You are on page 1of 12

Journal of Educational Psychology

© 2019 American Psychological Association 2019, Vol. 111, No. 6, 1119 –1130
0022-0663/19/$12.00 http://dx.doi.org/10.1037/edu0000340

Seeing Is Believing: Gender Diversity in STEM Is Related to


Mathematics Self-Concept
Christoph Niepel Matthias Stadler
University of Luxembourg Ludwig Maximilian University of Munich

Samuel Greiff
University of Luxembourg
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Although female students’ overall performance in mathematics is on a par with the performance of male
This document is copyrighted by the American Psychological Association or one of its allied publishers.

students, female students tend to report lower levels of mathematics self-concept (MSC) than their male
schoolmates. With the present study, we examined for the first time whether occupational gender diversity
(i.e., a balanced gender ratio) in science, technology, engineering, and mathematics (STEM) statistically
predicted students’ MSC beyond well-established sources of self-concept formation as described in the
big-fish-little-pond effect framework. To this end, we applied linear mixed-effects models to large, represen-
tative data sets comprising 120,270 students from 23 countries. After controlling for individual-level and
school-level achievement, we found an interaction between students’ gender and country-level STEM gender
diversity, with female students reporting higher MSC in countries characterized by greater gender diversity.
Our results therefore suggest that a lack of societal STEM gender diversity negatively affects female students’
MSC formation and that good mathematics performance in itself does not protect female students from this
adverse trend. We interpret our findings against the background of self-concept theory and research as well
as psychological theories of gendered socialization.

Educational Impact and Implications Statement


In our study, we examined gender differences in students’ mathematics self-concept (MSC)—that is, their
perception of their own mathematics ability—across 23 countries. Whereas female and male students tend
to perform similarly well in mathematics, female students tend to rate their own mathematics ability
considerably lower than their male counterparts do in almost all countries. Our results revealed that the
ratio of women to men in scientific, technical, engineering, or mathematics (STEM) research jobs in a
country was significantly related to female students’ MSC, but this finding did not hold for the male
students. Female students reported lower MSCs when they lived in countries in which female scientists
were scarce—regardless of the female students’ mathematics achievement. High achievement in math-
ematics alone thus does not seem to be sufficient for improving female students’ MSC in societies with
gender inequities in STEM, further indicating the need to enhance gender diversity in STEM fields around
the world in order to provide role models for girls and women.

Keywords: gender equity, occupational gender diversity, mathematics self-concept, big-fish-little-pond


effect, international large-scale data

Members of each sex are encouraged in, and become interested in and Although mathematics achievement today is characterized by a
proficient at, the kinds of tasks that are most relevant to the roles they high degree of gender similarity in general (Hyde, Lindberg, Linn,
fill currently or are expected to fill in the future Ellis, & Williams, 2008), female students tend to consider them-
—Eleanor Maccoby (1966). selves less competent in mathematics than male students consider

This article was published Online First February 7, 2019. Christoph Niepel and Samuel Greiff. All statements expressed in this
Christoph Niepel, Institute of Cognitive Science and Assessment, Uni- article are the authors’ and do not reflect the official opinions or policies of
versity of Luxembourg; Matthias Stadler, Department of Psychology, the authors’ host affiliations or any of the supporting institutions. We thank
Ludwig Maximilian University of Munich; Samuel Greiff, Institute of Lindie van der Westhuizen for her valuable support.
Cognitive Science and Assessment, University of Luxembourg. Correspondence concerning this article should be addressed to Christoph
Christoph Niepel and Matthias Stadler have joint first authorship. This Niepel, Institute of Cognitive Science and Assessment, University of
research was supported by a grant from the Luxembourg National Research Luxembourg, 11, Porte des Sciences, 4366 Esch-sur-Alzette, Luxembourg.
Fund (ATTRACT “ASKI21”; CORE “DynASCEL” C16/SC/11333571) to E-mail: christoph.niepel@uni.lu

1119
1120 NIEPEL, STADLER, AND GREIFF

themselves (OECD, 2015a). These gender differences even tend to of reference is available to them to evaluate their performance. For
hold when female students share the same competence level in instance, students enrolled in high-ability schools typically report
mathematics (OECD, 2015a, for recent data from the Programme lower levels of academic self-concept than equally able students
for International Student Assessment [PISA]) or obtain similar or enrolled in lower ability schools. This phenomenon is known as
better grades in mathematics compared with male students (Marsh the big-fish-little-pond effect (BFLPE; Marsh, 1987). The BFLPE
& Yeung, 1998). This gender gap is extremely problematic be- framework claims a social comparison effect of achievement
cause one’s mental representation of one’s own mathematics abil- on academic self-concept formation, which is typically studied
ity (i.e., one’s mathematics self-concept; MSC) is key not only for through multilevel regression models that take two different eco-
students’ well-being at school but also when it comes to predicting logical levels into consideration. Specifically, (a) while controlling
academic outcomes that enjoy a high level of societal recognition for the positive effect of achievement on academic self-concept on
and remuneration (Guo, Parker, Marsh, & Morin, 2015; Marsh, the individual student level (e.g., higher achieving students in
2006). This includes the pursuit of careers in the fields of science, mathematics will have higher MSC), (b) school-average achieve-
technology, engineering, and mathematics (STEM; Guo et al., ment negatively predicts academic self-concept (i.e., the BFLPE;
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

2015). Consequently, gender differences in MSC to the disadvan- e.g., the higher the mathematics achievement of one’s frame of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

tage of female students arguably hamper current societal and reference in general, the lower one’s MSC). At the core of the
political efforts that are geared toward fostering gender equity in BFLPE framework thus stand two assumed effects on academic
STEM fields; namely, STEM gender diversity (i.e., increasing self-concept formation: a positive effect of higher individual
women’s representation in STEM occupations). achievement and a negative frame-of-reference effect of higher
Not only is there convincing conceptual evidence (as articulated school-average achievement. The BFLPE framework has been
by different theories of gendered socialization; e.g., Eagly & extensively validated, among other approaches, by evaluating it
Wood, 2012; Eccles, 1994) but also initial empirical evidence that with cross-national large-scale data (e.g., Marsh & Hau, 2003;
the degree of gender diversity in STEM fields in a given society Seaton et al., 2009). It can thus explain important mechanisms of
may have a considerable impact on students’ MSC (Else-Quest, academic self-concept formation on the individual level and on the
Hyde, & Linn, 2010). Self-concept theory and research has exten- school level (see also Marsh, 2006). However, the BFLPE frame-
sively focused on social comparison and frame-of-reference ef- work alone cannot explain why female students with comparable
fects as determinants of MSC. Intriguingly, however, there have mathematics achievement tend to report lower levels of MSC than
been no comprehensive attempts in such research to incorporate male students do (see, e.g., OECD, 2015a).
societal-level gender diversity in STEM into established models of Turning to this conundrum, the notion that female students’
students’ MSC formation. Consequently, it is yet an open question gendered socialization experiences can largely explain such gender
whether and to what extent the experience of societal-level gender differences—through perceived societal norms or feedback and the
diversity in STEM is associated with students’ MSC beyond expectations of socializers such as parents or teachers—is virtually
well-known concurrent sources of MSC on an individual or school undisputed in self-concept theory and research (Trautwein &
level. Möller, 2016). Indeed, previous cross-national research has al-
In the current study, we drew on large cross-national data sets ready shed light on the potential impact of gender (in)equities in a
from PISA and the United Nations Educational, Scientific, and given society on its people’s cognitive ability (Bonsang, Skirbekk,
Cultural Organization (UNESCO) in an attempt to answer this & Staudinger, 2017), mathematics achievement (Else-Quest et al.,
compelling question. To this end, we brought together two re- 2010; Guiso, Monte, Sapienza, & Zingales, 2008), as well as
search strands that—to the best of our knowledge— have never STEM-related affect, attitudes, and beliefs (Else-Quest et al.,
before been studied simultaneously. These are (a) cross-national 2010; Miller et al., 2015; see also Stoet, Bailey, Moore, & Geary,
research on the formation of academic self-concept through social 2016) including MSC (Else-Quest et al., 2010). Else-Quest et al.
comparison effects (e.g., Marsh & Hau, 2003; Seaton, Marsh, & (2010) examined various relations between different indicators of
Craven, 2009) and (b) cross-national research on the impact of societal gender equity on the one hand and cross-national gender
societal STEM gender equity on STEM-related beliefs and atti- differences in students’ mathematics achievement, beliefs, and
tudes (e.g., Else-Quest et al., 2010; Miller, Eagly, & Linn, 2015). attitudes on the other. They found that women’s representation in
research jobs (operationalized as the percentage of women in
research positions across a number of different fields) emerged as
Theoretical Background
the strongest single predictor of gender differences in MSC (␤ ⫽
Shavelson, Hubner, and Stanton (1976) defined self-concept as |0.60|, p ⬍ .01; R2 ⫽ .36), such that countries with larger propor-
a personal perception of oneself that is shaped by one’s experience tions of women participating in research exhibited smaller gender
with the environment and thus strongly reinforced by the environ- gaps in MSC (Else-Quest et al., 2010). Naturally, the question that
ment and significant others (p. 411). Thus, academic self-concept arises is: What are the mechanisms behind the finding that socio-
can be defined as the personal perceptions of one’s own abilities, cultural factors such as gender diversity in research positions are
with MSC reflecting a person’s self-perception in mathematics related to students’ individual MSC? In the remainder of this
(Shavelson et al., 1976). Previous research has repeatedly demon- article, we refer primarily to Eagly and Wood’s (1999, 2012, 2016)
strated that social comparisons and frames of reference (Festinger, social role theory and Eccles and colleagues’ (Eccles et al., 1983;
1954; Mussweiler, 2003) play prominent roles in the formation of Eccles, 1994; Wigfield & Eccles, 1992) expectancy-value frame-
academic self-concepts such as MSC (e.g., Huguet et al., 2009; work to shed light on the psychological mechanisms behind gen-
Wheeler & Miyake, 1992). Accordingly, students with comparable dered socialization, which may cause STEM gender diversity on a
achievement develop different MSCs depending on which frame societal level to affect students’ individual MSC formation.
STEM GENDER DIVERSITY AND MATHEMATICS SELF-CONCEPT 1121

Social Role Theory Eccles et al.’s (1983) framework, young learners’ expectations of
success and the value they attach to mathematics are affected by
Social role theory focuses on the notion that similarities and the overall cultural milieu (e.g., women’s representation in STEM
differences in the behavior, beliefs, and affect of men and women occupations in a given society) as well as by the beliefs, stereo-
reflect gender role beliefs, which are primarily rooted in the types, and behaviors of important socializers (e.g., parents and
distributions of men and women across various social roles in their teachers). Young learners’ individual perceptions of socializers’
society (Eagly & Wood, 2012, 2016). Specifically, gender role attitudes and expectations for them as well as their perceptions of
beliefs are shaped when people encounter the behavior of women activity stereotypes and gender roles (e.g., through a gendered
and men in their everyday lives. This means that individuals division of labor) shape their expectations of success and values.
observe women and men’s behavior and conclude that women and Their individual aptitudes and previous achievement-related expe-
men are equipped with corresponding dispositions (Eagly & riences (e.g., mathematics grades or test scores) also exhibit an
Wood, 2012). According to social role theory, women and men’s impact on the expectations of success and values they attach to
division of labor in a society thereby serves as a prominent source mathematics as well as their own interpretations of these experi-
of such observations (Eagly & Wood, 2012). Gender role beliefs
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ences (e.g., seeing one’s success as a consequence of one’s hard


shape gender similarities and differences in affect, cognition, and
This document is copyrighted by the American Psychological Association or one of its allied publishers.

work vs. one’s talent; Eccles, 1994; Jacobs, Davis-Kean, Bleeker,


behavior through different processes; for instance, through inter- Eccles, & Malanchuk, 2005; Wigfield & Eccles, 1992).
nalized self-standards (Witt & Wood, 2010) or others’ expecta- Notably, expectations of success and academic self-concept
tions (Eagly, Makhijani, & Klonsky, 1992) against which individ- have been shown to be empirically directly linked to each other
uals regulate their behavior (Eagly & Wood, 2012). Accordingly, and can thus be collapsed into a single construct (Eccles, 2009;
people living in societies that are characterized by low STEM Trautwein & Möller, 2016). Accordingly, Eccles et al.’s (1983)
gender diversity should be more inclined to incorporate the gender framework has been frequently used to systematize various find-
role belief that STEM fields are reserved for men only, thus ings in self-concept research (e.g., Arens, Yeung, Craven, &
preventing female students from pursuing careers in such fields Hasselhorn, 2011). The expectancy-value framework has found
and therewith sustaining gender inequity in STEM occupations. In strong empirical support to date (Frome & Eccles, 1998; Jacobs et
the same vein, Miller and colleagues (2015) examined the relation al., 2005; Tenenbaum & Leaper, 2002), and with expectation of
between societal STEM gender diversity on the one hand (opera- success (academic self-concept) at its core encompassing different
tionalized as women’s representation in tertiary science education intertwined ecological levels, this framework may be particularly
and science careers) and individuals’ explicit and implicit gender- useful for researchers who would like to examine whether societal
science stereotypes on the other hand. They indeed found that gender inequity in STEM adversely predicts (female) students’
people living in countries that were characterized by a higher MSC formation. Thereby, Eccles et al.’s (1983) framework is
degree of gender diversity in STEM occupations endorsed fewer consistent with social role theory (Eagly & Wood, 2012) in pre-
gender-science stereotypes that connected science with men more dicting that female students should report lower levels of MSC
than with women (Miller et al., 2015). Miller and colleagues when they live in societies in which fewer women work in STEM
(2015) argued that in order to change persisting gender-science fields (Else-Quest et al., 2010). In addition, the Eccles et al. (1983)
stereotypes, people need to either directly observe (e.g., through framework also highlights mathematics aptitude and achievement-
social interactions) or indirectly observe (e.g., through mass me- related experiences as influential sources of MSC and thus offers
dia) multiple and mutually reinforcing examples of counterstereo- an integrative framework from which to examine both STEM
typical women. Thereby, female students’ exposure to only a few gender diversity and mathematics achievement (as claimed by the
female role models in STEM might be insufficient to change the BFLPE framework; Marsh, 1987) as predictors of (female) stu-
persisting stereotypes because such role models might be rejected dents’ MSC formation.
as atypical (Miller et al., 2015; see also Richards & Hewstone,
2001).
STEM Gender Diversity and the Formation of
Expectancy-Value Framework Mathematics Self-Concept
The expectancy-value framework by Eccles and colleagues In a nutshell, both social role theory (Eagly & Wood, 2012) and
(1983) focuses on students’ achievement-related decisions (i.e., the expectancy-value framework (Eccles et al., 1983) provide
whether or not to continue taking mathematics; Wigfield & Eccles, important insights into how the degree of gender diversity in
1992) as well as on gender disparities in (mathematics-related) STEM occupations in a given society might impact students’
educational and occupational choices (Eccles, 1994). On the basis MSC. A society’s gender diversity in STEM jobs might shape
of historically older motivational frameworks (e.g., Atkinson, people’s beliefs and attitudes (including stereotypes). Parents,
1957), Eccles et al. (1983) developed and tested this framework, family members, teachers, or other socializers might in turn com-
which explicitly adopts a social– cognitive perspective in encom- municate gendered expectations and attitudes toward STEM (e.g.,
passing multiple factors from different ecological levels such as attitudes toward women’s occupational roles and abilities in math-
sociocultural factors, characteristics of significant adult figures, as ematics) to the younger generation of learners in and outside of
well as students’ characteristics (Eccles, 1994; Wigfield & Eccles, school. In addition, female students may perceive that mathematics
1992). At the core lies the assumption that students’ performance, is a “male” domain through direct (e.g., lack of female family
persistence, and educational and occupational choices are deter- members working in STEM fields) or indirect (e.g., media appear-
mined by their expectations of success and the value they attach to ance of men in STEM) observations. Taken together, these mech-
the various options available (Eccles, 1994, p. 587). According to anisms should have an influence on students’ MSC formation,
1122 NIEPEL, STADLER, AND GREIFF

which favors gender differences in MSC to the disadvantage of way, students from lower SES families might underestimate their
female students. In addition, previous cross-national research mathematics abilities in comparison with their peers (e.g., Preckel,
(Else-Quest et al., 2010; Miller et al., 2015) has shown that Goetz, Pekrun, & Kleine, 2008). On the country level, we added
occupational gender diversity is indeed linked with STEM-related the countries’ gross domestic product (GDP) and human develop-
beliefs and attitudes such as MSC. However, cross-national re- ment index (HDI) as covariates to account for cross-national
search that has comprehensively examined the combined value of differences in economic development as well as in average edu-
societal gender diversity in STEM, mathematics achievement, and cation and development. Both GDP and HDI have been shown to
gender to develop an overall better understanding of academic be related to individual differences in PISA across countries and
self-concept formation as well as gender differences in MSC has could thus influence our results (Cheung & Chan, 2008; Yu, 2012).
yet to be conducted. We designed the present study to fill this gap. Finally, we controlled for the general participation of women in the
countries’ workforce in all models to avoid potentially confound-
ing women’s occupations in STEM with the general presence of
The Present Study
women in the labor market.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

In the present study, we drew on large and representative inter-


This document is copyrighted by the American Psychological Association or one of its allied publishers.

national data sets from 23 countries to examine for the first time Method
whether including gender diversity in STEM occupations at a
national level would enhance statistical predictions of students’
MSC in a BFLPE framework. Previous research has already dem-
Participants
onstrated the relation between occupational gender diversity and Data on individual and school levels were drawn from the 2012
students’ STEM-related beliefs and attitudes (Else-Quest et al., cycle of PISA (OECD, 2014), an ongoing international project in
2010; Miller et al., 2015). However, the unique contribution of the which the academic competencies of 15-year-old students around
present study is that we integrated gender, individual achievement, the world are assessed. PISA has been run by the Organisation for
school-level achievement, and societal gender diversity in STEM Economic Cooperation and Development (OECD) since 2000. The
into one comprehensive model in which we statistically predicted reading, mathematics, and science skills of a new international
MSC. In other words, we integrated STEM gender diversity (op- sample of 15-year-olds are assessed every 3 years. In addition,
erationalized as the percentage of women in STEM research po- students are surveyed with regard to their beliefs and attitudes
sitions in a given society) as a statistical predictor at the country toward one particular subject (reading, mathematics, or science),
level into the BFLPE framework. The present study thereby ex- with the subject rotating across cycles. The samples are fully
tends existing cross-national research on the BFLPE framework representative of the participating countries with regard to a vast
(e.g., Marsh & Hau, 2003) by adding societal STEM gender range of sociodemographic factors (see OECD, 2014, for full
diversity, and it extends existing cross-national research on soci- details on recruitment, sampling, and procedures). The complete
etal STEM gender equity (e.g., Else-Quest et al., 2010) by adding data set is available for free online (http://www.oecd.org/pisa/).
mathematics achievement (on an individual and school level) as a The PISA 2012 data were collected from March, 2012 to June,
statistical predictor of students’ MSC. To this end, we investigated 2012 and released in May, 2013. Mathematics was the focus of
potential determinants of MSC formation on three different eco- the 2012 wave, which is why we selected the 2012 data set for
logical levels (individual, school, and country) simultaneously, and our analyses. In total, the PISA 2012 assessment included
thus, we were able to formulate the following four hypotheses: 485,490 students from 65 countries and economies. In this
First, we hypothesized that we would replicate the previous study, we were able to use data from 23 countries (see the
finding in which female students reported lower levels of MSC Measures section), leaving us with a total sample size of
(Hypothesis 1). Second, we expected mathematics achievement to 120,270 students. The countries and their respective sample
be positively related to MSC on the individual level (Hypothesis sizes are presented in Table 1.
2). Third, while controlling for gender and individual-level
achievement, we expected to detect a BFLPE; that is, we expected
Measures
to find that school-average achievement in mathematics would be
negatively associated with students’ MSC on the school level Women’s representation in science, technology, engineering,
(Hypothesis 3). Fourth, while controlling for gender as well as and mathematics (STEM gender diversity). Data for women’s
individual-level and school-level achievement, we expected to find occupational representation in STEM were obtained from the
an interaction between students’ gender and societal STEM gender UNESCO website (http://data.uis.unesco.org/) from the table la-
diversity. Specifically, we expected female students to report beled “Researchers by age, sector of employment and sex (Head
higher levels of MSC in societies where women’s participation in Count)” in the year 2011. The UNESCO data were released in July
STEM research positions is higher than in countries where pre- 2011. This variable was used as a proxy for STEM gender diver-
dominantly men work in such positions (Hypothesis 4). sity on a societal level. The values represent the percentage of
In order to eliminate alternative explanations for the relation women among individuals employed as researchers in STEM
between women’s representation in STEM and students’ MSC, we fields in 2011. The academic domains included in the data are
added other individual and national attributes as covariates to all based on the definition of “science” as well as “engineering,
our analyses. First, we controlled for students’ socioeconomic manufacturing and construction” by UNESCO (1997), including
status (SES) because many of the various relations found in all subdisciplines of the life sciences, physical sciences, mathe-
educational research can be explained, at least partially, by the matics and statistics, computing, engineering and engineering
effects of SES (for a detailed summary, see Jeynes, 2002). In this trades, manufacturing and processing, and architecture and build-
STEM GENDER DIVERSITY AND MATHEMATICS SELF-CONCEPT 1123

Table 1
Descriptive Statistics for All Main Variables by Country

Average MSC Average mathematics achievement


% of women
Country N Female Male d Female Male d in STEM

Bulgaria 5,282 ⫺.07 .00 .18ⴱ 442.47 441.87 ⫺.01 41.25


Colombia 9,073 .02 .13 .29ⴱ 372.63 401.46 .39ⴱ 26.65
Costa Rica 4,602 ⫺.02 .14 .37ⴱ 394.52 419.38 .38ⴱ 33.80
Czech Republic 5,327 ⫺.14 .02 .31ⴱ 513.66 525.96 .13ⴱ 20.60
Estonia 4,779 ⫺.03 .06 .20ⴱ 519.66 523.99 .05ⴱ 32.00
Greece 5,125 ⫺.03 .13 .34ⴱ 449.80 458.02 .09ⴱ 30.10
Hungary 4,810 ⫺.12 .02 .31ⴱ 479.52 491.90 .14ⴱ 22.20
Japan 6,351 ⫺.36 ⫺.17 .47ⴱ 525.37 545.77 .22ⴱ 8.85
Kazakhstan 5,808 .20 .19 .00 431.34 431.74 .01 47.60
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Korea 5,033 ⫺.25 ⫺.13 .32ⴱ 544.94 562.31 .18ⴱ 18.55


⫺.07 .24ⴱ
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Lithuania 4,618 .05 478.80 478.57 .00 38.45


Latvia 4,306 ⫺.07 .00 .15ⴱ 496.14 494.74 ⫺.02 40.00
Montenegro 4,744 ⫺.11 .01 .26ⴱ 405.57 407.23 .02 46.85
Malaysia 5,197 .06 .04 ⫺.05 426.39 418.76 ⫺.09ⴱ 48.20
Netherlands 4,460 ⫺.06 .13 .42ⴱ 513.60 522.33 .09ⴱ 21.60
Poland 4,607 ⫺.09 .02 .22ⴱ 519.08 522.23 .03 28.65
Portugal 5,722 ⫺.11 .02 .29ⴱ 478.34 490.75 .13ⴱ 36.10
Romania 5,074 ⫺.01 .04 .16ⴱ 444.54 447.09 .03 42.10
Russian Federation 5,231 .02 .08 .13ⴱ 483.68 481.73 ⫺.02 39.20
Serbia 4,684 ⫺.12 ⫺.03 .17ⴱ 441.89 453.71 .13ⴱ 43.90
Slovak Republic 4,678 ⫺.16 ⫺.01 .36ⴱ 480.73 490.12 .09ⴱ 35.85
Slovenia 5,911 ⫺.10 .04 .33ⴱ 483.39 485.40 .02 29.00
Turkey 4,848 ⫺.05 .00 .10ⴱ 441.79 455.57 .15ⴱ 30.35
Average 5,229.13 ⫺.07 .03 .23ⴱ 468.17 476.11 .11ⴱ 33.12
Note. MSC ⫽ Mathematics Self-Concept (M ⫽ 0, SD ⫽ 1); STEM ⫽ Science, Technology, Engineering, and Mathematics; d ⫽ Cohen’s d.

p ⬍ .01.

ing. For a full list of the subdisciplines, see UNESCO (1997). For were asked to rate their agreement with statements such as “Math
the professions that were included, UNESCO adopted the OECD’s is one of my best subjects” on a 4-point Likert scale. The OECD
(2015b) definition by classifying researchers as “professionals aggregated the answers to the five items into a latent factor score
engaged in the conception or creation of new knowledge (who (with higher scores indicating higher MSC; see OECD, 2014, for
conduct research and improve or develop concepts, theories, mod- further details), which we used in our analyses. The scale showed
els, techniques, instrumentation, software or operational methods)” high internal consistency with a median Cronbach’s alpha of .89
(UNESCO, 2017). For our analyses, we used head counts rather across all OECD countries (OECD, 2014).
than full-time equivalents. Head counts were chosen rather than Mathematics achievement. PISA assessed students in multi-
full-time equivalents because we were interested in the actual ple areas of mathematics achievement. Because mathematics was
number of women in STEM. Because women are more likely to the major domain in 2012, the assessment included a total of 131
work part time (Ñopo, Daza, & Ramos, 2012; Weichselbaumer & items covering the subdomains of Change and Relationships,
Winter-Ebmer, 2005), the use of full-time equivalents would un- Quantity, Shape and Space, and Uncertainty and Data. The scores
derestimate STEM gender diversity. Only the 23 countries that were scaled so that the OECD average in each domain (mathe-
participated in PISA and provided data on STEM gender diversity matics, reading, and science) was 500, and the standard deviation
were included in our analyses. was 100 (see Stanat et al., 2002). The PISA data file provides five
Mathematics self-concept (MSC). The scale that was used to sets of plausible values for latent mathematics ability derived with
assess MSC in PISA 2012 was directly adopted from PISA 2003 item response theory (Kaplan & Su, 2016; OECD, 2014). All
(OECD, 2005, 2014) and has been used in various studies since it analyses including mathematics achievement were thus calculated
was first published. It was based on the Self-Description Ques- separately for each of the five plausible values, as suggested by the
tionnaire (SDQ; Marsh, 1990), which is considered to be one of the OECD (2009). The results were then aggregated to produce the
best self-concept instruments available (e.g., Byrne, 2002). As for final results reported in this study. The scale showed a very good
all PISA questionnaires, the items were translated into the respec- median internal consistency of ␣ ⫽ .91 for the complete scale
tive languages for each participating country. Originally, eight (OECD, 2014). In order to estimate the BFLPE, we calculated the
MSC items were piloted in the international field trial, and five mean of students’ mathematics achievement separately for each
were retained for the main study. Detailed information on the item school in the data set.
response theory parameters used for scaling can be found in the Control variables.
PISA 2003 technical report (OECD, 2005). Socioeconomic status (SES). Information on individual dif-
The PISA 2012 Student Questionnaire that we used in this study ferences in students’ socioeconomic status was obtained from the
included the final five items for assessing students’ MSC. Students PISA 2012 data, which contained an index of economic, social,
1124 NIEPEL, STADLER, AND GREIFF

and cultural status (OECD, 2014). The variables comprising this OECD (2014), in all analyses, we used the individual student
index included: (a) home possessions, comprised of all items on weights available in the PISA data set to account for unequal
the “family wealth possessions,” “cultural possessions,” and participation probabilities and explicit oversampling for national
“home educational resources” scales as well as “books in the reporting purposes. Students from densely populated areas were
home” recoded into a four-level categorical variable (fewer than or thus given slightly lower weights than students from less populated
equal to 25 books, 26 –100 books, 101–500 books, more than 500 areas.
books), (b) parents’ highest occupational level, and (c) parents’
highest educational level expressed as years of schooling. The Results
scale is well established as an indicator of SES and has been used
in PISA since 2003. The reliability for the SES score ranges from
.56 to .77, and the internal consistency for the pooled OECD
Descriptive Statistics
sample with equally weighted country data is .69. For more details, Table 1 shows the sample size, average MSC for male and
please refer to the section on SES in the PISA 2012 technical female students, average mathematics achievement, and STEM
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

manual (OECD, 2014), which provides exhaustive details on the gender diversity for all 23 countries represented in this study. As
This document is copyrighted by the American Psychological Association or one of its allied publishers.

computation, distribution, and psychometric quality of the index. can be seen from the table, female students exhibited a lower
Gross domestic product (GDP). Countries’ GDP in 2011 was average MSC than male students in all but one country. Female
obtained from data provided by the World Bank (https://data students had a higher average MSC (M ⫽ .06) than male students
.worldbank.org/). The GDP is a monetary measure of the market (M ⫽ .04) only in Malaysia. These mean differences were statis-
value of all final goods and services produced in that year in U.S. tically significant for all comparisons but Kazakhstan and Malay-
dollars (for more information, see http://www.worldbank.org/). sia. Cohen (1988) provided guidelines for the interpretation of
Human development index (HDI). Countries’ HDI values in effect sizes in which Cohen’s d values of 0.20, 0.50, and 0.80 are
2011 were obtained from data provided by the United Nations considered small, medium, and large, respectively. We character-
Development Programme (UNDP; http://hdr.undp.org/en/data). ized effect sizes of d ⫽ 0.10 as negligible or close to zero (Hyde,
The HDI was created to emphasize that people and their capabil- 2005). Accordingly, the effect sizes were mostly small to medium,
ities should be the ultimate criteria for assessing the development with the strongest effects to the advantage of male students found
of a country rather than economic growth alone. The index is a in Costa Rica, Japan, Netherlands, and the Slovak Republic. On
measure of average achievement in key dimensions of human the other hand, female students had higher average mathematics
development: a long and healthy life (life expectancy at birth), achievement scores than male students in five countries (Bulgaria,
being knowledgeable (expected and mean years of schooling), and Lithuania, Latvia, Malaysia, and the Russian Federation). More-
having a decent standard of living (gross national income per over, the effect sizes of the mean differences in average mathe-
capita). The HDI is the geometric mean of the normalized indices matics achievement scores were considerably lower than for MSC
for each of the three dimensions (for more information, see Klug- with only two moderate differences to the advantage of male
man, 2011). students (Colombia and Costa Rica). The percentage of women in
General participation of women in the workforce. Values for STEM varied from 8.85% (Japan) to 48.20% (Malaysia). The
the general participation of women in the workforce in 2011 for average percentage of women in STEM across all 23 countries was
each country were also obtained from data provided by UNDP 33.12%. Differences between male and female students’ MSC and
(http://hdr.undp.org/en/data). The data represent the participation occupational gender diversity by country are also visually repre-
of women over the age of 15 years in the overall workforce in sented in Figure 1.
percent (for more information, see Klugman, 2011).
Main Analyses
Statistical Analysis
We calculated four successive LMMs to test the four hypotheses
All analyses were conducted in R 3.4.0 (R Core Team, 2013). specified above. The dependent variable for all models was stu-
To account for the nested structure of the data (students nested in dents’ MSC. All models included students’ socioeconomic status
schools nested in countries), we modeled students’ MSC using as well as countries’ GDP, HDI, and the general participation of
linear mixed-effects models (LMMs) with maximum likelihood women in the workforce as control variables. Table 2 summarizes
estimation (lme4 package; Bates, Mächler, Bolker, & Walker, the models.
2015). LMM is an extension of the linear model in which the linear In Model 1, students’ gender was added as a fixed effect to
predictor contains random effects in addition to the usual fixed estimate overall gender differences in MSC. As expected (Hypoth-
effects, making it well-suited for analyzing nested data (for an esis 1), male students exhibited a significantly higher MSC than
introduction to LMM, see Demidenko, 2013). Students’ school and female students (␤ ⫽ 0.12, p ⬍ .001, a small effect size according
country were both entered into the model as random effects and all to Cohen, 1988). Thus, this result supported Hypothesis 1, indi-
other variables as fixed effects. Model fit was estimated by com- cating that female students reported lower levels of MSC. Model
puting changes in the explained variance for both the fixed effects 1 represented the baseline model for all consecutive analyses
(marginal R2) and the combination of fixed and random factors (Rm2 ⫽ .05; Rc2 ⫽ .11).
(conditional R2) using the formulas suggested by Nakagawa and In the next step (Model 2), students’ individual mathematics
Schielzeth (2013). The different model fits were compared with achievement was added to Model 1 as a fixed effect. In this way,
␹2-difference tests. The magnitude of effect sizes was categorized the overall gender differences in MSC found in Model 1 could be
in accordance with Cohen’s (1988) guidelines. As suggested by the controlled for actual mathematics performance. Students’ individ-
STEM GENDER DIVERSITY AND MATHEMATICS SELF-CONCEPT 1125
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 1. Difference between male and female students’ MSC by country (%); STEM ⫽ science,
technology, engineering, and mathematics; MSC ⫽ mathematics self-concept (factor score); BGR ⫽
Bulgaria; COL ⫽ Colombia; CRI ⫽ Costa Rica; CZE ⫽ Czech Republic; EST ⫽ Estonia; GRC ⫽ Greece;
HUN ⫽ Hungary; JPN ⫽ Japan; KAZ ⫽ Kazakhstan; KOR ⫽ Korea; LTU ⫽ Lithuania; LVA ⫽ Latvia;
MNE ⫽ Montenegro; MYS ⫽ Malaysia; NLD ⫽ Netherlands; POL ⫽ Poland; PRT ⫽ Portugal; ROU ⫽
Romania; RUS ⫽ Russian Federation; SRB ⫽ Serbia; SVK ⫽ Slovak Republic; SVN ⫽ Slovenia; TUR ⫽
Turkey.

ual mathematics achievement was strongly related to students’ their MSC. As expected, the percentage of women in STEM was
MSC (␤ ⫽ 0.44, p ⬍ .001). This result was in line with Hypothesis incrementally related to students’ MSC with a medium effect size
2, indicating that mathematics achievement was positively related (␤ ⫽ 0.17, p ⬍ .001) over and above the other variables in the model.
to MSC on the individual level. Individual differences in mathe- Moreover, the interaction between the percentage of women in STEM
matics achievement did not fully explain the gender differences in and gender was also significantly related to students’ MSC
MSC, though. Gender remained significantly related to MSC (␤ ⫽ (␤ ⫽ ⫺0.11, p ⬍ .001), with a small to medium effect size according
0.09, p ⬍ .001) even after we controlled for mathematics achieve- to Cohen’s (1988) conventions for moderation effects. Thus, country-
ment (Rm2 ⫽ .19; Rc2 ⫽ .28). Model 2 represented the data substan- level STEM gender diversity moderated the differences in MSC
tially better than Model 1 (⌬␹2 ⫽ 9384.30, ⌬df ⫽ 1, p ⬍ .001). between male and female students (Rm2 ⫽ .19; Rc2 ⫽ .28). This model
Average mathematics performance within each school was con- represented the data better than Model 3 (⌬␹2 ⫽ 115.67, ⌬df ⫽ 1, p ⬍
trolled for in Model 3. In line with the predictions of the BFLPE, .001). To explore this moderation further, we modeled the data for
school-mean mathematics achievement was negatively related to male and female students separately. We found a small positive
students’ individual MSC with a small effect size (␤ ⫽ ⫺0.02, p ⬍ relation between STEM gender diversity and female students’ MSC
.001). Thus, students in schools with higher average mathematics (␤ ⫽ 0.08, p ⬍ .001) but no statistically significant relation for male
performances exhibited slightly lower MSC, in line with Hypoth- students (␤ ⫽ 0.02, p ⫽ .346). Thus, our results were in line with
esis 3 (Rm2 ⫽ .19; Rc2 ⫽ .28). This model represented the data Hypothesis 4, indicating that female students reported higher levels of
slightly better than Model 2 (⌬␹2 ⫽ 16.03, ⌬df ⫽ 1, p ⬍ .001). MSC in societies where women’s participation in STEM research
In Model 4, STEM gender diversity in a given country was positions is higher than in countries where predominantly men work
added as a moderator of the relation between students’ gender and in such positions.
1126 NIEPEL, STADLER, AND GREIFF

Table 2 such that female students reported higher MSC in countries where
Summaries of All Linear Mixed Models more women held research positions in STEM fields. Else-Quest et
al. (2010) also found some unexpected results, which did not
Model 1 Model 2 Model 3 Model 4 support the general claim that gender equity (as assessed with
Fixed effects (␤) different indices) generally favors gender similarities in mathemat-
Gender .12ⴱⴱ .09ⴱⴱ .09ⴱⴱ .09ⴱⴱ ics achievement, beliefs, and attitudes. For instance, the authors
Individual MACH — .44ⴱⴱ .44ⴱⴱ .44ⴱⴱ found that gender differences in MSC were actually larger (to the
School-level MACH — — ⫺.02ⴱⴱ ⫺.02ⴱⴱ
Women in STEM — — — .17ⴱⴱ
disadvantage of female students) in countries that exhibited higher
Interaction — — — ⫺.11ⴱⴱ scores on the Gender Empowerment Measure (GEM; United Na-
Random effects (VC) tions Development Programme, 1995) and the Gender Gap Index
School .03 .05 .05 .06 (GGI; Hausmann, Tyson, & Zahidi, 2007). The GEM and GGI are
Country .03 .04 .04 .04
2
Rm .05 .19 .19 .19
both composite indices that are based on gender ratios in different
Rc2 .11 .28 .28 .28 domains, such as health, education, economics, and politics (for
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

AIC 214750 205367 205353 205242 further details as well as for a discussion of these findings, see
This document is copyrighted by the American Psychological Association or one of its allied publishers.

BIC 214833 205460 205455 205362 Else-Quest et al., 2010). In fact, similar findings have been made
Note. Dependent variable ⫽ Math self-concept; MACH ⫽ Math achieve- in related research using broader composite gender equity indices
ment; STEM ⫽ Science, Technology, Engineering, and Mathematics; at the country level (e.g., on mathematics anxiety: Stoet et al.,
VC ⫽ Variance components; R2m ⫽ marginal R2; Rc2 ⫽ conditional R2; 2016; see also Stoet & Geary, 2018). In their critical review of the
AIC ⫽ Akaike information criterion; BIC ⫽ Bayesian information crite-
rion.
use of different gender equity indices, Else-Quest and Grabe
ⴱⴱ
p ⬍ .001. (2012) stated that researchers must be mindful of how they use the
broad range of composite and nation-level indicators available for
multiple domains of gender equity. Thereby, they should choose
Discussion theoretically meaningful and relevant domain-specific indicators,
which reflect the mechanisms under investigation whenever pos-
The formation of students’ academic self-concept does not sible (Else-Quest & Grabe, 2012, p. 139; see also Else-Quest &
occur in a vacuum but rather in a larger social and structural Hamilton, 2018). Therefore, in our analyses, we decided to include
environment and is thus influenced by the culture students live in STEM gender diversity, which represents the country-level
(Cheryan, Ziegler, Montoya, & Jiang, 2017; Stephens, Markus, & domain-specific indicator that comes closest to assessing the so-
Fryberg, 2012). The aim of the present study was therefore to ciocognitive mechanisms suggested by social role theory and the
identify the combined value of a society’s gender diversity in expectancy-value framework (see also Else-Quest et al., 2010, p.
STEM, achievement, and gender in statistically predicting stu- 123, for a discussion of the indicator they used: “gender equity in
dents’ MSC. To this end, we drew on large representative data sets research positions”).
and brought together two research strands that had not previously The present study extended Else-Quest et al.’s (2010) study on
been studied simultaneously: cross-national research on the cross-national gender differences in MSC in multiple ways. Most
BFLPE framework and cross-national research on STEM gender
important, whereas Else-Quest and colleagues examined MSC and
equity. In all our analyses, we controlled for students’ socioeco-
mathematics achievement as separate outcome variables, in the
nomic status as well as countries’ GDP, HDI, and the general
current study, we entered students’ mathematics achievement into
participation of women in the workforce. In line with our hypoth-
the regression equation as an additional predictor variable that
eses, we found that female students reported higher MSC in
could go beyond occupational gender diversity in explaining dif-
countries characterized by higher gender diversity in STEM re-
ferences in students’ MSC. Specifically, we embedded STEM
search positions. Thus, the present study reveals the meaningful
gender diversity into the BFLPE framework and thus studied the
extent to which gender diversity in STEM on a societal level is
combined statistical effect of gender, individual-level achieve-
related to female students’ MSC across 23 countries, regardless of
ment, school-level achievement, and country-level STEM gender
how well they perform in mathematics or which school they are
diversity on students’ MSC. Further, Else-Quest et al. used several
enrolled in. In showing this, our results suggest that societal STEM
gender inequity adversely impacts female students’ MSC forma- (broader composite as well as more domain-specific) indicators of
tion across various countries and that good mathematics perfor- societal gender equity including occupational gender diversity,
mance by itself does not protect female students from this adverse which was operationalized as the percentage of research positions
trend. held by women (Else-Quest et al., 2010). Different from Else-
Quest et al., we used a STEM-specific gender diversity indicator;
namely, occupational gender diversity in STEM research positions
Societal Gender Diversity in STEM
(according to Klugman, 2011). Finally, we used more recent data.
and Students’ MSC More precisely, for MSC, we drew on data from PISA 2012,
Else-Quest et al. (2010) already examined the relations between whereas Else-Quest et al. used 2003 PISA data. Thus, the current
various societal-level gender inequities and students’ mathematics study not only extended Else-Quest et al.’s analyses on cross-
achievement, beliefs, and attitudes, including MSC. In doing so, national gender gaps in MSC (as highlighted in the two previous
Else-Quest et al. (2010) found that occupational gender diversity points) but also replicated one of its major findings in principle
was the only predictor that was meaningfully and negatively (i.e., MSC on an individual level is related to societal occupational
related to gender gaps in students’ MSC across different countries gender diversity) almost one decade later.
STEM GENDER DIVERSITY AND MATHEMATICS SELF-CONCEPT 1127

Our results are in line with academic self-concept theory and individual mathematics achievement, which is typically conceived
research (see, e.g., Marsh, 2006) as well as with social comparison as the strongest single predictor of MSC formation (see Marsh,
theory (Festinger, 1954; Mussweiler, 2003). In the current study, 2006). Furthermore, we did not find any evidence that an increase
we replicated the BFLPE framework cross-nationally (after con- in STEM gender diversity could have an adverse impact on male
trolling for a number of covariates and gender), in revealing students’ MSC. Our findings thus help provide a better under-
positive relations between individual achievement and MSC on the standing of allegedly paradoxical findings on cross-national
one hand and negative relations between school-level achievement mathematics-related gender gaps (cf. Stoet & Geary, 2018). For
and students’ MSC (i.e., the BFLPE) on the other (see, e.g., Seaton instance, although Dutch students live in a country characterized
et al., 2009; see also Dai, Rinn, & Tan, 2013). Furthermore, our by high similarity between the genders in mathematics achieve-
results are in line with previous research on STEM gender inequity ment (d ⫽ 0.09), female Dutch students reported considerably
and STEM-related beliefs and attitudes (Else-Quest et al., 2010; lower MSC than the male students did (d ⫽ 0.42; see Table 1 and
Miller et al., 2015) as well as with psychological theories of Figure 1). Our results suggest that this pattern can be explained at
gendered socialization experiences as expressed in Eagly and least in part by the highly imbalanced gender distribution in STEM
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Wood’s (2012) social role theory or Eccles et al.’s (Eccles, 1994; jobs at a national level in the Netherlands in that merely one out of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Wigfield & Eccles, 1992) expectancy-value framework. Both five STEM researchers was female (21.6%).
highlight the impact of STEM role models or societal norms and
expectations, which were transmitted to future generations of
Limitations
working citizens through important socializers or the mass media.
Finally, our research also stands on the shoulders of and is in line Even though we interpreted our findings from the perspective of
with the work initiated by Baker and Jones (1993) on the gender existing theory and research and our data were temporally ordered
stratification hypothesis. The authors stated that gender inequity in by design (the PISA data were gathered at a later time point than
occupational and educational opportunity structures (which be- the UNESCO data), causal inferences are unwarranted. Arguably,
come manifested in a lack of women in STEM occupations) shapes the underlying relation between societal gender diversity in STEM
numerous socialization processes, which causes gender differences on the one hand and individual MSC on the other can be regarded
in mathematics-related outcomes (such as MSC) to the disadvan- as reciprocally related across time; that is, students’ MSC predicts
tage of female students. The gender stratification hypothesis is students’ STEM career decisions at a later time point (e.g., Guo et
thereby consistent with the expectancy-value framework and so- al., 2015), and the gender ratio in STEM fields also predicts
cial role theory in maintaining that students’ ability perceptions students’ MSC formation. In a similar vein, previous research has
were shaped by the greater cultural milieu but focus more explic- repeatedly found that mathematics achievement and MSC are
itly on gender inequities in occupational and educational opportu- reciprocally related across time; that is, higher achievement pre-
nity structures (Baker & Jones, 1993; see also Else-Quest et al., dicts a higher subsequent self-concept, and a higher self-concept
2010). predicts higher subsequent achievement (e.g., Marsh & Martin,
When it comes to the role of same-gender models for the 2011; Valentine, DuBois, & Cooper, 2004). Thus, the underlying
formation of female students’ STEM-related beliefs and attitudes, interdependencies of the variables we examined are arguably much
we should note that previous research has in fact not always been more complex than depicted in the present study. Consequently,
unanimous in stating that mere exposure to same-gender role future research should make an effort to further combine longitu-
models necessarily evokes beneficial effects (Gilmartin, Denson, dinal models from academic self-concept research (e.g., reciprocal
Li, Bryant, & Aschbacher, 2007). However, the exposure to female internal/external frame of reference model; Möller, Retelsdorf,
role models, who were perceived as being similar by other girls Köller, & Marsh, 2011; Niepel, Brunner, & Preckel, 2014) with
and women, have been shown to enhance girls’ and women’s theories and frameworks of gendered socialization (e.g., Eccles,
achievements, beliefs, and attitudes (Marx & Ko, 2012; Zirkel, 1994).
2002). Indeed, as recently summarized by Cheryan et al. (2017), In the present study, we used the percentage of women in
there is ever-growing empirical evidence that exposure to role research positions in STEM in a given country made available by
models to whom other women can relate do increase women’s UNESCO. This variable served as a proxy for societal STEM
interest in STEM, and that low gender diversity in STEM facili- gender diversity and thus closely reflects the mechanisms under
tates students’ perceptions that STEM fields are masculine cultures investigation as claimed by Eagly and Wood’s (2012) social role
that are more welcoming to men than women (Cheryan et al., theory and Eccles et al.’s (1983; Eccles, 1994; Wigfield & Eccles,
2017). Thus, a lack of societal STEM gender diversity mirrors the 1992) expectancy-value framework. We found that STEM gender
prevailing masculine culture in STEM occupations, thereby pre- diversity seems to play an important role in the formation of
venting many girls and women (and arguably also counterstereo- students’ MSC in addition to the role that their actual mathematics
typical boys and men) from developing competence self- achievement plays. Our findings thus make a convincing case for
perceptions in STEM that are as high as (most of) their male adding STEM gender diversity at the country level to established
counterparts. models on academic self-concept formation (e.g., the original
In sum, the current study extends the understanding of gender two-level BFLPE framework) in order to better understand the
differences in MSC, demonstrating that even after a number of existing gender gap in MSC that functions to the disadvantage of
important variables and covariates were controlled for, national female students. However, we still found MSC variability that
differences in societal STEM gender diversity emerged as a gen- could not be statistically explained by the variables examined in
eralizable and meaningful statistical predictor of female students’ the present study. To further enhance the predictive validity of our
MSC across 23 countries. However, it was still outranked by model, future research could conduct more fine-grained analyses
1128 NIEPEL, STADLER, AND GREIFF

to differentiate among diverse STEM fields (see Cheryan et al., gender diversity in STEM on a societal level in order to help girls
2017, analysis on gender differences within STEM fields); for and women develop an adequate sense of their competences, or in
example, in examining how the percentages of women employed the words of Sally Ride (2012), physicist and astronaut: “You can’t
in jobs involving mathematics and statistics (or, e.g., in computer be what you can’t see.”
science) in a given country are related to students’ MSC (or to their
computer-science self-concept, respectively). Furthermore, future
research should also consider reanalyzing our models with more References
comprehensive STEM gender diversity data including STEM pro- Arens, A. K., Yeung, A. S., Craven, R. G., & Hasselhorn, M. (2011). The
fessions that are located outside of research positions. Moreover, twofold multidimensionality of academic self-concept: Domain speci-
women tend to be underrepresented in most areas of research ficity and separation between competence and affect components. Jour-
beyond STEM (see, e.g., Pashkova et al., 2013, for gender dispar- nal of Educational Psychology, 103, 970 –981. http://dx.doi.org/10
ities in medical research). Therefore, it would be highly relevant to .1037/a0025047
investigate how our findings generalize to other fields. Atkinson, J. W. (1957). Motivational determinants of risk-taking behavior.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Finally, we analyzed countries that participated in the 2012 Psychological Review, 64, 359 –372. http://dx.doi.org/10.1037/
This document is copyrighted by the American Psychological Association or one of its allied publishers.

PISA cycle and for which we had information on the percentage of h0043445
women in STEM research positions provided by UNESCO. The Baker, D. P., & Jones, D. P. (1993). Creating gender equality: Cross-
national gender stratification and mathematical performance. Sociology
resulting 23 countries were located in mainland Europe, Asia, and
of Education, 66, 91–103. http://dx.doi.org/10.2307/2112795
the Americas. As such, but inevitably because we drew on PISA
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting linear
data, our data were characterized by an imbalanced representation mixed-effects models using lme4. Journal of Statistical Software, 67,
of countries by continent (see http://www.oecd.org/pisa/aboutpisa/ 1– 48. http://dx.doi.org/10.18637/jss.v067.i01
pisa-2012-participants.htm). For some other countries, no match- Bonsang, E., Skirbekk, V., & Staudinger, U. M. (2017). As you sow, so
ing UNESCO data were available to us. Future cross-cultural shall you reap: Gender-role attitudes and late-life cognition. Psycho-
research is therefore indicated here. logical Science, 28, 1201–1213. http://dx.doi.org/10.1177/
0956797617708634
Byrne, B. M. (2002). Validating the measurement and structure of self-
Implications and Conclusion concept: Snapshots of past, present, and future research. American
Female students tend to underestimate their mathematical abil- Psychologist, 57, 897–909. http://dx.doi.org/10.1037/0003-066X.57.11
ities when they live in societies in which female scientist are .897
Cheryan, S., Ziegler, S. A., Montoya, A. K., & Jiang, L. (2017). Why are
scarce. Our results therefore suggest that enhancing mathematics
some STEM fields more gender balanced than others? Psychological
achievement per se is not sufficient for improving female students’
Bulletin, 143, 1–35. http://dx.doi.org/10.1037/bul0000052
MSC in societies with gender inequities in STEM. Barriers to Cheung, H. Y., & Chan, A. W. H. (2008). Understanding the relationships
gender diversity in STEM need to be redressed at all levels—not among PISA scores, economic growth and employment in different
just in school—to ensure that efforts to move toward equity are sectors. Research in Education, 80, 93–106. http://dx.doi.org/10.7227/
successful. Interventions must thus target various, intertwined lev- RIE.80.8
els comprising structural conditions, media coverage, as well as Cohen, J. (1988). Statistical power analysis for the behavioral sciences
the beliefs and attitudes of socializers and students. Along these (2nd ed.). Hillsdale, NJ: Erlbaum.
lines, Wang and Degol (2017) recently carved out various Dai, D. Y., Rinn, A. N., & Tan, X. (2013). When the big fish turns small:
evidence-based recommendations for practitioners and policymak- Effects of participating in gifted summer programs on academic self-
ers to improve STEM gender diversity and decidedly demanded concepts. Journal of Advanced Academics, 24, 4 –26. http://dx.doi.org/
that efforts be increased, inter alia, to combat negative gender 10.1177/1932202X12473425
Demidenko, E. (Ed.). (2013). Wiley series in probability and statistics:
STEM stereotypes or to increase the number of female role models
Mixed models. Hoboken, NJ: Wiley.
in STEM. For instance, at home and in school, parents and teachers
Eagly, A. H., Makhijani, M. G., & Klonsky, B. G. (1992). Gender and the
could communicate the idea that equivalent STEM performance evaluation of leaders: A meta-analysis. Psychological Bulletin, 111,
has been achieved between the genders and that STEM gender 3–22. http://dx.doi.org/10.1037/0033-2909.111.1.3
diversity has increased in recent years in most societies (Wang & Eagly, A. H., & Wood, W. (1999). The origins of sex differences in human
Degol, 2017). Objects stereotypically associated with male inter- behavior: Evolved dispositions versus social roles. American Psycholo-
ests should be eliminated from STEM classrooms so that students gist, 54, 408 – 423. http://dx.doi.org/10.1037/0003-066X.54.6.408
learn not to think of STEM as a typically “male” domain (Cheryan Eagly, A. H., & Wood, W. (2012). Social role theory. In P. van Lange,
et al., 2017). In addition, the mass media should strive to create A. W. Kruglanski, & E. T. Higgins (Eds.), Handbook of theories of
more positive role models involving female STEM researchers so social psychology (pp. 458 – 476). Thousand Oaks, CA: Sage. http://dx
that female students will more often encounter realistic images of .doi.org/10.4135/9781446249222.n49
Eagly, A. H., & Wood, W. (2016). Social role theory of sex differences. In A.
successful women in STEM to whom these students can relate
Wong, M. Wickramasinghe, R. Hoogland, & N. A. Naples (Eds.), The
(Wang & Degol, 2017). Moreover, policy and educational stake-
Blackwell encyclopedia of gender and sexuality studies (pp. 1–3). Chic-
holders must work to achieve STEM gender diversity in promoting ester, UK: Wiley and Sons. http://dx.doi.org/10.1002/9781118663219
network opportunities for women in STEM (e.g., career fairs, .wbegss183
career days at school, mentoring programs at universities) aiming Eccles, J. S. (1994). Understanding women’s educational and occupational
to install a support system as well as a sense of connectedness for choices: Applying the Eccles et al. model of achievement-related choic-
women pursuing careers in STEM fields (Wang & Degol, 2017). es. Psychology of Women Quarterly, 18, 585– 609. http://dx.doi.org/10
To conclude, our results strongly support the enhancement of .1111/j.1471-6402.1994.tb01049.x
STEM GENDER DIVERSITY AND MATHEMATICS SELF-CONCEPT 1129

Eccles, J. S. (2009). Who am I and what am I going to do with my life? Klugman, J. (2011). Human development report 2011. Sustainability and
Personal and collective identities as motivators of action. Educa- equity: A better future for all. UNDP-HDRO human development re-
tional Psychologist, 44, 78 – 89. http://dx.doi.org/10.1080/ ports. New York, NY: Palgrave Macmillan. Retrieved from https://ssrn
00461520902832368 .com/abstract⫽2294671
Eccles, J. S., Adler, T. F., Futterman, R., Goff, S. B., Kaczala, C. M., Maccoby, E. E. (1966). The development of sex differences. Palo Alto, CA:
Meece, J. L., & Midgley, C. (1983). Expectancies, values and academic Stanford University Press.
behaviors. In J. T. Spence (Ed.), Achievement and achievement motives: Marsh, H. W. (1987). The big-fish-little-pond effect on academic self-
Psychological and sociological approaches (pp. 75–146). San Fran- concept. Journal of Educational Psychology, 79, 280 –295. http://dx.doi
cisco, CA: W. H. Freeman. .org/10.1037/0022-0663.79.3.280
Else-Quest, N. M., & Grabe, S. (2012). The political is personal. Psychol- Marsh, H. W. (1990). Self Description Questionnaire (SDQ) II: A theoret-
ogy of Women Quarterly, 36, 131–144. http://dx.doi.org/10.1177/ ical and empirical basis for the measurement of multiple dimensions of
0361684312441592 adolescent self-concept: An interim test manual and a research mono-
Else-Quest, N. M., & Hamilton, V. (2018). Measurement and analysis of graph. San Antonio, TX: The Psychological Corporation.
nation-level gender equity in the psychology of women. In C. B. Travis Marsh, H. W. (2006). Self-concept theory, measurement and research into
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

& J. W. White (Eds.), Handbook of the psychology of women: Vol. 2. practice: The role of self-concept in educational psychology. Leicester,
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Perspectives on women’s private and public lives (pp. 545–563). Wash- UK: British Psychological Society.
ington, DC: American Psychological Association. http://dx.doi.org/10 Marsh, H. W., & Hau, K.-T. (2003). Big-fish-little-pond effect on aca-
.1037/0000060-029 demic self-concept. A cross-cultural (26-country) test of the negative
Else-Quest, N. M., Hyde, J. S., & Linn, M. C. (2010). Cross-national effects of academically selective schools. American Psychologist, 58,
patterns of gender differences in mathematics: A meta-analysis. Psycho- 364 –376. http://dx.doi.org/10.1037/0003-066X.58.5.364
logical Bulletin, 136, 103–127. http://dx.doi.org/10.1037/a0018053 Marsh, H. W., & Martin, A. J. (2011). Academic self-concept and aca-
Festinger, L. (1954). A theory of social comparison processes. Human demic achievement: Relations and causal ordering. British Journal of
Relations, 7, 117–140. http://dx.doi.org/10.1177/001872675400700202 Educational Psychology, 81, 59 –77. http://dx.doi.org/10.1348/
Frome, P. M., & Eccles, J. S. (1998). Parents’ influence on children’s 000709910X503501
achievement-related perceptions. Journal of Personality and Social Psy- Marsh, H. W., & Yeung, A. S. (1998). Longitudinal structural equation
chology, 74, 435– 452. http://dx.doi.org/10.1037/0022-3514.74.2.435 models of academic self-concept and achievement: Gender differences
Gilmartin, S., Denson, N., Li, E., Bryant, A., & Aschbacher, P. (2007). in the development of math and English constructs. American Educa-
Gender ratios in high school science departments: The effect of percent tional Research Journal, 35, 705–738. http://dx.doi.org/10.3102/
female faculty on multiple dimensions of students’ science identities. 00028312035004705
Journal of Research in Science Teaching, 44, 980 –1009. http://dx.doi Marx, D. M., & Ko, S. J. (2012). Superstars “like” me: The effect of role
.org/10.1002/tea.20179 model similarity on performance under threat. European Journal of
Guiso, L., Monte, F., Sapienza, P., & Zingales, L. (2008). Culture, gender, Social Psychology, 42, 807– 812. http://dx.doi.org/10.1002/ejsp.1907
and math. Science, 320, 1164 –1165. http://dx.doi.org/10.1126/science Miller, D. I., Eagly, A. H., & Linn, M. C. (2015). Women’s representation
.1154094 in science predicts national gender-science stereotypes: Evidence from
Guo, J., Parker, P. D., Marsh, H. W., & Morin, A. J. S. (2015). Achieve- 66 nations. Journal of Educational Psychology, 107, 631– 644. http://dx
ment, motivation, and educational choices: A longitudinal study of .doi.org/10.1037/edu0000005
expectancy and value using a multiplicative perspective. Developmental Möller, J., Retelsdorf, J., Köller, O., & Marsh, H. W. (2011). The recip-
Psychology, 51, 1163–1176. http://dx.doi.org/10.1037/a0039440 rocal internal/external frame of reference model. American Educational
Hausmann, R., Tyson, L. D., & Zahidi, S. (2007). The global gender gap Research Journal, 48, 1315–1346. http://dx.doi.org/10.3102/
report 2007. Geneva, Switzerland: World Economic Forum. 0002831211419649
Huguet, P., Dumas, F., Marsh, H., Régner, I., Wheeler, L., Suls, J., . . . Mussweiler, T. (2003). Comparison processes in social judgment: Mech-
Nezlek, J. (2009). Clarifying the role of social comparison in the anisms and consequences. Psychological Review, 110, 472– 489. http://
big-fish-little-pond effect (BFLPE): An integrative study. Journal of dx.doi.org/10.1037/0033-295X.110.3.472
Personality and Social Psychology, 97, 156 –170. http://dx.doi.org/10 Nakagawa, S., & Schielzeth, H. (2013). A general and simple method for
.1037/a0015558 obtaining R2 from generalized linear mixed-effects models. Methods in
Hyde, J. S. (2005). The gender similarities hypothesis. American Psychol- Ecology and Evolution, 4, 133–142. http://dx.doi.org/10.1111/j.2041-
ogist, 60, 581–592. http://dx.doi.org/10.1037/0003-066X.60.6.581 210x.2012.00261.x
Hyde, J. S., Lindberg, S. M., Linn, M. C., Ellis, A. B., & Williams, C. C. Niepel, C., Brunner, M., & Preckel, F. (2014). The longitudinal interplay
(2008). Gender similarities characterize math performance. Science, of students’ academic self-concepts and achievements within and across
321, 494 – 495. http://dx.doi.org/10.1126/science.1160364 domains: Replicating and extending the reciprocal internal/external
Jacobs, J., Davis-Kean, P., Bleeker, M., Eccles, J., & Malanchuk, O. frame of reference model. Journal of Educational Psychology, 106,
(2005). “I can, but I don’t want to”: The impact of parents, interests, and 1170 –1191. http://dx.doi.org/10.1037/a0036307
activities on gender differences in math. In A. Gallagher & J. Kaufman Ñopo, H., Daza, N., & Ramos, J. (2012). Gender earning gaps around the
(Eds.), Gender differences in mathematics: An integrative psychological world: A study of 64 countries. International Journal of Manpower, 33,
approach (pp. 73–98). New York, NY: Cambridge University Press. 464 –513. http://dx.doi.org/10.1108/01437721211253164
Jeynes, W. H. (2002). Examining the effects of parental absence on the OECD. (2005). PISA 2003 technical report. Paris, France: OECD Pub-
academic achievement of adolescents: The challenge of controlling for lishing. Retrieved from http://www.oecd.org/education/school/
family income. Journal of Family and Economic Issues, 23, 189 –210. programmeforinternationalstudentassessmentpisa/35188570.pdf
http://dx.doi.org/10.1023/A:1015790701554 OECD. (2009). PISA data analysis manual: SPSS (2nd ed.). Paris, France:
Kaplan, D., & Su, D. (2016). On matrix sampling and imputation of context OECD Publishing.
questionnaires with implications for the generation of plausible values in OECD. (2014). PISA 2012 technical report. Paris, France: OECD Publish-
large-scale assessments. Journal of Educational and Behavioral Statistics, ing. Retrieved from https://www.oecd.org/pisa/pisaproducts/PISA-
41, 57– 80. http://dx.doi.org/10.3102/1076998615622221 2012-technical-report-final.pdf
1130 NIEPEL, STADLER, AND GREIFF

OECD. (2015a). The ABC of gender equality in education. Paris, France: Developmental Psychology, 38, 615– 630. http://dx.doi.org/10.1037/
OECD Publishing. Retrieved from http://www.oecd.org/education/the- 0012-1649.38.4.615
abc-of-gender-equality-in-education-9789264229945-en.htm Trautwein, U., & Möller, J. (2016). Self-concept: Determinants and con-
OECD. (2015b). Frascati manual 2015. Paris, France: OECD Publishing. sequences of academic self-concept in school contexts. In A. A. Lipn-
Retrieved from http://www.oecd.org/publications/frascati-manual-2015- evich, F. Preckel, & R. D. Roberts (Eds.), Psychosocial skills and school
9789264239012-en.htm systems in the 21st century: The Springer series on human exceptionality
Pashkova, A. A., Svider, P. F., Chang, C. Y., Diaz, L., Eloy, J. A., & Eloy, (pp. 187–214). Cham, Switzerland: Springer International Publishing.
J. D. (2013). Gender disparity among U.S. anaesthesiologists: Are http://dx.doi.org/10.1007/978-3-319-28606-8_8
women underrepresented in academic ranks and scholarly productivity? UNESCO. (1997). International standard classification of education:
Acta Anaesthesiologica Scandinavica, 57, 1058 –1064. http://dx.doi.org/ ISCED 1997. Paris, France: Author. Retrieved from http://www.unesco
10.1111/aas.12141 .org/education/information/nfsunesco/doc/isced_1997.htm
Preckel, F., Goetz, T., Pekrun, R., & Kleine, M. (2008). Gender differences UNESCO. (2017). Concepts and definitions. Retrieved from https://
in gifted and average-ability students. Gifted Child Quarterly, 52, 146 – unstats.un.org/sdgs/metadata/files/Metadata-09-05-02.pdf
159. http://dx.doi.org/10.1177/0016986208315834 United Nations Development Programme. (1995). Human development
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

R Core Team. (2013). R: A language and environment for statistical report 1995. New York, NY: Oxford University Press.
computing (Version 3.4.0) [Computer software]. Vienna, Austria: R
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Valentine, J. C., DuBois, D. L., & Cooper, H. (2004). The relation between
Foundation for Statistical Computing. Retrieved from http://www self-beliefs and academic achievement: A meta-analytic review. Edu-
.R-project.org/ cational Psychologist, 39, 111–133. http://dx.doi.org/10.1207/
Richards, Z., & Hewstone, M. (2001). Subtyping and subgrouping: Pro- s15326985ep3902_3
cesses for the prevention and promotion of stereotype change. Person-
Wang, M.-T., & Degol, J. L. (2017). Gender gap in science, technology,
ality and Social Psychology Review, 5, 52–73. http://dx.doi.org/10.1207/
engineering, and mathematics (STEM): Current knowledge, implica-
S15327957PSPR0501_4
tions for practice, policy, and future directions. Educational Psychology
Ride, S. (2012). Interview for Harvard Business Review. Retrieved from
Review, 29, 119 –140. http://dx.doi.org/10.1007/s10648-015-9355-x
https://hbr.org/2012/09/sally-ride
Weichselbaumer, D., & Winter-Ebmer, R. (2005). A meta-analysis of the
Seaton, M., Marsh, H. W., & Craven, R. G. (2009). Earning its place as a
international gender wage gap. Journal of Economic Surveys, 19, 479 –
pan-human theory: Universality of the big-fish-little-pond effect across
511. http://dx.doi.org/10.1111/j.0950-0804.2005.00256.x
41 culturally and economically diverse countries. Journal of Educa-
Wheeler, L., & Miyake, K. (1992). Social comparison in everyday life.
tional Psychology, 101, 403– 419. http://dx.doi.org/10.1037/a0013838
Journal of Personality and Social Psychology, 62, 760 –773. http://dx
Shavelson, R. J., Hubner, J. J., & Stanton, G. C. (1976). Self-concept:
Validation of construct interpretations. Review of Educational Research, .doi.org/10.1037/0022-3514.62.5.760
46, 407– 441. http://dx.doi.org/10.3102/00346543046003407 Wigfield, A., & Eccles, J. S. (1992). The development of achievement task
Stanat, P., Artelt, C., Baumert, J., Klieme, E., Neubrand, M., Prenzel, M., values: A theoretical analysis. Developmental Review, 12, 265–310.
& Weiß, M. (2002). PISA 2000: Overview of the study. Retrieved from http://dx.doi.org/10.1016/0273-2297(92)90011-P
https://www.mpib-berlin.mpg.de/Pisa/PISA-2000_Overview.pdf Witt, M. G., & Wood, W. (2010). Self-regulation of gendered behavior in
Stephens, N. M., Markus, H. R., & Fryberg, S. A. (2012). Social class everyday life. Sex Roles, 62, 635– 646. http://dx.doi.org/10.1007/
disparities in health and education: Reducing inequality by applying a s11199-010-9761-y
sociocultural self model of behavior. Psychological Review, 119, 723– Yu, C. H. (2012). Beyond gross national product: An exploratory study of
744. http://dx.doi.org/10.1037/a0029028 the relationship between Program for International Student Assessment
Stoet, G., Bailey, D. H., Moore, A. M., & Geary, D. C. (2016). Countries scores and well-being indices. Review of European Studies. Advance
with higher levels of gender equality show larger national sex differ- online publication. http://dx.doi.org/10.5539/res.v4n5p119
ences in mathematics anxiety and relatively lower parental mathematics Zirkel, S. (2002). Is there a place for me? Role models and academic
valuation for girls. PLoS ONE, 11(4), e0153857. http://dx.doi.org/10 identity among White students and students of color. Teachers College
.1371/journal.pone.0153857 Record, 104, 357–376. http://dx.doi.org/10.1111/1467-9620.00166
Stoet, G., & Geary, D. C. (2018). The gender-equality paradox in science,
technology, engineering, and mathematics education. Psychological Sci-
ence, 29, 581–593. http://dx.doi.org/10.1177/0956797617741719 Received November 2, 2017
Tenenbaum, H. R., & Leaper, C. (2002). Are parents’ gender schemas Revision received December 1, 2018
related to their children’s gender-related cognitions? A meta-analysis. Accepted December 16, 2018 䡲

You might also like