Abstract
Theory predicts that individuals’ sexism serves to exacerbate inequality in their society’s gender hierarchy. Past research, however, has provided only correlational evidence to support this hypothesis. In this study, I analyzed a large longitudinal data set that included representative data from 57 societies. Multilevel modeling showed that sexism directly predicted increases in gender inequality. This study provides the first evidence that sexist ideologies can create gender inequality within societies, and this finding suggests that sexism not only legitimizes the societal status quo, but also actively enhances the severity of the gender hierarchy. Three potential mechanisms for this effect are discussed briefly.
Sexist ideologies have been classified as hierarchy-enhancing legitimizing myths that justify the creation of inequality (Sidanius & Pratto, 1999). These sexist ideologies are associated with a lower likelihood of voting for female political candidates (Swim, Aikin, Hall, & Hunter, 1995), less support for women in traditionally male (i.e., high-status) educational and occupational domains (Sakallı-Uğurlu, 2010; Swim et al., 1995), and opposition to public policies designed to attenuate male dominance (Sibley & Perry, 2010). However, the existing evidence for hierarchy-enhancing effects of sexist ideologies is indirect and only suggests that prejudice and discrimination lead to hierarchical intergroup relations at the societal level. The purpose of this study was to directly test the hypothesis that sexism is a hierarchy-enhancing ideology by examining the contribution of sexist ideologies to increases in gender inequality across 57 societies.
Ample evidence from a variety of samples and measures of sexism indicates that societal sexism correlates with societal measures of gender inequality (Glick et al., 2000, 2004; Inglehart & Norris, 2003; Napier, Thorisdottir, & Jost, 2010). These correlations are necessary but not sufficient to implicate sexism in the creation of inequality—a limitation noted in all of the sources just cited. It is possible, for example, that the association between sexism and gender inequality is the result of people endorsing sexist ideologies in an effort to justify existing gender inequality (Jost, Banaji, & Nosek, 2004; Kay et al., 2009), rather than the result of sexist ideologies contributing to increased gender inequality.
In the study reported here, I directly tested whether sexism enhances gender hierarchies. The measure of sexism used in this study was drawn from the fifth wave of the World Values Survey (WVS; World Values Survey Association, 2009), which included 57 countries. These data were collected between 2005 and 2007. Gender inequality was indexed by the United Nations Gender Empowerment Measure (GEM; United Nations Development Programme, 2005, 2006, 2007/2008, 2009b). The GEM is a composite measure summarizing the extent of women’s and men’s parity in their shares of managerial and professional jobs, parliamentary seats, and earned income in a society. Values for the GEM were calculated at the same time as WVS data were collected (Time 1) and after WVS data were collected (Time 2; the mean interval between Time 1 and Time 2 was 3 years). If greater sexism at Time 1 predicted an increase in gender inequality at Time 2, while controlling for gender inequality at Time 1, this result would suggest that sexism actively enhances the gender hierarchy. This approach can provide evidence that sexism temporally precedes gender inequality, and temporal precedence is key to demonstrating a causal effect (Cohen, Cohen, West, & Aiken, 2003).
Higher levels of societal health, wealth, and education have been related to lower levels of support for sexist gender ideologies (Inglehart & Norris, 2003; Napier et al., 2010) and lower levels of gender inequality (Inglehart & Norris, 2003). Therefore, changes in gender inequality may not be due to sexism, but instead may be due to levels of educational and economic development. To rule out this alternative explanation, I included societies’ levels of development in the analysis.
Method
Participants and procedure
Data were drawn from the fifth wave of the WVS (World Values Survey Association, 2009), which includes representative samples from 57 countries and highly autonomous regions (e.g., Taiwan). Data from 82,905 participants (47.9% men, 52.0% women, 0.1% no reply; mean age = 41.4 years, SD = 16.5) were collected using face-to-face interviews between 2005 and 2007. For a complete list of sample sizes and societies included, see Table 1.
Mean Sexism Among Women and Men at Time 1 and Societal Gender Inequality at Times 1 and 2 Across the 57 Societies in This Study
Note: Societies are ranked in order from least sexist to most sexist. Standard deviations are shown in parentheses. Gender inequality was indexed by Gender Empowerment Measure (GEM) scores, which can range from 0 (complete gender inequality) to 1 (complete gender equality; United Nations Development Programme, 2005, 2006, 2007/2008, 2009b). Sexism was measured on a scale from 1 to 4, with higher numbers indicating more sexism (World Values Survey Association, 2009). Independent t tests with listwise deletion were used to obtain significance values for the Cohen’s ds; negative numbers indicate greater sexism among men, whereas positive numbers indicate greater sexism among women. Data for the item “men make better business executives” were not available for Guatemala, Hong Kong, Iraq, and New Zealand. The means and standard deviations of sexism for these four countries were calculated using the scale’s remaining item.
p < .05. **p < .01. ***p < .001.
Measures
Sexism at Time 1
The measure of sexism included in the WVS is the same measure that Napier et al. (2010) used to represent hostile sexism in their cross-national examination of sexism and subjective well-being. This measure consists of two items that assess aversion and hostility to women in stereotypically male domains: “On the whole, men make better political leaders than women do” and “On the whole, men make better business executives than women do.” Responses were given on scales that ranged from 1 (strongly disagree) to 4 (strongly agree), where higher scores indicate more sexism. The responses on the two items were averaged to create a reliable scale (r = .68, p < .001).
To demonstrate that the measure of sexism in the WVS tapped the same construct of sexism as more common and well-validated measures do, I obtained a community sample via Amazon’s Mechanical Turk 1 (34 men, 52 women; mean age = 29.3 years, SD = 11.1; participants were paid $0.10), and I obtained a student sample from an introductory psychology course (31 men, 60 women; mean age = 20.1 years, SD = 3.1; participants received partial course credit). Participants in both samples completed the measure from the WVS (community sample: r = .54, p < .001; student sample: r = .49, p < .001) and five established measures of sexism.
The WVS measure was significantly and positively correlated with the Hostile Sexism Inventory (Glick & Fiske, 1996), r(79) = .55, p < .001, and r(85) = .55, p < .001; the Attitudes Toward Women Scale (Spence, Helmreich, & Stapp, 1973), r(77) = .64, p < .001, and r(83) = .58, p < .001; the Modern Sexism scale (Swim et al., 1995), r(78) = .43, p < .001, and r(86) = .47, p < .001; and the Old-Fashioned Sexism scale (Swim et al., 1995), r(83) = .56, p < .001, and r(87) = .60, p < .001 (for each measure, results for the community sample precede those for the student sample). Similar results were obtained when the Hostile Sexism Inventory was split into its three theoretical components: competitive gender differentiation, r(82) = .56, p < .001, and r(89) = .49, p < .001; dominant paternalism, r(83) = .49, p < .001, and r(89) = .58, p < .001; and heterosexual hostility, r(84) = .39, p < .001, and r(87) = .29, p = .01. With the exception of the Benevolent Sexism Inventory (Glick & Fiske, 1996), r(78) = .17, p = .13, and r(83) = .14, p = .21, in each sample the WVS sexism measure was significantly associated with these established measures of sexism. In sum, the WVS sexism measure taps into a form of hostile and modern sexism characterized by competitive gender differentiation and dominant paternalism. 2
Societal gender inequality at Time 1 and Time 2
Gender inequality was measured using the GEM (United Nations Development Programme, 2005, 2006, 2007/2008, 2009b). Each country’s GEM score at Time 1 was drawn from the United Nations Human Development Report that was released the same year that country’s WVS was completed (2005, 2006, or 2007); values for Time 2 were taken from the 2009 United Nations Human Development Report. Scores on the GEM range from 0 (complete gender inequality) to 1 (complete gender equality).
Control variable at Time 1
The Human Development Index (HDI; United Nations Development Programme, 2009a) at Time 1 was included as a control variable because a country’s wealth, health, and education (the three components of the HDI) have been related to gender attitudes in a country (Inglehart & Norris, 2003; Napier et al., 2010). The HDI ranges from 0 (completely undeveloped) to 1 (completely developed).
Primary analytic strategy
A cursory glance at Table 1 reveals that there are some missing data. Rather than using traditional listwise or pairwise deletion strategies, I used full-information maximum-likelihood estimation. This technique uses all of the available data to simultaneously estimate the model and the missing data on the basis of the values of the other parameters in the model. This technique provides more accurate and less biased results than listwise or pairwise deletion strategies and is highly recommended by statisticians (Peugh & Enders, 2004).
In the primary analysis, I regressed GEM at Time 2 on three variables—sexism at Time 1, GEM at Time 1, and HDI at Time 1—using a multilevel model estimated with MPlus Version 6 (Muthén & Muthén, 2010). In order to estimate societal levels of sexism, I decomposed the individual-level measure of sexism into two latent parts: one representing within-level variance and one representing between-level variance (Lüdtke et al., 2008). Practically, this means that sexism at the societal level can be interpreted as a latent representation of a society’s deviation from the cross-society mean level of sexism; this is conceptually equivalent to grand-mean centering. This strategy makes it possible to model the emergent effects of individual sexism on the societal-level measure of gender inequality (Preacher, Zyphur, & Zhang, 2010). Individual-level sexism is not discussed because no other variables are considered at the individual-level in this study.
Results
The means and standard deviations for the sexism and inequality measures are displayed in Table 1. Consistent with past sexism research (e.g., Glick & Fiske, 1996; Swim et al., 1995), the results indicated that men were more likely than women to report sexist attitudes in 56 out of the 57 societies sampled.
The findings of the multilevel model were consistent with my prediction: Higher sexism predicted higher GEM at Time 2 when controlling for the effects of GEM at Time 1 and HDI at Time 1 (b = −.06, SE = .03, p = .04). These findings show that greater sexism predicts decreases in gender equality over time and provide the first direct evidence of sexism as a hierarchy-enhancing ideology.
To test for gender differences in sexist attitudes, I calculated identical multilevel models separately for women and for men. I computed z scores for the difference between the slopes in the women’s and men’s models for all of the paths in the models, but no gender differences emerged (all zs < |0.23|; all ps > .82). This finding indicates that sexism may be a consensual legitimizing myth endorsed by both high-status and low-status groups in the creation of gender hierarchy (cf. Sidanius & Pratto, 1999).
Additionally, and consistent with past correlational research on societal sexism (e.g., Glick et al., 2000, 2004; Inglehart & Norris, 2003), sexism at Time 1 was negatively associated with both GEM at Time 1 (b = −.08, SE = .02, p < .001) and HDI at Time 1 (b = −.05, SE = .01, p < .001). These results show that sexism is more prevalent in countries that are less developed and have more gender inequality. HDI at Time 1 (b = .19, SE = .06, p = .001) and GEM at Time 1 (b = .82, SE = .08, p < .001) significantly predicted GEM at Time 2. These results indicate, respectively, that more developed countries are more likely to have increases in gender equality and that gender equality is relatively stable over time. Thus, the relationship between sexism at Time 1 and GEM at Time 2 is especially impressive because GEM at Time 1 and HDI at Time 1 were both strongly associated with sexism at Time 1 and GEM at Time 2, and the societal-level sample size was relatively small (N = 57 societies).
Discussion
Theorists have characterized sexist beliefs as a type of hierarchy-enhancing ideology; however, little work has demonstrated that sexism directly exacerbates the gender hierarchy. Consistent with past work (Glick et al., 2000, 2004; Inglehart & Norris, 2003; Napier et al., 2010), this study found an association between sexism and gender inequality. Going beyond past work, this study is the first to demonstrate that societal-level sexism predicts increases in systemic gender inequality. By demonstrating that sexism at Time 1 predicted gender inequality at Time 2 above and beyond gender inequality at Time 1, I was able to show the temporal precedence of sexism and provide a key piece of evidence for causality (Cohen et al., 2003). The association between sexism at Time 1 and gender inequality at Time 2 was found both for men and for women and even while controlling for a country’s level of development, allowing me to rule out an alternative explanation for the effect. Because it is unethical and infeasible to manipulate an entire society’s sexist ideology, the analytic strategy used in this study may be the only feasible method to test the causal link between these psychological and societal-level variables.
Before making too strong of a case for sexism as a cause of gender inequality, I should note that the prospective analysis approach is not without its shortcomings. First, it cannot rule out the possibility that gender inequality also causes sexism (cf. Jost et al., 2004; Kay et al., 2009), though this reverse causal relationship cannot account for the association between sexism at Time 1 and gender inequality at Time 2 when gender inequality at Time 1 was controlled for. Second, if unanticipated important variables were not included in the model, the model may not have provided accurate estimates of potential effects (a problem inherent to most nonexperimental approaches; Cohen et al., 2003). Therefore, until further research replicates this effect on a different sample of countries and using alternative measures of sexism (e.g., the Ambivalent Sexism Inventory; Glick & Fiske, 1996), the causal conclusions implied by this study should be tempered. Nonetheless, given the breadth of the analysis, the care taken to control for initial levels of gender inequality and development, and the difficulties of experimental research in this area, this analysis provides valuable insight into the role of sexism in creating gender inequality.
One important direction for future research is to determine the mechanism (or mechanisms) of this effect. Three possibilities have been identified. First, sexism is related to discrimination against women, and discrimination could directly account for sexism’s contribution to the creation of inequality (Sibley & Perry, 2010; Swim et al., 1995). Second, when adolescents (especially females) endorse sexist gender ideologies, they are less likely to aspire to a university degree, and women who endorse these ideologies earn lower wages than those who do not endorse these ideologies (Davis & Greenstein, 2009). These trends suggest that when women endorse sexist ideas, they are less likely to acquire social status and resources (perhaps because of a depressed sense of entitlement; Major, 1994). Third, experiencing sexism or even facing the possibility of experiencing sexism is associated with poor performance on objective measures of achievement (Adams, Garcia, Purdie-Vaughns, & Steele, 2006; Logel et al., 2009). Thus, sexism may exacerbate societal gender inequality by causing women to perform worse on a variety of tasks, thereby providing an “objective” basis to deny women jobs and promotions.
Conclusion
This study is the most expansive study of sexism conducted to date and is the first study to demonstrate the temporal precedence of sexism in enhancing gender inequality. By taking advantage of both individual and societal-level data, it was possible to examine the association between ideological beliefs and systemic outcomes. The results presented here suggest that sexism not only legitimizes gender inequality, but actively makes it worse. As people fight against the gender hierarchy, they need to pay attention not only to employment decisions, pay inequity, and violence against women, but also to the ideological forces that drive these effects and exacerbate the subjugation of women.
Footnotes
Acknowledgements
The author would like to thank Corinne Moss-Racusin, Jessica Good, and the Laboratory of Social Science Research at DePaul University for helpful comments on previous versions of this manuscript.
The author declared that he had no conflicts of interest with respect to his authorship or the publication of this article.