The Role of Math and Language Performance in Explaining the Gender Gap in STEM Major Choice. A Test for Germany

1 Introduction

For over a decade now, in most Western industrialized countries, women have outnumbered men in graduating from higher education (UNECE Statistical Database 2021). However, a continuing source of gender inequality lies in women’s and men’s distinct study field choices (DiPrete & Buchmann 2013). Across the globe, women persistently less often major in science, technology, engineering, and mathematics (STEM) than men (Charles & Bradley 2009; OECD 2017). This gender gap, which predominantly reflects a gender gap in STEM major choice rather than STEM persistence (Griffith 2010; OECD 2017)^[1], is troublesome from a societal perspective. The gender gap in STEM major choice increases existing economic inequalities between men and women because STEM professions are, on average, better paid, more job-secure, and with better career prospects than non-STEM professions (Anger et al. 2016; Black et al. 2008; Christie & Shannon 2001; OECD 2017). The gender gap in STEM major choice also increases shortages of STEM graduates in labor markets and limits economic growth (for the U. S., see National Academy of Sciences 2007; for Germany, see Expertenkommission Forschung und Innovation 2013).

Because of its importance, studies investigated a wide array of factors potentially contributing to the gender gap in STEM major choice. These range from structural factors, such as school performance, household financial constraints, and sex discrimination, to cultural factors, such as job and life goal preferences, gender-role expectations, and male-dominating cultures in STEM (for overviews see Blickenstaff 2005; Ganley et al. 2018; Kanny et al. 2014; Wang & Degol 2017). However, studies could not sufficiently account for gender gaps in STEM major choices with these factors. This also holds for studies testing the role of school performance, also known as “achievement” or “ability”, on which we focus in this study. Studies for the U. S. and the U. K., which cover most studies done on the role of school performance, show that math and language test competencies and grades account for at most one-fifth of the gender gap in STEM major choice (Hyde et al. 2008; Mann & DiPrete 2013; Morgan et al. 2013; Riegle-Crumb et al. 2012; Van der Werfhorst et al. 2003; Xie & Shauman 2003). Although math competencies and grades have a positive effect on the likelihood of STEM declaration, gender differences in math performance are too low in the Anglo-Saxon context to account for a substantial portion of the gender gap in STEM major choice (Hyde et al. 2008; Mann & DiPrete 2013; Xie & Shauman 2003). For recent cohorts in the U. S., math grades even favour women (Riegle-Crumb et al. 2012). Furthermore, language competencies are not associated with the STEM major choice (Mann & DiPrete 2013; Riegle-Crumb et al. 2012).^[2] Studies for the U. S. and the U. K. report that a student’s difference in grades for math and language (or difference in grades for science versus social sciences and humanities subjects; cf. Van der Werfhorst et al. 2003) can neither account for gender gaps in STEM declaration. Although a “comparative math grade advantage” (e. g., better math than language grades) is positively associated with the likelihood of a STEM major choice, gender differences in comparative math grade advantages are too small to contribute to the gender STEM gap (Riegle-Crumb et al. 2012; Van der Werfhorst et al. 2003). Gender differences in math-to-language tests are larger, but these comparative test advantages are not associated with STEM declaration (Riegle-Crumb et al. 2012). Therefore, scholars dismissed the school performance explanation of the gender gap in STEM major choice despite substantial school performance effects.

In this study, we aim to contribute to the literature on the gender gap in STEM major choice by assessing the role of school performance – math and language competencies and grades – for a context with considerable gender differences in math and language competencies and grades, namely Germany. Similar to the U. S. and the U. K., Germany displays gender-traditional math and language competency patterns. However, gender differences in math tests are considerably larger in Germany than in the U. S. and the U. K. during the period surveyed here (cf. OECD 2016). Grade patterns are also more gender-traditional in Germany: men achieve somewhat higher grades in math than women (Lippmann & Senik 2018), and women achieve notably higher grades in German than men (Bayer et al. 2021). German men, therefore, have a comparative advantage in math performance and German women in language performance (Lörz et al. 2011). This gender-traditional achievement pattern may be the outcome of Germany’s historically dominant male-breadwinner/female-carer model, its traditional gender institutions (e. g., extended maternity leaves and low childcare provision) and norms (Geisler & Kreyenfeld 2019; Lippmann & Senik 2018; Weinhardt 2017)^[3], and its early selection in education (Breda & Napp 2019; Lorenz & Schneebaum 2024). The gender-traditional achievement pattern may account to a stronger extent for the gender gap in STEM major choice than was found in the Anglo-Saxon context. Thus, our study may nuance conclusions from earlier studies on the school performance explanation of gender STEM gaps.

Prior studies for Germany indicate a substantial gender gap in STEM major choice of about 30 percentage points (Jacob et al. 2020, using data for 2003/2004; Uunk et al. 2019, using data for the school year 2010/2011). Uunk et al. (2019) found that this gap can be attributed to math and German language grades for one-fifth. Yet, Uunk et al. (2019) did not include competency measures and may have underestimated school performance’s role in accounting for gender differences in STEM major choice. In addition, their analyses were on students at general universities and did not include students of universities of applied sciences, students who also belong to higher education in Germany, are substantial in number, and may choose STEM and non-STEM majors (albeit a more limited range of possible subjects). Another study for Germany, by Lörz et al. (2011), observed that math and language grades accounted only for nine per cent of the gender gap in STEM. However, the study focused on study intentions (preference for a technical field of study in higher education among secondary school matriculates) instead of major choice. This may pose a problem as study intentions may differ from factual study choices, and this difference may be gender-specific (women less often realize their study plans than men; cf. Lörz et al. 2010). We improve on these studies for Germany by assessing the role of math and language competencies and grades in accounting for the gender gap in STEM major choice in higher education. The following section (section 2) discusses the German educational context, describing some relevant characteristics. Section 3 presents hypotheses, section 4 presents the data, measures, and methods, section 5 presents the findings, and section 6 ends with conclusions and a discussion.

2 The German Educational Context

The German educational system differs in some relevant respects from the Anglo-Saxon context and other advanced industrialized countries. First, Germany is characterized by early selection and a relatively high level of stratification in secondary education. For a long time, the German education system was characterized by a three-tier structure in which pupils were assigned to the Hauptschule, Realschule, or Gymnasium after primary school at the age of 10 or 12 (Helbig & Nikolai 2015). In recent decades, the German school system has changed, and an increasing number of federal states have introduced a two-tier education system structure, which only distinguishes between schools that prepare students for academic education at universities and those that provide a more vocationally oriented education (ibid.). The permeability of the German education system has also increased, making it easier to move between the educational tracks (ibid.). Yet, most first-year students come from upper secondary Gymnasium-level (about 80 per cent; Lörz 2013), and the choice of majors is restricted for those who come via vocational education. In such systems, with early selection and relatively high stratification, gender differences in subject specialization may be strengthened (Breda & Napp, 2019; Lorenz & Schneebaum, 2024). On the other hand, the high level of standardization in secondary education in Germany – there are compulsory courses in math, science, and German – may reduce such gender differences (cf. Ayalon & Livneh 2013).

Second, in Germany, test scores are hardly known to students and are not used for secondary and tertiary education selection. Grades, conversely, are known to students. They are essential for performance feedback (Bayer et al. 2021) and are also important for job applications. This feature of the German educational system may strengthen the role of grades in selecting study majors. On the other hand, for enrollment and study choice in higher education, grades do not matter once the entry qualification (Abitur) has been obtained, and neither does the subject choice at secondary school. Only when enrollment quotas exist are students selected based on grades (these quotas often exist for medicine, law, and psychology; in the case of medicine, students are also selected based on test competencies). That feature of German education may lower the impact of prior school performance on study field choices compared to systems that select stronger on these school performance measures (cf. Jacob et al. 2020).

Third, the major choice in Germany is already at the application stage. In some other systems, including the U. S., students select their major after college application. In contrast to the previous feature, this institutional feature may increase the relevance of prior (secondary) school performance for major choices from students’ perspective. Students weigh what they are good at, and choose a major accordingly.

3 Hypotheses

Generally, the choice of a study major can be seen as an outcome of an implicit, rational decision-making process (Jonsson 1999; Lörz et al. 2011). Prospective students weigh the expected costs and benefits of different majors and the likelihood of succeeding in these majors and choose the major that is expected to give the highest utility (benefits minus costs) and the greatest likelihood of success. These assumptions are also central to the expectancy-value theory, which addresses the role of individual expectations for study success and the subjective value of study fields for individuals (Eccles 2009). Of course, students will not consider all possible majors. Instead, they will choose from majors known to them and for which they may qualify, and they will rank these majors in attractiveness.

The school performance explanation of gender differences in STEM major choice centers on the likelihood of success argument. It assumes that math and language performance increase the probability of succeeding – or decrease the risk of failure – in STEM, respectively, non-STEM majors (Jonsson 1999; Eccles 2009), and therefore are important factors in choosing these majors.^[4] Higher performance in math will increase the likelihood of success in STEM majors and choosing these majors because of the generally higher math intensity of STEM majors. Vice versa, higher performance in the German language will increase the likelihood of success in non-STEM majors and choosing these majors because of the generally higher stress on language in non-STEM education. The literature indeed indicated positive math performance (test scores and grades) and negative language performance (grades but not test scores) on the odds of a STEM versus non-STEM major choice, both for the Anglo-Saxon area (Mann & DiPrete 2013; Riegle-Crumb et al. 2012) and for Germany regarding grades (Uunk et al., 2019). Furthermore, men and women may compare their grades in math and language and specialize in things they are good at because this gives them the highest probability of study success and minimal effort required to complete a major (Correll 2001; Jonsson 1999). Comparative math-to-language advantage is positively associated with STEM declaration when it concerns grades (Riegle-Crumb et al. 2012; Uunk et al. 2019), yet not competencies (Mann & DiPrete 2013; Riegle-Crumb et al. 2012).

The school performance explanation of gender differences in STEM major choice furthermore assumes gender differences in math and language performance. As stated, for Germany, these gender differences are considerable and do not only pertain to competencies – men score better in math tests and women in reading tests (OECD 2016) – but also to grades. During the period studied here, around 2013–2014, men achieved higher grades in math than women (Lippmann & Senik 2018), and women achieved notably higher grades in German than men in upper secondary school (Bayer et al. 2021). In addition, men’s grades in math are higher than in language and vice versa for women, which gives men a comparative advantage in math and women a comparative advantage in language at the upper secondary level (Lörz et al. 2011). These gender differences in math and language performance in Germany will – given the presumed effects of math and language performance on the choice of a STEM major – likely contribute to gendered STEM major choices. From the school performance perspective, therefore, we expect that the gender gap in STEM major choice in Germany can be attributed to performance in math and the German language (H1).

Next, we expect that the gender gap in STEM major choice in Germany can to a larger extent be explained by (math and German) grades than (math and German) competencies (H2). Although men and women in Germany substantially differ in math and language test competencies (as a proportion of the pooled standard deviation, this gender difference is even larger than in grades; see findings below), the effects of competencies on the STEM major choice can be assumed to be weaker than of grades. In Germany, students know their school grades, yet test scores are hardly known. In addition, school grades are relevant to job applications and for future labor market success. Even if grades do not reflect actual ability, for example, due to underassessment or overassessment by teachers, grades are relevant for study choices as one may expect such teacher bias in further education. Therefore, grades may be more important for STEM major declaration than competencies, even in the practical absence of grade selection in university admission in Germany. It is telling that in the U. S., where universities select students on test scores rather than grades, math grades have a stronger effect on STEM declaration than math test scores (Riegle-Crumb et al. 2012).

Finally, we expect that the gender gap in STEM major choice in Germany can to a larger extent be explained by German language grades than math grades (H3). Gender differences in math grades in Germany are low, though still at the disadvantage of women (Lippmann & Senik 2018; also see findings below). The relatively low gender difference in math grades may be attributed to women’s greater self-discipline and larger school investments in achieving good grades (cf. Duckworth & Seligman 2006). At the same time, women in Germany attain substantially higher grades in the German language than men, almost a full point on the 1-to-6 grade scale (Bayer et al. 2021; also see findings below). The larger gender differential in German than math grades implies – at presumably equal effects of math and German grades on STEM declaration – that German grades contribute stronger to the gender gap in STEM major choice than math grades, leading women more often into non-STEM majors than men.

4 Data, Measures, and Method

5 Findings

Table 1 lists means and standard deviations – pooled and by gender – of the potential explanatory factors of the gender gap in STEM major choice (math and language competencies and grades, advanced coursework, and other control variables). In line with prior studies for the U. S. and Germany (Riegle-Crumb et al. 2012; Lörz et al. 2011), we observe a small gender difference in absolute math grades in the last year of upper secondary school of 10 per cent of its standard deviation. However, unlike in the U. S., in Germany, the pattern is still gender-stereotypical, with men achieving higher grades in math than women (Lippmann & Senik 2018). Regarding language grades, the gender difference is larger and favours women (cf. Bayer et al. 2021). Women obtain considerably higher grades in German than men (the gender difference is 52 per cent of the standard deviation in German grades). In addition, since men’s grades in math are higher than in German (by 0.8 points) and women’s grades in German are higher than in math (by 0.8 points), men and women have a comparative grade advantage, respectively in math grades and German grades (cf. Lörz et al. 2011). The gender difference in the comparative grade advantage is also substantial (49 per cent of its standard deviation).

Test competencies in math and reading display the same gendered pattern, with male students achieving higher math and lower reading test scores than female students in the final year of upper secondary school. The gender difference in math test scores (61 per cent of the standard deviation) is larger than in reading test scores (23 per cent) – which is also shown for Germany in the Programme for International Student Assessment (PISA) 2015 study (cf. OECD 2016) – and also larger than in math grades (10 per cent). The gender difference in reading competencies is lower than in German grades (23 versus 52 per cent of the standard deviation). Thus, in Germany, men notably have stronger test competencies in math than women, and women have better grades in German than men.

Table 1 further shows a sizeable and statistically significant gender gap in the choice of a STEM major. While 54 per cent of the male Gymnasium students entering higher education choose STEM fields, only 23 per cent of female students do so. This difference implies a 31-percentage point gender gap in the choice of a STEM major or four times lower odds of women of a STEM rather than non-STEM major than men (cf. Jacob et al. 2020; Uunk et al. 2019).

The logistic regression in Table 2 shows that math and language performance affect the odds of a STEM versus non-STEM major choice. Of these performance measure, the grade in the German language has the strongest effect (recall that continuous variables are standardized): one standard deviation increase in the German grade (about 2.5 points) decreases the log odds of a STEM versus non-STEM major choice by 0.68 and the odds by a factor 2. This equals a 12 percentage points decrease in the likelihood of a STEM major choice, as shown by the average marginal effect. Math grades and math tests have the second and third strongest effects, with (logit) effect estimates of 0.44 and 0.34, respectively. This indicates that math grades and tests independently increase the odds of a STEM versus non-STEM major choice. However, reading competencies do not affect the likelihood of a STEM major choice, nor do they when the other performance measures are excluded (Appendix, Table A.5). Thus, math grades, math tests, and German grades matter for the choice of a STEM or non-STEM major in higher education in Germany but not reading competencies. Furthermore, the simultaneous existence of math and German grade effects suggests comparative advantage effects: for example, having a higher grade in math at a given German grade raises the odds of choosing a STEM versus non-STEM major.

Regarding the control variables, the effects of advanced coursework in math and German stand out. Taking an advanced math course raises the odds of choosing a STEM rather non-STEM major by a factor of two, and taking an advanced German course decreases these odds by a factor of 1.5. Additional analyses reported that it was important to account for these measures since the effect of math competencies on the STEM likelihood is substantially larger without coursework. This is hardly so for the other performance measures (Appendix Table A.6). Of the further control variables, the only statistically significant effect pertains to age, with older students having lower odds of STEM declaration. Additional analyses reported that being in a sample school in East Germany does not associate with the odds of a STEM major choice (Appendix Table A.7). The gender gap in STEM major choice is somewhat smaller in East than West Germany, but not overwhelmingly so (29 versus 32 percentage points). Finally, we observe a large effect of gender on the odds of choosing a STEM versus non-STEM major. The effect of gender is the largest of the covariates included. It indicates that women have substantially lower odds of entering a STEM major than men, even controlling for school performance and other measures. Women have 0.83 lower log odds and 2.3 lower odds of entering STEM majors than men at equal background and school performance characteristics. This equals a 17-percentage points corrected gender gap.^[9]

The decomposition analysis in the last column of Table 2 shows that math and language performance measures – grades and test scores – account, in total, for 45 per cent of the gender gap in STEM major choice in Germany. Although a significant gender gap remains, and therefore, a strong version of H1 is rejected, still nearly half of the gender gap in STEM major choice is explained by prior school performance. Yet, we also observe that mostly math competencies and German grades account for the gender gap in STEM major choice (17 and 24 per cent), and hardly math grades and reading competencies (both 2 per cent). That math competencies and German grades contribute so much to the gender gap in STEM major choice can be attributed to the large effects on the STEM major choice (cf. first column, Table 2) and the large gender differences in math competencies and German grades (cf. Table 1). The weak contribution of math grades and reading competencies to the gender gap in STEM major choice can be attributed to the small gender difference in math grades (cf. Table 1) and the insignificant effect of reading competencies on the STEM major choice (cf. first column, Table 2).

The above findings imply that the gender gap in STEM major choice can better be explained by grades (26 per cent in total) than competencies (19 per cent), in line with H2, but not overwhelmingly so. Regarding math performance, grades are even less important for the gender gap in STEM major choice than test scores (2 versus 17 per cent). That math test scores matter so much is surprising, given our arguments on the importance and visibility of grades. As stated, its strong contribution to the gender STEM gap can empirically be attributed to a substantial effect on the STEM major choice – though not larger than of math and German grades (cf. Table 2) – and a comparatively large gender difference in math test scores (cf. Table 1). Math grades are also associated with the likelihood of a STEM major choice, yet gender differences in math grades are too small to contribute to the gender gap in STEM major choices (cf. Table 1). Regarding language performance, on the other hand, grades are more important contributors of the gender STEM gap than competencies (24 per cent and 2 per cent), which is both due to a large effect on STEM declaration and large gender differences in German grades. The above findings imply that H3, stating a larger contribution of German than math grades to the gender gap in STEM major choice, is supported. This can be explained by the larger statistical effect of German than math grades on STEM declaration and the notably stronger gender differences in German than math grades.

We want to remark here that if we had only modelled competencies, we would have overestimated the role of competencies in explaining the gender gap in STEM major choice. This can be seen in Table 3. The explained gender STEM gap by competencies is 27 per cent when only modelling competencies (M6) and 19 per cent when modelling competencies and grades (M9). Overestimation does not pertain to the role of grades (cf. M5 and M9). We neither overestimate the role of math performance once we exclude language performance (cf. M7 and M9). Furthermore, we would underestimate the role of language performance in explaining the gender STEM gap if we had not controlled for math performance (18 versus 26 per cent; cf. M8 and M9). This is mainly because math grades suppress the effect of German grades on the likelihood of STEM major choice (compare M8 and M9 of Appendix, Table A.8). In addition, modelling the four performance measures separately (M1-M4, Table 3) underestimates the role of German grades and overestimates the role of math tests in accounting for the gender gap in STEM major choice. These findings underline that simultaneous estimation of school performance measures is important to obtain valid estimates of their contributions to the gender gap in STEM major choice.

Finally, our decomposition analyses show that advanced coursework in math and German forms an additional important explanation of the gender gap in STEM study choices. These measures account for 13 per cent of the gender gap, only six percentage points less than test competence measures (cf. Table 2). In addition, the role of math tests in accounting for the gender gap in STEM major choice is overestimated when excluding advanced coursework measures (Appendix, Table A.6). Together, gender differences in ‘school work’ (grades, tests, and advanced coursework) explain even more than half (58 per cent) of the gender gap in STEM major choice in Germany. This is a substantial share. On the other hand, at equal school work, women still opt substantially less for STEM majors than men, as evidenced by the remaining unexplained gender gap, so the school work explanation is not sufficient.

6 Conclusions and discussion

In Germany, as in all other countries, women are underrepresented in STEM majors in higher education and less often choose these majors than men. One explanation of this important form of gender segregation is that women have lower performance in math at secondary school than men and higher performance in language. However, this school performance explanation was rejected in studies for the Anglo-Saxon context (Hyde et al. 2008; Mann & DiPrete 2013; Morgan et al. 2013; Riegle-Crumb et al. 2012; Van de Werfhorst et al. 2003). The dismissal was not so much due to absent performance effects – on the contrary, the effect of math performance was among the strongest STEM determinants – but due to low gender differences in school performance in the Anglo-Saxon area. We aimed to advance the literature by testing the school performance explanation for a context, Germany, which has stronger gender differences in math and language performance, notably in grades. Our analyses of a longitudinally followed sample of 9^th graders in upper secondary school in Germany displayed that prior school performance –math grades, grades in the German language, math competencies, and reading competencies –can explain nearly half (45 per cent) of the gender gap in STEM major choice in higher tertiary education.

Our study indicates that in contexts with considerable gender differences in school performance, such as in Germany, school performance can be an important explanation of the gender gap in STEM major choice. This finding nuances earlier conclusions on the school performance explanation. We also find that grade and competence measures for math and language can better account for the gender gap in STEM major choice than grade measures only. An earlier study for Germany could explain one-fifth of the STEM major choice at general universities by math and German grades (Uunk et al. 2019). When using grades and competency measures, we explain nearly half of the gender gap. This is because competence measures, notably in math, are also associated with STEM major choices and show considerable gender differences. In addition, when estimating grade and competency measures, we find that the contribution of competencies to the gender STEM gap is reduced. We regard these findings as another advancement of the literature that focused either on testing effects of competencies or grades (but see Riegle-Crumb et al. 2012; Van der Werfhorst et al. 2003).

However, not all school performance measures could equally well account for the gender gap in STEM major choice. Our analyses showed that only math competencies and German grades mattered. These measures are relatively strongly associated with the likelihood of STEM declaration, compared with other STEM determinants, and display considerable gender differences (men achieving higher math test scores and women higher German grades). As to math competencies, the substantial effect on the STEM major choice surprises since these competencies are not directly known to students in Germany, and the competencies can be assumed to be less important for labor market success than grades. The math competence effect, which was also found in the Anglo-Saxon context (Hyde et al. 2008; Mann & DiPrete 2013; Xie & Shauman 2003), may, for Germany, be interpreted as students’ latent knowledge of their math and STEM abilities. The latent math knowledge and learning experiences may shape students’ self-confidence in math (“self-efficacy”) and subsequently steer them into STEM or non-STEM pathways (Lent et al. 2015). That math grades cannot account for the gender gap in STEM major choice is not so much due to its effect on the STEM major choice – it has a stronger positive effect than math competencies – but due to the low gender differences herein (cf. Riegle-Crumb et al. 2012; Uunk et al. 2019). The weak gender difference in math grades may be attributed to women’s greater self-discipline and efforts to achieve good school grades (Duckworth & Seligman 2006). Reading competencies neither accounted for the gender gap in STEM major choice. This is because reading competencies are not associated with STEM declaration, a finding which was also implied in studies for the Anglo-Saxon context (Mann & DiPrete 2013; Riegle-Crumb et al. 2012). One reason may be that reading competencies vary less among the sampled upper-secondary school students (i. e., most Gymnasium students have good reading competence).

Other noticeable findings from our study that appeared are that (a) advanced coursework in math and the German language can also account for the gender gap in STEM declaration (13 per cent in total), implying that “school work” (grades, competencies, coursework) can explain the gender STEM gap for more than half (58 per cent); (b) even earlier school performance, when students are 14 years of age, can account for a notable part of the gender gap in STEM major choice (36 per cent), a finding which has important policy implication (see below); (c) that in Germany it is math versus language rather than science versus language performance that explains the gender gap in STEM major choice, due to stronger gender differences in math than science competencies

Notwithstanding, our analyses and interpretations are not without limitations and raise questions for future research. The first limitation of our study concerns causality. Although school performance was measured before major choice, and we even find—in additional analyses—an effect of school performance in Grade 9 on STEM declaration in higher education, prior school performance is not likely to be the root cause of STEM major choice and the gender gap herein. Gender differences in math and language performance in Germany may themselves be an outcome of gender-stereotypical institutions and norms (Lippmann & Senik 2018). School performance may hence be seen as a more proximate cause of the gender gap in STEM major choice (cf. Stoet & Geary 2018). Future studies may test this suggestion, for example, by investigating whether in countries with more gender-traditional norms and institutions, gender gaps in STEM choices are greater due to more gendered school performance patterns.

A second limitation is that we cannot rule out that our main finding – school performance explaining the gender gap in STEM major choice strongly – is unique to Germany. Early selection in secondary education and the relatively high level of stratification may strengthen gender differences in subject specialization (Breda & Napp 2019; Lorenz & Schneebaum 2024). Therefore, the role of school performance in explaining the gender gap in STEM major choice may be distinct in other educational systems, as earlier studies for the Anglo-Saxon context seem to indicate (Hyde et al. 2008; Mann & DiPrete 2013; Morgan et al. 2013; Riegle-Crumb et al. 2012; Van der Werfhorst et al. 2003; Xie & Shauman 2003). Here, cross-comparative studies are needed that test the effects of distinct institutional features on the role of school performance in explaining gender gaps in STEM major choice (cf. Jacob et al. 2020).

A third and final limitation of our study is that although prior school performance may explain a large part of the gender gap in STEM major choice, it does not explain it entirely. Women still opt less for STEM majors than men at equal prior school performance, in Germany (this study) and the Anglo-Saxon context (e. g., Riegle-Crumb et al. 2012; Van de Werfhorst et al. 2003). Therefore, a question for further research is how to explain the remaining gender gap in STEM major choice. So far, research has not been able to explain much of the gender gap in STEM major choice by other individual-level factors. In line with our earlier suggestions, gendered socialization and the factors associated with this socialization, such as early subject interests and confidence, may seem a good candidate (e. g., Stearns et al. 2020). Cross-comparative studies further point to economic opportunities as a factor by which men and women are able to display gender in their study and occupational choices (“gendered self-indulgence”; Charles & Bradley 2009). Also, the flexibility of curricula may play a role (Mann & DiPrete 2013).

Our study implies that to decrease gender gaps in STEM major choice, policies should – next to using effective interventions such as intensive counselling (Erdmann et al. 2023; Piepenburg & Fervers 2021) – aim to increase women’s math and men’s language performance. Regarding language, this should be more than just reading competencies, as these competencies are not associated with STEM major choices. Efforts should be taken early in school since gender differences in math and language performance arise at this time in life (Freeman, 2004; Niklas & Schneider 2012) and are consequential for later study decisions. Examples are a more gender-neutral portrayal of math and STEM and language and non-STEM subjects at school (cf. Makarova et al. 2019) and systematic tutoring for low language and math achievers (Younger & Warrington 2005). These measures may alter gender differences in math and language performance, gender differences in major choices, and societal expectations regarding gender and STEM.