VoxEU Column Education Gender

It’s not just for boys! Understanding gender differences in STEM

Women are much less likely to study STEM degrees at university. This column reveals that in the case of Ireland, the gender gap is concentrated in the areas of engineering, technology, and mathematics. Subject choice in secondary school is the most important predictor of the portion of the gap that can be explained, with a small role for grades achieved in mathematics versus English. A gender gap of 9% remains even among students who studied the same subjects and achieved the same grades at secondary school.

While the education levels of women have increased dramatically relative to men in recent decades (Goldin et al. 2006), women are still greatly underrepresented in science, technology, engineering, and mathematics (STEM) college programmes and occupations. Card and Payne (2017) show that in the US and Canada, the gender gap in the likelihood of graduating with a STEM-related degree explains about 20% of the wage gap between younger college-educated men and women, suggesting that the gender gap in STEM is important for understanding the gender gap in earnings. This issue is also important in terms of aggregate productivity – much evidence suggests that qualified STEM workers play an important role in increasing productivity and driving economic growth (Peri et al. 2015).

Unfortunately, it is difficult to understand what determines college major choices. In our recent study (Delaney and Devereux 2019), we examine this question by using unique data from Ireland. Ireland’s centralised third-level admission system provides an ideal laboratory because students provide a preference ranking of college programmes, allowing us to observe the course preferences of all college applicants. Additionally, since college admission is almost completely determined by performance in a set of national examinations (the Leaving Certificate examinations), comparable information on prior preparation and relative performance across subjects is available for all applicants. We use this to examine whether the gender gap in STEM exists for boys and girls who have identical preparation at the end of secondary schooling (in terms of both subjects studied and grades achieved), or whether it is mostly due to differences in STEM-readiness that already exist at the end of secondary schooling.

Descriptive differences by gender

Overall, there exists a large gender gap in the fraction of applicants listing a STEM course as their first preference, with just over 40% of men listing a STEM course compared with roughly 19% of women. Figure 1 shows the proportion of each gender choosing STEM courses as first choice (broken down by science, technology, engineering, and mathematics). Clearly, the large STEM gap is driven by choices of engineering and technology courses, with women being slightly more likely to list science courses as first preference than men. Therefore, while the gender gap in STEM is large, the current focus on closing the gap in STEM might instead target the gaps in technology and engineering. We also find that the gender differences in acceptances of STEM courses are almost identical to gender differences in first preferences.

Figure 1 The proportion ranking a STEM college course as first preference

Figure 2 illustrates gender differences in subject choice in secondary school. Within the STEM category, boys are three times more likely to study physics and applied mathematics while girls are more likely to study chemistry and biology. There are also large gender differences in the take up of practical subjects, with less than 5% of girls taking subjects such as engineering, building construction, design graphics, and technology. Interestingly, we find similar gender differences in mixed-gender schools, where subject availability is the same for girls and boys.

Figure 2 The proportion choosing a STEM subject in secondary school

There are also gender differences in performance across subjects in the Leaving Certificate. On average, girls do better, scoring 17 more points in the Leaving Certificate (roughly equivalent to scoring 1.5 grades higher in one subject). However, boys score higher on average in mathematics (a compulsory subject) in the Leaving Certificate and do better in some more technical option subjects such as applied mathematics, engineering, and building construction. These differences suggest that male comparative advantage in STEM may account in part for their greater probability of choosing STEM courses in college.

Explaining the gender gap in STEM

We consider three broad explanations for the lower probability that girls enter STEM programmes. The first is differential general achievement – girls tend to obtain higher Leaving Certificate points than boys and this may provide them with a broader menu of choices and lead to a lower (or higher) probability of choosing STEM. We find that controlling for overall Leaving Certificate points does little to the gender gap, suggesting that it is not much influenced by differences in overall Leaving Certificate achievement. 

The second potential explanation is comparative advantage. Students may choose college courses that best utilise their talents. The previous literature has identified mathematical and verbal skills as being key predictors of STEM major choice (Anelli et al. 2017, Turner and Bowen 1999) and we measure these using grades in English and mathematics. While English and mathematics grades will not capture all aspects of comparative advantage, it is reassuring that they are strong predictors of STEM. Figure 3 shows the effect of English and mathematics grade indicators on the probability of doing STEM. The effects are largely monotonic with STEM probabilities increasing with mathematics grades and decreasing with English grades (the best grade is H1 and the worst grade is O8 – the omitted category is a grade of O4). We find that accounting for grades in English and mathematics reduces the STEM gender gap by 2.1%, suggesting a small role for comparative advantage. Furthermore, we find a total effect of about 1% from taking account of grades in other subjects. Overall, we conclude that about 3% of the STEM gender gap is due to gender differences in comparative advantage at the end of secondary schooling (as revealed by subject grades).

Figure 3 The effect of maths and English grades on ranking STEM first    

A third explanation for the gender gap is subject choices in secondary school. Leaving Certificate subject choices may in themselves have a subsequent causal effect on STEM entry (either through enabling students to meet programme requirements or by providing them with more information, expertise, or confidence in their STEM abilities), or they may simply reflect underlying preferences towards STEM. We find that after controlling for subjects the gender gap falls by over 9%, which is a substantial decrease. Choosing chemistry, physics, physics with chemistry, engineering, technology, design graphics, or applied mathematics as Leaving Certificate subjects is very strongly positively related to ranking a STEM degree as first preference (see Figure 4). However, there is a weaker relationship between doing biology or agricultural science and subsequently choosing STEM.

Figure 4 The effect of Leaving Certificate subjects on ranking STEM first


We find that there is a substantial gender gap in listing a STEM course as first preference (22%) that is concentrated in the areas of engineering, technology, and mathematics – boys and girls are about equally likely to list science. We find a negligible role for overall achievement in explaining the STEM gender gap, and a larger role for comparative advantage (as measured by differential achievement across subjects, particularly English and mathematics). These grade differences across gender need not be innate and may represent different interests and investments across subject areas throughout schooling. We find that subject choice in secondary school is the most important predictor of the portion of the gender gap that we can explain. While this may partly reflect the differing subjects that are available in girls’ versus boys’ schools, our finding of similar subject choice differences in mixed-gender schools suggests that the availability of subjects is not an important consideration. Boys are much more likely to do physics, design graphics, engineering, building construction, and applied mathematics in school – subjects that are strongly predictive of later doing STEM in college. So, even two years before college entry, there are systematic gender differences in decision-making that lead to boys being more likely to choose STEM subjects. 

Strikingly, we find that there remains a STEM gender gap of 9% even for people who have identical preparation at the end of secondary schooling (in terms of both subjects studied and grades achieved). Clearly, there are systematic gender differences in the tendency to list STEM courses even amongst observationally academically equivalent boys and girls. These differences could be influenced by biological or cultural factors, socialisation, role model effects, peer effects, expectations of future discrimination, job preferences, and many other factors. Understanding these better will be an important objective of future research.


Anelli, M, K Shih, and K Williams (2017), “Foreign Peer Effects and STEM Major Choice”, IZA Discussion Paper no. 10743.

Card, D and A A Payne (2017), “High school choices and the gender gap in STEM”, NBER working paper no. 23769.

Delaney, J, and P J Devereux (2019), “It's not just for boys! Understanding gender differences in STEM", CEPR Discussion Paper no. 13558.

Goldin, C, F K Lawrence, I Kuziemko (2006), “The homecoming of American college women: The reversal of the college gender gap”, Journal of Economic Perspectives 20(4): 133-156.

Peri, G, K Shih and C Sparber (2015), “Stem workers, h-1b visas, and productivity in us cities”, Journal of Labor Economics 33(S1): S225–S255.

Turner, S E, and W G Bowen (1999), “Choice of Major: The changing (unchanging) gender gap” ILR Review 52(2): 289–313.

4,409 Reads