Data

To measure patriarchal attitudes, we use data from the Integrated Values Surveys (IVS), a combination of the European Value Survey and the World Value Survey (WVS), covering over 115 countries between 1981 and 2022 (Haerpfer et al., Reference Haerpfer, Inglehart, Moreno, Welzel, Kizilova, Diez-Medrano, Lagos, Norris, Ponarin and Puranen2021); however, we include survey responses through 2019 to avoid biases from the COVID-19 pandemic.

Following the approach in Davis and Williamson (Reference Davis and Williamson2019) and Fike (Reference Fike2024), we use three IVS survey questions to measure patriarchal attitudes across three areas of social life: work, politics, and education. The questions are: (1) When jobs are scarce, men should have more right to a job than women, (2) On the whole, men make better political leaders than women, and (3) A university education is more important for a boy than for a girl. Each question is coded as the share of respondents in a country who agree or strongly agree with each respective statement and aggregated to the country level. Thus, a higher score indicates a more traditional perception of the role of women in society.

The patriarchal attitudes index (PAI) is created from extracting the first principal component from the three IVS patriarchal attitudes questions, reflecting the common variance among the attitudes. For PAI, the eigenvalue of the first component is 2.64, explaining 88% of the total variation across the three questions. The component weights are similar and substantial, ranging from 0.56 to 0.59, confirming that all three variables contribute meaningfully to the first component. The Cronbach’s alpha for the PAI is 0.93, indicating high internal consistency of the standardized components. The index is standardized for ease of interpretation. The PAI negatively correlates with measures of women’s empowerment such as female labour force participation (–0.50), the Women Business and the Law (WBL) legal gender parity index (–0.79), and the Cingranelli–Richards women’s political rights index (–0.52).

To measure individualism, we follow several recent papers using data from the WVS/IVS (Beugelsdijk et al., Reference Beugelsdijk, Maseland and Van Hoorn2015; Davis, Reference Davis2016; Cline and Williamson, Reference Cline and Williamson2017; Davis and Williamson, Reference Davis and Williamson2016, Reference Davis and Williamson2018, Reference Davis and Williamson2019, Reference Davis and Williamson2020, Reference Davis and Williamson2022; Pitlik and Rode, Reference Pitlik and Rode2017; Kramer, Reference Kramer2023). This approach offers advantages over established national measures of individualism. Hofstede (Reference Hofstede1980, Reference Hofstede2001) and Schwartz’s (Reference Schwartz2006) measures are based on older data and may not capture current values (Beugelsdijk et al., Reference Beugelsdijk, Maseland and Van Hoorn2015). The IVS expands the sample size and increases the representation of less developed countries. Compared to Hofstede’s measure, using the IVS-based measure expands the sample from 62 to 93 countries, a 50% increase, and it almost doubles the number of low- and middle-income countries.

Our method for measuring individualism closely follows Beugelsdijk, Maseland, and van Hoorn (Reference Beugelsdijk, Maseland and Van Hoorn2015) (BMH), who replicate Hofstede’s individualism measure. BMH chose four WVS questions measuring preferences for private versus government business ownership, making one’s parents proud, and views on homosexuality and abortion. While we use the first three questions, we omit the abortion question due to its potential link with patriarchal attitudes. Instead, we include a measure of individual versus government responsibility (Davis, Reference Davis2016; Di Tella et al., Reference Di Tella, Dubra and MacCulloch2007; Davis and Williamson, Reference Davis and Williamson2022).

Our questions capture individualistic values as described by Hofstede (Reference Hofstede2001), including self-reliance, personal independence, privacy, weaker family ties, reduced social conformity, and a preference for market competition. For instance, an individual who finds homosexuality justifiable expresses tolerance and the belief in a private life. Believing in private over government ownership reflects support for market competition and capitalism. Placing less importance on making parents proud indicates weaker family ties. Valuing individual over government responsibility highlights a belief in autonomy and individual incentives. Each question thus indirectly captures attitudes aligned with individualistic values.

The individualism index is created by extracting the first principal component from these four questions, reflecting common variation among the components. The eigenvalue of the first component is 2.56, explaining 64% of the total variation in the data. The component weights are all above 0.40 in absolute value, and Cronbach’s alpha for the individualism index is 0.81, confirming strong internal consistency of the standardized components. A higher score reflects greater individualism. The index is standardized for ease of interpretation.

Countries ranking highest on the PAI include Saudi Arabia, Egypt, and Iraq, suggesting these countries hold more traditional views of women in society. Ranking at the bottom of the PAI are Iceland, Denmark, and Sweden. Denmark, Switzerland, and Sweden are the most individualistic countries in our sample. Tunisia, Egypt, and Jordan rank at the bottom of the individualism index, representing more collectivist cultures.

Data are combined to create a cross-section for up to 93 countries from 1981 to 2019. The sample ends in 2019 to avoid biases from the COVID-19 pandemic. Summary statistics are reported in an online appendix, Appendix A.Footnote ¹ Variable definitions and sources are listed in online Appendix B. Online Appendix C lists all countries included in the estimations.

Individualism and support for patriarchal attitudes

Our empirical analysis examines the link between individualism and patriarchal attitudes in three parts. This section evaluates the total effect of individualism on patriarchal attitudes, demonstrating a strong, negative robust association. Section Addressing endogeneity and measurement error addresses issues related to the endogeneity of individualism, and Section Institutional channels of influence investigates institutional factors that mediate the association between individualism and patriarchal attitudes.

Baseline estimations

To investigate the relation between individualism and patriarchal attitudes, we estimate a series of models that take the form:

$$attitud{e_c} = \alpha + \beta individualis{m_c} + \gamma {X_c} + {\varepsilon _c}$$

in which c indexes countries, $attitud{e_c}$ is the patriarchal attitudes index or one of its components, $individualis{m_c}$ is a measure of individualism, ${X_c}$ is a vector of control variables, and ${\varepsilon _c}$ is an error term. The focal variable in this specification is individualism and, as noted earlier, we expect the coefficient on individualism to be negative, so that individualism is associated with less support for patriarchal attitudes. We use both Ordinary Least Squares (OLS) and IV methods.

In the baseline specification, we include four control variables as proxies for institutional quality, oil (% of Gross Domestic Product (GDP)), a country’s share of oil rents in 2019, latitude, the absolute value of latitude, landlocked, a dummy variable for landlocked status, and common law, a dummy variable for the common law tradition. Ross (Reference Ross2008) provides empirical evidence of a strong negative relationship between oil rents and women’s economic and political empowerment. The absolute value of latitude is included as an exogenous measure of institutional quality (Hall and Jones, Reference Hall and Jones1999). Being landlocked slows the spread of ideas (Voigt, Reference Voigt2024) and may insulate a country from international norms regarding gender roles. The common law tradition is associated with the protection of private property rights (La Porta et al., Reference La Porta, Lopez-de-Silanes and Shleifer2008).

Column 1 of Table 1 presents the baseline model. The coefficient on the individualism index is negative and significant at the one percent level, showing that individualism is associated with less support for patriarchal attitudes. The point estimate indicates that a one-standard deviation increase in individualism, equivalent to the cultural distance between Russia (individualism = –0.67) and Puerto Rico (individualism = 0.35) or Puerto Rico and Australia (individualism = 1.33), is associated with a decrease in the PAI of just over three-fourths of a standard deviation, which is roughly the difference in PAI between France (PAI = –1.31) and Guatemala (PAI = –0.56). Among the control variables, only oil rents are significant at conventional levels.

Table 1. Individualism and patriarchal attitudes

Notes: Detailed variable descriptions are provided in online Appendix B. Robust standard errors are shown in parentheses.

***, **, and *denote significance at 1%, 5%, and 10%, respectively.

Next, we use individual components of the PAI as the dependent variable. As seen in columns 2–4, the coefficient on individualism is negative and highly significant in all three regressions. Columns 5–8 provide evidence on the relationship between the PAI and the components of the individualism index. The coefficient on each component variable is significant at the one percent level and has the expected sign. In column 9, all four components of the individualism index are entered simultaneously. The coefficients on these variables have the expected sign and are significant at the five percent level or higher. These findings suggest that the association between individualism and patriarchal attitudes is not driven by any single component of the PAI or individualism indices. The strong empirical associations between support for patriarchal attitudes and the tastes for government ownership and government responsibility are particularly noteworthy, as these variables primarily reflect economic preferences rather than gender-related attitudes. This suggests that individualism’s influence on patriarchal attitudes stems from a broader cultural orientation that shapes both social and economic preferences.

In the final three columns, we consider the robustness of our results to using alternative measures of individualism from Beugelsdijk et al. (Reference Beugelsdijk, Maseland and Van Hoorn2015) and Hofstede (Reference Hofstede2001) and a measure of autonomy derived from Schwartz (Reference Schwartz2006). All three measures are negative and significant at the one percent level, indicating that our results are not driven by the construction of our measure of individualism. Our individualism index has a 0.95 and 0.74 correlation with the BMH and Hofstede individualism variables, respectively, and a 0.82 correlation with Schwartz’s autonomy index. Going forward, we focus primarily on the individualism index, preferring it to the BMH measure on a priori grounds and to the Hofstede and Schwartz measures because it is available for a larger and more representative sample.

Robustness: historical patriarchy

To examine the robustness of the association between individualism and patriarchal attitudes found above, we introduce a variety of control variables that are conceptually or empirically linked to patriarchal attitudes or gender inequality. We introduce most controls sequentially due to sample size constraints. Time-varying controls are measured using 2019 data or as close to that year as possible. See online Appendix B for variable descriptions.

In column 1 of Table 2, we include variables associated with structural shifts in the economy that may influence the economic opportunities available to women. These include the natural log of per capita income and its square, which capture modernization (Inglehart and Norris, Reference Inglehart and Norris2003) and the feminization U-Curve (Uberti and Douarin, Reference Uberti and Douarin2023; Davis, Reference Davis2024), sectoral shares of GDP (Boserup, Reference Boserup1970; Boelmann et al., Reference Boelmann, Raute and Schönberg2021), and the share of exports in GDP (Chen and Hu, Reference Chen and Hu2023). In column 2, we control for three variables that reflect the gender division of labour: women’s educational attainment, the female labour force participation rate, and the fertility rate. As seen in columns 1 and 2, individualism is robust to the inclusion of these variables.

Table 2. Individualism and patriarchal attitudes, total effects

Notes: Detailed variable descriptions are provided in online Appendix B. All specifications include controls for share of oil rents, absolute value of latitude, a landlocked dummy variable, and common law. Robust standard errors are shown in parentheses.

***, **, and *denote significance at 1%, 5%, and 10%, respectively.

We augment the baseline model with two variables related to the history of traditional agriculture, plough, which equals the share of a country’s population with ancestors that used the heavy plough (Alesina et al., Reference Alesina, Giuliano and Nunn2013), and agriculture revolution, which equals the number of years elapsed between the neolithic revolution and the year 2000 for the ancestors of a country’s contemporary population (Hansen et al., Reference Hansen, Jensen and Skovsgaard2015). Results are shown in column 3. While neither agricultural variable is significant, the coefficient on individualism is both negative and highly significant.

Next, we consider religion, which is empirically related to both patriarchal attitudes (Davis and Gao, Reference Davis and Gao2020; Davis, Reference Davis2024) and individualism (Davis, Reference Davis2021) and thus may confound the relationship between the two. To address this concern, we include shares of each country’s population that adhere to eight major religious traditions: Catholic, Protestant, Orthodox Christian, other Christian, Judaism, Islam, Hinduism, and Buddhism. As seen in column 4, individualism is robust to the inclusion of these variables.Footnote ²

Communism is empirically linked to both gender equality (Klasen, Reference Klasen2019) and collectivist attitudes (Alesina and Fuchs-Schundeln, Reference Alesina and Fuchs-Schündeln2007), raising the possibility that our earlier results are biased. To address this issue, we include a measure of a country’s history of communism, which equals the share of the 20th century during which a country had a communist government. As seen in column 5, its inclusion does not appreciably affect the estimated relationship between individualism and support for patriarchal attitudes.

Speaking a gendered language – one in which biological sex plays a significant role in the grammar of nouns or pronouns – is associated with significant consequences for gender equality and women’s empowerment (Mavisakalyan, Reference Mavisakalyan2015; Davis and Reynolds, Reference Davis and Reynolds2018; Santacreu-Vasut et al., Reference Santacreu-Vasut, Shoham and Gay2013, Reference Santacreu-Vasut, Shenkar and Shoham2014; Hicks et al., Reference Hicks, Santacreu-Vasut and Shoham2015; Davis et al., Reference Davis, Mavisakalyan and Weber2024). To see if this omission biases our results, we include measures that reflect the gender intensity of the grammar for nouns and pronouns in a country’s dominant language. As seen in column 6, the empirical relation between individualism and patriarchal attitudes is robust to the inclusion of these variables.

Finally, we consider the influence of additional cultural forces that might influence both individualism and patriarchal values. Columns 7–9 control for political ideology, three additional Hofstede cultural dimensions (power distance, uncertainty avoidance, and masculinity), and regional dummies. The coefficient on individualism is negative and highly statistically significant in all three specifications, reducing concern over cultural omitted variable bias.

Collectively, we find a strong negative association between individualism and support for patriarchal attitudes, robust to the use of alternative measures of key variables and to the inclusion of controls for factors that play an important role in determining patriarchal attitudes or gender inequality.

Addressing endogeneity and measurement error

This section presents instrumental variable analysis addressing two concerns regarding the results presented above. First, as is often the case with survey-based variables, the measurement of individualism is subject to considerable error. Classical measurement error tends to bias coefficient estimates towards zero, and survey-based measures are subject to additional sources of bias (Bertrand and Mullainathan, Reference Bertrand and Mullainathan2001). The second concern is that cultural variables are likely to be endogenous (Bowles, Reference Bowles1998). In particular, attitudes towards traditional gender roles may be codetermined with attitudes towards homosexuality or the subjective importance of making one’s parents proud, giving rise to simultaneity bias in our estimates of the coefficient on individualism.

We address these concerns by instrumenting for individualism using climatic and epidemiological instruments common in the literature. Davis (Reference Davis2016) contends that rainfall variability enhances the value of collectivist social bonds, strengthening informal risk-sharing mechanisms and increasing the benefits of collectivism in high-risk environments, and provides evidence of a robust negative relationship between historical rainfall variation and contemporary measures of individualism. Thus, our first instrument, ancestral rainfall variation, is a measure of rainfall variation experienced by the ancestors of a county’s contemporary population (Davis, Reference Davis2016).

The second instrument, historical disease, is a measure of the historical prevalence of infectious disease from Murray and Schaller (Reference Murray and Schaller2010), which is theoretically and empirically linked to collectivism. Thornhill and Fincher (Reference Thornhill and Fincher2014) argue that exposure to disease tends to empower authoritarian leaders and undermines social support for individual rights, which may be seen as conflicting with public health efforts. In addition, Nikolaev and Salahodjaev (Reference Nikolaev and Salahodjaev2017) argue that exposure to disease increases the intensity of insider-outsider distinctions, leading to greater group identification.

Table 3 presents second-stage results for two-stage least squares regressions in which we instrument for individualism using ancestral rainfall variation and the historical disease. Corresponding first-stage regression results are presented in online Appendix D. We begin with our baseline IV specification, including the four baseline controls. As seen in column 1, the coefficient on individualism is large, negative, and significant at the one percent level, suggesting there is a negative association between support for patriarchal attitudes and the exogenous variation in individualism identified by our instruments. As seen in column 2, we obtain similar results using Hofstede’s measure of individualism.

Table 3. Individualism and patriarchal attitudes, instrumental variable analysis

Notes: Instruments are historical disease and ancestral rainfall variation in columns 1, 2, 6, 7, 8, 9, 10, and 11; ancestral rainfall in column 3; historical disease in column 4. In column 5, instruments are historical disease, ancestral rainfall variation, pronoun drop, and LIFE. Column 12 instruments PAI with plough and ancestral agriculture revolution. First stage results for Panel A are presented in online Appendix D. Detailed variable descriptions are provided in online Appendix B. All specifications include controls for share of oil rents, absolute value of latitude, a landlocked dummy variable, and common law. Robust standard errors are shown in parentheses.

***, **, and *denote significance at 1%, 5%, and 10%, respectively.

Note also that the IV coefficient is roughly 30% larger than the corresponding OLS coefficient from column 1 of Table 2. This finding is consistent with significant attenuation bias in the OLS estimates arising from error in the measurement of individualism and is a common finding in the empirical literature investigating the impact of individualism on a wide array of social outcomes, including per capita income (Gorodnichenko and Roland, Reference Gorodnichenko and Roland2011), business regulation (Davis and Williamson, Reference Davis and Williamson2016), and Lesbian, Gay, Bisexual and Transgender rights (Davis, Reference Davisforthcoming).

The quality of our instruments is assessed using first-stage F-statistics and Sargan’s (Reference Sargan1958) overidentifying restrictions test, reported in the final two rows of Table 3. In column 1, the first-stage F-statistic is 29.2, exceeding the Stock and Yogo (Reference Stock, Yogo, Andrews and Stock2005) critical value of about 20 for 5% maximal IV size distortion with two instruments and one endogenous regressor, indicating strong instruments. Across models 6–11, most exceed this threshold, except column 7 and column 11, which are still above the critical value for 10% distortion. In column 2, when using Hofstede’s measure, the lower F-statistic suggests weaker instruments. Column 5 also reports an F-statistic slightly below the critical value, indicating potential weakness. Thus, these results should be interpreted with caution. Sargan’s test p-values (0.94 in column 1) suggest that we cannot reject instrument validity, conditional on at least one valid instrument.

Next, we consider a variety of specifications motivated by concerns regarding the exclusion restriction. We consider two specifications in which we exclude one instrumental variable and include the other instrument as a control. As seen in columns 3 and 4, neither instrument is significant when included as a control variable and the coefficient on individualism remains highly significant and is nearly identical in value to that in column 1. While not conclusive, these regressions fail to support the claim that the instruments in our baseline specification are invalid.

The fact that the two instruments for individualism used in column 1 are motivated by starkly different theoretical arguments makes it more likely that at least one of these instruments is valid and, thus, that conditions for Sargan’s test to provide information on instrumental validity are met. To bolster the claim that we have at least one valid instrument, we consider a specification in which we employ two additional instruments, pronoun drop (Davis and Abdurazokzoda, Reference Davis and Abdurazokzoda2016) and labour intensity of traditional agriculture (LIFE) (Ang, Reference Ang2019). As first pointed out by Kashima and Kashima (Reference Kashima and Kashima1998), the ability to drop subject pronouns reduces the importance of the speaker relative to the predicate of a sentence and is, thus, associated with collectivism. Ang (Reference Ang2019) provides evidence that labour-intensive farming practices are associated with collectivism.

As seen in column 5, the use of these additional instruments has little effect on the coefficient on individualism or the implications of the overidentifying restrictions test. Thus, the use of additional instruments motivated by very different theoretical arguments, based in linguistics and agricultural economics, bolsters empirical support for the notion that our primary instruments are in fact valid.

Next, we address a series of challenges to the exclusion restriction motivated by the existing empirical literature. Rainfall variation has been linked to social trust, church membership, and the values of religious traditions, each of which may have independent influence on patriarchal attitudes (Buggle and Durante, Reference Buggle and Durante2017; Davis, Reference Davis2021). Rainfall variation may also affect patriarchal attitudes through its impact on agricultural development. A final concern is that our instruments may be correlated with aspects of traditional agriculture that are known to affect patriarchal values. We address these issues by controlling for measures of social trust, religious affiliation, the frequency of attendance at religious services, the suitability of a country’s climate and soil for agricultural production, historical plough use, and time in agriculture. As seen in columns 6–10, the coefficient on individualism is negative and significant in each of these specifications.

As noted earlier, the acceptance of a gender hierarchy may serve to legitimize social hierarchies more broadly and, thus, support collectivist social values. In keeping with this perspective, we consider a two-equation model in which both individualism and patriarchal attitudes are treated as determinants of the other cultural variable. To identify this system of equations, we exclude ancestral rainfall variation and historical disease from the equation with the patriarchal attitudes index as the dependent variable and plough and agriculture revolution from the equation with the individualism index as the dependent variable. This system of equations is estimated using generalized least squares, which controls for the potential correlation of the error terms across equations.

Results are presented in Panel B, columns 11 and 12. As seen in column 11, our findings confirm the existence of a strong negative relationship between the exogenous component of the individualism index and patriarchal attitudes. In contrast, exogenous variation in the PAI does not have a significant impact on individualism, though given the low first-stage F-statistic in column 12, this finding may in part reflect the weakness of our instruments for the PAI. That said, given the absence of empirical evidence of reverse causation, we employ a single equation model for the remainder of the analysis.

This section provides evidence of a strong, negative, statistically significant relationship between the exogenous component of individualism and the patriarchal attitudes index. We consider a variety of specifications to address concerns over the exclusion restriction. While it is not possible to prove the validity of any particular set of instruments, none of the specifications we consider provides evidence that calls the exclusion restriction into question. Thus, our findings are consistent with individualism having a causal effect on patriarchal attitudes.

Institutional channels of influence

Individualism likely influences patriarchal attitudes through direct cultural mechanisms, promoting egalitarian values, and indirect institutional pathways that influence gender norms. This dual influence is rooted in the coevolution of culture and institutions, where cultural values shape institutional quality, and institutions, in turn, reinforce or modify social norms (Bisin and Verdier, Reference Bisin and Verdier2023; Greif and Tabellini, Reference Greif and Tabellini2017). Societies choose institutions to balance private disorder and state control, reflecting cultural preferences (Djankov et al., Reference Djankov, La Porta, Lopez-de-Silanes and Shleifer2003). In individualistic societies, there is a preference for formal institutions that protect individual freedoms, such as liberal democracies.

When formal institutions are perceived as legitimate, they influence behaviour as individuals conform to norms deemed appropriate within their cultural context (March and Olsen, Reference March and Olsen1998; Tyler, Reference Tyler1990). Thus, formal institutions gain legitimacy when aligned with individualism, encouraging norm internalization through repeated exposure to egalitarian practices. For example, democratic institutions increase women’s visibility in public roles, signalling gender equity (Welzel et al., Reference Welzel, Norris and Inglehart2002), while legal protections erode traditional views by equalizing economic opportunities.

We examine formal institutions with theoretical and empirical links to both individualism and patriarchal attitudes: liberal democracy (Coppedge et al., Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman, Bernhard, Cornell, Fish, Gastaldi, Gjerløw, Glynn, Good God, Grahn, Hicken, Kinzelbach, Krusell, Marquardt, McMann, Mechkova, Medzihorsky, Natsika, Neundorf, Paxton, Pemstein, Pernes, Rydén, von Römer, Seim, Sigman, Skaaning, Staton, Sundström, Tzelgov, Wang, Wig, Wilson and Ziblatt2023), economic freedom (Gwartney et al., Reference Gwartney, Lawson, Hall, Murphy, Mitchell, Grier, Grier and Mitchell2024), rule of law and control of corruption (Worldwide Governance Indicators, Reference Kaufmann and Kraay2021), the Women, Business, and the Law (WBL) index of legal gender parity (World Bank, 2024), and women’s political rights (WPR) (Cingranelli et al., Reference Cingranelli, Richards and Clay2014).

As a first step in our empirical analysis, we add each of these institutions to our baseline OLS regression to determine which are empirically related to PAI. Results are presented in online Appendix E. Individualism is negative and significant at the 1% level across all specifications. Liberal democracy, legal gender parity, and women’s political rights exhibit significant negative associations with the PAI. In contrast, control of corruption, economic freedom, and rule of law are insignificant, suggesting that they play a limited role in linking individualism to patriarchal attitudes. Given these findings, going forward, we focus the institutional analysis on liberal democracy, legal gender parity, and women’s political rights.

To quantify the extent to which institutions mediate the association between individualism and patriarchal attitudes, we turn to mediation analysis, which decomposes the total effect of individualism on PAI into the average direct effect and the average causal mediation effect (ACME), which captures the indirect effect of individualism operating through institutions. Thus, we quantify the proportion of individualism’s association with patriarchal attitudes that operates through institutions. We employ both OLS and instrumental variable (IV) mediation analyses, instrumenting individualism with ancestral rainfall variation and historical disease prevalence to address its endogeneity.

Table 4 reports the results.Footnote ³ In Panel A (OLS), the total effect of individualism on PAI is consistent, with coefficients around –0.77. The ACME varies by mediator, ranging from –0.06 to –0.17. Liberal democracy mediates 13.3% of the effect, the legal gender parity 22.2%, and women’s political rights 7.8%. Importantly, individualism’s direct effect remains statistically and economically significant.

Table 4. Individualism, institutions, and patriarchal attitudes, mediation analysis

Notes: ACME is Average Causal Mediation Effect. Panel A reports OLS single mediation analysis. Panel B shows instrumental variable single mediation analysis. Panel C shows instrumental variable simultaneous mediation analysis. Individualism is instrumented with historical disease and ancestral rainfall variation. Detailed variable descriptions are provided in online Appendix B. 95% confidence intervals are reported in parentheses.

Endogeneity most likely biases these estimations. Thus, we turn to IV mediation, estimating a series of equations using structural equation modelling (SEM), instrumenting for individualism, as presented below.

1. (1) First-Stage Regression (Instrumenting Individualism):

$${ individualism}_c=\alpha_0+\alpha_1 Z_{1c}+\alpha_2 Z_{2c}+\alpha_3 W_c+\epsilon_{1c}$$

where Z_1c is ancestral rainfall variation, Z_2c is historical disease, W_c is a vector of baseline controls, and is the error term.

1. (2) Mediator Equation:

$$ institution_c=\beta_0+\beta_1 \hat{X}_c+\beta_2 W_c+\epsilon_{2c}$$

where ${\hat X_c}$ is instrumented individualism, and ${\epsilon_{2c}}$ is the error term.

1. (3) Outcome Equation:

$$PAI_c=\delta_0+\delta_1\hat{X}_c+\delta_2 M_c+\delta_3 W_c+\epsilon_{3c}$$

where ${\hat X_{}}$ is instrumented individualism, ${M_c}{\rm{\;}}$ is the institutional mediator, and is the error term. The indirect effect (ACME) is β1⋅δ2, and the average direct effect is δ1. The total effect is estimated via: PAIc= π0+π1 ${\hat X_c}$ +γ4Wc+ε4c

Our IV mediation analysis instruments individualism with ancestral rainfall variation and historical disease prevalence, ensuring that the direct effect of individualism on patriarchal attitudes ( ${\delta _1}$ ) is robust to measurement error and reverse causality. Lacking valid instruments for formal institutions, indirect effects through institutions like democracy may be upwardly biased, as institutions likely correlate with unobserved factors (e.g., modernization) that also reduce patriarchal attitudes, inflating the mediation effect. Patriarchal attitudes may also affect institutions, introducing reverse causality. Thus, the direct effect is robust to endogeneity concerns, but indirect effects remain associational due to unaddressed institutional endogeneity.

The IV mediation results are presented in Panel B. The total effect is larger (–1.02), consistent with reduced measurement error. The proportion mediated also increases. Liberal democracy mediates 21.4%, legal gender parity mediates 37.3%, and women’s political rights mediates 12.3%, although women’s political rights are no longer statistically significant. Individualism’s direct effect also increases, ranging from –0.64 to –0.89.

To account for correlations among the institutional mediators, we estimate a simultaneous IV mediation model (Panel C, Table 4). Legal gender parity significantly mediates 29.5% and liberal democracy significantly mediates 12.7% of individualism’s effect on patriarchal attitudes. The direct effect remains statistically significant (–0.54), indicating individualism’s influence beyond institutions.

These findings clarify how individualism reduces patriarchal attitudes. The robust direct effect supports its role in promoting gender equality by reducing adherence to patriarchal norms, while indirect effects align with culture-institution coevolution, where culture shapes institutions that reinforce egalitarian norms.