Gender Gap in School Science: Are Single-Sex Schools Important?

Abstract

This paper compares science subject choices and science-related career plans of Australian adolescents in single-sex and coeducational schools. Data from the nationally representative Longitudinal Survey of Australian Youth collected from students who were 15 years of age in 2009 show that, in all schools, boys are overrepresented in physical science courses and careers, while girls are overrepresented in life science. It appears that students in all-girls schools are more likely to take physical science subjects and are keener on careers in physics, computing or engineering than their counterparts in coeducational schools. However, multi-level logit regressions reveal that most apparent differences between students in single-sex and coeducational schools are brought about by differentials in academic achievement, parental characteristics, student’s science self-concept, study time and availability of qualified teachers. The only differences remaining after introducing control variables are the higher propensity of boys in single-sex schools to plan a life science career and the marginally lower propensity of girls in girls-only schools to study life science subjects. Thus, single-sex schooling fosters few non-traditional choices of science specialization. The paper discusses the likely consequences of gender segregation in science and a limited potential of single-sex schools to reduce them. The results of the current analysis are contrasted with a comparable study conducted in Australia a decade ago to illustrate the persistence of the gender gap in science field choices.

Appendixes

Appendix 1 Coding of occupations and subjects

Science subjects listed below have been coded based on their content rather than titles. Online documentation for each subject available from state boards of secondary study has been used.

Physical Science Subjects

Chemistry, Earth and Environmental Science, Earth Science, Geology, Physical Sciences, Physics.

Life Science Subjects

Agricultural Science, Agriculture and Horticulture, Applied Science, Biological Science, Biology, Contemporary Issues and Science, Environmental Science, Geography, Human Biological Science, Life Science, Marine and Aquatic Practices, Marine Studies, Multi-Strand Science, Psychology, Science Life Skills, Science 21, Scientific Studies, Senior Science, Tasmanian Natural Resources.

Physical Science Occupations

These are occupations related to computing, engineering, mathematics or physical sciences. The numerics are the Australian Bureau of Statistics codes (ABS 2006).

1351 information and communication technology managers

2232 information and communication technology trainers

2241 actuaries, mathematicians and statisticians

2300 design, engineering, science and transport professionals

2310 air and marine transport professionals

2311 air transport professionals

2312 marine transport professionals

2320 architects, designers, planners and surveyors

2321 architects and landscape architects

2322 cartographers and surveyors

2326 urban and regional planners

2330 engineering professionals

2331 chemical and materials engineers

2332 civil engineering professionals

2333 electrical engineers

2334 electronics engineers

2335 industrial, mechanical and production engineers

2336 mining engineers

2339 other engineering professionals

2340 natural and physical science professionals

2344 geologists and geophysicists

2349 other natural and physical science professionals

2600 information and communication technology professionals

2610 business and systems analysts, and programmers

2611 information and communication technology business and systems analysts

2612 multimedia specialists and web developers

2613 software and applications programmers

2621 database and systems administrators, information and communication technology security specialists

2630 information and communication technology network and support professionals

2631 computer network professionals

2632 information and communication technology support and test engineers

2633 telecommunications engineering professionals

Life Science Occupations

2341 agricultural and forestry scientists

2343 environmental scientists

2345 life scientists

2346 medical laboratory scientists

2347 veterinarians

2500 health professionals

2510 health diagnostic and promotion professionals

2511 dieticians

2512 medical imaging professionals

2513 occupational and environmental health professionals

2514 optometrists and orthoptists

2515 pharmacists

2519 other health diagnostic and promotion professionals

2520 health therapy professionals

2521 chiropractors and osteopaths

2522 complementary health therapists

2523 dental practitioners

2524 occupational therapists

2525 physiotherapists

2526 podiatrists

2527 speech professionals and audiologists

2530 medical practitioners

2531 generalist medical practitioners

2532 anesthetists

2533 internal medicine specialists

2534 psychiatrists

2535 surgeons

2539 other medical practitioners

2540 midwifery and nursing professionals

2541 midwives

2542 nurse educators and researchers

2543 nurse managers

2544 registered nurses

Appendix 2 Details of measurement and methodology

Independent Variables

Student characteristics

Dummy (zero–one) variables

• 1. Female: coded 1 for females and 0 for males.

• 2. English spoken at home: coded 1 for students who spoke English at home and 0 for everyone else.

• 3. Australian born to Australian parents: coded 1 for students who were born in Australia and whose both parents were Australian born.

• 4. Foreign born student: coded 1 for students born overseas with both parents also born overseas.

• 5. Parent foreign born-coded 1 for students born in Australia with at least one parent born overseas.

• 6. Urban versus rural residence is denoted by a series of dummy variables: small town is up to 15, 000 inhabitants, town is up to 100,000 inhabitants, city-is up to 1 million, and large city denotes locations with over the population of over 1 million.

• 7. Aboriginal student is a self-report coded 1 for all Aboriginal students and 0 for everyone else.

Other variables

• 1. Economic & cultural status of family is the PISA Index of Educational, Social and Cultural Status (ESCS) (OECD 2012b). This composite construct comprises the International Socio-Economic Index of Occupational Status (ISEI); the highest level of education of the student’s parents, converted into years of schooling; the PISA index of family wealth, which denotes the availability of own room, internet and other possessions in the household; the PISA index of home educational resources which include textbooks, computer and educational software ownership; and the PISA index of cultural possessions including assets such as books of poetry or works of art in the family home (OECD 2012b). This index is standardised to the mean of 0 and the standard deviation of 1, across the OECD countries. The Cronbach’s alpha reliability of this index in 2009 for Australia was 0.59. ESCS is a conceptually strong measure of student socio-economic advantage as it includes a broad range of cultural resources pertinent to student educational outcomes.

• 2. Academic performance in science is measured by PISA’s five plausible values (OECD 2009) which indicate students’ ability to use science-related concepts in adult life. More detail on plausible value methodologies and the use of Balanced Repeated Replication (BRR) weights with Fay’s adjustment (OECD 2009) is in Methods of Estimation below, but for a comprehensive explanation of these methodologies the reader is referred to the PISA Data Analysis Manual (OECD 2009).

• 3. Minutes per week study science is science learning time at school computed by the OECD by multiplying the number of minutes on average in each science class by number of class periods per week (OECD 2012b). It was divided by 100 to facilitate the presentation of coefficients.

• 4. Self-confidence in science skills is a single question indicator of how well the student thought they did in science. Five answer categories ranged from ‘very poorly’ denoted by 0 to ‘very well’ denoted by 1.

School characteristics

Dummy (zero–one) variables

• 1. Boys-only school and Girls-only school are indicators identifying schools with 0 and 100 % of female students.

• 2. Government school, Independent school, Catholic school.

• 3. State or territory: New South Wales, Queensland, Australian Capital Territory, Victoria, Western Australia, Northern Territory, Tasmania.

Other variables

• 1. Selective admission to school is a three category question ‘How often student’s record of academic performance (including placement tests) is considered when students are admitted to your school?‘which was converted to two answer categories: ‘0’ Never and ‘1’ which combines Sometimes + Always.

• 2. Shortage of teachers is the OECD Index on Teacher Shortage constructed from four questions measuring the principal’s perceptions of potential factors hindering instruction at school: ‘Is your school’s capacity to provide instruction hindered by any of the following issues? A lack of qualified science teachers? A lack of qualified mathematics teachers? A lack of qualified English teachers? A lack of qualified teachers of other subjects? The Cronbach alpha for this index in Australia in 2009 was 0.84 (OECD 2012b).

Methods of Estimation

Multivariate analyses in this paper are two-level hierarchical logit models with school-level and student-level covariates (OECD 2012b; Raudenbush and Bryk 2002). The dependent variables denote the chances of studying 1) one or more life science subjects in Year 12) one or more physical science subjects in Year 12, 3) expectation at age 15 of a career related to life science, 4) expectation at age 15 of a career related to physical science. The two-level logit model, best suited to such variables, has the following functional form:

$$ \log \mathrm{it}\left({Y}_{ij}\right)={\gamma}_{00}+\mathbf{X}\boldsymbol{\upbeta } +\mathbf{Z}\boldsymbol{\upeta } +{u}_{0j} $$

where Y_ij denotes the dependent variable for student i in school j and γ ₀₀ is the average intercept across schools. X is a vector of student-level explanatory variables and β is a vector of regression coefficients corresponding to variables in vector X. Z is a vector of school‐level covariates corresponding to the vector of regression coefficients η. The error component u_0j varies between schools. In multilevel logit models, the individual error term, denoted by e_ij, is omitted due to identification problems (Raudenbush and Bryk 2002).

To measure student achievement Y09 uses PISA’s plausible value methodologies and an incomplete balanced matrix design, which means that students answer a sample of, rather than all science test questions. This is why descriptive estimates of student achievement in science in this paper are based on five plausible values for each student and computed by the OECD-recommended methods, including balanced-repeated replicate weights with Fay adjustment (OECD 2009).

Because of the use of plausible values and imputations of missing values (Mislevy et al. 1992), all estimates in multivariate analyses have been obtained using multiple imputation methodology. This involves fitting five sets of models, each with one plausible value, and then combining these values using the Rubin rule (Little and Rubin 1987) as per OECD recommendations (OECD 2012b). For estimations of multilevel models MPlus version 7 was used because of its ability to handle complex weights in hierarchical estimations.

The Y09 sample is representative of 15 years old, not of students in any particular grade. All analyses of career plans in this paper have been weighted back to the original PISA/Y09 population, while all analyses of subject choices have been weighted to such subpopulation of students, as remained after 1) those who failed to participate in the survey's subsequent waves and 2) who changed schools after 2009, or 3) who did not answer the question about changing school since 2009, were excluded from the analysis. Only student level weights have been used, as Y09 data have been collected with a sampling mechanism that is invariant across the sample clusters, so school weights are not necessary (Asparouhov 2004).