Sex segregation by field of study and the influence of labor markets: Evidence from 39 countries

Introduction

The growing number of women in higher education around the world over the last few decades is remarkable. Women now make up the majority of students receiving 2-year and 4-year degrees in countries all over the world: from developed countries such as the United States, Germany, and Japan, to developing and middle-income countries such as Algeria, Columbia, and Saudi Arabia ¹ (also see Buchmann and DiPrete, 2006; Schofer and Meyer, 2005; Shavit et al., 2007). This educational success is an important first step in achieving gender equality across other realms of society (Leach, 1998; World Bank, 2011). Although women’s achievements in higher education across the globe deserve celebration, if one examines these educational successes more closely, there is evidence that gender inequality persists under the surface.

Women continue to be overrepresented in some fields (humanities, services, and education) and underrepresented in others (math, science, and engineering), a situation referred to in the literature as sex segregation by field of study (Bradley, 2000; Charles and Bradley, 2002, 2009). What might be most surprising is which countries report the most integrated field choices by men and women in their higher education system. Charles and Bradley (2009) examined sex segregation by field across 44 economically and culturally diverse countries. The lowest overall levels of sex segregation by field were found in Columbia, Bulgaria, and Tunisia; the highest levels were in Finland, Hong Kong, South Africa, and Switzerland (pp. 940–942). Even countries well known for their strong support of gender equality, such as Sweden, Norway, and Denmark, reported significant levels of sex segregation by field. Embracing egalitarian norms does not seem to be enough to eliminate gendered choices when it comes to field of study. As Charles and Bradley (2002) acknowledge, ‘Sex segregation by field of study is generated and maintained by extremely resilient, taken-for-granted beliefs about gender differences that are not necessarily incompatible with mandates for gender equality’ (p. 575).

Sex segregation across college majors is of crucial importance for women’s equality outside of the academic world because different fields of study have access to specific labor market opportunities and rewards. Employment opportunities often follow from one’s choice of academic field, and consequently, wages and salaries are similarly linked to this choice (Organisation for Economic Co-operation and Development (OECD), 2006; Roska, 2005; Smyth and Steinmetz, 2008). Not surprisingly, a substantial portion of the wage gap between males and females can be explained by their often different academic backgrounds (Bobbitt-Zeher, 2007). Furthermore, many countries are concerned about the lack of women and minorities pursuing degrees in science and engineering because it is seen as a signal that their nation’s human capital resources are being underutilized, placing them at an economic and perhaps even a military disadvantage (Schofer et al., 2000).

Although we want to protect men’s and women’s freedom to choose an academic field according to his or her preferences (even if those preferences were formed by a gendered culture), we also need to acknowledge the important consequences of these choices. If women are to gain an equal footing with men in labor markets, political institutions, and larger society, we need to better understand what factors – economic, political, educational, and cultural – influence sex segregation by field of study. The aim of this article is to determine how these unique country characteristics, particularly those associated with a country’s labor market, may work to reduce or intensify sex segregation by field in higher education.

Other studies examine how the gender make-up of a country’s labor force influences sex segregation by field (Charles and Bradley, 2002, 2009); this study introduces two new labor market variables in its analysis: a country’s national unemployment rate and its economic protections for women. Both are important features of a country’s labor market and may help explain the variation in sex segregation levels across countries.

The second unique contribution of this article is its estimation method. This investigation seeks to explain sex segregation patterns by field across a diverse group of countries. The system generalized method of moments (system-GMM) is an estimation method best used with samples that have few time periods and a relatively large number of cross sections; these characteristics describe the sample for this study. Furthermore, the independent variables used in the model (features of a country’s labor market, as well as characteristics of its higher education system) are likely not strictly exogenous. The system-GMM estimation method is designed to handle models that face the challenge of endogeneity and unobserved heterogeneity. Thus, this estimation method is an ideal technique to carefully identify the macro-level factors that exert an important influence on sex segregation by field of study.

The article begins with a theoretical discussion and literature review of the macro-level factors that influence sex segregation by field. Section ‘Data and methodology’ outlines the sources of data and dependent and independent variables used in the analysis. This section also includes a discussion of the empirical methods. Section ‘Empirical results’ highlights the research results. The article concludes with a discussion of the main findings and draws some general conclusions.

Theoretical framework and previous findings

Although previous scholars highlight economic and educational factors that help explain cross-national variation in women’s representation in higher education, less evidence exists about the impact of labor market characteristics on sex segregation by field. After a brief review of studies that examine the influence of economic conditions on student field choice at the micro level, we turn our attention to the important theoretical considerations at the macro level.

Students’ expectations about their future labor market outcomes are an important determinant of their major choice in higher education (Altonji et al., 2012, 2015). After controlling for high school preparation and gender role attitudes, Frehill (1997) demonstrates that women are more likely than men to select jobs based on social meaning and intrinsic rewards than monetary rewards. Zafar (2013) confirms the conclusions of Frehill. Zafar (2013) finds that although male and female students have similar preferences regarding outcomes at college, their preferences differ with respect to employment outcomes. Nonpecuniary rewards in the workplace are more valued by females. These nonpecuniary determinants explain about half of the field of study choices for men and more than three-fourths for women (pp. 584–585). Gemici and Wiswall (2014) also examine the determinants of the gender gap in college major and find that males were more responsive than females to the increase in the relative prices of science and business skills during the 1980s and 1990s in the United States, and this contributed to a widening of the college major gap during this time period.

Rational choice theory offers an explanation for the divergent patterns of field choice by men and women. Rational choice contends that expected involvement with the formal labor market and family obligations is an important determinant of the choice of college major (Becker, 1991; Polachek, 1981). In this view, women anticipate that they will assume most of the responsibility for the family and the home and thus choose fields that accommodate their dual role of worker and caretaker (Becker, 1991). Women may also opt for an occupation in which reduced working hours and/or extended leave are not penalized (Polachek, 1981; Zellner, 1975). Thus, countries with higher numbers of working women might also experience higher levels of sex segregation by field of study since many of these working women choose fields (such as education and nursing) that help them juggle their home and work responsibilities. Consequently, economies with large numbers of working women are expected to have high levels of sex segregation by field as students choose fields that prepare them for these gender-divided labor markets.

A few studies formally investigate the link between sex segregation by field and labor markets at the national level. Carlo Barone examined the sex segregation patterns for eight European nations across three decades. He finds that sex segregation by field declined little over this period, and this lack of decline mirrors the lack of decline in occupational segregation in these countries. He concludes, ‘Postindustrial employment structures sustain and reinforce these cultural dynamics rather than counteract them’ (Barone, 2011: 173). As Charles and Bradley (2009) argue, postindustrialism is characterized by growth in female-labeled service occupations. According to rational choice theory, women will be aware of the high demand for female labor in these occupations when choosing their field of study and thus will be more likely to pursue degrees in social science, education, human development, or health degrees (Estèvez-Abe et al., 2003). These field choices often track women into female-dominated occupations (Frank and Meyer, 2007). Indeed, scholars find that countries with a high representation of women in their labor force often also experience high degrees of occupational segregation (Charles and Grusky, 2004; England, 2005; Hansen, 1997; Rubery et al., 2004; Smyth and Steinmetz, 2008). ²

Charles and Bradley (2002, 2009) examine the connection between women’s participation in labor markets and sex segregation by field of study. In their multivariate analysis of 12 industrialized countries, they found that after controlling for features of a country’s education system as well as national attitudes toward gender equality, countries with high levels of women’s labor force participation also experienced high levels of sex segregation by field of study (2002). The authors examined other labor market variables in a 2009 study that included both economically advanced and less developed countries. After controlling for economic development, features of a country’s educational system, and girls’ levels of math achievement, Charles and Bradley show that labor market variables exerted a stronger influence on sex segregation by field in developing and transitional countries than in advanced industrial countries. Particularly for developing and transitional countries, the percentage of the labor force that was female and the percentage of females in professional occupations were both associated with higher levels of sex segregation by field. Their findings confirm a connection between field choice and women’s role in labor markets.

Because this study is also considering the impact of national unemployment rates on sex segregation by field, women’s role in the labor market is measured with female employment rates, not female labor force participation rates, to minimize the correlation between these variables. ³ Building on the evidence presented above, countries with high levels of female employment rates are more likely to exhibit postindustrial economic features, such as large service sectors that employ large numbers of women. If, as the rational choice perspective contends, female students choose their college major, in part, based on likely employment opportunities following graduation, countries with higher female employment rates are expected to experience higher levels of sex segregation by field of study. Students typically select a college major a year or two before they graduate. Thus, the female employment rate (and corresponding structure of the labor market) that might influence their major decision is the rate that existed 2 years prior to graduation. Thus, this variable will be lagged 2 years to best capture the labor market conditions that actually influenced their major choice. ⁴ Against this background, I will test the following hypothesis:

There are other important features of labor markets that deserve examination for their impact on sex segregation by field. For instance, the overall health of a country’s labor market is expected to have implications for sex segregation by field of study. Human capital theory argues that individuals view higher education as an investment; the way to generate a return on this investment is to find employment and earn wages following graduation (García-Aracil, 2008). Previous studies found that labor market conditions have an important influence on other facets of human capital, including college enrollment (Betts and McFarland, 1995; Hershbein, 2012), college completion (Dynarski, 2008; Kahn, 2010), and graduate school attendance (Bedard and Herman, 2008; Johnson, 2013).

Beffy et al. (2012) propose that students consider both the consumption value of a potential major and its investment value. Assuming that each student knows her personal consumption value of a particular major, the student uses outside signals to form expectations regarding future labor market opportunities and earnings. These expectations help students assign investment values to potential majors and ultimately guides them to choose a major that maximizes their utility. The unemployment rate is a potential signal that students consider when forming these expectations about future labor market conditions.

Previous research demonstrates that unemployment rates vary widely among graduates of different fields of study (Thomas and Zhang, 2005). If the unemployment rate is a powerful influence on a student’s choice of field of study, it might also have an important impact on sex segregation by field. For instance, when jobs are plentiful (reflected in a low national unemployment rate), students are freer to indulge their preferences when making their field of study choices since they have a greater chance of finding employment and capitalizing on their investment. If students’ preferences are influenced by gender norms, these field choices act to increase sex segregation by field of study. When jobs are scarce (indicated by a high national unemployment rate), one would expect students to more seriously consider those fields with better employment prospects and place less emphasis on their individual preferences and traditional gender norms when choosing a college major and future career.

Two previous studies explore the relationship between unemployment and a country’s level of sex segregation by field of study. Blom et al. (2015) examined the relationship between economic conditions and choice of college major in the United States from 1960 to 2011. After controlling for the difficulty of the major, the gender balance of the major, breadth of job opportunities, pathways to graduate school, and subsequent geographic mobility, Blom et al. find that higher unemployment encourages women to enter male-dominated fields. In addition, students of both genders choose more difficult, higher paying fields such as science, technology, engineering, and mathematics, during recessionary economic conditions. Urrutia (2015) confirms the conclusions of Blom et al. (2015). Examining US economic conditions from 1980 to 2010, Urrutia also finds that students of both genders are more likely to choose a higher paying major when the unemployment rate at the major decision time increases.

The timing of a student’s choice of academic field also needs to be considered. Graduates from tertiary programs are likely influenced by the unemployment rates that existed when they chose their field of study, not necessarily the unemployment rate when they graduated. As graduation nears, students face high costs to change their field of study. In their empirical analysis, Blom et al. found that the unemployment rate that occurred when students were 20 years of age had the strongest influence on their ultimate choice of field. Urrutia assumed that the unemployment rate influenced students’ decisions by age 21, but noted that the age could be changed to 19 or 20 without significantly altering the results. (In the United States, most students graduate from bachelor’s programs around the age of 22.) Based on these arguments, it can be expected that

It is often suggested that greater gender equality in labor markets will promote gender integration in field choices. Human capital theory predicts that women are more likely to invest in their own human capital when they will reap the rewards from their time, energy, and monetary investments in their education (Charles and Bradley, 2009). Indeed, gender discrimination in pay and promotion opportunities reduces the return to female market work and tends to depress female labor supply (Jaumotte, 2003). Girls and women will more often seek training in male-dominated fields in contexts where they have reason to expect continuous employment and where they perceive more opportunities for women in high-status occupations (Estèvez-Abe et al., 2003; Polacheck,1978).

Evidence shows that countries that protect women’s economic rights (such as equal pay for equal work, the right to employment without obtaining permission from one’s husband or male relatives, and anti-discrimination laws) exhibit higher labor force participation by women (Doepke et al., 2011). After controlling for numerous labor supply characteristics, Beller’s (1982) empirical analysis demonstrates that Title VII of the US Civil Rights Act of 1964 (a federal equal opportunity law) increased a working woman’s probability of being employed in a male occupation relative to a man’s probability. Geddes et al. (2012) find that expanding women’s economic rights in the United States resulted in higher relative rates of school attendance for girls; this effect was largest for those between the ages of 15 and 19 years. So at least in the context of the United States, it seems that greater economic rights for women increase human capital investments by girls and women and encourage them to enter traditionally male-dominated fields.

Extending beyond the United States, Breen and Garcia-Penalosa (2002) develop a model to explain occupational segregation patterns across different countries. In their model, agents have imperfect information regarding their probability of success in different occupations and select careers based on these probabilities. The model shows that even if men and women have identical preferences, they will choose different careers based on their gender differences in beliefs about success in different occupations. These choices intensify occupational segregation. Equal protections for women in the form of equal opportunity law and other institutional protections of women’s economic rights might help change women’s perception of their probability of success in male-dominated occupations, reducing these segregation patterns.

Empirical evidence confirms Breen and Garcia-Penalosa’s predication. Farley Ordovensky Staniec (2004) investigated whether differences in major choice between men and women can be explained by expected labor market returns. Farley Ordovensky Staniec (2004) finds that ‘… a significant reason why women are less likely than men to choose SEM [science, engineering, and math] majors is that women’s expected returns to SEM, relative to majoring in other fields, are lower than those for men’ (p. 560). She concludes that one way to increase women’s representation in these SEM fields is to ensure equal returns for men and women employed in these areas after graduation.

This study builds upon this research by seeking the answer to a related but different question: Do countries that protect the economic rights of women experience more women seeking degrees in male-dominated, higher paying fields, and thus have lower levels of sex segregation by field of study? Women living in countries that have laws protecting women’s right to work can be reasonably assured that they will have equal access to these traditionally ‘male’ jobs and will earn a wage comparable to a similarly trained man. Consequently, it can be expected that

Although the central focus of this analysis is to determine the influence of labor market conditions on sex segregation by field of study, other characteristics that could exert a unique influence on sex segregation by field, or be correlated with labor market conditions, also deserve attention. One such national characteristic is the size of the government sector. Governments help determine the nature of employment opportunities in an economy (Steiber and Haas, 2012). Countries with larger government sectors (often measured by the amount of government spending) exhibit more female-friendly employment opportunities in public services such as care jobs for children, the elderly, and the disabled as well as administrative and other bureaucratic roles. Thus, increases in women’s labor force participation rates have ‘gone hand in hand with the expansion of welfare states’ (Datta Gupta et al., 2008: 66). Because of its potential correlation with women’s employment, the size of a country’s government is considered in the models that follow.

Labor market characteristics, such as the prevalence of women in the formal economy and economic protections for working women, might also be correlated with dominant religious practices. Formal religious institutions, which impact cultural norms, social rules, and behaviors, can have a significant impact on gender roles and attitudes and can contribute to persistent gender inequities in education as well as other spheres of life (Braunstein, 2014; Cooray and Potrafke, 2011; Inglehart and Norris, 2003; Norton and Tomal, 2009). A student’s choice of field of study might reflect this inequality, and thus, some of the models include controls for religion.

A country’s support of gender egalitarianism could also shape national sex segregation patterns. As one would expect, countries whose citizens value gender equality experience more women pursuing higher education, participating in labor markets, and engaging in politics (Corrigall and Konrad, 2007; Cotter et al., 2011; Fortin, 2005; Inglehart and Norris 2003). As Charles and Bradley (2002) argue, norms and beliefs regarding appropriate gender roles are deeply rooted in a country’s culture and reflect ‘the historical interaction of a broad array of attitudinal, institutional, and structural factors’ (p. 577). These gender attitudes can serve as the cultural background in which curricular choices are made. Charles and Bradley (2002) investigated the role of gender egalitarian values in their cross-sectional study of 12 industrialized countries. Their findings demonstrate that after controlling for features of a country’s educational system and women’s labor force participation rates, gender egalitarian attitudes modestly reduce sex segregation by field of study.

Previous research demonstrates that many other factors shape sex segregation by field of study across countries. For instance, modernization and societal development are identified as key macro-level influences. Modernization (often accompanied by urbanization, industrialization, and economic growth) can disrupt traditional ways of life, including gender roles, and encourage greater individualization for all. Society becomes more tolerant of women seeking fulfillment outside of their family roles and new channels develop that allow women to pursue higher education and employment outside the home (Goode, 1963; Inkeles and Smith, 1974). Developed countries might also have more robust science and engineering employment opportunities, which would provide another incentive for women to gain expertise in these fields. Thus, this line of reasoning argues that the more economically developed and ‘modern’ a country is, the lower their sex segregation by field of study.

Other scholars argue that even in modern, egalitarian societies, gender continues to exert an important influence on one’s life experiences, expectations, and ambitions (Charles and Grusky, 2004; Correll, 2004; Ridgeway, 2006). The influence of gender can be further intensified in those cultures that embrace the idea of individual fulfillment and thus value self-expression (Frank and Meyer, 2007; Meyer and Jepperson, 2000). Wealthy, modern societies most clearly exhibit this form of expression (Beck and Beck-Gernsheim, 2001; Inglehart, 1997). According to this view, because gender is often a central part of one’s identity, it plays a significant role when making important life choices, including choosing one’s field of study. Pursuing gender-conforming fields and occupations is a way for individuals to affirm their gender identity (Faulkner, 2007; Ridgeway, 2006; Xie and Shauman, 2003).

Ramirez and Wotipka (2001) investigated the percentage of women enrolled in science and engineering fields at the tertiary level across 67 countries in 1972 and 1992. They found that the level of societal development, measured using gross domestic product (GDP) per capita, had no influence on this percentage after controlling for political regime and men’s and women’s representation in higher education across all fields. Using models that also controlled for women’s role in the labor force, features of a country’s higher education system, and girls’ achievement in math, Charles and Bradley (2009) find a positive relationship of GDP to sex segregation by field but only for developing/transitional societies. The positive relationship fades when societies reach a certain level of economic prosperity. Bradley (2000) reached a similar conclusion.

These findings do not lend support to the idea that modern economies embrace new customs that disrupt traditional patterns of gender in curricular choice – quite the opposite. These authors conclude that material security and economic prosperity make it more possible for students, perhaps especially female students since they are less likely to be the family breadwinner, to sacrifice material rewards in order to pursue their passions. This privilege to make curricular choices based on one’s passions leads to more sex segregation by field. Charles and Bradley’s (2009) empirical results support the idea that countries that value self-expression and individual fulfillment (usually wealthier countries) exhibit greater sex segregation by field of study. In their words, ‘As individuals seek to express their essential (male and female) selves, the gender labeling of academic fields intensifies, and distributions across these fields become more closely aligned with gender-specific curricular dispositions’ (Charles and Bradley, 2009: 960). Other research findings agree with this claim. Several country-level case studies demonstrate that as students have more freedom to choose their own fields of study, sex segregation increases (Catsambis, 1994; Kontogiannopoulou-Polydorides, 1991; Plateau, 1991; Stolte-Heiskanen, 1991).

The nature of a country’s political system could also play a role in shaping cross-national patterns in sex segregation by field of study. Scholars argue that democracies allow women to enjoy greater freedoms, both politically and otherwise, and thus, they may act to disrupt traditional barriers (Cooray and Potrafke, 2011). According to this view, democracies and greater political freedoms for women are expected to have lower levels of sex segregation by field. However, following the arguments above, greater freedoms in the political spheres might intensify sex segregation by field if these freedoms encourage self-expression that is heavily influenced by gender.

The structural features of a country’s higher education system can also be a key factor that influences sex segregation by field of study (Charles and Bradley, 2002). Some argue that in those contexts where pursuing a tertiary degree is an exclusive activity reserved for intellectual or social elites, sex segregation by field declines. In these educational settings, students might more readily identify with their privileged place in society instead of with their gender (Della Fave, 1980; Gecas, 1991). Thus, the company of other elites when making field choices minimizes the influence of gender. As these programs expand to include students from more diverse backgrounds, gender might play a more important role in choice of field.

The empirical evidence is mixed and seems to depend on the level of economic development of the country as well as the field. Holding other factors constant, Charles and Bradley (2009) found that in less developed economies expanding the size of the higher education system acted to increase sex segregation by field by of study. Specifically, they found that women’s representation in engineering declined while it increased in humanities, social sciences, and health/other fields, acting to increase overall sex segregation by field (Charles and Bradley, 2009: 955–956). This result did not hold in advanced industrial countries. In their 2002 study that consisted of 12 industrialized countries, the size of the tertiary system did not have a significant impact on sex segregation by field after controlling for women’s labor force participation and national support of gender equality.

The diversity of tertiary programs is also posited as a potentially important influence of sex segregation by field. Highly diversified degree and field options maximize the ability of students to exercise their expressive choices and act to accommodate the increased numbers of female students (Bradley and Charles, 2004; Frank and Meyer, 2007). In this view, as 2-year academic and vocational programs become more widely available, field choices become more gendered. These programs often emphasize specific job training, and since labor markets are heavily segregated by sex, so too are these academic programs (Charles and Bradley, 2002; Smyth and Steinmetz, 2008).

In general, the evidence suggests that the more the degree options at the tertiary level of education, the more the segregated field choices are by sex. Charles and Bradley (2002) found that countries with large nonuniversity sectors have higher women’s representation in education, humanities, and, to a lesser extent, social sciences and lower women’s representation in engineering and natural sciences. Charles and Bradley (2002) as well as Rawlings (2007) find that an expansion of 2 years and vocational colleges increased sex segregation by field. Charles and Bradley (2009) found that the effects of a more diversified higher education system were more pronounced in advanced industrial countries than in developing ones, although this effect varied a great deal by field.

The degree of tertiary participation by women also influences sex segregation by field of study. As Charles and Bradley (2002) argue, when women at the tertiary level are relatively few, women’s curricular choices have to conform to the existing programs and thus women are channeled into less traditional fields, reducing sex segregation by field. When women start entering tertiary programs in larger numbers, new programs are established or existing ones expanded to accommodate women as a group. These more accommodating programs may be more compatible with traditional ideas about gender and thus may work to increase sex segregation by field. On the other hand, others claim that women’s increased participation in higher education may expand women’s access to traditionally male-dominated fields (Bradley and Ramirez, 1996; Davies and Guppy, 1997). In this view, simply incorporating women into the higher education experience can empower them to explore nontraditional fields (Ramirez and Wotipka, 2001).

Fewer empirical studies examine the connection between women’s participation in tertiary education and its impact on sex segregation by field. After controlling for a country’s economic and political features, Ramirez and Wotipka (2001) show that as women’s enrollment increased in non-science and engineering fields, their enrollment in science and engineering fields also increased. Their results provide some support for the idea that enhanced participation in higher education by women empowers them to pursue such male-dominated fields as science and engineering. In their multivariate analysis, Charles and Bradley (2002) found no significant influence of women’s tertiary participation on sex segregation by field.

Data and methodology

The data used in this analysis come from a variety of sources, including The United Nations and The World Bank. Refer to Table 6 in Appendix 1 for a more detailed discussion of the sources and precise variable descriptions. The diverse group of countries used in the analysis is listed in Table 7 in Appendix 1. The information on the number of male and female graduates by field of study comes from United Nations Educational, Scientific and Cultural Organization’s (UNESCO) Institute for Statistics and uses the International Standard Classification of Education (ISCED) 2011 classification system. These graduates are from programs classified as ISCED 5 (short-cycle tertiary education – general academic or vocational) or ISCED 6 (bachelor’s or equivalent). ⁵ The fields of study considered in this analysis are as follows: Education; Humanities and Arts; Social Science, Business, and Law; Science; Engineering, Manufacturing, and Construction; Agriculture; Health and Welfare; and Services. ⁶ Table 8 in Appendix 1 contains more detailed information on this field classification system (see also UNESCO, 2012). As others point out (see for instance, Charles and Grusky, 1995), a different field classification system would produce different values when measuring sex segregation by field. For our purposes, the key advantage of UNESCO’s system is that the field categories are consistent across countries, which allows for easy comparisons of estimated coefficients on the independent variables.

This article explores what factors account for the variation in sex segregation by field across countries; thus, a measure of sex segregation by field serves as the dependent variable in the regression analysis that follows. Two measures of sex segregation by field are common in the literature – the Index of Dissimilarity and the Index of Association. ⁷ The weakness of the Dissimilarity Index, especially in comparative research, is that the overall number of graduates in each field, not just their distribution, influences its value. As Karen Bradley (2000) explains, ‘… measures of the degree of programmatic differentiation by sex cannot be disentangled from variability in the sizes of fields’ (p. 6). Because there is considerable variation in the size of fields of study across countries, the Index of Association (A) is used instead.

A has the distinct advantage that it is invariant to the size of fields of study. It was developed by Charles (1992) and Charles and Grusky (1995) and used by Charles and Bradley (2009) and Bradley (2000) in their empirical studies. For each country and point in time, A is defined as

A = exp { 1 J × ∑ { In ( W j M j ) − [ 1 J × ∑ In ( W j M j ) ] } 2 } 1 2

where W_j and M_j are the numbers of women and men graduates, respectively, in field j, and J is the number of fields. The range for A is 1 (indicating a distribution across fields that corresponds to the gender distribution of graduates as a whole) to infinity (indicating near-perfect segregation). A gives the multiplicative factor by which women (or men) are overrepresented in the average field of study in the given country.

The influence of labor markets on sex segregation by field of study is tested using the following base model

A i t = α + β 1 A i t − 1 + β 2 Ln ( R G D P p c ) + β 2 F / M R a t i o Pr i m a r y E n r o l l i t + β 4 H u m a n C a p i t a l i t + β 5 S t r u c t D i v e r s i t + β 6 T e r i t a r y S i z e i t + β 7 F e m a l e P a r t i t + β 8 O E C D i + ε i

(1)

A country was included in the sample if UNESCO reported its number of graduates by sex and field and if data on the above control variables were also available. ⁸ The final sample includes 39 countries forming an unbalanced panel of 127 observations for the period 2008–2011. Because the primary objective of this study is to compare sex segregation by field of study across countries, not across time, this study considers a rather narrow range of years. Table 1 reports summary statistics for the sample.

OPEN IN VIEWER

Table 1.

Descriptive statistics.

Variable	Average	SD.	Min.	Max.
Index of Association	6.07	1.47	2.91	9.99
Real GDP per capita (2015 US dollars)	30,720	22,880	977	104,578
Females to males in primary enrollment	99.09	1.25	94.43	101.87
Human Capital Index	3.02	0.29	2.29	3.62
Structural divers. of tertiary system	20.31	15.00	0.05	64.27
Size of tertiary system	33.48	7.67	12.32	48.88
Female participation in tertiary system	59.70	5.96	45.73	71.53
Government spending as a % of GDP	19.44	5.75	6.93	29.48
Female employment rate	48.10	7.79	21.70	68.90
Unemployment rate	8.12	3.83	3.00	21.80
Women’s economic rights	1.99	0.74	1	3
Women’s political rights	2.27	0.48	1	3
Democracy	9.37	1.00	4	10
OECD	0.68	0.47	0	1
Islam	0.05	0.21	0	1
Percent Muslim	5.58	19.48	0	99.80
Percent Other Christians	19.85	26.79	0	87.00
Percent Catholic	38.59	35.97	0	94.00
Percent Orthodox	5.90	16.91	0	81.90
Percent Hindu	0.15	0.47	0	2.10
Percent Buddhist	4.31	14.89	0	66.80
Gender Inequality Index	0.19	0.12	0.02	0.51
Univ Education More Important for a Boy	15.96	10.02	2.60	41.00
Men Have More Right to a Job	22.01	13.46	2.00	59.40

GDP: gross domestic product; OECD: Organisation for Economic Co-operation and Development; SD: standard deviation.

The number of observations for each variable is 127 except in the case of the Democracy variable where the number of observations is 123 and Univ Education More Important for a Boy and Men Have More Right to a Job variables where the number of observations is 100.

The dependent variable A_it is the Index of Association and is calculated for every country (i) that reported male and female graduates by field of study for the 2008–2011 time period (t). Equation (1) also includes many control variables that influence sex segregation by field of study. Because macro-level data often display a degree of cyclical time trends, it is likely that today’s sex segregation by field is highly dependent on sex segregation levels in the past. Indeed, many previous studies highlight the stability of sex segregation by field in recent years within industrialized countries (Barone, 2011; Bradley, 2000). Excluding the lag of the index might lead to omitted variable bias, and thus, the 1-year lag of the Index of Association (A_it−1) is an important variable to include in the model. As other studies document, there is a strong link between a country’s economic development and field choices at the tertiary level. Real GDP per capita is commonly used to measure a country’s overall living standards and level of economic development. However, because many countries have relatively low levels of GDP per capita while a few have very high values, values of real GDP per capita across a diverse set of countries typically have skewed distributions. To correct for this skewness, the empirical analysis uses the natural log of real GDP per capita.

The model also controls for general educational features of a country that strongly correlate with a country’s level of economic development and living standards (Barro, 1991; Barro and Sala-i-Martin, 1995; Kabeer and Natali, 2013; Klasen and Lamanna, 2009). The first variable measures the female-to-male ratio of students enrolled in primary schools (F/MRatioPrimaryEnroll) and the second is an index of human capital, based on years of schooling and returns to education (HumanCapital). These variables are included in the model to control for these postindustrial features.

The work of Charles and Bradley (2002, 2009) highlighted that the characteristics of a country’s educational system at the tertiary level in particular can also influence sex segregation by field. Thus, this study controls for the structural diversification of a country’s tertiary education system (StructDivers), the overall size of the tertiary system (TertiarySize), and women’s participation in higher education (FemalePart). OECD is a dummy variable that includes all OECD member countries to capture any fixed effects associated with industrialized countries.

To this basic framework, I add three variables that capture features of a country’s labor market to test the hypotheses outlined above: the female employment rate, the national unemployment rate, and a measure for women’s economic rights. The CIRI Human Rights database ranks countries from 0 to 3 on their economic protections for women. Women’s economic rights include a number of internationally recognized rights, such as equal pay for equal work, the right to gainful employment without needing to obtain permission from her husband or a male relative, the freedom to work at night, and job security. A score of 0 indicates that there are no economic rights for women in law and that systematic discrimination based on sex might be built into law. A score of 3 indicates that all or nearly all of women’s economic rights are guaranteed by law and that the government fully and vigorously enforces these laws in practice.

As a robustness check, measurements of a country’s political system as well as the religious affiliation of the country are also included in the model. Each variable is measured in two different ways. The Center for Systemic Peace classifies countries by year according to their political regime type. Purely authoritarian governments receive a score of −10; pure democracies receive a +10. This variable is labeled Democracy in the tables that follow. The CIRI Human Rights database also ranks countries from 0 to 3 on their political protections for women. Women’s political rights include the right to vote, the right to run for political office, and the right to hold elected and appointed government positions, among others. The score also ranges from 0 to 3, and these scores carry similar definitions as those for women’s economic rights.

There are two ways the model controls for religion. Religious affiliation was measured by the percentage of the population practicing various religious beliefs and was collected from the CIA World Factbook. The dummy variable Islam was also created to control for religion. A country is coded as Islam if this religion has the most documented followers compared to all other religions. Government spending as a percentage of a country’s GDP is also included as an additional test of robustness.

Another source of data that quantifies the cultural features of a society is The World Values Survey. This survey asks about people’s attitudes across a range of topics, including family, environment, work, gender, religion, and politics in a diverse sample of countries. ⁹ To control for the general national sentiment regarding gender equality, the responses to two gender-related questions are used as independent variables. The variable Univ Education More Important for a Boy measures the percentage of respondents in each country who answered ‘Agree’ or ‘Strongly Agree’ to the following statement: ‘A university education is more important for a boy than for a girl’. Men Have More Right to a Job measures the percentage of respondents who agree (strongly agree was not an option) with this statement: ‘When jobs are scarce, men should have more right to a job than women’. An increase in either of these variables suggests that the country’s inhabitants are less tolerant of women’s participation in higher education and/or labor markets. Thus, one would expect that an increase in either of these variables would act to increase sex segregation by field.

Finally, some of the estimated models will also include the United Nations’ Gender Inequality Index (GII; United Nations Development Programme, 2015). The GII is calculated using three aspects of human development: reproductive health, empowerment (measured by proportion of parliamentary seats occupied by females and the proportion of adult females and males aged 25 years and older with at least some secondary education), and economic status (female and male labor force participation rates). It ranges between 0, where women and men fare equally, and 1, where one gender fares as poorly as possible in all measured dimensions. The advantage of the GII is that it simplifies the complex interrelationships of gender inequality into one number for each country. Including this variable in the models will help determine whether protecting women’s economic rights exerts a meaningful influence on sex segregation by field of study or whether what matters more for sex segregation patterns is how gender equality is manifested across multiple layers of society.

Table 1 shows descriptive statistics for these variables for all countries in the sample for the years 2008–2011. The average level of sex segregation by field of study was 6.07 (measured using the Index of Association) across all countries, with Romania reporting the highest level of sex segregation by field (9.99 in 2009) and Turkey the lowest (2.91 in 2009). The average real GDP per capita (in 2015 US dollars) was US$30,720 with an average unemployment rate of 8.12 percent. The female employment rate averaged 48.1 percent across this sample of countries, with Iceland reporting the highest female employment rate at 68.9 percent in 2008 and Turkey reporting the lowest female employment rate at 21.7 percent in 2008.

The variables capturing the characteristics of each country’s higher education system have high standard deviation values, suggesting that countries vary widely in terms of their system’s diversification, size, and female participation. Women’s dominance in the world of higher education is evident in these variables. Women made up almost 60 percent of all tertiary graduates over this period, supporting the idea of expanded post-secondary educational access for women across the globe. ¹⁰

Many of the independent variables measure the unique economic, political, and religious features of these diverse geographical settings. On average, the countries in this study tend to protect political rights for women more fully than they do economic rights. These countries earned an average ranking of 2.27 regarding their record on protecting women’s political freedoms and 1.99 for their protections of women’s economic rights. Only 2 of the 39 countries reported Islam as the religion with the most documented followers (Turkey and Kyrgyzstan). Across these countries, almost 40 percent identified themselves as Catholic; Other Christians made up the second largest group at about 20 percent. Muslim, Orthodox, and Buddhist each reported a following of between 4 and 6 % Hindu was the least followed religion at less than 1 percent.

On average, about 16 percent of respondents across all of the countries agreed with the notion that a university education is more important for boys than girls. About 22 percent of the respondents agreed that men have more right to a job than women when work opportunities are scarce although the responses varied a good deal by country. Only about 2 percent of Sweden respondents agreed with this statement compared to almost 60 percent of Turkish respondents. According to this source at least, it seems there is greater universal support for women to be educated than employed. On average, these countries report a GII of 0.19 (the higher the index, the more inequality exists in a country). According to this measure, Slovenia experiences the least gender-based disadvantages in its society (earning a score of 0.02); with a GII of 0.51, Panama experiences the most gender inequality.

Care must be taken when selecting an estimation method for this type of analysis. The ordinary least squares method is likely biased if there is a loop of causality between the dependent and independent variables. This problem is particularly relevant in this analysis since research studies establish a link between field choice in higher education and labor market conditions (Altonji et al., 2012, 2015; Blom, et al., 2015; Gemici and Wiswall, 2014; Jacobs, 1996; Roksa, 2005). For example, the unemployment rate is an independent variable in equation (1), suggesting that the health of labor markets influences the choice of college major in college students. Students are more likely to choose fields where jobs are plentiful regardless of male or female labels. Thus, according to this argument, high unemployment causes sex segregation by field to decline. Arguably, the causality might also flow in the other direction. If students do not factor in the availability of jobs into their choice of college major, their choices might contribute to an even higher unemployment rate.

The conventional response to endogeneity is to use instrumental variables. However, this technique has increasingly come under attack since many of the instruments used are weak (Bazzi and Clemens, 2013; Murray, 2006). Many cross-national studies that examine how a country’s unique features influence macro-level outcomes use fixed-effects econometric models (Berik et al., 2009; Kabeer and Natali, 2013). However, like ordinary least squares (OLS) models, fixed-effects models fail to control for endogeneity. Because the problem of endogeneity is particularly relevant in this context, an alternative modeling approach is preferred.

Manuel Arellano and Stephen Bond’s (1991) estimator that uses the GMM addresses the potential endogeneity of all the regressors and also incorporates fixed effects. This method estimates the dynamic model in first differences and instruments for current-period differences in the endogenous variables with their lagged values. Because this analysis includes some time-invariant variables, the ‘system-GMM’ estimator is preferred (Blundell and Bond, 1998). Although this method has its limitations (Bazzi and Clemens, 2013; Roodman, 2009), it is designed to handle models that face the challenge of endogeneity and unobserved heterogeneity. Because of these distinct advantages, the system-GMM is a better modeling technique than the more conventionally used fixed effects model. Furthermore, to control for autocorrelation in the dependent variable, the 1-year lag of the Index of Association is included as an independent variable.

Empirical results

The system-GMM estimations of equation (1) are presented in Tables 2 to 5. All independent variables are standardized except for categorical variables (women’s economic rights, women’s political rights, and democracy) and the Islam and OECD dummy variables. Although other scholars identified economic development and features of higher education systems to be important determinants of sex segregation by field of study, very few of these variables exert a significant influence on sex segregation by field in this analysis. By far, the most important determinant of a country’s sex segregation by field today is what segregation levels were 1 year ago. An increase of 1 standard deviation in last year’s sex segregation by field of study increases today’s sex segregation by 1.1–1.3 units depending on the model specification – this is the largest magnitude of any coefficient.

OPEN IN VIEWER

Table 2.

Impact of labor market variables on sex segregation by field of study.

	(1)	(2)	(3)	(4)	(5)
Control variables
Index of Associationt−1	1.293***	1.204***	1.218***	1.218***	1.256***
	(0.148)	(0.191)	(0.180)	(0.191)	(0.156)
Ln (RGDP per capita)	0.058	0.094	0.155	0.077	0.003
	(0.134)	(0.129)	(0.142)	(0.130)	(0.126)
Ratio of females to males in primary enrollment	0.123	0.119	0.133*	0.122	0.098
	(0.091)	(0.080)	(0.079)	(0.085)	(0.084)
Human Capital Index	−0.033	−0.071	−0.074	−0.057	−0.061
	(0.076)	(0.080)	(0.086)	(0.077)	(0.079)
Structural diversification of tertiary system	−0.141*	−0.151*	−0.170*	−0.149*	−0.113
	(0.077)	(0.087)	(0.089)	(0.085)	(0.079)
Size of tertiary system	0.043	0.057	0.084	0.049	0.027
	(0.119)	(0.109)	(0.114)	(0.112)	(0.114)
Female participation in tertiary system	−0.051	−0.099	−0.018	−0.137	−0.056
	(0.128)	(0.136)	(0.125)	(0.125)	(0.140)
OECD	−0.199	−0.246	−0.257	−0.238	−0.174
	(0.219)	(0.224)	(0.222)	(0.221)	(0.213)
Labor market variables
Female employment rate		0.225**		0.287***	0.118
		(0.104)		(0.091)	(0.095)
Unemployment ratet−2		−0.254	−0.476**		−0.251
		(0.279)	(0.231)		(0.279)
(Unemployment ratet−2)2		0.195	0.331*		0.210
		(0.225)	(0.193)		(0.222)
Women’s economic rights = 2		−0.254	−0.228	−0.250
		(0.203)	(0.202)	(0.198)
Women’s economic rights = 3		−0.659**	−0.573**	−0.665***
		(0.259)	(0.247)	(0.251)
Constant	6.207***	6.531***	6.504***	6.526***	6.190***
	(0.150)	(0.243)	(0.230)	(0.242)	(0.149)
Observations	127	127	127	127	127
Number of countries	39	39	39	39	39
Arellano–Bond statistic	0.33	0.04	0.42	−0.26	0.61
Hansen over-ID statistic	10.20	10.29	10.43	10.90	9.20

RGDP: real gross domestic product; OECD: Organisation for Economic Co-operation and Development; system-GMM: system generalized method of moments.

The dependent variable is the Index of Association. Estimation method is system-GMM. Instruments for first differences and levels equation: Ln (RGDP per capita), ratio of females to males in primary education, Human Capital Index, structural diversification of tertiary system, size of tertiary system, female participation in tertiary system, female employment rate, unemployment rate_t−2, (unemployment rate_t−2)², women’s economic rights, and OECD. GMM type: Index of Association_t−1. Robust standard errors in parenthesis. All independent variables have been standardized except for OECD and women’s economic rights.

***

,**, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

OPEN IN VIEWER

Table 3.

Robustness checks – change in control variables.

	(6)	(7)	(8)
Control variables
Index of Associationt−1	1.280***	1.226***	1.212***
	(0.158)	(0.169)	(0.194)
Ln (RGDP per capita)	0.041	0.055	0.098
	(0.121)	(0.116)	(0.136)
Ratio of females to males in primary enrollment		0.112	0.096
		(0.073)	(0.092)
Human Capital Index		−0.080	−0.064
		(0.069)	(0.086)
Structural diversification of tertiary system	−0.128		−0.123
	0.078		(0.083)
Size of tertiary system	0.034		0.038
	0.109		(0.099)
Female participation in tertiary system	−0.109		−0.133
	(0.129)		(0.156)
OECD	−0.101	−0.077	−0.228
	(0.222)	(0.156)	(0.221)
Government spending as percentage share of GDP			0.082
			(0.094)
Labor market variables
Female employment rate	0.225**	0.241**	0.250**
	(0.100)	(0.106)	(0.124)
Unemployment ratet−2	−0.279	−0.274	−0.246
	(0.294)	(0.272)	(0.288)
(Unemployment ratet−2)2	0.227	0.204	0.194
	(0.239)	(0.227)	(0.230)
Women’s economic rights = 2	−0.281	−0.281	−0.261
	(0.202)	(0.213)	(0.207)
Women’s economic rights = 3	−0.657**	−0.641**	−0.657***
	(0.257)	(0.254)	(0.255)
Constant	6.445***	6.424	6.522***
	(0.233)	(0.216)	(0.235)
Observations	127	127	127
Number of countries	39	39	39
Arellano–Bond statistic	0.05	0.23	0.05
Hansen over-ID statistic	10.98	9.35	9.88

RGDP: real gross domestic product; OECD: Organisation for Economic Co-operation and Development; system-GMM: system generalized method of moments.

The dependent variable is the Index of Association. Estimation method is system-GMM. Instruments for first differences and levels equation: Ln (RGDP per capita), ratio of females to males in primary education, Human Capital Index, structural diversification of tertiary system, size of tertiary system, female participation in tertiary system, government spending as percentage of GDP, female employment rate, unemployment rate_t−2, (unemployment rate_t−2)², women’s economic rights, and OECD. GMM type: Index of Association_t−1. Robust standard errors in parenthesis. All independent variables have been standardized except for OECD and women’s economic rights.

***

,**, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

OPEN IN VIEWER

Table 4.

Robustness checks – controlling for other features of gender equality.

	(9)	(10)	(11)	(12)
Labor market variables
Female employment rate	0.250***	0.223**	0.116	0.237**
	(0.092)	(0.104)	(0.095)	(0.103)
Unemployment ratet−2	−0.247	−0.257	−0.253	−0.246
	(0.282)	(0.278)	(0.280)	(0.273)
(Unemployment ratet−2)2	0.183	0.196	0.212	0.187
	(0.234)	(0.224)	(0.223)	(0.221)
Women’s economic rights = 2	−0.285	−0.251		−0.253
	(0.211)	(0.204)		(0.202)
Women’s economic rights = 3	−0.747***	−0.656**		−0.680**
	(0.273)	(0.259)		(0.268)
Other gender equality variables
Percent Muslim	−0.016
	(0.105)
Percent Catholic	0.016
	(0.125)
Percent Other Christians	0.042
	(0.135)
Percent Orthodox	0.036
	(0.141)
Percent Hindu	−0.008
	(0.048)
Percent Buddhist	−0.140
	(0.116)
Women’s political rights = 3		0.024
		(0.131)
Gender Inequality Index			−0.018	0.067
			(0.092)	(0.115)
Observations	127	127	127	127
Number of countries	39	39	39	39
Arellano–Bond statistic	0.09	0.05	0.63	0.00
Hansen over-ID statistic	10.63	10.20	9.12	10.30

RGDP: real gross domestic product; OECD: Organisation for Economic Co-operation and Development; system-GMM: system generalized method of moments.

The dependent variable is the Index of Association. Estimation method is system-GMM. Instruments for first differences and levels equation: Ln (RGDP per capita), ratio of females to males in primary education, Human Capital Index, structural diversification of tertiary system, size of tertiary system, female participation in tertiary system, female employment rate, unemployment rate_t−2, (unemployment rate_t−2)², women’s economic rights, women’s political rights, percent Muslim, percent Catholic, percent Other Christians, percent Orthodox, percent Hindu, percent Buddhist, Gender Inequality Index, and OECD. GMM type: Index of Association_t−1. Robust standard errors in parenthesis. All independent variables have been standardized except for OECD, women’s economic rights, and women’s political rights.

***

,**, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

OPEN IN VIEWER

Table 5.

Robustness checks – controlling for attitudes regarding gender equality.

	(13)	(14)	(15)
Labor market variables
Female employment rate	0.172**	0.180**	0.188**
	(0.84)	(0.083)	(0.087)
Unemployment ratet−2	−0.303	−0.289	−0.309
	(0.263)	(0.238)	(0.261)
(Unemployment ratet−2)2	0.259	0.258	0.268
	(0.222)	(0.196)	(0.215)
Women’s economic rights = 2	−0.172	−0.183	−0.165
	(0.219)	(0.216)	(0.219)
Women’s economic rights = 3	−0.260	−0.320	−0.264
	(0.235)	(0.252)	(0.251)
Variables controlling for attitudes regarding gender equality
Univ Education More Important for a Boy	0.001
	(0.087)
Men Have More Right to a Job		−0.089
		(0.064)
Observations	100	100	100
Number of countries	30	30	30
Arellano–Bond statistic	2.12	1.97	2.03
Hansen over-ID statistic	10.74	10.76	10.75

RGDP: real gross domestic product; OECD: Organisation for Economic Co-operation and Development; system-GMM: system generalized method of moments.

The dependent variable is the Index of Association. Estimation method is system-GMM. Instruments for first differences and levels equation: Ln (RGDP per capita), ratio of females to males in primary education, Human Capital Index, structural diversification of tertiary system, size of tertiary system, female participation in tertiary system, female employment rate, unemployment rate_t−2, (unemployment rate_t−2)², women’s economic rights, Univ Education More Important for a Boy, Men Have More Right to a Job, and OECD. GMM type: Index of Association_t−1. Robust standard errors in parenthesis. All independent variables have been standardized except for OECD and women’s economic rights.

***

,**, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

These results suggest that the historical legacy of previous sex segregation exerts a powerful influence on today’s field choices; in comparison, cross-national variations in economic growth, human capital, or characteristics of tertiary education systems exert little influence. It is worth noting that if the 1-year lag of the Index of Association is not included in the model, some of the other control variables become statistically significant. For example, across a variety of model specifications, the natural log of real GDP per capita is positive and statistically significant, and the structural diversification of a country’s tertiary system is negative and statistically significant, as is the OECD dummy variable (these results are not shown).

The influence of the labor market variables on sex segregation by field of study are presented in columns (2)–(5) of Table 2. The results show that an increase in the female employment rate by 1 standard deviation 2 years ago increases current levels of sex segregation by field of study by 0.225 units. This result also holds in model specification (4) when the variables measuring unemployment are not included in the model. However, the female employment rate loses its statistical significance when women’s economic rights are excluded (see column (5)), suggesting that the influence of female employment on sex segregation patterns is conditional on government protections for women’s economic rights.

Because the unemployment rate might have a negative but diminishing effect on sex segregation by field, the 2-year lag of the unemployment rate is included in the model as a quadratic term. Contrary to our expectations, the unemployment rate has no statistically significant influence on sex segregation by field of study; however, this nonsignificance only holds when the female employment rate is also included in the model. See, for example, column (3) of Table 2 where the female employment rate is not included in the estimation. All else equal, a 1 standard deviation increase in the national unemployment rate 2 years ago lowers the current year’s level of sex segregation by field for tertiary graduates by 0.476 units. The positive coefficient on the quadratic term indicates that the negative effect of unemployment does in fact diminish as the unemployment rate rises.

The negative impact of poor labor market opportunities on sex segregation by field of study is very consistent when the rate of female employment is excluded from the model. For every model specification presented in Tables 2 to 5, if the female employment rate is excluded, the 2-year lag of the unemployment rate is negative (and of a magnitude around −0.45) with a p-value of 0.05 or lower (these results are not shown). ¹¹ The lack of significance when the female employment rate is included in the models is likely due to the high degree of collinearity between these two variables (the correlation coefficient between these two variables is −0.56 and is significant at the 1% level). This strong negative relationship between these two labor market features seems quite reasonable. When employment prospects are good (indicating a low unemployment rate), the female employment rate likely increases: stay-at-home parents may seek paid employment and recent retirees might renter the labor market, thus increasing the female employment rate. The reverse scenario also seems realistic – when labor market opportunities are scarce, women may leave paid employment (voluntarily or not) and engage in unpaid work in the home thereby reducing the female employment rate. The evidence demonstrates that both variables have an important influence on sex segregation patterns. However, since the female employment rate retains its significance when the unemployment variables are also included in the model, female employment rates are better predictors of cross-national differences in sex segregation by field of study when other relevant factors are held constant.

The estimation results in Table 2 also support that stronger protections for women’s economic rights act to reduce sex segregation by field of study. Countries that were ranked the highest (a score of 3) in their protections and enforcement of economic protections for women had levels of sex segregation by field of study 0.573–0.665 units lower compared to countries that scored a ‘1’ in their economic protections for women (the excluded group). There was no difference between countries that ranked a ‘2’ for their record of economic protections for women and those in the ‘1’ group in terms of their influence on sex segregation by field of study. Of the three labor market variables considered in the model, the strong protection and enforcement of women’s economic rights had the largest and the most consistent impact on sex segregation by field of study.

In general, the regression results in Table 2 support the predictions stated in Hypotheses 1–3. As previous scholars note, countries with a greater presence of women in labor markets often exhibit high degrees of occupational segregation. ¹² This gender division in labor markets likely influences students’ choice of field, encouraging students to select careers and fields that are heavily populated with their own gender. Thus, all else equal, countries with higher rates of female employment in the recent past are also associated with higher levels of sex segregation by field of study. The general condition of labor markets also seems to make a difference. When female employment rates are not included in the estimations, the higher the national unemployment rate 2 years ago, the lower the sex segregation by field is today. Because field choice is closely linked to one’s future profession, this seems like a rational outcome. When labor conditions are poor (i.e. the unemployment rate is high), students likely choose fields based more on future employment prospects and less on their (gendered) preferences.

Finally, protecting women’s economic rights reduces sex segregation by field of study for this sample of countries. In countries with laws in place protecting women from labor market discrimination, men and women become more equally distributed across fields in higher education. This suggests that women have more incentive to become qualified for higher paying employment opportunities by studying fields that were traditionally dominated by men. Together, these results demonstrate the important links between labor markets and field choice in higher education. All three labor market variables (the rate of women’s employment in formal labor markets, the unemployment rate, and the level of protections for women pursuing work outside the home) have significant and meaningful effects on sex segregation by field of study.

Two tests were conducted to determine the validity of the estimation results. To test for serial correlation, the Arellano–Bond statistic is reported for every specification. For all estimations presented in Tables 2 to 4, this statistic fails to reject the null hypothesis of zero second-order autocorrelation at the 0.05 level. ¹³ Because system-GMM estimations can produce a large number of instrumental variables, I also conduct a test for overidentification and report the Hansen J statistic. In each specification, the statistic fails to reject the null hypothesis of no overidentification problem. Thus, the models are not weakened by problems of serial correlation or excessive instruments and overidentification.

I re-estimated the model using a variety of specifications to examine whether the results are robust to the choice of controls and the inclusion of other independent variables. The first round of these robustness checks is given in Table 3. The results in column (6) of Table 3 exclude two of the variables that are helping to control for the level of economic development – the ratio of females to males in primary enrollment and the Human Capital Index. Dropping these variables does very little to influence the estimation results for the labor market variables. In another specification (not shown), I also dropped the natural log of real GDP per capita along with the two other economic variables; the results from this specification are very similar to the results presented in column (6). In model specification (7), I dropped the variables that are controlling for the characteristics of the tertiary education systems of each country (the structural diversification of the tertiary system, the size of the tertiary system, and the level of female participation in the tertiary system). Again, excluding these control variables has very little impact on the estimation results of the labor market variables. ¹⁴ Column (8) includes a control for government spending as a percentage of the country’s GDP. The inclusion of this variable does little to change the estimated results.

Table 4 shows additional model specifications to determine the robustness of the results. These estimations include the full set of control variables from Table 2, but the presentation of Table 4 only reports the results for the labor market variables and additional variables that measure various features of gender equality. It is quite possible that important variables are missing from the estimated models in Tables 2 and 3 and these missing variables could be correlated with the labor market variables resulting in biased coefficients. The results in Table 4 column (9) include the religious affiliation variables. These variables measure the percent of the population in each country following various religions: Muslim, Catholic, Other Christians, Orthodox, Hindu, and Buddhist. The excluded religious group was ‘Other’. These variables had no unique influence on sex segregation by field of study and the labor market variables were largely unchanged (although the coefficient on women’s economic rights increased in absolute value). I also estimated the model with the inclusion of a dummy variable called Islam – a country is coded as Islam if this religion has the most documented followers compared to all other religions. The results from this estimation were very similar to the results in model specification (9).

Model specification (10) includes a control for women’s political rights. Countries were grouped into two categories: those earning a score of ‘1’ or ‘2’ for their protections of women’s political rights and those earning a ‘3’. ¹⁵ All else equal, countries with the highest levels of political protections for women (a score of ‘3’) exhibit similar levels of sex segregation by field as those countries that earned a ‘1’ or a ‘2’ for this variable. Replacing the women’s political rights variable with the Democracy variable (measured both as a continuous variable and as a categorical variable) added little to the model (results not shown). The coefficients and significance levels for the labor market variables were substantially unchanged from column (10). The Democracy variables were not statistically significant.

Model specification (11) includes the GII but does not control for women’s economic rights. Because equality in labor markets is one dimension of the GII, including both the GII and women’s economic rights might weaken the individual impact of each variable on sex segregation by field. As column (11) demonstrates, the GII does not exert a meaningful influence on sex segregation by field. The results are very similar to model specification (4) in Table 2 which implies that the GII adds little explanatory power to the model. Model specification (12) includes both the GII and women’s economic rights with women’s economic rights retaining its statistically significant negative influence on sex segregation patterns across countries. This suggests that gender equality in labor markets reduces sex segregation patterns in higher education more so than other features of gender equality (such as health outcomes and women’s political leadership) that are included in the GII.

Table 5 shows the estimated results for the models that include a control variable for a country’s general attitudes regarding gender equality. Model specification (13) includes the percentage of respondents who agreed that a university education is more appropriate for a boy; specification (14) includes the percentage of respondents who believe that during times of poor labor market conditions, men have more right to a job than women. As the results show, when these variables are included in the model, women’s economic rights lose its statistical significance. However, because not all countries in the sample were included in the World Values Survey, nine countries are excluded from the estimations. ¹⁶ Model specification (15) shows that when the sample is reduced to match the samples in specifications (13) and (14), women’s economic rights are no longer statistically significant, even when these ‘gender attitude’ variables are not included in the model. Thus, these results seem to be more of a function of the reduced sample than any unique impact originating from these gender attitude variables. Most importantly, all three of these specifications reject the null hypothesis of zero second-order autocorrelation at the 0.05 level indicating that the moment conditions used by the system-GMM estimation method are not valid. ¹⁷

Thus, the empirical results measuring the impact of the labor market variables on sex segregation by field are largely robust to changes in control variables as well as the addition of other potentially important variables. Removing controls for the level of a country’s economic development as well as the features of its tertiary system did little to change how labor markets affect sex segregation by field of study. Including additional variables that measure the degree of democratic practices in a country, measures of its people’s religious beliefs and attitudes about gender, and the general level of gender inequality did little to alter the effect of the female employment rate and the level of women’s economic rights on sex segregation by field. In addition, the unemployment rate also exerts an important influence on sex segregation patterns, although this effect is diminished when the female employment rate is also included. The empirical findings largely confirm Hypotheses 1 and 3 and provide modest support for Hypothesis 2. In sum, this analysis presents important new evidence that labor market conditions exert a meaningful influence on sex segregation by field of study.

Discussion

In the last few decades, the world witnessed substantial progress in women’s equality in a variety of venues – labor markets, politics, and, most dramatically, education. Despite women’s success in higher education in general, a great deal of research documents uneven progress regarding women’s share of degrees earned across fields. Women continue to be underrepresented in the higher paying, higher status fields of study in higher education. As other scholars highlight, the lack of women in these fields is of crucial importance – it is one of the key reasons the gender wage gap persists well into the 21st century. Understanding the macro-level factors that impact sex segregation by field can help countries identify policies that disrupt these gendered patterns of field choice.

Other scholars found an important relationship between both the level of economic development as well as the characteristics of a country’s educational system and the level of sex segregation by field of study. This analysis failed to find a significant connection between these features of a country and its level of sex segregation by field. Crucially, this is one of the few studies that controls for the previous year’s sex segregation of tertiary graduates. This study found that past sex segregation has much more explanatory power than the level of economic development or characteristics of tertiary education systems on current sex segregation by field. The results suggest that the norms and attitudes that create the cultural foundation on which sex segregation by field rests are quite powerful and seem resilient to fluctuations that take place in the larger economic or educational spheres.

This article also established new insights that highlight how a country’s labor market features can help reduce sex segregation by field. This analysis found that, all else equal, countries with laws and institutions in place that protect the economic rights of women also have lower levels of sex segregation by field of study. When all labor market opportunities are open to women, and they can be reasonably assured that they have the equal right to a job as a man, sex segregation by field declines. Governments, then, play an important role. By guaranteeing equal rights for women in labor markets, they can contribute to a lower level of sex segregation by field.

Other scholars argue that pressure to conform to gender stereotypes combined with norms of self-expression in post-modern economies intensify sex segregation levels (Charles and Bradley, 2009). This study contributes a deeper understanding of this process. Although cultures that value self-expression experience higher levels of sex segregation, both in field choice and in occupation, these societies are also more likely to establish rules and laws that support people’s ability to make these economic choices free from discriminatory influences. As this study demonstrates, this protection reduces sex segregation levels in field choice. The net effect of these competing influences warrants further scholarly attention.

The ability to find employment after graduation, though a real-world concern for the vast majority of tertiary graduates, was not considered in previous empirical studies seeking to understand the important determinants of sex segregation by field of study across different countries. This cross-national study included the 2-year lag of the country’s national unemployment rate measured as a quadratic term. Although not as consistent as the impact of protecting women’s economic rights, a higher national unemployment rate also acted to reduce sex segregation by field of study. However, this negative impact on sex segregation by field was only evident when the female employment rate was excluded from the model. Nonetheless, the influence of the unemployment rate on sex segregation by field seems to confirm the findings of Blom et al. (2015) and also adds more clarity. These researchers found that during times of high unemployment in the United States, both men and women are more likely to major in male-dominated fields so that the overall impact on sex segregation by field is unclear. This study found that on net, sex segregation levels decline following periods of poor labor market conditions. Thus, it seems that women’s field choices change more dramatically than men’s in response to high unemployment rates, leading to more equal representation of men and women across fields of study.

Although this study makes significant contributions to our understanding of why sex segregation by field of study varies across countries, the analysis is limited in a number of ways. First, due to the lack of available data, the models cannot control for two important features of countries that may have an important influence on sex segregation by field of study: occupational segregation and education tracking. How sex segregation by field of study contributes to occupational segregation in labor markets (and vice versa) is largely underexplored in this line of research. In addition, some educational systems group students according to ability and ‘track’ them into different educational programs (e.g. Germany and Austria have a high degree of tracking at the post-secondary education level). Thus, in countries that have tracking systems in place, student’s field choices might be constrained. This lack of freedom in choosing a college major may have important implications for sex segregation by field of study. Future research should examine how occupational segregation and education tracking influence sex segregation by field. Second, the Index of Association (the measurement of sex segregation by field for this study) will result in different values if different field classifications are used. Scholars in this area should empirically test the results of this article to see whether they are robust to different ways of measuring sex segregation by field. Finally, although this study included as many countries as possible in its analysis, some countries were excluded due to lack of available data. Thus, another area of future research could test the reliability of this article’s conclusions against a different sample of countries.

Concluding remarks

The current project sought to better understand how labor market characteristics influence sex segregation by field of study across a diverse sample of countries. This study was the first to use the system-GMM estimation method to analyze sex segregation by field. Given the possible endogeneity of the independent variables and the large number of cross sections relative to time periods, this estimation technique is an ideal method to understand this relationship. The analysis found that, all else equal, countries that protect the economic rights of women have lower levels of sex segregation by field. Although the evidence is less robust, the analysis also demonstrated that countries with higher rates of national unemployment have lower levels of sex segregation by field of study.

These results are promising. Although countries usually do their best to avoid periods of high unemployment, the business cycle seems to be a permanent feature of the global economy. This article supports the idea that sex segregation by field declines during these recessionary phases of the business cycle. Because the academic backgrounds of men and women explain a substantial portion of the wage gap, an unintended consequence of economic recessions is that men and women become more evenly distributed across academic fields, potentially leading to reduced wage gaps in the future. Determining whether gender pay gaps narrow after periods of high unemployment is an important area of focus for future research projects.

There is a second hopeful feature of the research findings. Countries that provide economic protections for women in the form of equal pay for equal work, the right to gainful employment without needing to obtain permission from her husband or a male relative, the freedom to work at night, and job security also experience lower levels of sex segregation by field of study. In these environments, men and women are more evenly distributed across academic fields. These findings support predictions from human capital theory. If women can be reasonably assured that there are employment laws in place protecting them from discriminatory practices, more women will choose fields of academic study that were historically dominated by male students. Governments can play an important role in encouraging equitable employment practices that do not favor men over women thereby creating incentives for more women to pursue a field that was historically categorized as male.

Across this diverse group of countries, when women’s economic rights are protected, sex segregation by field declines. This is a promising finding. Although the history of sex segregation by field casts a shadow on today’s higher education landscape, part of the answer seems to be simple: provide equal rights for women in employment. Treating women as rightful members of the labor market is a meaningful step to reduce sex segregation by field and enhance women’s equality in other aspects of life.