief Economist, South Asia Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at mbussolo@worldbank.org, jrexer@worldbank.org, and lhu6@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team From Patriarchy to Policy: Norms, Votes, and Gender Equal Laws * Maurizio Bussolo† Jonah M. Rexer‡ Lynn Hu§ The World Bank The World Bank The World Bank * We are grateful to Yurui Hu for excellent research assistance. The authors thank the participants of seminars at the School of Advanced International Studies, London School of Economics, Indian Statistical Institute and Indian School of Business for helpful comments. We thank Ana Maria Tribin Uribe, Alev Gurbuz Cuneo, and Tea Trumbic at the World Bank Women, Business, and the Law team for facilitating access to data and providing helpful assistance in resolving measurement issues. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors and do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. † Lead Economist, World Bank SARCE: mbussolo@worldbank.org ‡ Economist, World Bank SARCE: jrexer@worldbank.org § Economist, World Bank: lhu6@worldbank.org 1 Introduction Across the world, gender inequality in economic domains remains a persistent fact of life: women work less for pay, earn less when they do, and spend more time on childcare and household work (Goldin, 1994; Blau and Kahn, 2017; Van der Gaag et al., 2019). One way to improve gender equality in economic opportunity is to ensure that women enjoy the same legal protection of basic rights – freedom of mobility, marriage, asset ownership, employment – as men. Since Becker (1971, 1981), economic theory has revealed the inefficiencies of legal discrimination and has been highly influential in shaping legislation around labor market discrimination, childcare, divorce, and inheritance. Empirically, a legal equality index from the World Bank’s Women, Business, and the Law (WBL) project – which measures legalized gender discrimination – has recorded significant improvements across the world since 1970. And as Hyland et al. (2020a) show, legal equality is positively correlated with labor market equality, including smaller gender gaps in participation and wages. This cross-country evidence is primarily correlational. Even if improvements in laws were indeed causing a narrowing of gender gaps in economic opportunity, it is not clear that legal reforms alone can be relied upon to boost gender equality.1 There are two reasons for this, both driven by social norms. The first is that laws can be effective only if correctly imple- mented. There are numerous examples of well-intentioned gender-equality-enhancing legis- lation that encountered strong resistance and even social backlash in implementation because of incompatibility with prevailing social norms (Gulesci et al., 2024). Second, laws are en- acted through a political process, which is itself deeply shaped by prevailing social norms.2 A Downsian perspective suggests that in most democratic systems – and even in some au- thoritarian ones – the views of the majority tend to be represented in legislative bodies, and the median voter has the (indirect) power to pass laws (Downs, 1957; Meltzer and Richard, 1981). As such, the equilibrium existence of legalized gender discrimination is itself a function of social norms. Therefore, intervening on laws is helpful and can durably improve gender 1 Several recent studies leverage natural experiments to estimate the causal effects of gender equal legal reform on women’s economic outcomes. For example legal reforms around inheritance (Naaraayanan, 2019; Heath and Tan, 2020), work outside the home (Hallward-Driemeier and Gajigo, 2015), and divorce law (Stevenson, 2008) have been shown to increase women’s economic participation. 2 The relatively quick change in attitudes towards gay people and the ensuing shift in legislation towards same- sex marriage (Fernández et al., 2024) is a clear example of this link. 2 equality only if accompanied by changes in the underlying social norms. This paper studies the fundamental role of social norms in shaping legal and economic gender gaps, combining novel cross-country data on gender equality legislation and its im- plementation, social norms around gender roles, political outcomes, and long-run cultural traits. Specifically, we explore how social norms drive: (i ) the existence and implementation of gender-equal laws, and (ii ) the political processes through which such laws are proposed and enacted. Estimates of the impact of norms on laws in (i ) are likely to be affected by reverse causality, since laws can either formalize widely accepted social norms or signal ap- propriate behaviors, thereby influencing social norms (Lane et al., 2023). To overcome this challenge, we construct an instrument for social norms by leveraging the persistence of pre- modern patriarchal cultural traits, which are not influenced by current laws. To unpack (ii ), we combine data on electoral outcomes, political party orientations, and gender-equal leg- islation to estimate the relationship between social norms, political representation, and legal equality. We identify several core empirical results. First, the cross-country data reveal a robust positive correlation between social norms and gender equality. Countries with more liberal social norms tend to have more gender-equal laws, better implementation of those laws, and higher rates of female labor force participation. A unique feature of our social norms data – which distinguishes it from standard cross-country household opinion data such as the World Values Survey – is that it allows us to differentiate between personal beliefs and social expectations. Personal beliefs measure the extent to which a respondent agrees with a partic- ular normative statement about gender roles – for example, that women should be primarily homemakers. In contrast, social expectations measure the extent to which the respondent be- lieves that others hold that view. Personal beliefs and social expectations are often labeled in the literature as 1st and 2nd order beliefs, respectively. Consistent with results from Bussolo et al. (2024), we show that the correlation between social norms and legal and economic gender equality is strongest for social expectations, and somewhat weaker for personal beliefs. This result is consistent with the central role of per- ceived social pressure in supporting equilibrium adherence to social norms (Bicchieri, 2016; Mackie et al., 2015; Bursztyn et al., 2020). These results are robust to several measurement choices and controls for numerous country characteristics, including income per capita. 3 However, these OLS results are difficult to interpret, given the issues of reverse causality and omitted variables plaguing cross-sectional, cross-country relationships. To address this challenge, we construct an instrument for social norms that leverages the long-run persistence in patriarchal culture to account for reverse causality. Specifically, we use data from Giuliano and Nunn (2018) on the ancestral cultural characteristics of modern populations, which pro- vides the share of the modern population that is descended from pre-modern ancestors prac- ticing a specific cultural trait (e.g., sedentary agriculture). From this data, we construct an index of ancestral patriarchal culture (APC), averaging across the prevalence of several pa- triarchal traits such as polygamy and patrilocality to develop a country-level measure of the extent to which the current population is derived from ancestors practicing patriarchal cul- ture. The identification logic of this instrument is that it should be highly correlated, given the cultural persistence of gender norms (Alesina et al., 2013). However, the ancestral patriarchal culture index was determined long before contemporary economic and legal outcomes were realized. Therefore, variation in norms driven only by ancestral patriarchal culture cannot be affected by reverse causality.3 The instrument has a strong first stage. A 10 point increase in the patriarchy index is associated with a 1.8 p.p. and 5.4 p.p. increase in the social expectations and personal be- liefs, respectively, of the female homemaker norm. These estimates are insensitive to various country-level controls and yield a first-stage F statistic of nearly 50. This first stage estimate is accompanied by a strong reduced-form relationship with our economic and legal outcomes of interest. A 10 point increase in the ancestral patriarchy index is associated with a 4.9% re- duction in female labor force participation relative to the sample mean, as well as a reduction in legal equality of almost equal magnitude. This relationship is heteroskedastic, with greater variance in legal scores at high levels of the ancestral patriarchal culture index. Notable nega- tive outliers include South Asian and Middle Eastern countries, with particularly low female labor force participation and legal rights even given their high degrees of ancestral patriarchal culture. Combining the reduced-form and first-stage estimates yields our 2SLS estimate of inter- est. A 1 point increase in the instrumented social expectations of the female homemaker norm 3 Of course omitted variables remain a concern, as the exclusion restriction of ancestral patriarchal culture may not strictly hold. For example, patriarchal cultures may have fought more wars historically, affecting contempo- rary patterns of economic activity and gender roles through a mechanism outside social norms. 4 results in a 2 point decline in legal rights and a 1.4 p.p. fall in female labor force participation. These estimates are 42% and 22% higher than the OLS correlations, respectively, the former being significantly different from the OLS at the 5% level. Larger 2SLS estimates could arise for two reasons. First, it is likely that social norms are measured with error, since they are derived from survey responses. If this measurement error is classical, then the 2SLS solves the attenuation bias. Second, it is possible that the reverse causality effect is actually negative, so that purging it yields an increase in the coefficient magnitude. This is possible if, for exam- ple, social norms harden from a backlash effect in response to improved legal protections for women. Lastly, we investigate the political mechanism that links social norms and laws. We hy- pothesize that social norms represent underlying preferences over gender equality. Following standard political economy logic, these preferences are expressed through representative po- litical institutions, since more liberal societies tend to elect more liberal politicians with respect to gender equality on average. These liberal political parties, in turn, tend to pass legislation that supports gender equality, improving the WBL score. We test this logic in two regressions: (i ) regressing the ideological composition of national legislative bodies on social norms and (ii ) regressing legal equality on the ideological composition of legislative bodies. In short, the results show that societies with more liberal gender views elect parties more friendly to women’s rights (i ), and that these more liberal parties pass legislation that increases women’s legal equality (ii ). Several secondary results bolster our confidence in the validity of this mechanism. First, social expectations are more strongly correlated with political representation than personal beliefs, consistent with the initial OLS results showing the primacy of social expectations over beliefs. This suggests that social expectations impose constraints on voting behavior independent of individual beliefs, even under a secret ballot. Second, a placebo test reveals that while the parliamentary seat shares of pro and anti women parties are robustly positively and negatively associated with the legal index, the seat share of parties with neutral gender ideology is not associated with legal change. Lastly, we estimate a semi-parametric binned regression in which the legal equality index is regressed on indicators for bins of vote shares of pro-women parties. The results reveal substantial non-linearities: representation of pro- women parties is uncorrelated with legal equality until roughly 50%, at which point the WBL 5 index jumps substantially. These threshold effects suggest that political parties that support women’s rights require majority control of government to enact legislation to improve gender equality. In summary, societies that inherit stronger ancestral patriarchal cultural traits have more restrictive present-day social norms. In turn, these societies experience worse de facto and de jure legal frameworks, even after controlling for governance effectiveness and level of development. Passing more women’s rights protection laws is likely to be ineffective in these contexts, since hostile social norms can undermine implementation. In addition, we also show that in these social contexts it is more difficult to propose and enact laws that improve gender equality in the first place. This is because norms affect the legislative process through politics. Progressive social expectations, even more than personal beliefs, are correlated with electing more pro-women leaders or parties. Then these pro-women parties adopt more gender-equal legislation, particularly after passing the threshold of commanding a majority in government. This evidence indicates a strong link between social norms and gender equality that operates both through the implementation of existing laws and the legislative process itself. 2 Contribution to existing literature This paper bridges and contributes to four principal strands of the literature. The first strand examines separately the relationship between laws and gender equality and between social norms and gender equality, with a particular focus on gender gaps in economic partici- pation. The second strand explores the interplay between social norms and laws, as these two influence each other. The third strand investigates the political mechanisms through which shifts of social norms affect the enactment of formal rules and laws. Fourth, and finally, we contribute to the literature on cultural persistence and its implications for political and eco- nomic outcomes. A brief review of these bodies of research follows. 2.1 Laws, norms, and female labor force participation Descriptive and correlational studies consistently link both formal legal frameworks and social norms to women’s economic participation. Cross-country data, for instance, reveal a robust association between gender-discriminatory laws (e.g., barriers to signing contracts 6 or opening bank accounts) and lower female labor force participation (Hyland et al., 2020b; Gonzales et al., 2015). These patterns—observed in diverse contexts—underscore how le- gal inequalities restrict women’s access to markets, often magnifying wage gaps and limiting upward mobility (Doepke et al., 2012; Duflo, 2012). At the same time, social norms shape “ap- propriate” gender roles (Bicchieri, 2006) and are correlated with labor outcomes: in regions where prevailing beliefs discourage women’s paid work, female labor participation tends to remain low (Bussolo et al., 2024; Fortin, 2005; Jayachandran, 2021). Causal evidence confirms and refines these observations. Studies exploiting policy or legal reforms—such as inheritance rights in India (Sapkal, 2017) or family law revisions in Ethiopia (Hallward-Driemeier and Gajigo, 2015) —find that granting women stronger property and autonomy rights directly boosts their work activity. In the United States, historical legal shifts (e.g., access to birth control) also raised women’s labor market involvement (Goldin and Katz, 2002). Complementary analyses underscore the causal impact of social norms on female la- bor supply. The “epidemiological approach,” for example, controls for institutional and eco- nomic contexts by focusing on immigrant groups in the same host country. Results indicate that women’s labor behavior partly reflects inherited cultural beliefs (Fernandez and Fogli, 2009; Alesina et al., 2013). Experimental interventions confirm that shifting norms—often by correcting misperceptions about community attitudes—can lead to significant increases in women’s job search and employment (Bursztyn et al., 2020; Dhar et al., 2022). Overall, the evidence clarifies that both laws and norms matter, with norms often proving deeply entrenched and affecting women’s decisions long after formal legal barriers fall. 2.2 Interaction between norms and laws A significant body of descriptive work highlights correlations between more egalitarian attitudes and “good” laws (e.g., gender quotas), while case studies document how top-down legal changes sometimes precede noticeable shifts in cultural expectations (Beaman et al., 2012; Maida and Weber, 2022). Yet identifying the direction of influence calls for theoretical models and causal analyses. Theory frameworks (Doepke et al., 2012; Acemoglu and Jackson, 2017; Benabou and Tirole, 2025) posit a two-way feedback loop, where newly enacted laws can gradually displace older norms and, conversely, where supportive norms can accelerate both the adoption and the enforcement of progressive laws. Note that none of these frame- 7 works proposes a strict chronological approach or addresses the ’which came first?’ question. Instead, they emphasize that once a norm or law exists, it influences—yet is also influenced by—formal rules, informal beliefs, or economic and political incentives. Essentially, these frameworks are more concerned with how laws and norms co-evolve over time and reinforce (or undermine) each other, rather than providing an “origin” narrative that definitively says norms precede laws or vice versa. Similarly, empirical investigations provide mixed, yet illustrative, causal insights. On the one hand, female leadership quota in India (randomly assigned) helped dispel biases against women in governance and raised parents’ aspirations for their daughters (Beaman et al., 2012), suggesting that laws can catalyze norm change. On the other hand, rigid local norms often undermine legal reforms: families in patrilineal regions of India, for example, found informal ways to circumvent inheritance laws intended to bolster women’s property rights (Bahrami-Rad, 2021). These conflicting outcomes highlight that legal changes alone may fail to produce their intended results if cultural beliefs remain firmly opposed. In many cases, norms evolve more slowly than legislation, so a newly passed law can be left par- tially unenforced until local attitudes become more accepting. Nonetheless, consistent data from multiple countries show that well-implemented laws—especially those accompanied by public outreach and institutional support—can eventually nudge social expectations in favor of women’s autonomy (Lane et al., 2023; Chai et al., 2022). 2.3 Social norms and the political process Another strand of the literature focuses specifically on the political processes – primarily elections – through which social norms influence the expansion of legal protection of women’s rights. Descriptive studies suggest that societies with more egalitarian gender attitudes are more likely to see women elected, adopt pro-women legislation, and achieve higher female labor force participation (Htun et al., 2019). Causal research sheds light on how norms influ- ence politics: rising social acceptance of female work and leadership translates into greater electoral support for candidates and parties championing women’s rights. In turn, these coalitions enact laws to protect women against discrimination and expand their economic opportunities. For example, in India, women legislators have been found to deliver better local economic outcomes (Baskaran et al., 2024) and reduce gender gaps (Priyanka, 2020). 8 Similarly, Adnan and Miaari (2018) show that political preferences shape wage inequality: in Israel, localities favoring more progressive platforms tend to pass legislation that narrows the gender gap. Thus, this political path appears to follow a sequence: (i) changing social norms and at- titudes produce demand for gender equity, (ii) the electorate increasingly supports female politicians or male politicians with pro-women agendas, and (iii) legislation favorable to women’s interests is passed. Over time, these legal improvements can reinforce societal acceptance of women’s autonomy—although, as the literature also suggests, norms tend to move more slowly. 2.4 Cultural persistence A final and increasingly salient literature, surveyed in Lowes (2022a), emphasizes the en- during nature of social norms –particularly those related to gender roles – and their capacity to remain influential long after the initial economic or institutional conditions that shaped them have disappeared. Drawing on recent contributions in historical political economy, scholars find that pre-industrial agricultural technologies, historical imbalances in the sex ratio, and religious doctrines can have remarkably persistent impacts on women’s position in society. For example, reliance on the plow has been associated with lower participation of women in the labor force, even generations after agriculture ceased to be the mainstay of local production (Alesina et al., 2013). Similarly, male-biased sex ratios in eighteenth-century Australia appear to have left a lasting imprint on men’s attitudes toward women, resulting in more conservative gender norms and reduced female employment, decades later (Gros- jean and Khattar, 2018). Such examples reinforce the view that cultural beliefs, while not immutable, can remain effectively “locked in” when transmitted across generations through family practices, social institutions, and collective narratives. Persistent kinship systems also serve as a key mechanism through which these norms are maintained. Patrilineal versus matrilineal descent rules, the existence of polygynous mar- riages, and payments such as bride price all shape women’s bargaining power, exposure to domestic violence, and household decision-making for centuries (Alesina et al. 2013; Lowes 2022b). Even when overarching laws evolve or new political coalitions enact pro-women leg- islation, local communities can continue to adhere to entrenched customs that constrain the 9 real-world impact of formal reforms. A unifying theme across the reviewed strands—whether they focus on laws, on social norms, or on political processes —is that no definitive origin account emerges regarding whether formal rules precede cultural attitudes or vice versa. Most theoretical contributions underscore a two-way influence, with social norms and legislation co-evolving over time. However, even if this literature does not take a stance on whether norms “came first” or whether laws drove subsequent cultural changes, many studies have accumulated evidence on the persistence of social norms around gender roles. Even well-designed reforms may fail to achieve their aims unless norms become more receptive. Indeed, empirical studies that do isolate the influence of norms (for instance, by exploiting immigrant samples or natural experiments) show that entrenched cultural beliefs continue to play a salient role in both fe- male labor decisions and the adoption or enforcement of legal frameworks promoting gender equity. Furthermore, while scholars have highlighted how pro-women attitudes bolster the electoral success of parties favoring gender equality, much remains to be learned about the strength of this causal channel and about precisely which social norms or beliefs—versus purely personal attitudes—drive these political dynamics. Our paper contributes to this body of work by focusing on one direction of the norms–laws nexus: namely, how progressive social norms are linked to the legal protection of women’s rights, which in turn improves gender equality. We offer two specific contributions. First, building on the persistence literature, we instrument contemporary norms with data on pre- modern ancestral patriarchal cultures. This approach addresses the challenge of endogeneity, that is, that laws themselves might alter norms, and allows us to gauge the magnitude of the effect of (exogenous) variation in norms on gender-related legal frameworks. Second, we examine the political mechanism through which norms shape laws, documenting how more egalitarian attitudes manifest electorally, translate into legislative majorities, and ultimately drive pro-women legal reforms. By doing so, we provide direct evidence on how social norms influence both the likelihood of adopting gender-equal legislation and its subsequent impact on women’s economic participation. 10 3 Data 3.1 Social norms Data on social norms come from the Facebook Survey on Gender Equality at Home, con- ducted in partnership with CARE, the World Bank, and UNICEF. This online survey was rolled out in 2020 and 2021 through Facebook’s social media platform, and invited Facebook users to participate. One advantage of this survey is its extensive country coverage. The 2020 survey round received more than 461,000 responses from 126 locations around the world. Another key feature of the survey, which differentiates it from most global opinion surveys that collect data on gender attitudes, is that it contains information on both personal beliefs and social expectations, rather than just the former. The personal beliefs and social expectations of the respondents were elicited in two dimensions of gender roles: the “male breadwinner” and the “female homemaker” as follows: 1. Female Homemaker: • Personal belief: How much do you agree or disagree with the following statement? “Woman’s most important role is to take care of her home and children.” • Social expectation: Out of 10 of your neighbors, how many do you think believe that a woman’s most important role is to take care of her home and children? 2. Male Breadwinner: • Personal belief: How much do you agree or disagree with the following statement? “Household expenses are the responsibility of the man, even if his wife can help him.” • Social expectation: Out of 10 of your neighbors, how many do you think believe that household expenses are the responsibility of the man, even if his wife can help him? Surveyed individuals answered the questions about personal beliefs on a five-point Likert scale. From the Facebook data, we obtain four country-level average responses. For personal 11 beliefs, we measure the share of respondents who agree or strongly agree with the two nor- mative statements. For social expectations, we measure the average share of neighbors that respondents believe agree with the normative statements. One drawback of the survey is its lack of a representative national sample. However, the national-level estimates are re-weighted to represent the population with internet access, and not only the Facebook user population.4 To allay concerns about representativeness, we also collected data on personal beliefs around gender roles from the World Values Survey (WVS), a global public opinion dataset conducted in seven waves from 1981-2022. We use WVS sam- pling weights so that these survey responses are nationally representative. From the WVS we take national averages of the share of individuals agreeing with the statement: “when jobs are scarce, men have more right to work.” We also construct a WVS gender index, which averages the agreement rates across several questions relating to gender attitudes.5 We ob- tained 290 country-year observations for which the right-to-work question was asked. Since the WVS collects data only on personal beliefs, rather than distinguishing social expectations, we use WVS data primarily as a robustness test on our main results. 3.2 Legal framework and implementation Data on the legal framework come from the Women, Business, and the Law (WBL) ini- tiative at the World Bank. The WBL data contains two products. First is a composite index, ranging from 0 to 100, measuring de jure gender equality in the legal framework for 190 coun- tries from 1971–2024. This index is built from eight sub-indices: mobility, workplace, pay, marriage, parenthood, entrepreneurship, assets, and pension law. For example, workplace laws may cover gender discrimination or harassment law, marriage law may include gen- der inequality in divorce and remarriage rights, and assets include inheritance laws and land 4 Because the Facebook Survey covers only the online population, which is disproportionately young, urban, and highly educated, our country-norm estimates are likely to be more egalitarian than the true population av- erages. See, for example, Rosenzweig et al. (2025); Mercer et al. (2018); Collins et al. (2024); Hossain and Islam (2024). If this selection simply produces a constant level shift for all countries to more liberal (measured) attitudes, then it has no effect on our results, as this is captured in the regression constant. If instead the bias is more severe in the most conservative countries – a reasonable assumption if internet penetration is lower in these countries – then the selection generates a left-censoring problem, which would attenuate OLS estimates. This is because some conservative countries – which are likely to have high patriarchy and low gender equality – are mismeasured as being more liberal. Thus, our reported effects are probably lower bounds on the true causal relationships. 5 These are the share of the population agreeing with the statements: i ) when jobs are scarce, men have more right to work, ii ) being a housewife is fulfilling, iii ) university is more important for a boy than a girl, and iv) men make better political leaders than women. 12 tenure. For most specifications, we take the aggregate WBL legal gender equality index as our outcome of interest In addition, WBL has a newly-constructed index, covering only 2024, which measures the de facto supportive framework for the implementation of gender-equal laws. As an example, if a woman has a right to a passport, but in practice a male escort must accompany her to the office to obtain this document, then the implementing framework for that legal right is weak. Similarly, if anti-harassment laws are on the books, but police action against these crimes is limited, implementation lags legislation. From this we calculate the “implementation gap,” which is the measured as the distance of the country from the frontier of perfect implementa- tion, bounded between 0 and 1. 3.3 Ancestral culture Our instrumental variables strategy requires data on ancestral cultural traits. We obtained these data from Giuliano and Nunn (2018), who compile a database of ancestral character- istics of modern populations. This dataset contains the share of contemporary country-level populations that are descended from ethnic groups that engage in a wide range of traditional practices – from casteism, to endogamy, to intensive agriculture. To estimate these popula- tion shares, the authors combine data on the spatial distribution of ethnolinguistic groups with ethnographic data on ethnic group-level cultural practices from (Murdock, 1967).6 This data allows us to construct variables which measure, for each country, the share of the population descended from ethnic groups that engaged in patriarchal cultural practices in the pre-modern era – for example, polygamy or patrilocality. We construct an index of ancestral patriarchal culture (APC) by averaging across the following variables at the country level: i ) the share of the population practicing polygamy, ii ) the share of the population practicing patrilocality, iii ) the share of the population practicing early marriage, and iv) the share of the population practicing patrilineality. Since these variables are all shares, we obtain an index that varies from 0 to 1 and measures the extent to which contemporary populations derive from ancestral ethnic groups with patriarchal cultural institutions. We also take from this dataset several control variables capturing climatic, geographic, 6 Further details on the constructions of ancestral characteristics of modern populations, as well as access in- structions for the dataset, are available in Giuliano and Nunn (2018). 13 and historical controls, including the average year in which the ancestral patriarchal culture characteristics are observed, the share of the population historically practicing intensive agri- culture, the share of the population living in tropical climates, the average latitude, average distance to the coast, and average ruggedness. 3.4 Electoral outcomes and party ideology We obtained data on political ideology from the Global Party Survey (GPS), released in 2020, which contains party-level information on 1,043 political parties in 163 countries across the world (Norris, 2020). The GPS measures political ideology across a range of issue and rhetorical positions, including nationalism versus multilateralism, minority rights, economic and social conservatism, populism, and so forth. The survey is conducted on 1,861 party and election experts, with an average of 12.3 experts responding per party. Experts rank each party on a 10 point scale for each issue, and their responses are then averaged across experts within party-issue. Of particular interest for this study is the GPS question on party stances toward women’s rights, which is available for 954 parties. Experts were asked to rank parties’ support for women’s rights on a scale of 0-10, with 10 indicating strong opposition to women’s rights. We divide parties into three groups. Pro-women parties are defined as those with average stances of [0-2], neutral parties with stances (2-6), and anti-women parties with stances [6- 10]. We chose these cutoffs – instead of equal intervals – based on the distribution of party stances. Appendix Figure A1 plots a histogram of all the parties in our GPS analysis sample with vertical lines indicating cutoff points. Since the parties in the sample tend to be more in favor of women’s rights than not, we pick a relatively strict cutoff to define pro-women parties, and a relatively more generous one to define anti-women. We then match the GPS data on party ideology to data on political institutions and elec- toral outcomes from the Database of Political Institutions (DPI), from Scartascini et al. (2021). DPI, released in 2020, is a panel of 180 countries from 1975-2020, measuring a wide array of political characteristics, from institutional features such as electoral rules, checks and bal- ances, and term limits, to characteristics of political leadership, including the names and vote/seat shares of current governing and opposition parties. We take from DPI the number of parliamentary seats of all parties listed in the dataset, 14 as well as the total number of seats in the national chamber(s).7 We then match these parties with gender ideologies in GPS, in order to calculate the share of parliamentary seats held by pro, anti, and neutral parties in each country-year, as well as an indicator variable for whether such parties are part of a governing coalition. Since names, abbreviations, and translations of parties differ between the two datasets, we use manual and fuzzy merging techniques to maximize the match rate. In total, we are able to match 70% of the parties from GPS for which we have gender ideology to any party in DPI. For 2020, on average across 169 countries that are observed in both DPI and GPS, we are able to assign a gender ideology to 58% of all parliamentary seats.8 Of course, since, GPS data are observed for a cross-section of parties in 2020, the merge rate falls substantially as we go back in time (see Appendix Figure A2, Panel A). Worse still, nearly 15% of countries in 2020 have no parliamentary seats matched to a gender ideology at all (see Appendix Figure A2, Panel B). This presents a missing data problem, as we cannot assume the gender ideology of these missing seats. As such, for all analysis using the DPI and GPS merged data, we maintain a strict criterion that of less than a 25% unmatched seat share to minimize missing data and reduce bias imposed by measurement error. However, this shrinks our sample substantially – to only 76 countries in 2020, or 54 in the sample using Facebook data – reducing statistical power and representativeness. As such, we test robustness to other thresholds throughout the paper. 3.5 Additional cross-country data and sample Additional cross-country data are taken from the World Bank World Development Indi- cators. These include the female labor force participation rate, measured as a share of the working-age (15-64) female population, and log GDP per capita. In total, our sample for the bulk of our analysis is a single cross-section of 120 country-level observations observed in 2020 (or 2024, for the implementation gap), limited primarily by the availability of the Face- book social norms data. Summary statistics of key variables can be found in Appendix Table A1. 7 DPIonly provides the seats of the top three government and opposition parties. 8 Thismerge rate is slightly higher, at 61%, when we consider the subsample for which the Facebook social norms data are available. 15 3.6 Motivating analysis Figure 1 shows the unconditional positive correlation between gender equal laws and fe- male labor force participation in 2020. Places with greater gender equality in the legal frame- work tend to have higher participation of women in the labor force. Estimates in Table A2 suggest that a one standard deviation increase in the gender equality of the legal framework is associated with a 10.5 p.p. increase in female labor force participation, roughly an 18.6% gain relative to the mean (see Table A1 for means and standard deviations). Of course, this un- conditional comparison likely suffers from omitted variable bias. In particular, social norms are likely to be strongly correlated with laws and influence female labor force participation independently. Columns (2)-(8) introduce various combinations of GDP per capita and its square, to model the standard U-curve (Goldin, 1994), as well as social norms as measured by both personal beliefs and social expectations for the homemaker (Panel A) and breadwin- ner (Panel B) norms. Despite these controls, the magnitude of the relationship between laws and women’s labor market outcomes is remarkably robust. In the “horse race” between so- cial norms and laws, both emerge as important, independent predictors of female labor force participation. Figure 1 (a) shows a scatterplot of the correlation between laws and female labor force participation. Although there is clearly residual variation in female labor force participation for a given level of legal gender equality, the relationship is relatively tight. The univariate R2 is 0.37, substantially higher than the R2 for personal beliefs (0.18) or social expectations (0.27). And while this cross-sectional cross-country analysis is unlikely to uncover a causal relationship, it is certainly consistent with an emerging literature of carefully identified stud- ies linking gender bias in the legal system to economic outcomes (Naaraayanan, 2019; Heath and Tan, 2020; Stevenson, 2008). However, the relationship between laws and female labor force participation is unlikely to be homogeneous. In particular, social norms are likely to be an important catalyst variable: where norms are consistent with gender equality, improvements in laws are likely to result in improvements in outcomes. However, where norms and laws clash, improvements in the legal framework may not lead to improvements in gender equality in economic domains. We test this norms-laws concordance hypothesis by re-estimating the regression, interacting the 16 Figure 1: OLS correlations: laws and female labor force participation (a) Laws and FLFP (b) Laws and FLFP by APC MDG 1 80 Female labor force participation rate (2020) CHE SWE MOZ TZA LTU NLD ETH BLR FIN LVA DNK VNM GBRDEU KAZ AGO KHM JPN AUT AUS CAN AZEKEN SVN PRT MDA RUS ARM CZE THA ISR FRA Impact of legal index on FLFP CMR UGA URYBGR ESP GHA BTN USA SVK IRL HUN BEL HTI ARE JAM MWI BOL POL PER HRV SRB .5 60 QAT BWA LSO RWABFA CIV ZWE GEO LAO ALB PRY GRC BENKGZ BRA COL MNG MYS IDN ITA ARG PAN ZMBCRIMKD MLI CHL ZAF NGA DOM ECU SWZ NIC KWT MMR HND SLV BHR BIH MEX UZB GIN PHL PRI 0 40 BGD SEN GTM LBY TUR OMN SAU NPL LBN TUN IND MRT PAK MAR -.5 20 PSE DJI AFG DZA JOR EGY IRQ -1 0 20 40 60 80 100 0 .2 .4 .6 .8 Legal index Patriarchal culture index Note: Figure (a) shows unconditional correlation between WBL legal index and female labor force participation (FLFP), both measured in 2020. Figure (b) shows the predicted impact of the WBL legal index on FLFP along the distribution of the patriarchy index, controlling for log GDP and its square, as well as regional fixed effects. Patriarchy index (varying from 0 to 1) is measured as the average adherence across four ancestral patri- archal culture (APC) practices. FLFP is measured as the share of working-age women (15-64) active in the labor force within the seven days before the labor force survey, from ILO. Sample is 120 countries for which social norms data are available. WBL legal index with an exogenous measure of patriarchal social norms, the ancestral pa- triarchal culture (see Section 4.2). The results are in Figure 1 (b), which plots the predicted relationship between laws and female labor force participation over the distribution of an- cestral patriarchal culture, allowing for nonlinear interaction effects estimated with a kernel regression. Notably, the relationship is highly heterogeneous, creating an inverse-U. At the extremes of the norm spectrum – highly liberal and highly patriarchal societies – there is no relationship between laws and female labor force participation. In the intermediate cases, however, a significant correlation emerges. The right-hand part of this is consistent with the laws-norms dissonance argument of Acemoglu and Jackson (2017) – liberal laws in conser- vative countries do not produce the intended effects. Instead, the left part of the curve is consistent with the notion that in liberal societies, social norms already allow for women’s economic empowerment, and so legal rights are not a binding constraint. These results reveal that the relationship between laws, norms, and gender equality in economic outcomes is complex and multi-directional. In particular, we understand reason- ably well how gender-equal laws and supportive norms each affect women’s labor-market 17 outcomes. However, what remains less clear is how norms and laws interact, and whether social norms serve as a foundation for gender-equal legal frameworks, thereby amplifying their effect on female labor force participation through the legal system. How can we identify this mechanism, given the endogeneity and reverse causality in all of these variables? The remainder of this paper lays out an empirical approach to identify the pathway from norms to laws, and ultimately to gender inequality in economic outcomes. 4 Empirical strategy 4.1 OLS This paper studies the role of social norms in shaping gender-equal laws (both de jure and de facto) and female labor force participation. The naive approach is simply to regress the key outcomes – legal gender equality and female labor force participation – on the main ex- planatory variable, social norms, using OLS. In particular, the regression equation for country i is yi = α + βnormsi + Xi′ γ + ϵi (1) Where y is the female labor force participation rate as a share of the working age female pop- ulation, the WBL legal index, or the implementation gap. The variable norms is measured as either the share of survey respondents that agree with a particular statement about gender roles (personal beliefs), or the average expectation that respondents hold about their peers be- liefs with respect to a particular gender norm (social expectations). Throughout, we use both the female homemaker norm and the male breadwinner norm as our measures of social norms about gender roles. The vector Xi contains country-level characteristics, including log GDP per capita and its square, region fixed effects, and several other spatial characteristics includ- ing average latitude, distance to the coast, ruggedness (all subnational population-weighted), the share of the population living in tropical climates and traditionally practicing intensive agriculture. Robust standard errors correct for heteroskedasticity. Note that we only have a single cross section for 2020, since social norms are only observed in one survey. Even if social norms were observed over time, the inclusion of country fixed effects would be questionable, since norms tend to be intergenerationally correlated (Fernan- 18 dez and Fogli, 2009) and as such typically do not exhibit sufficient within-country temporal variation. Therefore, equation (1) presents a difficult endogeneity problem. First, there are ob- vious omitted variables. While some of these are subsumed in X , the selection-on-observables assumption is unlikely to hold. Second, the relationship between laws, norms, and economic outcomes is likely to be contaminated by reverse causality. While norms undoubtedly struc- ture the environment in which legal and economic institutions arise, it is equally plausible that rising female economic participation and empowerment may itself undermine traditional gender norms (Field et al., 2021). Similarly exogenous changes to the legal environment that empower women may either shift gender perceptions (Beaman et al., 2009), or cause backlash when they clash with local norms, worsening traditional gender beliefs. 4.2 Instrumental variables approach We address this identification problem in two complementary ways. First, we use an in- strumental variable, described in this section, to alleviate concerns about reverse causality. Second, in Section 7, we present additional evidence for a political economy driving the re- lationship between traditional gender norms and gender unequal laws. Given the inherent limitations of cross-country data, we are unable to rule out all omitted variables. We therefore caveat our results as correlational. However, the evidence we present is suggestive of a re- lationship between social norms and gender inequality in the legal framework that operates through a political mechanism. Our instrumental variables approach is as follows. We leverage the persistence of cultural traits in a two-stage least-squares (2SLS) framework to control for reverse causality in the relationship between social norms and contemporary gender equality outcomes in the legal system and the labor market. In particular, leveraging data from Giuliano and Nunn (2018), we use the degree of patriarchal culture of ancestral populations as an instrument for con- temporary gender norms. We construct an index of ancestral patriarchal culture (APC) by averaging across the following variables: i ) the share of the population practicing polygamy, ii ) the share of the population practicing patrilocality, iii ) the share of the population practic- ing early marriage, and iv) the share of the population practicing patrilineality. Since these variables are all shares, we obtain an index that varies from 0 to 1 and measures the extent to which contemporary populations derive from ancestral ethnic groups with patriarchal cul- 19 tural institutions. This instrument has two attractive properties. First, it is likely to provide a strong first stage and be highly correlated with present-day gender norms, given extensive evidence of the long-run persistence of certain cultural traits (Lowes, 2022a). Second, it allows us to con- trol for reverse causality in estimating the relationship between present-day gender norms and gender equality outcomes. This is because ancestral patriarchal culture is predetermined long before present-day legal or labor market outcomes. By using variation in present-day gen- der norms generated only by variation in ancestral culture, we purge all variation in gender norms that is generated by present day economic or legal conditions. Since ancestral patriar- chal culture predates contemporary economic and legal conditions, the instrument eliminates reverse causality from equation (1). However, while this instrument allows us to rule out reverse causality, it may not fully satisfy the exclusion restriction that ancestral patriarchal culture affects contemporary gender equality outcomes only through contemporary social norms. For example, ancestral patri- archal culture could affect broader development trajectories, levels of social conflict (Rexer, 2022), or the development of economic institutions that affect gender equality independent of social norms. While some of these channels might be controlled for directly, invariably some will be unobservable. In other words, while this instrument solves the reverse causal- ity problem between gender norms and laws or female labor force participation, it does not necessarily solve the omitted variables problem. As such, we should interpret 2SLS estimates with caution. The first stage regression, for country i, is: normsi = µ + π APCi + Xi′ φ + νi (2) Where normsi are either the personal beliefs or social expectations of gender roles, measured by male breadwinner or female homemaker norms, and APCi is the country-level ancestral patriarchal culture index. Xi is a vector of controls that contains, depending on specification, contemporary log GDP per capita, a region fixed effect, the average year in which APC is observed, the share of the population historically practicing intensive agriculture, the share of the population living in tropical climates, the average latitude, average distance to the 20 coast, and average ruggedness. We consider primarily historical and geospatial controls in order to minimize the bias introduced by conditioning on “bad” controls that are causally downstream of ancestral patriarchal culture (Cinelli et al., 2024). The reduced form equation for country i is yi = τ + κ APCi + Xi′ ϑ + ωi (3) Where, depending on specification, yi is the WBL legal gender equality index, the implemen- tation gap, or the female labor force participation rate. To complete the 2SLS estimation, the vector Xi contains the covariates as in (2). All standard errors are robust to heteroskedasticity. 5 OLS results: Laws, norms, and female labor force participation We begin our analysis with the results of OLS estimation of equation (1). The models in Table 1 regress female labor force participation (Panel A), the WBL index (Panel B), and the WBL implementation gap (Panel C) on social norms, measured as the homemaker (columns 1-4) and breadwinner (columns 4-8). Note that we use two key components of social norms: personal beliefs (also called 1st order beliefs) and social expectations (or 2nd order beliefs, i.e. what one believes other people believe should be the expected behavior in a given situation). In addition, columns (2), (4), (6), and (8) augment the specification with the standard u-curve controls for GDP per capita and its square, capturing the dynamics of gender equality over the country income distribution. Two key results emerge from Table 1. First, there is a strong correlation between social norms and all gender equality outcomes. A one p.p. increase in the social expectation of the female homemaker norm is associated with a 1.2 p.p. reduction in female labor force partic- ipation (2.1% of the sample mean), a 1.4 point reduction in the WBL legal gender equality index (1.8% of the sample mean), and a 1.7 p.p. increase in the implementation gap (3.1% of the sample mean). These estimates are similar when controls for GDP per capita are included, and are sim- ilar for both the female homemaker and male breadwinner norms. Controlling for region fixed effects, as well as additional country-level characteristics plausibly associated with so- 21 Table 1: Norms and outcomes: OLS correlations Variable Homemaker norm Breadwinner norm (1) (2) (3) (4) (5) (6) (7) (8) Panel A: Female labor force participation Social expectations -1.176*** -1.153*** -1.330*** -1.349*** (0.152) (0.206) (0.163) (0.221) Personal beliefs -0.367*** -0.353*** -0.491*** -0.462*** (0.066) (0.078) (0.075) (0.087) Per capita GDP (log) (2020) -33.239** -52.765*** -32.281** -46.090*** (14.947) (15.117) (13.743) (14.621) Per capita GDP (log squared) (2020) 1.858** 3.038*** 1.785** 2.642*** (0.836) (0.823) (0.775) (0.795) Observations 119 119 119 119 119 119 119 119 R2 0.268 0.309 0.182 0.286 0.357 0.398 0.251 0.329 Panel B: Legal index Social expectations -1.417*** -1.676*** -1.585*** -1.818*** (0.156) (0.304) (0.165) (0.259) Personal beliefs -0.619*** -0.694*** -0.719*** -0.774*** (0.059) (0.091) (0.071) (0.104) Per capita GDP (log) (2020) 13.332 -17.688 13.100 -5.006 (12.992) (11.961) (9.958) (11.966) Per capita GDP (log squared) (2020) -0.864 0.906 -0.850 0.212 (0.775) (0.668) (0.596) (0.682) Observations 120 119 120 119 120 119 120 119 R2 0.365 0.379 0.489 0.507 0.476 0.496 0.506 0.514 Panel C: Implementation gap Social expectations 0.017*** 0.011*** 0.017*** 0.011*** (0.002) (0.003) (0.002) (0.003) Personal beliefs 0.007*** 0.005*** 0.008*** 0.005*** (0.001) (0.001) (0.001) (0.001) Per capita GDP (log) (2020) -0.128 0.083 -0.109 -0.002 (0.125) (0.117) (0.116) (0.115) Per capita GDP (log squared) (2020) 0.004 -0.008 0.003 -0.003 (0.008) (0.007) (0.007) (0.007) Observations 119 118 119 118 119 118 119 118 R2 0.416 0.491 0.468 0.543 0.394 0.502 0.444 0.541 Note: Robust standard errors in parentheses. All variables are measured in 2020. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, ranging from 0 to 100. Imple- mentation gap measures the percentage-point gap between the WBL implementation score and full implementation of existing gender-equal laws, ranging from 0 to 1. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is 120 countries for which social norms data are available. *** p < 0.01, ** p < 0.05, * p < 0.1. cial norms in Appendix Table A3 – including latitude, distance to the coast, tropical climate, average ruggedness, and he share of the population traditionally practicing intensive agricul- ture – reduces the estimated coefficients only slightly. In Appendix Table A4, we estimate the robustness of the results to using different measures of gender attitudes taken from the World Values Survey (WVS). WVS data produces a very similar pattern of results, suggesting that we are not picking up a statistical artefact arising from measurement error in the World Bank 22 - Facebook survey data. Figure 2: OLS correlations: social norms and legal equality 120 120 100 100 WBL Legal Index (2020) WBL Legal Index (2020) 80 80 60 60 40 40 20 20 0 20 40 60 80 20 40 60 80 Breadwinner norm Homemaker norm Social expectations Personal beliefs Social expectations Personal beliefs Note: Figure shows the unconditional country-level correlations between social norms components – i.e. social expectations (blue) and personal beliefs (red) – and the WBL legal index, all measured in 2020. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is 120 countries for which social norms data are available. The second key result from the OLS analysis is that the estimated coefficients are con- sistently larger for social expectations than for personal beliefs. This is consistent with the literature on social norms and gender equality, which shows that it is the expectation of social sanctions, rather than private beliefs, that constrain behavior (Bursztyn et al., 2020; Bussolo et al., 2024). Specifically, a one-percentage-point shift in social expectations generates an effect on female labor-force participation that is 3.2 times larger than an equivalent shift in personal beliefs; the corresponding multipliers are 2.3 for the legal-equality index and 2.4 for the im- plementation gap. These relative magnitudes are made clear in Figure 2, which plots the OLS correlations between social expectations, personal beliefs, and the WBL legal index in 2020 (similar plots for female labor force participation and the implementation gap are can be found in Appendix Figures A3 and A4). For both dimensions of norms, the slope of the relationship between social expectations and the legal index is much steeper than the slope 23 for personal beliefs, in part because social expectations tend to be much more conservative than personal beliefs and therefore cluster more tightly to the right of the distribution. 6 2SLS results: Laws, norms, and female labor force participation The results in Section 5 reveal strong relationships between norms and gender equality: countries with more conservative social norms have lower female employment, less gender equal legal systems, and weaker implementation of gender equality legislation. But inter- preting these correlations as causal in a single cross-country cross section is clearly unjusti- fied. Norms, employment, and the legal system are jointly determined in equilibrium. As such, these correlations reflect not only the causal effect of norms, but also reverse causality of outcomes in shaping norms, as well as omitted variables shaping both outcomes and social norms. We attempt to identify the pathway of norms to laws in a two-part strategy. First, in this section, we use an instrumental variable predetermined long before contemporary out- comes that allows us to, at the very least, rule out reverse causality. Second, in Section 7 we provide additional evidence in support of a political mechanism linking regressive social norms to gender unequal laws. Table 2: Patriarchal culture and contemporary social norms: first stage Variable Homemaker norm Breadwinner norm Type Expectations Beliefs Expectations Beliefs (1) (2) (3) (4) (5) (6) (7) (8) Patriarchy index 0.188*** 0.115*** 0.531*** 0.404*** 0.176*** 0.107*** 0.416*** 0.290*** (0.028) (0.026) (0.056) (0.067) (0.026) (0.027) (0.050) (0.060) Per capita GDP (log) (2020) 10.663** -20.133* 10.004* -0.284 (4.532) (10.644) (5.096) (10.636) Per capita GDP (log squared) (2020) -0.781*** 0.782 -0.729** -0.302 (0.272) (0.621) (0.299) (0.602) Observations 118 117 118 117 118 117 118 117 R2 0.325 0.605 0.369 0.554 0.273 0.493 0.299 0.460 Note: Robust standard errors in parentheses. Patriarchy index is measured as the average across four ancestral culture patriarchal characteristics, varying from 0 to 1. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is 120 countries for which social norms data are available. *** p < 0.01, ** p < 0.05, * p < 0.1. Table 2 presents the first stage of our instrumental variables approach, showing estimates from equation (2) of the correlation between pre-colonial patriarchal culture and contempo- rary gender norms measured in 2020. A 10-point increase in the country-level patriarchal 24 culture index – which ranges from 0 to 100 – is associated with a 1.8 p.p. increase in the social expectations of the homemaker norm (column 1) and a 5.3 p.p. increase in reported personal beliefs in the homemaker norm (column 3), both significant at the 1% level. These estimates are not sensitive to controls for GDP per capita in columns (2) and (4) or the choice of gender norm (homemaker vs. breadwinner). The results are also robust to the inclusion of region fixed effects and additional country-level controls in Appendix Table A5, remaining significant at 1% though the magnitudes are somewhat reduced. The results suggest a robust first-stage relationship between ancestral pre-colonial cultural characteristics and present-day gender norms. Figure 3, top panel, shows the first stage relationship in graphical form. The tight correlation around the line of best fit (with an unconditional R2 of 0.33) yields a Kleiber- gen and Paap (2006) first stage F-statistic of 49.1 for the homemaker norm and 45.6 for the breadwinner norm (see Table 4), well above the critical values for the (Stock and Yogo, 2005) weak instruments test.9 Controlling for national per capita income and its square reduces the estimate of the first stage F by more than half, to 19.5, though it remains large enough to clear the critical value thresholds and avoid weak instrument inference. Table 3: Ancestral patriarchal culture and legal and economic outcomes: reduced form Outcome FLFP Legal index Implementation gap (1) (2) (3) (4) (5) (6) Patriarchy index -0.279*** -0.204*** -0.379*** -0.344*** 0.004*** 0.003*** (0.063) (0.061) (0.059) (0.080) (0.001) (0.001) Per capita GDP (log) (2020) -43.790*** -1.649 -0.019 (14.764) (12.775) (0.124) Per capita GDP (log squared) (2020) 2.625*** 0.213 -0.003 (0.817) (0.739) (0.007) Observations 117 117 118 117 118 117 R2 0.140 0.242 0.258 0.299 0.232 0.482 Note: All outcome variables are measured in 2020. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, ranging from 0 to 100. Implementation gap measures the percentage-point gap between the WBL implementation score and full implementation of existing gender-equal laws, ranging from 0 to 1. Patriarchy index is measured as the average across four ancestral culture patriarchal characteristics, varying from 0 to 1. Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. Table 3 contains reduced form results from the estimation of equation (3), relating pre- colonial patriarchal culture to present-day economic and legal gender equality outcomes. The estimates reveal a strong relationship between gender equality outcomes and patriarchal cul- 9 Wenote, however, that the first-stage F still falls below the minimum required to ensure a test with size 0.05, according to (Lee et al., 2022). 25 ture. A 10-unit increase in the patriarchy index on the 100-point scale is associated with a 2.8 p.p. reduction in the female labor force participation rate, or a 4.9% reduction on the sample mean, significant at the 1% level. The estimate is robust to controlling for the participation- income per capita u-curve in column (2). Figure 3: First-stage and reduced form correlations: culture, social norms, laws, and economic outcomes DJI AFG 80 70 IDN UZB MYS MLI PAK PHL PAK SEN MRT EGY KWT SAU PHL BGD DJI SEN BFA BGD AZE QAT LSO MDG OMN BEN MLI IND LBN BEN IND CIV UZB MDG NPL SAU GEO IDN AFG MMR HND CIV ARM GTM ZAF KGZ GIN BHR ZWE JOR ARE SWZ TUR 70 ARM VNM ZAF MYS GIN BHR AZE BIH DZA ZWE BWA QAT KWT SWZ ETH 60 GEO TUNDZA LBN ARE OMN EGY Breadwinner norm MEX Homemaker norm ZMB PER HTI NGA BFAUGA IRQ ETH CMR NIC DOM ECU THA KAZ TZA LBY ALB NPLLSO BTN HND KAZ KGZ MRT ALB JOR KEN ZMB JAM GHA MWI PRY SLV VNM POL CMR MAR TWN RWA GTM TUN CHL LAO BOL BWA KEN DOM PRYGRC MKD ECU HTI AGO NGAKHM LBY IRQ BTN TUR RWA JAM NIC MWI UGA MAR RUS CRI KHM MDA MMR MEX GHA BRA COL ARG MOZ MOZ BLR BGR SLV TZA HRV MKD GRC FRA LTU BIH COL LAO BOL ITA PAN MDAURY HUN JPN 60 ITA RUS CRIARG PAN CHL PER FRA HUN LTU TWN 50 BLR BGR AGO SVK CZE ISR BRA THA CZE LVA POL MNG CHE BEL LVA HRV IRL SVK DEUAUS MNG PRI AUS SVN ISR AUT USA PRI URY USA IRL NLD PRT PRT GBR ESP DEU CHE JPN SWE CAN GBR 50 SWE AUT CAN ESP 40 FIN SVN BEL NLD FIN 40 30 DNK DNK 0 .2 .4 .6 .8 0 .2 .4 .6 .8 Patriarchal culture index Patriarchal culture index Female labor force participation rate (2020) 100 SWE DEU CAN DNK BEL FRA LVA MDG GBR NLD IRL ITA AUS AUT ESP PRT GRC PER PRY FIN HUN SVK LTU POL SVN CZE HRV 80 SWE CHE NLD MOZ FIN TZA LTU LVA DNK DEU CAN BLR VNM ETH USA BGR ALB TWN GBR AUT AUS AGO KAZ JPN URY MEXECU SLV KHM SVN MDADOM LAO BOL ZAF PRT AZE KEN CHE NIC GEO ZWE RUS MDA FRABGR ARM CZE ISR MKD ARM MNG BIH THA WBL legal index (2020) USA URY ESP SVK UGA CMR CRI MOZ CIV BFA TUR IRL BEL GHA HUN BTN COL BRA PRI VNMTZA 80 PAN PHL ZMB NPL KEN ISR RWA MWI HTI PER JAMGEO POL HRV ARE MWI ARG HND CHL GHA THA BLR KHM AZE KGZ BEN UGA JPN LSO MAR BTN 60 PRY GRC LAO CIV BOL MNG BFA ZWE QAT BWA ALB LSO RWA RUS AGO GTM GIN ITA BRA COL IDN KGZ BEN MYS DJI IND ETH CRIARG ZMB MKD PAN KAZ SAU CHL NGA ZAF MLI JAM DOM NIC ECU UZB MDG HND MMR SWZ KWT SEN IDNNGA TUN MLI BWA SLV BIHBHR MEX PHL 60 MMR HTI DZA CMR ARE 40 PRI SEN UZB GTM GIN BGD LBY TUR LBN PAK LBY MYS BGD OMN SAU NPL MRT LBN BHR SWZ TUN IND EGY IRQ MRT PAK 40 JOR AFG 20 DJI MAR OMN DZA AFG JOR EGY IRQ QAT KWT 20 0 0 .2 .4 .6 .8 0 .2 .4 .6 .8 Patriarchal culture index Patriarchal culture index Note: Figure shows first stage and reduced form relationships for the instrumental variables strat- egy. All outcome variables are measured in 2020. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, rang- ing from 0 to 100. Patriarchal culture index is measured as the average across four ancestral culture patriarchal characteristics, varying from 0 to 1. Sample is 120 countries for which social norms data are available. This correlation suggests minimal convergence in gender equality over time. Countries that are initially conservative in their gendered cultural practices still have large gender gaps in economic outcomes today. We argue that this persistence operates through social norms, which in turn shape the legal environment. Columns (3) and (4) estimate the reduced form relationship between pre-colonial patriarchal culture and the WBL legal equality index. A 10- 26 unit increase in the patriarchy index is associated with a 3.8 point reduction in country-level legal gender equality score, or a 4.9% reduction on the sample mean, significant at the 1% level. Again, the inclusion of controls for log income per capita does not affect the results. Figure 3, bottom panel, shows the reduced form relationship between legal equality (left) or female employment (right) and the instrumental variable in a scatterplot. The legal equality plot suggests a relatively tight, relationship, reflecting the unconditional R2 if 0.26, though the relationship appears heteroskedastic, with greater variance in legal scores at higher levels of patriarchal culture. Notable negative outliers are South Asian and Middle Eastern countries, with particularly low levels of legal equality even given their high degrees of ancestral patri- archal culture. In contrast, many Sub-Saharan African countries, with relatively high levels of legal equality and female labor force participation despite their high APC scores. Lastly, in columns (5)-(6), we test the reduced form association between culture and the present-day implementation of gender-equal legislation. Societies with ancestral patriarchal cultures have weaker implementation of gender equal legislation: a 10-point increase in the patriarchy index is associated with a 7.3% increase in the “implementation gap” of the legal framework. Appendix Table A6 contains robustness tests of the reduced form results to the inclusion of region fixed effects and additional controls. Though the estimates are somewhat weakened in magnitude and significance, they remain of the correct sign in nearly all specifi- cations. Table 4 contains two stage least squares estimates of the instrumental variables model in which ancestral patriarchal culture serves as an instrument for present-day gender norms. All estimates are shown with and without controls for the per capita income u-curve, across the three main economic and legal outcomes, and for both the homemaker norm (Panel A) and the breadwinner norm (Panel B). The models use social expectations as the main indepen- dent variable, rather than personal beliefs, following the theoretical and empirical literature demonstrating the primacy of expectations in defining social norms (Bursztyn et al., 2020). However, the results are replicated in Appendix Table A8 using personal beliefs, yielding similar results, as personal beliefs and social expectations are highly correlated in practice. Across all outcomes, the results are similar in magnitude to the OLS estimates in Section 5. In the 2SLS estimates, a 1 point increase in the share of the population believed to hold conservative beliefs about the homemaker norm is associated with a 1.4 p.p. reduction in fe- 27 Table 4: Norms and legal and economic gender equality: 2SLS Outcome FLFP Legal index Implementation gap (1) (2) (3) (4) (5) (6) Panel A: Homemaker norm Social expectations -1.438*** -1.775*** -2.016*** -2.986*** 0.023*** 0.024*** (0.297) (0.502) (0.303) (0.567) (0.004) (0.006) First-stage F-statistic 49.069 19.467 46.402 19.467 46.402 19.467 Observations 117 117 118 117 118 117 R2 0.245 0.262 0.289 0.205 0.378 0.384 Panel B: Breadwinner norm Social expectations -1.559*** -1.916*** -2.158*** -3.223*** 0.024*** 0.026*** (0.302) (0.516) (0.294) (0.567) (0.004) (0.006) First-stage F-statistic 45.641 15.402 45.433 15.402 45.433 15.402 Observations 117 117 118 117 118 117 R2 0.337 0.348 0.400 0.244 0.306 0.309 GDP Controls No Yes No Yes No Yes Note: Robust standard errors in parentheses. All outcome variables are measured in 2020. Estimates are from a 2SLS model where social expectations is instrumented with the patriarchal culture index. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, ranging from 0 to 100. Implementation gap measures the percentage-point gap between the WBL implementation score and full implementation of existing gender-equal laws, ranging from 0 to 1. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Patriarchal culture index is measured as the average across four ancestral culture patriarchal characteristics, varying from 0 to 1. Sample is 120 countries for which social norms data are available. *** p < 0.01, ** p < 0.05, * p < 0.1. male labor force participation (2.5% of the sample mean), an estimate roughly 22% larger than the unconditional correlation in Table 1 Panel A column (1), though we cannot rule out that these estimates are statistically identical to each other. The similarity in magnitudes suggests that the reverse causality channel is a not particularly large component of the unconditional correlation, and may even be positive. With respect to the legal index in Table 4 column (3), in the 2SLS model we find that a 1 point increase in the social expectation of the female homemaker norm is associated with a 2 point reduction in legal gender equality (2.5% of the sample mean), significant at the 1% level. This estimate is 42% larger than the OLS estimate in Table 1 Panel B column (1), and significantly different at the 5% level. Similar, slightly larger effects are observed in Panel B for 28 the male breadwinner norm. The 2SLS estimates are also robust to controlling for region fixed effects and country-level covariates (Appendix Table A7) as well as accounting for missing data in the components of the patriarchal index (Appendix Table A9). 7 Laws and norms: Political mechanism We now turn to the mechanism linking social norms to legal equality. We hypothesize that socially conservative countries will tend to select political leaders that reflect the under- lying social norms of the country. These leaders, in turn, will pass legislation that enshrines gender inequality in the legal framework. Ultimately, this inequality translates into the eco- nomic sphere in the form of reduced participation in economic life. This effect is likely to be particularly strong for democratic countries, where elections provide a clear mechanism for translating gender preferences into political outcomes. But the effect may also obtain in non-democracies, which remain responsive to citizen preferences and election results despite the lack of direct accountability mechanisms (Miller, 2015). We test this hypothesis in two steps. First, we use data on national legislative electoral outcomes and party ideology to estimate the relationship between social norms and political outcomes, in order to demonstrate that a nation’s gender norms are reflected in the persuasion of its political leadership. In this step, we run the following regression for country i in 2020. pi = δ + ψnormsi + γlog( GDPi ) + θr + υi (4) Where pi gives the political representation of pro-women’s rights parties in country i’s na- tional parliament in 2020, measured as either the share of seats held by these parties, or an indicator variable specifying whether a pro-women’s rights party is part of the national gov- erning coalition. θr represents a fixed effect for World Bank regions, included in some spec- ifications to control for time-invariant regional heterogeneity. Our main variable of interest, normsi , is a vector including both personal beliefs and social expectations of either the male breadwinner or female homemaker norm. In robustness tests, we also allow for electoral out- comes to be averaged over a longer time period, from 2015-2020, to smooth out variations in political cycles. 29 In the second step, we show that national legislative control by political parties hostile to women’s rights is associated with significantly weaker WBL scores. In this step, we estimate the following two-way fixed effects (TWFE) panel regression for country i at time t WBLit = κ + ϕ pit + ηi + ηt + uit (5) Where WBLit is country i’s WBL score at time t and pit is the share of seats in country i’s national parliament held by parties for, against, or neutral with respect to women’s rights, depending on specification. Lastly, ηi and ηt and country and year fixed effects, which control for country-specific heterogeneity and aggregate time trends, respectively. Standard errors are clustered at the country level. Both equations (4) and (5) suffer from a measurement challenge. The party-level match rate between the DPI and the GPS is far from 100% and varies substantially over time and countries. Parties for which gender ideology is missing cannot be assumed neutral. As such, we calculate the share of parliamentary seats for which we can match a gender ideology in a given country-year. We then restrict the sample to country-years where 75% or more of the parliamentary seats are matched. This restriction cuts the sample significantly, to only 53 countries in 2022. However, we test robustness to this threshold. The estimation in equation (5) uses country fixed effects. Given the small number of obser- vations after imposing the 75% matched seat share sample restriction, it is reasonable to assess the extent of residual within-country variation in the pro-woman seat share. We regress the parliamentary seat share of pro, anti, and neutral parties on country and year fixed effects and obtain the R2 of this regression, plotting these values in Appendix Figure A6. The results re- veal that the two-way fixed effects absorb a substantial share of the variation in parliamentary seat shares, particularly in the more restrictive sample where the 25% missingness criteria is imposed. The R2 in the sample with no restriction ranges between 0.8 and 0.85, monotonically increasing in the sample restrictiveness. However, even in the most restrictive sample, the R2 of the fixed effects remains below 0.9 for the seat share of pro-women parties, suggesting that there remains enough residual variation to identify the parameter of interest ϕ in the TWFE model. Table 5 provides estimates for equation (4). Panel A uses the female homemaker norm 30 Table 5: Gender norms and political outcomes Outcome Seat share In government (1) (2) (3) (4) (5) (6) Panel A: Homemaker norm Homemaker expectation -0.871 -2.096*** -2.206*** -0.027** -0.036*** -0.042*** (0.857) (0.657) (0.796) (0.011) (0.011) (0.012) Homemaker beliefs 0.478 1.025* 1.082 0.004 0.009 0.010 (0.403) (0.602) (0.674) (0.006) (0.008) (0.009) GDP per capita (log) (constant 2015 USD) 4.883 -2.233 0.008 -0.108 (5.291) (7.723) (0.072) (0.104) Region FE No Yes Yes No Yes Yes Observations 53 53 52 53 53 52 R2 0.061 0.225 0.221 0.129 0.246 0.258 Panel B: Breadwinner norm Breadwinner expectations 0.148 -1.330** -1.239* -0.019 -0.034*** -0.036*** (0.959) (0.576) (0.624) (0.013) (0.010) (0.010) Breadwinner beliefs 0.298 0.460 0.472 0.005 0.009 0.009 (0.525) (0.501) (0.503) (0.007) (0.007) (0.007) GDP per capita (log) (constant 2015 USD) 6.691 2.674 0.043 -0.063 (5.402) (7.467) (0.076) (0.111) Region FE No Yes Yes No Yes Yes Observations 53 53 52 53 53 52 R2 0.043 0.145 0.140 0.098 0.230 0.231 Note: Robust standard errors in parentheses. Seat share is the share of parliamentary seats belonging to political parties expressing an ideology in favor of gender equality. In government is an indicator variable for whether any of these pro-equality parties is a part of the ruling coalition. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is all countries with political leadership data in 2022 for which 75% or more of the parliamentary seats can be matched to a political gender ideology. All variables are measured in 2020. *** p < 0.01, ** p < 0.05, * p < 0.1. while Panel B considers the male breadwinner. The estimates reveal that conservative social norms are significantly negatively associated with representation of pro-women parties. In the most rigorous specification in column (3), which includes both controls for income per capita and regional fixed effects, a one-point increase in the female homemaker norm is as- sociated with a 2.2 percentage point reduction in the parliamentary seat share of pro-women parties, equivalent to 10.8% reduction on the sample mean. This estimate is significant at the 1% level, and is stronger for the homemaker norm than the breadwinner norm (Panel B), where it is only marginally significant at the 10% level. Considering participation in govern- ing coalitions in column (6), a one p.p. increase in the prevalence of the homemaker norm is associated with a 4.2 p.p. reduction in the probability of participation for pro-women parties, 31 or 16.2% on the sample mean, again significant at 1%. Importantly, the estimate on personal beliefs is quantitatively small and never statistically significant in any of the specifications. This is consistent with results throughout this paper, including in Table A2 and Figure 2, which point to the primacy of social expectations, rather than personal beliefs, in driving behavior. This result suggests that voters select conservative leaders because of expected social sanctions for deviating from traditional social norms, rather than because of individual beliefs about women’s rights. Figure 4 plots the relationship between political outcomes and the female homemaker norm in a binned scatterplot, controlling for region fixed effects, log GDP per capita, and personal beliefs. The estimates reveal a largely linear conditional correlation in our sample. Appendix Figure A5 recreates the same plot for the breadwinner norm, confirming the weaker relationship than the homemaker norm. Figure 4: Correlation between gender norms and political orientation: homemaker norm 60 .6 Seat share of pro-women parties Pro-woman party in govt 40 .4 .2 20 0 0 -.2 50 55 60 65 70 50 55 60 65 70 Homemaker social expectations Homemaker social expectations Note: Figure shows the correlation between social expectations of gender roles and political out- comes, binning at 20 quantiles of the distribution of social expectations and controlling for log GDP per capita, average personal beliefs in the homemaker norm, and region fixed effects. Seat share is the share of parliamentary seats belonging to political parties expressing an ideology in favor of gender equality. In government is an indicator variable for whether any of these pro-equality par- ties is a part of the ruling coalition. Social expectations are measured as the country-level average belief about the share of peers that agree with the homemaker norm, ranging from 0 to 100. Sample is 53 countries with political leadership data in 2022 for which 75% or more of the parliamentary seats can be matched to a political gender ideology. All variables are measured in 2020. We conduct several robustness tests on this result. Appendix Table A10 compares the re- 32 sults in the 25 and 75% samples without controls. Appendix Table A12 uses a time window of 2015-2020 to allow for greater political turnover. None of these robustness tests meaningfully affects the result. Next, we consider the second step of the political economy mechanism in equation (5), which is estimated in Table 6. All models include country and year fixed effects and cover a country-year panel from 1975 to 2020. In each model, we vary the definition of pit to measure the parliamentary seat share of pro-women parties (1), anti-women parties (2), or neutral parties (3). Note that the panel is unbalanced and the country composition of the sample may change over time, since the 75% threshold may bind differentially depending on the match rate for a given country-year. Table 6: Political representation and gender-equal laws Outcome WBL index (1) (2) (3) Seat share (pro-women) 0.069*** (0.022) Seat share (anti-women) -0.099*** (0.036) Seat share (neutral-women) -0.028 (0.028) Country FE Yes Yes Yes Year FE Yes Yes Yes Observations 2576 2576 2576 R2 0.916 0.916 0.915 Note: Standard errors in parentheses clustered at the country level. Out- come variable is the WBL legal score, ranging from 0 to 100. Seat share is the share of parliamentary seats belonging to political parties expressing a given gender ideology. Sample is all country-years with political leader- ship data in 2022 for which 75% or more of the parliamentary seats can be matched to a political gender ideology. *** p < 0.01, ** p < 0.05, * p < 0.1. The results present clear evidence in favor of the hypothesis that party ideology shapes the legal framework with respect to gender equality. An additional percentage point of seat share allocated to a party in favor of women’s rights is associated with a 0.069-unit improve- ment on the WBL index. In contrast, allocating this percentage point in representation to an anti-woman party is associated with a 0.099 point reduction in the WBL index, with both es- timates significant at 1%. As a placebo test, we consider the impact of parties classified as 33 neutral in their ideology with respect to women’s rights in column (3). We find that political representation of these neutral parties has no significant association with gender equality in the legal framework. We test the robustness of these results to several different potential threats. Appendix Ta- ble A13 shows that the results are robust to using the less stringent 25% cutoff for our match rate, though the anti-women coefficient loses significance in this sample. Appendix Table A14 show that the results are unaffected by controlling for time-varying log GDP per capita and region-by-year fixed effects, for both the 25% and 75% missingness thresholds, respectively. Finally, it is plausible that progressive gender ideology is correlated with overall left-wing orientation, in which case our estimates would be contaminated by omitted variable bias. Appendix Table A15 controls for variables indicating whether a left-wing party is in govern- ment or the executive, as well as the left-wing parliamentary seat share, for the 25% and 75% samples. The coefficient on pro-women parties remains robustly positive and significant. An additional piece of evidence provides further support to our mechanism. In Figure 5, we plot coefficients from a modified estimation of equation (5), in which the linear term pit , the share of seats for pro-women parties, is replaced by indicators for seat shares binned in ten p.p. intervals. The coefficients are normalized relative to the omitted group, country- years with [0-10] percent representation of pro-women parties. The estimates reveal a jump in the WBL score above 50%, when pro-women parties attain parliamentary majorities. This nonlinear relationship provides additional support for the political mechanism driving the relationship between norms and laws: more liberal societies elect more progressive political parties, which translate into more gender-equal laws. However, the final stage of this causal chain should exhibit the type of threshold effects observed in Figure 5, since popular prefer- ences are more easily translated into policy outcomes under majority rule. What kinds of laws are most affected by progressive politics? The WBL data contains detailed scores on gender inequality across 8 sub-categories, which together are aggregated up to the total index: workplace, marriage, mobility, pay, parenthood, assets, entrepreneur- ship, and pension laws. Appendix Figure A8 plots coefficients from estimating equation 5 for the seat share of pro-women parties, varying the outcome variable across all WBL sub- indices. The results reveal that workplace, marriage, and mobility laws see the largest gains from increasing representation of pro-women parties. In contrast, laws around female en- 34 Figure 5: Impact of political regimes on gender-equal laws: threshold effects 15 10 Impact on WBL 5 0 -5 0 10 20 30 40 50 60 70 80 90 Pro-women seat share Note: Figure shows coefficient estimates and 95% confidence intervals from a binned regression of WBL index on indicators for 10 p.p. bins of the seat share of pro-women parties, omitting the 0-10 bin and controlling for country and year fixed effects. Outcome variable is the WBL legal score, ranging from 0 to 100. Seat share is the share of parliamentary seats belonging to political parties expressing a given gender ideology. Sample is all country-years with political leadership data in 2022 for which 75% or more of the parliamentary seats can be matched to a political gender ideology. Standard errors are clustered at the country level. trepreneurship and pensions are not significantly affected. These results are reasonable, as the former are politically salient dimensions of gender inequality in the legal system, while the latter are not. Taken together, these results establish a clear link between social norms and gender equality via a political economy mechanism. 8 Conclusion Legal institutions structure the economic “rules of the game” and in turn play an impor- tant role in determining the functioning of factor, asset, and output markets. Gender inequal- ity in the legal framework – restrictions on women’s inheritance and asset ownership rights, 35 physical mobility, labor market participation, or entrepreneurship – begets gender inequality in economic outcomes. But where do gender-unequal laws come from? This paper studies how social norms inherited from the pre-modern ancestral culture shape the political process, patterns of representation, and, in turn, gender equality in legal institutions. Our analysis reveals three key findings. First, consistent with existing literature, we iden- tify strong persistence in cultural norms – countries with ancestral cultures that engaged in patriarchal practices such as patrilocality, polygamy, and male inheritance, tend to have less progressive gender norms today, even conditional on their level of economic development. Since this ancestral patriarchal culture predates present-day legal and economic outcomes, we use it as an instrument for social norms in identifying the impact of norms on laws, purged of reverse causality. Second, using this instrumental variables approach, we identify a positive relationship between liberal social norms, legal equality, and ultimately female labor force participation. Finally, we provide evidence for a political economy mechanism linking social norms to gender-equal laws. Social norms translate into political preferences, and more lib- eral countries elect representatives from parties more supportive of gender equality. When the vote share of these parties is sufficiently high that they are able to form a governing ma- jority, countries experience large increases in the gender equality of the legal framework. The results highlight that legalized gender inequality has deep roots. The contemporary social norms that support regressive laws are linked to long-run cultural traits and therefore exhibit a high degree of persistence. These norms are expressed politically through represen- tative government, and so the resulting laws that these governments pass should be seen as a reflection of underlying preferences as mediated through the political system. Simply chang- ing laws in a political context of conservative norms will result in weak implementation, at best, and social backlash at worst. Despite this, an emerging literature shows that it is possible to shift social norms over a relatively short period of time, particularly when personal beliefs are more permissive than social expectations. Our research shows that in addition to their direct effect on economic inequality, such norm shifts will filter through the political system, and ultimately result in better legal and economic outcomes for women. Still, more research is needed on how to durably change social norms toward gender equality. 36 References Acemoglu, D. and Jackson, M. O. (2017). Social norms and the enforcement of laws. Journal of the European Economic Association, 15(2):245–295. Adnan, W. and Miaari, S. H. (2018). Voting patterns and the gender wage gap. Journal of Economic Behavior Organization, 146:222–247. Alesina, A., Giuliano, P., and Nunn, N. (2013). On the Origins of Gender Roles: Women and the Plough. Quarterly Journal of Economics, 128(2):469–530. Bahrami-Rad, D. (2021). Keeping it in the family: Female inheritance, inmarriage, and the status of women. Journal of Development Economics, 153:102714. Baskaran, T., Bhalotra, S., Min, B., and Uppal, Y. (2024). Women legislators and economic performance. Journal of Economic Growth, 29(2):151–214. Beaman, L., Chattopadhyay, R., Duflo, E., Pande, R., and Topalova, P. (2009). Powerful Women: Does Exposure Reduce Bias? Quarterly Journal of Economics, 124(4):1497–1540. Beaman, L., Duflo, E., Pande, R., and Topalova, P. (2012). Female leadership raises aspirations and educational attainment for girls: A policy experiment in india. science, 335(6068):582– 586. Becker, G. S. (1971). The economics of discrimination. University of Chicago Press. Becker, G. S. (1981). A treatise on the family. Harvard University Press. Benabou, R. and Tirole, J. (2025). Laws and Norms. Technical Report w17579, National Bureau of Economic Research, Cambridge, MA. Bicchieri, C. (2006). The Grammar of Society: The Nature and Dynamics of Social Norms. Cam- bridge University Press. Google-Books-ID: 4N1FDIZvcI8C. Bicchieri, C. (2016). Norms in the Wild: How to Diagnose, Measure, and Change Social Norms. Oxford University Press. Google-Books-ID: 7PuSDQAAQBAJ. Blau, F. D. and Kahn, L. M. (2017). The gender wage gap: Extent, trends, and explanations. Journal of economic literature, 55(3):789–865. 37 Bursztyn, L., González, A. L., and Yanagizawa-Drott, D. (2020). Misperceived Social Norms: Women Working Outside the Home in Saudi Arabia. American Economic Review, 110(10):2997–3029. Bussolo, M., Ezebuihe, J. A., Munoz Boudet, A. M., Poupakis, S., Rahman, T., and Sarma, N. (2024). Social Norms and Gender Disparities with a Focus on Female Labor Force Partici- pation in South Asia. The World Bank Research Observer, 39(1):124–158. Chai, Y., Ríos-Salas, V., Stek, P., and Heymann, J. (2022). Does Enhancing Paid Maternity Leave Policy Help Promote Gender Equality? Evidence from 31 Low- and Middle-Income Countries. Gender Issues, 39(3):335–367. Cinelli, C., Forney, A., and Pearl, J. (2024). A crash course in good and bad controls. Sociological Methods & Research, 53(3):1071–1104. Collins, E., Contreras, I., Henderson, S., Lamke, C., Rosenbaum, M., and Warren, S. S. (2024). Can remote survey methods yield nationally representative samples in lmics? a cross- national analysis of pandemic-era studies. Technical report, Innovations for Poverty Action, Washington, DC. Working paper. Dhar, D., Jain, T., and Jayachandran, S. (2022). Reshaping adolescents’ gender attitudes: Evi- dence from a school-based experiment in india. American economic review, 112(3):899–927. Doepke, M., Tertilt, M., and Voena, A. (2012). The economics and politics of women’s rights. Annu. Rev. Econ., 4(1):339–372. Downs, A. (1957). An economic theory of democracy. Harper Collins, New York. Duflo, E. (2012). Women empowerment and economic development. Journal of Economic Literature, 50(4):1051–79. Fernandez, R. and Fogli, A. (2009). Culture: An Empirical Investigation of Beliefs, Work, and Fertility. American Economic Journal: Macroeconomics, 1(1):146–177. Fernández, R., Parsa, S., and Viarengo, M. (2024). Coming out in america: thirty years of cultural change. The Journal of Law, Economics, and Organization, page ewae010. 38 Field, E., Pande, R., Rigol, N., Schaner, S., and Troyer Moore, C. (2021). On Her Own Account: How Strengthening Women’s Financial Control Impacts Labor Supply and Gender Norms. American Economic Review, 111(7):2342–2375. Fortin, N. M. (2005). Gender role attitudes and the labour-market outcomes of women across oecd countries. oxford review of Economic Policy, 21(3):416–438. Giuliano, P. and Nunn, N. (2018). Ancestral characteristics of modern populations. Economic History of Developing Regions, 33(1):1–17. Goldin, C. (1994). The U-Shaped Female Labor Force Function in Economic Development and Economic History. Technical Report w4707, National Bureau of Economic Research, Cambridge, MA. Goldin, C. and Katz, L. F. (2002). The power of the pill: Oral contraceptives and women’s career and marriage decisions. Journal of political Economy, 110(4):730–770. Gonzales, C., Jain-Chandra, S., Kochhar, K., and Newiak, M. (2015). Fair Play: More Equal Laws Boost Female Labor Force Participation. Grosjean, P. and Khattar, R. (2018). It’s raining men! hallelujah? the long-run consequences of male-biased sex ratios. The Review of Economic Studies, 86(2):723–754. Gulesci, S., Leone, M., and Zafar, S. (2024). Domestic Violence Laws and Social Norms: Evi- dence from Pakistan. Trinity Economics Papers. Hallward-Driemeier, M. and Gajigo, O. (2015). Strengthening Economic Rights and Women’s Occupational Choice: The Impact of Reforming Ethiopia’s Family Law. World Development, 70:260–273. Heath, R. and Tan, X. (2020). Intrahousehold Bargaining, Female Autonomy, and Labor Supply: Theory and Evidence from India. Journal of the European Economic Association, 18(4):1928–1968. Hossain, D. and Islam, R. (2024). A systematic literature review on the relationship between education and gender ideology. Discover Global Society, 2(1):39. 39 Htun, M., Jensenius, F. R., and Nelson-Nuñez, J. (2019). Gender-discriminatory laws and women’s economic agency. Social Politics: International Studies in Gender, State Society, 26(2):193–222. Hyland, M., Djankov, S., and Goldberg, P. K. (2020a). Gendered laws and women in the workforce. American Economic Review: Insights, 2(4):475–490. Hyland, M., Islam, A., and Muzi, S. (2020b). Discriminatory Environment, Firms’ Discriminatory Behavior, and Women’s Employment in the Democratic Republic of Congo. Number 9224 in Policy Research Working Paper. World Bank, Washington, DC. Jayachandran, S. (2021). Social Norms as a Barrier to Women’s Employment in Developing Countries. IMF Economic Review, 69(3):576–595. Kleibergen, F. and Paap, R. (2006). Generalized reduced rank tests using the singular value decomposition. Journal of Econometrics, 133(1):97–126. Lane, T., Nosenzo, D., and Sonderegger, S. (2023). Law and Norms: Empirical Evidence. American Economic Review, 113(5):1255–1293. Lee, D. S., McCrary, J., Moreira, M. J., and Porter, J. (2022). Valid t-ratio inference for iv. American Economic Review, 112(10):3260–90. Lowes, S. (2022a). Culture in historical political economy. Working Paper 30511, National Bureau of Economic Research. Lowes, S. (2022b). Kinship structure and the family: Evidence from the matrilineal belt. Work- ing Paper 30509, National Bureau of Economic Research. Mackie, G., Moneti, F., Shakya, H., and Denny, E. (2015). What are social norms? How are they measured? University of California at San Diego-UNICEF Working Paper. Maida, A. and Weber, A. (2022). Female leadership and gender gap within firms: Evidence from an italian board reform. Ilr Review, 75(2):488–515. Meltzer, A. H. and Richard, S. F. (1981). A rational theory of the size of government. Journal of Political Economy, 89(5):914–927. 40 Mercer, A., Lau, A., and Kennedy, C. (2018). For weighting online opt-in samples, what matters most? Technical report, Pew Research Center, Washington, DC. Miller, M. K. (2015). Elections, information, and policy responsiveness in autocratic regimes. Comparative Political Studies, 48(6):691–727. Murdock, G. P. (1967). Ethnographic atlas: A summary. Ethnology, 6(2):109–236. Naaraayanan, L. (2019). Women’s Inheritance Rights and Entrepreneurship Gender Gap. Norris, P. (2020). Measuring populism worldwide. Party Politics, 26(6):697–717. Priyanka, S. (2020). Do female politicians matter for female labor market outcomes? evidence from state legislative elections in india. Labour Economics, 64:101822. European Association of Labour Economists, 31st annual conference, Uppsala Sweden, 19-21 September 2019. Rexer, J. M. (2022). The brides of boko haram: Economic shocks, marriage practices, and insurgency in nigeria. The Economic Journal, 132(645):1927–1977. Rosenzweig, L. R., Bergquist, P., Hoffmann Pham, K., Rampazzo, F., and Mildenberger, M. (2025). Survey sampling in the global south using facebook advertisements. Political Science Research and Methods, pages 1–17. FirstView article. Sapkal, R. S. (2017). From mother to daughter: Does equal inheritance property laws reform improve female labor supply and educational attainments in india? Asian Journal of Law and Economics, 8(1):20150028. Scartascini, C., Cruz, C., and Keefer, P. (2021). The database of political institutions 2020 (dpi2020). Stevenson, B. (2008). Divorce Law and Women’s Labor Supply. Journal of Empirical Legal Studies, 5(4):853–873. Stock, J. H. and Yogo, M. (2005). Testing for Weak Instruments in Linear IV Regression, page 80–108. Cambridge University Press. Van der Gaag, N., Heilman, B., Gupta, T., Nembhard, C., and Barker, G. (2019). State of the World’s Fathers: Unlocking the Power of Men’s Care, 2019. Promundo-US. 41 ONLINE APPENDIX Appendix Tables Table A1: Summary statistics Mean SD Min Max p25 p50 p75 N Personal beliefs in breadwinner norm 0.383 0.174 0.086 0.757 0.239 0.367 0.525 120 Personal beliefs in homemaker norm 0.518 0.199 0.130 0.827 0.317 0.541 0.695 120 Normative expectations in breadwinner norm 0.567 0.077 0.282 0.715 0.514 0.582 0.613 120 Normative expectations in homemaker norm 0.635 0.075 0.358 0.768 0.595 0.648 0.698 120 Labor force participation, female (ILO 15-64 years)(2020) 56.358 17.141 11.374 83.626 48.334 59.490 69.189 119 WBL legal index 77.292 17.625 26.250 100.000 68.750 80.938 90.938 120 Per capita GDP (log) (2020) 8.686 1.331 6.073 11.346 7.691 8.583 9.757 119 Per capita GDP (log squared) (2020) 77.209 23.421 128.735 73.676 36.877 59.154 95.209 119 Summary statistics 42 Table A2: Female labor force participation, laws and norms Outcome Female labor force participation rate (1) (2) (3) (4) (5) (6) (7) (8) Panel A: Homemaker norm WBL legal index 0.594*** 0.594*** 0.567*** 0.457*** 0.482*** 0.543*** (0.080) (0.109) (0.100) (0.103) (0.095) (0.104) Personal beliefs -36.659*** 0.050 4.051 16.178 (6.626) (7.920) (10.201) (11.214) Social expectations -117.566*** -53.059** -34.610 -58.150 (15.196) (18.056) (26.406) (31.134) Per capita GDP (log) -42.741** -39.660** -33.906* (13.441) (13.429) (15.687) Per capita GDP (log square) 2.525*** 2.274** 1.964* (0.737) (0.773) (0.890) Observations 119 119 119 119 119 119 119 119 R2 0.374 0.182 0.268 0.374 0.454 0.408 0.462 0.470 Panel B: Breadwinner norm WBL legal index 0.594*** 0.501*** 0.496*** 0.369** 0.394*** 0.405*** (0.080) (0.109) (0.105) (0.112) (0.105) (0.114) Personal beliefs -49.068*** -13.105 -7.800 3.049 (7.549) (9.181) (12.012) (11.941) Social expectations -132.976*** -74.439*** -63.356* -66.149* (16.277) (21.057) (28.495) (30.454) Per capita GDP (log) -43.605*** -37.439** -37.139** (12.823) (13.171) (13.757) Per capita GDP (log square) 2.537*** 2.119** 2.107** (0.706) (0.756) (0.779) Observations 119 119 119 119 119 119 119 119 R2 0.374 0.251 0.357 0.382 0.456 0.432 0.481 0.481 Note: Robust standard errors in parentheses. All variables are measured in 2020. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, ranging from 0 to 100. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is 120 countries for which social norms data are available. *** p < 0.01, ** p < 0.05, * p < 0.1. 43 Table A3: Norms and outcomes: OLS correlations, robustness to controls Variable Homemaker norm Breadwinner norm (1) (2) (3) (4) (5) (6) (7) (8) Panel A: Female labor force participation Social expectations -1.004*** -0.446* -1.311*** -0.626*** (0.255) (0.233) (0.237) (0.217) Personal beliefs -0.334*** -0.164 -0.440*** -0.232** (0.090) (0.107) (0.091) (0.098) Per capita GDP (log) (2020) -29.691* -29.071* -42.970** -34.913** -26.102 -30.294** -36.667** -34.599** (17.274) (15.645) (16.483) (14.445) (15.771) (14.582) (16.439) (14.694) Per capita GDP (log squared) (2020) 1.973** 1.853** 2.757*** 2.205*** 1.716* 1.896** 2.329** 2.146** (0.981) (0.900) (0.916) (0.823) (0.904) (0.845) (0.914) (0.835) Region FE No Yes No Yes No Yes No Yes Observations 117 118 117 118 117 118 117 118 R2 0.429 0.628 0.424 0.625 0.512 0.648 0.455 0.634 Panel B: Legal index Social expectations -1.070*** -0.564*** -1.275*** -0.894*** (0.224) (0.203) (0.201) (0.181) Personal beliefs -0.534*** -0.445*** -0.585*** -0.448*** (0.072) (0.070) (0.077) (0.088) Per capita GDP (log) (2020) -3.720 15.615 -18.437 5.838 -1.367 14.555 -9.774 7.730 (13.526) (13.469) (12.987) (12.829) (11.371) (11.923) (13.171) (13.312) Per capita GDP (log squared) (2020) 0.367 -1.013 1.140 -0.462 0.201 -0.997 0.603 -0.634 (0.762) (0.752) (0.712) (0.715) (0.643) (0.657) (0.726) (0.740) Region FE No Yes No Yes No Yes No Yes Observations 117 118 117 118 117 118 117 118 R2 0.575 0.662 0.663 0.713 0.638 0.708 0.664 0.713 Panel C: Implementation gap Social expectations 0.004 0.003 0.004** 0.003 (0.003) (0.003) (0.002) (0.002) Personal beliefs 0.003*** 0.002** 0.004*** 0.003*** (0.001) (0.001) (0.001) (0.001) Per capita GDP (log) (2020) 0.038 -0.006 0.098 0.047 0.035 0.009 0.044 0.038 (0.115) (0.144) (0.112) (0.141) (0.112) (0.143) (0.112) (0.142) Per capita GDP (log squared) (2020) -0.008 -0.002 -0.010 -0.005 -0.008 -0.002 -0.007 -0.004 (0.007) (0.008) (0.006) (0.008) (0.006) (0.008) (0.006) (0.008) Region FE No Yes No Yes No Yes No Yes Observations 117 118 117 118 117 118 117 118 R2 0.655 0.665 0.686 0.676 0.659 0.666 0.696 0.678 Note: Robust standard errors in parentheses. All variables are measured in 2020. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, ranging from 0 to 100. Implementation gap measures the percentage-point gap between the WBL implementation score and full implementation of existing gender-equal laws, ranging from 0 to 1. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is 120 countries for which social norms data are available. Controls are the date at which cultural data is recorded, latitude, distance to coast, average ruggedness, share of population living in tropical climate, and share of population traditionally practicing intensive agriculture. *** p < 0.01, ** p < 0.05, * p < 0.1. 44 Table A4: Norms and outcomes: OLS correlations, robustness World Values Survey data (1) (2) (3) (4) Panel A: Female labor force participation Men have more right to work -55.793*** -47.485*** (4.218) (8.210) WVS gender index -72.642*** -54.342*** (6.748) (11.714) Per capita GDP (log) -23.488 -21.518 (13.478) (13.671) Per capita GDP (log squared) 1.425 1.414 (0.735) (0.736) Observations 287 284 288 285 R2 0.418 0.452 0.320 0.388 Panel B: Legal index Men have more right to work -58.628*** -56.165*** (3.507) (7.143) WVS gender index -81.851*** -72.614*** (5.697) (10.255) Per capita GDP (log) 7.040 11.402 (10.888) (11.774) Per capita GDP (log squared) -0.334 -0.486 (0.597) (0.644) Observations 290 283 291 284 R2 0.557 0.571 0.490 0.525 Panel C: Implementation gap Men have more right to work 0.549*** 0.338*** (0.045) (0.082) WVS gender index 0.717*** 0.405*** (0.072) (0.115) Per capita GDP (log) -0.016 -0.045 (0.145) (0.161) Per capita GDP (log squared) -0.003 -0.002 (0.008) (0.009) Observations 239 235 240 236 R2 0.392 0.537 0.296 0.508 Note: Standard errors clustered at the country level in parentheses. Female labor force participa- tion is measured as the share of working-age women (15-64) in the labor force. “Men have more right to work” captures the share of men agreeing with this statement in a given WVS round. WVS gender index averages the agreement shares across several questions relating to gender attitudes: i ) when jobs are scarce, men have more right to work, ii ) being a housewife is fulfilling, iii ) univer- sity is more important for a boy than a girl, and iv) men make better political leaders than women. Sample is 290 country-year observations for which the right-to-work question is asked and legal index data is available. *** p < 0.01, ** p < 0.05, * p < 0.1. 45 Table A5: Patriarchal culture and contemporary social norms: first stage, robustness to con- trols Variable Homemaker norm Breadwinner norm Type Expectations Beliefs Expectations Beliefs (1) (2) (3) (4) (5) (6) (7) (8) Patriarchy index 0.084*** 0.060** 0.340*** 0.147*** 0.065** 0.062** 0.262*** 0.090 (0.024) (0.025) (0.067) (0.052) (0.029) (0.027) (0.063) (0.054) Per capita GDP (log) (2020) 10.368** 6.446 -10.442 -16.971 10.663** 1.776 6.314 -9.709 (4.508) (4.847) (10.663) (11.335) (5.254) (5.266) (10.225) (11.972) Per capita GDP (log squared) (2020) -0.756*** -0.448 0.220 0.866 -0.773** -0.197 -0.798 0.301 (0.267) (0.287) (0.612) (0.622) (0.301) (0.308) (0.572) (0.651) Region FE No Yes No Yes No Yes No Yes Observations 117 117 117 117 117 117 117 117 R2 0.680 0.705 0.658 0.756 0.588 0.616 0.593 0.678 Note: Robust standard errors in parentheses. Patriarchy index is measured as the average across four ancestral culture patriarchal characteristics, varying from 0 to 1. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is 120 countries for which social norms data are available. Controls are the date at which cultural data is recorded, latitude, distance to coast, average ruggedness, share of population living in tropical climate, and share of population traditionally practicing intensive agriculture. *** p < 0.01, ** p < 0.05, * p < 0.1. Table A6: Patriarchal culture and legal/economic outcomes: reduced form, robustness to controls Outcome FLFP Legal index Implementation gap (1) (2) (3) (4) (5) (6) Patriarchy index -0.109 0.030 -0.184*** -0.111** 0.002** 0.001 (0.075) (0.064) (0.062) (0.052) (0.001) (0.001) Per capita GDP (log) (2020) -39.573** -34.777** -12.817 15.495 0.055 -0.030 (16.708) (15.349) (14.135) (13.826) (0.116) (0.149) Per capita GDP (log squared) (2020) 2.690*** 2.224** 1.019 -0.976 -0.009 -0.000 (0.937) (0.873) (0.794) (0.775) (0.007) (0.008) Region FE No Yes No Yes No Yes Observations 117 117 117 117 117 117 R2 0.369 0.616 0.530 0.655 0.664 0.668 Note: Robust standard errors in parentheses. All outcome variables are measured in 2020. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, ranging from 0 to 100. Implementation gap measures the percentage-point gap between the WBL implementation score and full implementation of existing gender-equal laws, ranging from 0 to 1. Patriarchy index is measured as the average across four ancestral culture patriarchal characteristics, varying from 0 to 1. Controls are the date at which cultural data is recorded, latitude, distance to coast, average ruggedness, share of population living in tropical climate, and share of population traditionally practicing intensive agriculture. Sample is 120 countries for which social norms data are available. *** p < 0.01, ** p < 0.05, * p < 0.1. 46 Table A7: Norms and legal/economic outcomes: 2SLS, robustness to controls Outcome FLFP Legal index Implementation gap (1) (2) (3) (4) (5) (6) Panel A: Homemaker norm Social expectations -1.299* 0.506 -2.181*** -1.849* 0.018** 0.018 (0.739) (1.083) (0.690) (1.042) (0.007) (0.012) Region FE No Yes No Yes No Yes First-stage F-statistic 12.023 5.948 12.023 5.948 12.023 5.948 Observations 117 117 117 117 117 117 R2 0.423 0.571 0.488 0.560 0.557 0.563 Panel B: Breadwinner norm Social expectations -1.672* 0.492 -2.808*** -1.797** 0.023** 0.018 (0.853) (1.097) (0.988) (0.872) (0.012) (0.012) Region FE No Yes No Yes No Yes First-stage F-statistic 5.227 5.293 5.227 5.293 5.227 5.293 Observations 117 117 117 117 117 117 R2 0.500 0.543 0.430 0.639 0.433 0.537 Note: Robust standard errors in parentheses. All outcome variables are measured in 2020. Estimates are from a 2SLS model where social expectations is instrumented with the patriarchal culture index. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, ranging from 0 to 100. Implementation gap measures the percentage-point gap between the WBL implementation score and full implementation of existing gender-equal laws, ranging from 0 to 1. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Patriarchal culture index is measured as the average across four ancestral culture patriarchal characteristics, varying from 0 to 1. Sample is 120 countries for which social norms data are available. Controls are the date at which cultural data is recorded, latitude, distance to coast, average ruggedness, share of population living in tropical climate, and share of population traditionally practicing intensive agriculture. *** p < 0.01, ** p < 0.05, * p < 0.1. 47 Table A8: Norms and legal/economic outcomes: 2SLS (personal beliefs) Outcome FLFP Legal index Implementation gap (1) (2) (3) (4) (5) (6) Panel A: Homemaker norm Personal beliefs -0.505*** -0.506*** -0.714*** -0.851*** 0.008*** 0.007*** (0.111) (0.141) (0.089) (0.137) (0.001) (0.002) First-stage F-statistic 104.515 36.024 88.911 36.024 88.911 36.024 Observations 117 117 118 117 118 117 R2 0.146 0.257 0.482 0.491 0.455 0.519 Panel B: Breadwinner norm Personal beliefs -0.654*** -0.704*** -0.913*** -1.185*** 0.010*** 0.009*** (0.136) (0.194) (0.114) (0.190) (0.002) (0.002) First-stage F-statistic 71.241 23.387 68.834 23.387 68.834 23.387 Observations 117 117 118 117 118 117 R2 0.206 0.269 0.452 0.377 0.395 0.455 Note: Robust standard errors in parentheses. All outcome variables are measured in 2020. Estimates are from a 2SLS model where social expectations is instrumented with the patriarchal culture index. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, ranging from 0 to 100. Implementation gap measures the percentage-point gap between the WBL implementation score and full implementation of existing gender-equal laws, ranging from 0 to 1. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Patriarchal culture index is measured as the average across four ancestral culture patriarchal characteristics, varying from 0 to 1. Sample is 120 countries for which social norms data are available. *** p < 0.01, ** p < 0.05, * p < 0.1. 48 Table A9: Norms and legal/economic outcomes: 2SLS, robustness to missing data Outcome FLFP Legal index Implementation gap (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A: Homemaker norm Social expectations -1.924*** -1.966*** -2.059*** -3.835*** -3.621*** -3.399*** 0.028*** 0.028*** 0.026*** (0.622) (0.620) (0.578) (0.738) (0.697) (0.602) (0.006) (0.006) (0.005) First-stage F-statistic 15.379 18.454 24.902 15.379 18.454 24.902 15.379 18.454 24.902 Observations 112 117 117 112 117 117 112 117 117 R2 0.221 0.233 0.217 -0.095 -0.002 0.079 0.268 0.302 0.339 Panel B: Breadwinner norm Social expectations -1.677*** -1.794*** -2.011*** -3.342*** -3.304*** -3.320*** 0.025*** 0.025*** 0.025*** (0.471) (0.501) (0.530) (0.495) (0.510) (0.495) (0.005) (0.005) (0.005) First-stage F-statistic 21.115 22.604 24.939 21.115 22.604 24.939 21.115 22.604 24.939 Observations 112 117 117 112 117 117 112 117 117 R2 0.366 0.364 0.333 0.199 0.215 0.209 0.312 0.316 0.313 Threshold (%) 0 50 100 0 50 100 0 50 100 Note: Robust standard errors in parentheses. All outcome variables are measured in 2020. Estimates are from a 2SLS model where social expectations is instrumented with the patriarchal culture index. Female labor force participation is measured as the share of working-age women (15-64) in the labor force. Legal index is the WBL legal score, ranging from 0 to 100. Implementation gap measures the percentage-point gap between the WBL implementation score and full implementation of existing gender-equal laws, ranging from 0 to 1. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Patriarchal culture index is measured as the average across four ancestral culture patriarchal characteristics, varying from 0 to 1. Threshold refers to the inclusion criterion for calculating ancestral patriarchal culture, wherein a variable is only included in the index if it is nonmissing for greater than x share of the population, according to Giuliano and Nunn (2018). *** p < 0.01, ** p < 0.05, * p < 0.1. 49 Table A10: Norms and politics: robustness to threshold Outcome Seat share In government (1) (2) (3) (4) Panel A: 75% threshold Homemaker social expectations -1.356* -0.028** (0.700) (0.011) Homemaker personal beliefs 0.393 0.004 (0.358) (0.005) Breadwinner social expectations -0.463 -0.023** (0.786) (0.011) Breadwinner personal beliefs 0.085 0.004 (0.450) (0.006) Observations 54 54 54 54 R2 0.043 0.007 0.129 0.093 Panel B: 25% threshold Homemaker social expectations -1.200*** -0.029*** (0.436) (0.008) Homemaker personal beliefs 0.374* 0.007* (0.209) (0.003) Breadwinner social expectations -0.574 -0.024*** (0.497) (0.009) Breadwinner personal beliefs 0.189 0.007* (0.235) (0.004) Observations 92 92 92 92 R2 0.051 0.012 0.110 0.075 Note: Robust standard errors in parentheses. Seat share is the share of parliamentary seats be- longing to political parties expressing an ideology in favor of gender equality. In government is an indicator variable for whether any of these pro-equality parties is a part of the ruling coalition. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is all countries with political leadership data in 2022 for which 75% (Panel A) or 25% (Panel B) or more of the parliamentary seats can be matched to a political gender ideology. All variables are measured in 2020. *** p < 0.01, ** p < 0.05, * p < 0.1. 50 Table A11: Norms and politics (25% threshold): robustness to income and region Outcome Seat share In government (1) (2) (3) (4) (5) (6) Panel A: Homemaker norm Social expectations -0.765 -1.314** -1.259** -0.024*** -0.029*** -0.031*** (0.555) (0.563) (0.578) (0.009) (0.010) (0.010) Personal beliefs 0.476* 0.688* 0.690* 0.008* 0.011** 0.011** (0.243) (0.362) (0.377) (0.004) (0.005) (0.005) GDP per capita (log) (constant 2015 US$) 4.964 1.649 0.053 -0.057 (3.891) (4.814) (0.055) (0.081) Region FE No Yes Yes No Yes Yes Observations 91 92 91 91 92 91 R2 0.073 0.139 0.137 0.118 0.198 0.200 Panel B: Breadwinner norm Social expectations -0.055 -0.420 -0.334 -0.017 -0.021** -0.021* (0.611) (0.634) (0.640) (0.010) (0.010) (0.011) Personal beliefs 0.364 0.392 0.426 0.010** 0.013** 0.012** (0.279) (0.268) (0.280) (0.005) (0.005) (0.005) GDP per capita (log) (constant 2015 US$) 6.188* 3.458 0.094* -0.009 (3.695) (5.081) (0.054) (0.086) Region FE No Yes Yes No Yes Yes Observations 91 92 91 91 92 91 R2 0.051 0.087 0.090 0.105 0.178 0.175 Note: Robust standard errors in parentheses. Seat share is the share of parliamentary seats belonging to political parties express- ing an ideology in favor of gender equality. In government is an indicator variable for whether any of these pro-equality parties is a part of the ruling coalition. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is all countries with political leadership data in 2022 for which 75% or more of the parliamentary seats can be matched to a political gender ideology. All variables are measured in 2020. *** p < 0.01, ** p < 0.05, * p < 0.1.*** p < 0.01, ** p < 0.05, * p < 0.1. 51 Table A12: Norms and politics (75% threshold): 2015-2020 Outcome Seat share In government (1) (2) (3) (4) (5) (6) Panel A: Homemaker norm Social expectations -0.456 -1.684** -1.629** -0.005 -0.018* -0.019* (0.893) (0.670) (0.775) (0.011) (0.010) (0.011) Personal beliefs 0.437 1.071** 1.086** 0.004 0.012* 0.012* (0.367) (0.509) (0.529) (0.005) (0.006) (0.006) GDP per capita (log) (constant 2015 US$) 5.831 0.963 0.070 -0.010 (5.637) (6.289) (0.071) (0.085) Region FE No Yes Yes No Yes Yes Observations 323 329 323 323 329 323 R2 0.059 0.295 0.291 0.042 0.209 0.205 Panel B: Breadwinner norm Social expectations 0.255 -1.213* -1.135* -0.007 -0.026** -0.026** (0.855) (0.627) (0.653) (0.013) (0.010) (0.010) Personal beliefs 0.274 0.422 0.466 0.004 0.009 0.009 (0.481) (0.455) (0.455) (0.007) (0.006) (0.006) GDP per capita (log) (constant 2015 US$) 6.517 3.478 0.062 0.009 (5.346) (6.311) (0.070) (0.090) Region FE No Yes Yes No Yes Yes Observations 323 329 323 323 329 323 R2 0.042 0.217 0.216 0.040 0.203 0.198 Note: Standard errors in parentheses are clustered at the country level. Seat share is the share of parliamentary seats belong- ing to political parties expressing an ideology in favor of gender equality. In government is an indicator variable for whether any of these pro-equality parties is a part of the ruling coalition. Social expectations are measured as the country-level av- erage belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is all countries with political leadership data in 2022 for which 75% or more of the parliamentary seats can be matched to a political gender ideology. Outcome variable is measured from 2015-2020. *** p < 0.01, ** p < 0.05, * p < 0.1. 52 Table A13: Politics and gender-equal laws (25% threshold) Outcome WBL index (1) (2) (3) Seat share (pro) 0.063*** (0.019) Seat share (anti) -0.021 (0.030) Seat share (neutral) 0.013 (0.017) Country FE Yes Yes Yes Year FE Yes Yes Yes Observations 4231 4231 4231 R2 0.901 0.900 0.900 Note: Standard errors in parentheses clustered at the coun- try level. Outcome variable is the WBL legal score, ranging from 0 to 100. Seat share is the share of parliamentary seats belonging to political parties expressing a given gender ideol- ogy. Sample is all country-years with political leadership data in 2022 for which 25% or more of the parliamentary seats can be matched to a political gender ideology. *** p < 0.01, ** p < 0.05, * p < 0.1. 53 Table A14: Politics and gender-equal laws: robustness to region and income Outcome WBL index (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A: 75% threshold Seat share (pro) 0.061*** 0.081*** 0.076*** (0.018) (0.021) (0.020) Seat share (anti) -0.087** -0.098*** -0.091** (0.037) (0.036) (0.037) Seat share (neutral) -0.033 -0.044 -0.050 (0.030) (0.031) (0.032) GDP per capita (log) (constant 2015 US$) 2.954 2.624 3.068 2.345 1.977 2.592 (3.308) (3.207) (3.379) (3.696) (3.550) (3.761) Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes No No No No No No Year × Region FE No No No Yes Yes Yes Yes Yes Yes Observations 2231 2231 2231 2534 2534 2534 2192 2192 2192 R2 0.919 0.919 0.918 0.927 0.927 0.926 0.926 0.926 0.925 Panel B: 25% threshold Seat share (pro) 0.055*** 0.062*** 0.061*** (0.016) (0.018) (0.015) Seat share (anti) -0.013 -0.021 -0.015 (0.028) (0.029) (0.027) Seat share (neutral) -0.000 0.000 -0.012 (0.017) (0.017) (0.017) GDP per capita (log) (constant 2015 US) 3.125 3.173 3.187 4.129 4.145 4.249 (2.717) (2.732) (2.745) (3.315) (3.311) (3.361) Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes No No No No No No Year × Region FE No No No Yes Yes Yes Yes Yes Yes Observations 3727 3727 3727 4209 4209 4209 3714 3714 3714 R2 0.908 0.907 0.907 0.909 0.908 0.908 0.915 0.914 0.914 Note: Standard errors in parentheses clustered at the country level. Outcome variable is the WBL legal score, ranging from 0 to 100. Seat share is the share of parliamentary seats belonging to political parties expressing a given gender ideology. Sample is all country-years with political leadership data in 2022 for which 75% (Panel A) or 25% (Panel B) or more of the parliamentary seats can be matched to a political gender ideology. *** p < 0.01, ** p < 0.05, * p < 0.1. 54 Table A15: Politics and gender-equal laws: robustness to left-wing parties Outcome WBL index Party orientation Pro Anti Neutral (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A: 75% threshold Share of seats 0.072*** 0.064*** 0.057*** -0.104*** -0.098*** -0.075** -0.028 -0.029 -0.037 (0.024) (0.021) (0.020) (0.037) (0.036) (0.032) (0.028) (0.028) (0.027) Left-wing party in government 0.360 0.241 0.645 (0.768) (0.754) (0.739) Left-wing executive 0.408 0.187 0.623 (0.799) (0.753) (0.746) Left-wing seat share 4.781* 4.238* 5.612** (2.569) (2.512) (2.542) Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 2571 2555 2565 2571 2555 2565 2571 2555 2565 R2 0.916 0.917 0.918 0.916 0.918 0.918 0.914 0.916 0.917 Panel B: 25% threshold Share of seats 0.062*** 0.056*** 0.053*** -0.017 -0.021 -0.004 0.013 0.013 0.005 (0.020) (0.018) (0.018) (0.031) (0.030) (0.027) (0.017) (0.017) (0.017) Left-wing party in government 1.110* 1.222* 1.290** (0.621) (0.635) (0.608) Left-wing executive 1.204* 1.255* 1.344** (0.618) (0.637) (0.606) Left-wing seat share 5.102*** 5.477*** 5.464*** (1.706) (1.681) (1.731) Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 4209 4158 4202 4209 4158 4202 4209 4158 4202 R2 0.901 0.903 0.903 0.900 0.902 0.902 0.900 0.902 0.902 Note: Standard errors in parentheses clustered at the country level. Outcome variable is the WBL legal score, ranging from 0 to 100. Seat share is the share of parliamentary seats belonging to political parties expressing a given gender ideology. Left wing party variables are indicators for left-wing political control in various branches of national government, according to Scartascini et al. (2021). Sample is all country-years with political leadership data in 2022 for which 75% (Panel A) or 25% (Panel B) or more of the parliamentary seats can be matched to a political gender ideology. *** p < 0.01, ** p < 0.05, * p < 0.1. 55 Appendix Figures Figure A1: Gender ideology among matched parties .1 .08 .06 Fraction .04 .02 0 0 2 4 6 8 10 Issues: Party favors (0) or opposes (10) women's rights Note: Figure shows party-level distribution of gender ideology from Norris (2020) among sample of 659 parties that have information on gender ideology and can be matched to a party in the Dataset of Political Institutions. Parties rank ideology on a scale of 0 to 10, with 10 indicating more opposition to women’s rights. Red lines indicate cutoff points for pro, neutral, and anti-women categories. 56 Figure A2: Match rates between DPI and GPS across countries and over time (a) Match rates over time (b) Distribution of match rates in 2020 .2 Average share of parliamentary seats matched 60 50 .15 40 Fraction .1 30 20 .05 10 0 1975 1985 1995 2005 2015 0 20 40 60 80 100 Year Share of parliamentary seats matched, 2020 Note: Figure shows the annual cross-country average share of matched parliamentary seats between DPI and GPS over time (left) and the distribution of this country-level match rate in 2020 (right). Figure A3: OLS correlations: social expectations, personal beliefs, and female labor force participation 100 100 Female labor force participation rate (2020) Female labor force participation rate (2020) 80 80 60 60 40 40 20 20 0 0 0 20 40 60 80 20 40 60 80 Breadwinner norm Homemaker norm Social expectations Personal beliefs Social expectations Personal beliefs Note: Figure shows the unconditional country-level correlations between social norms (blue), per- sonal beliefs (red), and female labor force participation, all measured in 2020. Social expectations are measured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is 120 countries for which social norms data are available. 57 Figure A4: OLS correlations: social expectations, personal beliefs, and implementation gaps .8 .8 WBL implementation gap .6 WBL implementation gap .6 .4 .4 .2 .2 0 0 0 20 40 60 80 20 40 60 80 Breadwinner norm Homemaker norm Social expectations Personal beliefs Social expectations Personal beliefs Note: Figure shows the unconditional country-level correlations between social norms (blue), per- sonal beliefs (red), and the implementation gap, all measured in 2020. Social expectations are mea- sured as the country-level average belief about the share of peers that hold a particular gender norm, ranging from 0 to 100. Personal beliefs measures the share of respondents who agree with a given gender norm, ranging from 0 to 100. Sample is 120 countries for which social norms data are avail- able. 58 Figure A5: Correlation between norms and political orientation: breadwinner norm 50 .6 Seat share of pro-women parties 40 Pro-woman party in govt .4 30 .2 20 10 0 0 -.2 45 50 55 60 65 45 50 55 60 65 Breadwinner social expectations Breadwinner social expectations Note: Figure shows the correlation between social expectations of gender norms and political out- comes, binning at 20 quantiles of the distribution of social norms and controlling for log GDP per capita, personal beliefs, and region fixed effects. Seat share is the share of parliamentary seats be- longing to political parties expressing an ideology in favor of gender equality. In government is an indicator variable for whether any of these pro-equality parties is a part of the ruling coalition. Social expectations are measured as the country-level average belief about the share of peers that agree with the male breadwinner norm, ranging from 0 to 100. Sample is 53 countries with political leadership data in 2022 for which 75% or more of the parliamentary seats can be matched to a polit- ical gender ideology. All variables are measured in 2020. 59 Figure A6: Residual variation in parliamentary seat shares: two-way fixed effects Seat share (anti) 100 75 50 25 Seat share (neutral) 100 Unmatched share < 75 50 25 Seat share (pro) 100 75 50 25 .8 .85 .9 .95 R-squared Note: Figure shows the estimated R2 values from a regression of each seat share outcome on country and year fixed effects. This regression is estimated separately for each unmatched share threshold from 25 to 100%. Unmatched share refers to the share of parliamentary seats in a given country-year that can be matched to gender ideology. 60 Figure A7: Impact of political regimes on gender-equal laws by matched seat share thresholds .15 .1 Impact on WBL score .05 0 -.05 -.1 -.15 25 50 75 100 Unmatched seat threshold Note: Figure shows coefficients from a regression of WBL score on the parliamentary seat share of pro-women parties controlling for country and year fixed effects. Each estimate is from a regression on the subsample of country-years for which the share of seats that cannot be matched to a gender ideology is less than the threshold. Standard errors are clustered at the country level. 61 Figure A8: Impact of political regimes on gender-equal laws: WBL sub-indices .4 Impact on WBL score .2 0 -.2 -.4 Pension Entrepreneurship Assets Parenthood Pay Mobility Marriage Workplace Note: Figure shows estimates from regressions of each WBL score sub-index on the parliamen- tary seat share of pro-women parties controlling for country and year fixed effects. Sample is all country-years for which the unmatched parliamentary seat share is less than 25%. Standard errors are clustered at the country level. 62