Average female wages in traditionally male occupations have steeply risen over the past couple of decades in Germany. This trend led to a new and substantial pay gap between women working in male-typed occupations and other women. I dissect the emergence of these wage disparities between women, using data from the German Socio-Economic Panel (1992–2015). Compositional change with respect to education is the main driver for growing inequality. Other factors are less influential but still relevant: marginal returns for several wage-related personal characteristics have grown faster in male-typed occupations. Net of individual-level heterogeneity, traditionally male occupations have also become more attractive because of rising returns to task-specific skills. Discrimination of women in typically male lines of work seems to have declined, too, which erased part of the wage penalty these women had previously experienced. In sum, I document changes in the occupational sorting behavior of women as well as shifts in occupation-level reward mechanisms that have had a profound impact on the state of inequality between working women.
The declining gender segregation of the occupational landscape has been one of the big structural changes in the economy within the past 60 years. Although rates of desegregation lagged behind change in other areas of gender equality (Charles and Grusky 2004; England 2010), they were still substantial in size, and desegregation was observable in various country contexts (Blau and Kahn 2017; Hakim 1994; Hausmann and Kleinert 2014; Jacobs 1989). The dissimilarity index (Duncan and Duncan 1955) for the United States, for instance, went down from 64.48 to 51.04 between 1970 and 2009 (Blau et al. 2013).1 Analyses of the demographic gradients in desegregation revealed for various developed economies that the largest share of these observed trends can be attributed to women with particularly high educational attainment and from younger cohorts (Blau et al. 2013; Jacobs 1999; Pettit and Hook 2009; but also see Pearlman 2019). Furthermore, evidence from the United States and Germany notes that much of the change was confined to specific occupational groups. Women integrated a few sectors that were previously dominated by men, particularly white collar, service, and sales occupations. At the same time, desegregation was practically inexistent in other fields, such as blue collar work (Blau et al. 2013; Busch 2013b; Cotter et al. 2004; England 2010; Hausmann and Kleinert 2014).
Several studies documented how these changes contributed to the convergence of average wages between women and men (Blau and Kahn 2017; Mandel and Semyonov 2014). However, what is missing so far is a structured analysis of how the demographic and occupation-specific dynamics of desegregation led to growing inequality within the female population. The concern for this issue arises from the intuition that desegregation was primarily achieved through the sorting of highly qualified women into a set of increasingly well-paid, traditionally male occupations. This resulted in the socioeconomic differentiation between women in male-typed occupations and other women, who are less qualified and who work in less-advantaged occupational fields. Knowledge about the extent of growing wage disparities between women is very limited because existing evidence is mostly confined to empirical snapshots, showing group differences between women at single points in time (Cotter et al. 2004; De Ruijter et al. 2003; Levanon et al. 2009; Magnusson 2013; Murphy and Oesch 2016). A few articles reported on the changing demographic profile of women in male-typed occupations (e.g., Mandel 2013, 2018), but they did not deliver a systematic analysis of how total wage differences between groups developed over time. Other studies focused on wage penalties (and changes thereof) in predominantly female occupations (Busch 2018; Levanon et al. 2009), mostly ignoring the rising wage advantages for women in predominantly male occupations.
In this article, I address this gap in the extant literature by studying changes in the wage difference between women in typically male occupations and other women. I focus on outcomes in the German labor market between 1992 and 2015, an interesting case given shifts in policies, gender norms, and female labor force participation (see the next section). In addition to estimating the magnitude of changing wage differences, I also focus on the channels through which this change occurred.
Analyzing data from the German Socio-Economic Panel (SOEP), I find that the wage gap between women in male-typed occupations and other women has strongly grown over the past quarter-century. Because of the parallel unfolding of several processes starting in the 1990s, average female wages in typically male occupations skyrocketed, while economy-wide wages increased moderately at best. The cumulative effect of these processes has been substantial. In the early 1990s, women in male-typed work still earned slightly less than women in gender-integrated occupations and only slightly more than those in female-typed occupations. Comparatively, in 2015, women’s mean hourly wages in male-typed occupations were 25% (11%) higher than in female-typed (mixed) occupations. Wage disparities between women in different occupational fields are thus a recent and pervasive phenomenon. Analysis reveals that the increased sorting of highly educated women into male-typed occupations is the most important driver behind the growth of wage differences. Rising returns to task-specific skills and to other wage-related personal characteristics also added to growing wage disparities albeit to a lesser extent. Contributions of unobserved components suggest that a decline in discriminatory practices against women in male-typed work has improved their economic standing, too.
In sum, this study demonstrates how the differentiation of women in the labor market has been connected with increasing wage inequality between groups. The growing representation of women in well-paid, typically male domains can be seen as great progress toward gender equality that was possible only because of women’s success in higher education. At the same time, this trend created new disparities in the female labor force that were further amplified by structural change between occupations. In the following section, I lay out the economic, political, and normative context of these developments in Germany, and I and trace women’s pathways into paid work and traditionally male occupations.
Women in the German Labor Market
Germany is well-known as a country in which women’s role in paid work was long restricted by institutional and normative constraints. Incremental change in the economic, cultural, and political arenas have, however, slowly eroded part of this picture in important ways.
In the second half of the twentieth century, women’s labor market integration moved rather slowly in West Germany, while being more dynamic in the East (Busch 2013b:30–32). Yet, because of the German reunification, and particularly because of a considerable rise in female employment during the early 2000s, female labor force participation rose to 74% among working-age women (aged 15–64) until 2017 (Organisation for Economic Co-operation and Development (OECD) 2019). In international comparison, the occupational landscape in Germany is strongly segregated by gender, in East Germany more so than in the West (Busch 2013a; Trappe and Rosenfeld 2001). Desegregation occurred only slowly over the past decades (Charles and Grusky 2004; Hausmann and Kleinert 2014). Furthermore, most of the decline in aggregate segregation can be attributed to structural change rather than adapted sorting behavior of women and men (Busch 2013b:158; Hausmann and Kleinert 2014). This is, however, not true for desegregation in the professional and managerial occupations that can largely be explained by changing patterns of occupational selection (Busch 2013b:165–167). In fact, Hausmann and Kleinert (2014:7) found that between the mid-1970s and 2010, women were particularly successful in increasing their representation in these two occupational groups, along with engineering.2 This suggests that although aggregate desegregation has been slow in Germany, visible progress was made in some high-skill, high-wage occupations.
The activation of women for paid work as well as their integration into traditionally male work domains must be viewed against the backdrop of Germany’s institutional and normative setup and of this setup’s reconfiguration in recent times. Within a typically conservative welfare regime (Esping-Andersen 1990; Korpi and Palme 1998), social policy in Germany has long been an impediment to stronger female involvement in the labor market (Gauthier 1996; Korpi 2000; Pettit and Hook 2005). In particular, women’s reactivation rates after childbirth were negatively affected by the conservative policy-making in the 1980s and 1990s (Gangl and Ziefle 2015; Pettit and Hook 2009; Schönberg and Ludsteck 2014). This tradition held until the mid-2000s, when politics pushed more in the direction of a dual-earner, dual-carer model (Bünning 2015; Kluve and Tamm 2012; Schober 2014). In line with this policy context, gender norms in the society shifted from a comparatively strong preference for women’s involvement in the family (over paid work) around 1990, to a more liberal setup until the early 2000s (Braun and Scott 2009; Braun et al. 1994). At the end of the 2000s, average intrinsic commitment to paid work was relatively high in international comparison, and it was significantly higher for German women than for men (Steiber 2013:208).3 This pathway of women toward an increasingly active role in the labor market would have been unthinkable without the vast expansion of free or subsidized education, particularly in the secondary and tertiary sectors (Becker and Mayer 2019:155), that Germany experienced in the postwar era (Göggel 2007:2–4).
Through the trends of activation, higher education, and desegregation, women in Germany earned a strong footing in the labor market. Yet, we know little about how this mix of socioeconomic processes affected the overall (or between-occupation) wage structure in the female population. Some relevant knowledge comes from studies that produced cross-sectional evidence showing that women’s earnings in Germany (as in other countries) are similar in gender-integrated and in male-typed occupations. At the same time, earnings are substantially lower in typically female sectors (Busch 2013a; Cotter et al. 2004; Murphy and Oesch 2016) and somewhat lower in occupation cells with 90% to 100% male incumbents (Cotter et al. 2004; Murphy and Oesch 2016). Furthermore, in Germany, women’s wage differences in occupations with 20% to 60% female workers seem to be very small, if they exist at all (Murphy and Oesch 2016:13). Even when an extensive set of control variables is used, this nonlinearity persists (Cotter et al. 2004; Murphy and Oesch 2016). The previous literatures have not delivered a systematic analysis of how these wage disparities between women in different occupational fields have changed over time, so it remains our task to describe such trends and their underlying factors.
In this section, I develop my main arguments about recent shifts in the female wage structure. Building on the previous discussion, I argue that in the process of women’s entry into typically male occupations, the female workforce has become increasingly differentiated with respect to important wage-related personal characteristics. In tandem with structural changes in the economy that influenced the wage distribution in the same period, this differentiation led to a growing wage advantage of women in male-typed occupations over other women in the labor market.
Sorting and Differentiation
Germany has experienced a growing representation of women in traditionally male occupations, but this trend was unevenly distributed over the female population. The disproportionate entry of highly educated women (Blau et al. 2013; Jacobs 1999; Pettit and Hook 2009) into a set of high-skill professional and managerial occupations (Busch 2013b; Cotter et al. 2004; England 2010; Hausmann and Kleinert 2014; Percheski 2008) led to fundamental change in the composition of women working in male-typed occupations. I expect that women’s increasing differentiation with respect to their socioeconomic background contributed to growing wage disparities.
Hypothesis 1.1 (H1.1): The increased selection of highly educated women into male-typed occupations has elevated their wages relative to other women.
Another potential source of differentiation is selection on unobserved characteristics. Institutional and cultural barriers decrease the baseline propensity of women to enter a male-typed field, a mechanism already well visible in education or in the transition to work (Charles and Bradley 2009; Shauman 2009). Several literatures suggest that psychological factors play an important role in influencing women’s tendency to avoid typically male career paths (e.g., Cech et al. 2011; Correll 2001, 2004). As a corollary, I argue, it is evident that those women who are nevertheless successful in entering and persisting in a male-typed occupation are also positively selected on personal traits, such as confidence, tenacity, or ability. A possible result of the increased inclusiveness of women in male-typed occupations in recent decades (Hausmann and Kleinert 2014) is that these women are less positively selected on unobserved traits than they were before.
Hypothesis 1.2 (H1.2): Decreasing positive selection on unobserved characteristics has contributed to a decreasing wage advantage of women in male-typed occupations relative to other women.
Returns to Education and Skills
In parallel with compositional change in occupations, several shifts in the economy had an additional impact on the opportunities in traditionally male domains of work. Skill-biased technological change (SBTC) theory predicts job creation in high-paying analytical jobs (Violante 2008) as well as increasing returns to education if growing demand for such labor is not met with sufficient supply (Goldin and Katz 2009). In West Germany, the growth of returns to education started only in the 2000s (Anger et al. 2010; Gebel and Pfeiffer 2007; Lauer and Steiner 2000), later than in several other developed economies, but a steady upward trend could be observed for East Germany after reunification (Anger et al. 2010). Given that many of the traditionally male occupations fall into the spectrum of high-skill work, I expect to see that rising educational premiums in this group also contributed to growing wage inequality between women.
Hypothesis 2.1 (H2.1): Rising returns to education have contributed to a wage advantage of women in male-typed occupations relative to other women.
Whereas SBTC theory predicts occupational upgrading, its theoretical expansion proclaims the hollowing-out of the labor force resulting from routine-biased technological change (Autor et al. 2003; Fernández-Macías 2012; Goos and Manning 2007).4 The idea was that computerized systems would be able to substitute jobs that require a high degree of routinized work. Further theoretical contributions in the field identified specific skill sets that would be complementary with recent technological advances. Liu and Grusky (2013:1358), for instance, demonstrated the increasing value of analytical, computer, managerial, and nurturing skills. The authors attributed the increasing value of analytical and computer skills in large part to technological change. On the other hand, rising returns to managerial and nurturing skills are explained through patterns of labor demand that might be specific to the United States. There is not much evidence on this trend within the German context. To convert the existing evidence on task-specific skills and levels of routine at work in this study, I rely on the broad idea that male-typed occupations are mostly found in the highest ranks of the wage distribution. These occupations often require analytical and managerial skills that are very low in routine contents.
Hypothesis 2.2 (H2.2): Both rising returns to task-specific labor market skills and the decreasing value of routine work have contributed to a wage advantage of women in male-typed occupations relative to other women.
Despite wide consensus that typically male occupations often generate larger salaries (England 1992; Levanon et al. 2009), some scholars have noted that women who work in these occupations are less likely to profit from such an advantage. Within organizational contexts, token women stand out as a minority, they are more isolated, and they are confronted with stereotypical views of female inferiority (Kanter 1977).5 Similar predictions can be made on the basis of status characteristics theory (Berger et al. 1977), when employers make inferences of women’s weaker ability to perform in typically male occupations (Wagner and Berger 1997:12–14). Such discriminatory standards are based on cultural beliefs that justify people’s tendency to attribute less value to what is considered female work against typically male work (Auspurg et al. 2017; England et al. 1988; Ridgeway 2011). Despite a surprising degree of stability in gender beliefs (Lueptow et al. 2001), there is a consensus that Germany, just as other developed economies, has experienced a shift away from traditionalism toward more gender-egalitarian cultural norms (Bolzendahl and Myers 2004; Brewster and Padavic 2000; Ciabattari 2001; Davis and Greenstein 2009; Inglehart and Norris 2003; Knight and Brinton 2017). Given that traditional beliefs are a key component of the undervaluation of women, I hypothesize that the discrimination of women in typically male occupations has declined over time. On the other hand, a comparable discriminatory effect would not have existed in more gender-integrated or female-typed fields, given the lack of a culture viewing women as inferior within fields that had already been culturally marked as more typical for women. Hence, the declining penalty for women in male-typed occupations would have led to increasing average wages for them relative to other women.
Hypothesis 3 (H3): Declining discrimination of women in male-typed occupations contributed to their wage advantage relative to other women.
If we cared only about the individual components of growing inequality, it would be easy to dismiss some of the discussed changes as rather trivial or inconsequential. For instance, differences in educational or unobserved characteristics exist independently of occupational sorting. In other words, highly educated women would also profit from their educational achievements if they were not employed in a typically male sector. This argument, however, would miss the cumulative advantage that is likely to emerge from the joint effect of multiple mechanisms. With the exception of changes in the structure of unobserved characteristics, all discussed hypotheses point toward a growing aggregate wage advantage for women in male-typed occupations over other women. The additive effects of discussed mechanisms imply, for instance, that women with strong educational credentials increasingly selected into occupations that have generated growing wages for highly educated workers. I thus expect to observe that average wages of women in traditionally male occupations have outpaced the wages of other women over the past decades.
Hypothesis 4 (H4): During the past several decades, average wages of women in male-typed occupations have outpaced the wages of other women.
I use individual-level data from the Socio-Economic Panel Study (SOEP) (Wagner et al. 2007), a representative survey of the residential population living in private households in Germany that started in 1984. Throughout the study, the main dependent variable is the natural logarithm of gross hourly wages deflated to 2015 euros.6 The sample is restricted to observations made between 1992 and 2015, and the self-employed are not included. I employ weights in all computations to contain discontinuities in the data arising from, for example, oversampling of certain subpopulations.
Occupations are captured on the basis of the KldB-1992,7 and they are grouped into one of three clusters that designate the occupations’ gender type: male, mixed, or female. The choices made when aggregating occupations into these three groups are not trivial as previous studies have demonstrated (e.g., Levanon et al. 2009). I build on a study by Murphy and Oesch (2016), who suggested that occupational gender types in Germany are best captured by looking at the 60% threshold of sex ratios (p. 18–19). Hence, female-typed occupations include at least 60% women, male-typed occupations include at least 60% men, and mixed occupations are in between.8 Furthermore, it is generally a good idea to capture occupation-level sex ratios on a very detailed level because precision is quickly lost when similar occupations that have different gender types are conflated. Thus, I start from the most detailed four-digit classification level, merging groups with a similar occupation only if fewer than 20 observations are available. This leaves 1,147 occupation clusters for which I compute individual gender types. These occupational gender types (male, female, mixed) are kept constant over the 1992–2015 period.9
To derive the main independent variables, I rely on two data sources. The SOEP delivers information on individuals’ formal education in the CASMIN standard for educational achievement. I condense the CASMIN categories into three groups for analysis.10 Occupational task components are derived on the basis of the Employment Surveys by the BIBB/IAB (1991/1992 and 1998/1999) and by the BIBB/BAuA (2005/2006 and 2011/2012) (Hall et al. 2015). Tasks are useful to approximate the level of productive skills—that is, the extent to which a person’s skills are required in a given job. I compute average task measures for occupation cells and marry the results with the SOEP data (for details, see Table A1 in the online appendix). Staying close to domain-specific previous work (e.g., Liu and Grusky 2013), I extract values for seven tasks: verbal, quantitative, analytical, creative, computer, management, and care. Furthermore, I quantify the level of routine at work to account for the substitution effect via automated systems.
I also employ a set of control variables in the analyses. Differences in acquired human capital are held constant by including individuals’ labor market experience (years in full-time employment and their second order polynomial) and firm tenure (also measured in years). Previous research has produced evidence on rising premiums for very long work hours, which I control with a dummy variable for those working 50 hours or more (Cha and Weeden 2014). Conflict potential between family and work is controlled by variables detailing whether a woman has a child below age 6 in the household and indicating the number of hours spent in the household and with childcare. Finally, I add control variables on the workplace (firm size, the industry, and a dummy variable for public vs. private sector) and demographic background variables (east vs. west, marital status, and living with partner). The final data set includes 88,138 women with at least 2,177 observations per year and with at least 296 observations per gender type–year.
In the theory section, I develop a framework to analyze the wage advantage built by women in male-typed occupations over other women. Although this suggests a twofold comparison (male-typed occupations vs. other), I split the “other” group into women in mixed and female-typed occupations, aligning this research with the broader literature on gender segregation. I proceed by looking at the changing wage inequality between women in two ways. First, I decompose changes in wage gaps between women in male-typed occupations and women in mixed and female-typed occupations, respectively. The decomposition technique applied in this step was introduced by Juhn et al. (1991) and has since been used in different applied studies in the social sciences (e.g., Blau and Kahn 2017; Cha and Weeden 2014). Second, I implement occupation switch models via fixed-effect regression that allow me to control for stable idiosyncratic traits of individuals and to estimate the contribution of occupational characteristics to intergroup wage differences. Although I have yearly data from the SOEP ranging from 1992 until 2015, I construct biennial data cells for all statistical models to increase the stability of estimates through a larger N per period.
The observed quantities component summarizes changes in the wage difference that can be attributed to changes in average characteristics of women (e.g., educational upgrading). The observed prices part details changes due to changing marginal returns (e.g., increasing returns to education). The unobserved quantities component captures whether women in mixed/female-typed occupations are moving up or down in the distribution of women working in male-typed work. This component captures two substantive changes of interest: changes in unobserved traits (e.g., tenacity, ability) and changes in discrimination. In this framework, I cannot isolate these two mechanisms statistically. Hence, the estimated component can be interpreted as a cumulative outcome of both changes in unobserved traits and discrimination. Furthermore, the unobserved prices component summarizes aggregate wage inequality in traditionally male occupations after accounting for observed characteristics.13
When the within-transformation is applied to the data, αi is eliminated, enabling me to estimate the effect of switching between occupational gender-type groups (e.g., switching from a female- to a male-typed occupation) while controlling for plausible differences in performance-relevant unobserved factors. I use panel-robust standard errors with the fixed-effect models.
Figure 1 shows median hourly wages in male-typed, mixed, and female-typed occupations over the period 1992–2015 for women (panel a) and men (panel b). Women working in traditionally male occupations have outpaced other women in terms of wage growth since at least the early 1990s. Their median wage surpassed that of women working in mixed occupations in the early 2000s and has gained more ground since then until 2015. The main interest here is in the wages of women, but the comparison with male wages is also instructive: men in typically male occupations experienced much less wage growth, and their growth rates were less steep than those observed for men in mixed occupations. This suggests that the underlying factors for women’s increasing wage performance in male-typed occupations are likely located on the individual rather than occupational level.
Turning now to potential mechanisms driving the trend of increasing wages for women in male-typed occupations, I show in Table 1 average characteristics (pooled means and deltas between 1992 and 2015) that summarize how the socioeconomic compositions of women differ among the three groups. Although shifts over time (deltas) are of most interest in this research design, it is worthwhile to appreciate that women working in traditionally male domains have been distinct in various ways over the entire observed period. Two factors that must positively affect their income relative to the other two groups are their high educational attainment and their tendency to work long hours. Other notable differences are that women in male-typed occupations work much more often in manufacturing—a typically male field—and that they are more often located in the eastern part of Germany, where wages are generally lower. Even though these differences exist toward both women in the mixed and women in the female-typed groups, the latter group also shows some distinct characteristics: those employed in female-typed occupations have lower education and experience, and they spend considerably less time at work (where they are often employed in small firms) and more at home caring for the household and children. Overall, the differences in characteristics between the three groups illustrate well that women in male-typical occupations are most strongly geared toward monetary success in the labor market. On the other hand, women working in female-typed occupations are clearly much closer to the traditional stereotype of the homemaker, taking on work in the family and orienting less toward a professional career. Differences in the presence of young kids in the household seem to be rather small.
Finally, I want to highlight the temporal dynamics observed in the displayed variables. Notably, compared with the mixed and female-typed occupations, in male-typed occupations, the proportion with a tertiary degree has strongly increased (from 26% in 1992, up to 50% in 2015), work hours have declined less sharply, and the representation in large firms has grown. All these changes could potentially explain why women in traditionally male domains were able to increase their wage advantages over the past couple of decades. At the same time, they have seen a relative decline in their average labor market experience, perhaps because younger women were more prone to enter traditionally male occupations, which pushed down average age and experience relative to other women. This pattern (declining relative age) could also explain why women in male-typed occupations have shown slightly elevated rates of parenthood over the observed period.
Decomposing the Change in Wage Differences
Table 2 summarizes the aggregate components of changing wage differences between women in typically male occupations and women working in mixed or female-typed domains. The first two columns detail components of change between male- and female-typed occupations, and columns 3 and 4 display results for the comparison between male-typed and mixed occupations. The mean wage difference between traditionally male and female occupations changed by 0.244 log points (0.146 points vs. mixed occupations).14 In both cases, changes in the observed components were most influential, explaining more than 60% of total changes in differences. Also notable is that the change in unobserved quantities had a substantial positive impact on the opening of the wage gaps.
Table 3 lists detailed observed components of changing wage patterns. The quantities components for education confirm that the entry of highly educated women into predominantly male occupations has strongly elevated their wages relative to other women. More than 40% of the growth in the wage difference between the male- and female-typed groups (30% vs. the mixed group) can be attributed to increasing discrepancies in average educational attainment. Changes in the presence of young children in the household as well as changes in the time committed to work at home have not contributed appreciably. On the other hand, growing wage disparities can also be attributed to women’s unevenly distributed growing representation in larger firms, in specific industrial sectors, and in the former West German states. Furthermore, it is interesting to see that educational premiums (prices) in male-typed occupations have slightly decreased relative to prices in other occupations. This somewhat limited the increasingly stratifying effect of educational differences. Finally, prices for overwork have increased particularly in male-typed fields, further increasing wage disparities.15
The identical analysis is further applied to three subperiods of the overall 1992 to 2015 time span: 1992–1999, 1999–2007, and 2007–2015 (Table 4). This choice of periods is motivated by idiosyncratic sociodemographic changes occurring during these episodes in Germany.16 There are three main takeaways from this analysis. First, the growth of wage disparities was strongest up to 2007, slowing down in the third period. Second, in absolute terms, the differentiation by educational achievement was not specific to one of the subperiods, but its relative contribution to period-specific changes in wage gaps increased over time. From this perspective, educational sorting by occupational fields has become more relevant in recent years. Third, there has only been a small relative increase in educational premiums in male-typed occupations in the early 2000s, whereas relative premiums declined in the 1990s and post-2007.
The previous section highlights how compositional change transformed the sociodemographic profile of women who work in male-typed occupations and how this transformation has impacted wages. This section addresses how occupation-level mechanisms have contributed to increasing wage disparities between women. I implement models capturing women’s switching between occupational categories, based on Eq. (5). Here, the idea is to measure wage changes when individuals switch from one gender type field to another (e.g., from a male- to a female-typed occupation). This way, I filter out any stable characteristics (such as educational attainment or motivation) that would confound the relationship between the occupational gender types and wages.
To begin, I implement a model that includes only binary variables for the occupational gender types (male-typed, mixed, female-typed) and biennial dummy variables, yielding the wage differences after switching to another occupational group as an average over all years. The results (not displayed) show that women receive about 1% higher wages when switching from a mixed to male-typed occupation (not significant) and 3.6% higher wages when switching from a female-typed occupation (highly significant). In the next step, I disaggregate the average gender type coefficient of this “empty” model into biennial estimates that are displayed in Table 5, Model 1.17 For most years, women in the sample profited from switching to a traditionally male field (see positive coefficient estimates), but these effects are mostly insignificant up to the mid-2000s. From 2006/2007 onward, the wage bonus associated with switching from a female- to a male-typed occupation became larger and significant (4.1% in 2006/2007 and 7.1% in 2014/2015). The figures for Model 2 display results after individual-level control variables (same as in the decomposition models, including educational attainment) are introduced. The decline of some of the estimates suggests that part of the wage benefits obtained when switching to a traditionally male occupation is due to nonconstant person-level heterogeneity.
In the next step, I add variables to the wage equation to control for variation in occupational tasks between the three groups. This includes seven task types and the degree of routine at work. Model 3 includes an average coefficient over all years for each variable. In Model 4, the same coefficients are allowed to vary over periods, which accounts for growing or diminishing returns to skills over time. Overall variation in occupational tasks as well as growing differences in how much employers are paying for them explain most of the wage bonus experienced by those who switch to a male-typed occupation. Summing up, the fixed-effects models show a wage bonus of up to 4.6% when switching to a typically male work domain (Model 2). Yet, this bonus shapes up less like an advantage in male-typed occupations and more like a recent disadvantage in female-typed occupations given their inherent task structure. Tasks that I found to be critical for bumping up wages in recent years were related to computers, care, and management (coefficients displayed in Figs. A1 and A2, online appendix). With some nuance, this confirms previous findings on computer, managerial, and nurturing tasks from the U.S. context (Liu and Grusky 2013) and findings on the value of management tasks in Germany (Liebeskind 2004).
Before proceeding with the conclusions of this article, I want to address the additional question of how to explain the strong educational upgrading in male-typed occupations that I identified as the prime factor of growing wage disparities between women. Three principal mechanisms plausibly could have led to such upgrading. First, increased demand for highly skilled workers in traditionally male occupations would have elevated the average educational attainment in the group. Second, demographic and cultural change could have led more highly educated women to choose male-typed careers in order to pursue goals of increased status and pay, which would be a supply-side mechanism. Third, if women faced increasing difficulties in entering stereotypically male work—for instance, because of discriminatory barriers—then stronger positive selection could have also led to a rise in educational levels.
Figure 2 depicts the proportion of women and men with a tertiary degree (CASMIN = 3) for each of the three occupational gender types. Panel a shows a strong increase in the presence of highly educated women in male-typed fields (as in Table 1). The comparison with the trend for men illustrates that upgrading was mostly confined to the female population. Hence, the scenario of demand-driven upgrading is unlikely given that this would have resulted in similarly strong changes in educational attainment for men.18 Furthermore, the proportion of women in occupations categorized as typically male has increased from 6.2% in 1985 (only West Germany) to 9.1% in 1992 (reunified Germany), up to 13.5% in 2015. Such desegregation indicates that rather than facing stronger barriers, women have been more successful in pursuing traditionally male careers. This also makes the reinforced selection argument much less plausible. Overall, I believe that the data best support the supply-side narrative in which particularly highly educated women had the motivation and the opportunity to choose more male-typed careers.
This article documents the growing wage advantage of women in male-typed occupations over women who work in traditionally female and gender-integrated fields. The analysis is central to evaluate growing social inequality between women in strongly segregated labor markets. Some 25 years ago, average wages of the three groups were relatively similar. In 2015, women’s mean hourly wages in male-typed occupations were 25% and 11% higher than in female-typed and gender-integrated fields, respectively. Estimating monthly labor income on the basis of observed hourly rates and work hours, I find that in 2015, women in female-typed and mixed occupations had only 63% and 84%, respectively, of the expected income of women working in traditionally male occupations.
Beyond highlighting this trend, the main objective of this article is to identify its driving forces. Results of formal decompositions strongly suggest that women’s educational upgrading in male-typed occupations has been the prime factor for observed changes. Nowadays, about 50% of women in these occupations have earned a tertiary degree—much more than men in comparable jobs or other women. In 2014/2015, educational advantages in male-typed occupations accounted for a wage bonus of 12% and 7% over women in female-typed and mixed domains, respectively.19 Hence, wage inequality between women has emerged primarily because of increased sorting of highly educated women into well-paid, male-typed occupations.
Occupation-level mechanisms seem to be negligible between women in traditionally male and mixed occupations. On the other hand, I provide evidence that the gap between male- and female-typed occupations has also been growing because of increasing returns to task-specific skills. This is a recent phenomenon observed since the mid-2000s. In line with previous work (Liu and Grusky 2013), the value of several task-specific skill types (analytical, management, and care) seems to have increased over time,20 while wages in routine-heavy jobs declined. So far, very few studies have employed multidimensional task measures to explain the wage differences between occupational gender categories. The results here extend previous research on the feminization of occupations that rely on one-dimensional measures of skill and that fail to explain wage disadvantages in female-typed occupations (e.g., Murphy and Oesch 2016).
I furthermore exploit the longitudinal setup of the study to look at changes in unobserved factors that drive women’s success in male-typed work. To my knowledge, previous research has not produced any evidence on unobserved differences between male-typed, mixed, and female-typed occupational groups. To that extent, my results are novel and should be interpreted with caution. Overall, I find that unobserved factors had a sizable impact on growing disparities in wages between women. In my view, the most plausible explanation for this result is an underlying decline in discrimination against women in traditionally male occupations. More work is required to corroborate this finding. The balance of work and family commitments, and its potential effect on gender segregation in the labor market, is a classical topic in the literature (Becker 1985, 1993; Polachek 1981). Germany has long exhibited relatively conservative views on gender roles (see the earlier section, Women in the German Labor Market), penalties for motherhood are still found to be much larger than in other countries (e.g., recently, Kleven et al. 2019), and fertility among highly educated women is low (Kreyenfeld and Konietzka 2008). In this study, I do not, however, detect that changing patterns in the occurrence of (or the penalties for) motherhood, work in the household, or time spent with childcare played an important role for increasing wage disparities between women. A reason for this finding could be that I do not further differentiate occupational groups by skill level (Görlich and de Grip 2009). Furthermore, it could be argued that this study does not address the known issue that mothers in Germany are particularly prone to being economically inactive, which leads to endogenous sample selection (Pettit and Hook 2009:62).21
A finding that goes against expectations is that between 1992 and 2015, educational returns in male-typed occupations have declined relative to educational returns in other occupations.22 This has, in fact, slightly attenuated the growth of wage differences between these groups. One possible source of the devaluation of educational credentials is the massive inflow of highly educated women into male-typed occupations. I show that within the course of a quarter-century, the percentage of women with a tertiary degree in this group increased from about 20% to 50%. If this landslide change in the educational composition in male-typed occupations was not caused by demand-side mechanisms, then this likely would have created an oversaturation of highly educated workers, depressing the marginal returns to education. In other words, there was not enough high-skilled work for all the high-skilled female workers who started to work in a typically male occupation.
This narrative resonates well with an important side finding regarding what has been behind the increasing rate at which highly educated women sorted into male-typed occupations. Given the practical non-existence of similar changes in the educational backgrounds among men in male-typed occupations, I find it very unlikely that demand-side occupational upgrading is behind the trend. My findings are most consistent with the notion that cultural change opened career paths for high-achieving women, who increasingly started to target traditionally male lines of work in order to maximize status and monetary returns. This cultural change did not take place just in the heads of those who pursue such careers (Goldin 2006). It also occurred among firms and the wider public, which have become more accepting of the idea that women work in fields previously dominated by men. To further these ideas, it would be necessary to design studies that more directly test the theory about the causes of changes in women’s occupational sorting.
I am grateful for financial support of the Economic and Social Research Council (Grant No. ES/J500112/1). Thank you to Colin Mills, Richard Breen, Paula England, and Helen Buchs for their helpful feedback on earlier drafts of the article. Thank you to Marlis Buchmann, who generously supported me during the writing of the article at the University of Zurich. I also benefitted from discussions at the 2018 European Consortium for Sociological Research conference and at the 2018 Zurich Sociology thesis workshop. I also thank the Editors and the anonymous reviewers for their very constructive comments and help throughout the publication process. All errors are my own.
Data sets used in this study are the Socio-Economic Panel (DOI: https://doi.org/10.5684/soep.v32.1) and the Employment Surveys of BIBB/IAB (DOIs: https://doi.org/10.4232/1.2565 and https://doi.org/10.4232/1.12247) and BIBB/BAuA (DOIs: https://doi.org/10.4232/1.11072 and https://doi.org/10.7803/501.12.1.1.40). These can be accessed through the Research Data Centers of the SOEP and of the BIBB, respectively. Replication source code is deposited on https://www.github.com/febusch/research.
Compliance With Ethical Standards
Ethics and Consent
Analyses in this article are based solely on secondary survey data. The author followed good practice standards and the data providers’ regulations to protect data privacy and to prevent unauthorized proliferation of microdata.
Conflict of Interest
The author declares that he has no conflict of interest.
Substantively, this means that in 2009, 51% of all workers would have had to switch into another occupational group in order for each group’s sex compositions to be identical with the overall sex composition in the labor market.
Other groups that saw a marked inflow of women were higher-skill administration, technical occupations, and semiprofessional occupations (Hausmann and Kleinert 2014:7).
Multidimensional approaches to gender norms show that a dominant culture of liberal egalitarianism has developed in Germany but that more traditional views still prevail in parts of the society (Grunow et al. 2018; Knight and Brinton 2017).
Although some countries have exhibited a hollowing-out of the wage structure (Autor et al. 2003; Goos and Manning 2007), evidence for this phenomenon is weak in the German case (Oesch and Rodríguez Menés 2011; but see also Spitz-Oener 2006:261–263).
A strongly related concept is the “glass escalator effect” that potentially advantages men in female-typed work environments (Budig 2002; Williams 1995; Wingfield 2009).
This variable is constructed using the current monthly labor income and weekly work hours multiplied by 4.35 (roughly the average number of weeks in a month). I also restrict the leverage of outliers by top- and bottom-coding hourly wages. This affects about 0.37% of all observations and makes the wage gap estimates slightly more conservative.
The German KldB-1992 (Statistisches Bundesamt 1992) is a typical hierarchical classification of occupations, comparable with the U.S. Standard Occupational Classification.
Results in this article are similar but a bit attenuated if I use slightly different sex thresholds (e.g., 70%) to compute occupational gender groups.
I hold them constant for two reasons. First, changes in gender types can be expected to have very minor effects over shorter time spans (Busch 2018; England et al. 2007). Second, I would have to merge many more occupational cells if I wanted to compute a sex ratio per cell-year, which would make the analysis much more imprecise.
Category 1: basic vocational or elementary education and below; category 2: intermediate education (general or vocational); category 3: tertiary education.
The choice about which group is represented by A and B is consequential for estimation results. Therefore, it is customary to display decomposition results by changing group assignment between A and B in a second run of all analyses. However, this makes sense only if we are strictly interested in comparing two groups. In these analyses, I make comparisons among three groups: outcomes in male-typed versus mixed-typed occupations and in male- versus female-typed occupations. To make these results comparable between each other, I fix A to represent women in male-typed occupations.
See Blau and Kahn (1997:6–8) for excellent insight into what is behind ∆θt.
In some studies, the observed and unobserved interactions are included in one of the other components, but I ignore them for two reasons: (1) these interactions do not have a substantively useful interpretation for this study, and (2) I would find it misleading to add the joint contribution of quantities and prices to either the quantities or the prices component.
These changes are substantial in size. In 2015 euros, changes in wage differences amount to 4.55 EUR (vs. female-typed occupations) and 2.5 EUR (mixed occupations). In comparison, mean wages of the entire female population in 2015 were about 16 EUR.
Following Pannenberg (2005), I also run analyses on a restricted data set until 2014, showing that uncompensated overtime contributed similarly to increasing wage disparities as overwork (results not displayed).
As discussed earlier, several changes occurred in the early to mid-2000s: women started to integrate more quickly into paid work; returns to education were rising; changes in family policy were implemented; and women who had grown up in a more gender-egalitarian society entered work.
Results are based on an interaction term of the gender type variable and the biennial period dummy variables. Bold figures designate significance at p < .05.
Although the supply of women with a tertiary degree has increased steeply since 1992, this was also the case for men (see Figure A3, online appendix). Hence, it is not plausible that there was demand-driven upgrading skewed toward female workers because of a lack of men with sufficient qualifications.
This result is based on an Oaxaca-Blinder decomposition (Kitagawa 1955) of current wage differences. Results are not displayed.
A wage bonus for those in caring activities is found in the early 2000s. Personal care can be considered a strongly female-typed activity, and its positive effect on wages is another indication that female-typed work has not been culturally devalued in Germany in the observed period.
In pooled regressions (results not displayed), I find that across all occupational groups, having children is positively associated with wages, which is another indicator for the positive selection of employed mothers.
A slight inverse trend is observed for only the 1999–2007 subperiod.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.