Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Occupational Devaluation Due to Feminization? Causal Mechanics, Effect Heterogeneity, and Evidence from the United States, 1960 to 2010

Occupational Devaluation Due to Feminization? Causal Mechanics, Effect Heterogeneity, and... Abstract Proponents of the devaluation hypothesis claim that occupations experiencing a significant inflow of female workers are devalued, both in status and in pay. We suggest that devaluation is an essentially cultural phenomenon that can be subject to change over time and that is not constant with varying contexts. Our theoretical model connects changing gender compositions in occupations with the formation of occupational gender stereotypes. In combination with a cultural bias that attributes less value to female work, these stereotypes can lead to devaluation. This is a novel view on the mechanics at play in the devaluation process. With US census data from 1960 to 2010, we show that devaluation was restricted to sometimes very specific contexts. A trend toward declining or disappearing devaluation is observed over the entire time span. Given potential cultural inertia and the stability of stereotypes, this result could, however, be an artifact of a deficient testing strategy that focuses solely on changes in occupational gender compositions without taking into account the normative power of the past. Introduction When occupations experience a larger inflow of female workers—meaning that they feminize—this leads to their devaluation, translating into a lower status and a decline in wages. The basic premise of the devaluation hypothesis is that this link between feminization and devaluation is indeed causal, and that it is due to cultural norms that attribute more economic value to the work done by men than to the work done by women. In the cross-section, this translates to the phenomenon that predominantly female occupations are paid less than gender integrated or predominantly male occupations, which puts all of those at disadvantage who work in a “female sector” at a disadvantage. Evidence for devaluation exists for various developed economies (United States, UK, Germany, Sweden, Netherlands, Spain, Switzerland). Moreover, recent advances have applied panel data methods to cope with unobserved heterogeneity on the individual or occupational level, and to shed light on change over time (England, Allison, and Wu 2007; Levanon, England, and Allison 2009; Murphy and Oesch 2016). Estimations of the devaluation effect have shown to be robust when being tested against diverse competing hypotheses coming from human capital or power resources theory. Overall, the existence of devaluation hence comes close to being an established fact and it is a prominent feature in explanations of wage inequality between occupations. While empirical results thus speak in favor of the devaluation hypothesis, there is persistent insecurity about its underlying causal mechanisms. Even though cross-sectional data shows relatively large associations between the proportion of female workers (from here on: percent F) and wages, panel data models using occupational fixed effects can hardly explain variation in occupational wages (e.g., England, Allison, and Wu 2007). Hence, the question emerges of what mechanisms drive this large cross-sectional association, if it is not the assumed relationship between occupational feminization and wages. Moreover, there have been several findings of heterogeneous effects over time (De Ruijter, van Doorne-Huiskes, and Schippers 2003; Levanon, England, and Allison 2009; Mandel 2013) and countries (Grönlund and Magnusson 2013; Hausmann, Kleinert, and Leuze 2015; Murphy and Oesch 2016), for which mere ad hoc explanations could be offered. Devaluation hence seems to be embedded in time- and space-specific contexts. Given the lack of research addressing these contexts, authors have consequently called for a “refinement of the theoretical discussion” and for putting more emphasis on “the search for mechanisms” (Grönlund and Magnusson 2013, p. 1015).1 More to the point, the devaluation hypothesis seems to be undertheorized, which undermines our ability to understand devaluation from a more sociologically founded perspective. Tapping into this debate, we argue that the relationship between occupational feminization and devaluation is governed by two main mechanisms. The first mechanism is the link between feminization (increases in the share of women) and the formation of occupational gender stereotypes.2 This is the process in which occupations are attributed a gender label, conditional on the share of women within that occupation. The second mechanism is the interaction of occupational stereotypes with a cultural bias that attributes a lower value to female labor, which then results in devaluation. As a consequence, changes in occupational gender compositions are linked with devaluation conditionally on the cultural context defined by said two mechanisms. Adding ideas on how these mechanisms could play out differently for men and women, and between higher- and lower-pay occupations, we derive several hypotheses about the effect heterogeneity in the devaluation process. Testing these hypotheses with census data from the United States ranging from 1960 to 2010, we show that the effect of feminization on wages via devaluation has been restricted mostly to the bottom 80 percent of occupations between 1960 and 1980, where mostly women received a penalty. Yet this is not true for the highest-paying occupations, where men have even received rewards after sectoral feminization. Overall, we observe significant declines in the devaluation effect from 1960 to 2010, to which our theoretical model offers two competing explanations: On the one hand, the spread of egalitarian gender norms and the equal pay movement could have undermined the bias that attributes less value to female work. This way, devaluation could be a phenomenon of the past. On the other hand, if unmeasured occupational gender stereotypes would have become resistant to underlying changes in the labor market’s gender composition, then the declining effects could simply be an artifact of a deficient testing strategy that solely identifies changes in the gender composition, neglecting stability in stereotypes, on which devaluation is truly based. This latter argument is supported by our finding that past levels of percent F are equally good predictors of current wages as more recent percent F, showing that occupations that previously had a large proportion of female workers are still lacking behind in wages, even though their gender composition has changed in the meantime. Keeping in mind that previous findings as well as ours do not provide enough evidence to adjudicate between those two competing stories, we strongly suggest that causal narratives with respect to the devaluation hypothesis need to be made with much care toward both theoretical considerations and modeling strategies. The following section (Devaluation so far) shortly summarizes the standard devaluation model and previous empirical results on the topic. Thereafter, in the Constructive criticism section, we propose a theoretical model and we discuss mechanisms and the potential for heterogeneous effects. We then formulate several hypotheses about effect heterogeneity over time, genders, and occupations. Following, we present the data and our empirical results, that we further embed in a broader context in the Discussion section. Devaluation so far The devaluation hypothesis has emerged as a theoretical instrument to explain the gender wage gap and to shed light on inequality between occupations. As such, this framework could be seen as complementary to more conventional approaches to the wage function, for instance building on human capital theory. At the same time, scholarly research on devaluation has received objections toward its empirical testing strategies, which has resulted in much of the field’s resources being focused on identifying a robust devaluation effect, net of potential confounding and selection mechanisms. First, reverse causality has been suggested by the crowding hypothesis following Bergmann (1974), where decreasing wages could follow if women are discriminated in higher-paying and more prestigious occupations by male employers. Second, if male occupations were protected better by closure strategies (Bol and Weeden 2015; Weeden 2002), then the correlation between percent F and wages could be spurious. Third, neoclassic economic theory suggests that in particular the division of labor in the traditional household could result in women’s lower investment in high-return, specific labor market skills, which would explain why occupations with a higher share of women pay less in general.3 On the empirical side, there is a relatively rich set of findings on devaluation from different countries, the majority of which support the devaluation hypothesis at least to some extent. For the United States, studies consistently show that—net of confounding factors—if a completely male occupation hypothetically becomes completely female, this would result in a wage penalty of 4 to 9 percent for women and a penalty of 9 to 19 percent for men (England et al. 1988; England, Allison, and Wu 2007; Gerhart and El Cheikh 1991; Karlin, England, and Richardson 2002; Kilbourne et al. 1994a, 1994b; Levanon, England, and Allison 2009; Macpherson and Hirsch 1995).4 In the most recent of these studies, Levanon, England, and Allison (2009) for instance find that female-typed occupations’ wages were 6 to 10 percent below those of mixed occupations. Evidence from other countries than the United States also speaks in favor of the devaluation hypothesis, where similar effect types are estimated to be in the range of 3 to 13 percent.5 Not all of these articles’ methods perform equally well in controlling for confounders and selection mechanisms, but many are good in at least addressing one or two of the mechanisms listed above. Piecing them together, we know that devaluation is very likely not a phantom, and that it exists despite variation in human capital requirements (Murphy and Oesch 2016; Perales 2013),6 institutional barriers (Bol and Weeden 2015; Murphy and Oesch 2016), and potential crowding (De Ruijter, van Doorne-Huiskes, and Schippers 2003; Karlin, England, and Richardson 2002; Levanon, England, and Allison 2009).7 Constructive criticism Whereas the literature is rich in studies showing that female occupations are being devalued, we know relatively little about how devaluation really works. The following discussion is built around a theoretical model that helps shed light 1) on the mechanics behind devaluation; and 2) on the plausibility of heterogeneous effects over time, between different types of occupations, and between genders. Causal effects and mechanisms The basic premise of the devaluation hypothesis is that occupational wages and prestige drop with an influx of female workers. This, in turn, leads to social inequality between occupational groups. In figure 1, this relationship is described through link (a). However, research has remained surprisingly silent with regard to the mechanisms that produce this effect. In the most general way, previous views on devaluation have assumed that “gendered cultural beliefs ... portray men as more competent and status-worthy than women,” and that “the value assigned to work in different occupations depends on the characteristics of the occupations’ incumbents” (Levanon, England, and Allison 2009, p. 868). Or, as England, Allison, and Wu (2007, p. 1238) put it: “Somehow, the low status of women ‘rubs off’ on employers’ evaluation of the occupation, and they set a lower pay level for both men and women in the occupation.” Yet, this “somehow,” as important as it seems to be in this process, has received far too little attention in the past, which is why we lack a better procedural understanding of how occupations are stigmatized and devalued in the first place. Figure 1. View largeDownload slide A theoretical model of occupational devaluation due to feminization Figure 1. View largeDownload slide A theoretical model of occupational devaluation due to feminization We hence propose a theoretical model to account for the process underlying female occupational devaluation. This model includes two main mechanisms that are believed to govern the way in which feminization leads to devaluation, and ultimately to between-occupation wage inequality. The first mechanism is the formation of occupational gender stereotypes, given the share of women within occupations (link x in figure 1). This is the step in which an individual characteristic—being female—is ascribed to the entirety of an occupational group. This has so far been widely ignored in the literature, even though it seems like an absolute necessity if we are talking about the gender trait of a group that almost always includes mixed genders. In research, the notion “female occupation” is often used as a reference to both occupations that have a lot of female incumbents and occupations that are seen as “typically female” in public perception. We believe that this is sufficiently imprecise, as it conflates an objective demographic characteristic of a group (the gender composition) with people’s view on the typical representative of this group (the stereotype). In the long run, these can be two very different things, as demographic accounts are updated every time a person joins or leaves the occupation, while stereotypes often remain sticky even though they might have become a very inaccurate representation of reality since the time they emerged (see also discussion under the section “Change over time”). The second mechanism is the interaction of gender stereotypes with a cultural bias against the value of female labor, leading to devaluation. This link (labeled y in figure 1) follows the intuition that devaluation can only emerge if the perception of an occupation to be female is paired with the judgment that women’s work is worth less than men’s work. The idea that male and female work can be differently valued due to cultural norms has previously been featured, for example in work on the drivers of segregation (Charles and Grusky 2004; Levanon and Grusky 2016) and in status characteristics theory (Berger et al. 1977), where the bias we describe would be understood as a “status belief” (see also Ridgeway and Correll [2004]). We pick up this idea and apply it to the framework that explains how occupational wages are generated. One important distinction between the two variables we include in our model (differential valuation of work and stereotypes) is that cultural norms, which define the intrinsic worth of male and female labor, are not tied to specific jobs, as they operate on a broader societal level. To the contrary, we are interested in stereotypes only insofar as they are tied to specific occupations. This distinction also makes clear why we believe that the two variables can vary independently from each other: It is, for example, not given that the spread of egalitarian gender norms, which would result in the equalization of the perceived worth of male and female work, would automatically diminish the extent to which society views specific occupations as typical for one or the other gender. We discuss this claim in more detail in the next section, where we elaborate on how such changes over time might have occurred in the United States over the past several decades. Change over time In the conventional view on devaluation, when female occupational shares change, this also affects wages. But has this link been constant over time? The theoretical model indicates that for the effect to be constant, the core mechanisms leading to devaluation must stay in place. Yet, the research suggests that both the formation of stereotypes and cultural valuations of female work are embedded in historical contexts that have changed significantly at least since the immediate postwar era. The valuation of female work and its changes The differential valuation of male and female work is well documented for the United States. Lalive and Stutzer (2010) present a review of evidence based on vignette studies indicating the existence and extent to which such differential valuations persist. In one study, for instance, Jasso and Webster (1997) find that the female-to-male ratio of perceived fair pay is 0.85 for men and 0.88 for women. These beliefs imply that it would be fair to pay women more than 10 percent less in wages, all else being equal. The systemic relevance of these beliefs is demonstrated by the intriguing findings that both men and women throughout societies have similar biases that attribute less value to women in general and to female labor specifically (Eagly, Wood, and Diekman 2000; Lalive and Stutzer 2010). On the bright side for the proponents of the equal pay movement, there is little doubt that the support for gender equality has significantly risen over the second half of the past century (Bolzendahl and Myers 2004; Charles and Grusky 2004; Mason, Czajka, and Arber 1976). This is true in a wide array of life domains, including work and family responsibilities, where women have traditionally been expected to care for the household. While this can be seen as progress in various ways, it is also widely known that such progress comes at a slow pace: Cultural views on gender are learned in early life stages, while they prove to be relatively stable throughout the life course (e.g., Alwin and Krosnick 1991). This is why generational change is often a prerequisite for wider cultural shifts.8 Furthermore, gender systems are built on firmly institutionalized beliefs, and unequal distributions of resources help maintain those systems (Ridgeway and Correll 2004). Without major external shocks, it thus takes decades for culturally induced inequality to vanish. Occupational stereotypes: changes and resilience Evidence on occupational stereotypes in the United States is not abundant, mainly because they are difficult to measure, but we nevertheless know a lot about them. The first extensive surveys found that the most male-typed occupations were, for example, miner, construction worker, or engineer.9 Female-typed occupations were manicurist, nurse, or secretary, among others (Albrecht, Bahr, Chadwick 1977; Panek, Rush, Greenawalt 1977; Shinar 1975). This all sounds familiar, but it is less clear if stereotypes have remained stable over time, given the sometimes strong changes in the gender composition within these occupations. White et al. (1989) find that there have been significant changes in stereotypes between the mid-1970s and the late 1980s. By that time, occupational stereotypes had become less strongly perceived, although they were still an important feature of how people thought about occupations. Other findings suggest that compositional changes in occupations also result in adapted stereotypes (e.g., Diekman and Eagly 2000). This makes sense in particular if we accept the premise of sex role theory (Eagly 1987) that the roles in which people act on a daily basis shape their ideas about their appropriate place in society. To the contrary, several studies show that despite desegregation in various socio-economic domains, female stereotypes and the gender stereotypes attached to occupations have remained surprisingly stable (see Lueptow, Garovich-Szabo, and Lueptow [2001] for a meta-study). Although people nowadays avoid alluding to stereotypical ideas when being asked about them, it is plausible that stereotypes are hardwired in our brains to the extent that implicit processes still bias our thoughts and actions along the lines of traditional gender views, without us knowing about it (White and White 2006).10 This line of thought is also reminiscent of the evidence on the stability of gender essentialism, the belief that women and men have innately different abilities and interests that make them more suitable for one or the other occupation (Charles and Grusky 2004; Levanon and Grusky 2016). Indeed, essentialist views, such as the belief that women are particularly suited for caring activities, do not only result from social role-derived knowledge, but they are also based on more deep-rooted biological claims (Epstein 2007), religious beliefs, traditionalism, and political views (Knight and Brinton 2017). Summary of ideas In state-of-the-art research, correlations between occupational gender compositions and wages, all else equal, are interpreted as evidence for the devaluation of female occupations. The previous discussion has its merit in defining the potential mechanisms that must be in place for this interpretation to be valid when we look at variation over time. More specifically, we see two scenarios in which declining correlations could be indicative of two very different processes: First, if the spread of egalitarian norms undermines the bias that justifies the differential valuation of male and female work, then declining effects could be interpreted as a decline of devaluation altogether. On the other hand, with a stability of occupational stereotypes (despite shifts in gender compositions), declining correlations could be viewed as an artifact of our focus on gender compositions, whereas the correlation between stereotypes and wages persists.11 Variation between occupations and gender Our understanding of devaluation as an essentially cultural phenomenon enables us to appreciate its dynamics over time. In this section, we argue that in addition to that, some occupations might be more prone to devaluation than others: We look at the occupations’ rank in the wage distribution, speculating that higher-paying occupations might be more susceptible to devaluation. Moreover, zooming in from the perspective on inequality between occupations to broadening gaps within groups, we discuss the idea that men and women, both working in the same feminizing occupation, might not receive the same penalty, and that either group might even profit from post-feminization wage adjustments. Several studies have pointed out that the wage penalty in female occupations varies between different occupational types or “classes.” De Ruijter, van Doorne-Huiskes, and Schippers (2003) show that in the Netherlands, the share of female workers is a stronger negative correlate of wages in occupations with high education and skill requirements.12Mandel (2013) produces evidence for the United States (1970 to 2007), suggesting that the feminization effect was strongest in high-wage male-typed occupations. In general, the intuition so far has been that this group of occupations is most susceptible to devaluation because it has the highest status attached to it and thus it has the “most to lose” from an entry of women who are seen as low-productivity, low-status workers (ibid., pp. 1187ff). Furthermore, occupational gender stereotypes may not have the same inherent relevance for both genders. This means that feminization could play out quite differently for male and female workers in the same occupation. In fact, due to a methodological byproduct in part of the devaluation literature,13 we already have evidence on such variation. Previously, the wage penalty after feminization was often found to be larger for men than for women (England et al. 1988; England, Allison, and Wu 2007; Gerhart and El Cheikh 1991; Macpherson and Hirsch 1995). Yet, such evidence has not been complemented with much theoretical accounts that could explain it. A more recent study furthermore does not find large effect differentials between genders (Levanon, England, and Allison 2009). We offer, once again, two competing views on this problem: Williams’s (1992) work suggests that, due to the male stereotype of high competence and leadership qualities, men are more prone to be selected into supervisory and managerial positions when working in predominantly female work environments. Hence, it might be to the advantage of male workers when the sector they work in feminizes, as this increases their propensity to move up in the wage hierarchy. Importantly, such promotions can occur within one and the same occupation, so that these changes would be observed between genders and not between occupational groups. Even though this “glass escalator” phenomenon might only partially apply in today’s world of work (Williams 2013), it strongly suggests that men’s wages might be less negatively affected by feminization or that they even might receive a net benefit from it. To the contrary, status characteristics theory (Berger et al. 1977) offers the view that a general status characteristic (here: being male or female) can lose part of its relevance in situations where a specific characteristic (working in a male or female task setting) is more salient. In other words, although men might be seen as more competent than women in general, this must not be true in settings requiring skills that are typically seen as being in the domain of female work (see also Busch [2013]). Indeed, the ensuing specific status expectations could tip the scales in favor of women in female and feminizing sectors, while men would profit from working in roles that are typically seen as more suitable for their specific skills and abilities. Hypotheses The theory section allows us to derive hypotheses about how the effect of feminization on occupational wages has shaped up over time and about how it varies between different groups. H1.1: Increasing female shares in occupations have constantly led to a decline in wages since the postwar era. Constant effects are plausible if no change has occurred in the way in which feminization is translated into stereotypes, and if those stereotypes interact with a constant cultural bias that attributes less value to female work. This would undermine vast empirical evidence that the United States has experienced a shift toward egalitarian gender norms. H1.2: The negative effect of increasing female shares in occupations on wages has slowly declined since the postwar era. This hypothesis is motivated through two conflicting points of view: First, a net decrease in the effect would be given if the rise of egalitarian norms outweighs any change in the importance of stereotypes. The slow pace of such change would be the result of the stability of cultural norms within cohorts that precludes fast cultural revolutions. Second, if unmeasured stereotypes have become stable and resistant to changes in the occupational gender structure, then we would also observe declining effects, even though devaluation could persist through the correlation between stereotypes and wages. H2: Declines in wages due to increases in female occupational shares are located primarily in the high-paying sectors. It is possible that high-status, high-wage occupations are most susceptible to an influx of lower status (female) workers. However, this theoretical idea is not very well developed and it might require further refinement. H3.1: After an occupation experiences an increase in its female share, only women receive a wage penalty in this group, while men’s wages are not negatively affected. This hypothesis is about additional wage inequality within occupations. If men can indeed profit from their stereotypical perception as better leaders and higher-competence workers, then they are more likely to be propelled into managerial and more specialized positions (within the same occupation) when an increasingly large proportion of their occupation becomes female. Hence, men would not be negatively affected and they even could profit after occupational feminization. H3.2: After an occupation experiences an increase in its female share, men’s wages are more negatively affected than women’s wages. If views on the suitability to fulfill job requirements are based on occupational skills (rather than generic skills), then men could be evaluated less fit to perform in feminizing occupations. H4: Current wage levels are determined by earlier levels of feminization, not by current levels. It is not clear if stereotypes are updated with changes in the group’s gender composition, or if they are established at a given point in time, remaining relatively constant thereafter. If the latter is true, then the stereotypes of the past would still govern wage determination processes at later stages, even if occupational segregation has declined. Data and Methods Data are drawn from the Integrated Public Use Microdata Series (IPUMS) (Ruggles et al. 2015). The sample used for this study includes six waves of US census data and ranges from 1960 to 2010, where data are available every 10 years. Data from 1950 are available but excluded, as per methodology, this year offered a far too small number of occupational observations (see clustering method below). In order to generate a quasi-panel data set from the micro-level census files, I aggregate data using occupation-industry groups for men and women, respectively. The key variables are the share of women (measured as percent of women at a given point in time) and the median wage within each group (log-transformed). As control variables, I include the mean years of education, mean of potential labor market experience,14 mean working hours, percent working in the public sector, percent ratios of ethnic minorities/non-Hispanic white, percent married, percent with young children in the household, and the relative size of the group within the labor market. The latter variable is used to control for structural economic change that influences the demand for workers in the given sector.15 Whereas most of the variables are derived by gender, this is not the case for female shares, ethnic compositions, and the relative size of groups, as their influence on wages is believed to be located purely at the occupational level. The occupation-industry clusters are derived from the IPUMS’s SOC1990 and IND1990 variables that sort individuals into harmonized occupational and industrial categorizations.16 The method used for this harmonization is backward and forward coding. As a result, some occupations that are observed in 1990 are not observed in earlier or later decades. Levanon, England, and Allison (2009) use the same data and attempt to solve this problem by merging occupations containing no observations in some cells (over years) with similar occupations. If this is done correctly, this could increase the precision in the median and mean values per occupation. However, this step potentially also introduces bias, as it might conflate occupational groups that are different in their female proportion or wage determination mechanisms. We thus do not merge any cells. However, we drop occupation-industry clusters for which we have below 20 observations for either gender. This leaves us with 1,455 occupation-industry groups that are observed on average 3.9 times. With an average of 3.9 observations per group, there are some groups that are not observed the maximum of 6 times. If possible, we also tested all models excluding groups with missing year-cells, which left the results virtually unchanged. See table 1 for the total N within each wage quintile and period cell. Table 1. Total N of Observed Occupation-Industry Cells by Quintile and PeriodQuintile   Year  1960  1970  1980  1990  2000  2010  1st  127  126  231  269  253  170  2nd  104  128  239  281  249  154  3rd  100  126  242  278  258  154  4th  90  103  241  277  258  160  5th  60  79  228  277  263  195    Year  1960  1970  1980  1990  2000  2010  1st  127  126  231  269  253  170  2nd  104  128  239  281  249  154  3rd  100  126  242  278  258  154  4th  90  103  241  277  258  160  5th  60  79  228  277  263  195  Source: IPUMS USA 1960–2010. The aggregation of occupation-industry data is based on a restricted sample, dropping individuals not at work, having zero income, and being younger than 25 or older than 65. Whereas some studies only include full-time employees in the sample (e.g., England, Allison, and Wu 2007), we believe that this could artificially distort the measured profile of occupations, as women are particularly prone to working less than full-time. Consequently, a control for average working hours is included in order to avoid bias due to potential higher returns in jobs with large volumes of hours worked. Statistical tests of all but the last hypothesis will be based on the fixed effects method (Allison 2009; Wooldridge 2010). This method allows us to exploit the quasi-panel structure of the data set and draw all inference from variation within the occupation-industry groups over time. The fixed effects models difference out constant heterogeneity between group-level clusters. In order to control for the potentially large changes in period-specific average wages from 1960 until today, I include binary time variables for each year. Interacting these period effects with the gender composition variable allows us to test effect heterogeneity over time (Allison 2009, pp. 19ff). Panel robust standard errors are employed to correct for serial correlation (Wooldridge 2010). The last hypothesis (H4) is tested via standard OLS models of 2010 occupational wages. Here, wages are regressed on current percent female, and subsequently on the same variable in 1980, testing the predictive power of both approaches. Results In table 2, we observe the results of male and female fixed effects regressions, testing the hypothesis that feminization leads to a decline in occupational wages. “P.c. female” is the main effect, and all included control variables are displayed in the table. In this first step, feminization is assumed to be constant, which leaves us with a single coefficient for the male and female wage regression, respectively. The male model results in a highly significant positive effect, whereas the effect for women is close to zero and not significant. We do not find a negative average effect, as predicted within the devaluation framework.17 Substantively, the effect implies a 0.12 percent raise in median wages for men with every percent increase of women in an occupation. The within-occupation standard deviation from 1960 to 2010 is 8.23, which leaves us with a standardized effect of 1 percent per unit change in percent female. The hypothetical change from zero to 100 percent women in an occupation would be expected to result in a wage increase of 12.75 percent for men. Table 2. Coefficients (t-values in parentheses) from Fixed Effects Regressions of Occupational Log Wages   Male wages  Female wages  Coefficients  t-values  Coefficients  t-values  P.c. female  0.0012***  (4.23)  0.0004  (1.49)  1960 (ref.)           1970  0.2137***  (26.94)  0.1632***  (20.87)   1980  0.1740***  (12.33)  0.0114  (1.06)   1990  0.1210***  (8.03)  −0.0231  (−1.84)   2000  0.1103***  (6.54)  0.0009  (0.06)   2010  0.0958***  (4.82)  0.0037  (0.20)  Working hours  0.0050*  (2.34)  0.0076***  (4.85)  Public sector  −0.0013**  (−2.82)  0.0000  (0.11)  Ratio Black  0.0042  (0.17)  0.0311  (0.98)  Ratio Native  −0.4976***  (−2.27)  −0.5567**  (−2.83)  Ratio Asian  0.2316**  (2.63)  0.2245**  (3.25)  Ratio Hispanic  −0.1018*  (−2.45)  −0.0923**  (−2.95)  Years edu.  0.0753***  (14.60)  0.1063***  (14.29)  LM exp.  0.0403***  (7.11)  0.0281***  (5.47)  LM exp.2  −0.0006***  (−5.01)  −0.0005***  (−4.18)  P.c. married  0.0070***  (16.68)  0.0017***  (4.60)  P.c. child  0.1181**  (2.81)  0.2364***  (4.75)  Sector size  0.0096  (0.97)  −0.0031  (−0.25)  Constant  0.3442*  (2.51)  0.3183**  (2.63)  Ntotal  5,720  5,720  Ngroups  1,455  1,455    Male wages  Female wages  Coefficients  t-values  Coefficients  t-values  P.c. female  0.0012***  (4.23)  0.0004  (1.49)  1960 (ref.)           1970  0.2137***  (26.94)  0.1632***  (20.87)   1980  0.1740***  (12.33)  0.0114  (1.06)   1990  0.1210***  (8.03)  −0.0231  (−1.84)   2000  0.1103***  (6.54)  0.0009  (0.06)   2010  0.0958***  (4.82)  0.0037  (0.20)  Working hours  0.0050*  (2.34)  0.0076***  (4.85)  Public sector  −0.0013**  (−2.82)  0.0000  (0.11)  Ratio Black  0.0042  (0.17)  0.0311  (0.98)  Ratio Native  −0.4976***  (−2.27)  −0.5567**  (−2.83)  Ratio Asian  0.2316**  (2.63)  0.2245**  (3.25)  Ratio Hispanic  −0.1018*  (−2.45)  −0.0923**  (−2.95)  Years edu.  0.0753***  (14.60)  0.1063***  (14.29)  LM exp.  0.0403***  (7.11)  0.0281***  (5.47)  LM exp.2  −0.0006***  (−5.01)  −0.0005***  (−4.18)  P.c. married  0.0070***  (16.68)  0.0017***  (4.60)  P.c. child  0.1181**  (2.81)  0.2364***  (4.75)  Sector size  0.0096  (0.97)  −0.0031  (−0.25)  Constant  0.3442*  (2.51)  0.3183**  (2.63)  Ntotal  5,720  5,720  Ngroups  1,455  1,455  Source: IPUMS USA 1960–2010. *p < 0.5 **p < 0.1 ***p < 0.01 This finding is interesting in itself, including some initial evidence that men are favored in the wake of occupational feminization (H3.1). However, assuming a constant feminization effect on wages is not satisfactory given our theoretical model. In a further step, I hence decompose the previous results by period and occupational wage quintiles. Figure 2 depicts marginal effects of feminization over the entire time span for each quintile. Estimations include identical control variables, but marginal effects are derived from including an interaction term between P.c. female and periods. Furthermore, the sample of 1,455 occupation-industry groups was split in quintiles according to average wages in each group over the full time span. Figure 2. View largeDownload slide Marginal effects of feminization on wages, by occupational wage quintiles and gender (1960–2010) Figure 2. View largeDownload slide Marginal effects of feminization on wages, by occupational wage quintiles and gender (1960–2010) The results suggest that feminization can cause occupational devaluation. Seen in historical context, however, this relationship has been tied to specific periods (1960s–1980s), was stronger for women than for men, and was rather located in the second and fourth wage quintile. These findings are partly in line with our ideas on effect heterogeneity and the theoretical model, but they are also more diffuse than we expected. I summarize in three points: First, effect sizes are relatively similar between all wage groups, excluding the very top quintile. In quintiles one to four, effect sizes are mostly either negative or around zero (in earlier decades), becoming slightly positive in more recent years. Negative coefficients, which are evidence for the predicted wage effect via feminization, are largest in the fourth quintile. In the fifth, however, we find exclusively positive wage effects, for both women and men. As such, wage developments at the top of the income ladder are related to feminization in a very different way, as it is the case for all remaining groups. At the top, positive effect sizes for men range between 0.19 percent per unit increase in P.c. female in 1980, and 0.53 percent in 1970. This partial view also explains the positive wage effect found in the previous step: Replicating the male model shown in table 2 excluding the fifth quintile, the previously significant positive effect comes very close to zero and is non-significant. Taken together, this undermines the plausibility of H2, according to which devaluation would be located primarily in the top sectors. In the highest quintile in 1960 and 1970, confidence intervals are relatively large for both genders, which is due to the lower N of observations within these period-cells (see also table 1). The magnitude of the errors makes our analysis less reliable for these years, but an estimation of similar models including only groups without any missing period-cells does not lead to any substantively different results (not shown here). Table 3. Correlations of Percent F and Current/Subsequent Occupational Log Wages   p.c. F1960  p.c. F1970  p.c. F1980  p.c. F1990  p.c. F2000  p.c. F2010  ln(wages)1960  −0.45            ln(wages)1970  −0.49  −0.50          ln(wages)1980  −0.54  −0.55  −0.56        ln(wages)1990  −0.44  −0.45  −0.45  −0.39      ln(wages)2000  −0.36  −0.37  −0.36  −0.30  −0.27    ln(wages)2010  −0.31  −0.32  −0.31  −0.24  −0.20  −0.17    p.c. F1960  p.c. F1970  p.c. F1980  p.c. F1990  p.c. F2000  p.c. F2010  ln(wages)1960  −0.45            ln(wages)1970  −0.49  −0.50          ln(wages)1980  −0.54  −0.55  −0.56        ln(wages)1990  −0.44  −0.45  −0.45  −0.39      ln(wages)2000  −0.36  −0.37  −0.36  −0.30  −0.27    ln(wages)2010  −0.31  −0.32  −0.31  −0.24  −0.20  −0.17  Source: IPUMS USA 1960–2010. Second, effect sizes are relatively similar between genders and most trends seem to apply to both men and women. We implement z-tests, following Clogg, Petkova, Haritou (1995), confirming that in every single period-quintile cell except one, male and female coefficients are not significantly different from each other with p < 0.5 (results not shown). In figure 2, we yet also observe that when devaluation existed, it was mostly impacting the wages of women. This is clearest in the second and fourth quintiles, where women were affected by feminization the most. Another twist is observed in the top wage group: Here, the effect on male wages is positive and significant over the entire range of years, while this is the case for women only in two periods. The lack of more significant female coefficients is because of both smaller effect sizes and wider confidence intervals, describing more variation in wages overall. Evidence thus speaks more in favor of H3.1 (men are losing less and they are gaining more after feminization), while H3.2 finds no support at all. Third, on visual inspection of the marginal effects, we would believe that the devaluation effect has been decreasing in time at least for the second, third, and fourth wage quintiles. This trend seems to be following a relatively linear pattern for women, while it is bumpier for men. Figure 3 depicts results of a pairwise comparison of marginal effects in 2010 and 1960, by wage quintile and gender. The positive and statistically significant differences for men and women in quintiles 2–4 suggest that effect sizes have in fact decreased over time. The largest significant decline in the feminization effect is observed for women in the fourth quintile (0.36 points difference), and the smallest one is observed for men in the third quintile (0.12). According to this analysis, no significant change happened in the highest- and lowest-paying sectors. Evidence thus speaks—at least for part of the wage distribution—in favor of H1.2: The effect of occupational feminization on wages has slowly decreased over time. Figure 3. View largeDownload slide Differences in marginal effect sizes, 2010 and 1960, by wage quintile and gender, and 95 percent confidence intervals Figure 3. View largeDownload slide Differences in marginal effect sizes, 2010 and 1960, by wage quintile and gender, and 95 percent confidence intervals Finally, we address the question of whether stereotypical gender views of the past are still affecting current wages. Here, I employ a strategy using the full sample, including male and female observations, and occupational median wages as a dependent variable. This is equivalent to the methodology in cross-sectional studies that use OLS wage regressions to quantify the adverse effect of working in a female occupation. Table 3 lists cross-correlations of P.c. female and occupational median log wages. The bold figures in the diagonal show the correlation of current levels. A striking feature of the results is that the correlation between P.c. female and wages was strongest in the 1970s and 1980s, while it dropped severely in 2000 and 2010. This result is fostering our previous assertion that the effect of feminization on wages has significantly declined over time. Moreover, we observe that the female percentages of previous decades are much stronger correlates of 2010 wages than the 2010 percentages themselves: corr(wages2010,p.c. F1970) = 0.32 vs. corr(wages2010,p.c. F2010) = 0.17. This finding would support the hypothesis that past gender segregation is still influencing the current pay levels in occupations, even though the distribution of male and female workers over occupations has changed (H4). Table 4. Coefficients (t-values in parentheses) from OLS Regression of 2010 Occupational Log Wages   Only:p.c. F2010  Only:p.c. F1980  Added:Education  Added:Other controls  p.c. F2010  −0.0036***    −0.0046***    −0.0017***      (−5.98)    (−15.65)    (−6.45)    p.c. F1980    −0.0058***    −0.0043***    −0.0017***      (−11.22)    (−16.15)    (−7.11)  Education      0.1946***  0.1822***  0.1583***          (48.56)  (45.60)  (23.17)    N  736  736  736  736  736  736  Adj. R2  0.0451  0.1452  0.7733  0.7769  0.8768  0.8783    Only:p.c. F2010  Only:p.c. F1980  Added:Education  Added:Other controls  p.c. F2010  −0.0036***    −0.0046***    −0.0017***      (−5.98)    (−15.65)    (−6.45)    p.c. F1980    −0.0058***    −0.0043***    −0.0017***      (−11.22)    (−16.15)    (−7.11)  Education      0.1946***  0.1822***  0.1583***          (48.56)  (45.60)  (23.17)    N  736  736  736  736  736  736  Adj. R2  0.0451  0.1452  0.7733  0.7769  0.8768  0.8783  Source: IPUMS USA 1980, 2010. *p < 0.5 **p < 0.1 ***p < 0.01 After the implementation of a multivariate OLS regression of 2010 log wages, this initial judgment can be put into perspective. In table 4, I depict the results of six models, where three models use current P.c. female as predictor and the remaining three models include P.c. female from 1980.18 The first two models, including only our main independent variable, simply replicate the findings from the cross-correlations table where the 1980 percentages are a much stronger predictor of 2010 wages levels than its 2010 equivalents. However, the subsequent two models critically show that, holding constant the educational levels in occupations, the two correlations with P.c. female are approximately level. This suggests that the initial finding is biased by a selection mechanism that has changed in the period between 1980 and 2010: Nowadays, women are much more likely to select into high-education, high-paying jobs, while this was not the case in 1980 (cf. Mandel 2013). In the two last models, I include all remaining control variables, which lowers the effect of percent female in both instances, but this does not change the overall evaluation that both current and past levels of feminization are similarly good predictors of current wage levels. This finding can be interpreted as partly confirming H4. Discussion Previous results showing that in the United States occupational feminization leads to devaluation were confirmed in this study. Yet, this is only true if we look at sometimes relatively specific contexts in which devaluation seems to play a larger role than in others. Relying on a single coefficient for the United States of the past fifty or sixty years is not nearly enough to capture the complex and culturally contingent mechanisms that produce devaluation. This was shown by decomposing the average feminization effect by periods, by occupational wage groups, and by gender. This attempt, especially when added up with similar research from the past, should caution following studies against ignoring effect variation, especially when describing a phenomenon working through the channels of cultural bias and stereotyping. The latter are neither stable over time, nor do they always affect different subgroups in the population in the same way. We have observed a decline in occupational devaluation in the middle part of the wage ladder. This decline seems to have taken place as part of a slow and steady development from 1960 to 2010, which fits well with our theoretical expectations that incremental cultural change toward gender egalitarian norms would have slowly eradicated devaluation. However, if we believe the theoretical model, then we should be cautioned that the declining effect over time could also be due to an increasing stability in gender stereotypes. In such a scenario, ongoing change in the occupational gender structure would have less or no effect on stereotypes, which would logically weaken the observed correlation between feminization and wages, even though devaluation would persist. This interpretation finds some evidentiary support in our results showing that previous levels of occupational segregation are about equally good predictors of today’s wages as current segregation levels. As this paper does not offer a more refined strategy to isolate one causal narrative from the other, we yet caution the reader not to derive a strong causal claim from our findings. We clearly require further tests of how the link between feminization and wages has shaped up over time. Currently, we see two ways to address this issue: We either build statistical models in which we test the composite effect of current and past feminization at the same time. or we find a way to measure occupational stereotypes—possibly even over time—including them as a new control in our wage regressions. The former approach would possibly need to deviate from standard fixed effects models, as those measure changes in key variables that make it difficult to test any type of inertia in wage determination mechanisms. The latter approach would enable us to directly test the model’s mechanisms, but data on stereotypes by occupation is very scarce, if available at all. The theory section included some vague ideas on how the feminization effect could differ between higher- and lower-status occupations. Our intuition was that at the top of the wage ladder, occupations are prestigious and mostly dominated by men, which could make them more susceptible to an inflow of lower-status female workers (Mandel 2013). We have tested these ideas by estimating separate models by wage quintile, showing that the interesting part of the variation might in fact be located between the fifth quintile and the rest of all occupations. In the fifty years before 2010, trends were relatively similar in the bottom 80 percent of occupations. The fact that wage penalties were strongest in the fourth quintile could, however, be interpreted as some support of the high-status-susceptibility hypothesis. The most striking result yet was found for the top 20 percent: For this group, female devaluation does not seem to have existed at all. Men have profited from feminization all along, while female effects were also positive, even though rather non-significant. This is forcedly a novel finding, because—to our knowledge—previous studies have not looked at such a granularity when dividing occupations into wage or status groups. One might be tempted to think that this finding could be explained by an increasing shortage of specific skills located in high-paying jobs. Liu and Grusky (2013) demonstrate that particular analytic skills, but also computer, managerial, and social skills, have seen a rise in payoff between 1980 and 2010. However, this account would not allow to us to explain the constant rewards for men in these high-paying jobs, or the large feminization coefficients in the 1960 and 1970, where much of the skill-biased structural change would not have happened yet. At this point, we certainly require more research—including different data sources and possibly cross-country insights—to confirm this peculiarity, and to help better understand the impact of feminization on wage generation mechanisms at the top of the wage ladder.19 In the past, surveys have sometimes shown different coefficients for the devaluation of male and female wages. For the United States, there is, however, no clear tendency showing either of the genders to be more disadvantaged than the other. In the most similar study to the one at hand, Levanon, England, and Allison (2009) show very similar effect sizes for men and women, which is confirmed by our own results. Significant devaluation effects were found mostly for female wages, and in the top wage group, benefits were significant over all periods only for men. The results yet seem insufficiently skewed in favor of men to speak of any healthy evidence for the glass escalator effect. Budig (2002) has argued that previous evidence for men riding glass escalators in female occupations might be misleading due to a lack of appropriate counterfactual reasoning. In such cases, studies often focus on predominantly female sectors (such as nursing), neglecting a more meaningful comparison of results with male advantage in mixed or predominantly male sectors. Furthermore, we can also not find any evidence for Mandel’s (2013) judgment that women are “moving up the down staircase”—meaning that women have reached high-skilled jobs yet tend to experience declining wages. Even though this study was able to shed light on various under-researched features of devaluation, it also neglects an important criticism that certainly deserves more attention. Several authors have highlighted that the assumption of a linear relationship between feminization and devaluation could be too simplistic (Magnusson 2013). The most popular alternative approach is to divide occupations into clusters of predominantly male, female, and gender-mixed groups, where the event of switching to the female-dominated sector is operationalized as a main predictive variable. From a statistical point of view, this approach could be favorable due to its relaxation of the linearity assumption. Moreover, assuming that the cultural process in which occupational stereotypes are formed follows a stepwise (not continuous) adaption where certain thresholds in the share of women must be reached in order to change a prevailing stereotype, then splitting up occupations into more or less female groups might make a lot of sense. Unfortunately, the data used in the present paper does not allow to further break up occupational groups without seriously undermining statistical efficiency. Finally, studies using similar statistical models have previously been criticized for a lack of more refined controls for specific vocational training and changes in skill requirements within occupations. This is due to a trade-off where we chose very long time series over data sets that include richer background variables that are at the same time being limited to a time span in which the slow change in culturally determined processes could hardly be observed. Conclusion The idea that occupations devalue due to their feminization has received a lot of attention—not just because it challenges dominant views of how wages are determined, but also because it was backed up by ample empirical support. In this paper, we claim that if further progress is to be made on this topic, then we need to more explicitly address the causal mechanics of female devaluation. Without explaining in more detail how feminization impacts occupational wages, we will also be unable to understand why this effect seems to be heterogeneous over time, between countries, and between different occupational groups. The proposed theoretical model is an attempt to address these concerns. It captures the mechanisms by which occupational gender stereotypes are formed and through which those stereotypes interact with a cultural bias against female labor to result in devaluation. Given the very plausible changes within these mechanics over time, we have shown that feminization might be linked to devaluation in various different ways. Moreover, if stereotypes freeze at any given point in time, devaluation might be entirely decoupled from ongoing changes in gender compositions, as it would be governed by past segregation processes. In the empirical section, we have shown that feminization did result in a decline of wages in the United States from 1960 to 2010, but that the effect was restricted to sometimes very specific contexts in which they occurred. Devaluation was mostly observed in the 1960s to 1980s. Men seem less affected than women, even though this result has received only weak evidentiary support. In the top wage quintile, all of the above does not apply. Here, no devaluation could be observed in any period or for either gender. This finding undermines the notion that in particular high-status occupations would be susceptible to an inflow of lower-status (female) workers. Over and above the results on effect heterogeneity, this paper raises important questions pointing at the culturally contingent contexts in which devaluation might be embedded. This is missing in the canonical view on devaluation and—in our view—deserves more rigorous testing in future research. Notes 1 Mandel (2013, p. 1202) also urges that “further investigation is required to explicate the mechanisms underlying the over-time trends” of the percent F effect. 2 The term “occupational gender stereotype” is very central in the following debate. The general idea behind this is that there are broad societal views on what gender is typical for an occupation. This is a subjective category, which is not the same as the actual or objective gender composition of an occupation, which is usually measured in studies on devaluation. We also refer to research on sex-typing of occupations that reaches back to the 1970s (cf. White and White 2006, p. 259) and where stereotypes are defined as “unconscious habits of thought that link personal attributes to group membership” (Reskin 2000, p. 322). 3 For an excellent full-scale discussion of potential confounders and selection mechanisms, see, for example, recently Murphy and Oesch (2016). 4 Even though these are seemingly impressive figures, they also tend to inflate the importance of devaluation by assuming the irrelevant case in which an occupation switches from zero to 100 percent female. 5 See evidence for the UK (Brynin and Perales 2016; Murphy and Oesch 2016; Perales 2013), for Germany (Busch 2013; Murphy and Oesch 2016; but also Hausmann, Kleinert, and Leuze 2015; Leuze and Strauß 2016), for Sweden (Grönlund and Magnusson 2013; Magnusson 2013), and for the Netherlands (De Ruijter, van Doorne-Huiskes, and Schippers 2003), Switzerland (Murphy and Oesch 2016), and Spain (Polavieja 2008, effect non-significant). 6 Some evidence points into the opposite direction, for example Tam (1997) or Polavieja (2008). 7 Controlling for crowding is notoriously difficult to do. De Ruijter, van Doorne-Huiskes, and Schippers (2003) employ a measure of how concentrated women are to a restricted number of occupations. A clear problem with this measure is that women could be concentrated in some occupations for many reasons that are not based on discrimination of distorted preferences. Levanon, England, and Allison (2009) simply show for the United States that if wages in occupations decrease, this does trigger a larger influx of women. This is probably a better attempt of controlling for crowding, as it rules out reverse causality in the devaluation story altogether. 8 One study addressing attitudes on fair pay directly, and also within the context of generational change, is presented by Auspurg, Gatskova, Hinz (2013), who show that older cohorts in West Germany put more weight on a person’s gender when judging on fair pay in work. 9 Research on stereotypes has employed different test strategies to probe gender-typical views attached to occupations. They hence do not rely on the objective gender composition in occupations to identify female- and male-typed occupations. For an introduction of the instruments, see White and White (2006). 10 For a distinction between explicit and implicit stereotypes in this context, see White and White (2006, pp. 259ff). 11 A similar reasoning can be found in England, Allison, and Wu (2007), who discuss if inertia, particularly in the wage structure, might be a reason for the observation that feminization has not led to much change in pay levels after all. 12 Grönlund and Magnusson (2013) test if this is also true in the Swedish case but do not find any significant effects. 13 Several surveys using occupation-level data include separate statistical models for men and for women in order to control for the individual-level wage gap. 14 Following approaches in the literature, we proxy labor market experience by taking individuals’ age, subtracting their years of education, and adding 6 (the typical age where children enter school in the United States). 15 Changes in the group sizes could also reflect supply-side factors. One variant discussed in the paper is crowding, where women are shifted toward less favorable segments in the market. 16 There are different approaches to group individuals into occupation and industry clusters. The approach in this paper was previously found to deliver good results. See Levanon, England, and Allison (2009) for a thorough discussion and tests with other classifications. 17 Although our study uses a similar methodology and data than Levanon, England, and Allison (2009), we receive contrary results in this first step: They find a negative effect of changes in percent F on wages for both genders (p. 881, table 4), while our estimates are positive and significant for men but non-significant for women. In order to probe if the different findings can be explained by our methodological choices, we have attempted to fully replicate the findings by Levanon et al. This included, inter alia, the inclusion of 1950 to 2000 IPUMS data, a different treatment of occupation-industry clustering, the conversion of percent F into logits, slight differences in sample restrictions, the inclusion of lagged independent variables including a 10-year lag of logit(percent F), and the implementation of a dynamic fixed effects model correcting for endogeneity. With these specifications, we were unable to replicate the authors’ findings of negative coefficients. To the contrary, results were similar as shown in our table 2. Hence, even though it is difficult to suggest what has led to the different findings, we are confident that none of the methodological choices listed above play an important role. 18 We also could have used the 1970 or 1960 values as a comparison group, but the number of occupation-industry groups observed in 2010 and 1980 was much larger than for the combinations with earlier decades. 19 The deviation of our results by occupational group from those by Mandel (2013) are strong and underline the merit of replication studies using different data. A reason for the differences could lie in the choice of occupational clusters, as Mandel looks at under 400 occupations, while we divide the sample into over 1,400 groups altogether. See Levanon, England, Allison (2009) for the implications of these choices. About the Author Felix Busch is a DPhil candidate in sociology at Nuffield College, University of Oxford. His research interests are in social inequality, labor markets, and gender. In his dissertation, he studies the role of occupational gender segregation in the production of social inequalities. References Albrecht, Stan L., Howard M. Bahr, and Bruce A. Chadwick. 1977. “ Public Stereotyping of Sex Roles, Personality Characteristics, and Occupations.” Sociology & Social Research  61( 2): 223– 40. Allison, Paul D. 2009. Fixed Effects Regression Models . Thousand Oaks, CA: SAGE Publications. Google Scholar CrossRef Search ADS   Alwin, Duane F., and Jon A. Krosnick. 1991. “ Aging, Cohorts, and the Stability of Sociopolitical Orientations over the Life Span.” American Journal of Sociology  97( 1): 169– 95. Google Scholar CrossRef Search ADS   Auspurg, Katrin, Ksenija Gatskova, and Thomas Hinz. 2013. “ Vorstellungen von Lohngerechtigkeit in West-und Ostdeutschland und in der Ukraine.” WSI-Mitteilungen  66( 2): 77– 88. Google Scholar CrossRef Search ADS   Berger, Joseph, Hamit Fisek, Robert Z. Norman, and Morris Zelditch. 1977. Status Characteristics and Social Interaction: An Expectation-States Approach . New York: Elsevier. Bergmann, Barbara R. 1974. “ Occupational Segregation, Wages and Profits When Employers Discriminate by Race or Sex.” Eastern Economic Journal  1( 2): 103– 10. Bol, Thijs and Kim A. Weeden. 2015. “ Occupational Closure and Wage Inequality in Germany and the United Kingdom.” European Sociological Review  31( 3): 354– 69. Google Scholar CrossRef Search ADS   Bolzendahl, Catherine I., and Daniel J. Myers. 2004. “ Feminist Attitudes and Support for Gender Equality: Opinion Change in Women and Men, 1974–1998.” Social Forces  83( 2): 759– 89. Google Scholar CrossRef Search ADS   Brynin, Malcolm, and Francisco Perales. 2016. “ Gender Wage Inequality: The De-Gendering of the Occupational Structure.” European Sociological Review  32( 1): 162– 74. Google Scholar CrossRef Search ADS   Budig, Michelle J. 2002. “ Male Advantage and the Gender Composition of Jobs: Who Rides the Glass Escalator?” Social Problems  49( 2): 258– 77. Google Scholar CrossRef Search ADS   Busch, Anne. 2013. “ Der Einfluss der beruflichen Geschlechtersegregation auf den Gender Pay Gap.” Kölner Zeitschrift für Soziologie und Sozialpsychologie  65( 2): 301– 38. Google Scholar CrossRef Search ADS   Charles, Maria, and David B. Grusky. 2004. Occupational Ghettos: The Worldwide Segregation of Women and Men . Stanford, CA: Stanford University Press. Clogg, Clifford C., Eva Petkova, and Adamantios Haritou. 1995. “ Statistical Methods for Comparing Regression Coefficients between Models.” American Journal of Sociology  100( 5): 1261– 93. Google Scholar CrossRef Search ADS   De Ruijter, Judith M P, Anneke van Doorne-Huiskes, and Joop J. Schippers. 2003. “ Size and Causes of the Occupational Gender Wage-Gap in the Netherlands.” European Sociological Review  19( 4): 345– 60. Google Scholar CrossRef Search ADS   Diekman, Amanda B., and Alice H. Eagly. 2000. “ Stereotypes as Dynamic Constructs: Women and Men of the Past, Present, and Future.” Personality and Social Psychology Bulletin  26( 10): 1171– 88. Google Scholar CrossRef Search ADS   Eagly, Alice H. 1987. Sex Differences in Social Behavior: A Social-Role Interpretation . Hillsdale, NJ: Lawrence Erlbaum. Eagly, Alice H., Wendy Wood, and Amanda B. Diekman. 2000. “Social Role Theory of Sex Differences and Similarities: A Current Appraisal.” In The Developmental Psychology of Gender , edited by T. Eckes and H. M. Trautner, pp. 123– 174. Mahwah, NJ: Lawrence Erlbaum. England, Paula, Paul Allison, and Yuxiao Wu. 2007. “ Does Bad Pay Cause Occupations to Feminize, Does Feminization Reduce Pay, and How Can We Tell with Longitudinal Data?” Social Science Research  36( 3): 1237– 56. Google Scholar CrossRef Search ADS   England, Paula, George Farkas, Barbara Stanek Kilbourne, and Thomas Dou. 1988. “ Explaining Occupational Sex Segregation and Wages: Findings from a Model with Fixed Effects.” American Sociological Review  53( 4): 544– 58. Google Scholar CrossRef Search ADS   Epstein, Cynthia F. 2007. “ Great Divides: The Cultural, Cognitive, and Social Bases of the Global Subordination of Women.” American Sociological Review  71( 1): 1– 22. Google Scholar CrossRef Search ADS   Gerhart, Barry, and Nabil El Cheikh. 1991. “ Earnings and Percentage Female: A Longitudinal Study.” Industrial Relations: A Journal of Economy and Society  30( 1): 62– 78. Google Scholar CrossRef Search ADS   Grönlund, Anne, and Charlotta Magnusson. 2013. “ Devaluation, Crowding or Skill Specificity? Exploring the Mechanisms behind the Lower Wages in Female Professions.” Social Science Research  42( 4): 1006– 17. Google Scholar CrossRef Search ADS   Hausmann, Ann-Christin, Corinna Kleinert, and Kathrin Leuze. 2015. “ Entwertung von Frauenberufen oder Entwertung von Frauen im Beruf?” Kölner Zeitschrift für Soziologie und Sozialpsychologie  67( 2): 217– 42. Google Scholar CrossRef Search ADS   Jasso, Guillermina, and Murray Webster Jr. 1997. “ Double Standards in Just Earnings for Male and Female Workers.” Social Psychology Quarterly  60( 1): 66– 78. Google Scholar CrossRef Search ADS   Karlin, Carolyn Aman, Paula England, and Mary Richardson. 2002. “ Why do Women’s Jobs Have Low Pay for Their Educational Level?” Gender Issues  20( 4): 3– 22. Google Scholar CrossRef Search ADS   Kilbourne, Barbara, Paula England, George Farkas, Kurt Beron, and Dorothea Weir. 1994a. “ Returns to Skill, Compensating Differentials, and Gender Bias: Effects of Occupational Characteristics on the Wage of White Women and Men.” American Journal of Sociology  100( 3): 689– 719. Google Scholar CrossRef Search ADS   Kilbourne, Barbara, Paula England, and Kurt Beron. 1994b. “ Effects of Individual, Occupational, and Industrial Characteristics on Earnings: Intersections of Race and Gender.” Social Forces  72( 4): 1149– 76. Google Scholar CrossRef Search ADS   Knight, Carly R., and Mary C. Brinton. 2017. “ One Egalitarianism or Several? Two Decades of Gender-Role Attitude Change in Europe.” American Journal of Sociology  122( 5): 1485– 1532. Google Scholar CrossRef Search ADS   Lalive, Rafael, and Alois Stutzer. 2010. “ Approval of Equal Rights and Gender Differences in Well-Being.” Journal of Population Economics  23( 3): 933– 62. Google Scholar CrossRef Search ADS   Leuze, Kathrin, and Susanne Strauß. 2016. “ Why Do Occupations Dominated by Women Pay Less? How Female-Typical Work Tasks and Working-Time Arrangements Affect the Gender Wage Gap among Higher Education Graduates.” Work, Employment & Society  30( 1): 1– 19. Levanon, Asaf, and David Grusky. 2016. “ The Persistence of Extreme Gender Segregation in the Twenty-First Century.” American Journal of Sociology  122( 2): 573– 619. Google Scholar CrossRef Search ADS   Levanon, Asaf, Paula England, and Paul Allison. 2009. “ Occupational Feminization and Pay: Assessing Causal Dynamics Using 1950–2000 US Census Data.” Social Forces  88( 2): 865– 91. Google Scholar CrossRef Search ADS   Liu, Yujia, and David Grusky. 2013. “ The Payoff to Skill in the Third Industrial Revolution.” American Journal of Sociology  118( 5): 1330– 74. Google Scholar CrossRef Search ADS   Lueptow, Lloyd B., Lori Garovich-Szabo, and Margaret B. Lueptow. 2001. “ Social Change and the Persistence of Sex Typing: 1974–1997.” Social Forces  80( 1): 1– 36. Google Scholar CrossRef Search ADS   Macpherson, David A. and Barry T. Hirsch. 1995. “ Wages and Gender Composition: Why Do Women’s Jobs Pay Less?” Journal of Labor Economics  13( 3): 426– 71. Google Scholar CrossRef Search ADS   Magnusson, Charlotta. 2013. “ More Women, Lower Pay? Occupational Sex Composition, Wages and Wage Growth.” Acta Sociologica  56( 3): 227– 45. Google Scholar CrossRef Search ADS   Mandel, Hadas. 2013. “ Up the Down Staircase: Women’s Upward Mobility and the Wage Penalty for Occupational Feminization, 1970–2007.” Social Forces  91( 4): 1183– 1207. Google Scholar CrossRef Search ADS   Mason, Karen Oppenheim, John L. Czajka, and Sara Arber. 1976. “ Change in US Women’s Sex-Role Attitudes, 1964–1974.” American Sociological Review  41( 4): 573– 96. Google Scholar CrossRef Search ADS   Murphy, Emily and Daniel Oesch. 2016. “ The Feminization of Occupations and Change in Wages: A Panel Analysis of Britain, Germany, and Switzerland.” Social Forces  94( 3): 1221– 55. Google Scholar CrossRef Search ADS   Panek, Paul E., Michael C. Rush, and James P. Greenawalt. 1977. “ Current Sex Stereotypes of 25 Occupations.” Psychological Reports  40( 1): 212– 14. Google Scholar CrossRef Search ADS   Perales, Francisco. 2013. “ Occupational Sex-Segregation, Specialized Human Capital and Wages: Evidence from Britain.” Work, Employment & Society  27( 4): 600– 620. Google Scholar CrossRef Search ADS   Polavieja, Javier G. 2008. “ The Effect of Occupational Sex-Composition on Earnings: Job-Specialization, Sex-Role Attitudes and the Division of Domestic Labour in Spain.” European Sociological Review  24( 2): 199– 213. Google Scholar CrossRef Search ADS   Redbird, Beth. 2017. “ The New Closed Shop? The Economic and Structural Effects of Occupational Licensure.” American Sociological Review , online first. Reskin, Barbara F. 2000. “ The Proximate Causes of Employment Discrimination.” Contemporary Sociology  29( 2): 319– 28. Google Scholar CrossRef Search ADS   Ridgeway, Cecilia L. and Shelley J. Correll. 2004. “ Unpacking the Gender System. A Theoretical Perspective on Gender Beliefs and Social Relations.” Gender & Society  18( 4): 510– 31. Google Scholar CrossRef Search ADS   Ruggles, Steven, Katie Genadek, Ronald Goeken, Josiah Grover, and Matthew Sobek. 2015. Integrated Public Use Microdata Series: Version 6.0 [Machine-readable database] . Minneapolis: University of Minnesota. Shinar, Eva H. 1975. “ Sexual Stereotypes of Occupations.” Journal of Vocational Behavior  7( 1): 99– 111. Google Scholar CrossRef Search ADS   Tam, Tony. 1997. “ Sex Segregation and Occupational Gender Inequality in the United States: Devaluation or Specialized Training?” American Journal of Sociology  102( 6): 1652– 92. Google Scholar CrossRef Search ADS   Weeden, Kim A. 2002. “ Why Do Some Occupations Pay More Than Others? Social Closure and Earnings Inequality in the United States.” American Journal of Sociology  108( 1): 55– 101. Google Scholar CrossRef Search ADS   White, Michael J., Theresa A. Kruczek, Michael T. Brown, and Gwendolen B. White. 1989. “ Occupational Sex Stereotypes among College Students.” Journal of Vocational Behavior  34( 3): 289– 98. Google Scholar CrossRef Search ADS   White, Michael J., and Gwendolen B. White. 2006. “ Implicit and Explicit Occupational Gender Stereotypes.” Sex Roles  55( 3): 259– 66. Google Scholar CrossRef Search ADS   Williams, Christine L. 1992. “ The Glass Escalator: Hidden Advantages for Men in the ‘Female’ Professions.” Social Problems  39( 3): 253– 67. Google Scholar CrossRef Search ADS   ———. 2013. “ The Glass Escalator, Revisited Gender Inequality in Neoliberal Times.” Gender & Society  27( 5): 609– 29. CrossRef Search ADS   Wooldridge, Jeffrey M. 2010. Econometric Analysis of Cross Section and Panel Data . Cambridge, MA: MIT Press. Author notes Thank you to the participants of the 2016 workshop on occupational regulation at the Centre of the Study of Professions at Oslo and Akershus University College of Applied Sciences, to all the commentators at the session on “Gender differences in the labor market” at the ECSR Conference 2016, and special thanks to Colin Mills for his advice. I gratefully acknowledge the work and effort that three referees have put into improving the quality of this paper. This research was generously supported by the Economic and Social Research Council (award ES/J500112/1). Please direct correspondence to Felix Busch, Nuffield College, 1 New Rd, OX1 1NF Oxford, Oxfordshire, UK; e-mail: felix.busch@nuffield.ox.ac.uk © The Author 2017. Published by Oxford University Press on behalf of the University of North Carolina at Chapel Hill. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Social Forces Oxford University Press

Occupational Devaluation Due to Feminization? Causal Mechanics, Effect Heterogeneity, and Evidence from the United States, 1960 to 2010

Social Forces , Volume 96 (3) – Mar 1, 2018

Loading next page...
 
/lp/ou_press/occupational-devaluation-due-to-feminization-causal-mechanics-effect-VmH18gI32a

References (56)

Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press on behalf of the University of North Carolina at Chapel Hill. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
0037-7732
eISSN
1534-7605
DOI
10.1093/sf/sox077
Publisher site
See Article on Publisher Site

Abstract

Abstract Proponents of the devaluation hypothesis claim that occupations experiencing a significant inflow of female workers are devalued, both in status and in pay. We suggest that devaluation is an essentially cultural phenomenon that can be subject to change over time and that is not constant with varying contexts. Our theoretical model connects changing gender compositions in occupations with the formation of occupational gender stereotypes. In combination with a cultural bias that attributes less value to female work, these stereotypes can lead to devaluation. This is a novel view on the mechanics at play in the devaluation process. With US census data from 1960 to 2010, we show that devaluation was restricted to sometimes very specific contexts. A trend toward declining or disappearing devaluation is observed over the entire time span. Given potential cultural inertia and the stability of stereotypes, this result could, however, be an artifact of a deficient testing strategy that focuses solely on changes in occupational gender compositions without taking into account the normative power of the past. Introduction When occupations experience a larger inflow of female workers—meaning that they feminize—this leads to their devaluation, translating into a lower status and a decline in wages. The basic premise of the devaluation hypothesis is that this link between feminization and devaluation is indeed causal, and that it is due to cultural norms that attribute more economic value to the work done by men than to the work done by women. In the cross-section, this translates to the phenomenon that predominantly female occupations are paid less than gender integrated or predominantly male occupations, which puts all of those at disadvantage who work in a “female sector” at a disadvantage. Evidence for devaluation exists for various developed economies (United States, UK, Germany, Sweden, Netherlands, Spain, Switzerland). Moreover, recent advances have applied panel data methods to cope with unobserved heterogeneity on the individual or occupational level, and to shed light on change over time (England, Allison, and Wu 2007; Levanon, England, and Allison 2009; Murphy and Oesch 2016). Estimations of the devaluation effect have shown to be robust when being tested against diverse competing hypotheses coming from human capital or power resources theory. Overall, the existence of devaluation hence comes close to being an established fact and it is a prominent feature in explanations of wage inequality between occupations. While empirical results thus speak in favor of the devaluation hypothesis, there is persistent insecurity about its underlying causal mechanisms. Even though cross-sectional data shows relatively large associations between the proportion of female workers (from here on: percent F) and wages, panel data models using occupational fixed effects can hardly explain variation in occupational wages (e.g., England, Allison, and Wu 2007). Hence, the question emerges of what mechanisms drive this large cross-sectional association, if it is not the assumed relationship between occupational feminization and wages. Moreover, there have been several findings of heterogeneous effects over time (De Ruijter, van Doorne-Huiskes, and Schippers 2003; Levanon, England, and Allison 2009; Mandel 2013) and countries (Grönlund and Magnusson 2013; Hausmann, Kleinert, and Leuze 2015; Murphy and Oesch 2016), for which mere ad hoc explanations could be offered. Devaluation hence seems to be embedded in time- and space-specific contexts. Given the lack of research addressing these contexts, authors have consequently called for a “refinement of the theoretical discussion” and for putting more emphasis on “the search for mechanisms” (Grönlund and Magnusson 2013, p. 1015).1 More to the point, the devaluation hypothesis seems to be undertheorized, which undermines our ability to understand devaluation from a more sociologically founded perspective. Tapping into this debate, we argue that the relationship between occupational feminization and devaluation is governed by two main mechanisms. The first mechanism is the link between feminization (increases in the share of women) and the formation of occupational gender stereotypes.2 This is the process in which occupations are attributed a gender label, conditional on the share of women within that occupation. The second mechanism is the interaction of occupational stereotypes with a cultural bias that attributes a lower value to female labor, which then results in devaluation. As a consequence, changes in occupational gender compositions are linked with devaluation conditionally on the cultural context defined by said two mechanisms. Adding ideas on how these mechanisms could play out differently for men and women, and between higher- and lower-pay occupations, we derive several hypotheses about the effect heterogeneity in the devaluation process. Testing these hypotheses with census data from the United States ranging from 1960 to 2010, we show that the effect of feminization on wages via devaluation has been restricted mostly to the bottom 80 percent of occupations between 1960 and 1980, where mostly women received a penalty. Yet this is not true for the highest-paying occupations, where men have even received rewards after sectoral feminization. Overall, we observe significant declines in the devaluation effect from 1960 to 2010, to which our theoretical model offers two competing explanations: On the one hand, the spread of egalitarian gender norms and the equal pay movement could have undermined the bias that attributes less value to female work. This way, devaluation could be a phenomenon of the past. On the other hand, if unmeasured occupational gender stereotypes would have become resistant to underlying changes in the labor market’s gender composition, then the declining effects could simply be an artifact of a deficient testing strategy that solely identifies changes in the gender composition, neglecting stability in stereotypes, on which devaluation is truly based. This latter argument is supported by our finding that past levels of percent F are equally good predictors of current wages as more recent percent F, showing that occupations that previously had a large proportion of female workers are still lacking behind in wages, even though their gender composition has changed in the meantime. Keeping in mind that previous findings as well as ours do not provide enough evidence to adjudicate between those two competing stories, we strongly suggest that causal narratives with respect to the devaluation hypothesis need to be made with much care toward both theoretical considerations and modeling strategies. The following section (Devaluation so far) shortly summarizes the standard devaluation model and previous empirical results on the topic. Thereafter, in the Constructive criticism section, we propose a theoretical model and we discuss mechanisms and the potential for heterogeneous effects. We then formulate several hypotheses about effect heterogeneity over time, genders, and occupations. Following, we present the data and our empirical results, that we further embed in a broader context in the Discussion section. Devaluation so far The devaluation hypothesis has emerged as a theoretical instrument to explain the gender wage gap and to shed light on inequality between occupations. As such, this framework could be seen as complementary to more conventional approaches to the wage function, for instance building on human capital theory. At the same time, scholarly research on devaluation has received objections toward its empirical testing strategies, which has resulted in much of the field’s resources being focused on identifying a robust devaluation effect, net of potential confounding and selection mechanisms. First, reverse causality has been suggested by the crowding hypothesis following Bergmann (1974), where decreasing wages could follow if women are discriminated in higher-paying and more prestigious occupations by male employers. Second, if male occupations were protected better by closure strategies (Bol and Weeden 2015; Weeden 2002), then the correlation between percent F and wages could be spurious. Third, neoclassic economic theory suggests that in particular the division of labor in the traditional household could result in women’s lower investment in high-return, specific labor market skills, which would explain why occupations with a higher share of women pay less in general.3 On the empirical side, there is a relatively rich set of findings on devaluation from different countries, the majority of which support the devaluation hypothesis at least to some extent. For the United States, studies consistently show that—net of confounding factors—if a completely male occupation hypothetically becomes completely female, this would result in a wage penalty of 4 to 9 percent for women and a penalty of 9 to 19 percent for men (England et al. 1988; England, Allison, and Wu 2007; Gerhart and El Cheikh 1991; Karlin, England, and Richardson 2002; Kilbourne et al. 1994a, 1994b; Levanon, England, and Allison 2009; Macpherson and Hirsch 1995).4 In the most recent of these studies, Levanon, England, and Allison (2009) for instance find that female-typed occupations’ wages were 6 to 10 percent below those of mixed occupations. Evidence from other countries than the United States also speaks in favor of the devaluation hypothesis, where similar effect types are estimated to be in the range of 3 to 13 percent.5 Not all of these articles’ methods perform equally well in controlling for confounders and selection mechanisms, but many are good in at least addressing one or two of the mechanisms listed above. Piecing them together, we know that devaluation is very likely not a phantom, and that it exists despite variation in human capital requirements (Murphy and Oesch 2016; Perales 2013),6 institutional barriers (Bol and Weeden 2015; Murphy and Oesch 2016), and potential crowding (De Ruijter, van Doorne-Huiskes, and Schippers 2003; Karlin, England, and Richardson 2002; Levanon, England, and Allison 2009).7 Constructive criticism Whereas the literature is rich in studies showing that female occupations are being devalued, we know relatively little about how devaluation really works. The following discussion is built around a theoretical model that helps shed light 1) on the mechanics behind devaluation; and 2) on the plausibility of heterogeneous effects over time, between different types of occupations, and between genders. Causal effects and mechanisms The basic premise of the devaluation hypothesis is that occupational wages and prestige drop with an influx of female workers. This, in turn, leads to social inequality between occupational groups. In figure 1, this relationship is described through link (a). However, research has remained surprisingly silent with regard to the mechanisms that produce this effect. In the most general way, previous views on devaluation have assumed that “gendered cultural beliefs ... portray men as more competent and status-worthy than women,” and that “the value assigned to work in different occupations depends on the characteristics of the occupations’ incumbents” (Levanon, England, and Allison 2009, p. 868). Or, as England, Allison, and Wu (2007, p. 1238) put it: “Somehow, the low status of women ‘rubs off’ on employers’ evaluation of the occupation, and they set a lower pay level for both men and women in the occupation.” Yet, this “somehow,” as important as it seems to be in this process, has received far too little attention in the past, which is why we lack a better procedural understanding of how occupations are stigmatized and devalued in the first place. Figure 1. View largeDownload slide A theoretical model of occupational devaluation due to feminization Figure 1. View largeDownload slide A theoretical model of occupational devaluation due to feminization We hence propose a theoretical model to account for the process underlying female occupational devaluation. This model includes two main mechanisms that are believed to govern the way in which feminization leads to devaluation, and ultimately to between-occupation wage inequality. The first mechanism is the formation of occupational gender stereotypes, given the share of women within occupations (link x in figure 1). This is the step in which an individual characteristic—being female—is ascribed to the entirety of an occupational group. This has so far been widely ignored in the literature, even though it seems like an absolute necessity if we are talking about the gender trait of a group that almost always includes mixed genders. In research, the notion “female occupation” is often used as a reference to both occupations that have a lot of female incumbents and occupations that are seen as “typically female” in public perception. We believe that this is sufficiently imprecise, as it conflates an objective demographic characteristic of a group (the gender composition) with people’s view on the typical representative of this group (the stereotype). In the long run, these can be two very different things, as demographic accounts are updated every time a person joins or leaves the occupation, while stereotypes often remain sticky even though they might have become a very inaccurate representation of reality since the time they emerged (see also discussion under the section “Change over time”). The second mechanism is the interaction of gender stereotypes with a cultural bias against the value of female labor, leading to devaluation. This link (labeled y in figure 1) follows the intuition that devaluation can only emerge if the perception of an occupation to be female is paired with the judgment that women’s work is worth less than men’s work. The idea that male and female work can be differently valued due to cultural norms has previously been featured, for example in work on the drivers of segregation (Charles and Grusky 2004; Levanon and Grusky 2016) and in status characteristics theory (Berger et al. 1977), where the bias we describe would be understood as a “status belief” (see also Ridgeway and Correll [2004]). We pick up this idea and apply it to the framework that explains how occupational wages are generated. One important distinction between the two variables we include in our model (differential valuation of work and stereotypes) is that cultural norms, which define the intrinsic worth of male and female labor, are not tied to specific jobs, as they operate on a broader societal level. To the contrary, we are interested in stereotypes only insofar as they are tied to specific occupations. This distinction also makes clear why we believe that the two variables can vary independently from each other: It is, for example, not given that the spread of egalitarian gender norms, which would result in the equalization of the perceived worth of male and female work, would automatically diminish the extent to which society views specific occupations as typical for one or the other gender. We discuss this claim in more detail in the next section, where we elaborate on how such changes over time might have occurred in the United States over the past several decades. Change over time In the conventional view on devaluation, when female occupational shares change, this also affects wages. But has this link been constant over time? The theoretical model indicates that for the effect to be constant, the core mechanisms leading to devaluation must stay in place. Yet, the research suggests that both the formation of stereotypes and cultural valuations of female work are embedded in historical contexts that have changed significantly at least since the immediate postwar era. The valuation of female work and its changes The differential valuation of male and female work is well documented for the United States. Lalive and Stutzer (2010) present a review of evidence based on vignette studies indicating the existence and extent to which such differential valuations persist. In one study, for instance, Jasso and Webster (1997) find that the female-to-male ratio of perceived fair pay is 0.85 for men and 0.88 for women. These beliefs imply that it would be fair to pay women more than 10 percent less in wages, all else being equal. The systemic relevance of these beliefs is demonstrated by the intriguing findings that both men and women throughout societies have similar biases that attribute less value to women in general and to female labor specifically (Eagly, Wood, and Diekman 2000; Lalive and Stutzer 2010). On the bright side for the proponents of the equal pay movement, there is little doubt that the support for gender equality has significantly risen over the second half of the past century (Bolzendahl and Myers 2004; Charles and Grusky 2004; Mason, Czajka, and Arber 1976). This is true in a wide array of life domains, including work and family responsibilities, where women have traditionally been expected to care for the household. While this can be seen as progress in various ways, it is also widely known that such progress comes at a slow pace: Cultural views on gender are learned in early life stages, while they prove to be relatively stable throughout the life course (e.g., Alwin and Krosnick 1991). This is why generational change is often a prerequisite for wider cultural shifts.8 Furthermore, gender systems are built on firmly institutionalized beliefs, and unequal distributions of resources help maintain those systems (Ridgeway and Correll 2004). Without major external shocks, it thus takes decades for culturally induced inequality to vanish. Occupational stereotypes: changes and resilience Evidence on occupational stereotypes in the United States is not abundant, mainly because they are difficult to measure, but we nevertheless know a lot about them. The first extensive surveys found that the most male-typed occupations were, for example, miner, construction worker, or engineer.9 Female-typed occupations were manicurist, nurse, or secretary, among others (Albrecht, Bahr, Chadwick 1977; Panek, Rush, Greenawalt 1977; Shinar 1975). This all sounds familiar, but it is less clear if stereotypes have remained stable over time, given the sometimes strong changes in the gender composition within these occupations. White et al. (1989) find that there have been significant changes in stereotypes between the mid-1970s and the late 1980s. By that time, occupational stereotypes had become less strongly perceived, although they were still an important feature of how people thought about occupations. Other findings suggest that compositional changes in occupations also result in adapted stereotypes (e.g., Diekman and Eagly 2000). This makes sense in particular if we accept the premise of sex role theory (Eagly 1987) that the roles in which people act on a daily basis shape their ideas about their appropriate place in society. To the contrary, several studies show that despite desegregation in various socio-economic domains, female stereotypes and the gender stereotypes attached to occupations have remained surprisingly stable (see Lueptow, Garovich-Szabo, and Lueptow [2001] for a meta-study). Although people nowadays avoid alluding to stereotypical ideas when being asked about them, it is plausible that stereotypes are hardwired in our brains to the extent that implicit processes still bias our thoughts and actions along the lines of traditional gender views, without us knowing about it (White and White 2006).10 This line of thought is also reminiscent of the evidence on the stability of gender essentialism, the belief that women and men have innately different abilities and interests that make them more suitable for one or the other occupation (Charles and Grusky 2004; Levanon and Grusky 2016). Indeed, essentialist views, such as the belief that women are particularly suited for caring activities, do not only result from social role-derived knowledge, but they are also based on more deep-rooted biological claims (Epstein 2007), religious beliefs, traditionalism, and political views (Knight and Brinton 2017). Summary of ideas In state-of-the-art research, correlations between occupational gender compositions and wages, all else equal, are interpreted as evidence for the devaluation of female occupations. The previous discussion has its merit in defining the potential mechanisms that must be in place for this interpretation to be valid when we look at variation over time. More specifically, we see two scenarios in which declining correlations could be indicative of two very different processes: First, if the spread of egalitarian norms undermines the bias that justifies the differential valuation of male and female work, then declining effects could be interpreted as a decline of devaluation altogether. On the other hand, with a stability of occupational stereotypes (despite shifts in gender compositions), declining correlations could be viewed as an artifact of our focus on gender compositions, whereas the correlation between stereotypes and wages persists.11 Variation between occupations and gender Our understanding of devaluation as an essentially cultural phenomenon enables us to appreciate its dynamics over time. In this section, we argue that in addition to that, some occupations might be more prone to devaluation than others: We look at the occupations’ rank in the wage distribution, speculating that higher-paying occupations might be more susceptible to devaluation. Moreover, zooming in from the perspective on inequality between occupations to broadening gaps within groups, we discuss the idea that men and women, both working in the same feminizing occupation, might not receive the same penalty, and that either group might even profit from post-feminization wage adjustments. Several studies have pointed out that the wage penalty in female occupations varies between different occupational types or “classes.” De Ruijter, van Doorne-Huiskes, and Schippers (2003) show that in the Netherlands, the share of female workers is a stronger negative correlate of wages in occupations with high education and skill requirements.12Mandel (2013) produces evidence for the United States (1970 to 2007), suggesting that the feminization effect was strongest in high-wage male-typed occupations. In general, the intuition so far has been that this group of occupations is most susceptible to devaluation because it has the highest status attached to it and thus it has the “most to lose” from an entry of women who are seen as low-productivity, low-status workers (ibid., pp. 1187ff). Furthermore, occupational gender stereotypes may not have the same inherent relevance for both genders. This means that feminization could play out quite differently for male and female workers in the same occupation. In fact, due to a methodological byproduct in part of the devaluation literature,13 we already have evidence on such variation. Previously, the wage penalty after feminization was often found to be larger for men than for women (England et al. 1988; England, Allison, and Wu 2007; Gerhart and El Cheikh 1991; Macpherson and Hirsch 1995). Yet, such evidence has not been complemented with much theoretical accounts that could explain it. A more recent study furthermore does not find large effect differentials between genders (Levanon, England, and Allison 2009). We offer, once again, two competing views on this problem: Williams’s (1992) work suggests that, due to the male stereotype of high competence and leadership qualities, men are more prone to be selected into supervisory and managerial positions when working in predominantly female work environments. Hence, it might be to the advantage of male workers when the sector they work in feminizes, as this increases their propensity to move up in the wage hierarchy. Importantly, such promotions can occur within one and the same occupation, so that these changes would be observed between genders and not between occupational groups. Even though this “glass escalator” phenomenon might only partially apply in today’s world of work (Williams 2013), it strongly suggests that men’s wages might be less negatively affected by feminization or that they even might receive a net benefit from it. To the contrary, status characteristics theory (Berger et al. 1977) offers the view that a general status characteristic (here: being male or female) can lose part of its relevance in situations where a specific characteristic (working in a male or female task setting) is more salient. In other words, although men might be seen as more competent than women in general, this must not be true in settings requiring skills that are typically seen as being in the domain of female work (see also Busch [2013]). Indeed, the ensuing specific status expectations could tip the scales in favor of women in female and feminizing sectors, while men would profit from working in roles that are typically seen as more suitable for their specific skills and abilities. Hypotheses The theory section allows us to derive hypotheses about how the effect of feminization on occupational wages has shaped up over time and about how it varies between different groups. H1.1: Increasing female shares in occupations have constantly led to a decline in wages since the postwar era. Constant effects are plausible if no change has occurred in the way in which feminization is translated into stereotypes, and if those stereotypes interact with a constant cultural bias that attributes less value to female work. This would undermine vast empirical evidence that the United States has experienced a shift toward egalitarian gender norms. H1.2: The negative effect of increasing female shares in occupations on wages has slowly declined since the postwar era. This hypothesis is motivated through two conflicting points of view: First, a net decrease in the effect would be given if the rise of egalitarian norms outweighs any change in the importance of stereotypes. The slow pace of such change would be the result of the stability of cultural norms within cohorts that precludes fast cultural revolutions. Second, if unmeasured stereotypes have become stable and resistant to changes in the occupational gender structure, then we would also observe declining effects, even though devaluation could persist through the correlation between stereotypes and wages. H2: Declines in wages due to increases in female occupational shares are located primarily in the high-paying sectors. It is possible that high-status, high-wage occupations are most susceptible to an influx of lower status (female) workers. However, this theoretical idea is not very well developed and it might require further refinement. H3.1: After an occupation experiences an increase in its female share, only women receive a wage penalty in this group, while men’s wages are not negatively affected. This hypothesis is about additional wage inequality within occupations. If men can indeed profit from their stereotypical perception as better leaders and higher-competence workers, then they are more likely to be propelled into managerial and more specialized positions (within the same occupation) when an increasingly large proportion of their occupation becomes female. Hence, men would not be negatively affected and they even could profit after occupational feminization. H3.2: After an occupation experiences an increase in its female share, men’s wages are more negatively affected than women’s wages. If views on the suitability to fulfill job requirements are based on occupational skills (rather than generic skills), then men could be evaluated less fit to perform in feminizing occupations. H4: Current wage levels are determined by earlier levels of feminization, not by current levels. It is not clear if stereotypes are updated with changes in the group’s gender composition, or if they are established at a given point in time, remaining relatively constant thereafter. If the latter is true, then the stereotypes of the past would still govern wage determination processes at later stages, even if occupational segregation has declined. Data and Methods Data are drawn from the Integrated Public Use Microdata Series (IPUMS) (Ruggles et al. 2015). The sample used for this study includes six waves of US census data and ranges from 1960 to 2010, where data are available every 10 years. Data from 1950 are available but excluded, as per methodology, this year offered a far too small number of occupational observations (see clustering method below). In order to generate a quasi-panel data set from the micro-level census files, I aggregate data using occupation-industry groups for men and women, respectively. The key variables are the share of women (measured as percent of women at a given point in time) and the median wage within each group (log-transformed). As control variables, I include the mean years of education, mean of potential labor market experience,14 mean working hours, percent working in the public sector, percent ratios of ethnic minorities/non-Hispanic white, percent married, percent with young children in the household, and the relative size of the group within the labor market. The latter variable is used to control for structural economic change that influences the demand for workers in the given sector.15 Whereas most of the variables are derived by gender, this is not the case for female shares, ethnic compositions, and the relative size of groups, as their influence on wages is believed to be located purely at the occupational level. The occupation-industry clusters are derived from the IPUMS’s SOC1990 and IND1990 variables that sort individuals into harmonized occupational and industrial categorizations.16 The method used for this harmonization is backward and forward coding. As a result, some occupations that are observed in 1990 are not observed in earlier or later decades. Levanon, England, and Allison (2009) use the same data and attempt to solve this problem by merging occupations containing no observations in some cells (over years) with similar occupations. If this is done correctly, this could increase the precision in the median and mean values per occupation. However, this step potentially also introduces bias, as it might conflate occupational groups that are different in their female proportion or wage determination mechanisms. We thus do not merge any cells. However, we drop occupation-industry clusters for which we have below 20 observations for either gender. This leaves us with 1,455 occupation-industry groups that are observed on average 3.9 times. With an average of 3.9 observations per group, there are some groups that are not observed the maximum of 6 times. If possible, we also tested all models excluding groups with missing year-cells, which left the results virtually unchanged. See table 1 for the total N within each wage quintile and period cell. Table 1. Total N of Observed Occupation-Industry Cells by Quintile and PeriodQuintile   Year  1960  1970  1980  1990  2000  2010  1st  127  126  231  269  253  170  2nd  104  128  239  281  249  154  3rd  100  126  242  278  258  154  4th  90  103  241  277  258  160  5th  60  79  228  277  263  195    Year  1960  1970  1980  1990  2000  2010  1st  127  126  231  269  253  170  2nd  104  128  239  281  249  154  3rd  100  126  242  278  258  154  4th  90  103  241  277  258  160  5th  60  79  228  277  263  195  Source: IPUMS USA 1960–2010. The aggregation of occupation-industry data is based on a restricted sample, dropping individuals not at work, having zero income, and being younger than 25 or older than 65. Whereas some studies only include full-time employees in the sample (e.g., England, Allison, and Wu 2007), we believe that this could artificially distort the measured profile of occupations, as women are particularly prone to working less than full-time. Consequently, a control for average working hours is included in order to avoid bias due to potential higher returns in jobs with large volumes of hours worked. Statistical tests of all but the last hypothesis will be based on the fixed effects method (Allison 2009; Wooldridge 2010). This method allows us to exploit the quasi-panel structure of the data set and draw all inference from variation within the occupation-industry groups over time. The fixed effects models difference out constant heterogeneity between group-level clusters. In order to control for the potentially large changes in period-specific average wages from 1960 until today, I include binary time variables for each year. Interacting these period effects with the gender composition variable allows us to test effect heterogeneity over time (Allison 2009, pp. 19ff). Panel robust standard errors are employed to correct for serial correlation (Wooldridge 2010). The last hypothesis (H4) is tested via standard OLS models of 2010 occupational wages. Here, wages are regressed on current percent female, and subsequently on the same variable in 1980, testing the predictive power of both approaches. Results In table 2, we observe the results of male and female fixed effects regressions, testing the hypothesis that feminization leads to a decline in occupational wages. “P.c. female” is the main effect, and all included control variables are displayed in the table. In this first step, feminization is assumed to be constant, which leaves us with a single coefficient for the male and female wage regression, respectively. The male model results in a highly significant positive effect, whereas the effect for women is close to zero and not significant. We do not find a negative average effect, as predicted within the devaluation framework.17 Substantively, the effect implies a 0.12 percent raise in median wages for men with every percent increase of women in an occupation. The within-occupation standard deviation from 1960 to 2010 is 8.23, which leaves us with a standardized effect of 1 percent per unit change in percent female. The hypothetical change from zero to 100 percent women in an occupation would be expected to result in a wage increase of 12.75 percent for men. Table 2. Coefficients (t-values in parentheses) from Fixed Effects Regressions of Occupational Log Wages   Male wages  Female wages  Coefficients  t-values  Coefficients  t-values  P.c. female  0.0012***  (4.23)  0.0004  (1.49)  1960 (ref.)           1970  0.2137***  (26.94)  0.1632***  (20.87)   1980  0.1740***  (12.33)  0.0114  (1.06)   1990  0.1210***  (8.03)  −0.0231  (−1.84)   2000  0.1103***  (6.54)  0.0009  (0.06)   2010  0.0958***  (4.82)  0.0037  (0.20)  Working hours  0.0050*  (2.34)  0.0076***  (4.85)  Public sector  −0.0013**  (−2.82)  0.0000  (0.11)  Ratio Black  0.0042  (0.17)  0.0311  (0.98)  Ratio Native  −0.4976***  (−2.27)  −0.5567**  (−2.83)  Ratio Asian  0.2316**  (2.63)  0.2245**  (3.25)  Ratio Hispanic  −0.1018*  (−2.45)  −0.0923**  (−2.95)  Years edu.  0.0753***  (14.60)  0.1063***  (14.29)  LM exp.  0.0403***  (7.11)  0.0281***  (5.47)  LM exp.2  −0.0006***  (−5.01)  −0.0005***  (−4.18)  P.c. married  0.0070***  (16.68)  0.0017***  (4.60)  P.c. child  0.1181**  (2.81)  0.2364***  (4.75)  Sector size  0.0096  (0.97)  −0.0031  (−0.25)  Constant  0.3442*  (2.51)  0.3183**  (2.63)  Ntotal  5,720  5,720  Ngroups  1,455  1,455    Male wages  Female wages  Coefficients  t-values  Coefficients  t-values  P.c. female  0.0012***  (4.23)  0.0004  (1.49)  1960 (ref.)           1970  0.2137***  (26.94)  0.1632***  (20.87)   1980  0.1740***  (12.33)  0.0114  (1.06)   1990  0.1210***  (8.03)  −0.0231  (−1.84)   2000  0.1103***  (6.54)  0.0009  (0.06)   2010  0.0958***  (4.82)  0.0037  (0.20)  Working hours  0.0050*  (2.34)  0.0076***  (4.85)  Public sector  −0.0013**  (−2.82)  0.0000  (0.11)  Ratio Black  0.0042  (0.17)  0.0311  (0.98)  Ratio Native  −0.4976***  (−2.27)  −0.5567**  (−2.83)  Ratio Asian  0.2316**  (2.63)  0.2245**  (3.25)  Ratio Hispanic  −0.1018*  (−2.45)  −0.0923**  (−2.95)  Years edu.  0.0753***  (14.60)  0.1063***  (14.29)  LM exp.  0.0403***  (7.11)  0.0281***  (5.47)  LM exp.2  −0.0006***  (−5.01)  −0.0005***  (−4.18)  P.c. married  0.0070***  (16.68)  0.0017***  (4.60)  P.c. child  0.1181**  (2.81)  0.2364***  (4.75)  Sector size  0.0096  (0.97)  −0.0031  (−0.25)  Constant  0.3442*  (2.51)  0.3183**  (2.63)  Ntotal  5,720  5,720  Ngroups  1,455  1,455  Source: IPUMS USA 1960–2010. *p < 0.5 **p < 0.1 ***p < 0.01 This finding is interesting in itself, including some initial evidence that men are favored in the wake of occupational feminization (H3.1). However, assuming a constant feminization effect on wages is not satisfactory given our theoretical model. In a further step, I hence decompose the previous results by period and occupational wage quintiles. Figure 2 depicts marginal effects of feminization over the entire time span for each quintile. Estimations include identical control variables, but marginal effects are derived from including an interaction term between P.c. female and periods. Furthermore, the sample of 1,455 occupation-industry groups was split in quintiles according to average wages in each group over the full time span. Figure 2. View largeDownload slide Marginal effects of feminization on wages, by occupational wage quintiles and gender (1960–2010) Figure 2. View largeDownload slide Marginal effects of feminization on wages, by occupational wage quintiles and gender (1960–2010) The results suggest that feminization can cause occupational devaluation. Seen in historical context, however, this relationship has been tied to specific periods (1960s–1980s), was stronger for women than for men, and was rather located in the second and fourth wage quintile. These findings are partly in line with our ideas on effect heterogeneity and the theoretical model, but they are also more diffuse than we expected. I summarize in three points: First, effect sizes are relatively similar between all wage groups, excluding the very top quintile. In quintiles one to four, effect sizes are mostly either negative or around zero (in earlier decades), becoming slightly positive in more recent years. Negative coefficients, which are evidence for the predicted wage effect via feminization, are largest in the fourth quintile. In the fifth, however, we find exclusively positive wage effects, for both women and men. As such, wage developments at the top of the income ladder are related to feminization in a very different way, as it is the case for all remaining groups. At the top, positive effect sizes for men range between 0.19 percent per unit increase in P.c. female in 1980, and 0.53 percent in 1970. This partial view also explains the positive wage effect found in the previous step: Replicating the male model shown in table 2 excluding the fifth quintile, the previously significant positive effect comes very close to zero and is non-significant. Taken together, this undermines the plausibility of H2, according to which devaluation would be located primarily in the top sectors. In the highest quintile in 1960 and 1970, confidence intervals are relatively large for both genders, which is due to the lower N of observations within these period-cells (see also table 1). The magnitude of the errors makes our analysis less reliable for these years, but an estimation of similar models including only groups without any missing period-cells does not lead to any substantively different results (not shown here). Table 3. Correlations of Percent F and Current/Subsequent Occupational Log Wages   p.c. F1960  p.c. F1970  p.c. F1980  p.c. F1990  p.c. F2000  p.c. F2010  ln(wages)1960  −0.45            ln(wages)1970  −0.49  −0.50          ln(wages)1980  −0.54  −0.55  −0.56        ln(wages)1990  −0.44  −0.45  −0.45  −0.39      ln(wages)2000  −0.36  −0.37  −0.36  −0.30  −0.27    ln(wages)2010  −0.31  −0.32  −0.31  −0.24  −0.20  −0.17    p.c. F1960  p.c. F1970  p.c. F1980  p.c. F1990  p.c. F2000  p.c. F2010  ln(wages)1960  −0.45            ln(wages)1970  −0.49  −0.50          ln(wages)1980  −0.54  −0.55  −0.56        ln(wages)1990  −0.44  −0.45  −0.45  −0.39      ln(wages)2000  −0.36  −0.37  −0.36  −0.30  −0.27    ln(wages)2010  −0.31  −0.32  −0.31  −0.24  −0.20  −0.17  Source: IPUMS USA 1960–2010. Second, effect sizes are relatively similar between genders and most trends seem to apply to both men and women. We implement z-tests, following Clogg, Petkova, Haritou (1995), confirming that in every single period-quintile cell except one, male and female coefficients are not significantly different from each other with p < 0.5 (results not shown). In figure 2, we yet also observe that when devaluation existed, it was mostly impacting the wages of women. This is clearest in the second and fourth quintiles, where women were affected by feminization the most. Another twist is observed in the top wage group: Here, the effect on male wages is positive and significant over the entire range of years, while this is the case for women only in two periods. The lack of more significant female coefficients is because of both smaller effect sizes and wider confidence intervals, describing more variation in wages overall. Evidence thus speaks more in favor of H3.1 (men are losing less and they are gaining more after feminization), while H3.2 finds no support at all. Third, on visual inspection of the marginal effects, we would believe that the devaluation effect has been decreasing in time at least for the second, third, and fourth wage quintiles. This trend seems to be following a relatively linear pattern for women, while it is bumpier for men. Figure 3 depicts results of a pairwise comparison of marginal effects in 2010 and 1960, by wage quintile and gender. The positive and statistically significant differences for men and women in quintiles 2–4 suggest that effect sizes have in fact decreased over time. The largest significant decline in the feminization effect is observed for women in the fourth quintile (0.36 points difference), and the smallest one is observed for men in the third quintile (0.12). According to this analysis, no significant change happened in the highest- and lowest-paying sectors. Evidence thus speaks—at least for part of the wage distribution—in favor of H1.2: The effect of occupational feminization on wages has slowly decreased over time. Figure 3. View largeDownload slide Differences in marginal effect sizes, 2010 and 1960, by wage quintile and gender, and 95 percent confidence intervals Figure 3. View largeDownload slide Differences in marginal effect sizes, 2010 and 1960, by wage quintile and gender, and 95 percent confidence intervals Finally, we address the question of whether stereotypical gender views of the past are still affecting current wages. Here, I employ a strategy using the full sample, including male and female observations, and occupational median wages as a dependent variable. This is equivalent to the methodology in cross-sectional studies that use OLS wage regressions to quantify the adverse effect of working in a female occupation. Table 3 lists cross-correlations of P.c. female and occupational median log wages. The bold figures in the diagonal show the correlation of current levels. A striking feature of the results is that the correlation between P.c. female and wages was strongest in the 1970s and 1980s, while it dropped severely in 2000 and 2010. This result is fostering our previous assertion that the effect of feminization on wages has significantly declined over time. Moreover, we observe that the female percentages of previous decades are much stronger correlates of 2010 wages than the 2010 percentages themselves: corr(wages2010,p.c. F1970) = 0.32 vs. corr(wages2010,p.c. F2010) = 0.17. This finding would support the hypothesis that past gender segregation is still influencing the current pay levels in occupations, even though the distribution of male and female workers over occupations has changed (H4). Table 4. Coefficients (t-values in parentheses) from OLS Regression of 2010 Occupational Log Wages   Only:p.c. F2010  Only:p.c. F1980  Added:Education  Added:Other controls  p.c. F2010  −0.0036***    −0.0046***    −0.0017***      (−5.98)    (−15.65)    (−6.45)    p.c. F1980    −0.0058***    −0.0043***    −0.0017***      (−11.22)    (−16.15)    (−7.11)  Education      0.1946***  0.1822***  0.1583***          (48.56)  (45.60)  (23.17)    N  736  736  736  736  736  736  Adj. R2  0.0451  0.1452  0.7733  0.7769  0.8768  0.8783    Only:p.c. F2010  Only:p.c. F1980  Added:Education  Added:Other controls  p.c. F2010  −0.0036***    −0.0046***    −0.0017***      (−5.98)    (−15.65)    (−6.45)    p.c. F1980    −0.0058***    −0.0043***    −0.0017***      (−11.22)    (−16.15)    (−7.11)  Education      0.1946***  0.1822***  0.1583***          (48.56)  (45.60)  (23.17)    N  736  736  736  736  736  736  Adj. R2  0.0451  0.1452  0.7733  0.7769  0.8768  0.8783  Source: IPUMS USA 1980, 2010. *p < 0.5 **p < 0.1 ***p < 0.01 After the implementation of a multivariate OLS regression of 2010 log wages, this initial judgment can be put into perspective. In table 4, I depict the results of six models, where three models use current P.c. female as predictor and the remaining three models include P.c. female from 1980.18 The first two models, including only our main independent variable, simply replicate the findings from the cross-correlations table where the 1980 percentages are a much stronger predictor of 2010 wages levels than its 2010 equivalents. However, the subsequent two models critically show that, holding constant the educational levels in occupations, the two correlations with P.c. female are approximately level. This suggests that the initial finding is biased by a selection mechanism that has changed in the period between 1980 and 2010: Nowadays, women are much more likely to select into high-education, high-paying jobs, while this was not the case in 1980 (cf. Mandel 2013). In the two last models, I include all remaining control variables, which lowers the effect of percent female in both instances, but this does not change the overall evaluation that both current and past levels of feminization are similarly good predictors of current wage levels. This finding can be interpreted as partly confirming H4. Discussion Previous results showing that in the United States occupational feminization leads to devaluation were confirmed in this study. Yet, this is only true if we look at sometimes relatively specific contexts in which devaluation seems to play a larger role than in others. Relying on a single coefficient for the United States of the past fifty or sixty years is not nearly enough to capture the complex and culturally contingent mechanisms that produce devaluation. This was shown by decomposing the average feminization effect by periods, by occupational wage groups, and by gender. This attempt, especially when added up with similar research from the past, should caution following studies against ignoring effect variation, especially when describing a phenomenon working through the channels of cultural bias and stereotyping. The latter are neither stable over time, nor do they always affect different subgroups in the population in the same way. We have observed a decline in occupational devaluation in the middle part of the wage ladder. This decline seems to have taken place as part of a slow and steady development from 1960 to 2010, which fits well with our theoretical expectations that incremental cultural change toward gender egalitarian norms would have slowly eradicated devaluation. However, if we believe the theoretical model, then we should be cautioned that the declining effect over time could also be due to an increasing stability in gender stereotypes. In such a scenario, ongoing change in the occupational gender structure would have less or no effect on stereotypes, which would logically weaken the observed correlation between feminization and wages, even though devaluation would persist. This interpretation finds some evidentiary support in our results showing that previous levels of occupational segregation are about equally good predictors of today’s wages as current segregation levels. As this paper does not offer a more refined strategy to isolate one causal narrative from the other, we yet caution the reader not to derive a strong causal claim from our findings. We clearly require further tests of how the link between feminization and wages has shaped up over time. Currently, we see two ways to address this issue: We either build statistical models in which we test the composite effect of current and past feminization at the same time. or we find a way to measure occupational stereotypes—possibly even over time—including them as a new control in our wage regressions. The former approach would possibly need to deviate from standard fixed effects models, as those measure changes in key variables that make it difficult to test any type of inertia in wage determination mechanisms. The latter approach would enable us to directly test the model’s mechanisms, but data on stereotypes by occupation is very scarce, if available at all. The theory section included some vague ideas on how the feminization effect could differ between higher- and lower-status occupations. Our intuition was that at the top of the wage ladder, occupations are prestigious and mostly dominated by men, which could make them more susceptible to an inflow of lower-status female workers (Mandel 2013). We have tested these ideas by estimating separate models by wage quintile, showing that the interesting part of the variation might in fact be located between the fifth quintile and the rest of all occupations. In the fifty years before 2010, trends were relatively similar in the bottom 80 percent of occupations. The fact that wage penalties were strongest in the fourth quintile could, however, be interpreted as some support of the high-status-susceptibility hypothesis. The most striking result yet was found for the top 20 percent: For this group, female devaluation does not seem to have existed at all. Men have profited from feminization all along, while female effects were also positive, even though rather non-significant. This is forcedly a novel finding, because—to our knowledge—previous studies have not looked at such a granularity when dividing occupations into wage or status groups. One might be tempted to think that this finding could be explained by an increasing shortage of specific skills located in high-paying jobs. Liu and Grusky (2013) demonstrate that particular analytic skills, but also computer, managerial, and social skills, have seen a rise in payoff between 1980 and 2010. However, this account would not allow to us to explain the constant rewards for men in these high-paying jobs, or the large feminization coefficients in the 1960 and 1970, where much of the skill-biased structural change would not have happened yet. At this point, we certainly require more research—including different data sources and possibly cross-country insights—to confirm this peculiarity, and to help better understand the impact of feminization on wage generation mechanisms at the top of the wage ladder.19 In the past, surveys have sometimes shown different coefficients for the devaluation of male and female wages. For the United States, there is, however, no clear tendency showing either of the genders to be more disadvantaged than the other. In the most similar study to the one at hand, Levanon, England, and Allison (2009) show very similar effect sizes for men and women, which is confirmed by our own results. Significant devaluation effects were found mostly for female wages, and in the top wage group, benefits were significant over all periods only for men. The results yet seem insufficiently skewed in favor of men to speak of any healthy evidence for the glass escalator effect. Budig (2002) has argued that previous evidence for men riding glass escalators in female occupations might be misleading due to a lack of appropriate counterfactual reasoning. In such cases, studies often focus on predominantly female sectors (such as nursing), neglecting a more meaningful comparison of results with male advantage in mixed or predominantly male sectors. Furthermore, we can also not find any evidence for Mandel’s (2013) judgment that women are “moving up the down staircase”—meaning that women have reached high-skilled jobs yet tend to experience declining wages. Even though this study was able to shed light on various under-researched features of devaluation, it also neglects an important criticism that certainly deserves more attention. Several authors have highlighted that the assumption of a linear relationship between feminization and devaluation could be too simplistic (Magnusson 2013). The most popular alternative approach is to divide occupations into clusters of predominantly male, female, and gender-mixed groups, where the event of switching to the female-dominated sector is operationalized as a main predictive variable. From a statistical point of view, this approach could be favorable due to its relaxation of the linearity assumption. Moreover, assuming that the cultural process in which occupational stereotypes are formed follows a stepwise (not continuous) adaption where certain thresholds in the share of women must be reached in order to change a prevailing stereotype, then splitting up occupations into more or less female groups might make a lot of sense. Unfortunately, the data used in the present paper does not allow to further break up occupational groups without seriously undermining statistical efficiency. Finally, studies using similar statistical models have previously been criticized for a lack of more refined controls for specific vocational training and changes in skill requirements within occupations. This is due to a trade-off where we chose very long time series over data sets that include richer background variables that are at the same time being limited to a time span in which the slow change in culturally determined processes could hardly be observed. Conclusion The idea that occupations devalue due to their feminization has received a lot of attention—not just because it challenges dominant views of how wages are determined, but also because it was backed up by ample empirical support. In this paper, we claim that if further progress is to be made on this topic, then we need to more explicitly address the causal mechanics of female devaluation. Without explaining in more detail how feminization impacts occupational wages, we will also be unable to understand why this effect seems to be heterogeneous over time, between countries, and between different occupational groups. The proposed theoretical model is an attempt to address these concerns. It captures the mechanisms by which occupational gender stereotypes are formed and through which those stereotypes interact with a cultural bias against female labor to result in devaluation. Given the very plausible changes within these mechanics over time, we have shown that feminization might be linked to devaluation in various different ways. Moreover, if stereotypes freeze at any given point in time, devaluation might be entirely decoupled from ongoing changes in gender compositions, as it would be governed by past segregation processes. In the empirical section, we have shown that feminization did result in a decline of wages in the United States from 1960 to 2010, but that the effect was restricted to sometimes very specific contexts in which they occurred. Devaluation was mostly observed in the 1960s to 1980s. Men seem less affected than women, even though this result has received only weak evidentiary support. In the top wage quintile, all of the above does not apply. Here, no devaluation could be observed in any period or for either gender. This finding undermines the notion that in particular high-status occupations would be susceptible to an inflow of lower-status (female) workers. Over and above the results on effect heterogeneity, this paper raises important questions pointing at the culturally contingent contexts in which devaluation might be embedded. This is missing in the canonical view on devaluation and—in our view—deserves more rigorous testing in future research. Notes 1 Mandel (2013, p. 1202) also urges that “further investigation is required to explicate the mechanisms underlying the over-time trends” of the percent F effect. 2 The term “occupational gender stereotype” is very central in the following debate. The general idea behind this is that there are broad societal views on what gender is typical for an occupation. This is a subjective category, which is not the same as the actual or objective gender composition of an occupation, which is usually measured in studies on devaluation. We also refer to research on sex-typing of occupations that reaches back to the 1970s (cf. White and White 2006, p. 259) and where stereotypes are defined as “unconscious habits of thought that link personal attributes to group membership” (Reskin 2000, p. 322). 3 For an excellent full-scale discussion of potential confounders and selection mechanisms, see, for example, recently Murphy and Oesch (2016). 4 Even though these are seemingly impressive figures, they also tend to inflate the importance of devaluation by assuming the irrelevant case in which an occupation switches from zero to 100 percent female. 5 See evidence for the UK (Brynin and Perales 2016; Murphy and Oesch 2016; Perales 2013), for Germany (Busch 2013; Murphy and Oesch 2016; but also Hausmann, Kleinert, and Leuze 2015; Leuze and Strauß 2016), for Sweden (Grönlund and Magnusson 2013; Magnusson 2013), and for the Netherlands (De Ruijter, van Doorne-Huiskes, and Schippers 2003), Switzerland (Murphy and Oesch 2016), and Spain (Polavieja 2008, effect non-significant). 6 Some evidence points into the opposite direction, for example Tam (1997) or Polavieja (2008). 7 Controlling for crowding is notoriously difficult to do. De Ruijter, van Doorne-Huiskes, and Schippers (2003) employ a measure of how concentrated women are to a restricted number of occupations. A clear problem with this measure is that women could be concentrated in some occupations for many reasons that are not based on discrimination of distorted preferences. Levanon, England, and Allison (2009) simply show for the United States that if wages in occupations decrease, this does trigger a larger influx of women. This is probably a better attempt of controlling for crowding, as it rules out reverse causality in the devaluation story altogether. 8 One study addressing attitudes on fair pay directly, and also within the context of generational change, is presented by Auspurg, Gatskova, Hinz (2013), who show that older cohorts in West Germany put more weight on a person’s gender when judging on fair pay in work. 9 Research on stereotypes has employed different test strategies to probe gender-typical views attached to occupations. They hence do not rely on the objective gender composition in occupations to identify female- and male-typed occupations. For an introduction of the instruments, see White and White (2006). 10 For a distinction between explicit and implicit stereotypes in this context, see White and White (2006, pp. 259ff). 11 A similar reasoning can be found in England, Allison, and Wu (2007), who discuss if inertia, particularly in the wage structure, might be a reason for the observation that feminization has not led to much change in pay levels after all. 12 Grönlund and Magnusson (2013) test if this is also true in the Swedish case but do not find any significant effects. 13 Several surveys using occupation-level data include separate statistical models for men and for women in order to control for the individual-level wage gap. 14 Following approaches in the literature, we proxy labor market experience by taking individuals’ age, subtracting their years of education, and adding 6 (the typical age where children enter school in the United States). 15 Changes in the group sizes could also reflect supply-side factors. One variant discussed in the paper is crowding, where women are shifted toward less favorable segments in the market. 16 There are different approaches to group individuals into occupation and industry clusters. The approach in this paper was previously found to deliver good results. See Levanon, England, and Allison (2009) for a thorough discussion and tests with other classifications. 17 Although our study uses a similar methodology and data than Levanon, England, and Allison (2009), we receive contrary results in this first step: They find a negative effect of changes in percent F on wages for both genders (p. 881, table 4), while our estimates are positive and significant for men but non-significant for women. In order to probe if the different findings can be explained by our methodological choices, we have attempted to fully replicate the findings by Levanon et al. This included, inter alia, the inclusion of 1950 to 2000 IPUMS data, a different treatment of occupation-industry clustering, the conversion of percent F into logits, slight differences in sample restrictions, the inclusion of lagged independent variables including a 10-year lag of logit(percent F), and the implementation of a dynamic fixed effects model correcting for endogeneity. With these specifications, we were unable to replicate the authors’ findings of negative coefficients. To the contrary, results were similar as shown in our table 2. Hence, even though it is difficult to suggest what has led to the different findings, we are confident that none of the methodological choices listed above play an important role. 18 We also could have used the 1970 or 1960 values as a comparison group, but the number of occupation-industry groups observed in 2010 and 1980 was much larger than for the combinations with earlier decades. 19 The deviation of our results by occupational group from those by Mandel (2013) are strong and underline the merit of replication studies using different data. A reason for the differences could lie in the choice of occupational clusters, as Mandel looks at under 400 occupations, while we divide the sample into over 1,400 groups altogether. See Levanon, England, Allison (2009) for the implications of these choices. About the Author Felix Busch is a DPhil candidate in sociology at Nuffield College, University of Oxford. His research interests are in social inequality, labor markets, and gender. In his dissertation, he studies the role of occupational gender segregation in the production of social inequalities. References Albrecht, Stan L., Howard M. Bahr, and Bruce A. Chadwick. 1977. “ Public Stereotyping of Sex Roles, Personality Characteristics, and Occupations.” Sociology & Social Research  61( 2): 223– 40. Allison, Paul D. 2009. Fixed Effects Regression Models . Thousand Oaks, CA: SAGE Publications. Google Scholar CrossRef Search ADS   Alwin, Duane F., and Jon A. Krosnick. 1991. “ Aging, Cohorts, and the Stability of Sociopolitical Orientations over the Life Span.” American Journal of Sociology  97( 1): 169– 95. Google Scholar CrossRef Search ADS   Auspurg, Katrin, Ksenija Gatskova, and Thomas Hinz. 2013. “ Vorstellungen von Lohngerechtigkeit in West-und Ostdeutschland und in der Ukraine.” WSI-Mitteilungen  66( 2): 77– 88. Google Scholar CrossRef Search ADS   Berger, Joseph, Hamit Fisek, Robert Z. Norman, and Morris Zelditch. 1977. Status Characteristics and Social Interaction: An Expectation-States Approach . New York: Elsevier. Bergmann, Barbara R. 1974. “ Occupational Segregation, Wages and Profits When Employers Discriminate by Race or Sex.” Eastern Economic Journal  1( 2): 103– 10. Bol, Thijs and Kim A. Weeden. 2015. “ Occupational Closure and Wage Inequality in Germany and the United Kingdom.” European Sociological Review  31( 3): 354– 69. Google Scholar CrossRef Search ADS   Bolzendahl, Catherine I., and Daniel J. Myers. 2004. “ Feminist Attitudes and Support for Gender Equality: Opinion Change in Women and Men, 1974–1998.” Social Forces  83( 2): 759– 89. Google Scholar CrossRef Search ADS   Brynin, Malcolm, and Francisco Perales. 2016. “ Gender Wage Inequality: The De-Gendering of the Occupational Structure.” European Sociological Review  32( 1): 162– 74. Google Scholar CrossRef Search ADS   Budig, Michelle J. 2002. “ Male Advantage and the Gender Composition of Jobs: Who Rides the Glass Escalator?” Social Problems  49( 2): 258– 77. Google Scholar CrossRef Search ADS   Busch, Anne. 2013. “ Der Einfluss der beruflichen Geschlechtersegregation auf den Gender Pay Gap.” Kölner Zeitschrift für Soziologie und Sozialpsychologie  65( 2): 301– 38. Google Scholar CrossRef Search ADS   Charles, Maria, and David B. Grusky. 2004. Occupational Ghettos: The Worldwide Segregation of Women and Men . Stanford, CA: Stanford University Press. Clogg, Clifford C., Eva Petkova, and Adamantios Haritou. 1995. “ Statistical Methods for Comparing Regression Coefficients between Models.” American Journal of Sociology  100( 5): 1261– 93. Google Scholar CrossRef Search ADS   De Ruijter, Judith M P, Anneke van Doorne-Huiskes, and Joop J. Schippers. 2003. “ Size and Causes of the Occupational Gender Wage-Gap in the Netherlands.” European Sociological Review  19( 4): 345– 60. Google Scholar CrossRef Search ADS   Diekman, Amanda B., and Alice H. Eagly. 2000. “ Stereotypes as Dynamic Constructs: Women and Men of the Past, Present, and Future.” Personality and Social Psychology Bulletin  26( 10): 1171– 88. Google Scholar CrossRef Search ADS   Eagly, Alice H. 1987. Sex Differences in Social Behavior: A Social-Role Interpretation . Hillsdale, NJ: Lawrence Erlbaum. Eagly, Alice H., Wendy Wood, and Amanda B. Diekman. 2000. “Social Role Theory of Sex Differences and Similarities: A Current Appraisal.” In The Developmental Psychology of Gender , edited by T. Eckes and H. M. Trautner, pp. 123– 174. Mahwah, NJ: Lawrence Erlbaum. England, Paula, Paul Allison, and Yuxiao Wu. 2007. “ Does Bad Pay Cause Occupations to Feminize, Does Feminization Reduce Pay, and How Can We Tell with Longitudinal Data?” Social Science Research  36( 3): 1237– 56. Google Scholar CrossRef Search ADS   England, Paula, George Farkas, Barbara Stanek Kilbourne, and Thomas Dou. 1988. “ Explaining Occupational Sex Segregation and Wages: Findings from a Model with Fixed Effects.” American Sociological Review  53( 4): 544– 58. Google Scholar CrossRef Search ADS   Epstein, Cynthia F. 2007. “ Great Divides: The Cultural, Cognitive, and Social Bases of the Global Subordination of Women.” American Sociological Review  71( 1): 1– 22. Google Scholar CrossRef Search ADS   Gerhart, Barry, and Nabil El Cheikh. 1991. “ Earnings and Percentage Female: A Longitudinal Study.” Industrial Relations: A Journal of Economy and Society  30( 1): 62– 78. Google Scholar CrossRef Search ADS   Grönlund, Anne, and Charlotta Magnusson. 2013. “ Devaluation, Crowding or Skill Specificity? Exploring the Mechanisms behind the Lower Wages in Female Professions.” Social Science Research  42( 4): 1006– 17. Google Scholar CrossRef Search ADS   Hausmann, Ann-Christin, Corinna Kleinert, and Kathrin Leuze. 2015. “ Entwertung von Frauenberufen oder Entwertung von Frauen im Beruf?” Kölner Zeitschrift für Soziologie und Sozialpsychologie  67( 2): 217– 42. Google Scholar CrossRef Search ADS   Jasso, Guillermina, and Murray Webster Jr. 1997. “ Double Standards in Just Earnings for Male and Female Workers.” Social Psychology Quarterly  60( 1): 66– 78. Google Scholar CrossRef Search ADS   Karlin, Carolyn Aman, Paula England, and Mary Richardson. 2002. “ Why do Women’s Jobs Have Low Pay for Their Educational Level?” Gender Issues  20( 4): 3– 22. Google Scholar CrossRef Search ADS   Kilbourne, Barbara, Paula England, George Farkas, Kurt Beron, and Dorothea Weir. 1994a. “ Returns to Skill, Compensating Differentials, and Gender Bias: Effects of Occupational Characteristics on the Wage of White Women and Men.” American Journal of Sociology  100( 3): 689– 719. Google Scholar CrossRef Search ADS   Kilbourne, Barbara, Paula England, and Kurt Beron. 1994b. “ Effects of Individual, Occupational, and Industrial Characteristics on Earnings: Intersections of Race and Gender.” Social Forces  72( 4): 1149– 76. Google Scholar CrossRef Search ADS   Knight, Carly R., and Mary C. Brinton. 2017. “ One Egalitarianism or Several? Two Decades of Gender-Role Attitude Change in Europe.” American Journal of Sociology  122( 5): 1485– 1532. Google Scholar CrossRef Search ADS   Lalive, Rafael, and Alois Stutzer. 2010. “ Approval of Equal Rights and Gender Differences in Well-Being.” Journal of Population Economics  23( 3): 933– 62. Google Scholar CrossRef Search ADS   Leuze, Kathrin, and Susanne Strauß. 2016. “ Why Do Occupations Dominated by Women Pay Less? How Female-Typical Work Tasks and Working-Time Arrangements Affect the Gender Wage Gap among Higher Education Graduates.” Work, Employment & Society  30( 1): 1– 19. Levanon, Asaf, and David Grusky. 2016. “ The Persistence of Extreme Gender Segregation in the Twenty-First Century.” American Journal of Sociology  122( 2): 573– 619. Google Scholar CrossRef Search ADS   Levanon, Asaf, Paula England, and Paul Allison. 2009. “ Occupational Feminization and Pay: Assessing Causal Dynamics Using 1950–2000 US Census Data.” Social Forces  88( 2): 865– 91. Google Scholar CrossRef Search ADS   Liu, Yujia, and David Grusky. 2013. “ The Payoff to Skill in the Third Industrial Revolution.” American Journal of Sociology  118( 5): 1330– 74. Google Scholar CrossRef Search ADS   Lueptow, Lloyd B., Lori Garovich-Szabo, and Margaret B. Lueptow. 2001. “ Social Change and the Persistence of Sex Typing: 1974–1997.” Social Forces  80( 1): 1– 36. Google Scholar CrossRef Search ADS   Macpherson, David A. and Barry T. Hirsch. 1995. “ Wages and Gender Composition: Why Do Women’s Jobs Pay Less?” Journal of Labor Economics  13( 3): 426– 71. Google Scholar CrossRef Search ADS   Magnusson, Charlotta. 2013. “ More Women, Lower Pay? Occupational Sex Composition, Wages and Wage Growth.” Acta Sociologica  56( 3): 227– 45. Google Scholar CrossRef Search ADS   Mandel, Hadas. 2013. “ Up the Down Staircase: Women’s Upward Mobility and the Wage Penalty for Occupational Feminization, 1970–2007.” Social Forces  91( 4): 1183– 1207. Google Scholar CrossRef Search ADS   Mason, Karen Oppenheim, John L. Czajka, and Sara Arber. 1976. “ Change in US Women’s Sex-Role Attitudes, 1964–1974.” American Sociological Review  41( 4): 573– 96. Google Scholar CrossRef Search ADS   Murphy, Emily and Daniel Oesch. 2016. “ The Feminization of Occupations and Change in Wages: A Panel Analysis of Britain, Germany, and Switzerland.” Social Forces  94( 3): 1221– 55. Google Scholar CrossRef Search ADS   Panek, Paul E., Michael C. Rush, and James P. Greenawalt. 1977. “ Current Sex Stereotypes of 25 Occupations.” Psychological Reports  40( 1): 212– 14. Google Scholar CrossRef Search ADS   Perales, Francisco. 2013. “ Occupational Sex-Segregation, Specialized Human Capital and Wages: Evidence from Britain.” Work, Employment & Society  27( 4): 600– 620. Google Scholar CrossRef Search ADS   Polavieja, Javier G. 2008. “ The Effect of Occupational Sex-Composition on Earnings: Job-Specialization, Sex-Role Attitudes and the Division of Domestic Labour in Spain.” European Sociological Review  24( 2): 199– 213. Google Scholar CrossRef Search ADS   Redbird, Beth. 2017. “ The New Closed Shop? The Economic and Structural Effects of Occupational Licensure.” American Sociological Review , online first. Reskin, Barbara F. 2000. “ The Proximate Causes of Employment Discrimination.” Contemporary Sociology  29( 2): 319– 28. Google Scholar CrossRef Search ADS   Ridgeway, Cecilia L. and Shelley J. Correll. 2004. “ Unpacking the Gender System. A Theoretical Perspective on Gender Beliefs and Social Relations.” Gender & Society  18( 4): 510– 31. Google Scholar CrossRef Search ADS   Ruggles, Steven, Katie Genadek, Ronald Goeken, Josiah Grover, and Matthew Sobek. 2015. Integrated Public Use Microdata Series: Version 6.0 [Machine-readable database] . Minneapolis: University of Minnesota. Shinar, Eva H. 1975. “ Sexual Stereotypes of Occupations.” Journal of Vocational Behavior  7( 1): 99– 111. Google Scholar CrossRef Search ADS   Tam, Tony. 1997. “ Sex Segregation and Occupational Gender Inequality in the United States: Devaluation or Specialized Training?” American Journal of Sociology  102( 6): 1652– 92. Google Scholar CrossRef Search ADS   Weeden, Kim A. 2002. “ Why Do Some Occupations Pay More Than Others? Social Closure and Earnings Inequality in the United States.” American Journal of Sociology  108( 1): 55– 101. Google Scholar CrossRef Search ADS   White, Michael J., Theresa A. Kruczek, Michael T. Brown, and Gwendolen B. White. 1989. “ Occupational Sex Stereotypes among College Students.” Journal of Vocational Behavior  34( 3): 289– 98. Google Scholar CrossRef Search ADS   White, Michael J., and Gwendolen B. White. 2006. “ Implicit and Explicit Occupational Gender Stereotypes.” Sex Roles  55( 3): 259– 66. Google Scholar CrossRef Search ADS   Williams, Christine L. 1992. “ The Glass Escalator: Hidden Advantages for Men in the ‘Female’ Professions.” Social Problems  39( 3): 253– 67. Google Scholar CrossRef Search ADS   ———. 2013. “ The Glass Escalator, Revisited Gender Inequality in Neoliberal Times.” Gender & Society  27( 5): 609– 29. CrossRef Search ADS   Wooldridge, Jeffrey M. 2010. Econometric Analysis of Cross Section and Panel Data . Cambridge, MA: MIT Press. Author notes Thank you to the participants of the 2016 workshop on occupational regulation at the Centre of the Study of Professions at Oslo and Akershus University College of Applied Sciences, to all the commentators at the session on “Gender differences in the labor market” at the ECSR Conference 2016, and special thanks to Colin Mills for his advice. I gratefully acknowledge the work and effort that three referees have put into improving the quality of this paper. This research was generously supported by the Economic and Social Research Council (award ES/J500112/1). Please direct correspondence to Felix Busch, Nuffield College, 1 New Rd, OX1 1NF Oxford, Oxfordshire, UK; e-mail: felix.busch@nuffield.ox.ac.uk © The Author 2017. Published by Oxford University Press on behalf of the University of North Carolina at Chapel Hill. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.

Journal

Social ForcesOxford University Press

Published: Mar 1, 2018

There are no references for this article.