Do single-sex schools help Australians major in STEMM at university?

ABSTRACT

Single-sex schooling is believed to benefit students’ academic achievement and girls’ engagement in science, technology, engineering, mathematics, and medicine (STEMM). The latter is assumed because single-sex environments are meant to neutralise gender stereotypes. Little is known, however, about longer term effects of such schooling. Therefore, we consider whether graduating from a single-sex school increases the uptake of university majors in physical or life sciences. Using data from the 2003 cohort of the Longitudinal Survey of Australian Youth and multinomial logistic regressions, we find that girls who graduated from single-sex schools did not major in physical sciences at university at rates higher than their peers from coeducational schools. Likewise, there are no differences in the take up of life science majors, irrespective of gender or type of school. By contrast, fewer boys who graduated from single-sex schools went on to study physical sciences at university. We discuss the implications of these findings.

KEYWORDS:

• Single-sex schools
• gender segregation in science
• gender and STEMM
• school selectivity
• Australian education

Introduction

Single-sex schooling has been a part of the Australian education system since its inception (Campbell et al., 2009). Although the numbers of students educated in sex-segregated settings are falling (Dix, 2017), all-boys and all-girls schools continue to market themselves to parents not only on their capacity to generate academic excellence but also on their unique potential to foster high-level science and mathematics skills among girls (Forgasz, 2016). As single-sex learning environments are believed to counteract gender stereotypes which adversely affect adolescents’ engagement in secondary science, the argument that such schools might play a role in offsetting the shortages of women in science, technology, engineering, mathematics, and medicine (STEMM) has received serious consideration (Feniger, 2011; Forgasz & Leder, 2019; Park et al., 2018). One of the most prominent theories, the expectancy value theory, suggests that women are less likely to engage in physical science fields because they have less confidence in their mathematical and physical science abilities and they place less subjective task value on them (Eccles, 2011). More recently, the “male over-persistence” argument suggests that the stereotypical beliefs that men are better at mathematics lead them to feel more confident and invest in mathematics and physical sciences (Marsh et al., 2019; Penner & Willer, 2019). As a result, they over-persist in these fields even when they have little aptitude or receive few positive returns to their persistence. Research based on these arguments pays more attention to individual characteristics than to school effects (Penner & Willer, 2019). Nevertheless, a wide range of school characteristics has been shown to influence girls’ and boys’ STEMM enrolments both at secondary and post-secondary levels (Legewie & DiPrete, 2014; Smyth & Hannan, 2006; van Hek et al., 2018).

One question that received less attention is how single-sex schools might affect longer term gender differences in STEMM persistence and the related outcomes. This could be because the potential benefits of single-sex schooling are not easy to disentangle from other school effects. Australia is a case in point here, given that most of its sex-segregated secondary education is not provided by the government but by Catholic or independent schools. These schools charge tuition fees and are typically inaccessible to families with moderate material resources (Forgasz & Hill, 2013). Thus, it is not clear to what extent the previously reported advantages are attributable not to single-sex learning environments but to teacher quality, student admission policies that favour advantageous socioeconomic background, or prior academic achievement. This knowledge gap pertains particularly to longer term benefits of single-sex schooling for students’ engagement with science because prior studies in Australia focused mostly on how single-sex education influenced participation in school science subjects (Ainley & Daly, 2002; Sikora, 2014). Moreover, the Australian case is interesting, because mathematics and science are optional in secondary study pathways that lead to obtaining the Australian Tertiary Admission Rank (ATAR; Sikora & Pitt, 2019). Students can study any combination of subjects with reasonable hopes of securing a high ATAR, which captures individual performance relative to the relevant peer cohort (Pilcher & Torii, 2018) and is used by universities to admit students on competitive basis.

To contribute an insight about its longer term effects, in this study we explore how single-sex secondary education may act as an institutional factor that facilitates the pursuit of specific science majors at university. We use data from the 2003 cohort of the nationally representative Longitudinal Survey of Australian Youth (LSAY; National Centre for Vocational Education Research, 2011). Given that men and women dominate physical sciences and life sciences, respectively (Sikora, 2015), we ask whether single-sex secondary schooling reduces, net of other school and student characteristics, gender differences in the pursuit of physical sciences or life sciences at university. We conceptualize STEMM as comprising life sciences and physical sciences, with the former encompassing medicine, biology, and environmental sciences and the latter involving engineering, mathematics, technology, and physics (Feniger, 2011; Sikora & Pokropek, 2012; Wegemer & Eccles, 2019).

Can single-sex education foster students’ engagement in gender-atypical STEMM fields?

The expectancy value theory suggests that, regardless of the type of school attended, women are underrepresented in physical sciences and overrepresented in life sciences. The key reason is women’s lower self-concept in mathematics and less subjective task value placed on physical sciences (Eccles, 2011). In other words, many young women have less confidence in their mathematical abilities than comparable men, show less interest in mathematics and physical sciences, and believe that performance in these fields is unimportant as it is unrelated to their future occupational objectives. Correspondingly, in contrast to boys, few adolescent girls expect careers in physical sciences, but many expect careers in life sciences (Sikora & Pokropek, 2012). Such occupational expectations are not mere correlates but rather potential causes of the gender gap in the STEMM major choices (Morgan et al., 2013).

The advocates of single-sex schooling claim that it promotes gender equality by encouraging boys and girls to take up and develop vocational interests in gender-atypical fields and occupations (Salomone, 2003). Their arguments usually involve several components. First, all-girls schools provide effective female role models with greater proportions of women who teach mathematics and science (Riordan, 2015). Second, the absence of boys in all-girls schools creates an environment that is free of competition with male peers (Watson, 1997). This environment reduces the pressure for girls to view mathematics and physical sciences as the domains in which only men can excel, and to act on these traditional gender role expectations (Cherney & Campbell, 2011). By analogy, in all-boys schools, students may experience less pressure to comply with gender role expectations, and thus engage more in gender-atypical fields, such as English literature and drama (Norfleet James & Richards, 2003; Salomone, 2003). It is also possible that in single-sex schools, men’s over-persistence at mathematics could be offset effectively. Over-persistence is a culturally reinforced tendency of men to elect mathematics-related tasks, even in the absence of aptitude or positive returns (Penner & Willer, 2019). By contrast, in coeducational schools, girls might struggle to overcome gender stereotype threats (Spencer et al., 1999) while boys may be subject to stronger over-persistence.

Indeed, studies in Britain and in Australia reported that single-sex schooling enhanced students’ confidence and participation in gender-atypical subjects. In the 1970s, among 16-year-old British students, smaller gender gaps in English and mathematics self-concepts were found in single-sex schools than in coeducational schools (Sullivan, 2009). In the 1980s, across 16 schools in Melbourne, girls in single-sex schools were more likely to study science and achieve good results in mathematics and science than girls in coeducational schools (Foon, 1988). In the mid-1990s in England, boys and girls in single-sex schools were more likely than their same-sex peers in coeducational schools to study high-level mathematics (Spielhofer et al., 2004).

Yet, the balance of evidence in Britain and Australia has been far from unequivocal. Studies devoted to mathematics and science teaching in single-sex and coeducational settings often reached mixed conclusions. Two Australian studies that considered a variety of school characteristics showed that during the late 1990s and mid-2000s, students in single-sex schools were just as likely as students in coeducational schools to study physical science subjects in the last year of secondary school, 12th grade, which in Australia is referred to as Year 12 (Ainley & Daly, 2002; Sikora, 2014). In the mid-2000s, girls educated in single-sex schools were marginally less likely than girls in coeducational settings to study life science subjects in Year 12 (Sikora, 2014).

With respect to the longer term effects of single-sex schooling, some research from English-speaking countries concludes that students with experiences of gender-segregated secondary schooling are more likely to engage in gender-atypical fields than students who graduated from coeducational schools. A study based on a sample of British people who were born in 1958 found that those who had single-sex education were more likely to gain post-secondary qualifications in gender-atypical areas (Sullivan et al., 2010). Along the same lines, Norfleet James and Richards (2003) reported that from the 1970s to the 1990s in the United States, boys in single-sex schools took more interest in the humanities than their same-sex peers in coeducational schools. These boys also retained their early interests in their post-secondary education and careers. In the same country, men who went to all-boys schools in the 1990s were more likely to graduate with a bachelor’s degree in gender-neutral fields than men who attended coeducational schools (Karpiak et al., 2007).

However, this body of research is also inconclusive. In Australia, for example, two studies based on a sample of university undergraduates in the 1990s found no evidence that all-girls school graduates went on to pursue mathematics or science at university at higher rates than their peers in coeducational schools (Forgasz, 1998; Lumley, 1992). Similarly, in the United States, women who attended all-girls schools in the 1980s and 1990s declared a major in gender-neutral fields at higher rates, but were only as likely as women from coeducational schools to complete these majors (Karpiak et al., 2007; Thompson, 2003).

This pattern of mixed evidence suggests that the effectiveness of single-sex schooling in counteracting gender stereotypes in the choice of study varies over time and from place to place (Kim & Law, 2012; Law & Kim, 2011; Sikora, 2014). Apart from historical and geographic contingencies, an argument has been put forward that this variability also arises from methodological problems.

Confounding school- and student-level characteristics in research on single-sex education and science

Analytical separation of the effects that single-sex schooling might have on student outcomes requires meeting certain data and sampling criteria which many previous studies have not met. Not only is much of the existing research based on samples without random assignment, but it also confounds the effects of single-sex schooling with a range of omitted factors, such as the socioeconomic status of individual students and the student population within schools, school resources, and selective admission procedures (Halpern et al., 2011; Smyth, 2010). As a result, the apparent benefits of single-sex schooling could be attributed to other school characteristics that coexist with sex segregation. One notable exception is the study by Park and his colleagues (2018), which involved students who were randomly assigned into coeducational and single-sex high schools in Seoul. This prevented confounding the effects of single-sex schooling with other school characteristics. Yet, such “natural experiment” data are not available in other locations. More importantly, no longitudinal data based on such experimental designs are available to establish whether the experience of single-sex schooling produces long-lasting gender differentials. In the absence of such longitudinal and experimental data, the optimal strategy is to control for as many pertinent selection effects as possible.

School-level confounders

First, most single-sex schools in Australia belong to the Catholic or independent sectors that charge tuition fees (Forgasz & Hill, 2013). These schools tend to be located in affluent communities and metropolitan areas and attract students from families of high socioeconomic status (Sikora, 2014). Schools in affluent communities tend to offer more advanced academic subjects, including advanced mathematics, physical science, and life science, than schools in disadvantaged communities (Murphy, 2018; Perry & Southwell, 2014). As students from advantageous backgrounds tend to perform better academically, single-sex schools have more resources and more captive internal markets for advanced science subjects than coeducational schools, which creates “economies of scale”. Single-sex schools usually have fewer difficulties in recruiting qualified teachers because most of them are Catholic and independent schools that can offer more competitive salaries and better work conditions compared to the government schools which are mostly coeducational (Sikora, 2014; Tsolidis & Dobson, 2006). Finally, in Australia single-sex schools may admit more students of Asian background because middle-class Asian parents, who often settle in Australia as skilled migrants, strive to send their children to selective schools located in affluent communities (Butler et al., 2017; Watkins, 2017). Some selective single-sex schools are in the government sector, and they admit students solely based on a placement test. Other schools, outside of the government sector, utilise a range of admission policies.

Over time, single-sex education in Australia has been shrinking (Dix, 2017), which poses challenges in the task of separating the effects of school sector, that is, government, independent, or Catholic, from single-sex versus coeducational schooling. While more than half of the students from the private school sector attended single-sex secondary schools in 1985, the proportion dropped to about 43% in 1995 (Australian Bureau of Statistics [ABS], 1997) and a mere 12% in 2018 (Forgasz & Leder, 2019). The decline in the proportion of students enrolling in single-sex schools merits attention, as early research suggests that when single-sex schools become rare, they become academically selective, which brings about superior educational outcomes (Baker et al., 1995). This might make the detection of unique effects of single-sex schooling even more difficult. Given these considerations, in the analysis presented in this study we make special effort to avoid confounding factors by paying attention to the differences between education sectors, that is, government, Catholic, or independent as well as student admission policies and teacher quality (Halpern et al., 2011; Pahlke et al., 2014; Signorella et al., 2013).

Student-level confounders

Apart from motivational factors such as mathematics self-concept and occupational expectations, university major choices are likely to depend on other characteristics. Secondary students of high socioeconomic status tend to attend schools in affluent communities, and more often enrol in advanced mathematics, physical science, and life science subjects than students from disadvantaged backgrounds (Ainley et al., 2008; Lamb et al., 2001; Teese, 2007). Furthermore, students of East Asian descent tend to outperform their peers at school, particularly in mathematics (Jerrim, 2015). Such differences in socioeconomic background and performance are likely to affect university major choices. As these factors might operate differently for men and women, the best strategy is to analyse student outcomes separately by gender, making sure that all pertinent control variables are considered.

Research questions

To understand the potential longer term effects of single-sex schooling with respect to gendered choices of STEMM majors at university, we distinguish physical science majors from life science majors and consider the following research questions:

1. Does graduating from a single-sex secondary school raise the likelihood that young women will elect a physical science major as their specialisation at university?

   •  (1a) Is this likelihood explained by students’ socioeconomic and ethnic backgrounds combined with school resources and characteristics, that is, school-level confounders? Or is it explained by students’ prior academic achievement histories, occupational motivation, and secondary school subject choices, that is, student-level confounders?

2. Does graduating from a single-sex secondary school increase the likelihood that young women will choose a life science major as their specialisation at university? Can this be attributed to the factors listed in 1a above?

3. Does graduating from a single-sex secondary school raise the likelihood that young men will enrol in a physical science major as their specialisation at university? Can this be attributed to the factors listed in 1a above?

4. Does graduating from a single-sex secondary school increase the likelihood that young men will select a life science major as their specialisation at university? Can this be attributed to the factors listed in 1a above?

Data

LSAY 2003 was built on the Australian sample from the Programme for International Student Assessment (PISA) 2003 conducted for the Organisation for Economic Co-operation and Development (OECD, 2005). The primary focus of PISA 2003 was an assessment of mathematical literacy among 15-year-old students. A total of 10,370 Australian adolescents who participated in PISA 2003 were included in LSAY 2003.

LSAY 2003 extends the PISA survey by annual collections of information on students’ educational and occupational experiences until 2013. More recent cohorts of LSAY (2006, 2009, and 2015) lack comprehensive data either on mathematics self-concept, which prior research singles out as essential for later physical science interest, or on students’ histories of university study. Out of 6,747 participants who completed the last year of secondary school, that is, Year 12, we focused on 3,712 who enrolled in a bachelor’s degree between 2004 and 2013. However, we had to exclude 119 students who changed schools after 2003 because LSAY did not collect information on the schools they moved to. In two out of eight Australian states and territories, Tasmania and the Australian Capital Territory (ACT), students who complete Year 10 in the government sector must enrol in another senior secondary school in order to proceed to Year 11. We had to exclude 225 such students from the analysis, 209 of whom went to coeducational schools. Tasmania and the ACT account for less than 14% of the entire sample, so the impact on our analytical sample is negligible. A total of 192 students did not report their fields of study, and two students came from schools that did not report the proportions of male and female students. In sum, we had to exclude 75 students who, at age 15, went to a single-sex school and 463 students who attended coeducational schools, as we had no information on their later history of schooling. The resulting analytical sample involves 3,174 students who remained in the same secondary school from age 15 until graduation.

Key variables

Dependent variable

The dependent variable has three categories: students’ enrolment in a bachelor’s degree in (a) non-science fields, which represents the baseline for comparisons, (b) physical science, and (c) life science fields. Our definition of physical and life sciences follows prior research in Australia that used detailed information on tertiary fields of study. Physical science majors involve all subfields within the following broad fields listed in the Australian Standard Classification of Education (ASCED; ABS, 2001a): mathematical sciences, physics and astronomy, chemical sciences, earth sciences, information technology, and engineering and related technologies. Life science majors include all subfields listed within the broad ASCED categories of biological sciences, studies related to agriculture and environment, health, behavioural science, and related disciplines. The details of coding are in Appendix 1.

Independent variables: individual characteristics

The array of individual student characteristics we include denote two groups of factors: (a) students’ sociodemographic and socioeconomic characteristics and (b) their motivation combined with school experiences. In the former, we include Asian descent, and the economic, social, and cultural status of family of origin. In the latter, we control for students' occupational expectations related to science, mathematics achievement, and self-concept – all measured at age 15. Moreover, we control for the history of science subject choices in the final year of secondary school because they are electives, that is, students do not have to study mathematics or science to qualify for university entry. The details of these variables are in Appendix 2.

Independent variables: school characteristics

Single-sex schools (all-girls and all-boys schools)

Single-sex schools are identified based on the proportion of girls in each school. Schools with 100% girls are all-girls schools, and those with 0% are all-boys schools. Others are treated as coeducational schools.

Control variables

We include school sector, metropolitan versus provincial location, selective student admission policy, a shortage of qualified teachers reported by principals, whether the school is likely to offer advanced mathematics and science subjects, and the proportion of students of Asian descent. The measurement of all these variables is detailed in Appendix 2.

Method

Weights to adjust for the sampling design of LSAY 2003

Appropriate weights are necessary to account not only for the two-stage stratified sampling of PISA but also for the attrition of respondents in each subsequent follow-up survey of LSAY (Lim, 2011). The LSAY/PISA sample is age based, but, in Australia, students of the same age attend different grade levels across states and territories. Students also commence their university degrees at different ages due to a variety of reasons. Therefore, we must obtain the information about students’ enrolment at university from different LSAY waves, that is, between 2004 and 2013. Likewise, we must collect the history of school science enrolment from different waves. Thus, neither the PISA nor the LSAY weights, which are wave specific, are suitable for the analysis of our sample. To obtain unbiased estimates, the best procedure is to follow the strategy suggested in the LSAY technical report (Lim, 2011). This involves including as controls the variables that were used to construct the LSAY weights. They were the state or territory in which the school is located, the family structure (denoted by an indicator of whether a family is nuclear or has some other forms, such as a single-parent family), and students’ immigration status that distinguished youth born to Australian parents and those born to foreign-born parents. In regression models, we used all theoretically motivated predictors and all weight-related controls. We do not report coefficients for weight-related control variables to conserve space, but they are available upon request.

Multinomial logistic regressions and predicted probabilities

Given that our dependent variable has three possible outcomes, we use multinomial logistic regression. We conduct our analyses separately for men and women to allow for differences in factors that matter for each gender. We rely on predicted probabilities from our models to highlight the gaps in the outcomes between graduates of single-sex schools and of coeducational schools separately for men and women. To minimize the adverse effect of nonresponse on our independent variables, we base our analyses on multiply imputed data with the details of the imputation procedure in Appendix 2. Here, we note that the results on non-imputed data lead to substantively equivalent results.

Results

How different are students in single-sex schools from students in coeducational schools?

Table 1 shows that students in Australian single-sex schools differ from their peers in coeducational settings on some crucial characteristics. Single-sex school graduates hail from families of higher socioeconomic status (men: 0.82 vs. 0.60; women: 0.79 vs. 0.48 of a standard deviation above the mean), and they are more likely to be of Asian descent (men: 20% vs. 11%; women: 20% vs. 11%). These differences are significant, and when expressed in Cohen’s d, which is a standardized mean difference, are of moderate size, with d > 0.20 (Algina et al., 2005). Other differences tend to be smaller.

Table 1. Student characteristics (only graduates who went to university at some point between 2004 and 2013) in single-sex and coeducational schools: proportions, means, and Cohen’s d.

	Men		d	Women		d	Min.	Max.
	Single-sex schools	Co-ed schools		Single-sex schools	Co-ed schools
Dependent variables
Entry into a physical science degree^a	0.18	0.28	0.24	0.06	0.04	−0.11	0	1
Entry into a life science degree	0.23	0.21	−0.04	0.32	0.33	0.02	0	1
Family background
Socioeconomic status ^a b	0.82	0.60	−0.30	0.79	0.48	−0.40	−2.86	2.15
Asian descent ^a b	0.20	0.11	−0.28	0.20	0.11	−0.29	0	1
Mathematics
Mathematics achievement at age 15	613	602	−0.14	588	569	−0.22	259	842
Mathematics self-concept at age 15 ^a	0.50	0.66	0.19	0.26	0.26	0.01	−2.12	2.42
Occupational expectations
Expected a physical science career at age 15	0.17	0.23	0.15	0.05	0.03	−0.11	0	1
Expected a life science career at age 15	0.18	0.14	−0.13	0.33	0.34	0.02	0	1
Subject studied in Year 12 (at about age 18)
Advanced mathematics and physical science	0.15	0.18	0.08	0.07	0.04	−0.20	0	1
Physical science without advanced mathematics	0.25	0.28	0.04	0.21	0.22	0.02	0	1
Advanced mathematics without physical science	0.03	0.03	−0.00	0.03	0.02	−0.07	0	1
Studied life science	0.28	0.32	0.07	0.44	0.57	0.27	0	1
N	399	959		539	1277

Data: LSAY 2003, weighted estimates after multiple imputations of missing data.

^a indicates that the difference between men from all-boys schools and men from coeducational schools is statistically significant at p < 0.01.

^b indicates that the difference between women from all-girls schools and women from coeducational schools is statistically significant at p < 0.01.

Students from single-sex schools do not outstrip their peers in mathematics achievement. On a scale ranging from 259 to 842, women from all-girls schools appear to perform better by 19 points in mathematics than women from coeducational schools (with Cohen’s d of −0.22). But neither this difference nor the difference for men of 12 points reaches statistical significance.

With respect to motivational factors, men educated in all-boys schools had a marginally lower self-concept in mathematics at age 15 than their peers who attended coeducational schools (with Cohen’s d = 0.19). Although women educated in all-girls schools appear to also have lower mathematics self-concept than their female peers graduated from coeducational schools, this difference is not significant, and Cohen’s d is close to zero.

Moving to teenage occupational expectations, men graduated from all-boys schools are as likely as men educated in coeducational schools to expect a career either in physical science or in life science. Similarly, female students in single-sex schools resemble their peers in coeducational schools with regard to their occupational expectations. Next, we find few differences in science and mathematics subject choices in Year 12. Most Cohen d values are low, and there are no significant contrasts by either gender or type of school.

Upon school graduation, men educated in all-boys schools were less likely than men from coeducational schools to engage in physical science university studies (18% vs. 24% in Table 1 with Cohen’s d at 0.24). By contrast, the take-up of life science majors at university happened at rates that did not differ between single-sex and coeducational schools.

Considering the potential school-level confounders is also crucial because single-sex schools and coeducational schools differ (see Appendix 3). Most single-sex schools are Catholic and independent schools, located in large cities with over 1 million inhabitants. They are less likely to experience shortages of qualified teachers but are more likely to offer advanced academic subjects thanks to their access to more human capital, superior material resources, as well as the student market that is captive of high-level subjects.

Multinomial logit models

We present the results of our multinomial logit models, separately for women and men, in the form of relative risk ratios (RRR) in Tables 2 and 3. RRRs show whether, relative to students pursuing non-science qualifications at university, each independent variable raises (RRR is greater than 1) or lowers (RRR is less than 1) the likelihood of majoring in either physical science or life science. We first modelled the total effect that graduating from a single-sex secondary school had on pursuing science majors (not shown in the tables), and later added the potential confounders, or our controls. Because the full models did not explain the total effects of sex-segregated education, we focus our discussion on multinomial logit models with the full suite of predictors.

Table 2. Relative risk ratios of young women choosing a physical science or life science major relative to a non-science major.

	Physical science		Life science
School characteristics
Gender composition of school (reference = coeducational school)
All-girls school	1.093	(0.367)	1.135	(0.210)
School sector (reference = government school)
Catholic school	0.887	(0.263)	0.976	(0.146)
Independent school	1.528	(0.503)	1.165	(0.213)
School location (reference = large city with over 1 million inhabitants)
City (under 1 million inhabitants)	0.644	(0.269)	1.121	(0.197)
Village, small town, and town (under 100,000 inhabitants)	1.169	(0.543)	1.195	(0.205)
School policies and resources
Selective admission to school (reference = no selective admission)	0.563*	(0.159)	0.879	(0.121)
Teacher shortage	0.967	(0.177)	0.999	(0.088)
Proportion of students of Asian descent	3.897	(4.858)	0.621	(0.446)
School that might not offer advanced mathematics, physical science, or life science subject for Year 12 students	0.177*	(0.127)	0.889	(0.176)
Student characteristics
Family background
Family socioeconomic status	0.822	(0.130)	0.840*	(0.073)
Asian descent	0.532	(0.287)	1.005	(0.215)
Mathematics
Mathematic achievement at age 15	0.885	(0.169)	0.957	(0.074)
Mathematics self-concept at age 15	1.659**	(0.255)	1.217*	(0.093)
Occupational expectations
Expected a physical career at age 15	5.681***	(2.319)	3.319***	(0.994)
Expected a life science career at age 15	1.051	(0.308)	2.757***	(0.380)
Relevant subject choice in Year 12
Studied advanced mathematics and physical science	13.822***	(6.521)	5.325***	(1.291)
Studied physical science without advanced mathematics	6.017***	(1.705)	3.442***	(0.530)
Studied advanced mathematics without physical science	1.390	(1.271)	1.043	(0.384)
Studied life science	1.260	(0.319)	3.084***	(0.428)
Intercept	0.014***	(0.008)	0.107***	(0.027)

Data: LSAY 2003.

Note: The sample for this multinomial logistic regression analysis contains 1,816 young women from 249 schools with multiple imputations of missing data. The analysis was undertaken using appropriate weights (as described in the main text in the method section), and the standard errors were adjusted for school clustering. The relative risk ratios of the following weight-related control variables were not reported to conserve space, but they are available upon request: the state or territory in which the school are located, the family structure, and students’ immigration status that distinguished between youth born to Australian parents and those born to foreign-born parents.

*p < 0.05, **p < 0.01. ***p < 0.001.

Table 3. Relative risk ratios of young men choosing a physical science or life science major relative to a non-science major.

	Physical science		Life science
School characteristics
Gender composition of school (reference = coeducational school)
All-boys school	0.604*	(0.125)	0.947	(0.179)
School sector (reference = government school)
Catholic school	1.020	(0.202)	0.995	(0.208)
Independent school	0.925	(0.193)	0.957	(0.202)
School location (reference = large city with over 1 million inhabitants)
City (under 1 million inhabitants)	0.899	(0.184)	0.975	(0.206)
Village, small town, and town (under 100,000 inhabitants)	0.885	(0.203)	1.372	(0.302)
School policies and resources
Selective admission to school (reference = no selective admission)	1.133	(0.198)	0.901	(0.156)
Teacher shortage	1.051	(0.104)	0.841*	(0.072)
Proportion of students of Asian descent	1.474	(0.855)	2.054	(0.994)
School that might not offer advanced mathematics, physical science, or life science subject for Year 12 students	1.538	(0.351)	1.577	(0.379)
Student characteristics
Family background
Family socioeconomic status	0.971	(0.108)	0.950	(0.116)
Asian descent	0.321***	(0.098)	0.762	(0.216)
Mathematics
Mathematic achievement at age 15	0.974	(0.096)	0.981	(0.118)
Mathematics self-concept at age 15	1.506***	(0.145)	0.861	(0.094)
Occupational expectations
Expected a physical career at age 15	3.057***	(0.501)	2.555***	(0.571)
Expected a life science career at age 15	0.930	(0.260)	4.250***	(0.945)
Relevant subject choice in Year 12
Studied advanced mathematics and physical science	11.913***	(3.423)	9.529***	(2.980)
Studied physical science without advanced mathematics	4.447***	(1.011)	4.531***	(1.011)
Studied advanced mathematics without physical science	0.686	(0.370)	0.659	(0.448)
Studied life science	0.911	(0.190)	2.797***	(0.493)
Intercept	0.110***	(0.034)	0.065***	(0.023)

Data: LSAY 2003.

Note: The sample for this multinomial logistic regression analysis contains 1,358 young men from 239 schools with multiple imputations of missing data. The analysis was undertaken using appropriate weights (as described in the main text in the method section), and the standard errors were adjusted for school clustering. The relative risk ratios of the following weight-related control variables were not reported to conserve space, but they are available upon request: the state or territory in which the school are located, the family structure, and students’ immigration status.

*p < 0.05. **p < 0.01. ***p < 0.001.

Does single-sex schooling benefit women by increasing their chances of specialising in physical science and life science at university?

In response to our first and second research questions, Table 2 demonstrates that graduating from an all-girls school has no effect on the pursuit of physical science or life science university majors among women (RRR = 1.093 and 1.135, respectively). As the interpretation of RRRs is not always intuitive, we show in Figure 1 the predicted probabilities derived from RRRs in Table 2. Graduates of all-girls schools and female graduates of coeducational schools were equally likely to enrol in a physical science degree (5% vs. 2%). The same was the case for a life science degree (30% vs. 31%). Thus, we observe no benefits of sex-segregated schooling for young women, in terms of increasing the likelihood of specialising in science at university.

Figure 1. Predicted probabilities of young women enrolling in a physical science or life science major by gender composition of school.

Data: LSAY 2003.

Note: The predicted probabilities are based on the multinomial logistic regression models presented in Table 2 and computed with other variables held at the mean for each type of school.

Does single-sex schooling enhance the chances of men engaging in physical and life science fields at university?

With respect to our third research question, graduating from an all-boys secondary school reduces, rather than raises, the likelihood that young men will choose physical science majors. The RRR = 0.604 in Table 3 translates to 27% of coeducational graduates versus only 16% of single-sex school graduates (see Figure 2).

Figure 2. Predicted probabilities of young men enrolling in a physical science or life science major by gender composition of school.

Data: LSAY 2003.

Note: The predicted probabilities are based on the multinomial logistic regression models presented in Table 3 and computed with other variables held at the mean for each type of school.

Concerning our fourth research question, men do not differ in their choices of life science majors regardless of the type of secondary school they attended (RRR = 0.947 in Table 3). Figure 2 shows that the predicted proportions of male graduates from single-sex and coeducational schools choosing life science majors at university are the same: 21%.

Effects of other school and student characteristics

Turning back to Tables 2 and 3, some school-level predictors in our models affect the likelihood that young men and women pursue university science majors, but they do not explain why attending an all-boys school lowers the chances of young men specialising in physical sciences. First, women who graduated from schools with selective student admission are more likely to pursue a degree in non-science fields than in physical sciences (0.563 in Table 2). Physical sciences are also less likely to be selected by women if they attended a school that most likely did not offer advanced mathematics, physical science, or life science subjects (0.177 in Table 2). The only school characteristics that matter for men other than attending an all-boys school is teacher shortage: Men who went to schools with a shortage of qualified teachers had a lower chance of pursuing life science studies at university (0.841 in Table 3).

Relative to school-level predictors, student characteristics influence more the likelihood that men and women choose science at university. Students of both genders who expected a career in physical sciences had a high level of mathematics self-concept, and studied physical sciences, with or without advanced mathematics in secondary school, were more likely to pursue physical sciences at university. Adolescents with plans to work in either life or physical sciences were more likely to pursue life sciences at university than teenagers who did not plan a science career. Thus, adolescent plans to work in life sciences did not affect the chance of pursuing physical sciences at university for either gender. Coming from a family of high socioeconomic status steered some of the young women, but none of the men, away from life science by the time they reached university. Finally, men of Asian descent were distinctly less likely to major in physical sciences, while this characteristic made no difference for women.

Taking physical sciences with or without mathematics at school kept university science options relatively open, while early specialisation in life sciences created more subsequent path dependency. Students of both genders who took physical sciences in conjunction with advanced mathematics or on its own in the final year of secondary school were more likely to elect either physical science or life science majors. The study of life science subjects in Year 12, however, only raised the likelihood of pursuing a life science major at university.

Discussion

To what extent can single-sex secondary schooling counteract gender stereotypes and benefit young men and women in the longer term by enhancing their chances of pursuing STEMM studies at university in Australia? This analysis demonstrates that for Australians born between 1988 and 1989, graduation from a single-sex secondary school did not increase the likelihood of pursuing either physical sciences or life sciences.

As many existing studies confound the advantages of single-sex education with those of private, selective, or well-resourced schools (Halpern et al., 2011; Smyth, 2010), this study considered a broad array of pre-existing differences between single-sex and coeducational settings. Our results contribute to the growing body of evidence which concludes that the apparent advantages of single-sex schooling are brought about by a range of confounding factors because single-sex schools which cater to students from privileged socioeconomic backgrounds are mostly Catholic or independent tuition-charging institutions, which possess more resources than government schools. In the Australian context, it also matters that single-sex schools are more likely to admit students of Asian descent and of high socioeconomic status due to the selective immigration policies in Australia. Finally, elite schools, including single-sex schools, are more likely to offer advanced academic subjects than other schools.

Nevertheless, single-sex education leads young men to take up physical science majors at university at rates lower than what is typical for their peers from coeducational schools. Given that additional analysis (available upon request) shows that graduates of all-boys schools are more likely to go for business and economics than physical sciences, future research should explore the reasons for this variation to the stereotypical choice of tertiary specialisations. It is likely that all-boys schools provide a more favourable atmosphere for adolescent boys to pursue gender-mixed or gender-atypical fields of study, as suggested by previous studies (Norfleet James & Richards, 2003; Sullivan et al., 2010). It is also possible that the cultural pressures to over-persist in fields of specialisation (Penner & Willer, 2019) that require advanced quantitative skills are lessened in all-boys schools.

Thus far, however, only a small section of the literature on single-sex education focused on the outcomes for young males. Most of the focus has been on young females, as it is widely believed that all-girls secondary schools raise the number of young women who proceed to study physical or life science majors at university. We find no evidence of that.

This study has no benefit of having direct measures for school cultures, gender beliefs of students or teachers, and practices oriented towards counteracting gender stereotypes attached to science. Therefore, we cannot identify the specific reasons for the lower uptake of physical science majors and a higher uptake of business and economics among young men who experienced sex-segregated schooling. Furthermore, this study does not benefit from an experimental design, and, as such, we cannot assert that what we find are causal effects. Nevertheless, our conclusions are based on careful controls for the pre-existing differences between schools and as such are more reliable than the conclusions of studies that failed to include them.

While Australian single-sex schools can still be seen as beneficial through virtue of being a form of elite education, the historical trend for this type of schooling in Australia is a slow and steady decrease in the proportion of students whose parents see sex segregation as a desirable learning environment for their children. In fact, the expectation is that in 20 or 30 years there might be no single-sex schools in the country. In light of our results, it seems likely that parents no longer perceive single-sex schooling as having much potential to reduce the impact of the widely shared gender stereotypes associated with STEMM. Thus, even though the Australian economy has an enormous demand for skilled workers in STEMM (Australian Government, Department of Education, Skills and Employment, 2019), single-sex schools should not be regarded as a panacea to the falling interest in science among Australian teenagers.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Helen Law is a postdoctoral researcher in the Department of Sociology at the University of Tübingen. She is a recipient of the Marie Skłodowska-Curie Actions Individual Fellowship from the European Commission. She completed her PhD in sociology at the Australian National University. Her research interests focus on gender differences in educational and occupational choices, educational inequalities and quantitative methods. Her recent publications include “Gender and Mathematics: Pathways to Mathematically Intensive Fields of Study in Australia” in Advances in Life Course Research and “Why Do Adolescent Boys Dominate Advanced Mathematics Subjects in the Final Year of Secondary School in Australia?” in the Australian Journal of Education.

Joanna Sikora is a senior lecturer in the School of Sociology at the Australian National University. Her research interests are educational inequalities, social stratification and mobility, social inequality in comparative perspective, women in science, and adolescents and their occupational expectations. She has published in a wide range of journals, including Social Science Research, Sex Roles, Science Education, Mathematics Educational Research Journal, the International Journal of Science Education, the Australian Journal of Education, and the International Journal of Educational Development.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

This work was supported by the Marie Skłodowska-Curie Actions Individual Fellowship (MSCA-IF) of the European Commission under Grant number 749068.

Appendix 1. Coding of study fields

Table

ASCED code	Field of study
Physical science
0101	Mathematical sciences
0103	Physics and astronomy
0105	Chemical sciences
0107	Earth sciences
02	Information technology
03	Engineering and related technologies
Life science
0109	Biological sciences
0199	Other natural and physical sciences
019901	Medical science
019903	Forensic science
019905	Food science and biotechnology
019907	Pharmacology
019909	Laboratory technology
05	Agriculture, environmental and related studies
06	Health
0907	Behavioural science

Note: Fields of study are defined by the ASCED (ABS, 2001a). All ASCED subfields within the listed two-digit or four-digit fields have been included unless otherwise indicated.

Appendix 2. Details of measurement and method

Independent variables: individual characteristics

Family’s socioeconomic status

We measure the socioeconomic status of a student’s family by the PISA index of economic, social, and cultural status. This index was standardised to a mean of 0 and a standard deviation of 1 across the member countries of the OECD that participated in PISA 2003. Larger values indicate higher socioeconomic status. In Australia, Cronbach’s alpha for this index is 0.61 (OECD, 2005).

Asian descent

Students are regarded as having Asian descent if at least one of their parents was born in these continents or countries: Mainland Southeast Asia, Maritime Southeast Asia, Chinese Asia, Japan and the Koreas, Southeast Asia, and Central Asia. This classification follows the birthplace listed in the 2001 Australian census (ABS, 2001b).

Occupational expectations – expected a physical science or life science career at age 15

In PISA 2003, students were asked what occupations they expected to have when they are about 30 years old. The responses were coded to four-digit International Standard Classification of Occupations (ISCO-88) codes (International Labour Office, 1990). Using the ISCO-88 codes, Sikora and Pokropek (2012) classified some of the occupations under the field of computing, engineering, or mathematics (physical science) and the field of biology, agriculture, or health (life science). We include only those managerial or professional occupations because they are likely to require a bachelor’s degree.

Mathematics achievement at age 15

We measure students’ mathematics achievement by PISA’s five plausible values that capture students’ numeracy at age 15. They represent the likely distribution of a student’s proficiency in mathematics, and they have a mean of 500 and a standard deviation of 100 (OECD, 2005). In the regression models, we standardise mathematics achievement to a mean of 0 and a standard deviation of 1.

Mathematics self-concept at age 15

We measure mathematics self-concept by a PISA scale that comprises students’ self-evaluation in response to the following five statements: “I am just not good at mathematics,” “I get good marks in mathematics,” “I learn mathematics quickly,” “I have always believed that mathematics is one of my best subjects,” and “In my mathematics class, I understand even the most difficult work.” Higher values indicate a more positive self-concept in mathematics. In Australia, Cronbach’s alpha for this scale is 0.89 (OECD, 2005).

Relevant subject choice in Year 12 (at about age 18)

Australian students who are engaged in physical science tertiary education tend to have studied physical science subjects together with advanced mathematics in Year 12 (Lamb & Ball, 1999). Therefore, we classified subject choice into four categories: (1) students who enrolled in at least one subject in advanced mathematics and at least one subject in physical science, (2) students who took at least one subject in physical science but did not enrol in advanced mathematics, (3) students who took at least one subject in advanced mathematics but did not enrol in physical science, and (4) students who took at least one subject in life science.

Every Australian state and territory adopts its own subject labels with different curriculum content (Ainley et al., 2008). Across all states and territories, advanced mathematics subjects contain significant calculus content. Following Ainley et al. (2008) and Sikora (2014, 2015), school subjects with an emphasis on physics, chemistry, or geology are treated as physical science, whereas subjects with a focus on biology, health, or environment content are treated as life science. Table A2.1 lists all the advanced mathematics, physical science, and life science subjects between 2003 and 2006, that is, in the time period in which the LSAY 2003 cohort were attending Year 12.

Table A2.1 Advanced mathematics, physical science, and life science subjects in Year 12 (2003–2006).

Advanced mathematics

Calculus, mathematics C, mathematics extension, mathematics specialised, specialist mathematics

Physical science

Chemistry, earth science, earth and environmental science, geology, physical science, physics

Life science

Biological science, biology, contemporary issues and science, earth and environmental science, environmental science, human biology, life sciences, marine and aquatic practices, marine studies, multi-strand science, psychology, science life skills, senior science, science of natural resources

Data: LSAY 2003.

Note: This coding follows the categorisation made by Ainley et al. (2008) and Sikora (2014, 2015) and is based on the curriculum contents rather than the name of the subject.

Independent variables: school characteristics that capture the pre-existing differences between single-sex and coeducational schools

School sector

The school sector is indicated by Catholic, independent, and government schools.

School location

We measure the school location by the community in which the schools were located: large city (over 1 million inhabitants), city (under 1 million inhabitants), and village, small town, and town (under 100,000 inhabitants).

School policies and resources – (1) selective admission to school

This variable refers to the school principal’s report on whether the school considers students’ academic records as a relevant criterion for admission.

School policies and resources – (2) shortage of qualified teachers

We specify the shortage of qualified teachers by an index of five items based on school principals’ reports that the shortage of the following teachers hinder instruction at school: qualified mathematics teachers, qualified science teachers, qualified English teachers, foreign-language teachers, and experienced teachers. Positive values of this index indicate shortages. Cronbach’s alpha for this index in Australia is 0.78 (OECD, 2005).

Proportion of students of Asian descent

We consider the proportion of students of Asian descent at school by obtaining such a proportion in the first wave of LSAY 2003. We have discussed how we identify students as having Asian descent in the previous section on individual characteristics.

Schools that might not offer advanced mathematics, physical science, or life science subjects for Year 12 students

Single-sex schools may have more resources than coeducational schools to offer advanced academic subjects because they tend to locate in affluent communities and belong to the Catholic or independent sector (Perry & Southwell, 2014; Sikora, 2014). LSAY did not provide any information on what school subjects were available for students in a particular school. However, we count the number of students who selected advanced mathematics, physical science, and life science subjects in Year 12. We regard those schools that did not have any student choosing those subjects in Year 12 as schools that might not offer any advanced mathematics, physical science, or life science subjects for Year 12 students. It is possible that schools offered those academic subjects, but students did not take up any of them.

Method: multiple imputations of missing values

We use Stata 15 to impute missing values on the independent variables resulting from nonresponses by chained equations. In the sample, 2,663 participants provide complete information on the independent variables. Among the remaining 511 participants, missing values are present in at least one of these variables: family’s socioeconomic status, mathematics self-concept, immigration status, family structure, expectation of a career in physical sciences or life sciences, and relevant subject choice in Year 12 (see Table A2.2).

As PISA allocates five plausible values to each student to denote mathematics achievement, we create five sets of imputed data and assign a different plausible value to each set of imputed data. We follow the PISA recommendations on analyses with plausible values by performing analyses independently on each set of imputed data and aggregating the results to obtain the final estimates of the statistics and their respective standard errors (OECD, 2005, 2009).

Table A2.2 Number of missing values in each independent variable.

	N	Number of missing values
		Total	Men	Women
School characteristics
All-boys or all-girls school (ref. = co-ed school)	3,174	0	0	0
School sector (ref. = government school)
Catholic school	3,174	0	0	0
Independent school	3,174	0	0	0
School location (ref. = large city with over 1 million inhabitants)
City (under 1 million inhabitants)	3,174	0	0	0
Village, small town, and town (under 100,000 inhabitants)	3,174	0	0	0
School policies and resources
Selective admission to school (reference = no selective admission)	3,169	5	4	1
Teacher shortage	3,174	0	0	0
Proportion of students of Asian descent	3,174	0	0	0
School that might not offer advanced mathematics, physical science, or life science subject for Year 12 students	3,174	0	0	0
Control variables: state/territory (reference = New South Wales)
Australian Capital Territory	3,174	0	0	0
Northern Territory	3,174	0	0	0
Queensland	3,174	0	0	0
South Australia	3,174	0	0	0
Tasmania	3,174	0	0	0
Victoria	3,174	0	0	0
Western Australia	3,174	0	0	0
Student characteristics
Family socioeconomic status	3,165	9	5	4
Asian descent	3,148	26	8	18
Occupational expectations
Expected a physical career at age 15	2,941	233	110	123
Expected a life science career at age 15	2,941	233	110	123
Mathematics
Mathematics achievement at age 15	3,174	0	0	0
Mathematics self-concept at age 15	3,172	2	1	1
Relevant subject choice in Year 12
Studied advanced mathematics and physical science	2,922	252	95	157
Studied physical science without advanced mathematics	2,922	252	95	157
Studied advanced mathematics without physical science	2,922	252	95	157
Studied life science	2,922	252	95	157
Control variables
Other family structure (ref. = nuclear family)	3,158	16	9	7
Immigration status (ref. = native students)
First-generation students	3,143	31	11	20
Second-generation students	3,143	31	11	20

Data: LSAY 2003.

Appendix 3. School characteristics of LSAY 2003

Table

	University entrants among graduates between 2004 and 2013 N = 3,174				All secondary schools – first wave of LSAY 2003 N = 10,355 ^a
	All-boys schools	Coeducational schools	All-girls schools		All-boys schools	Coeducational schools	All-girls schools
	(J = 24)	(J = 233)	(J = 25)	J	(J = 24)	(J = 263)	(J = 26)	J
Proportion of schools by sector
Government school	0.02	0.95	0.03	176	0.02	0.95	0.03	207
Catholic school	0.23	0.60	0.17	58	0.23	0.60	0.17	58
Independent school	0.17	0.64	0.19	48	0.17	0.64	0.19	48
Proportion of schools by location
Large city (over 1 million inhabitants)	0.67	0.35	0.80	282	0.67	0.31	0.77	313
City (under 1 million inhabitants)	0.33	0.19	0.16	282	0.33	0.24	0.19	313
Village, small town, and town (under 100,000 inhabitants)	0.00	0.46	0.04	282	0.00	0.45	0.04	313
Proportion or mean for schools (min., max.)
Schools that consider students’ academic record in admission (0, 1)	0.63	0.48	0.32	281	0.63	0.46	0.31	312
Shortage of qualified teachers (−1.20, 2.46) ^b c	−0.16	0.27	−0.60	282	−0.16	0.30	−0.56	313
Schools that might not offer advanced mathematics, physical science or life science subjects ^b c	0.13	0.29	0.00	282	0.13	0.32	0.00	313
Proportion of students of Asian descent (0, 1) ^b c	0.16	0.08	0.19	282	0.16	0.08	0.18	313

Data: PISA 2003, LSAY 2003, unweighted estimates.

Note: We show the information about the first wave if LSAY 2003 to illustrate that the school characteristics from the pooled sample remain similar to the first wave even though LSAY 2003 has the problem of attrition. N refers to the number of students while J refers to the number of schools.

^a One school did not provide information on its proportion of girls, and therefore the school as well as 15 students who attended this school were omitted from this table.

^b indicates that the difference between all-boys schools and coeducational schools is statistically significant at the 0.05 level.

^c indicates that the difference between all-girls schools and coeducational schools is statistically significant at the 0.05 level.