A Cultural Psychological Model of Cross-National Variation in Gender Gaps in STEM Participation

Contextual Variation in the Gender Gap in Math/Science Performance

Researchers have examined cross-national variation in the gender gap in STEM performance using data from the Programme for International Student Assessment (PISA) or the Trends in International Mathematics and Science Study (TIMSS, see Table 1). ³ Several studies documented that the gender gap in math and science performance was smaller in nations that scored higher on certain indicators of gender equality or economic development, in accordance with the freedom hypothesis (Baker & Jones, 1993; Else-Quest et al., 2010; Fryer & Levitt, 2010; Guiso et al., 2008; Hyde & Mertz, 2009; Nosek et al., 2009; Penner, 2008; Reilly, 2012; Riegle-Crumb, 2005). However, the majority of these studies also reported findings contradicting the freedom hypothesis. For instance, Else-Quest and colleagues (2010) found that the gender gap in math performance was smaller in nations with greater gender equality in share of research positions, participation in economic activities, and share of parliamentary seats, as well as higher global gender equality indicator scores, based on PISA 2003 scores. However, these indicators were unrelated to the gender gap based on TIMSS 2003 scores (Else-Quest et al., 2010). Fryer and Levitt (2010) found that the relationship between global gender equality measures and the gender gap in math performance that emerged with the PISA 2003 data (Else-Quest et al., 2010) disappeared when a group of majority-Muslim countries, which have low levels of equality but also small gender gaps, were added into the data set. Stoet and Geary (2015) concluded that the relationship between gender equality and the gender gap in math performance was not robust, as it did not emerge in the four exams that they examined other than the PISA 2003. They found that the gender gap in math performance increased with greater gender inequality based on some of the domain-specific measures (e.g., women’s share of research positions), but was unrelated to it based on several others (e.g., percentage of women in the parliament) across four data sets. Overall, there seems to be evidence both supporting and contradicting the freedom hypothesis for math/science performance, depending on the specific exam scores, countries, and indices used.

OPEN IN VIEWER

Table 1.

Cross-National Variation in the Gender Gap in Math and Science Performance.

Authors	Sample (countries included)	Findings that support the freedom hypothesis	Findings that fail to support the freedom hypothesis
Baker and Jones (1993)	N = 19; SIMS 1982	Gender gap in math performance was smaller in nations with higher: • percentage of women in higher education, r(10) = −.54 • percentage of women in the labor force, r(19) = −.55 • percentage of women in industrial work, r(19) = −.59 • percentage of women in service jobs, r(19) = −.40	—
Riegle-Crumb (2005)	N = 47; TIMSS 1995	Gender gaps in math and science performance (respectively) were smaller in nations with higher: • women’s share of government positions (bs = 2.86 and 3.78)	Gender gaps in math and science performance (respectively) were unrelated to: • gender equality in the home and the labor force (bs = 0.46 and −5.45) • level of economic development (bs = −2.27 and 0.04)
Penner (2008)	N = 22; TIMSS 1995	Gender gap in math performance was smaller in nations with higher a : • gender equality in labor force position (bs between −0.26 and 0.28) at all levels of the distribution of scores • overall status of women at the top of the distribution (b = 0.14) • gender equality in education at the bottom of the distribution (bs between −0.18 and −0.11)	Gender gap in math performance was larger in nations with higher: • representation of women in the labor force (bs between −0.21 and 0.19)
Guiso et al. (2008)	N = 40; PISA 2003	Decrease in the gender gap in math performance was larger in nations with higher: • GGI (b = 26.9) • female economic activity rate (b = 0.14) • women’s political participation (b = 10.05) b	—
Hyde and Mertz (2009)	N = 41; PISA 2003	The gender gap above the 95th percentile was smaller in nations with higher: • GGI (r = −.34) Percentage of girls on national IMO teams was larger in nations with higher: • GGI among the top 30 teams, r (30)= .44	—
Kane and Mertz (2012)	N = 46; TIMSS 2003, TIMSS 2007, PISA 2009	Percentage of girls in IMO teams was smaller in nations with higher: • GGI (rs range between .041 and .047 among the top 40, 50, and 60 countries)	The gender gap in math performance was unrelated to: • the GGI in 2003 among eighth graders (r = −.03) and fourth graders (r = .14), and in 2009 (r = .08) The gender gap in math performance was larger in nations with higher: • GGI in 2007 among eighth graders(r = .29) and fourth graders (r = .58)
Nosek et al. (2009) c	N = 34; TIMSS 1999 and 2003	Gender gap in math performance was smaller in nations with higher: • GGI in the 1999 data set	Gender gaps in math and science performance were unrelated to: • GDP per capita in both data sets The gender gap in science performance was: • unrelated to the GGI in 2003 • larger in nations with higher GDP ranking in both data sets
Else-Quest et al. (2010)	N = 46 and N = 41; TIMSS 2003 and PISA 2003	Gender gap in math performance was smaller in nations with higher (in the PISA data set): • GEM (β = −.33; R2 = .11) • GEQ (β = −.38; R2 = .14) • SIGE (β = −.31; R2 = .09) • GGI (β = −.38; R2 = .14) • women’s share of research positions (β = −.32; R2 = .10) • male/female economic activity rate ratio (β = −.37; R2 = .13) • women’s share of parliamentary seats (β = −.30; R2 = .09)	Gender gap in math performance was unrelated to (in the PISA data set): • male/female primary enrollment ratio (β = .08; R2 = .01) • male/female secondary enrollment ratio (β = −.10; R2 = .01) • male/female tertiary enrollment ratio (β = −.29; R2 = .08) • women’s share of higher labor market positions (β = −.29; R2 = .08)
		Gender gap in math performance was smaller in nations with high (in the TIMMS data set):• male/female primary enrollment ratio (β = −.36; R2 = .13)• male/female secondary enrollment ratio (β = −.41; R2 = .17)• male/female tertiary enrollment ratio (β = −.37; R2 = .14)	Gender gap in math performance was unrelated to (in the TIMMS data set):• GEM (β = .20; R2 = .04)• GEQ (β = −.19; R2 = .04)• SIGE (β = −.06; R2 = .00)• GGI (β = .01; R2 = .00)• male/female economic activity ratio (β = .07; R2 = .00)• women’s share of higher labor market positions (β = −.29; R2 = .08)
Fryer and Levitt (2010)	N = 41 and N = 46; TIMSS 2003 and PISA 2003	Decrease in the gender gap in math performance was larger in nations with high:• GGI in the PISA data set (βs = .60, .82, and .80) d	Decrease in the gender gap in math performance was unrelated to• the GGI in the TIMSS data set (βs = .02, .01, and .07)
Reilly (2012)		Gender gap in math performance was smaller in nations with higher:• women in research, r(65) = −.38• relative status of women, r(65) = −.14	Gender gap in math performance was unrelated to:• the GGI• Gender gap in math performance was smaller in nations with higher:• power distance, r(65) = −.28• Gender gap in math performance was larger in nations with higher:• economic development, r(65) = .31
Stoet and Geary (2015)	N = 42, N = 41, N = 57, and N = 74; PISA 2000, 2003, 2006, and 2009	Gender gap in math performance was smaller in nations with higher:• GGI in the 2003 data set, r(35) = −.41• GEM in the 2003 data set, r(35) = −.41• women’s share of research positions in the 2000, r(31) = −.68, 2003, r(29) = −.35, and 2006 data sets, r(40) = −.29 (marginally)• female economic activity in the 2000, r(39) = .35, and 2003 data sets,• r(35) = .38 e	Gender gap in math performance was unrelated to:• the GGI in the 2000, 2006, and 2009 data sets f • the GEM in the 2000, r(30) = .005, and 2006 data sets, r(41) = −.03• the GEM in the 2009 data set, r(60) = .26• women’s share of research positions in the 2009 data set, r(46) = −.16• female economic activity in the 2006, r(49) = −.05, and 2009 data sets, r(62) = −.002• percentage of women in the parliament in any data set, rs between r(37) = −.23 and r(64) = .18
Stoet and Geary (2018)	N = 67; PISA 2015	—	• Gender gap in science performance was unrelated to the GGI, r(62) = .23• Gender gap in strength in science was larger in nations with higher GGI, r(62) = .42

Note. Regions of the countries included in the data sets and their range of scores on GDP per capita and the GGI are listed below. Scores on these indices are based on data for the year when each exam was conducted, or the closest year if scores for the exact year are unavailable.

TIMSS 1995: 26 European, 8 Eurasian/Asian, 3 Central/South American, 2 North American, 2 Pacific, 1 Middle Eastern/North African.

GDP per capita ranges between 8,443 for Colombia and 81,017 for Kuwait; GGI ranges between 0.580 for Iran and 0.813 for Sweden.

TIMSS 1999: 16 European, 10 Eurasian/Asian, 1 Central/South American, 2 North American, 2 Pacific, 6 Middle Eastern/North African, 1 African.

GDP per capita ranges between 5,806 for Indonesia and 51,706 for Singapore; GGI ranges between 0.585 for Turkey and 0.796 for Finland.

TIMSS 2003: 18 European, 10 Eurasian/Asian, 2 Central/South American, 2 North American, 2 Pacific, 10 Middle Eastern/North African, 3 African.

GDP per capita ranges between 2,539 for Ghana and 61,974 for Singapore; GGI ranges between 0.580 for Iran and 0.813 for Sweden.

PISA 2000: 27 European, 6 Eurasian/Asian, 5 Central/South American, 2 North American, 2 Pacific, 1 Middle Eastern/North African.

GDP per capita ranges between 5,669 for Albania and 81,690 for Luxembourg; GGI ranges between 0.616 for Korea and 0.813 for Sweden.

PISA 2003: 24 European, 7 Eurasian/Asian, 3 Central/South American, 2 North American, 2 Pacific, 2 Middle Eastern/North African.

GDP per capita ranges between 8,803 for Tunisia and 88,610 for Luxembourg; GGI ranges between 0.585 for Turkey 0.813 for Sweden.

PISA 2006: 33 European, 9 Eurasian/Asian, 6 Central/South American, 2 North American, 2 Pacific, 5 Middle Eastern/North African.

GDP per capita ranges between 8,800 for Jordan and 109,802 for Qatar; GGI ranges between 0.585 for Turkey and 0.813 for Sweden.

PISA 2009: 36 European, 15 Eurasian/Asian, 10 Central/South American, 2 North American, 2 Pacific, 7 Middle Eastern/North African, 1 Caribbean, 1 African.

GDP per capita ranges between 6,734 for Georgia and 125,141 for Qatar; GGI ranges between 0.583 for Turkey and 0.828 for Iceland.

PISA 2015: 37 European, 13 Eurasian/Asian, 9 Central/South American, 2 North American, 2 Pacific, 8 Middle Eastern/North African, 1 Caribbean.

GDP per capita ranges between 5,555 for Vietnam and 119,749 for Qatar; GGI ranges between 0.593 for Jordan and 0.881 for Iceland.

SIMS = Second International Mathematics Survey; TIMSS = Trends in International Mathematics and Science Study; PISA = Programme for International Student Assessment; GGI = gender gap index; IMO = International Math Olympics; GDP = gross domestic product; GEM = gender empowerment measure; GEQ: gender equality index; SIGE: standardized index of gender equality.

a

Positive (negative) coefficients mean overrepresentation (underrepresentation) of females above or below each percentile. ^bThe gender gap is calculated as the difference between mean score of females and males; gap becomes smaller as this value increases from negative to zero. Relationship was also found for cultural attitudes toward women (b = 13.21). ^cStatistics for the relationships were not provided in the Supporting Information for the paper. ^dThe gender gap is calculated as the difference between weighted mean score of females and males; gender gap becomes smaller as this value increases from negative to zero. ^ePositive correlations here mean that the gender gap is smaller where more women participate in economic activity. ^fStatistics were not provided for these analyses.

Contextual Variation in the Gender Gap in Math/Science Attitudes

Studies have used a variety of variables, including math self-concept and self-efficacy (e.g., Else-Quest et al., 2010; Sikora & Pokropek, 2012), liking and enjoyment of math/science domains, and STEM-related aspirations (e.g., Charles et al., 2014; Riegle-Crumb, 2005) to examine cross-national variation in STEM-related attitudes (see Table 2). Only two studies provided support for the freedom hypothesis, where researchers found smaller gender gaps in nations with greater gender equality or economic development (Else-Quest et al., 2010; Riegle-Crumb, 2005). Several findings showed the opposite relationship. For instance, contrary to the freedom hypothesis, Else-Quest and colleagues (2010) found that the size of the gender gap in five of the seven math attitudes they examined was larger in nations with greater global gender equality. Similarly, gender gaps in math self-concept, math self-efficacy, and math anxiety were larger in more gender-equal nations (based on domain-specific such as women’s share of parliamentary seats and women’s share of research positions; Else-Quest et al., 2010). Likewise, Sikora and Pokropek (2012) reported that the gender gap in science self-concept was larger in advanced industrial countries than in developing countries. In a similar vein, larger gender differences (favoring boys) in liking of math and aspirations for math-related careers appeared in countries with high compared to medium or low Human Development Index (HDI) scores (Charles et al., 2014). Furthermore, the odds of female students reporting strong aspirations for a math-related job (relative to male students) were lower in countries with higher HDI scores (Charles, 2017). In a recent analysis, Stoet and colleagues (2016) found that gender gaps in math anxiety, as well as self-efficacy, enjoyment of, interest in, and tendency to overestimate competence in science were larger in nations with higher gender equality and societal development levels (Stoet & Geary, 2018). Overall, the majority of evidence contradicts the freedom hypothesis for math/science attitudes.

OPEN IN VIEWER

Table 2.

Cross-National Variation in the Gender Gap in Math and Science Attitudes.

Authors	Sample (countries included)	Findings that support the freedom hypothesis	Findings that fail to support the freedom hypothesis
Riegle-Crumb (2005)	N = 47; TIMSS 1995	Gender gaps in math and science attitudes (respectively) were smaller in nations with higher:• gender equality in the home and the labor force combined (bs = 0.46 and 0.91)Gender gap in science attitudes was smaller in nations with higher:• economic development (b = 0.78)	Gender gaps in math and science attitudes (respectively) were unrelated to:• women’s share of government positions (bs = −0.00 and −0.06)
Else-Quest et al. (2010)	N = 46; N = 41; PISA 2003 and TIMSS 2003	Gender gap in math attitudes was smaller in nations with higher (in the PISA data set):• GEM for anxiety (R2 = .41)• GGI for anxiety (R2 = .19)• male/female tertiary enrollment ratio for intrinsic motivation (R2 = .12)• women’s share of research positions for self-confidence (R2 = .15), extrinsic motivation (R2 = .37), intrinsic motivation (R2 = .33), self-concept (R2 = .36), and self-efficacy (R2 = .15)• women’s share of parliamentary seats for anxiety (R2 = .17)Gender gap in math attitudes was smaller in nations with higher (in the TIMMS data set):• male/female secondary enrollment ratio for valuing math (R2 = .11) and self-confidence (R2 = .16)• male/female tertiary enrollment ratio for self-confidence (R2 = .12)• women’s share of research positions for valuing math (R2 = .23)• women’s share of higher labor market positions for valuing math (R2 = .11) and self-confidence (R2 = .15)	Gender gap in math attitudes was larger in nations with higher (in the PISA data set):• GEM for extrinsic motivation (R2 = .20), intrinsic motivation (R2 = .12), self-concept (R2 = .27), and self-efficacy (R2 = .16)• GGI for self-concept (R2 = .12) and self-efficacy (R2 = .11)• women’s share of parliamentary seats for extrinsic motivation (R2 = .15), intrinsic motivation (R2 = .11), self-concept (R2 = .21), and self-efficacy (R2 = .17)• women’s share of research positions for anxiety (R2 = .23)• male/female primary enrollment ratio for extrinsic motivation (R2 = .11)Gender gap in math attitudes was larger in nations with higher (in the TIMMS data set):• GEM for valuing math (R2 = .16)• All other relationships between indicators and attitude variables were non-significant (R2s between .00 and .11)
Sikora and Pokropek (2012)	N = 50; PISA 2006	—	Gender gap in science self-concept was larger in advanced industrial countries (−0.30 of a standard deviation) compared to transforming countries (−0.11 of a standard deviation)
Charles et al. (2014)	N = 53; TIMSS 2003, 2007, and 2011	—	Gender gap in math attitudes was larger in nations with higher HDI• Boys were 36% more likely than girls to report enjoying math in high-HDI, 24% in medium-HDI, and 13% low-HDI countries• Boys were 75% more likely than girls to aspire for a math-related job in high-HDI, 62% in medium-HDI, and 49% in low-HDI countriesGender gap in math attitudes was larger in nations with higher self-expressiveness• One standard deviation increase in self-expressiveness predicted an increase in boys’-to-girls’ odds of enjoying math from 1.24 to 1.34• One standard deviation increase in self-expressiveness predicted an increase boys’-to-girls’ odds of aspiring to a math-related job from 1.64 to 1.74
Stoet et al. (2016)	N = 41; N = 68; PISA 2003 and 2012	—	Gender gap in math anxiety was positively correlated with:• the GGI, r(55) = .43, and HDI, r(59) = .48, in 2003• with HDI, r(36) = .40, in 2012Gender gap in math anxiety was unrelated to:• the GGI, r(36) = .06, in 2012
Charles (2017)	N = 32; TIMSS 2003 and 2011	—	Girls’ odds of reporting strong aspirations for math-related jobs relative to boys were lower in nations with higher HDI, r(32) = −.76
Stoet and Geary (2018)	N = 64; PISA 2015	—	Gender gap in math attitudes was larger in nations with higher GGI for:• science self-efficacy, r(61) = .60• science interest, r(50) = .41• science enjoyment, r(61) = .46

Note. Regions of the countries included in the data sets and their range of scores on GDP per capita and the GGI are listed below. Scores on these indices are based on data for the year when each exam was conducted, or the closest year if scores for the exact year are unavailable.

TIMSS 2007: 16 European, 12 Eurasian/Asian, 2 Central/South American, 1 North American, 1 Pacific, 16 Middle Eastern/North African, 2 African.

GDP per capita ranges between 2,539 for Ghana and 109,802 for Qatar; GGI ranges between 0.451 for Yemen and 0.825 for Sweden.

TIMSS 2011: 11 European, 12 Eurasian/Asian, 1 Central/South American, 1 North American, 2 Pacific, 14 Middle Eastern/ North African, 1 African.

GDP per capita ranges between 3,059 for Ghana and 125,141 for Qatar; GGI ranges between 0.575 for Saudi Arabia and 0.840 for Norway.

PISA 2012: 34 European, 12 Eurasian/Asian, 8 Central/South American, 2 North American, 2 Pacific, 6 Middle Eastern/North African.

GDP per capita ranges between 4,408 for Vietnam and 125,141 for Qatar; GGI ranges between 0.601 for Turkey and 0.864 for Iceland.

TIMSS = Trends in International Mathematics and Science Study; PISA = Programme for International Student Assessment; GEM = gender empowerment measure; GGI = gender gap index; GDP = gross domestic product; HDI = Human Development Index.

Contextual Variation in the Gender Gap in STEM Participation

Researchers have focused on women’s representation in STEM departments at higher education institutions and professional STEM jobs to examine the gender gap in STEM participation (see Table 3). The only piece of evidence supporting the freedom hypothesis showed that representation of women in engineering programs was higher in nations with greater gender equality in share of professional jobs (Charles & Bradley, 2009). Several studies failed to find a relationship between the gender gap in STEM participation and economic development or gender equality (Charles & Bradley, 2006, 2009; Ramirez & Wotipka, 2001). Contrary to the freedom hypothesis, Bradley (2000) documented that women constituted a larger proportion of engineering students in less economically developed countries between the years 1965 and 1990. Charles and Bradley (2009) observed that the gender gap in participation in math/science and engineering fields was larger in nations with greater economic development and higher percentage of women in the labor force. Charles (2011a) also reported that the most male-dominated engineering programs in the world between 2005 and 2008 were in the highly industrialized countries of Japan, Switzerland, Germany, and the United States, whereas the least male-dominated ones were in less economically developed nations like Mongolia, Greece, Serbia, and Panama. Representation of women in science fields was also higher in nations that scored lower on gender equality indices (Charles, 2011a). In a recent analysis, Stoet and Geary (2018) found that the percentage of women among STEM graduates decreased as gender equality increased (Stoet and Geary, 2018). Overall, almost all of the evidence contradicts the freedom hypothesis for STEM participation.

OPEN IN VIEWER

Table 3.

Cross-National Variation in the Gender Gap in STEM Participation.

Authors	Sample (countries included)	Findings that support the freedom hypothesis	Findings that fail to support the freedom hypothesis
Bradley (2000)	UNESCO (1965–1993)	—	Proportion of women among engineering graduates was higher in less economically developed countries
Ramirez and Wotipka (2001)	N = 67; UNESCO (1972, 1982, and 1992)	—	Representation of women in natural science and engineering (number of women enrolled in these fields divided by the 20–24 age cohorts) was unrelated to:• GDP per capita in 1972 and 1992 (βs range between −.17 and .07), except in one of the eight models estimated (β = .38), for the respective years• GDP per capita in 1972 (βs range between −.001 and .02) for the year 1992
Charles and Bradley (2006)	N = 21; OECD (2001)	—	Women’s representation among computer science degree holders was unrelated to:• per capita GNP (r = −.04)• women’s share of the labor force (r = −.27)• women’s share of professional occupations (r = .05)• women’s share of tertiary students (r = −.02)
Charles and Bradley (2009)	N = 44; UNESCO (1992–1998)	Representation of women in engineering programs was larger in nations with higher:• women’s share of professional jobs (β = .10)	Representation of women in engineering and math/natural science fields was smaller in nations with higher:• GDP per capita (rs = −.45 and −.48, respectively)Representation of women in engineering fields was unrelated to:• percentage of women in the labor force (β = −.01)Representation of women in math/natural science fields:• was unrelated to percentage of women in professional jobs (β = −.04)was smaller in nations with higher percentage of women in the labor force (β = −.11)
Stoet and Geary (2018)	N = 64; PISA 2015, UNESCO	—	Percentage of degrees conferred to women in STEM was smaller in nations with higher:• GGI, r(50)= −.47

Note. STEM = Science, Technology, Engineering, and Mathematical; UNESCO = United Nations Educational, Scientific and Cultural Organization; GDP = gross domestic product; OECD = Organisation for Economic Co-operation and Development; GNP = gross national product; PISA = Programme for International Student Assessment; GGI = gender gap index.

We conducted additional analyses to examine the relationship between WEIRD or Majority World status and women’s representation in STEM and social science fields using 2018 (or most recent available) data from UNESCO’s database (http://data.uis.unesco.org/). ⁴ Whereas WEIRD status is positively related to the percentage of degrees conferred to women in higher education fields overall, r(77) = .33, p = .003, and in social sciences, r(68) = .34, p = .005, it is negatively correlated with the percentage of degrees conferred to women in computer science, r(55) = −.62, p < .001. The relationships are non-significant for physics, r(68) = .14, p = .27, math, r(68) = −.15, p = .23, and engineering, r(70) = −.15, p = .22. The data suggest that although women’s representation in higher education may be greater in WEIRD than Majority World settings, their representation in STEM is not; women who enter higher education in WEIRD settings may be preferring non-STEM fields such as social sciences.

Why Is the Freedom Hypothesis Unsupported?

Based on studies reviewed in this section, evidence supporting the freedom hypothesis is very limited, whereas evidence contradicting it has been accumulating. There are some inconsistencies across findings in the math and science performance domain, which researchers consider to be a function of different gender equality measures being used, or outlier country groups (e.g., Stoet et al., 2016). Most studies reveal a pattern of gender similarities in math and science performance; where differences exist, effect sizes are generally small (e.g., Hyde, 2005; Hyde et al., 1990, 2008; Kane & Mertz, 2012; Lindberg et al., 2010). When it comes to STEM attitudes and participation, the majority of evidence suggests that, contrary to the freedom hypothesis, the gender gaps are larger in (generally WEIRD) settings with higher economic development and greater gender equality, compared to Majority World settings.

Why is the freedom hypothesis unsupported in terms of the gender gap in STEM attitudes and participation in particular? The assumption implicit in the freedom hypothesis is that gendered belief systems that constrain women’s choices and push them away from STEM would erode with increased gender equality. In the reviewed studies, gender equality is operationalized using indicators that document representation of women and men across economic, political, and education domains, as well as their access to resources such as health care. Researchers have noted that these global indicators of gender equality cannot directly explain variation in the gender gaps in STEM participation, as they are not designed to provide causal explanations for such phenomena (Richardson et al., 2020). These gender equality measures do not account for various sociopolitical and historical features of societies that may actually underlie women’s rates of representation across political, economic, or education domains to begin with (Richardson et al., 2020). Indeed, several researchers have emphasized the multifaceted nature of gender equality (e.g., Charles & Grusky, 2004; Knight & Brinton, 2017; Richardson et al., 2020). For instance, Knight and Brinton (2017) have argued that rejection of traditional gender roles emphasizing male primacy, gender essentialism (i.e., beliefs in innate gender differences), and a cultural emphasis on free choice are intersecting dimensions that lead to different varieties of gender egalitarianism. Based on their argument, systems that uphold gender egalitarian norms may still support beliefs in inherent gender differences, and encourage their enactment through free choice. Gender differences in participation in certain academic or professional fields may therefore be acceptable in an otherwise gender egalitarian system. Charles and Bradley (2009) have similarly argued that gender essentialist ideologies can justify unequal representation of men and women in certain domains. As a result, whereas vertical forms of gender inequality (e.g., unequal participation in higher education overall) become less acceptable in gender egalitarian societies, horizontal forms (e.g., unequal participation across domains) may persist (Charles & Bradley, 2002). In light of the literature, our aim is to move beyond global indicators of gender equality as the cause of gender gaps and consider cultural patterns across societies to understand the roots of contextual variation in the gender gap in STEM. We outline our model in more detail in the following sections.

The Gendering of Academic Attitudes Through the STEM=Male Stereotype

Researchers challenging the freedom hypothesis have emphasized the cultural shaping of seemingly personal preferences within highly gendered social structures (Cech, 2013; Charles, 2011a, 2017; Charles et al., 2014). Social psychological research provides ample support that the STEM=male stereotype can affect performance in STEM-related domains and play a central role in the gendering of interest, anxiety, self-efficacy, and judgments of ability and performance (see path 1 in Figure 2). Stereotypes permeate into children’s and adolescents’ lives through the behaviors of their parents, teachers, and peers (Sha et al., 2016; Simpkins et al., 2012). Parents may have different expectations of their sons and daughters in math-related domains, which can affect the learning opportunities they provide for their children (Catsambis, 2005). Differential treatment of boys and girls by parents, teachers, and counselors shapes children’s self-beliefs, interests, abilities, and performance in accordance with stereotypical expectations (e.g., Frome & Eccles, 1998; Gunderson et al., 2012; Jacobs et al., 2005; Lee & Burkam, 1996; Saucerman & Vasquez, 2014; Zeldin & Pajares, 2000). For instance, girls and women tend to report lower efficacy beliefs or confidence in STEM domains than do boys and men, even when they perform equally well or better than them (Correll, 2001, 2004; Else-Quest et al., 2010, 2013; Hackett, 1985; Herbert & Stipek, 2005; MacPhee et al., 2013; Pajares, 2005; Sikora & Pokropek, 2012). Beliefs about efficacy and confidence relate to gender discrepancies in academic aspirations, interests, and persistence in general (Bandura et al., 2001) and in STEM domains in particular (Correll, 2001; Hackett, 1985; Jagacinski, 2013; Zeldin & Pajares, 2000).

Even women and girls who develop positive attitudes toward STEM domains may experience concerns about confirming negative stereotypes, which can harm performance and attitudes (Steele, 1997; Steele et al., 2002). For example, reminders of stereotypes about gender differences in math ability (Spencer et al., 1999) or situational cues in STEM-related contexts such as reminders of sexism or subtle sexist behaviors (Adams et al., 2006; Logel et al., 2009) interfere with women’s performance on a relevant task. Meta-analyses document small to moderate effects of gender-based stereotype threat on math performance (Doyle & Voyer, 2016; Flore & Wicherts, 2015; Nguyen & Ryan, 2008; Picho et al., 2013; Shewach et al., 2019; Stoet & Geary, 2012; Walton & Spencer, 2009), which may contribute to gender differences in standardized test scores (Spencer et al., 2016). Repeated experience of stereotype threat can lead women to distance from math-intensive fields altogether (Spencer et al., 1999; Steele, 1997). It is also possible for men to experience a boost in performance or attitudes due to the same stereotypes that disadvantage women (e.g., Walton & Cohen, 2003), although research on men’s experiences in STEM is much more limited.

Importantly, cultural patterns such as power distance (PD) may influence the extent to which stereotypes impact STEM attitudes by shaping habitual social comparison tendencies (see path 2 in Figure 2). PD is defined by the acceptance of hierarchies in a society such as status differences between women and men (Hofstede, 2001) and is lower in WEIRD settings (i.e., PD is negatively related to gender equality and economic development; Stoet et al., 2016). Researchers have argued that low (high) PD can encourage habitual engagement in across-(within-)category comparisons (Guimond et al., 2006). For instance, women in low PD settings may consider it appropriate to compare themselves with men, whereas in high PD settings, women may be less likely to include men in the set of relevant others with whom they compare themselves, and restrict their comparisons to other women (Guimond et al., 2006). Gender stereotypes become more likely to inform self-evaluations in (mostly WEIRD) low PD settings where women consider men as a relevant comparison category for themselves (Guimond et al., 2006, 2007). The shifting standards model likewise posits that women may evaluate their own attitudes toward math to be more positive when they compare themselves to “the average woman” (who, stereotypically, has negative math attitudes, and sets a low standard), but relatively negative when they compare themselves to “the average man” (Biernat, 2012; Biernat & Manis, 1994; Biernat et al., 1991, also see Heine et al., 2002). Consistent with this reasoning, gender gaps in math anxiety (Stoet et al., 2016) and math performance (Hamamura, 2012) are larger in low PD settings. ⁶ These patterns are consistent with the larger gender differences in self-reported personality characteristics and preferences found in more gender-equal societies (Costa et al., 2001; Lippa et al., 2010).

It is also possible for the effects of stereotype threat to be amplified in low PD settings. Research has shown that girls and women are less likely to face stereotype threat in single-sex settings (Dasgupta et al., 2015; Huguet & Regner, 2007; Inzlicht & Ben-Zeev, 2000), where they are less likely to compare themselves with men. For instance, women in engineering felt less threatened and more challenged by a task (in a positive sense) when they worked in teams that had gender parity or were composed mostly of women, compared to teams where women were in the minority (Dasgupta et al., 2015). Women in the former groups also reported greater self-confidence and more ambitious career aspirations, even when stereotypes associating engineering with males were activated (Dasgupta et al., 2015). Weaker habitual tendencies to engage in comparisons with men may protect women against the adverse effects of negative stereotypes on performance and attitudes.

The Cultural Construction of Academic Choice

Although evidence for the gendering of attitudes is abundant, cultural variation in the relationship between attitudes and academic choice is not commonly considered in the literature (see path 3 in Figure 2). The relationship between attitudes and choice is perhaps taken for granted due to the assumption that a choice is simply the enactment of personal preferences that make up the core of attitudes (Riemer et al., 2014; Stephens et al., 2007). However, cultural patterns in WEIRD and Majority World settings inform the normative relationship between attitudes and choice in different ways. More specifically, the cultural patterns specific to WEIRD and Majority World settings foster different understandings of the self and agency. These understandings, in turn, foster a particular construction of academic choice that resonates with them. In the following sections, we discuss self-expressive and security-oriented constructions of choice that emerge in WEIRD and Majority World settings, respectively, in terms of their implications for women’s and men’s tendency to choose STEM or not.

Implications of the Gender Gap for the STEM=Male Stereotype

The final part of our model considers how the observed gender gap in STEM participation feeds back into the STEM=male stereotype, which underlies the contextual variation in such internalized associations (path 6 in Figure 2). Based on the Implicit Association Test, the majority (70%) of participants from 34 countries showed a tendency to implicitly associate men with science and women with liberal arts more than vice versa (Greenwald et al., 1998, 2003; Nosek et al., 2009). The pattern of cross-national variation in the strength of stereotypical associations matches the pattern of variation in the gender gap in STEM participation. This is not surprising, because the strength of stereotypical associations is not only a cause but also a reflection of existing gender gaps in STEM participation (Nosek et al., 2009; Nosek & Smyth, 2011). In a study using data from 66 nations, women’s representation in science fields in tertiary education and in research jobs predicted strength of implicit and explicit stereotyping (D. I. Miller et al., 2015). For instance, stereotypical associations between men and STEM are very strong in the Netherlands, where women have very low levels of representation in science (D. I. Miller et al., 2015). An analysis we conducted based on global indicators provided by the UNDP (2018) and recent data on the strength of implicit associations between STEM and men (and between women and liberal arts, relative to vice versa, retrieved from https://osf.io/y9hiq/) shows that internalized associations between STEM and men are stronger in WEIRD compared to Majority World settings, r(104) = .28, p = .004.

Not only the strength but also the content of stereotypes about gender and academic fields vary across settings as a function of the size of the gender gap in STEM participation. For instance, in the United States, where women’s representation in CS is very low, stereotypes represent CS as a “geeky” domain that involves working in isolation for the most part and is suitable for males who lack social skills (e.g., Cheryan et al., 2015). In contrast, such a stereotype is not common in countries where there is gender parity in participation in CS. For instance, local norms in Malaysian and Indian contexts consider CS as a field that offers opportunities for jobs in comfortable and safe office spaces, which are considered particularly suitable for women (Lagesen, 2008; Varma, 2009). Similarly, CS is a field that appeals to both men and women in Armenia due to the lucrative career opportunities it offers, and it is not strongly gendered (Gharibyan & Gunsaulus, 2006).

An important point to emphasize is that stereotypes are not simply reflections of traditional gender ideologies; they are also internalized associations based on observed material realities (e.g., gender gaps in STEM participation). As long as gender differences in representation across academic fields persist, implicit stereotypical associations between gender categories and academic fields are also likely to persist, even though traditional gender ideologies may be eroding. These internalized associations may continue to fuel the gendering of attitudes in a subtle way, and to shape choices, particularly in WEIRD settings. On the contrary, in Majority World settings, attitudes may be less gendered for those STEM fields where women are represented more or less equally with men to begin with.

Summary

Our model strives to provide an explanation for the pattern of cross-national variation in the gender gap in STEM participation. In the first part of our model, we documented social psychological research showing that academic attitudes are gendered through the STEM=male stereotype. Furthermore, we discussed the reasons why the impact of stereotypes on attitudes may be larger in more gender egalitarian, low PD settings, where women habitually compare themselves with men. The story is incomplete, however, if one fails to consider the extent to which these gendered attitudes inform academic choices as a function of the particular cultural construction of choice. Therefore, in the second part of our model, we drew upon cultural psychological perspectives to examine cultural understandings of self, agency, and choice, which inform the normative standards for making an academic choice in a particular setting. We presented research evidence for the processes underlying the larger gender gap in STEM participation in WEIRD settings that define academic choice as self-expressive, compared to Majority World settings that define academic choice as security-oriented. In the final part of our model, we considered how the gender gap in STEM participation feeds back into the STEM=male stereotype, fueling the reproduction of gender similarities or differences in academic choices.

Considering the Role of Cultural Patterns in Gender Gaps

Most research in the area has considered the experiences of (White) women in North American and European settings to understand the roots of the gender gaps in STEM. Our analysis highlights that it is necessary to consider ways of being outside of hegemonic WEIRD settings to reach a comprehensive understanding of the sociocultural bases of these gender gaps. We acknowledge that the WEIRD-Majority World distinction that we use has limitations, and runs the risk of essentializing cultural patterns. Although it would be ideal to consider the specific characteristics of each cultural setting, the dichotomy that we use provides a starting point to challenge certain assumptions inherent in mainstream research.

One implication of our analysis is that is important to extend research on the gender gap in STEM in a way to consider the role of cultural patterns in academic choices of both women and men. It is necessary to keep in mind the dynamic nature of the cultural patterns that we have discussed in our analysis (e.g., Adams & Markus, 2004), such that they are subject to change over time. Furthermore, people often experience the influence of various cultures, and are able to switch between different cultural frames (Hong et al., 2000). Relatedly, students may consider self-expression as well as security as important goals of academic choice. Future research might consider how people reconcile self-expressive and security-oriented goals when making their choices. For instance, younger generations in security-oriented settings may be exposed to messages about such ideals as pursuit of dreams and desires through the media in an increasingly globalizing world. At the same time, they may be exposed to messages about the importance of security from older family members. Likewise, students in settings that emphasize self-expression may also be concerned about security due to perceived societal or global instability, or disadvantaged status due to social class or racial/ethnic identity. Future research might also examine whether concepts such as authenticity and identity fit that we have discussed as part of a self-expressive construction of choice may resonate with people in Majority World settings to a lesser extent than they do with people in WEIRD settings. Again, the influence of such concerns may vary as a function of exposure to foreign cultures that emphasize self-expression.

For a direct test of the effects of self-expressive or security-oriented constructions of academic choice on the gender gap in STEM, researchers might experimentally activate them. For instance, in an experimental study in the United States, both male and female college student participants were assigned to self-expression, economic security, or control conditions to examine the influence of activation of these different constructions (Soylu Yalcinkaya, 2017). Reading a passage on the importance of self-expression through academic choice led women to report lower motivation to pursue STEM compared to a control condition, but did not affect men’s responses (Soylu Yalcinkaya, 2017). On the contrary, activation of concerns about future financial security led male participants to show greater motivation to pursue STEM compared to a control condition, but did not affect women’s responses (Soylu Yalcinkaya, 2017). Future research could look into different ways of activating financial security concerns and relational obligations among both women and men across cultural settings. For instance, prospects of future instability and insecurity at the societal level can be effective in activating security goals in academic choice. Activating a promotion focus (i.e., considering past and current hopes, dreams, and aspirations) or prevention focus (i.e., considering duties, responsibilities, and obligations; Higgins, 2005; Kirmani & Zhu, 2007; Pham & Avnet, 2004) may also lead people toward different goals in academic choice.

Challenging the Androcentric Standard

In our analysis, we have emphasized the importance of bringing the experiences of men, in addition to women, under research focus. The focus of research on women’s avoidance of STEM is understandable to the extent that women’s exclusion from many STEM fields deprives them of important opportunities to achieve prestige and economic power. However, the focus on women in STEM as the phenomenon of investigation may inadvertently elevate men’s experiences to an unquestioned standard, against which women’s experiences become a “deviation” that requires explanation and intervention (Hegarty & Pratto, 2001).

In contrast, an important strategy of cultural psychological perspectives is to interrogate apparently natural truths as told from dominant standpoints (Adams & Salter, 2007; Adams et al., 2015). One implication of this strategy is to illuminate the extent to which men’s experience in STEM is not a natural expression of inherent ability or interest. Instead, it is a product of sociocultural forces that enhance their performance and motivation to participate in STEM. For instance, research on stereotype lift suggests that gender gaps in math performance result partly from an awareness of stereotypes about men’s greater math ability that artificially inflate their performance (Walton & Cohen, 2003). In a similar fashion, we have argued that men’s experience in STEM is not a natural expression of inherent ability or interest; instead, just like women’s, it is a product of sociocultural forces that enhance their performance and participation in STEM. Bringing men’s choices under focus is an important step in challenging the taken-for-granted assumption that men’s STEM outcomes constitute the natural standard or ideal.

A second implication of this strategy is to re-think assumptions about the desirability of STEM as a career goal. Some research suggests that high rates of participation of men in a particular field can increase its societal value, leading to an association between male domination and the social valuation of fields over time (Levanon et al., 2009). As a result, androcentric standards about the value of different academic fields emerge. In that case, efforts to push women into male-dominated fields as a means to improve their access to power may be somewhat limited. Instead, efforts might focus on achieving equal gender representation across all fields and challenging the devaluation of female-dominated fields and over-valuation of male-dominated fields (see Croft et al., 2015).

Advancing the Cultural Psychology of Preference and Choice

One assumption our analysis has aimed to challenge is that choice is a direct reflection of personal preferences, which are stable and authentic entities. Some readers may interpret the observation that greater opportunity to express personal preferences leads to more gender-stereotypical choices as evidence for natural or innate sex differences, such that men have a natural interest or ability in STEM pursuit that women may not inherently possess (e.g., Schmitt et al., 2008; Su et al., 2009). Furthermore, people often believe that women are motivated by wishes and aspirations rather than obligations and duties (Johnston & Diekman, 2015). Therefore, they may consider the roles that women undertake in the society to be a result of their own preferences (Johnston & Diekman, 2015). However, without affirming or denying the possibility of innate, sex-linked differences in interests, cultural psychological perspectives emphasize the dynamic, sociocultural constitution of personal preference. Previous researchers have raised precisely the point that personal preferences are not just natural, but instead develop as people engage with (gendered) sociocultural ecologies that channel their interests and identifications in particular directions (Cech, 2013; Charles & Bradley, 2009; Falk & Hermle, 2018; Schmitt et al., 2008). Humans inhabit cultural ecologies shaped by design (or less conscious selection) to make material particular beliefs and desires. From this perspective, it is not surprising that young girls and adult women develop little interest in or motivation to pursue STEM, given hyper-masculine constructions of these fields that constitute a hostile climate for them (e.g., Cheryan et al., 2009, 2011), and that men are drawn to these fields. One important implication of the dynamic relationship between psyche and culture concerns the malleability of preference: academic preferences of women and men can change if cultural constructions of STEM (and other fields) change. Such a transformation would be particularly consequential for STEM participation in settings where preferences form the primary basis of choice.

Considering the Experiences of Marginalized Populations

An important contribution of intersectional perspectives (e.g., Shields, 2008; Warner & Shields, 2013), which are informed by decolonial and women-of-color feminisms (Crenshaw, 1991; Mohanty, 1991), to the present topic is to highlight how structures of domination along one identity dimension (e.g., gender) can have qualitatively different consequences as a function of another identity dimension (e.g., racial, social class, or cultural identity). This can help illuminate the experience of the marginalized majority of women who are typically invisible in hegemonic accounts (e.g., Cole, 2009) and challenge the tendency to treat the experience of White women in WEIRD settings as a default standard for an assumed monolithic, essential category of “women.”

Intersectional perspectives can be instrumental for the purposes of acknowledging cultural variation in constructions of choice along the lines of social class (e.g., Kraus et al., 2011) and racial/ethnic spaces, as they intersect with gender. For instance, a self-expressive construction of choice may be more common in upper social class settings (Stephens et al., 2007), where the abundance of material and social resources allows people to prioritize the pursuit of individual goals or desires (Kraus et al., 2012; Kraus & Stephens, 2012; Stephens et al., 2011, 2012). This may particularly influence women’s choices, as men may still face obligations to provide for their families, as we have discussed earlier (e.g., Mullen, 2014; Stone & McKee, 2000). On the contrary, a security-oriented construction of academic choice may be typical of lower social class settings, which foster an interdependent self-construal (Grossmann & Varnum, 2011), and a greater focus on future material security as a primary goal of college education (e.g., Eagan et al., 2015; Snibbe & Markus, 2005; Stephens et al., 2007). Hence, the gender gap in STEM choice may be larger in upper social class settings. However, one may also consider a possible effect of social class in the opposite direction, such that the resources of upper social class families may help students develop interests and capabilities related to STEM (e.g., Wang & Degol, 2013). Future research could examine the gender gap in STEM across social classes. Likewise, future research might focus on constructions of academic choice across racial spaces within societies. For instance, researchers have found weaker implicit “STEM=male” associations among African American students (women and men) than among European American students, which statistically accounted for European American women’s lower tendency to participate in STEM majors than African American women (O’Brien et al., 2015). It is possible for different constructions of choice to shape men’s and women’s academic decisions across these racial spaces (Riegle-Crumb et al., 2011).

Implications for Education, Intervention, and Policy

Our analysis has several potential implications for practice. Wang and Degol (2013, 2017) caution readers against forcing women or men into fields they are not interested in, advising instead that interventions focus on teaching young girls and boys that they are completely free to choose any pursuit they may wish. Their approach is typical of research based in WEIRD settings that emphasize the removal of barriers, with the assumption that it would empower women to pursue any career without being constrained by their gender. However, our analysis emphasizes that the path to gender equality in representation in STEM is far less straightforward. Still, Wang and Degol’s (2013) arguments do raise a relevant question: Should educators or policy-makers “push” women to pursue fields such as STEM that they have no interest in, or that potentially constitute a hostile climate for them? One answer to this question emphasizes that preferences are not simply innate but gendered through cultural stereotypes, artificially precluding women from entering lucrative and high-status fields (Beede et al., 2011; Charles, 2011a, 2011b; Cheryan et al., 2009; Freeman, 2004; Hill et al., 2010), which contributes to the reproduction of women’s disempowerment relative to men (Charles, 2011a, 2011b; Correll, 2004; Croft et al., 2015; Riegle-Crumb, 2005; Wood & Eagly, 2012). This point is very important but incomplete without an attempt to re-think assumptions about the inherent value or desirability of STEM fields. Efforts to push women into male-dominated fields as a means to improve their access to power may be somewhat limited without challenging androcentric standards about the value of occupations, as we have discussed earlier (see Croft et al., 2015; Riegle-Crumb et al., 2016).

One implication of our analysis is that increasing participation of women in STEM takes more than a one-size-fits-all approach; interventions need to take into consideration the dominant construction of choice in a setting to maximize effectiveness. For instance, social psychological research on the gender gap in STEM participation seldom considers the role of variables other than performance and attitudes, such as financial security and fulfilling one’s family’s needs and wishes. Research in vocational psychology has indeed considered family influence as a factor in career selection (Fouad et al., 2010, 2016). Moving beyond the assumption that choices reflect personal preferences can pave the way for researchers and practitioners to consider other factors such as relational expectations and duties, which would be important for intervention purposes.

Another important implication concerns the role of teachers and counselors in guiding students in the process of major selection. Without ignoring interests or proclivities, teachers and counselors may encourage students to consider the broader societal context and future career opportunities when making their academic choices. For instance, Diekman and colleagues (2017) have argued that an abundance of options in higher education may lead women to choose fields where they can use their strengths in the reading domain (i.e., more gender-stereotypical domains) over math-related domains, even if they might also be strong performers in the latter. Broadening the considerations that shape students’ academic choices can make STEM options more appealing.

Furthermore, re-structuring education systems in a way to require greater STEM-related courses to facilitate early exposure to these domains in a gender-balanced environment can also help eradicate gender gaps (Cheryan et al., 2017). When provided with greater freedom of choice, girls may tend to select fewer math and science courses than do boys in middle school (Lee & Burkam, 1996) and be less likely to take courses in physics, computer programming, and engineering than boys in high school (Barron, 2004; Hazari et al., 2010; Nord et al., 2011; Tyson et al., 2007). This is particularly the case in settings like the United States, where education systems require few courses and provide abundant freedom for students to choose their courses (Cheryan et al., 2017; Cunningham et al., 2015). Not surprisingly, the fields that have largest gender gaps in higher education in the United States are those that students are generally not required to engage with during high school (Cheryan et al., 2017). On the contrary, education systems that are highly standardized and require equal exposure to all content areas regardless of personal preference foster smaller gender gaps in participation in CS, for instance (e.g., Turkey, South Korea, China, and India; Ayalon & Livneh, 2013; Charles & Bradley, 2006; Tsui, 2007; Varma, 2009).

Concluding Remarks: Re-Thinking Liberty and Liberation

Our review of cultural-ecological variation in STEM gender gaps suggests that freedom from financial or relation-maintenance concerns and freedom to pursue personal dreams may insidiously constrain women, who seem otherwise liberated, to conform to particular stereotypes about gender and academic pursuit. The point of this analysis is neither to blame women who inhabit settings of privilege for opting out of STEM nor to imply that the gender gap is a frivolous, “first-world” problem that does not deserve attention. Instead, the point concerns appropriate responses to (gender) oppression in STEM and otherwise. An understanding of liberation at the individual level, defined as liberty to enact personal goals and desires free of societal constraints, may resonate with people in WEIRD and other societies that value self-expression (Markus & Schwartz, 2010; Savani et al., 2008; Stout et al., 2011). Drawing on this conception of liberation, conventional responses to the gender gap in STEM emphasize removal of gendered barriers to increase women’s freedom of academic and professional choice. Whereas these efforts are worthwhile, cultural psychological approaches to the gender gap in STEM suggests that they may not automatically produce women’s liberation in the sense of collective relief from structural inequality (e.g., Cheryan et al., 2017). From this perspective, the key to liberation from oppression may not lie in the expansion of individual liberty to pursue self-expression and personal fulfillment; instead, individual liberation in this sense may inadvertently contribute to societal gender stratification by reproducing the male-dominated status of prestigious and well-paid STEM occupations.