Full Terms & Conditions of access and use can be found athttps://www.tandfonline.com/action/journalInformation?journalCode=nere20

Educational Research and EvaluationAn International Journal on Theory and Practice

ISSN: 1380-3611 (Print) 1744-4187 (Online) Journal homepage: https://www.tandfonline.com/loi/nere20

Gender norms and STEM: the importance offriends for stopping leakage from the STEMpipeline

Maaike van der Vleuten, Stephanie Steinmetz & Herman van de Werfhorst

To cite this article: Maaike van der Vleuten, Stephanie Steinmetz & Herman van de Werfhorst(2019): Gender norms and STEM: the importance of friends for stopping leakage from the STEMpipeline, Educational Research and Evaluation, DOI: 10.1080/13803611.2019.1589525

To link to this article: https://doi.org/10.1080/13803611.2019.1589525

© 2019 The Author(s). Published by InformaUK Limited, trading as Taylor & FrancisGroup

Published online: 14 Mar 2019.

Submit your article to this journal

Article views: 113

View Crossmark data

Gender norms and STEM: the importance of friends forstopping leakage from the STEM pipelineMaaike vander Vleuten a, Stephanie Steinmetz b and Herman van de Werfhorst b

aDepartment of Sociology, Radboud University, Nijmegen, the Netherlands; bDepartment of Sociology,University of Amsterdam, Amsterdam, the Netherlands

ABSTRACTMore women are now entering male-dominated fields, yet, science,technology, engineering, and mathematics (STEM) remaindominated by men. We examined the association between boys’and girls’ STEM choices after secondary education and friends’gender norms, and whether pressure to conform to traditionalgender norms differs depending on the gender composition ofthe friend group. Drawing on 3 waves of longitudinal data (N =744) from the Netherlands, our sample consists of adolescents inSTEM trajectories in secondary education. Their retention in STEMafter secondary education gives us a better understanding ofgender-specific “leakage” from the STEM pipeline. We found thatgirls’ likelihood of choosing STEM decreased drastically whenfriends had more traditional gender norms. Friends withtraditional gender norms had less effect on boys. Nonetheless,boys with only same-sex friends were more likely to enter STEM.Our findings indicate that an environment with gender-normativeideas pushes girls out of the STEM pipeline.

ARTICLE HISTORYReceived 7 June 2018Accepted 27 February 2019

KEYWORDSSTEM; gender; field of study;friends’ gender norms;gender composition

Introduction

Women have made tremendous inroads into higher education and the labour market.Nevertheless, women are still underrepresented in fields involving science, technology,engineering, and mathematics (STEM) (Charles & Bradley, 2009; Mann & DiPrete, 2013).Women’s tendency to drop out of STEM fields during their educational and occupationalcareers has been referred to as the “leaky pipeline” (Mann & DiPrete, 2013; Morgan, Gelb-giser, & Weeden, 2013; Riegle-Crumb, King, Grodsky, & Muller, 2012). This leaking of thepipeline is a problem, as women’s participation in STEM fields is deemed crucial for econ-omic innovation and productivity (Corbett & Hill, 2015). Moreover, many talented girls aremissing out on the higher professional status and earnings offered by STEM careers(Hill, Corbett, & Rose, 2010; Reimer & Steinmetz, 2009).

Explanations of gender differences in STEM choices have traditionally focused on aca-demic performance and/or ability, but ample research shows that the gender gap in STEMcannot be attributed solely to such disparities (Ceci, Williams, & Barnett, 2009; Hyde,

© 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis GroupThis is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License(http://creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in anymedium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way.

CONTACT Maaike van der Vleuten M.vandervleuten@maw.ru.nl

EDUCATIONAL RESEARCH AND EVALUATIONhttps://doi.org/10.1080/13803611.2019.1589525

Lindberg, Linn, Ellis, & Williams, 2008; Mann & DiPrete, 2013). Research has therefore goneon to examine other characteristics (for an overview, see Eccles, 2011). Particularly, gender-role socialization practices within adolescents’ environment have been hypothesized asrelevant (Alon & DiPrete, 2015; Correll, 2001; Hyde & Mertz, 2009; Van der Vleuten,2018). Gender-normative ideas hold STEM fields to be congruent with male and incongru-ent with female gender-role behaviour (Buck, Plano Clark, Leslie-Pelecky, Lu, & Cerda-Lizarraga, 2008; Hilliard & Liben, 2010). If adolescents base their educational decisionson these norms, gender differences in STEM fields will persist. While researchers haveexamined the consequences of parental socialization practices for children’s field ofstudy choices (Dryler, 1998; Van der Vleuten, Jaspers, Maas, & Van der Lippe, 2018), littleresearch has examined gender norms among friends in the school context. This line ofstudy is important, however, as there is evidence that during adolescence friends growmore important compared to parents or teachers (Ganotice & King, 2014). Moreover, class-mates are instrumental in many educational outcomes, such as grade point average, mathtest scores, and algebra placement (Cook, Deng, & Morgano, 2007); reading test scores(Legewie & DiPrete, 2012); and college aspirations and attendance (Hallinan & Williams,1990). Less is known about how friends affect STEM choices. Although the influence offriends’ gender norms on field of study choices is often acknowledged, it is assumedrather than formally tested (Crosnoe, Riegle-Crumb, Field, Frank, & Muller, 2008; Franket al., 2008; Gabay-Egozi, Shavit, & Yaish, 2015; Riegle-Crumb, Farkas, & Muller, 2006).The aim of our study was to examine whether and to what extent the gender norms ofclass friends matter in boys’ and girls’ STEM choices after secondary education.

An additional puzzle is whether the association between friends’ gender norms andSTEM choices differs depending on the gender composition of the friend group. Evidencein this regard is contradictory. Boys and girls have been found to be more likely to complywith gender-typed norms both when they have more same-sex friends (Robnett & Leaper,2013) and when they have more opposite-sex friends (Dasgupta, McManus Scircle, & Hun-singer, 2015; Schneeweis & Zweimüller, 2012). Most of these studies, however, againassume rather than test the role of gender norms (i.e., Riegle-Crumb et al., 2006; Schnee-weis & Zweimüller, 2012). We tested how the gender composition of the class friend groupinteracts with the gender normativity of the friend group for girls and for boys.

We used longitudinal data collected from adolescents in secondary school in the Neth-erlands in 2011–2012 (age 16) and after secondary education in 2014 and 2015 (ages 18–19;N = 744). Dutch students must make a STEM-related trajectory choice in secondary edu-cation (in 9th or 10th grade; ages 14–15). Such a choice prepares them to continue ona STEM path after secondary education. To increase our understanding of the gender-specific leakage from the STEM pipeline, we focused specifically on students who chosea STEM-related trajectory in secondary education. In other words, we examined therelation between friends’ characteristics and field of study choices after secondary edu-cation specifically for students in the STEM pipeline. However, like most research onpeer influence, we cannot make causal inferences about whether friends influence eachother to enter STEM fields. Indeed, friend effects may result in part from a tendency tochoose friends with certain characteristics, like particular gender norms and STEM inter-ests. For example, adolescents who are interested in STEM might seek friends whoshare this interest, as shown by recent experimental evidence (Dasgupta et al., 2015).Our study therefore examined whether friends’ characteristics were associated with

2 M. VAN DER VLEUTEN ET AL.

STEM choices after secondary education. The longitudinal nature of the data allowed us toevaluate the relation between friends’ characteristics and the actual fields of study adoles-cents entered, instead of STEM aspirations (Dasgupta et al., 2015), intentions (Riegle-Crumb & Morton, 2017), or preferences (Francis, Hutchings, Archer, & Melling, 2003), likemost previous research. This is an important advance as behaviours (such as enteringSTEM) are constrained more by cultural norms than by attitudes or orientations (Alon &DiPrete, 2015). This implies that the fields of study students enter might be moreclosely aligned with friends’ gender norms than with their own aspirations, intentions,or preferences among fields.

There is another noteworthy feature of the educational system in the Netherlands:Whereas in other countries STEM choices are made mostly in tertiary education (e.g., inthe choice of college major in the USA; Legewie, & DiPrete, 2014), students in the Nether-lands who intend to continue their education after secondary school must choose a fieldof study, regardless of the educational level they followed in secondary education. Thedata from the Netherlands thus provide the unique opportunity to evaluate STEMchoices for a wider group than only students in tertiary education.

Theory

The gender normativity of the environment

The environment in which boys and girls are socialized is crucial in shaping cultural beliefsabout “appropriate” male and female behaviour. Traditional gender norms hold that girlsare more talented verbally, have better social skills (communal), and are more focused onchildren and family (Konrad, Ritchie, Lieb, & Corrigall, 2000). By contrast, boys are held tobe good at mathematics and science, agentic (e.g., in acquiring skills and competence),and more focused on financial gain and status (Diekman, Brown, Johnston, & Clark,2010). STEM fields are therefore often considered a natural match for male gender-rolebehaviour and incongruent with female gender-role behaviour (Buck et al., 2008;Cheryan, Plaut, Handron, & Hudson, 2013; Hill et al., 2010). Living in an environment inwhich traditional gender-role beliefs are prevalent may push young women away fromSTEM fields, while drawing boys in towards these fields (Charles & Bradley, 2009;Legewie & DiPrete, 2014; Morgan et al., 2013). In particular, friends may play a crucialrole, by approving or disapproving of the adoption of gender-conforming behaviour.Studies show that friends do reinforce gender-stereotypical behaviour and penalizenon-conformity (Hannover & Kessels, 2004; Kessels, 2005). To fit in or to avoid penalties,adolescents may conform to traditional gender-role behaviour if their friends have tra-ditional gender-normative ideas. We therefore hypothesize that having friends with moretraditional gender norms is associated with a lower likelihood of girls choosing STEM fields(H1A) and with a higher likelihood of boys choosing STEM fields (H1B).

Friends’ gender norms and gender composition of the friend group

The gender composition of an adolescent’s circle of friends may affect the relationshipbetween friends’ gender norms and STEM choices in two ways. First, scholars haveargued that the pressure to conform to gender-stereotypical norms is greater in same-

EDUCATIONAL RESEARCH AND EVALUATION 3

sex groups than in mixed-sex groups (Drury, Bukowski, Velásquez, & Stella-Lopez, 2013;Leaper & Smith, 2004; Martin & Fabes, 2001). This is because gender becomes moresalient among same-sex friends, since the division between their “own” gender and the“other” gender is more pronounced. If same-sex classmates reinforce compliance withgender norms, then the gender normativity of the circle of friends would have a strongereffect in same-sex friend circles. We, therefore, expect that the negative relationshipbetween having friends with more traditional gender norms and girls’ likelihood of choosingSTEM fields is stronger when they have a higher share of same-sex friends (H2A), and the posi-tive relationship between having friends with more traditional gender norms and boys’ like-lihood of choosing STEM fields is stronger when they have a higher share of same-sexfriends (H2B).

Studies in line with these hypotheses, however, have only assumed the role of gendernorms, mainly by extrapolating on the wider environment (same-sex schools or classes) orby focusing on other education-related outcomes (Francis et al., 2003; Thompson, 2003).Studies that consider gender norms and focus on a more immediate environment, suchas Robnett and Leaper (2013; US), suggest that – in line with H2A – girls are less interestedin STEM careers when their friend group does not support STEM and is predominantlycomprised of members of the same sex. Other studies, contrary to H2A or H2B, findthat a male-dominated environment enforces traditional gender norms among girls (Das-gupta et al., 2015; Riegle-Crumb & Morton, 2017). For instance, Dasgupta et al. (2015) con-cluded from an experiment that implicit masculine stereotypes about engineering led toless engineering career aspirations among women who were assigned to primarily oppo-site-sex (male-dominated) groups. In primarily same-sex (female-dominated) groups andgroups with equal numbers of boys and girls, effects of such stereotypes were absent.Much less research has focused on boys. A recent meta-analysis found no substantialadvantages of same-sex education for boys in terms of math, science, and verbal perform-ance; general achievement; school attitudes; gender stereotyping; math attitudes; andeducational aspirations (Pahlke, Hyde, & Allison, 2014). Boys’ educational choices,however, can be influenced by gender norms. Boys make more gender-stereotypical edu-cational choices, like for STEM fields, if they have more traditional gender-role beliefs (Vander Vleuten, Jaspers, Maas, & Van der Lippe, 2016). Moreover, research has found that boysare rebuked more severely than girls for exhibiting gender-atypical behaviour, and boysfeel more social pressure (at least from their male friends) to behave in gender-normativeways (Simpson, 2005). It is therefore worthwhile to explore whether having male friendswith more traditional gender norms draws boys into the STEM pipeline.

It is also possible that a larger group of same-sex friends might lessen the need toconform to gender norms, resulting in less gender-stereotypical educational choicesamong both boys and girls. In a male-dominated environment – like the STEM pipeline– the presence of more boys could reinforce girls’ feeling that they “do not belong”(Dasgupta et al., 2015). In a female-dominated environment, girls might feel morefreedom to explore gender-atypical interests and abilities, because differencesbetween boys’ and girls’ behaviour will be less in evidence. If gender differences areless pronounced in same-sex groups, pressure to conform to friends’ traditionalgender norms may be weaker among students with more same-sex friends. On thebasis of this argument, we expect that the negative relationship between having friendswith more traditional gender norms and girls’ likelihood of choosing STEM fields is

4 M. VAN DER VLEUTEN ET AL.

weaker when they have a higher share of same-sex friends (H3A). This argument hasmainly been put forward for girls, but we assume that it applies similarly to boys. There-fore, we hypothesize that the positive relationship between having friends with more tra-ditional gender norms and boys’ likelihood of choosing STEM fields is weaker when theyhave a higher share of same-sex friends (H3B).

Method

The educational system in the Netherlands

In the Netherlands, secondary education begins at age 12 and is compulsory until obtain-ment of a “starting qualification” at the upper secondary level (age 17 or 18). Dependingon their grades, test results, and teacher recommendations, Dutch students can enter oneof three possible levels of secondary education: (a) the VMBO or vocational level (4 years),which provides access to further vocational training (MBO); (b) the HAVO or general level(5 years), which provides access to universities of applied sciences offering bachelor’sdegrees; and (c) the VWO or academic level (6 years), which prepares students to entera research university offering academic bachelor’s and master’s degrees.

Furthermore, during secondary education, Dutch students must choose one of four coresubject areas which emphasize a certain field of study, in addition to satisfying general edu-cational requirements. Vocational students make this choice at the end of their second year(in Grade 8, age 14) and choose between four trajectories: “health & wellbeing”, “econ-omics”, “agriculture”, and “technology”. Students in the general and academic levelmake the choice at the end of their third year of secondary school (Grade 9, age 15).They have four core trajectories to choose from: “culture & society”, “economics &society”, “science & health”, and “science & technology”. Or they may elect to combinethese trajectories, usually “culture & society” combined with “economics & society” or“science & health” together with “science & technology”. STEM trajectories are “technol-ogy” at the vocational level and “science & technology” and “science & health” at thegeneral and academic levels. In the vocational “technology” trajectory, mathematics ismandatory and schoolwork focuses on science and physics. The “science & technology” tra-jectory at the general and academic levels focuses onmathematics, chemistry, and physics,whereas the “science & health” trajectory focuses on mathematics, chemistry, and biology.

After secondary school, vocational students go on to further vocational education(MBO), choosing their field of study when they enter MBO (age 16). Students in thegeneral level of secondary education finish at age 17, and those in the academic levelfinish at age 18. Most then go on to tertiary education, choosing a field of study in doing so.

It is very difficult to enter STEM after secondary education in the Netherlands, comparedto other countries. Most STEM-related study programmes require incoming students tohave completed a STEM trajectory in secondary school. To examine the mechanism of“leakage” from the pipeline, and also because secondary school students who do notchoose a STEM trajectory are very unlikely to end up in a STEM field later, we focusedon students who had already opted for a STEM trajectory in secondary education; thatis, students who chose “technology” at the vocational level and “science & technology”or “science & health” (or a combination of these two) at the general level or the academiclevel.

EDUCATIONAL RESEARCH AND EVALUATION 5

Data

We used Dutch data from the Children of Immigrants Longitudinal Study in Four EuropeanCountries (CILS4EU; Kalter et al., 2014, 2016a, 2016b) and the follow-up to this project inthe Netherlands, the Children of Immigrants Longitudinal Survey in the Netherlands(CILSNL; Jaspers & Van Tubergen, 2014, 2015). These projects explored the structural, cul-tural, and social integration of immigrant and non-immigrant children. From the CILS4EU,we used the second wave, collected in 2011–2012, when adolescents were 16 years oldand in the 10th grade (4th year of secondary education). From CILSNL, we used thefourth and fifth waves of data, collected in 2014 and 2015, when adolescents were 18or 19 years old. In Wave 4 (2014), adolescents from the general secondary school level(which lasts 5 years) had left secondary education; most had continued to tertiarystudies, entering a specific field of study. In Wave 5 (2015), adolescents from the academiclevel (which lasts 6 years) had left secondary education, with most entering university edu-cation and, in so doing, choosing a field of study. Students from the vocational level (whichlasts 4 years) chose their field of study in Wave 3. However, because they were not queriedabout their field of study in Wave 3, we used the field of study designation fromWave 4 foradolescents in the vocational level. Thus, the dependent variable (i.e., field of study) wasmeasured in Wave 4 or Wave 5, depending on the timing of the field of study choices,while the independent variables (e.g., friends’ characteristics) and control variables weremeasured in Wave 2. Because classes in the Netherlands remain relatively stable after stu-dents make their trajectory choice in secondary education, we could associate the charac-teristics of class friends gathered in Wave 2 with the STEM choices adolescents made inWave 4 or Wave 5.

In Wave 1, students were selected based on a three-stage stratified sample design (byeducational level and by percentage of non-Western immigrants in a school). First, schoolswith a higher proportion of immigrant children were oversampled, otherwise the samplingof the schools was random. The initial school-level response rate was 34.9%. To increasethe response rate, non-participating schools were replaced with similar alternativeschools, leading to a response rate of 91.7% at the school level. Second, within eachschool, two classes were randomly sampled (class participation rate = 94.5%). Third, all stu-dents who were present in these classes were asked to participate (student participationrate = 91.1%). In total, 4,363 students participated across 222 classes in 100 schools.

In Wave 2, 98% of all previous schools and 72.5% of all students at these schools par-ticipated again (N = 3,614). We added an extra sample of students who were not part ofthe original sampling frame for two main reasons (Nstudents_new = 2,307). First, somerespondents participated in the first wave even though their class had not beensampled then. This was, for example, because schools wanted to participate with morethan the two sampled classes. Most of these students also participated in Wave 2 andcould therefore be included in our sample. Second, in Wave 2 students were groupedinto classes according to their chosen trajectory, meaning that classes changed consider-ably between Wave 1 and Wave 2. One goal of the CILS project was to survey wholeclasses, which meant that new students who were not part of the original samplingframe were also surveyed.

The response rates for Waves 4 and 5 were, 55.5% and 54.4%, respectively, both calcu-lated as the ratio between the number of respondents who participated and the number

6 M. VAN DER VLEUTEN ET AL.

of adolescents who were approached and did not refuse participation before thestart of Wave 4 or Wave 5. In both waves, a mixed mode approach was used. In Wave4, approximately 69% completed an electronic questionnaire, 10% filled in a print ques-tionnaire, and 21% responded by telephone. In Wave 5, approximately 85% completedan electronic questionnaire, 1% filled in a print questionnaire, and 14% responded bytelephone.

In total, 5,921 respondents participated in Wave 2 (Nclasses = 301), 1,600 of whomfulfilled our sample requirement of having chosen a STEM trajectory in secondary edu-cation. However, to obtain our final analytical sample, we excluded respondents whodid not participate in Wave 4 or Wave 5 (n = 439) and those who had not indicated aselected field of study (n = 324; most of these students were still in secondary education).After excluding missing values on all variables of interest (n =93; 51 due to not havingfriends in the class), our analytical sample comprised 744 students from 174 classes. Seethe section on selectivity of the sample for more information on how we adjusted forselective sample attrition and non-response.

Operationalization

The dependent variable is categorical and measures individual STEM field choices in post-secondary education. It is based on the question, “What is your field of study?” The originalresponse categories were coded using the 3-digit International Standard Classification ofEducation (ISCED97; United Nations Educational Scientific and Cultural Organization,2006). These were recoded into two categories: (1) STEM field or (0) non-STEM field (seeAppendix 1 for an overview of the coding of the dependent variable). Due to our interestin gender, we focused on STEM fields in which women are underrepresented (Diekmanet al., 2010) and defined fields such as medicine as non-STEM fields.1 Moreover, wechose to define STEM fields in line with recent Dutch legislation to promote entry ofgirls in the hard science fields of mathematics, natural sciences, engineering, and technol-ogy (Booy, Jansen, Joukes, & Van Schaik, 2012; Jansen & Joukes, 2012).

The main independent variables are sex (girl = 1 vs. boy = 0) and traditional gender normsof friends. The latter indicates the norms upheld by class friends regardingmale and femalegender-role behaviour. “Who are your best friends in class?” was used to identify respon-dents’ friends in the class, with respondents allowed to name up to five friends. All respon-dents were asked about their own gender norms. Complete classes participated in Wave 2,giving us the gender norms of the respondents’ class friend group. Traditional gender-rolebehaviour was measured by the question, “Who do you think should do the followingtasks?” The tasks presented were taking care of children, cooking, earning money, andcleaning the house. Response categories were “mostly the man”, “mostly the woman”,and “both about the same”. For the more “feminine” tasks – taking care of children,cooking, and cleaning – we assigned a score of 2 to respondents who answered“mostly the woman”, a score of 1 if they answered “both about the same”, and a scoreof 0 if they answered “mostly the man”. For the more “masculine” item – earningmoney – we assigned a score of 0 to respondents who answered “mostly the woman”,a score of 1 if they answered “both about the same”, and a score of 2 if they answered“mostly the man”. A mean score was calculated (Cronbach’s α = .70), with higher scoresindicating more traditional gender norms. To analyse how traditional friends’ gender

EDUCATIONAL RESEARCH AND EVALUATION 7

norms were, we averaged the traditional gender norms score (the mean scale of all fouritems) of the respondents’ friend group (not including own gender norms).

To measure the sex composition of the circle of friends, we first constructed a variableindicating the proportion of same-sex friends in the friend group. Among the respondentswho fulfilled our sample requirements of having chosen a STEM trajectory in secondaryeducation, indicating a post-secondary education field of study and having no missingvalues on other variables, the average number of friends in the class was 3 (N = 795; M= 3.19; SD = 0.85). Of these respondents, 51 students had no friends in the class. As wewere interested in how characteristics of the friend group correlated with STEM choicesafter secondary education, we excluded these respondents. This somewhat increasedthe average number of class friends in our sample (N = 744; M = 3.27; SD = 0.76). The pro-portion of same-sex friends represents for boys the proportion of males and for girls theproportion of females in the friend group. This variable, however, was skewed. Only 176(24%) of all students indicated having an opposite-sex friend in their class. We thus con-structed the variable at least one opposite-sex friend indicating if the respondent had onlysame-sex friends (0) or at least one opposite-sex friend (1). A disadvantage of our data wastheir limitation to within the class, meaning that we could not examine the role of friendselsewhere.

Controls

We controlled for individual math achievement, considering it an important indicator forstudents’ choice for a STEM field. Mathematics achievement refers to the respondent’smath grade on his or her latest progress report (“What was your math grade on yourlatest progress report?”). It can vary between 1 (low achievement) and 10 (high achieve-ment). In addition, we controlled for the math achievement of friends, reflecting howwell the respondent’s group of class friends did in mathematics. For this, we averagedthe math achievement of respondents’ best friends in the class.

Traditional gender norms refer to the respondent’s own gender norms. We held this con-stant because we were interested in how a traditional gender-normative environmentaffected STEM choices irrespective of the respondent’s own gender norms. We performedadditional analyses to explore how friends’ characteristics interact with students own tra-ditional gender norms (see the section on additional analyses).

As children from higher educated families are more likely to make gender-atypicalstudy choices (Støren & Arnesen, 2007), we controlled for the highest educational levelof parents, referring to the highest level attained by either the father or mother, or bythe one parent if the respondent was from a single-parent household. This variableranges from no education (0) to university (5).

As the data contain an oversampling of respondents with a non-Western immigrantbackground, we controlled for non-Western immigrant background. This variable indicateswhether one of the respondent’s parents was (1) or was not (0) born in a non-Westerncountry.2

Finally, we included a set of dummy variables: science & health-general level; science &health & technology-general level; science & technology-general level and science & health-academic level; science & health & technology-academic level; science & technology-academiclevel. These six dummies indicate which trajectory students chose in, respectively, the

8 M. VAN DER VLEUTEN ET AL.

general level and the academic level of secondary school. The reference category is stu-dents who chose technology in the vocational level; technology-vocational level. Table 1presents the (unweighted) descriptive statistics of all variables in our analyses.

Selectivity of the sample

Several types of selectivity may have occurred in our sample. First, we only included stu-dents who chose a STEM trajectory in secondary education, but gender norms might haveinfluenced those choices. It might therefore be that the girls and/or boys in our analyseswere selectively included on the basis of their gender norms as well as those of theirfriends. Tests showed this to be true only for the girls, not the boys. The girls in oursample were less traditional (M = 1.21, SD = 0.29) and had fewer traditional friends (M =1.23, SD = 0.21) than the girls who did not choose a STEM-related core subject area in sec-ondary education (Mownnorms = 1.25, SDownnorms = 0.31; Mfriendnorms = 1.26, SDfriendnorms =0.22). Moreover, the girls in our sample were less traditional (M = 1.21, SD = 0.29) thanthe boys in our sample (M = 1.42, SD = 0.35), and the girls in our sample had less traditionalfriends (M = 1.23, SD = 0.21) than the boys in our sample had (M = 1.40, SD = 0.23). The girlsin our sample, as well as their friends, can therefore be considered a selective group basedon their gender-role attitudes, which might lead to underestimation of the effect of(friends’) gender norms for girls.

Table 1. Descriptive statistics of all the variables in our analyses for all respondents (N = 744), and forboys (n = 444) and girls (n = 300) separately.

MEAN (SD) MIN MAX

Total Boys Girls Total Boys Girls Total Boys Girls

Dependent variableNon-STEM fields .60 .44 .83 0 0 0 1 1 1STEM fields .40 .56 .17 0 0 0 1 1 1

Independent variablesSex (girl = 1) .40 0 0 0 1 1 1Traditional gender 1.33 1.40 1.23 .63 .63 .75 2 2 2norms of friends (0.24) (0.23) (0.21)At least one opposite- .24 .23 .25 0 0 0 1 1 1sex friend

ControlsMathematics 6.89 6.84 6.97 2 2 3 10 10 10achievement (1.34) (1.39) (1.27)Math achievement of 6.75 6.72 6.79 3 3 3 10 9 10friends in class (0.96) (0.92) (1.01)Traditional gender 1.34 1.42 1.21 0.5 0.5 0.75 2 2 2norms (0.34) (0.35) (0.29)Highest educational 3.37 3.33 3.43 0 0 0 5 5 5parents (1.14) (1.11) (1.18)Non-Western immigrant .14 .11 .18 0 0 0 1 1 1backgroundTechnology-vocational level .29 .43 .09 0 0 0 1 1 1Science & Health-general level .19 .11 .30 0 0 0 1 1 1Science & Health & Technology-general level .04 .03 .05 0 0 0 1 1 1Science & Technology-general level .10 .14 .04 0 0 0 1 1 1Science & Health-academic level .16 .09 .26 0 0 0 1 1 1Science & Health & Technology-academic level .10 .07 .14 0 0 0 1 1 1Science & Technology-academic level .12 .12 .12 0 0 0 1 1 1

Source: Wave 2 of Children of Immigrants Longitudinal Survey in Four European Countries and Waves 4 and 5 of Children ofImmigrants Longitudinal Survey in the Netherlands, own calculations.

Note: For categorical variables, proportions are given. The standard deviation (SD) is indicated in brackets.

EDUCATIONAL RESEARCH AND EVALUATION 9

Second, our sample may be selective due to panel attrition (school dropout) or non-response on the field of study variable. These adolescents may have somewhatdifferent characteristics than those who remained in education and went on to indicatea post-secondary field of study. For example, children of higher educated parents mightbe more likely to stay in school (Sewell, 1971). To test for this type of selectivity, werepeated all analyses using multinomial logit analyses and a categorical dependent vari-able indicating if students were in our data (0; N = 744), had dropped out (1; N = 367),or were in our data but had not indicated a field of study (2; N = 292). Overall, boyswith a non-Western immigration background are underrepresented in our data. This isnot necessarily problematic, as our goal was not specifically to generalize our findingsto non-Western immigrants. Furthermore, students from the vocational level are slightlyunderrepresented in our data, as they were more likely to drop out and less likely tohave indicated a field of study compared to students from higher education (generaland academic level). Given that lower educated families often have more traditionalgender norms, this may lead to underestimation of the role of gender norms.

To adjust for selective sample attrition and non-response, we used inverse probabilityweighting (IPW). First, we obtained estimated probabilities from logit models predictinginclusion in our analytical sample of demographic characteristics, family background, andmath grade. Second, we assigned each case the inverse of this probability, so that caseswhich were less likely to be in our sample had a greater weight than cases that were morelikely to be in our sample (for a review of this procedure, see Seaman & White, 2013). Notweighting our analyses produced highly similar results and did not alter our conclusions.

Analyses

Given the binary nature of our dependent variable, we performed logit regression to testour hypotheses. We clustered standard errors at the class level (Wave 2) to account for thenesting of students within classes. To prevent three-way interactions, we performed ana-lyses separately for boys and for girls.

We started our analyses with a model including both boys and girls as well as all controlvariables, to get an indication of the size of the gender differences in the choice for STEMfields. This model is presented in Appendix 2. To test our hypotheses, we estimated sixmodels that included the control variables (Models 1, 2, and 3 for girls and Models 4, 5,and 6 for boys; see Table 2). The first hypothesis states that having friends with more tra-ditional gender norms is associated with a lower likelihood of girls choosing STEM fields(H1A), and with a higher likelihood of boys choosing STEM fields (H1B). We used fourmodels to test these hypotheses. The first two concern the effect of friends’ traditionalgender norms for girls (Model 1) and for boys (Model 4). The second two, Models 2 and5, add the gender composition of the friend group. These four models combined giveus insight into the association between friends’ traditional gender norms and students’likelihood of entering STEM fields.

Model 3 tests Hypotheses 2 and 3 for girls. These state that the negative relationshipbetween having friends with more traditional gender norms and girls’ likelihood ofchoosing STEM fields (H1) is stronger (H2A) or weaker (H3A) if they have a larger shareof same-sex friends. Similarly, Model 6 tests Hypotheses 2 and 3 for boys. These statethat the positive relationship between having friends with more traditional gender

10 M. VAN DER VLEUTEN ET AL.

norms and boys’ likelihood of choosing STEM fields (H1) is stronger (H2B) or weaker (H3B) ifthey have a larger share of same-sex friends. We tested these hypotheses by including aninteraction between the variable traditional gender norms of friends and at least one oppo-site-sex friend in Models 2 and 3. Table 2 presents the results in terms of log odds. To facili-tate interpretation, we also report average marginal effects (AME). The AME gives theaverage change in the probability of choosing STEM fields for a one-unit change in theexplanatory variable. Multiplied by 100, this is the average change in percentage points.

We performed Wald tests to assess the models and determine whether the variablescontribute significantly to the model. For model fit, we used generalized Hosmer andLemeshow goodness-of-fit measures (Fagerland & Hosmer, 2012).

Results

To determine the size of the gender difference in STEM fields in the Netherlands, wefirst estimated a model that included both boys and girls as well as all control variables(Appendix 2). This model shows the gender gap to be quite large. Girls were onaverage 31 percentage points less likely than boys to enter STEM fields.

In line with Hypothesis 1A, Model 1 in Table 2 shows that girls were less likely to chooseSTEM fields if their class friends had more traditional gender norms. Calculated in AMEs,when friends were more traditional (by 1 point), the average likelihood of a girl choosinga STEM field decreased by 31%. This effect hardly changed when we controlled for gendercomposition (Model 2). Girls remained on average 30% less likely to choose STEM fieldswhen their friends’ traditional gender ideology increased by 1 point. This confirms Hypoth-esis 1A. Contrary to Hypothesis 1B, Models 4 and 5 show no association between friends’traditional gender norms and STEM choices for boys. Model 5 does show that among boyswith non-traditional friends, the gender composition of the friend group was linked toSTEM choices. Boys who had at least one opposite-sex friend were 16% less likely tochoose a STEM field compared to boys with only same-sex friends.

Models 3 and 6 test Hypotheses 2 and 3. These state that the relationship between havingfriends with more traditional gender norms and students’ likelihood of choosing STEM fieldsis stronger (H2) or weaker (H3) when they have a larger share of same-sex friends. The inter-actions in Models 3 and 6, however, make no significant contribution to either model (Model3: Wald χ(2)² = 0.71; p = .40; Model 6: Wald χ(2)² = 1.76; p = .18), meaning that these modelspecifications are not significantly better than Models 2 and 5, respectively. Moreover, theinteraction effects are not significant, refuting Hypotheses 2 and 3.

Table 2 also shows that, compared to the technological-vocational level, boys and girlsin the science & health-general level were less likely to choose STEM fields. Girls were alsoless likely to choose STEM fields in the science & health-academic level and the science &health & technology-academic level compared to the technological-vocational level.3

For all models, the goodness-of-fit tests show no significant difference between theobserved and predicted models, indicating good model fit.

Additional analyses: individual gender norms and class characteristics

In Table 2, we see that students’ own gender norms did not affect STEM choices after sec-ondary education for boys or girls. Moreover, for all analyses, we found no evidence that

EDUCATIONAL RESEARCH AND EVALUATION 11

the effect of students’ own traditional gender norms on choosing STEM fields was strongeror weaker depending on whether friends had more traditional gender norms (interactioneffect). For boys, we found no evidence that class friends’ gender norms were important,although for girls having friends with more traditional gender norms was associated withSTEM choices, regardless of the girls’ own gender norms.

Previous research suggests that, besides friends, the class context can play a role ingender differences in the choice for STEM fields (Dryler, 1999). To ensure that ourfindings can be attributed to class friends and not to class characteristics, we ran all ofthe analyses in Table 2 again including three class context variables: traditional gendernorms within the class (excluding own gender norms and the gender norms of the friendgroup), proportion of females in the class (excluding own sex and sex of friends), andaverage classmath achievement (excludingownmath achievement andmath achievementof friendgroup). Overall, the results differed little from those reported in this paper. Thus, the

Table 2. Results of logit regression (log-odds) models that test the effect of friends’ gender normsand the gender composition of the friend group on choosing STEM fields for girls (n = 300) andboys (n = 444) separately.Independent variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Traditional gender norms of friends −2.53** −2.50** −2.87** −0.57 −0.78 −1.11*(0.84) (0.85) (0.95) (0.46) (0.48) (0.55)

At least one opposite-sex friend −0.35 −2.12 −0.68** −2.87(0.46) (2.23) (0.26) (1.66)

Traditional gender norms of friends 1.44 1.62*at least one opposite-sex friend Controls (1.71) (1.22)Mathematics achievement 0.12 0.11 0.12 0.20* 0.20* 0.21*

(0.16) (0.16) (0.16) (0.08) (0.08) (0.08)Math achievement of friends 0.03 0.03 0.03 −0.02 −0.04 −0.04

(0.16) (0.16) (0.16) (0.11) (0.11) (0.11)Traditional gender norms −0.01 0.01 0.02 −0.00 −0.02 −0.03

(0.56) (0.57) (0.57) (0.27) (0.28) (0.27)Highest educational level parents 0.11 0.12 0.13 −0.16 −0.17 −0.17

(0.15) (0.16) (0.15) (0.10) (0.10) (0.10)Non-Western immigrant 0.62 0.65 0.70 −0.26 −0.21 −0.24background (0.41) (0.43) (0.42) (0.33) (0.33) (0.34)Technology-vocational level (ref) – – – – – –Science & Health-general level −1.91*** −1.87*** −1.85*** −1.00** −0.98** −1.01**

(0.53) (0.55) (0.55) (0.35) (0.36) (0.36)Science & Health & Technology- 0.07 0.17 0.13 1.03 1.27 1.18general level (0.74) (0.78) (0.77) (0.71) (0.80) (0.80)Science & Technology-general −0.04 0.02 0.10 0.06 0.15 0.15level (0.77) (0.75) (0.75) (0.34) (0.35) (0.34)Science & Health-academic level −1.81** −1.84** −1.88** −0.83 −0.66 −0.65

(0.60) (0.62) (0.63) (0.43) (0.43) (0.44)Science & Health & Technology- −1.46* −1.45* −1.47* 0.07 0.04 0.04academic level (0.65) (0.65) (0.65) (0.44) (0.44) (0.44)Science & Technology-academic −0.85 −0.91 −0.96 0.16 0.26 0.23level (0.62) (0.64) (0.64) (0.35) (0.35) (0.35)Constant 1.11 1.09 1.47 0.58 1.09 1.53

(2.05) (2.01) (2.18) (1.18) (1.20) (1.30)Model fitHosmer–Lemeshow chi2 8.36 13.41 10.24 5.39 4.95 8.27df 8 8 8 8 8 8p value 0.40 0.10 0.25 0.71 0.76 0.41

Source: Wave 2 of Children of Immigrants Longitudinal Survey in Four European Countries and Waves 4 and 5 of Children ofImmigrants Longitudinal Survey in the Netherlands, own calculations. *p < 0.05. **p < 0.01. ***p < 0.001.

Note: Ntotal = 744; standard errors clustered by class (N = 174). Model estimates use inverse-probability weighting toaccount for selective sample attrition and non-response. Results are shown in log odds, and the standard deviation(SD) is indicated in brackets.

12 M. VAN DER VLEUTEN ET AL.

class context had no significant effect. These extra analyses highlight the importance ofclass friends (i.e., their gender normativity) in adolescents’ choices for STEM fields.

Conclusion and discussion

This study examined the role of class friends in boys’ and girls’ choices to pursue STEM fields.We tested whether having friends with more traditional gender norms was associated withchoosing STEM fields for boys and for girls, and whether pressure to conform to traditionalgender norms differed depending on the gender composition of the friend group. Wefocused on students in the STEM pipeline; that is, boys and girls who had already chosena STEM-related trajectory in secondary education. This allowed us to better understandpossible causes of gender-specific STEM dropout in continued education. Using threewaves of longitudinal data, we analysed 744 Dutch students using logistic regression.

Despite various government efforts to open up STEM fields to girls, a substantial gendergap remains in STEM choices in the Netherlands. In our sample, girls were on average 31%less likely to choose STEM fields than boys. Our findings provide evidence of the influenceof the gender normativity of class friends on girls’ STEM choices, but not on boys’ STEMchoices. Girls were substantially less likely to pursue STEM fields when their friendsupheld more traditional gender norms, irrespective of their own (traditional) gendernorms. Our finding is in line with research showing that women are deterred fromSTEM fields by ideas about what is “appropriate” male or female gender-role behaviour(Charles & Bradley, 2009; Legewie & DiPrete, 2014; Morgan et al., 2013). Our resultsshow the persuasiveness of class friends’ traditional gender norms in shaping STEMchoices. The fact that girls, but not boys, were influenced by their friends’ traditionalgender norms agrees with research showing that girls are more responsive to socialnorms in their networks (Frank et al., 2008).

We found no evidence that gender norms were more influential among same-sexgroups (Robnett & Leaper, 2013) or opposite-sex groups (Dasgupta et al., 2015; Riegle-Crumb et al., 2012; Schneeweis & Zweimüller, 2012) for either class friends or the classcontext. Our findings do indicate that the gender composition of the friend group wasimportant in boys’ STEM choices. Boys who had at least one opposite-sex friend in theirgroup were less likely to choose STEM fields. In other words, having only male friends isassociated with gender-typical male behaviour for boys – in this case a higher likelihoodof entering STEM fields. This, however, is not because same-sex friends’ gender normspressure boys into gender-conforming behaviour. For boys, other explanations for gen-dered network effects are more plausible. For example, same-sex friends likely sharegender-typical interests and activities (Martin et al., 2013). As STEM fields are consideredto coincide with male-typical interests, boys with more same-sex friends might be morelikely to engage in and share STEM-related interests. This may be why having moresame-sex friends increases boys’ likelihood of choosing STEM fields.

The findings of this study should be viewed in the context of its limitations. Like moststudies on peer influence, we cannot be sure that the effects measured were not, at least inpart, the result of individuals choosing friends based on certain characteristics, often thoseshared in common (Mouw, 2006). Unfortunately, our data are unsuited to social networkanalyses. But future research on the role of friends in gender differences in entry into STEMfields would benefit from advanced statistical tools that disentangle the processes of

EDUCATIONAL RESEARCH AND EVALUATION 13

selection and influence within friendship circles (Steglich, Snijders, & Pearson, 2010). More-over, our study measured math achievement based on self-reported grades, thereforerelying on recall. Future research might replicate our results using objective measures,like national test scores or secondary school entry assessment. We must also be cautiousin generalizing our results to other countries or educational systems. The Netherlands hasa distinctive educational system. Adolescents are divided into different educational levelsin secondary school, and they choose a core area of study during the secondary schoolperiod. This suggests the value of future research that looks similarly at other educationalsystems for a cross-national comparison of the effects of gender norms and the gendercomposition of class friends on STEM choices. Moreover, although class friends are impor-tant in shaping STEM choices, future research might investigate the effect of friends ingeneral (i.e., including friends outside the class), since adolescents likely have friendsand peer groups outside the class and outside school as well.

Overall, we can conclude that in secondary education adolescents’ friends play a key rolein shaping their STEM choices. We demonstrated that having friends with more traditionalgender-normative ideas may push girls out of the STEM pipeline, whereas same-sex friendsmay retain boys in the STEM pipeline. In our sample, same-sex or opposite-sex class friendsdid not reinforce or weaken the effects of traditional gender norms in the friend group. Ourresearch suggests that STEM fields are still thought of as an “inappropriate” choice for girls inschool environments. We might reduce the leakage of girls from the STEM-pipeline by tack-ling norms of what is “appropriate” male and female gender-role behaviour.

Notes

1. To ensure the robustness of our results, we ran all analyses again with biology, health, andmedicine-related fields as a separate dependent category. This category included life sciences,biology, and biochemistry, as well as health, medicine, dental studies, medical diagnostics andtreatment technology, and pharmacy. Because there were too few girls in life sciences,biology, and biochemistry, we were unable to run the analyses on biology-related fields asa separate category from health & medicine-related fields. In total, 106 adolescents (nboys =34; ngirls = 72) chose biology, health, and medicine-related fields. The results were similar tothose reported in this paper. Neither friends’ traditional gender norms nor the gender compo-sition of the friend group affected the likelihood of boys or girls choosing biology, health, ormedicine-related fields.

2. On the basis of the definition given by Statistics Netherlands (CBS) and consistent with howthe CILS4EU sample was drawn, Western societies are defined as Europe (excluding Turkey),North America, Oceania, Indonesia, and Japan (Indonesia and Japan are consideredWestern based on their sociocultural and socioeconomic position). Non-Western countriesare Turkey, Morocco, Surinam, Dutch Antilles, and Aruba, alongside the countries of Africa,Asia (excluding Indonesia and Japan), and Latin America.

3. For boys and for girls, including an interaction between secondary education level (controllingfor the trajectory chosen in secondary education) and friends’ gender norms did not indicatethat the effect of friends’ gender norms was different for the different secondary educationlevels. We also included an interaction between the combination of trajectory and levelchoices (e.g., technology-vocational level) and friends’ gender norms. For girls (not boys),there is evidence that friends’ gender norms are less important in the science & technol-ogy-general level, the science & health & technology-general level, and the science &health-academic level than in the technology-vocational level. In other words, friends’gender norms appear to matter most for students in the technology-vocational level.

14 M. VAN DER VLEUTEN ET AL.

Importantly, in all analyses the main effect of friends’ gender norms remained important andsignificant for girls and unimportant for boys.

Acknowledgements

We are thankful for the helpful suggestions of participants attending the CILS4EU conference inStockholm, Sweden (December 2015), the round table session “STEM Education” at the annual Amer-ican Sociological Association meeting in Seattle, WA, USA (August 2016), the “Influence of Peers andClassmates” session at the European Consortium of Sociological Research (ECSR) conference: “Stra-tification and Population Processes in European Societies” in Oxford, UK (September 2016), theexpert meeting “Gender differences in field of study choices” in Oxford, UK (September 2016), theICS forum day in Nijmegen, the Netherlands (November 2016), and the Work-Family Seminar atUtrecht University, the Netherlands (January 2016). We also thank Ineke Maas, Tanja van derLippe, and Eva Jaspers for their helpful suggestions.

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

This research was supported by the Dutch Scientific Organisation (NWO) research talent grantentitled “Gendered Choices: school and field of study trajectories of male and female adolescentsin four European countries” [grant number 406-12-018]. Additionally, financial support from theNORFACE research programme on Migration in Europe – Social, Economic, Cultural and PolicyDynamics and from “NWO middelgroot” [480-11-013] is acknowledged.

Notes on contributors

Maaike van der Vleuten is an assistant professor at the Department of Sociology and ICS (Inter-uni-versity Centre for Social Science Theory and Methodology) at Radboud University Nijmegen, theNetherlands. Her research interests include gender inequality, gender roles, field of study choices,and same-sex couples.

Stephanie Steinmetz is an associated professor at the Institute for Social and Political Sciences at theUniversity of Lausanne. Her main research interests are social stratification, gender and educationalinequalities, as well as quantitative (web-based) research methods. She has published on topicsrelated to educational gender inequalities, women’s labour-market participation, and occupationalsegregation from a cross-national perspective.

Herman van de Werfhorst is professor of Sociology at the University of Amsterdam, and director ofthe Amsterdam Centre for Inequality Studies (AMCIS). His work concentrates on education and socialinequalities. One strand of research focuses on within-level differences in inequalities across fields ofstudy, and another strand concentrates on cross-national differences in inequalities in education, tostudy how inequalities are related to the institutional setup of educational systems. His work hasappeared in many journals including the European Sociological Review, American Journal of Sociology,Demography, Social Forces, Comparative Education Review, and the American Sociological Review.

ORCID

Maaike van der Vleuten http://orcid.org/0000-0003-1108-2446Stephanie Steinmetz http://orcid.org/0000-0001-7136-6622Herman van de Werfhorst http://orcid.org/0000-0003-2323-3782

EDUCATIONAL RESEARCH AND EVALUATION 15

References

Alon, S., & DiPrete, T. A. (2015). Gender differences in the formation of a field of study choice set.Sociological Science, 2(5), 50–81. doi:10.15195/v2.a5

Booy, C., Jansen, N., Joukes, G., & Van Schaik, E. (2012). Trend analysis: Gender in higher STEM edu-cation. Amsterdam: VHTO Landelijk Expertise Bureau.

Buck, G. A., Plano Clark, V. L., Leslie-Pelecky, D., Lu, Y., & Cerda-Lizarraga, P. (2008). Examining the cog-nitive processes used by adolescent girls and women scientists in identifying science role models:A feminist approach. Science Education, 92(4), 688–707. doi:10.1002/sce.20257

Ceci, S. J., Williams, W. M., & Barnett, S. M. (2009). Women’s underrepresentation in science:Sociocultural and biological considerations. Psychological Bulletin, 135(2), 218–261. doi:10.1037/a0014412

Charles, M., & Bradley, K. (2009). Indulging our gendered selves? Sex segregation by field of study in44 countries. American Journal of Sociology, 114(4), 924–976. doi:10.1086/595942

Cheryan, S., Plaut, V. C., Handron, C., & Hudson, L. (2013). The stereotypical computer scientist:Gendered media representations as a barrier to inclusion for women. Sex Roles, 69(1–2), 58–71.doi:10.1007/s11199-013-0296-x

Cook, T. D., Deng, Y., & Morgano, E. (2007). Friendship influences during early adolescence: Thespecial role of friends’ grade point average. Journal of Research on Adolescence, 17(2), 325–356.doi:10.1111/j.1532-7795.2007.00525.x

Corbett, C., & Hill, C. (2015). Solving the equation: The variables for women’s success in engineering andcomputing. Washington, DC: Aauw.

Correll, S. J. (2001). Gender and the career choice process: The role of biased self-assessments.American Journal of Sociology, 106(6), 1691–1730. doi:10.1086/321299

Crosnoe, R., Riegle-Crumb, C., Field, S., Frank, K., & Muller, C. (2008). Peer group contexts of girls’ andboys’ academic experiences. Child Development, 79(1), 139–155. doi:10.1111/j.1467-8624.2007.01116.x

Dasgupta, N., McManus Scircle, M., & Hunsinger, M. (2015). Female peers in small work groupsenhance women’s motivation, verbal participation, and career aspirations in engineering.Proceedings of the National Academy of Sciences, 112(16), 4988–4993. doi:10.1073/pnas.1422822112

Diekman, A. B., Brown, E. R., Johnston, A. M., & Clark, E. K. (2010). Seeking congruity between goalsand roles: A new look at why women opt out of science, technology, engineering, and mathemat-ics careers. Psychological Science, 21(8), 1051–1057. doi:10.1177/0956797610377342

Drury, K., Bukowski, W. M., Velásquez, A. M., & Stella-Lopez, L. (2013). Victimization and gender iden-tity in single-sex andmixed-sex schools: Examining contextual variations in pressure to conform togender norms. Sex Roles, 69(7–8), 442–454. doi:10.1007/s11199-012-0118-6

Dryler, H. (1998). Parental role models, gender and educational choice. The British Journal ofSociology, 49(3), 375–398. doi:10.2307/591389

Dryler, H. (1999). The impact of school and classroom characteristics on educational choices by boysand girls: A multilevel analysis. Acta Sociologica, 42(4), 299–318. doi:10.1177/000169939904200402

Eccles, J. (2011). Gendered educational and occupational choices: Applying the Eccles et al. model ofachievement-related choices. International Journal of Behavioral Development, 35(3), 195–201.doi:10.1177/0165025411398185

Fagerland, M. W., & Hosmer, D. W. (2012). A generalized Hosmer–Lemeshow goodness-of-fit test formultinomial logistic regression models. The Stata Journal, 12(3), 447–453. doi:10.1177/1536867X1201200307

Francis, B., Hutchings, M., Archer, L., & Melling, L. (2003). Subject choice and occupational aspirationsamong pupils at girls’ schools. Pedagogy, Culture & Society, 11(3), 425–442. doi:10.1080/14681360300200182

Frank, K. A., Muller, C., Schiller, K. S., Riegle-Crumb, C., Mueller, A. S., Crosnoe, R., & Pearson, J. (2008).The social dynamics of mathematics coursetaking in high school. American Journal of Sociology,113(6), 1645–1696. doi:10.1086/587153

16 M. VAN DER VLEUTEN ET AL.

Gabay-Egozi, L., Shavit, Y., & Yaish, M. (2015). Gender differences in fields of study: The role of signifi-cant others and rational choice motivations. European Sociological Review, 31(3), 284–297. doi:10.1093/esr/jcu090

Ganotice, F. A., & King, R. B. (2014). Social influences on students’ academic engagement and scienceachievement. Psychological Studies, 59(1), 30–35. doi:10.1007/s12646-013-0215-9

Hallinan, M. T., & Williams, R. A. (1990). Students’ characteristics and the peer-influence process.Sociology of Education, 63(2), 122–132. doi:10.2307/2112858

Hannover, B., & Kessels, U. (2004). Self-to-prototype matching as a strategy for making academicchoices. Why high school students do not like math and science. Learning and Instruction, 14(1),51–67. doi:10.1016/j.learninstruc.2003.10.002

Hill, C., Corbett, C., & Rose, A. S. (2010). Why so few? Women in science, technology, engineering, andmathematics. Washington, DC: Aauw.

Hilliard, L. J., & Liben, L. S. (2010). Differing levels of gender salience in preschool classrooms: Effectson children’s gender attitudes and intergroup bias. Child Development, 81(6), 1787–1798.

Hyde, J. S., Lindberg, S. M., Linn, M. C., Ellis, A. B., & Williams, C. C. (2008). Gender similarities charac-terize math performance. Science, 321(5888), 494–495. doi:10.1126/science.1160364

Hyde, J. S., & Mertz, J. E. (2009). Gender, culture, and mathematics performance. Proceedings of theNational Academy of Sciences of the United States of America, 106(22), 8801–8807. doi:10.1073/pnas.0901265106

Jansen, N., & Joukes, G. (2012). Long term, interrelated interventions to increasewomen ‘s participationin STEM in the Netherlands. International Journal of Gender, Science and Technology, 5(3), 305–316.

Jaspers, E., & Van Tubergen, F. (2014). Children of Immigrants Longitudinal Survey in the Netherlands(CILSNL) – Wave 4. Full version v4.0.0 [Data file]. DANS.

Jaspers, E., & Van Tubergen, F. (2015). Children of Immigrants Longitudinal Survey in the Netherlands(CILSNL) – Wave 5. Full version v5.0.0 [Data file]. DANS.

Kalter, F., Heath, A. F., Hewstone, M., Jonsson, J. O., Kalmijn, M., Kogan, I., & Van Tubergen, F. (2014).Children of Immigrants Longitudinal Survey in four European countries (CILS4EU) – Full version [Datafile for on-site use]. GESIS Data Archive, Cologne, ZA5353 Data file Version 1.0.0. doi:10.4232/cils4eu.5353.1.0.0

Kalter, F., Heath,A. F., Hewstone,M., Jonsson, J.O., Kalmijn,M., Kogan, I., &VanTubergen, F. (2016a).Childrenof Immigrants Longitudinal Survey in four European countries (CILS4EU) – Full version [Data file for on-siteuse]. GESIS Data Archive, Cologne, ZA5353 Data file Version 1.2.0. doi:10.4232/cils4eu.5353.1.2.0

Kalter, F., Heath, A. F., Hewstone, M., Jonsson, J. O., Kalmijn, M., Kogan, I., & Van Tubergen, F. (2016b).Children of Immigrants Longitudinal Survey in four European countries (CILS4EU) – Full version [Datafile for on-site use]. GESIS Data Archive, Cologne, ZA5353 Data file Version 2.3.0. doi:10.4232/cils4eu.5353.2.3.0

Kessels, U. (2005). Fitting into the stereotype: How gender-stereotyped perceptions of prototypicpeers relate to liking for school subjects. European Journal of Psychology of Education, 20(3),309–323. doi:10.1007/BF03173559

Konrad, A. M., Ritchie, J. E., Jr., Lieb, P., & Corrigall, E. (2000). Sex differences and similarities in jobattribute preferences: A meta-analysis. Psychological Bulletin, 126(4), 593–641. doi:10.1037/0033-2909.126.4.593

Leaper, C., & Smith, T. E. (2004). A meta-analytic review of gender variations in children’s languageuse: Talkativeness, affiliative speech, and assertive speech. Developmental Psychology, 40(6),993–1027. doi:10.1037/0012-1649.40.6.993

Legewie, J., & DiPrete, T. A. (2012). School context and the gender gap in educational achievement.American Sociological Review, 77(3), 463–485. doi:10.1177/0003122412440802

Legewie, J., & DiPrete, T. A. (2014). The high school environment and the gender gap in science andengineering. Sociology of Education, 87(4), 259–280. doi:10.1177/0038040714547770

Mann, A., & DiPrete, T. A. (2013). Trends in gender segregation in the choice of science and engin-eering majors. Social Science Research, 42(6), 1519–1541. doi:10.1016/j.ssresearch.2013.07.002

Martin, C. L., & Fabes, R. A. (2001). The stability and consequences of young children’s same-sex peerinteractions. Developmental Psychology, 37(3), 431–446. doi:10.1037/0012-1649.37.3.431

EDUCATIONAL RESEARCH AND EVALUATION 17

Martin, C. L., Kornienko, O., Schaefer, D. R., Hanish, L. D., Fabes, R. A., & Goble, P. (2013). The role of sexof peers and gender-typed activities in young children’s peer affiliative networks: A longitudinalanalysis of selection and influence. Child Development, 84(3), 921–937. doi:10.1111/cdev.12032

Morgan, S. L., Gelbgiser, D., & Weeden, K. A. (2013). Feeding the pipeline: Gender, occupational plans,and college major selection. Social Science Research, 42(4), 989–1005.

Mouw, T. (2006). Estimating the causal effect of social capital: A review of recent research. AnnualReview of Sociology, 32, 79–102.

Pahlke, E., Hyde, J. S., & Allison, C. M. (2014). The effects of single-sex compared with coeducationalschooling on students’ performance and attitudes: A meta-analysis. Psychological Bulletin, 140(4),1042–1072. doi:10.1037/a0035740

Reimer, D., & Steinmetz, S. (2009). Highly educated but in the wrong field? Educational specialisationand labour market risks of men and women in Spain and Germany. European Societies, 11(5), 723–746. doi:10.1080/14616690802326400

Riegle-Crumb, C., Farkas, G., & Muller, C. (2006). The role of gender and friendship in advanced coursetaking. Sociology of Education, 79(3), 206–228. doi:10.1177/003804070607900302

Riegle-Crumb, C., King, B., Grodsky, E., &Muller, C. (2012). Themore things change, themore they stay thesame? Prior achievement fails to explain gender inequality in entry into STEM college majors overtime. American Educational Research Journal, 49(6), 1048–1073. doi:10.3102/0002831211435229

Riegle-Crumb, C., & Morton, K. (2017). Gendered expectations: Examining how peers shape femalestudents’ intent to pursue STEM fields. Frontiers in Psychology, 8, 329. doi:10.3389/fpsyg.2017.00329

Robnett, R. D., & Leaper, C. (2013). Friendship groups, personal motivation, and gender in relation tohigh school students’ STEM career interest. Journal of Research on Adolescence, 23(4), 652–664.doi:10.1111/jora.12013

Schneeweis, N., & Zweimüller, M. (2012). Girls, girls, girls: Gender composition and female schoolchoice. Economics of Education Review, 31(4), 482–500. doi:10.1016/j.econedurev.2011.11.002

Seaman, S. R., & White, I. R. (2013). Review of inverse probability weighting for dealing with missingdata. Statistical Methods in Medical Research, 22(3), 278–295. doi:10.1177/0962280210395740

Sewell, W. H. (1971). Inequality of opportunity for higher education. American Sociological Review,36(5), 793–809. doi:10.2307/2093667

Simpson, R. (2005). Men in non-traditional occupations: Career entry, career orrientation and experienceof role strain. Gender, Work & Organization, 12(4), 363–380. doi:10.1111/j.1468-0432.2005.00278.x

Steglich, C., Snijders, T. A. B., & Pearson, M. (2010). Dynamic networks and behavior: Separating selectionfrom influence. Sociological Methodology, 40(1), 329–393. doi:10.1111/j.1467-9531.2010.01225.x

Støren, L. A., & Arnesen, C. Å. (2007). Women’s and men’s choice of higher education –What explainsthe persistent sex segregation in Norway? Studies in Higher Education, 32(2), 253–275. doi:10.1080/03075070701267293

Thompson, J. S. (2003). The effect of single-sex secondary schooling on women’s choice of collegemajor. Sociological Perspectives, 46(2), 257–278. doi:10.1525/sop.2003.46.2.257

United Nations Educational Scientific and Cultural Organization. (2006). International StandardClassification of Education: ISCED 1997. Montreal, Canada: Author.

Van der Vleuten, M., Jaspers, E., Maas, I., & Van der Lippe, T. (2016). Boys’ and girls’ educational choicesin secondary education. The role of gender ideology. Educational Studies, 42(2), 181–200. doi:10.1080/03055698.2016.1160821

Van der Vleuten, M. (2018). Gendered choices: Fields of study of adolescents in the Netherlands (ICS-dis-sertation). Utrecht University, Utrecht.

Van der Vleuten, M., Jaspers, E., Maas, I., & Van der Lippe, T. (2018). Intergenerational transmission ofgender segregation: How parents’ occupational field affects gender differences in field of studychoices. British Educational Research Journal, 44(2), 294–318. doi:10.1002/berj.3329

18 M. VAN DER VLEUTEN ET AL.

Appendices

Appendix 1. Coding of STEM fields

Non-STEM fields

10 Basic/broad, general programmes 761 Child care and youth services80 Literacy and numeracy 762 Social work and counselling90 Personal skills 811 Hotel, restaurant and catering140 Teacher training and education science 812 Travel, tourism and leisure142 Education science 813 Sports143 Training for pre-school teachers 814 Domestic services144 Training for teachers at basic levels 815 Hair and beauty services145 Training for teachers with subject specialization 840 Transport services146 Training for teachers of vocational subjects 850 Environmental protection200 Humanities and Arts 851 Environmental protection technology210 Arts 852 Natural environments and wildlife211 Fine arts 861 Protection of persons and property212 Music and performing arts 862 Occupational health and safety213 Audio-visual techniques and media production 863 Military and defense214 Design

STEM Fields215 Craft skills220 Humanities 400 Science, Mathematics and Computing221 Religion 440 Physical science222 Foreign languages 441 Physics223 Mother tongue 420 Life science225 History and archaeology 421 Biology and biochemistry226 Philosophy and ethics 422 Environmental science310 Social and behavioural science 442 Chemistry311 Psychology 443 Earth science312 Sociology and cultural studies 461 Mathematics313 Political science and civics 462 Statistics314 Economics 481 Computer science320 Journalism and information 482 Computer use321 Journalism and reporting 500 Engineering, Manufacturing & Construction322 Library, information, archive 520 Engineering and engineering trades340 Business and administration 521 Mechanics and metal work341 Wholesale and retail sales 522 Electricity and energy342 Marketing and advertising 523 Electronics and automation343 Finance, banking, insurance 524 Chemical and process344 Accounting and taxation 525 Motor vehicles, ships and aircraft345 Management and administration 541 Food processing346 Secretarial and office work 542 Textiles, clothes, footwear, leather380 Law 543 Materials (wood, paper, plastic, glass)620 Agriculture, forestry and fishery 544 Mining and extraction621 Crop and livestock production 581 Architecture and town planning622 Horticulture 582 Building and civil engineering623 Forestry624 Fisheries640 Veterinary700 Health and Welfare720 Health721 Medicine723 Nursing and caring724 Dental studies725 Medical diagnostic and treatment technology726 Therapy and rehabilitation727 Pharmacy760 Social services

EDUCATIONAL RESEARCH AND EVALUATION 19

Appendix 2. Results of logit regression models that test gender differencesin STEM fields (N = 744). Model 0A is in terms of log odds and Model 0B interms of average marginal effects

Independent variables Model 0A Model 0B

Sex (girl = 1) −1.66*** −0.31***(0.21) (0.03)

ControlsMathematics achievement 0.19* 0.04*

(0.08) (0.01)Math achievement of friends −0.02 −0.00

(0.09) (0.02)Traditional gender norms −0.04 −0.01

(0.24) (0.04)Highest educational level parents −0.08 −0.01

(0.08) (0.02)Non-Western immigrant 0.02 0.00background (0.24) (0.05)Technology-vocational level (ref) – –Science & Health-general level −1.14*** −0.22***

(0.30) (0.06)Science & Health & Technology-general level 0.81 0.17*

(0.43) (0.09)Science & Technology-general level 0.02 0.00

(0.30) (0.06)Science & Health-academic level −0.99** −0.20**

(0.07)Science & Health & Technology-academic level −0.33 −0.07

(0.35) (0.07)Science & Technology-academic level 0.03 0.01

(0.29) (0.06)Constant −0.34 –

(0.82)Model fitHosmer–Lemeshow chi2 1.74 1.74df 8 8p value 0.99 0.99

Source: Wave 2 of Children of Immigrants Longitudinal Survey in Four European Countries and Waves 4 and 5 of Children ofImmigrants Longitudinal Survey in the Netherlands, own calculations. *p < .05. **p < .01. ***p < .001.

Note: Ntotal = 744; standard errors clustered by class (N = 174). Model estimates use inverse-probability weighting toaccount for selective sample attrition and non-response. The standard deviation (SD) is indicated in brackets.

20 M. VAN DER VLEUTEN ET AL.