Abstract
Empirical studies show that though there is more room for improvement, much progress has been made toward gender equality since the second wave of feminism. Evidence also suggests that women’s advancements have been more dramatic in the public sphere of work and politics than in the private sphere of family life. We argue that this lopsided gender progress may be traced to uneven changes in gender attitudes. Using data from more than 27,000 respondents who participated in the General Social Survey from 1977 through 2016, we show that gender attitudes have more than one underlying dimension and that these dimensions have changed at different rates over time. Using latent class analysis, we find that the distribution of respondents’ attitudes toward gender equality has changed over the past 40 years. There has been an increase in the number of egalitarians who support equality in public and private spheres, while the traditionals who historically opposed equality in both domains have been replaced by ambivalents who feel differently about gender equality in the public and private spheres. Meanwhile, successive birth cohorts are becoming more egalitarian, with Generation-Xers and Millennials being the most likely to hold strong egalitarian views. The feminist revolution has succeeded in promoting egalitarian views and decreasing the influence of gender traditionalism, but has yet to convince a substantial minority that gender equality should extend to both public and private spheres of social life
Despite major gains for women since the feminist movement of the 1960s, a wealth of research has shown that progress toward gender equality has stalled in recent decades, a phenomenon known as the stalled gender revolution (Cotter, Hermsen, and Vanneman 2011; England 2010; Goldscheider, Bernhard, and Lappegård 2015). One area of the stalled gender revolution that has attracted considerable attention from scholars is gender attitudes. Consistent with other trends, egalitarian progress in gender attitudes slowed in the 1990s (Cotter, Hermsen, and Vanneman 2011). Yet recent research (Fate-Dixon 2017; Knight and Brinton 2017; Pepin and Cotter 2018) suggests that the stalled progress in gender attitudes may be more complicated than initially thought. While there has been a major increase in the number of people holding gender egalitarian attitudes, the trends over time are still unclear, particularly when examining attitudes directed toward different social spheres.
In this study, we build on recent research on gender attitudes and the stalled gender revolution by focusing on attitudinal change over time and whether gender attitudes differ in relation to the public and private realms of social life. We use latent class analysis to examine the varying dimensions of gender attitudes among respondents in the General Social Survey from 1977 through 2016. This approach allows us to identify clusters of gender attitudes in the United States and how they have changed over time and across generations.
In the following section, we review the existing research on gender attitudes which has moved from conceptualizing attitudes toward gender equality as a unidimensional construct to viewing it as multidimensional. Next, we review our source of data and the methods used in analyzing longitudinal trends in gender attitudes. Our findings support the public/private split in gender attitudes discussed by Pepin and Cotter (2018) and Fate-Dixon (2017), but also add significant insight by showing how this attitudinal split is a relatively recent phenomenon replacing the traditional gender attitudes of the past. That is, rather than a return to the past, our findings show attitudes that used to be traditional are now ambivalent, supporting equality in one sphere of social life while opposing it in another. The revolution may be stalled, but there is little evidence that public sentiment is returning to the levels of traditionalism observed in the past.
Gender Attitudes: From Unidimensional to Multidimensional
The majority of studies examining gender attitudes measure this characteristic by combining several survey items into a single scalar measure (Bolzendahl and Myers 2004; Brewster and Padavic 2000; C. Brooks and Bolzendahl 2004; Cotter, Hermsen, and Vanneman 2011; Mason, Czajka, and Arber 1976; also see Davis and Greenstein 2009 for a review). Cotter, Hermsen, and Vanneman (2011), for example, combined four General Social Survey items into a single gender attitude scale in their study highlighting the stall in egalitarian gender attitudes during the 1990s. Indeed, it has been the norm in social science research to calculate the standardized mean of multiple survey questions in order to obtain a reliable measure of respondents’ gender attitudes. This method often requires combining survey items that ask about gender in separate domains of social life. The scale used by Cotter, Hermsen, and Vanneman (2011), for example, included items about women’s suitability for politics alongside questions about domestic family roles and mothering.
Combining survey items into a single construct to measure gender attitudes offers certain advantages, but it also has important limitations. One benefit of this approach is that gender attitude scales tend to be more reliable when multiple items are combined, since responses are less subject to volatile shifts in single survey questions. Furthermore, most research has found that gender attitude survey items are strongly correlated, prompting researchers to assume they share an underlying construct. Yet, one limitation of unidimensional scalar measures is that they limit the researcher’s ability to capture the complexity of gender attitudes. By combining gender attitude questions that focus on the different realms of public and private life, researchers are unable to tease out the relative disparity in attitudinal change in each sphere across time. When researchers use unidimensional scales to measure gender ideologies, they sacrifice the ability to detect heterogeneity within gender attitudes in order to gain greater reliability.
Recent research suggests, however, that progress in gender equality may be distinct between the public sphere of work/politics and the private sphere of the family. On several accounts, the advances made toward gender equality in the labor force far outshine the steps made in the family. In the United States, women now outnumber men in higher education, occupy almost half of the labor force and management positions, and are increasingly visible in top management and politics (Blau, Brinton, and Grusky 2006; BLS 2015; Goldscheider, Bernhard, and Lappegård 2015; Scarborough 2017; Thomas and Wilcox 2014). Much of these gains occurred through shifting industrial conditions, where a growing service economy in the United States, as in other industrialized countries, created new occupational opportunities for women (Charles and Grusky 2004; England 2010). Gendered divisions in the family with regard to housework and child care, on the other hand, have proven to be much more rigid over time, with persistently large gender gaps in time spent on child care and household labor (Bianchi et al. 2006; Parker and Wang 2013). The distinction between the public realm of work/politics and the private realm of family is not just an analytic framework used in research. Young adults interviewed by Gerson (2010) were keenly aware of the different challenges they faced in their career and family goals.
Conceptualizing gender as multidimensional across public and private spheres is hardly groundbreaking. In fact, the public/private split was foundational to the functionalist theories of gender developed by Parsons (1951; Parsons and Bales 1955), which long since have been rejected for legitimating gendered divisions of household labor (Lopata and Thorne 1978). What constitutes a theoretical development, however, is the more recent attention to the different cultural meanings attached to public (work/politics) and private (family) social spheres. While public and private spheres are most certainly interrelated in individuals’ lives, contemporary research has shown that work and family have distinct cultural meanings where people hold separate life plans, values, and opinions (Cech 2013; England 2010; Gerson 2010; Jacobs and Gerson 2015; Pepin and Cotter 2018). What characterizes the difference between public and private spheres across these empirical domains is the increased support of gender egalitarianism in the public realm of work/politics and the ongoing acceptance of gendered responsibilities in the private realm of families where, for example, women are viewed as more naturally nurturant and, therefore, responsible for caregiving.
Divergent trends in opinions about gender in work and family also have been found in research on gender attitudes. It has been long established that support for gender equality at work has been greater than gender equality in the family (Kane and Sanchez 1994; Mason and Lu 1988). Many scholars attribute this to the fact that support for women in the public sphere of work does not necessarily challenge gender hierarchy if women are concentrated in less rewarding and more feminized occupations than men (Charles and Grusky 2004; England 2010). Furthermore, support for women in work resonates strongly with contemporary ideals of liberalism and free choice (Cech 2013; England 2010).
Cotter, Hermsen, and Vanneman (2011) argued that the contrast of increasing support for women at work—with stagnating support for gender equality in the home—represents a new cultural frame of egalitarian essentialism that combines the feminist ideal of free choice with the cultural belief in women’s special nurturing abilities. Within this cultural logic, women have the choice to choose to work for pay if they wish, but most women are expected to assume a primary parenting role as well. Recent research on gender attitudes in the United States provides evidence for this emergent cultural frame by showing increasing support for women’s and mother’s labor force participation alongside waning support for equal gender responsibilities in the family (Fate-Dixon 2017; Pepin and Cotter 2018). According to the egalitarian essentialist framework, the rise of essentialist attitudes in recent years in younger generations is a major contributor to the stalled gender revolution. Other studies, however, have provided alternative explanations. Cross-national research focusing on clusters of gender attitudes have found little evidence to suggest that egalitarian essentialism has recently emerged. Instead, these studies suggest that these types of gender attitudes have existed for decades, and only the proportion of individuals espousing such views has changed over time (Brinton and Lee 2016; Grunow, Begall, and Buchler 2018; Knight and Brinton 2017). According to these scholars, the stalled gender revolution can be explained, in part, by the shrinking number of those who hold traditional attitudes alongside the growing number of those who are indifferent or who support equality at work but not at home.
In short, research on gender attitudes has gone from analyzing gender attitudes as a unidimensional construct to increasingly examining the multiple dimensions of gender attitudes and their changing form over time. Recent research has proposed two frameworks for understanding the shifting dimensions of gender attitudes. First, the egalitarian essentialist framework suggests an emergent gender ideology characterized by the resurgence of traditional attitudes toward the family occurring alongside increasing acceptance of gender equality in the public sphere (Cotter, Hermsen, and Vanneman 2011; Fate-Dixon 2017; Pepin and Cotter 2018). This research has primarily focused on the United States and has used linear regression models treating gender attitudes (and their dimensions) as scalar measures. In the second framework, scholars have focused on clusters of gender attitudes over time, arguing that various configurations of gender attitudes have existed for decades, while the proportion of individuals in each cluster has changed over time (Brinton and Lee 2016; Grunow, Begall, and Buchler 2018; Knight and Brinton 2017). This body of research focused on European or OECD 1 samples and used latent class analysis to identify clusters of gender attitudes.
In this study, we examine changing gender attitudes in the United States from 1977 to 2016. We use latent class analysis (LCA)—defined in the following section—to explore the characteristics and shifting composition of clusters of gender attitudes. To our knowledge, latent class analysis has not yet been used to examine gender attitudes in the United States. Instead, this method has been primarily used in cross-national research. Therefore, one contribution of our paper is to test whether the egalitarian essentialist interpretation of U.S. gender attitudes is supported when using latent class analysis that does not presuppose conceptual dimensions of gender attitudes. In addition to describing clusters of gender attitudes and tracking their historical trajectory, our study also uses multinomial logistic regression to examine the characteristics of individuals who belong to each cluster of gender attitudes. This allows us to test differences between birth cohorts to determine if millennials are, as some have suggested, stalling the revolution in gender attitudes.
Methods
We analyzed the pooled General Social Survey (GSS) (Smith et al. 2017) from 1977 to 2016. The GSS is a repeated cross-sectional survey drawing nationally representative samples of people aged 18 and over who live in non-institutionalized settings within the United States. We used 20 surveys from 1977 through 2016 as data because the GSS transitioned from an annual to a biennial survey from 1994 onwards. Because our analysis revealed two primary dimensions of gender attitudes relating to the public and private spheres, we dropped respondents with missing data on the question measuring public sphere gender attitudes in addition to those with missing data on all of the private sphere attitude items (see below). Our total sample is comprised of 27,032 respondents. On average, the response rates for the GSS are more than 70 percent for each survey year.
Two finite mixture modeling approaches—latent class analysis (LCA) and multinomial logit latent class regression (MNLLCR)—are used in our analysis and are represented in Figure 1 as Blocks A (LCA) and B (MNLLCR), respectively. These will be elaborated below. Finite mixture modeling is an ideal approach for this study because it enables us to identify unobserved heterogeneity where respondents are distributed into classes based on a latent construct (in this case, gender attitudes). LCA is used to identify latent classes and MNLLCR is used to examine the effects of covariates on latent class membership. The descriptive statistics for all study variables are included in the online supplement associated with this article. Next, we elaborate on how this model is measured and operationalized.

Working Conceptual Framework (Finite Mixture Model).
Indicators and Analytic Approach for LCA (Block A)
Latent class analysis (LCA) is a form of multidimensional scaling that reduces the multitude of response patterns across indicators by partitioning the data into smaller number of probabilistic clusters where respondents who reside in the same cluster are assumed to share some underlying association (Clogg 1995; Collins and Lanza 2010). As a result, LCA is more flexible than variable-based modeling (such as confirmatory factor analysis or the use of alpha reliability coefficients) because configurations of gender attitudes are allowed to emerge empirically from the data and are identified based on item response patterns. Relatedly, the main advantage is that these clusters are not defined a priori but emerge empirically from the data, thus making our hypotheses falsifiable (Popper 1959).
We constructed the latent classes based on indicators of four gender attitudes: attitudes toward women in public office, gendered expectations affixed to family social roles, perceptions of the relationship a working mother has with her children, and opinions toward the impact maternal employment has on a young child. Question wording is provided in Table 2. The indicator measuring attitudes about women in public office is a binary variable, while the rest are ordinal variables. 2 We recoded all variables so higher values indicate greater support for gender equality.
Theoretically speaking, the first indicator taps into gendered beliefs about the public sphere, specifically gendered status expectations, that is, the evocation of gender stereotypes in the assessment of women’s emotional competency in leadership in relation to men (D. J. Brooks 2011; Dolan 2014; Thomas and Wilcox 2014). On first blush, this item should be limited to women in political office and should not be generalized to other aspects of the public sphere such as the workplace. However, as scholars who study gender in politics attest, attitudes about women in public offices reflect generalized gendered expectations about the public sphere, particularly about sex-based stereotypes on the presumed incompetence of women, because women lawmakers are publicly elected (D. J. Brooks 2011; Dolan 2014; Thomas and Wilcox 2014). Thus, while our measure of gender attitudes toward the public sphere focuses on politics and not work, findings from this study may inform future research analyzing attitudes toward women in the labor force. 3
The remaining three indicators correspond to two different types of gendered expectations about the private sphere: one concerning gendered family social roles (economic provider/home-maker) based on the structural-functionalist model (Parsons and Bales 1955) and the last two concerning working motherhood (Ciabattari 2001). There is ambiguity about whether the indicator for gendered family social roles (“the man is the achiever outside the home and the woman takes care of the home and family”) measures gender attitudes toward the public or the private sphere. For instance, Yamaguchi (2000) used this indicator to measure gender attitudes on the public sphere, while, conversely, Donnelly, Twenge, and Clark (2015) used the indicator to measure attitudes toward family roles. By virtue of using LCA, we avoid making any presumptions about whether this item measures gender attitudes in the public or private sphere. Instead, our analysis revealed that responses to this item followed similar patterns as the questions on motherhood. Therefore, we positioned this variable as a measure of private sphere gender attitudes.
There are two sets of parameters that are estimated in LCA: class membership probabilities and item-response probabilities. With regard to the former, instead of assuming that each person belongs exclusively to one latent class, the model accounts for a degree of uncertainty in class membership by generating posterior probabilities, that is, the probability of each respondent belonging to each identified latent class. The item-response probabilities are the conditional probabilities that a particular individual within a latent class would respond affirmatively to each of the four gender attitude questions. We use the item-response probabilities to help assess the qualitative nature—or meaning—of each latent class.
The first step in LCA is to identify the number of classes. We examined models with one to seven latent classes by analyzing the pooled dataset containing 20 survey years from 1977 to 2016. We controlled for year in each of these models. Class enumeration depends on several criteria: theory, substantive meaning, and fit indices (Collins and Lanza 2010). Regarding fit statistics, we relied on log-likelihood ratio statistics (G2), the Vuong-Lo-Mendell-Rubin likelihood ratio test (VLMR), the Bayesian Information Criterion (BIC), the adjusted BIC, and Entropy (Asparouhov and Muthén 2012; Nylund, Asparouhov, and Muthén 2007). In instances where fit statistics did not agree on the best model, we relied on the recommendation by LCA experts to choose the number of latent classes based on the most conceptually defensible and meaningful model that is theoretically informed (Collins and Lanza 2010). Table 1 provides the LCA fit statistics.
Latent Class Model Fit Statistics, n = 27,032
NOTE: Shaded latent class indicates model with best fit. AIC = Akaike information criterion; BIC = Bayesian information criterion; df = degrees of freedom; G2 = log-likelihood ratio statistics; VLMR = Vuong-Lo-Mendell-Rubin likelihood ratio test.
Conventionally, the lower the AIC, BIC, and adjusted BIC indices, the better the fit. In contrast, entropy reflects how distinct the classes are from one another, and the higher the value, the better the fit. VLMR compares the improvement in fit between k and k – 1 class models. A significant p value indicates that the k class model is a better fit than the k – 1 model. When comparing models with a varying number of classes, Table 1 reports that the model with six classes had the lowest BIC and adjusted BIC, indicating best model fit. While the AIC and log-likelihood ratio statistic improved when moving from the six-class model to the seven-class model, the BIC and adjusted BIC, which balance fit and parsimony, indicated that the six-class model outperformed the seven-class model. The VLMR provided little information because the results were significant across the models. Following previous research using this method (Knight and Brinton 2017), we gave greater weight to the BIC and adjusted BIC and chose the 6-class model for further analysis.
Covariates and Analytical Approach for Multinomial Logistic Regression (Block B)
Multinomial logit latent class regression (MNLLCR) is an extension of LCA with the purpose of predicting latent class membership (Yamaguchi 2000). Conceptually, latent classes are analogous to a multinomial variable with discrete unordered categories. 4 Hence, we used covariates (sociodemographic variables) in a multinomial logistic regression predicting class membership. 5 The covariates included race, gender, education, birth cohort, employment status, marital status, parenthood, and region, and they have been identified by prior research as predictors of gender attitudes (Brewster and Padavic 2000; C. Brooks and Bolzendahl 2004; Cotter, Hermsen, and Vanneman 2011).
After using LCA to identify class membership for the pooled sample of survey years from 1977 through 2016, we conducted multinomial logistic regression models to identify the likelihood of class membership for key predictors. In these models, we removed respondents with missing data on independent variables, slightly reducing our sample to 26,820. Because a major motivation of this study is examining differences between birth cohorts, we first examined predicted probabilities of class membership comparing pre-Baby-boomers, Baby-boomers, Generation-Xers, and Millennials 6 using the pooled sample of survey years from 1977 through 2016. Then, to examine change over time in differences between birth cohorts, we performed separate multinomial logistic regression models for each survey year. 7 Finally, we analyzed other predictors of class membership using the pooled sample of survey years.
There are some limitations to our analytic approach that should be noted. First, at its inception, ethnoracial categories in the GSS were reduced to a tripartite system—white, black, and others—and the race/ethnicity of the participants were determined by the interviewer. The ethnoracial categories were eventually expanded to include other categories, including Asians and Hispanic/Latino/Latinas in the late 1990s and early 2000s and amended to be self-reported. Because our analysis drew data from 1977 onwards, we hesitantly used the tripartite racial categories in this analysis for consistency’s sake, while remaining cognizant of its limitations. Second, our analysis does not thoroughly examine gender differences across race and class. Small sample sizes in intersectional categories (race by gender by class) prevent us from including these analyses here.
Changing Gender Attitudes from 1977 through 2016
How Persistent?
Across the years, six distinct types of latent classes emerged. The item response patterns used to describe these classes are reported in Table 2. Strong egalitarians and egalitarians were respondents who, respectively, strongly agreed and agreed with gender equality across all questions. In the opposite direction, strong traditionals and traditionals were respondents who, respectively, strongly disagreed and disagreed with gender equality on all items. 8 Finally, the third pair of latent classes to emerge in our analysis consists of a group we call ambivalents; they expressed different levels of support for gender equality in the public sphere of work/politics and the private sphere of the home. Within this pair, pro-public anti-private ambivalents (henceforth referred to as pro-public ambivalents) supported gender equality in the public sphere and opposed it in the private sphere. The second group, anti-public pro-private ambivalents (henceforth, anti-public ambivalents) reported lower levels of support for gender equality in the public sphere, while supporting equality in the private sphere. In general, pro-public ambivalents made up a larger share of respondents than anti-public ambivalents. Interestingly, both groups of ambivalents had bimodal responses for the question about family roles, with nearly equal shares supporting and opposing equality. While this item was clearly supported or opposed by egalitarians and traditionals, both classes of ambivalents were uncertain. Yet ambivalents’ responses to this question were closer to their responses to the other private sphere items than to the item measuring public sphere gender attitudes.
Conditional Probabilities for Each Item Response by Class
NOTE: Bold indicates largest conditional item-response probability. Egal. = egalitarian; Ambiv. ProPub = pro-public anti-private ambivalents; Ambiv. AntiPub = anti-public pro-private ambivalents; Trad. = traditional.
Figure 2A and B illustrates the changing proportion of respondents within each latent class from 1977 to 2016. There are two salient results. The first is the steady increase in the proportion of Americans who were strong egalitarians and egalitarians, increasing from 5 to 25 percent and 17 to 44 percent, respectively. Two-thirds of Americans now hold egalitarian attitudes toward both public and private spheres. There were also major changes in the proportion of respondents who were traditionals and strong traditionals. In 1977, traditionals made up the largest latent class (at 35 percent), followed by strong traditionals (24 percent). By 2016, these two groups were virtually non-existent, with only 4 percent of respondents classified as traditionals and 3 percent classified as strong traditionals. Looking closer at the year-to-year change in class proportions, major declines in the proportion of traditional and strong traditional respondents occurred between 1977 and the early 1990s. By 1993, traditionals and strong traditionals were the smallest classes, and remained so through 2016. As traditionals disappeared, the proportion of ambivalents grew. Between 1977 and 1993, the proportion of pro-public ambivalents increased from 6 to 28 percent. In other words, by the early 1990s, pro-public ambivalents had largely replaced traditionals. The proportion of respondents classified as anti-public ambivalents, meanwhile, was relatively stable, fluctuating around 10 percent throughout the years included in this study.

(A) Proportion of Respondents in Each Class by Year; (B) Collapsed Classes into Egalitarians, Ambivalents, and Traditionals
As the gender revolution in attitudes unfolded, there were two general trends that can be observed in Figure 2B that collapses the six latent classes into three broad groups of egalitarians, traditionals, and ambivalents. First, there was growth in the proportion of Americans who support gender equality in both spheres. By 2016, egalitarians and strong egalitarians together constituted 69 percent of the U.S. public, increasing by 47 percentage points since 1977. Second, the replacement of traditionals and strong traditionals by pro-public ambivalents, along with the consistent size of the anti-public ambivalent class, indicates that the gender revolution had success among Americans who hold traditional views about gender by changing their worldview to see women as having equal rights as men, if only in one sphere and not the other. Our results show that gender attitudes neither have improved uniformly between the public and private spheres across the years nor have they stalled completely. This differential pace in the improvement of gender attitudes, leading to the emergence and persistence of ambivalents, might be interpreted as evidence for a new cultural frame labeled egalitarian essentialism (Cotter, Hermsen, and Vanneman 2011; Pepin and Cotter 2018) where respondents report egalitarian attitudes about gender in work/politics while continuing to espouse traditional opinions about gender in the family. Yet, two pieces of evidence support another interpretation that Americans with very traditional worldviews have changed their ideas about the place of women in society, but not in all social spheres. First, the decline in traditionals coupled with the rise and/or persistence of ambivalents suggests that the traditionals of earlier periods are the ambivalents of today. Thus, rather than the emergence of a new cultural frame, we may instead be observing a reconfiguration of traditionalism where respondents maintain reluctance to gender equality by holding on to traditional norms in one social sphere, while accepting egalitarian changes in the other—a trend previously observed in Europe (Grunow, Begall, and Buchler 2018; Knight and Brinton 2017). Second, while the emergence of egalitarian essentialism may be applied to pro-public ambivalents who support equality in the public, but not private, spheres, it is less applicable to anti-public ambivalents who oppose equality in public spheres but support it in the private sphere.
How Different?
We now turn the analysis to investigate sociodemographic characteristics of each latent class of gender attitudes. In particular, we focus on differences between birth cohorts to determine if Millennials as a cohort are moving backwards to less egalitarian beliefs. First we present predicted probabilities of latent class membership based on respondent birth cohort using the pooled sample that includes survey years from 1977 through 2016. Second, we present predicted probabilities of class membership calculated with separate multinomial logistic regression models for each of the 20 survey years included in the data. Predicted probabilities were calculated following the marginal effects at the mean (MEMS) (Long and Freese 2014, 341) or the “average then predict” approach (Gordon 2012, 592).
Figure 3 presents the marginal effects of cohort on predicted class for the pooled sample. In general, these results show that gender egalitarianism has grown with each successive generation. Baby-boomers were more likely to be (strong) egalitarians than pre-Baby-boomers, Generation-Xers were more likely to be strong egalitarians than Baby-boomers, and Millennials were more likely to be strong egalitarians than Generation-Xers. These findings support the notion of steady progress toward liberal gender attitudes. Yet, other indicators reported in Figure 3 add to this narrative. Baby-boomers, Generation-Xers, and Millennials all were less likely than pre-Baby-boomers to be (strong) traditionals and pro-public ambivalents. Comparing more recent generations, differences in the likelihood of being a pro-public ambivalent emerged as a key differentiator between birth cohorts. Generation-Xers were significantly less likely to be pro-public ambivalents than Baby-boomers, and Millennials were significantly less likely to be pro-public ambivalents than Generation-Xers. These findings indicate that while traditionalism marked the primary point of divergence between pre-Baby-boomers and later generations, differences in pro-public ambivalence constitute the major point of differentiation between Baby-boomers, Generation-Xers, and Millennials.

Predicted Probabilites of Birth Cohort on Latent Class Membership
Successive generations are more likely to be egalitarian, less likely to be traditional, and less likely to be pro-public ambivalent. These three trends further support the notion that gender attitudes are becoming more egalitarian with each passing generation. However, differences between birth cohorts in the likelihood of being anti-public ambivalents (those who oppose gender equality in the public sphere but support it in the private sphere) provide insight into a separate dimension of inter-generational change. Generation-Xers were significantly more likely to be anti-public ambivalents than Baby-boomers, while Millennials were more likely to be anti-public ambivalents than pre-Baby-boomers, Baby-boomers, and Generation-Xers. These findings suggest an increase in anti-public ambivalent attitudes in recent generations, although it is important to contextualize these results. The size of the anti-public ambivalent class never exceeds 15 percent across years included in this sample. Thus, while there may be more anti-public ambivalents in Generation-X and Millennial generations, they remain a minority, especially compared to egalitarians whose numbers have increased with each successive birth cohort.
We also analyzed the marginal effects of birth cohort on predicted class probabilities for each year to determine the stability of generational differences in gender attitudes. The results of this analysis are included in the online supplement (Figure 1A through F), and we briefly summarize key themes here. Baby-boomers were consistently more likely to be egalitarians than pre-Baby-boomers, while being equally as likely to be either pro-public or anti-public ambivalents from 1977 through 2016. Baby-boomers were less likely than pre-Baby-boomers to be traditionals, although the size of this effect diminished after the early 1990s and is non-significant for most years following 2000. Comparing Generation-Xers and pre-Baby-boomers across survey years, there were decreasing marginal effects of class membership for traditional and ambivalent classes and a growing significance for the strong egalitarian class, indicating that the primary difference between pre-Baby-boomers and Generation-Xers is that the latter are more likely to be strong egalitarians. Across years where Millennials are represented in the data, we found that Millennials were generally more likely to be strong egalitarians, less likely to be pro-public ambivalents, and less likely to be (strong) traditionals than pre-Baby-boomers.
Examining differences in predicted class membership between Baby-boomers and the two later birth cohorts, we found that across most years in the data, Generation-Xers had similar probabilities of membership as had Baby-boomers in (strong) traditional, (strong) egalitarian, and anti-public ambivalent classes. There were, however, differences with regard to the pro-public ambivalent class. With few exceptions, Generation-Xers were consistently less likely than Baby-boomers to belong to the class of respondents supporting equality in the public sphere but not in the private sphere. Similar differences were observed when comparing Millennials and Baby-boomers. Millennials were equally as likely to be egalitarians, (strong) traditionals and, with the exception of 2012, anti-public ambivalents as Baby-boomers. However, Millennials were less likely than Baby-boomers to be pro-public ambivalents and more likely to be strong egalitarians across nearly all years in the data.
Finally, examining the marginal effects of predicted class membership for Millennials compared to Generation-Xers, we found no significant effects for membership in either traditional class, while differences in predicted membership in anti-public ambivalent and (strong) egalitarian classes were non-significant for most years. With regard to the pro-public ambivalent class, only in 2006 and 2016 were Millennials less likely than Generation-Xers to belong to this group.
In summary, our analysis reveals that Millennials are not stalling the gender revolution. Like Generation-Xers and Baby-boomers, Millennials are less likely to be traditionals than pre-Baby-boomers. This resonates with our earlier finding that traditional gender attitudes are increasingly rare—so much so that only pre-Baby boomers hold these attitudes at higher rates than other birth cohorts. Consistent with this, Baby-boomers, Generation-Xers, and Millennials are about equally as likely to belong to either of the traditional classes. Instead, the traditionals of earlier generations are the pro-public ambivalents of more recent times. Both Generation-Xers and Millennials are less likely than Baby-boomers to be classified as pro-public ambivalents, while being more likely to be classified as strong egalitarians. Thus, if the generational break between pre-Baby-boomers and later birth cohorts is defined by declining traditionalism, the difference between Baby-boomers and later birth cohorts is characterized by declining pro-public ambivalence and increasing strong egalitarianism.
We did not identify a similar break in attitudes between Generation-Xers and Millennials. There was some evidence to suggest that Millennials have greater anti-public ambivalent attitudes than previous birth cohorts, but the size of the effects were small and only significant when pooling survey years. In general there were little differences between birth cohorts in terms of likelihood of belonging to the anti-public ambivalent class, and this is consistent with the trends in Figure 2 showing stability in the proportion of respondents in this class over time. While membership in egalitarian, traditional, and pro-public ambivalent classes are subject to cohort replacement, anti-public ambivalence is not. We explore reasons for this in the next section, where we examine other predictors of class membership.
Predictors of Latent Class Membership
To describe the general characteristics of respondents belonging to each latent class, we conducted multinomial logistic regression models for the pooled data set of survey years from 1977 to 2016. Figure 4 reports the marginal effects predicting class membership for key respondent characteristics included in these models. For brevity, we report findings only for variables with significant effects across classes.

Predicted Probabilities of Respondent Characteristics on Latent Class Membership
Respondents in the strong egalitarian class were more likely to be women, have some college education or a college degree, reside in New England, 9 and work part-time (compared to nonworkers) or full-time (compared to part-time and nonworkers). Meanwhile, compared to whites and blacks, other-race respondents were less likely to be strong egalitarians. Membership in the egalitarian class also was more likely for women, blacks, those with more education, and New England residents. While being less likely to be egalitarian than blacks, whites were more likely to belong to this class than other-race respondents.
Predictors of traditional class membership mirrored those of egalitarian class membership. Men, the less educated, those residing in the East South Central Region, and non-workers were more likely to be traditionals and strong traditionals than women, the highly educated, New England residents, and workers. Across race, blacks were less likely than both whites and other-race respondents to be traditionals or strong traditionals, while other-race respondents were more likely than whites to be traditionals or strong traditionals.
Predictors of membership in the pro-public ambivalent class were similar to those of the traditional classes, resonating with the finding that generally pro-public ambivalence has replaced traditionalism in recent years. Women were less likely than men to be pro-public ambivalents, while the college educated and those with some college were less likely to be pro-public ambivalents than respondents whose highest level of education was a high school degree or less. Blacks were less likely than whites and other race respondents to be pro-public ambivalents. Part-time and non-working respondents also were more likely to be pro-public ambivalents than full-time workers.
The final latent class, anti-public ambivalents, is perhaps the most difficult to describe because it has been given significantly less attention in previous literature. While the concept of gender essentialism (Cotter, Hermsen, and Vanneman 2011; Pepin and Cotter 2018) resonates with pro-public ambivalence where respondents support equality in the public but not the private spheres, there has been little acknowledgement of the small, but persistent group of individuals who oppose gender equality in the public sphere but support it at home. The results reported in Figure 4 indicate that women and men are equally as likely to be anti-public ambivalents. Blacks and other-race respondents are both more likely to be anti-public ambivalents than whites. Respondents with some college or a college degree are less likely to be anti-public ambivalents than those with only a high school or less than high school education. One reason for these race and education effects could be labor market competition. Cross-national research has found that high unemployment is related to lower acceptance of women’s workforce participation (Azmat, Güell, and Manning 2006; Fortin 2005), in part because unemployment causes greater competition for jobs and some feel that the few remaining work opportunities should be reserved for men. While the small number of unemployed persons in the GSS prevents us from directly examining the relationship between unemployment and anti-public ambivalence, it is probable that discrimination toward minorities and lack of jobs for the less educated results in competitive labor markets for these groups that may make respondents less likely to support women’s equality in the public sphere while endorsing equality in the private sphere.
Examining the results in Figure 4 as a whole, we find that marginal effects for class membership in traditional, strong traditional, and public-ambivalent classes tend to cluster together, while egalitarian and strong egalitarian classes are almost always plotted near each other. Meanwhile, the coefficients for membership in the anti-public ambivalent class is sometimes similar to traditional classes (education), while other times having coefficients near to egalitarian classes (race). These findings suggest that traditional, strong traditional, pro-public ambivalent, egalitarian, and strong egalitarian classes are each affected by the shared social phenomena of shifting gender norms. Meanwhile anti-public ambivalent gender attitudes appear to be influenced by something separate. While examining these dynamics extends beyond the scope of this study, Figure 5 presents preliminary analysis of the relationship between job competitiveness, measured with the monthly unemployment rate (BLS 2018), and the proportion of respondents in each GSS survey who are anti-public ambivalents. When unemployment was low in the late 1990s and early 2000s, so too was the proportion of anti-public ambivalents. When unemployment peaked in the mid-1980s and in the recession years of 2007 to 2010, there was also an increase in anti-public ambivalent attitudes. Apparent lags in these effects may be due to characteristics of the data, since the unemployment rate is reported monthly while gender attitudes from the GSS are annual estimates occurring at varying intervals. Lag effects also may be due to cultural anticipation of economic shifts, where anxiety toward increasing unemployment stimulates anti-public ambivalent attitudes before unemployment actually rises. Regardless, these findings present preliminary evidence that labor market competitiveness may be a key driver of anti-public gender attitudes, as respondents feel that jobs should be reserved for men in times of scarcity. Previous research on working class occupations support these initial reflections, showing that workers in these jobs often have more gender egalitarian family arrangements than upper-class occupations because of a greater reliance on women’s wages (Gerstel and Clawson 2014). Future research is needed to further explore the relationship between shifting employment/labor market conditions and changes in predominant gender attitudes.

Unemployment Rate and the Proportion of Anti-Public Ambivalents
Conclusion
Our results support previous research that points to the differential pace of change toward gender equality in the public sphere of work/politics versus the private sphere of the family (Cotter, Hermsen, and Vanneman 2011; Knight and Brinton 2018; Pepin and Cotter 2018). Our use of latent class analysis to study gender attitudes in the United States adds significant depth to this previous research. First, gender attitudes were found to cluster into six categories. Egalitarians and strong egalitarians supported gender equality at home and in the public sphere. The size of these egalitarian classes has consistently increased since 1977. Traditionals and strong traditionals oppose gender equality in both public and private spheres. While these classes were the largest in 1977, they became virtually nonexistent by the early 1990s, mirroring a trend found in international research (Knight and Brinton 2017; Grunow, Begall, and Buchler 2018). At that point, traditionals were replaced by pro-public ambivalents who supported equality in the public sphere and opposed it in the private sphere. Rather than the pro-public ambivalent class being composed of young people endorsing a new cultural frame of egalitarian essentialism, as suggested by previous scholars (Cotter, Hermsen, Vanneman 2011; Pepin and Cotter 2018), we found that individuals in this group were likely to be older and less educated people who had left behind more traditional attitudes. Finally, our analysis revealed a sixth cluster of gender attitudes composed of people who opposed gender equality in the public sphere and supported it in the private sphere. This group consistently makes up a small share of total respondents, and our findings suggest that these attitudes may be related to labor market competition where individuals feel that public-sphere opportunities should be reserved for men.
These analyses challenge the notion that there has been a stalled revolution in gender attitudes. Instead, our results support previous research (Grunow, Begall, and Buchler 2018; Knight and Brinton 2017) suggesting that there has been ongoing change, albeit slow, in an egalitarian direction since 1977. At the macro level of the gender structure, egalitarian ideology continues to become more common as traditionals have been replaced by ambivalents. While ambivalent attitudes are certainly not emblematic of gender egalitarianism, they are markedly more liberal than the traditionalism of the past.
There may be more good news. Millennials are less likely to be pro-public ambivalents than previous birth cohorts. While they are slightly more likely to be anti-public ambivalents, this effect is diminishing as time progresses. These findings support qualitative research on Millennials that suggests that while they are a diverse generation, overall they support gender egalitarianism at home and at work (Risman 2018). Millennials and Generation-Xers are the most likely to be strong egalitarians. The gender revolution may be uneven and unfinished at this moment in history, but our results show that feminism is well on its way to becoming the American normative cultural logic as the majority now endorse gender equality in both the public and private spheres.
Supplemental Material
GandS809604_Online_Appendix_CLN – Supplemental material for Attitudes And The Stalled Gender Revolution: Egalitarianism, Traditionalism, and Ambivalence from 1977 through 2016
Supplemental material, GandS809604_Online_Appendix_CLN for Attitudes And The Stalled Gender Revolution: Egalitarianism, Traditionalism, and Ambivalence from 1977 through 2016 by William J. Scarborough, Ray Sin and Barbara Risman in Gender & Society
Footnotes
Authors’ Note:
The authors would like to thank Shannon Davis, Bill Bielby, Lorena Garcia, Rachel Gordon, Anthony Paik, Brian Powell, and the editors and anonymous reviewers at Gender & Society for feedback on earlier versions of this article.
Notes
William J. Scarborough is a PhD candidate in sociology at the University of Illinois at Chicago and research assistant at the Institute for Research on Race and Public Policy. His research examines cultural and economic change across the United States and its relation to inequality. His work has been published in multiple academic journals including Social Currents, Socius, and The Sociological Quarterly. With Barbara Risman and Carissa Froyum, he is also the co-editor of the Handbook of the Sociology of Gender.
Ray Sin, PhD, is a behavioral scientist at Morningstar, Inc. He applies theories and concepts from multiple disciplines (primarily sociology, economics, and psychology) to better learn how to help investors make better decisions. To that end, he has conducted research and published on sustainable investments and the perception of fees. His research has been covered by InvestmentNews, Kiplinger, and Reuters. Ray holds a bachelor’s degree in sociology from National University of Singapore, and received his PhD, also in sociology, from the University of Illinois at Chicago.
Barbara Risman is a College of Liberal Arts & Sciences Distinguished Professor of Sociology at the University of Illinois at Chicago. Her work includes Where The Millennials Will Take Us: A New Generation Wrestles with the Gender Structure (Oxford University Press, 2018), a second edition of her textbook, Families As They Really Are (Norton 2015 co-edited with Virginia Rutter), and the Handbook on the Sociology of Gender (Springer 2018) co-edited with Carissa Froyum Roise and William Scarborough. She is a public intellectual whose editorials have appeared in the Chicago Tribune, The Seattle Times,
, and the Huffington Post and is frequently quoted in the press, including in the Economist, LA Times, New York Times, and The Atlantic. Her awards include the 2011 American Sociological Association’s Award for the Public Understanding of Sociology and the 2005 Katherine Jocher Belle Boone Award from the Southern Sociological Society for lifetime contributions to the study of gender.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
