The Role of Sex Segregation in the Gender Wage Gap Among University Graduates in Germany

1 Introduction

In industrialized countries, women’s economic situation in the labor market has substantially changed over the last half century (see, e.g. Blau and Kahn 2017). Despite the persistence of social gender roles that assign men and women not only different characteristics but also obligations (women still bear the main responsibility for household chores and childcare), women have significantly changed their work behaviour and, e.g. strongly increased their labor market participation. Women have also not only caught up with men in terms of education, they are now even more likely to have a university degree than men. Less progress has been made with respect to sex segregation. While women have entered many occupational fields as well as fields of study, segregation persists until today. This is the case despite political efforts to increase the number of women in male-dominated fields, in particular STEM-fields, which are also motivated by a shortage of workers in these domains. In this paper, we focus on Germany, a country where the segregation into fields of study is exceptionally high (Charles and Bradley 2009). Germany is also one of the EU-countries with strong occupational segregation (Sparreboom 2014). Women’s low numbers in male-dominated fields are of concern because earnings tend to be higher there (see, e.g. Harris 2022; Levanon et al. 2009). Thus, occupational segregation is an important contributor to the gender wage gap, which is particular large in Germany. For example, in 2019, with a wage differential of 19.2%, Germany was the EU-country with the fourth highest gender pay gap after Estonia, Latvia and Austria (Eurostat 2021).

In this paper we examine data from the German Centre for Higher Education Research and Science Studies (DZHW, formerly HIS) for the graduate cohorts 1997–2013 to evaluate the wage effects of female-dominated fields. There are two main reasons why it is of particular interest to examine the gender wage gap among university graduates. First, at the stage of labor market entry, university graduates are very homogeneous with respect to their human capital.^[1] Typically, these young adults have no family obligations and, thus, had no career interruptions yet; they have little job experience and, thus, do not differ much with respect to their on-the-job training. Because they only enter the labor market, their earnings are not affected by previous promotion decisions or wage negotiation outcomes. Consequently, when examining this group, we have to worry less about differences in characteristics that may be observed by employers and determine wages while they remain unobserved by the researcher. Second, apart from this methodological reason, entry wage differences are also economically important because first jobs set the tone for future careers and are therefore consequential for future earnings. This is not only because lower wages typically come with lower raises. Also, pay increases are added to the initial base wage. Hence, starting one’s work career with a discriminatory low wage can accumulate to a marked income gap over one’s lifetime. Finally, outside offers from other firms are also often based on previous earnings, additionally amplifying the importance of entry wages (Graham et al. 2000: 11).

Our paper is inspired by Leuze and Strauß (2009) who focused on the 1997 graduate cohort in the DZHW data to examine the effect of female-dominated fields of study as well as occupations on the gender wage gap. Apart from a strong negative association between female-dominated subjects/occupations and hourly wages, the authors found their hypotheses confirmed that, first, men experience a stronger wage penalty than women when graduating in a female-dominated field of study, and, second, women are more penalized for entering female-dominated professions than men.

In this paper we revisit these issues also for later cohorts. We do so for multiple reasons: First, over the last decades attempts have been made to increase the number of women in particular in STEM-fields, which, if successful, may have led to some desegregation and a decrease in the gender wage gap among university graduates. Second, attempts have been made to revalue the work primarily done by women and recent literature suggests to some success. For example, using U.S. census data, Busch (2018) found the devaluation of female-dominated occupations to slowly decline from 1960 to 2010. Third, Francesconi and Parey (2018) have shown that in the DZHW data there is some fluctuation concerning the gender wage gap among university graduates over the years that does not quite match the relative stability, and periodically moderate decline, of the overall gender wage gap in Germany since the late 1990s (Gallego Granados and Wrohlich 2018). For this reason, results based on one particular cohort of graduates may not hold more generally and, thus, lead to erroneous conclusions.

In our study that is based on multiple cohorts of graduates represented in the DZHW data, we confirm the very strong negative association between female presence in a subject or occupation and wages. However, no consistent pattern emerges with regard to whether men’s or women’s wages suffer larger penalties. There is also no time trend discernable with respect to the wage penalty for female-dominated fields. We further show that significant wage gaps exist within fields of study, especially in some male-dominated fields like engineering and natural science.

Our paper is structured as follows: In the next section, we review the literature on the gender wage gap among university graduates. We then give an overview over theories that have been used to explain why men and women choose different fields, why female-dominated occupations pay low wages, and why pay penalties may differ for men and women. We also refer to some related empirical literature. Next, we describe our data, also with respect to female shares within fields of study and occupations over time. After a short methods section, we present our results. The final section concludes.

2 Gender Gaps Among University Graduates

In recent years, the potential advantages of examining gender wage gaps among graduates at the start of their careers motivated several studies. Only few studies found similar wages upon graduation that drifted apart over the years to come (for example, Albrecht et al. (2018) and Bertrand et al. (2010) for professionals with a business (or economics) degree in Sweden and the USA). Most studies identified gender wage gaps already upon labor market entry, for the U.S. (e.g. Graham et al. 2000; McDonald and Thornton 2007) as well as for Germany (Behr and Theune 2018; Bredtmann and Otten 2014; Francesconi and Parey 2018; Leuze and Strauß 2009; Reimer and Schröder 2006). For example, Reimer and Schröder (2006) showed that within the particularly homogenous group of social science graduates from just one German university, females earned 5% less at their first job – a number that actually increased to 7% once controlling for human capital characteristics. Bredtmann and Otten (2014) examined business and economics graduates from one other university. From an approx. 8% wage differential, about 2 percentage points could be explained by differences in endowments. Using data from the nationwide 2001 DZHW graduate panel, Behr and Theune (2018) found that upon labor market entry, men earn higher wages at all percentiles of the wage distribution. Concerning other labor market outcomes than wages, Wieschke (2018) showed for the German state of Bavaria that female university graduates enter the labor market at less favourable terms (e.g. they are less likely to hold a permanent contract or an executive position at their first job), so that their job satisfaction is lower. This explains why female graduates change their job more frequently.

The importance of subject of degree on the gender wage differential has been emphasized by various studies (e.g. Chevalier 2007; Graham et al. 2000; McDonald and Thornton 2007). For example, Sánchez-Mangas and Sánchez-Marcos (2021) examined first job wages for college graduates in 11 European countries and found that the gender wage gap strongly differs between fields, with subjects from STEM and Health yielding substantial unexplained wage penalties for women. Studies did not only look at gender wage gaps in different fields, but also sought to explain to what degree subject choice could explain gender wage gaps. For instance, Machin and Puhani (2003) found for the United Kingdom and Germany that subject choice makes a significant part (8–20%) of the gender wage gap (looking at people at different stages of their career). Focusing at first jobs only, Braakmann (2013) identified that field of study even explained up to 70% of the gender wage gap of German university graduates from the cohort of 1997.^[2]

To the best of our knowledge, Francesconi and Parey (2018) are the only authors who previously made use of the entire set of student cohorts who have been surveyed by the DZHW in Germany from 1989 to 2009.^[3] They estimated various gender gaps among recent university graduates and found that, amongst others, women started their studies with somewhat better school grades, but finished with slightly poorer grades than men. This may be due to a larger number of university dropouts among men. With regards to income, the authors found an average earnings penalty for women of 29 log points, with the field of study being the most important factor explaining this gap.

In this paper we revisit the findings from Leuze and Strauß (2009), who used the DZHW data on the 1997 graduate cohort to examine the effect of female-dominated fields of study as well as occupations on the gender wage gap. They found a negative association between female-dominated majors/occupations and hourly wages, with a male advantage among individuals who graduated in a female-dominated major, and a female advantage among individuals working in female-dominated occupations. In this paper, we examine for a series of graduate cohorts whether these findings hold more generally.

3 Theoretical and Empirical Background on Sex Segregation and Pay

3.5 Segregation Over Time

In the following, we illustrate to what degree the level of gender segregation has changed within subjects and occupations over time. Figure 1 depicts the female shares for subjects for the years 1998 and 2013 (approximating the graduation years of our first and last student cohort).^[13] The figure illustrates the female shares (as provided by the German Federal Statistical Office) for all subjects which were represented in the DZHW data. The dashed line depicts the identity line. Subjects above the 45° line increased their female share between 1998 and 2013. In line with increasing numbers of women striving for higher education, Figure 1 documents that over the observed period, most subjects increased their relative numbers of female students. Across subjects, the female share increased by 4 percentage points on average.

Figure 1:

Sex segregation in subjects 1998 and 2013.Notes: The data used to calculate the female shares within subjects come from the German Federal Statistical Office.

The colored circles depict those subjects in which the female share of students increased by at least 8 percentage points. The highest gain was made in dentistry (+14 pp.) followed by high gains in architecture, economics and business administration, physics, medicine and industrial engineering (+11 pp.). The subject “natural sciences interdisciplinary” sticks out by its originally high share of female students and its large decline over time (−22 pp.). However, note that it represents only a residual category, where interdisciplinary subjects which also cover mathematics and/or natural science fields are grouped together. It is also populated by only few students and the number of students has further declined over time.^[14]

Figure 2 visualizes the change of female shares within occupations (as provided by the German Federal Employment Agency) between 1997 and 2009.^[15] The figure includes only occupations in which respondents of the DZHW graduates panel worked at the time of the survey. Again, we see an increase in female representation in the majority of occupations. Across occupations, the female share rose by 1 percentage point on average. The colored circles mark those occupations with the highest changes in female shares (by at least 10 percentage points). The occupations with the highest relative increase of women were veterinarian (+19 pp.) and machine cleaning (+18 pp.). In contrast to the fields of study, there is also a large fraction of occupations with a decreasing share of females. The occupation photo lab technician (−28 pp.) and the participation in the family business (“help FB not agriculture”) (−19 pp.) experienced the largest drop.

Figure 2:

Sex segregation in occupations 1997 and 2009.Notes: “Help FB” stands for help in family business. The data used to calculate the female shares within occupations come from the Federal Employment Agency.

We also calculated the Duncan Segregation Index (Duncan and Duncan 1955), which provides the most well-known indicator for the un/equal distribution of the sexes, for both subjects and occupations. It indicates how many men or women would have to switch fields to reach an equal gender representation. A Duncan index of 0 means that women and men are represented equally, 1 indicates that there is a full segregation between women and men. We find that the Duncan Segregation Index did not change much over the considered years for both subjects and occupations. In our sample, across years, the Duncan Index for Segregations is around 0.3 for subjects and 0.6 for occupations. While the index is virtually constant over time as far as subjects of studies are concerned, the index for occupations experienced a slight decline from 0.623 in 1997 to 0.581 in 2009. Thus, despite the rising number of women in higher education and a large fraction of women participating in the labor market, sex segregation persists today.

4 Method

We test whether there is a correlation between female-dominated occupations/subjects and wages by estimating the following linear regression model, first for each year separately, then for the pooled sample:

As a depended variable we use the logarithmic hourly wage.^[16] F i is a female dummy and its coefficient α 1 captures the wage penalty for women. In our model, the coefficients of interest are γ₁–γ₄, as they depict the relationship between the representation of women in subjects/occupations and wages. We use the dummies female-dominated subject (FS), integrated subject (IS), female-dominated occupation (FO) and integrated occupation (IO) to model this relationship. We further control for a large set of covariables X _i capturing the field of study, various job and firm specific characteristics, education and work history as well as family life. For example, the job specific characteristics include information on whether the job is an internship, a part-time job, or a management job. Firm specific characteristics capture whether an employer belongs to the public sector, has more than 500 employees, and is located in East Germany. Education specific characteristics include information about the highest tertiary degree obtained (Bachelor, PhD or other^[17]), whether a graduate is still enrolled in a PhD program, his/her final grade at university and age at graduation. Concerning the work history of a graduate, the covariables capture whether an individual worked prior or during his/her studies, his/her months unemployed, months as a homemaker and months of overall work experience. Additionally, the family specific characteristics include dummy variables for marital status, for having children and for whether (at least) one parent has a high school diploma (Abitur).

To examine whether men or women are penalized more for participating in a female-dominated or integrated field, we extend model (1) by including a set of interaction terms:

Our coefficients of interest in this regard are η₁–η₄, as they examine to what degree women’s wages are affected differently than those of men by a particular representation of women in a field. A negative coefficient indicates that women are penalized more strongly than men for being in a more female-dominated field. We also include controls for our covariables X _i interacted with female sex. The variable ε _i depicts the error term.

5 Results

In a first step, we plot the average conditional wages (in log points) based on our regression model (2): Figure 3 shows these wages for the different subject types, Figure 4 for different occupational types. We estimated our regression model as depicted in equation (2) and calculated the predicted wages for each of our observations. The average wages presented in Figures 3 and 4 are based on these predicted wages and calculated for six different subcategories: Men or women in female-, male-dominated or integrated subjects (or occupations). In Figure 3, we see that, according to our model, men earn more than women in all subject types (male-dominated, integrated and female-dominated). Also, within one sex, wages are always highest for graduates from male-dominated subjects, followed by integrated and female-dominated subjects. As a result, the average conditional wages are highest for men who graduated in male-dominated subjects and lowest for women who graduated in female-dominated subjects. For occupations, we find a similar scenario. In Figure 4, we see that men earn the highest wages when working in male-dominated occupations, lower wages in integrated occupations, and the lowest wages in female-dominated occupations. Our model also predicts that women tend to earn most in male-dominated followed by integrated and female-dominated occupations. Only in the years 2009 and 2013, female average conditional wages are similar in male-dominated and integrated occupations. In almost all years, women earn less than men within one occupational type. Only in 2013, our model predicts that men and women received similarly low wages in female-dominated occupations.

Figure 3:

Average conditional log wages for different subject types by sex, 1997–2013.

Figure 4:

Average conditional log wages for different occupational types by sex, 1997–2013.

While Figures 3 and 4 give an overview of different average conditional wages depending on the type of occupation or subject over the years, in the next step, we present the findings from our linear regression model. Table 3 presents the results for the five different cohorts of graduates (1997–2013) and provides detailed information on how the female wage penalty changes across different specifications and which of the coefficients of interest are statistically significant. Column (1) shows the “raw” gap, while column (2) includes the field of study (education, health, agriculture, engineering, arts, natural sciences, law, social sciences, economics and business administration, ref: humanities) in which individuals have graduated. Column (3) further includes dummies capturing whether the subject in which an individual has graduated is female-dominated or integrated (ref: male-dominated), in column (4) dummies capture whether the individual is employed in a female-dominated or integrated job (ref: male-dominated). Column (5) includes both sets of “sex segregation” dummies and column (6) additionally controls for a large set of individual characteristics (concerning education and work history as well as family specific characteristics), job specific characteristics (internship, part-time, management position) and firm specific characteristics (public/private sector, company location East/West Germany, large firm). In the final column (7), the full specification additionally includes interaction terms between the female dummy and all covariables.

As column (1) shows, the “raw” gender gap among university graduates fluctuated between 25 and 28 log points (or 22–24%)^[18] for the cohorts 1997–2005 and has declined to 23.5 log points (21%) for the 2009 and 2013 cohorts. Controlling for the field of study explains a large portion of that gap (column 2). In most years, adding this variable reduces the coefficient of the female dummy by approximately 10 log points; in 1997 it is reduced by half. As column (3) illustrates, wages are substantially lower for graduates who studied a female-dominated or integrated subject instead of one that is male-dominated. The effects are very large, but also fluctuate significantly between the different time periods. While, for example, in 1997, graduating in a female-dominated subject reduced wages by 16 log points (15%), this effect amounted to 44 log points (35%) in 2013. Also, the sex segregation of occupations strongly affects wages. According to column (4), female-dominated jobs pay 26–35 log points (23–29%) less than male-dominated occupations. Penalties also exist for integrated occupations, but are much lower. Including the sex-type of field of study and occupation simultaneously (column 5) reduces the gender wage gap (compared to column 2) by approximately 5–7 log points. The remaining wage differential amounts to 7–10% for the different time periods. Other individual or firm-specific characteristics explain relatively little. After their inclusion (column 6), the gender wage gap amounts to 5–6% and is remarkably constant over time. While this wage differential is much lower than the raw gap, it is still substantial considering that it compares young graduates with identical qualification and similar individual characteristics who work at similar firms.

When we include the interaction terms in column 7, we find no clear gendered pattern concerning the effects of female-dominated subjects: in some years the interaction term female subject*female is insignificantly positive, in others negative. Only in one year, 2013, there is a statistically significant gender difference, as women who completed a female- instead of a male-dominated subject experience a wage penalty of 27 log points while there is no statistically significant penalty for men. Also, when comparing the penalties that men and women experience for studying an integrated subject, there is no consistent discernible gender difference over the years.^[19] Therefore, we find no support in our empirical setting for the hypothesis that men are more stigmatized than women when they graduate in a female-dominated academic subject as suggested by Leuze and Strauß (2014).

Concerning the effect of sex-segregated occupations, if there is a “glass escalator” (Williams 1992) for men, we would expect that they are punished less for entering a female-dominated occupation. This would also be the case if men are pushed into somewhat higher positions than women (unobservable in the data) because they are considered more status worthy (Correll and Ridgeway 2003). This is not what our results in column (7) show. In no single year do we find statistically significant differences between the effects that a female-dominated occupation has on female versus male wages (i.e. a significant coefficient for the interaction term [FO*female]). We therefore cannot confirm the “glass escalator” or “status expectation” hypothesis for entry wages in our setting..^[20] There is only some evidence that women are punished more than men for entering an “integrated occupation” (for the years 1997, 2001, and 2005).

Table 4 shows the full set of results for the pooled data (1997–2013) including year dummies.^[21] Columns (1)–(6) follow the same structure as in Table 3 and confirm for the pooled sample that female-dominated and integrated subjects/occupations pay less than male-dominated ones. The results for the fully interacted model are presented in the last two columns: Column (7) “Main Effects” displays the main effects of the covariables, while column (7 cont.) “IA Effects” presents the interaction effects between the respective covariable and the female dummy. Concerning the interaction terms of being female and having graduated in a female-dominated subject or working in a female-dominated occupation (our variables of interest), we do not find significant sex differences also in this pooled sample. However, while there is no significant sex difference for graduating in an integrated subject, we find a wage penalty for working in an integrated occupation only for women (column 7 cont.) but not for men (column 7).

Concerning the covariates, the findings show that most fields of study yield higher earnings than the humanities (reference category). Incomes are particular high for engineering, economics and business administration, natural science and health (column 2). Note that because only university graduates are examined, the less qualified health workers who are known for being poorly paid are not included here. In their first year after graduation, graduates with a law degree earn less because after graduating from university, they still have to complete a two-year legal clerkship (Rechtsreferandariat) and pass another state examination before they fully finish their education. With respect to education specific characteristics (column 6), Bachelor graduates earn approx. 14 and PhDs 9 log points less than master and other graduates (that together constitute the reference category). Individuals, who are older upon graduation earn somewhat more. This effect is mainly driven by women (column 7 cont.). Concerning job specific characteristics, internships pay lower hourly wages, while management positions pay more – the latter is particularly true for women, who receive a premium of 10 log points for working in management, while men only make additional 5 log points. With regard to firm specific characteristics, we find that women benefit from being employed in the public sector, while there is no effect for men. Wages are significantly higher for large firms (+16 log points) and lower in the region of the former German Democratic Republic (approx. −14 log points). Concerning the work history of employees, months of unemployment are associated with somewhat lower wages and months of work experience are related to higher wages. If individuals worked during (or prior to) their studies, their wages are higher. The effect of months as homemakers is significantly negative. Finally, our results show that the people in our sample who are married and/or have children earn significantly higher wages (the interaction effects for women are relatively large and negative, but statistically insignificant).

Note that in our procedure we used listwise deletion to handle missing values, which means that we dropped observations from our sample if they had missing values for one of the covariables. Because in each column in both, Tables 3 and 4, we added new covariables to our specification, our sample decreases from column (1) to (7).^[22] However, to rule out the possibility that the results of a particular specification depend on the particular sample examined, in Tables A3 and A4 in the Appendix, we also included the results for different specifications based on the reduced sample from column (7) and found similar dynamics as we detected in Tables 3 and 4.

6 Wage Gaps Within Fields

We have previously established that fields of study explain a large part of the gender wage gap among university graduates. In our last step, we now analyze the wage gaps within different fields of study. Table 5 summarizes the results for the years 1997–2013.^[23]

In Table 5 we report average male and female log wages within each field as well as the size of the observed wage gap. We further show which part of the gap can be explained by using the full set of covariables described before. However, different from the original setting, we do not include female shares of occupations and subjects, but control for the different subjects themselves.^[24] The findings in Table 5 are estimated using a Kitagawa–Blinder–Oaxaca decomposition (Blinder 1973; Kitagawa 1955; Oaxaca 1973)^[25] with a pooled regression as the baseline for the coefficients (Jann 2008).

We find the highest raw wage gaps in those two fields that have the lowest share of women: engineering (18 log points) and natural sciences (22 log points). These are followed by high wage gaps in education (15 log points) as well as in economics and business administration (13 log points). The smallest observed wage gaps are within the field of law (5 log points), in which overall wages are also the lowest.^[26] The highest unexplained wage gap can be found in economics and business administration (9 log points), engineering as well as in natural sciences (7 log points). For graduates from the natural sciences a particularly large part of the wage gap can be explained by gender specific differences in endowments. Interestingly, the unexplained wage gap in law is quite high at 6 log points, despite the low raw gap in the field. This is the case because in law women have similar endowments than men, so nothing of the raw gap can be explained. While the unexplained wage gap cannot be equated with the level of labor market discrimination, it captures the size of the wage gap which cannot be explained with the human capital variables used in our model. If our model misses crucial variables of interest, the size of the unexplained gap may overstate the level of discrimination. Variables that might be relevant in this context are preferences for flexible working hours or overtime work, the ability to commute, as well as individual characteristics such as competitiveness or negotiation skills (for a detailed discussion of potential factors affecting the wage gap see Blau and Kahn 2017).^[27] However, if covariables within our model are themselves affected by discrimination, the unexplained gap would underestimate the level of discrimination. For example, entry barriers into specific occupations or study subjects, as well as discrimination within professions or fields of study may affect human capital variables and, therefore, influence the wage gap. We cannot rule out that these concerns affect the wage gap in some fields more than others.

Overall, our findings show that women make relatively high wages when studying a subject in the field of engineering or natural sciences, in particular when compared to, e.g. the humanities or the arts. However, the gender wage gap within the field is among the highest for STEM subjects. When women compare themselves to their peers and worry about their relative earnings (Duesenberry 1949), these large gender wage gaps – in addition to other potential job disamenities – may make these fields unattractive to women despite their overall high wages.^[28]

7 Conclusion and Discussion

Our findings support the hypothesis that a higher female share of workers within an occupation is associated with lower wages. The same is true for subjects with a high share of female students. This negative correlation of female shares of workers/students and wages that we find, is in accordance with devaluation theory and a lower status that is given to work conducted by women. However, also other reasons may be responsible for the negative association between female shares and wages. First, even though women’s labor market participation has drastically increased since the 1970s, sex differences in labor market commitment remain, e.g. because women more often take a career break due to parenthood. If the depreciation of human capital is lower in some fields than others, these may be more attractive for women who expect to interrupt their labor market participation at some stage (Polachek 1981). For example, if female-dominated-jobs require less specialized training (Tam 1997), they may be more attractive for women, who expect to interrupt their careers.

Second and relatedly, omitted variables could bias the results. For example, different career attitudes and job characteristics could be responsible for the differences in earnings. According to the theory of compensating differentials (Rosen 1986), if a job offers non-financial amenities such as flexible hours or working from home (which are especially valued by mothers), it may pay lower wages – as long as there are enough workers who sufficiently value these non-financial benefits. Unpleasant working conditions, on the other hand, have to be compensated by higher wages. For example, jobs that require long hours pay significantly higher wages as remuneration (Goldin 2014). Long hours, thus, may not only drive sex segregation but also cause differences in wages between male and female-dominated professions. For example, Leuze and Strauß (2016) confirmed for Germany that occupational long hours increase wages. Reimer and Schröder (2006), however, did not find that controlling for job attitudes of employees (that measured what they valued in a job, e.g. income, work/life balance, work content) decreased the gender wage gap.

Third, while devaluation theory suggests that an occupation is devalued and wages fall when women enter the profession, it may also be that this causality is reversed, and low wages lead to the feminization of a profession – because men are not available for these jobs. According to queuing theory (Reskin and Roos 1990) employers rank potential employees according to their place in the societal hierarchy and hire people with higher social status first. Therefore, high paying jobs are filled with men first – women may then have to turn to occupations with lower pay. Some longitudinal studies have tested the “queuing” against the “devaluation” theory and typically found stronger evidence for the latter (England et al. 2007; Levanon et al. 2009). Last, low wages in female-dominated occupations can also occur because of crowding (Bergmann 1971, 1974). If women are crowded into a few occupational fields, the oversupply of female employees in these occupations pushes down wages. Such crowding may result, e.g. from discrimination or traditional gender norms.

In our setting, we cannot distinguish between these potential reasons. However, we are able to confirm a negative association between female-dominated subjects/occupations and wages for German university graduates across multiple cohorts that finished their studies between 1997 and 2013. We do not find a clear time trend but in our data, there is no indication that the devaluation of female-dominated fields has decreased over the years as suggested by Busch (2018).

In our study, we were also interested in whether men’s or women’s wages suffer larger penalties as a result of a high female share in a field. In contrast to Leuze and Strauß (2014), we did not find conclusive evidence that wage penalties accompanied with a high share of females are different for women and men. Overall, we found that women earn less than men in male-dominated fields, but it is not true that women earn more than men in female-dominated fields, as would be suggested by a “lack of fit” framework that attributes women a worse fit to male-dominated fields but a better fit to female-dominated fields. Indeed, women tend to earn less than men more generally, so a different explanation is required. Apart from unobserved heterogeneity, discrimination may be one possible reason for our findings. Employers may still regard women as “secondary earners” and pay them relatively little as a result. It may also be that because women’s work is socially devalued, not only female-dominated fields are more poorly paid compared to male-dominated fields, but also women are paid less than men.

We finally showed that significant wage gaps exist within fields of study, especially in some male-dominated STEM fields like engineering and natural science. If women care about their relative incomes and compare themselves to their peers, these large gender wage gaps will not help make these fields attractive to women despite their high wages.