Is it a Matter of Skills? High School Choices and the Gender Gap in STEM

1 Introduction

Gender differences in educational choices, particularly in STEM fields, have been extensively studied due to their long-term impact on life opportunities, job prospects, and the gender pay gap. Reducing this gap would not only expand opportunities but also boost economic growth, increasing jobs and GDP (EIGE 2018).

Women remain significantly underrepresented in STEM education and careers, particularly in ICT and engineering. In 2015, only 21 % of ICT and 28 % of engineering graduates in the EU were women (EIGE 2018). Conversely, men are underrepresented in education, health, and social work. This segregation persists in the labor market, where women account for 75 % of education, health, and social work roles but only 15 % of STEM-related occupations (EIGE 2018).

Research has examined factors behind these gaps, including gender stereotypes, competitiveness differences, and peer influences (e.g. Buser, Peter, and Wolter 2017; Mouganie and Wang 2020; Card and Payne 2021; among many others). While there is considerable evidence that these mechanisms are at play, the extent to which they contribute to the gender gap in STEM educational choices has not been assessed.

This paper explores the role of academic performance in shaping gendered choices. Boys’ stronger performance in mathematics (Fryer and Levitt 2010; OECD 2023) – possibly influenced by stereotypes (Boaler 2009; Carlana 2019) – aligns with their preference for STEM fields, particularly at higher achievement levels. In contrast, girls’ comparative advantage in literacy may steer them toward non-STEM paths. These patterns suggest academic performance is a key driver of educational segregation.

Most of the existing literature does not seem to support this hypothesis. However, research on gendered educational choices has largely focused on university fields of study, where decisions are mostly influenced by career aspirations and considerations like work-family balance. However, at earlier stages of education, ability and school performance may be more important in shaping choices because they are key to determining the likelihood of success and, if performance and enjoyment of subjects are correlated, also to ensuring better time at school. Moreover, the gender gap in education likely originates in childhood (Fryer and Levitt 2010), suggesting that policies targeting younger students, when preferences and skills are still malleable, could be more effective.

This paper addresses the gap by examining the educational choices of Italian boys and girls during the transition from lower to upper secondary school at age 14. At this stage, students choose among various high school types differing in difficulty, prestige, and subject focus. Italy’s system is particularly relevant because students are tracked early and can freely choose their preferred school type without restrictions based on merit or binding teacher recommendations. This allows preferences, shaped by multiple factors, to be fully expressed without formal barriers.

Another contribution of this paper is that studying the gender gap at the transition from lower to upper secondary education is crucial for children who do not go on to university and can help to understand the horizontal gender segregation among those employed in low-medium level occupations. This mainly involves low socio-economic backgrounds, disproportionately represented in non-academic high school types, as there are significant differences in occupational outcomes between students who attend schools with STEM and non-STEM focus in terms of career opportunities and wages (Alma Diploma 2021).

We use a detailed longitudinal dataset linking two administrative sources (Anagrafe Nazionale Studenti-MIUR and INVALSI) to analyze the school choices of Italian children in grade 8 during the 2015–16 school year. This allows us to examine initial high school enrolment rather than later attendance or completion, which may be influenced by endogenous school failures. The dataset includes academic performance metrics (teacher grades and standardized test scores), family background, and socio-demographic characteristics of schools.

While gender differences primarily manifest as horizontal differentiation (field of study), we contend that analyzing educational choices without considering vertical differentiation is inadequate, given the highly tracked nature of the Italian upper secondary school system and the stark social disparities in these choices (Checchi and Flabbi 2013). Since gender and social strata may significantly interact, our empirical analysis employs a cross-classification of high school types encompassing both horizontal and vertical dimensions. The assumption is that when selecting a high school, girls and boys simultaneously decide between humanities or STEM (or other fields) and the degree of academic rigor. By applying suitable aggregations, we analyze choices across seven school types (detailed in Section 4).

Our study examines various channels influencing gendered educational choices, including individual performance in mathematics and language, class ranks, and exposure to gender role models. Class rank is particularly relevant as self-perception of abilities often hinges on comparisons with peers (Delaney and Devereux 2021b). Additionally, exposure to high-achieving peers of the opposite gender may undermine confidence in specific subjects (Mouganie and Wang 2020).

Descriptive analyses reveal a significant gender gap in high school choices, with girls less likely to select STEM tracks, particularly for school types in the middle of the vertical hierarchy. Multinomial logit models show that achievement variables reduce this gap only minimally. Both genders respond to absolute and relative performance in math and language, but girls require stronger signals of mathematical ability to choose STEM fields.

Further analysis reveals that parental education shapes the gender gap differently. For children of tertiary-educated parents, school performance explains a significant portion of the gap, while it has limited explanatory power for children of less-educated parents. This suggests that socio-economic factors amplify the barriers faced by girls from disadvantaged backgrounds.

Finally, we simulate the effect of equalizing math test scores between genders. This would reduce the gender gap in STEM Lyceum choices by about 20 %, though it has little effect on STEM technical school choices. These findings suggest that prior performance contributes to some extent to the perpetuation of gendered educational pathways, particularly in schools with academic curricula.

2 Related Literature

In this section, we examine the existing literature, which considers various hypotheses to explain the gender gap in educational choices.

3 The Italian Education System

In Italy, formal education is divided into three stages: five years of primary school (ages 6–11), three years of middle school (11–14), and five years of high school (14–19). Education is compulsory until age 16. The school system is predominantly public, with public schools generally being of higher quality, particularly at the upper secondary level. Private schools, on the other hand, are often attended by low-achieving pupils from middle and upper social classes who struggle in public schools (MIUR 2022).

Up to middle school, the system is comprehensive and standardized, meaning that syllabuses and learning targets are defined at the national level and there are generally no elective subjects available. At the end of middle school, students take a national exam before progressing to high school, where tracking is a key feature. High schools vary significantly along two key dimensions. The first dimension, horizontal differentiation, relates to the curriculum focus (e.g. STEM, humanities, business-related studies) and is especially relevant for understanding gender disparities. The second dimension, vertical differentiation, pertains to academic content and prestige, playing a crucial role in social stratification.

The most prestigious schools are Traditional Lyceums (Classical and Scientific), offering rigorous academic preparation as a pathway to university. Non-Traditional Lyceums (e.g. Linguistic, Artistic, and Human Sciences) are less academically demanding and prepare students for middle-level white-collar jobs. Technical and Vocational schools provide more applied education, with technical tracks being more demanding than vocational ones. These tracks are often chosen by students from middle-low social backgrounds who are uncertain about pursuing higher education. Conversely, students from advantaged backgrounds are overrepresented in lyceums, particularly in Traditional Lyceums, reflecting strong social stratification in high school choices (Jackson 2013).

High schools are usually housed in separate buildings, meaning students from different tracks rarely interact during school hours. This segregation highlights the importance of high school choice in shaping peer groups and, consequently, future educational and professional outcomes.

Students select their high school type in grade 8, typically at age 13. Teachers provide non-binding recommendations, and pupils are free to apply to any school. High schools cannot select students based on prior academic performance; in cases of excess demand, priority is given to those in the local catchment area. School guidance is generally informal, with local authorities offering basic information through websites and fairs. Personalized counseling services are almost non-existent (Argentin, Barbieri, and Barone 2017), and families often base decisions on interests, aptitudes, academic performance, and career goals.

Despite its openness at enrolment, the Italian system becomes selective post-entry. Grade retention is common, particularly at the end of grade 9, often leading students to transfer to less demanding schools (Contini and Salza 2024).

University admission in Italy is unrestricted by merit, allowing students from any high school type to enroll. However, university transition rates vary significantly: 76 % of lyceum graduates, 46 % of technical school graduates, and 25 % of vocational school graduates proceed to higher education. Degree completion rates are higher for Lyceum graduates, who have better preparation and stronger academic skills (ANVUR 2023). Despite this, the proportion of tertiary-educated individuals in Italy remains low (28 % of the 25–34 age group, OECD 2021).

Labor market outcomes show significant disparities between students with and without a STEM background, especially when they do not continue to university (Alma Diploma 2021). For instance, over 25 % of students from STEM-focused technical schools have stable contracts one year after graduation, compared to around 10 % from business-oriented technical schools or humanities-focused non-traditional lyceums. Starting salaries for STEM-focused technical graduates are approximately 15 % higher (Alma Diploma 2021).

University paths also reflect prior educational choices. Most graduates from technological schools or scientific lyceums pursue STEM degrees, with significant gender disparities favoring boys. Over 70 % of technological school graduates who attend university choose STEM fields, compared to less than 10 % from business technical schools or non-traditional lyceums (Alma Diploma 2021). Consequently, both direct entry into the labor market and university choices reveal that a STEM background increases the likelihood of STEM-related careers.

The trade-offs in high school choice can be summarized as follows:

Lyceums versus non-academic school types: Traditional Lyceums offer high academic prestige, better university preparation, and favorable peer groups. However, they demand greater effort, have higher grade retention risks, and offer limited job prospects for those not pursuing university. These disadvantages are particularly pronounced for low-SES students, who have fewer resources to compensate for academic struggles, are more risk-averse, and tend to have lower aspirations (Breen and Golthorpe 1997).

STEM versus non-STEM focused schools: STEM schools provide better labor market opportunities and higher wages. They also align well with university STEM programs. However, they require strong mathematical skills and interest in STEM subjects, which may deter students with weaker math abilities, including many girls.

In conclusion, the Italian education system is characterized by early tracking, strong social stratification, and significant disparities in outcomes based on high school type and curriculum focus. These factors shape students’ educational trajectories and career prospects, reinforcing existing inequalities.

4 Data and Descriptive Statistics

This study utilizes a longitudinal dataset linking two administrative archives: the Italian National Register of Students (Anagrafe Nazionale Studenti), which tracks students’ educational paths, and INVALSI, which provides standardized test scores from national assessments. This linkage offers a comprehensive view of students’ achievements before high school, including test scores, grades, and socioeconomic background.

The dataset includes detailed information on students’ academic careers, such as dropouts, grade repetitions, school and class attended, and high school choice, the study’s primary focus. Class and school identifiers allow us to incorporate peer and school-level characteristics, such as classmates’ achievements and backgrounds.

The analysis focuses on students enrolled in grade 6 in 2013/14 across three Italian regions: Piedmont, Lombardy, and Veneto. By 2016/17, these students begin high school (grade 9), where tracking starts. The dataset covers 173,684 students, with 97 % traceable to their high school enrollment (168,445 students across 1837 middle schools).

High school choices are categorized into seven groups: Traditional STEM Lyceum (Scientific Lyceums), Traditional Humanities Lyceum (Classical lyceum), Non-Traditional Humanities Lyceums, Technical STEM schools, Technical Business schools, Vocational STEM schools, and Vocational Business schools. This categorization reflects differences in curricular focus and provides a framework for analyzing the determinants of high school choice in the Italian context (Table 1).

Table 2 presents descriptive statistics on educational choices, showing that boys are over-represented in STEM-focused schools. Over a third of boys attend Technical STEM schools, compared to just 7 % of girls. The gap is also significant in Traditional STEM Lyceums (27 % boys, 19 % girls) and Vocational-STEM schools (5 % boys, 2 % girls). In contrast, girls are more likely to attend schools focusing on humanities, social sciences, and business. The largest gaps, in both absolute numbers and ratios, are seen in Non-Traditional Lyceums with a humanity focus and Technical-STEM schools.

Table 3 presents gender gaps in academic performance in Italian and mathematics, based on test scores and teachers’ grades. Grades refer to teachers’ evaluations at the end of grade 7, before students choose high school in grade 8, while test scores represent performance in the INVALSI tests at the end of grade 8. Although the test results are released after the high school choice, they provide valuable insight into actual skills, distinct from the effort-based teacher grades.

Girls slightly outperform boys in Italian and mathematics according to teachers’ grades, but in standardized math tests, boys significantly outperform girls. This gap in test scores may reinforce the perception that boys are better at math (“he’s good at math” versus “she works hard”). As noted in international research (OECD 2023), the gap is especially large at the middle and top of the distribution, which is crucial for subsequent school choices. High achievers in math are more likely to pursue STEM fields, a trend that influences high school enrollment decisions.

We take advantage of the richness of this dataset and add a detailed list of control variables in the main model, including individual-level factors like socio-economic background (measured by parental education, migration status, and ESCS) and school-level characteristics (average student performance in Italian and mathematics, socio-economic background, share of females, and immigrant students). A full description of these explanatory variables and their statistics is available upon request.

5 Setup and Empirical Strategy

Our goal is to determine whether the gender gap in high school choices can be attributed to ability levels at the time of decision-making. To this end, we use a multinomial logit model to analyze high school choices. We first examine the entire sample to assess the extent to which the gender gap is influenced by mechanisms related to individual and peer performance. Then, we conduct separate analyses for girls and boys to evaluate whether they respond differently to ability measures.

We model the probability of opting for any of the M = 7 high school options described in Section 4 using a multinomial logit model, for which the simplest version is:

where the vector of explanatory variables X includes the different measures of performance and several control variables, describing family background (parental education; migrant status and an index of socio-economic disadvantage) and middle school-level characteristics (percentage of students whose parents have a university degree, percentage of students with migrant background in the school, percentage of girls in the school, average and standard deviation of test scores in Italian and Mathematics).

We estimate several versions of this model to explore how achievement influences educational choices. First, we examine the impact of students’ grades and test scores in Italian and mathematics. To determine whether the key factor is performance in each subject or a comparative advantage (e.g. being better at math or Italian), we replace the two grade variables with one measuring the difference between these grades. Next, we incorporate individual class rankings in both subjects to assess the role of relative versus absolute performance. Finally, we include the class proportion of high achievers in mathematics, interacted with gender, to test for the presence of gender role models.

6 The Role of Performance: Full Sample Analysis

In this section, we present results from the estimation of multinomial logit models on the complete dataset, to assess how high school choices are influenced by various measures of school performance. First, we focus on the effect of gender to determine whether the inclusion of performance variables and other mechanisms fully or partially explains the gender gap in choices (Section 6.1). Second, we analyze the samples of girls and boys separately to uncover potential differences in how performance affects their choices (Section 6.2). In the following section (Section 7), we replicate some analyses by parental education to examine whether the role of performance varies across family backgrounds.

6.1 Is the Gender Gap in High School Choices Explained by Individual School Performance?

To answer this question, we estimate a set of multinomial logit models for educational choices on the complete sample of students and observe if and how the gender gap in educational choices is explained as we progressively add different control variables capturing different mechanisms. Main results are shown in Figure 1 (details are available in Appendix A).

Figure 1:

Marginal effects of gender (female vs males) on school choice, different models. Note: Model 1 includes only the binary gender variable. Model 2 also includes individual grades and test scores in Italian and mathematics. Models 3 to 6 include the following individual control variables: a measure of students’ socio-economic status (ESCS index, calculated considering parental occupation; parental education; and availability of a series of items that can favor quiet study time, such as a computer, a quiet place to study, an internet connection etc.); parental education (ref. no high school diploma; high school diploma; higher education); migrant status (ref. citizen; first- or second-generation immigrant); they also include middle school-level variables: proportion of students with parents with a university degree; proportion of migrants; proportion of girls; average and standard deviation of test scores in Italian and Math. Model 5 does not include grades in year 7 but includes the difference between Italian and Mathematics grades, in addition to the above mentioned variables. Model 6 includes two variables for the percentage of girls and boys who are high achievers in the class in Mathematics. Models are estimated using the mlogit routine in Stata.

At first, we visualize the raw gap by estimating a model that only includes the gender binary variable (model 1, M1). In model 2 (M2), we add test scores and teachers’ grades, and in model 3 (M3) we add individual family background (parental education, ESCS and migrant origin) and school characteristics (share of girls, share of students with migrant background, share of students with parents with university degree, average and standard deviation of test scores in language and math) as control variables. In model 4 (M4), we add individual class rankings in all performance measures (a variable ranging between 0 and 1, depending on the student’s ranking in the class in the specific subject). In model 5 (M5), we analyze the role of comparative advantage in one of the two subjects by including, instead of grades in Italian and mathematics, the difference between the two. Finally, in model 6 (M6) we add to model 4 two variables representing the share of girls and boys with top grades in math in the class. Average marginal effects of the female indicator can be visualized in Figure 1.

As already mentioned, girls are significantly less likely to select schools with an emphasis on STEM subjects, including the more academically demanding ones (Traditional STEM Lyceum) and, to an even higher degree, the Technical-STEM paths. On the other hand, they are disproportionately more likely to select Humanities-focused schools (in particular, non-Traditional Lyceums and, to a lesser degree, business related Technical schools). The interesting finding is that these gaps remain quite stable when all the different mechanisms involving performance – teacher grades and standardized test scores in Italian and math, individual rankings in the classroom and the share of high achieving girls and boys – are taken into account.

Altogether, we conclude that past achievements only marginally explain the gender gap in high school educational choices.

6.2 Do Performances in Italian and Math Matter for High School Choices? Do They Matter Differently for Girls and Boys?

Prior achievement does not seem to explain the gender gap in high school educational choices. However, as we now show, performance in Italian and math are very important predictors of high school choices. In this section we focus on how the different dimensions of school performance matter for the educational choices of boys and girls.

Results relative to the estimation of model (1), version M3, are summarized in Figure 2. The effects of teachers’ grades and standardized test scores on educational choices are almost always in the same direction, although their magnitudes are quite different. In some instances, grades matter more than test scores, because grades are always disclosed to pupils and parents, and are the most visible measure of performance, but this is not always the case.

Figure 2:

Effect of school grades and test scores in Italian and math on school choices (AME) by gender. Note: From estimation of model (1), version M3.

For both girls and boys, the probability of choosing a Lyceum with a focus on STEM increases as mathematics performance increases, and the effect is even larger for girls than for boys. The same direction, but much smaller effects, apply to the choice of technical STEM. Instead, the probability of choosing schools with other focuses decreases with mathematics proficiency. On the other hand, the probability of choosing schools with a focus on humanities increases with performance in Italian for both sexes. However, while boys are also more likely to choose a Traditional STEM Lyceum as their Italian test scores increase, this is not the case for girls (for this reason, if girls – who outperform boys in Italian – were to make their choices as boys do, they would choose the Traditional STEM Lyceum even more than boys, as Figure 2 suggests). In other words, boys are more likely to choose a STEM Lyceum if they perform well in both Italian and mathematics, and girls are more likely to make this choice if they perform well in mathematics but not so well in Italian. Due to its lower prestige, the probability of choosing the vocational track decreases with performance in both subjects, regardless of the school’s focus.

Next, we examine the role of other mechanisms related to performances.

First, we analyze the role played by comparative advantage in math or Italian. To do so, instead of both grades, we include in the model a variable measuring the difference between Italian and Math grades. Results are presented in Table A5 in the Appendix, and show that increasing the comparative advantage in Italian, students are less likely to choose STEM-oriented schools. These effects seem to be larger for girls.

Second, students may also consider their own ability in relative terms. To understand whether relative performance plays a role in educational choices, in addition to absolute performance, in the next model we include four variables measuring students’ class rank in terms of teachers’ grades and test scores in both Italian and mathematics. We find that class rankings play an important role in shaping educational choices – in the same direction of absolute measures of performance. However, these effects are very similar for boys and girls (Table A6 in the Appendix). Ranking in teachers’ grades seem more relevant than ranking in test scores, especially in Italian (results for ranking in test scores are hardly significant), and this is true for both boys and girls.

Finally, we test the importance of exposure to “gender role models”, conceived as the presence on the class of high achieving students of the same or opposite gender. We find no evidence of empirical relevance of either variable, for both boys and girls (Table A7 in the Appendix).

In Appendix B we report some additional results on the role of the performance of classmates for educational choices by gender and the role of individual’s family background and school context. We also show the results of a decomposition of the gender gap in choices using an Oaxaca-Blinder-like procedure for categorical dependent variables (Fairlie 2005). Consistently with the results displayed in Figure 2, the explained component plays only a very minor role. Finally, we include the results of a robustness check where we use grade 5 instead of grade 7 and 8 test scores and grades, as a possible way to address the issue of anticipatory effects.

7 The Role of Performance: Analyses by Parental Education

We now turn to another question: do gendered choice patterns vary by socio-economic background? First, the value or utility that girls and boys attach to different options within the high school system may vary depending on their family background. Furthermore, as vertical differentiation in terms of academic content and prestige intersects with horizontal differentiation shaped by the emphasis on various subjects, the primary choice set may also differ based on parental education.

Specifically, we categorize students into three groups: those with at least one parent holding a higher education qualification (high parental education), those with at least one parent holding a high school diploma (medium parental education), and those whose parents do not possess a high school diploma (low parental education). Descriptive statistics are presented in Table C1 in the Appendix.

Building on the analyses conducted on the full sample (see Section 6.1), we investigate, for each level of parental education, whether the gender gap in school choices can be explained by a set of performance-related variables that capture the various mechanisms through which academic performance may influence educational decisions. This is done by comparing the average marginal effects of gender on the probability of choosing each school option, as estimated by six different versions of model (1). The analysis reveals distinct patterns by parental education (Figure 3). While the finding that performance plays only a very minor role in explaining the gender gap in high school choices holds for children of medium- and low-educated parents, the influence of performance seems more pronounced for children of highly educated parents.

Figure 3:

Marginal effects of gender (F–M) on school choices, by parental education, different models. Note: Parental levels of education are defined in Section 7 (p. 18). Models 3 to 6 include the following individual control variables: students’ socio-economic status (ESCS index); parental education migrant status; middle school-level variables: proportion of students with parents with a university degree; proportion of migrants; proportion of girls; average and standard deviation of test scores in Italian and Math. Model 5 does not include grades in year 7 but includes the difference between Italian and Mathematics grades, in addition to the above mentioned variables. Model 6 includes two variables for the percentage of girls and boys who are high achievers in the class in Mathematics. Models are estimated using the mlogit routine in Stata.

For this more advantaged group, over half of the gender gap in enrolment in the Traditional Humanities Lyceum and 16–30 % of the gap in the Scientific Lyceum can be attributed to differences in past academic performance (for example, for the latter the raw gap favoring boys stands at 19 percentage points but when grades and test scores are accounted for, this gap narrows to 13–16 percentage points). In contrast, for children of medium- and low-educated parents, the gender gaps remain largely unchanged. A similar pattern emerges for the Technical STEM track: when accounting for grades and test scores the gender gap decreases by 20 % among children of highly educated parents, compared to only a 10 % reduction for children of less-educated parents.

Predicted probabilities as a function of grades in mathematics and Italian, derived from our preferred model (M3 in Section 6.1) separately by gender and parental education, are shown in Figures C1 and C2 in the Appendix. The key results can be summarized as follows.

Among boys from low-educated backgrounds, Technical-STEM schools are consistently the most likely choice, regardless of academic performance. The Traditional STEM lyceum becomes a competitive option only for those who perform well in both Italian and mathematics. For boys from highly educated families, Traditional STEM lyceums consistently dominate. Technical-STEM schools are a common alternative only for those with low Italian performance, while non-Traditional humanities attract those with poor grades in math.

Girls’ choices are more diverse. Those from low-educated families often prefer Technical-business or non-Traditional humanities lyceums. Vocational-business schools are the predominant choice for girls with low performances, whereas Scientific lyceums are chosen only by those with high math scores. Overall, STEM school types are rarely selected, even by top-performing math students. Instead, girls from highly educated families tend to favor humanities-focused lyceums when they have low-medium math results, while Scientific lyceums are preferred by those excelling in math.

We now explore STEM-oriented school choices in greater detail. Figure 4 illustrates how the predicted probability of enrolling in the Scientific Lyceum and Technical STEM schools changes as math grades increase. For all student groups, the probability of enrolment in the Scientific Lyceum steadily increases with math grades, with a steeper slope for students with highly educated parents, particularly for girls. The relationship is more nuanced for enrolment in the Technical STEM track. As noted earlier, there are huge gender differences, especially among students from families with lower levels of parental education, where this option is notably more common. Furthermore, the enrolment probability is much less sensitive to math grades: as performance improves, it increases slightly for girls from less educated families, rises and then falls gently for boys from less educated families, decreases for boys and is constant for girls from highly educated families.

Figure 4:

Predicted probabilities of enrolling in scientific lyceum and technical stem. Note: From estimation of model (1), version M3. All other variables at their means.

Figure 5 illustrates the gender gap in STEM school choices, measured as the ratio of the predicted probabilities of boys to girls, across different levels of performance in math and Italian, and disaggregated by parental education.

Figure 5:

Gender ratio of predicted probabilities, for scientific lyceum and technical stem. Note: From estimation of model (1), version M3. All other variables at their means.

For Scientific Lyceums, the gender gap favoring boys is weak or disappears entirely at the top of the math grade distribution. In contrast, at the lower end of the math performance scale, the divide is much more pronounced, with significantly fewer girls opting for this school type compared to boys.^[5] Additionally, the gap is more substantial among students with high grades in Italian, likely because students in this group (and girls in particular) are more easily diverted toward school types that emphasize the study of humanities. Somewhat surprisingly, the gender ratio is higher among students with highly educated parents.

The gender gap in choosing Technical STEM schools is significantly larger and is most pronounced among children of less educated parents. As with the Scientific Lyceum, the gap is wider among students who perform poorly in math but do well in Italian.

Taken together, these findings confirm that girls require much stronger signals of math ability to pursue academic-oriented schools with a focus on STEM, while they challenge the idea that gender norms exert the strongest influence among socially disadvantaged groups.

8 What if the Gender Gap in Math Performance Closed?

The previous findings show that – except for the students from high educated backgrounds – the gender gap in high school choices is only marginally explained by absolute and relative achievements. At first glance, this result seems to imply that if girls were to close the gender gap in math not much would change in terms of the gender gap in choices. However, this is not necessarily the case. This question is relevant because it has often been argued that reducing the gender gap in math could have a substantial impact on STEM choices (see for example OECD, 2019).

Our analysis considers both teacher grades and test scores in Italian and mathematics. Although girls perform slightly better in teacher grades, they underperform in math test scores compared to boys. Since math achievements strongly influence STEM choices, girls’ choices could change if their math test scores improved, while other abilities remained constant.

To estimate this impact, we simulate an increase in girls’ math test scores to match the average score difference between boys and girls with the same grade in math. The probability of making each high school choice are calculated for different levels of parental education.

Results, shown in detail in Appendix D, reveal that improving girls’ math test scores would increase their likelihood of choosing the Traditional STEM Lyceum, reducing the current gender gap from 10.1 percentage points (pp) to 7.8 pp (a 22 % decrease). However, the effect on Technical STEM schools is minimal, with a 2 % reduction in the gender gap.

The analysis by parental education shows that the potential reduction in the gender gap in STEM Lyceum would be driven mainly by girls of highly educated parents for whom this choice is more common. The gap would fall from 18.6 to 15.2 pp for children with highly educated parents, from 8.6 to 6.3 pp for the medium-educated group, and from 3.2 to 1.8 pp for the low-educated group. On the other hand, when it comes to the choice of Technical STEM schools, the reduction in the gender gap is always almost negligible, suggesting that this choice is strongly related to other factors, presumably related to gender norms in education and in the labor market.

9 Discussion and Conclusions

Women remain significantly underrepresented in STEM fields in higher education, particularly in high-demand areas like ICT and engineering. This disparity often stems from decisions made earlier, during upper secondary school. Moreover, many high school graduates, especially those from lower socio-economic backgrounds, forgo tertiary education and enter the workforce directly.^[6] For this group, gendered patterns in high school subject choices reinforce horizontal labor market segregation, with women disproportionately concentrated in lower-paid, less promising careers.^[7]

This paper adds to the existing literature on gender gaps in education, which is largely focused on the choice of the field of study at university, by shedding light on the gendered pattern of choices at a younger age. We examine Italy, where children are tracked into very different school types at age 14, following a decision made at 13 during their final year of middle school. The Italian case is particularly interesting because it is an open choice system, both at the secondary school and university level, and therefore preferences are not constrained by ability restrictions imposed on the decision-making process.

In tracked education systems, students are streamed into distinct school types with centrally defined curricula and objectives, unlike systems where students select electives or the level of core subjects, such as language and mathematics, at the end of their school cycle. These early decisions are pivotal for future opportunities. Understanding the factors influencing gendered choices at this stage – and how they differ from those shaping later decisions – could inform effective policy, as earlier interventions often yield better outcomes. We hypothesize that choices made in early adolescence are more influenced by academic performance and less by labor market considerations compared to decisions made later in life.

Our findings reveal a significant gender gap in STEM-focused Lyceums, with even larger disparities in other school types with lower academic focus and prestige. For example, technical STEM schools are predominantly attended by boys, while non-traditional humanities-focused Lyceums attract mostly girls. Hence, these gaps are most pronounced in school types common among low-SES students.

Teacher grades and test scores strongly influence choices, but the impact of ability differs by gender. Girls are less likely to opt for STEM high schools unless they excel in mathematics or have a strong comparative advantage in the subject. Conversely, boys often choose STEM pathways regardless of weak math performance and continue to do so even when they excel in Italian. Consistent with prior research, our findings suggest that girls require stronger evidence of their mathematical ability than boys to pursue STEM studies.

Our results also indicate that mathematical and Italian skills partly account for the gender gap among high-SES children but not among low-SES children. A potential explanation of this finding is that parents from advantaged backgrounds – whose children are more likely to attend traditional Lyceums – tend to hold more gender-egalitarian views (Dryler 1998). As a result, they are more inclined to base educational decisions on objective factors, such as their child’s abilities in various subjects. However, this view is challenged by the finding that gender differences in the choice of the Scientific Lyceum are wider among the highly educated parents’ group.

An alternative explanation lies in the institutional structure of the education system. In academically oriented Lyceums, STEM subjects primarily include mathematics, physics, and natural sciences, while technical or vocational STEM schools – less academic and prestigious – emphasize technical subjects. As argued by Barone (2011), there is a notable technology/care divide in STEM fields of study at university: the largest gender gaps are found in engineering and ICT, because other STEM fields (such as math, science or health) retain a symbolic affinity with care work, including teaching. For girls from disadvantaged backgrounds, the decision may be framed less as a choice between humanities and sciences and more as one between humanities and technology, offering little incentive to pursue STEM-focused schools.

A third explanation concerns the shorter perceived path to the labor market for low-SES students, as many do not pursue university. This makes them and their parents prioritize high school choices based on immediate job prospects, even at a young age.

It is often suggested that closing the math skills gap could help reduce the gender disparity in STEM educational choices and careers. To explore this, we simulated a scenario where we raised girls’ math test scores to match those of boys and predicted their high school choices. The results showed that the gender gap in choosing the STEM Lyceum would decrease by about 20 %, while the gap in choosing the STEM technical school would remain largely unchanged. However, our findings do not imply that improving girls’ math skills is unimportant; closing the gender gap in math could boost girls’ confidence, interest in math-related fields, and participation in STEM studies at all levels.

It is beyond the scope of this paper to examine the role of other channels discussed in literature, such as stereotypes and gender norms. While these mechanisms have been shown to matter, there is limited evidence on the extent to which they account for the gender gap. In this context, our finding that differences in individual performance – even at an early age – can only marginally explain the gender gap in STEM choices suggests that other explanations, particularly those linked to behavioral differences, are overall more significant.

In summary, this paper finds that, similar to research on later educational choices, the gender gap in early educational choices is not largely driven by gender differences in ability. Thus, our hypothesis that school ability and performance would be more decisive at the transition between lower and upper secondary education is not confirmed. Our findings add to the existing literature by providing further evidence of the need to invest in policies aimed at deconstructing gender stereotypes and prejudices from an early age, when abilities and interests are still relatively malleable. Effective interventions could target both parents’ and teachers’ expectations or biases (Alesina et al. 2024; Carlana 2019; Carlana and Corno 2024), as well as directly influence children’s attitudes through female role models promoting the link between women and STEM (Breda et al. 2023), or male role models strengthening the connection between men and the humanities or social work.

Another channel, less explored in the literature, is that both girls and boys may be discouraged from enrolling in schools where the opposite sex predominates (as is the case in Italian high schools, with girls highly underrepresented in Technical STEM schools and boys in Non-Traditional Humanities lyceums). This could significantly influence educational choices and limit the role of individual achievements. However, the available data make it difficult to disentangle this mechanism from the effects of preferences and stereotypes. This presents an interesting area for future research.