Content uploaded by Melanie C. Steffens
Author content
All content in this area was uploaded by Melanie C. Steffens on Jun 05, 2014
Content may be subject to copyright.
On the Leaky Math Pipeline:
Comparing Implicit Math-Gender Stereotypes
and Math Withdrawal in Female and Male Children and Adolescents
Melanie C. Steffens, Petra Jelenec, & Peter Noack
Friedrich Schiller University, Jena
Address of corresponding author:
Melanie Steffens
Institut für Psychologie der Friedrich-Schiller-Universität Jena
Am Steiger 3, Haus 1
D-07743 Jena, Germany
E-mail: melanie.steffens@uni-jena.de
Phone:+49-3641-945 111
Fax:+49-3641-945 112
Author Notes
Please address correspondence to Melanie Steffens, Institut fuer Psychologie,
Friedrich-Schiller-Universitaet Jena, Am Steiger 3, Haus 1, D-07743 Jena, Germany. E-mail:
melanie.steffens@uni-jena.de. This research is part of the doctoral dissertation of the second
author, funded by grants from the German Science Foundation (DFG Ste 938/6-1, 9-1) and by
a doctoral dissertation scholarship by the Ministerium für Bildung, Wissenschaft und
Weiterbildung in Rhineland-Palatinate, Germany. We want to thank Janette Schult for
valuable comments on an earlier draft of this paper, and Julia Anheuser, Kristine N.
Goergens, Julia Lichau, Yvonne Still, Miriam Edelmann, Jennie Graef, Saskia Lucht, Katrin
Meinerling, Sarah Nase, Mirco Peveling, Elena Redwanz, Björke Roos, and Kathrin Utz for
their help with regard to planning and running the studies.
Running head: Implicit math-gender stereotypes
Implicit math-gender stereotypes 2
Abstract
Many models assume that habitual human behavior is rather guided by spontaneous,
automatic, or implicit processes than by deliberate, rule-based, or explicit processes. Thus,
math ability self-concepts and math performance could be related to implicit math-gender
stereotypes, in addition to explicit stereotypes. Two studies assessed at what age implicit
math-gender stereotyping can be observed and what the relations between these stereotypes
and math-related outcomes are in children and adolescents. Implicit math-gender stereotypes
could be detected with Implicit Association Tests (IATs) already among 9-year-old girls.
Adolescent girls showed stronger implicit gender stereotypes than adolescent boys, who, on
average, did not reveal implicit gender-stereotypic associations. Girls also showed an implicit
affinity to language versus math already at the age of 9. In a regression analysis, implicit
math-gender stereotypes predicted academic self-concepts, academic achievement, and
enrolment preferences above and beyond explicit math-gender stereotypes for girls, but (with
the exception of achievement) not for boys. These findings suggest implicit gender
stereotypes are an important factor in the dropout of females from math-intensive fields.
Keywords: math-gender stereotypes; implicit stereotypes; IAT; gender differences;
mathematics
Implicit math-gender stereotypes 3
On the Leaky Math Pipeline:
Comparing Implicit Math-Gender Stereotypes
and Math Withdrawal in Female and Male Children and Adolescents
When Ruth Lawrence graduated in math at Oxford University at the age of 13, her
success received much attention in terms of media coverage (BBC News Archive, 1985).
Several years later, this exceptional story found its entrance to the second author’s English
class. Still today, even less spectacular female math or science role models cannot be taken
for granted. The present research investigated stereotypes as factors contributing to this
gender gap in math-related careers. Gender stereotypes stressing the incompetence of females
in math appear to have a great impact on women by lowering their performance and interest
in math (e.g., Davies, Spencer, Quinn, & Gerhardstein, 2002; Spencer, Steele, & Quinn,
1999). Such stereotypes are not necessarily conscious and open to introspection (Greenwald
& Banaji, 1995; Nisbett & Wilson, 1977). Therefore, assessments by way of implicit
measures promise to provide instructive insights (cf. Muzzatti & Agnoli, 2007). Moreover, as
crucial educational and career decisions are made during school years, our focus was on
implicit math-gender stereotypes in preadolescent and adolescent students.
During the last decades women have caught up with men on post-secondary education,
outnumbering male students by the turn of the millenium (e.g., U.S. Department of Education,
2000). However, fewer women than men enter math-intensive fields such as engineering or
computer science, with percentages ranging below one third in the U.S. and one sixth in
Germany (National Science Foundation, 2006; Ramm & Bargel, 2005). Females are much
less motivated to excel in these fields, including physics, than males are (e.g., Taasoobshirazi
& Carr, 2009). According to the Expectancy-Value model of achievement-related choices (for
a review, see Eccles, 2005), socialization processes linked to gender play a crucial role for
emerging domain-specific ability self-concepts. Ability self-concepts in math, in turn, are one
factor explaining the math gender gap (Eccles, 1994). These self-concepts have been shown
to exert an influence on math achievement (Marsh & Yeung, 1997), and they can have a
greater impact on subsequent course selections than math grades (Köller, Daniels, Schnabel,
& Baumert, 2000). Boys’ higher math self-concepts relative to girls’ are particularly
Implicit math-gender stereotypes 4
pronounced in adolescence, and they exceed by far actual performance differences (Hyde,
Fennema, & Lamon, 1990; Hyde, Fennema, Ryan, & Frost, 1990; Hyde, Lindberg, Linn,
Ellis, & Williams, 2008). These gender-specific variations of math self-concepts can already
be observed in grades 3 or 4 of elementary school (e.g., Marsh, 1989).
According to the Expectancy-Value model, factors affecting gender differences in
math-related ability self-concepts are gender role stereotypes, cultural stereotypes of jobs and
school subjects, and gender roles (Eccles, 2005). Students are confronted with math-gender
stereotypes at various occasions, for example, stereotypic beliefs expressed by teachers or
parents (Bhanot & Jovanovic, 2005; Jacobs & Eccles, 1992). The detrimental effect of math-
gender stereotype salience on females’ performance has been demonstrated, referred to as
stereotype threat (Ambady, Shih, Kim, & Pittinsky, 2001; Davies et al., 2002; Keller &
Dauenheimer, 2003; Muzzatti & Agnoli, 2007; Shih, Pittinsky, & Ambady, 1999; Spencer et
al., 1999). However, children and adolescents often disavow math-gender stereotypes when
asked directly (Ambady et al., 2001; Hyde, Fennema, Ryan et al., 1990; Muzzatti & Agnoli,
2007), and young children may reveal ingroup bias instead, stating that their own gender is
more successful in math (Heyman & Legare, 2004; Muzzatti & Agnoli, 2007). Generally,
questionnaires capturing deliberate (or explicit) stereotyping can be influenced by systematic
errors such as self-enhancement and impression management strategies (Strack, 1994).
Moreover, it is possible that students respond in ways that hide their views on these socially
sensitive topics, and questionnaires may be insensitive to more insidious, not necessarily
conscious aspects of stereotyping (Greenwald & Banaji, 1995; Nisbett & Wilson, 1977). Most
importantly, even when math-gender stereotypes were not explicitly endorsed, they negatively
affected girls’ performance (Ambady et al., 2001; Muzzatti & Agnoli, 2007). These findings
stress that questionnaire responses do not capture all aspects of stereotypes that affect girls’
math performance.
This is in line with many current models assuming that human behavior is guided by
spontaneous, automatic, impulsive, or implicit processes in addition to deliberate, rule-based,
controlled, or explicit ones (e.g., Fazio, 1990; e.g., Jacoby, 1991; Strack & Deutsch, 2004).
Implicit processes can be defined as quick reactions triggered automatically and
Implicit math-gender stereotypes 5
unintentionally upon encounter of environmental stimuli. They are often assumed to change
only gradually through learning. Indeed, Rudman and colleagues (2007) have presented
evidence that the sources of implicit attitudes are in childhood. Importantly, overt behavior
may be activated by implicit processes, and “behavioral schemata and their links to other
contents in the impulsive system can be understood as habits.” (Strack & Deutsch, p. 229)
Indeed, it has recently been demonstrated that implicit measures are appropriate for assessing
implicit representations guiding habitual behavior (e.g., Conner, Perugini, O’Gorman, Ayres,
& Prestwich, 2007; Kawakami, Phills, Steele, & Dovidio, 2007).
In order to assess these association-based implicit representations, reaction-time based
measures are typically used that require fast responses. It is still debated whether the primary
distinction should be made between implicit and explicit measures or between implicit and
explicit constructs, and what their relations are. One possibility to explain empirically found
dissociations between measures (e.g., Asendorpf, Banse, & Mücke, 2002) is to assume that
implicit and explicit constructs should be conceptually distinguished (Wilson, Lindsey, &
Schooler, 2000). Another possibility is that implicit and explicit measures generally tap the
same construct, but that motivations to control prejudiced reactions distort explicit measures
(Akrami & Ekehammar, 2005). Also, the difference between what is assessed with implicit
and explicit measures could be gradual rather than absolute (Cunningham & Zelazo, 2007).
Indeed, implicit-explicit correlations depend on several factors (Nosek, 2005), and the
distinction between spontaneous and deliberate reactions may even be more meaningful than
that between implicit and explicit measures (Ranganath, Smith, & Nosek, 2008). For the
present purposes, the take-home message from the ongoing implicit-explicit debate is that
implicit measures predict behavior—including behavior related to one’s self-concept
(Steffens & Schulze-König, 2006), and that the predictive validity of implicit measures
exceeds that of explicit measures particularly with regard to constructs where socially
desirable responding is likely (for a review, see Greenwald, Poehlman, Uhlmann, & Banaji,
2009).
Applied to children’s and adolescents’ domain-specific achievement-related behavior,
it has been assumed “that the conscious and nonconscious choices people make about how to
Implicit math-gender stereotypes 6
spend their time and effort lead, over time, to marked differences between groups and
individuals in life-long achievement-related patterns” and “… that these choices are heavily
influenced by socialization pressures and cultural norms” (Eccles, 2005, p. 10). It is
reasonable to assume that socialization and culture primarily affect individuals’ habitual
reactions. Thus it is possible that they are most adequately assessed with implicit measures.
Setting the stage for the present research, Muzzatti and Agnoli (2007) recently found that
explicit gender stereotyping of mathematics was present in fourth grade, but disappeared in
Italian eighth grade boys and girls. However, the math performance of these very girls
suffered when the gender stereotype was activated, clearly demonstrating the stereotype’s
influence. Those authors speculated that “the stereotype is internalized at this age”, and
“participants are not aware of (or deny) the stereotype, but it is present implicitly” (p. 757-
758). Unfortunately, there is hardly any research assessing implicit gender stereotyping in
children and adolescents, whereas first, implicit math-gender stereotypes have been
investigated in adults, and second, implicit measures of other constructs have been used with
children.
Against this backdrop, the present research set out to study implicit gender stereotypes
of children and adolescents. Using the terms “implicit stereotypes” and “implicit self-
concepts” in the following, we do not mean to imply that implicit and explicit constructs
should necessarily be distinguished, but more modestly refer to stereotypes and self-concepts
assessed with implicit measures. Implicit stereotypes can be viewed as associations between
gender and stereotypic attributes, for example, of math with male or language with female.
Individuals are expected to vary in the extent to which they link math with male so that
implicit associations may differ in strength (Dasgupta & Asgari, 2004). Stereotypic
associations can be activated automatically without intention or control. For example, when
thinking about physicists, male exemplars may come to mind more easily than female ones.
Implicit stereotypes may influence behavior in the absence of awareness of that impact—a
person could be even unaware of holding the stereotype (Greenwald & Banaji, 1995).
Implicit stereotypes are measured using techniques such as Implicit Association Tests
(IATs) (Greenwald, McGhee, & Schwartz, 1998) typically administered in a computerized
Implicit math-gender stereotypes 7
procedure.i IATs were developed to capture the strength of associations between two pairs of
concepts. The general idea is that reactions are fast if associated concepts require the same
reaction, rather than different ones. For instance, sorting a deck of cards into two piles, hearts
and diamonds on the left versus spades and clubs on the right (same color), is faster than
sorting hearts and clubs onto one pile, and diamonds and spades onto another one (different
color). Similarly, people are asked to sort words like computation, boy, poem, girl that are
related to math, male, language, and female, respectively. People who think of math as a male
and language as a female domain should be faster sorting onto one pile the math words with
the male words and onto another the language words with the female words than sorting onto
one pile the math with the female and onto the other, the language with the male words.
Implicit measures are assumed to capture automatically activated cognitions because
due to time pressure, response speed cannot be manipulated as easily as responses to
questionnaires if participants become aware of the purpose of the study. Even if they already
are experienced with IATs and are rewarded for faking, their ability to do so is very limited
(Steffens, 2004). IATs have shown good measurement properties in many studies. Internal
consistencies of IATs are often high, with Cronbach’s α > .80 (see Lane, Banaji, Nosek, &
Greenwald, 2007; Steffens & Buchner, 2003).
Nosek, Banaji, and Greenwald (2002b) assessed implicit math-related cognitions in
college students. Both men and women showed strong math-gender stereotypes, and women
revealed more negative attitudes towards math than men. On a measure of implicit self-
concept, women identified more with arts than math; men as a group did not show a tendency
towards identifying with math or arts. Implicit—but not explicit—math-gender stereotypes
were related to implicit and explicit math attitudes, math self-concepts, and performance. Men
with stronger implicit math-gender stereotypes showed more positive math attitudes, higher
math ability self-concepts, and better performance; conversely, stronger implicit math-gender
stereotypes in women were related to lower math preferences, ability self-concepts, and
performance. In a huge internet sample, women and men showed similar levels of implicit
gender stereotyping (Nosek, Banaji, & Greenwald, 2002a). Another study, with 11th graders
and undergraduates, showed that girls held stronger implicit gender stereotypes than boys
Implicit math-gender stereotypes 8
with regard to physics, and girls also held more negative implicit attitudes toward physics
than boys (Kessels, Rau, & Hannover, 2006). Moreover, women’s implicit attitudes towards
math were more negative after they were subtly reminded of gender as compared to a control
condition (Steele & Ambady, 2006), indicating that a stereotyped identity can affect attitudes
consistent with that identity (cf. also Devos, Blanco, Rico, & Dunn, 2008). In a prospective
study, stronger implicit math-gender stereotypes predicted worse math performance and lower
interest in math-related careers in female college students (Kiefer & Sekaquaptewa, 2007b).
In sum, implicit gender stereotypes seem to play an important role in math-related outcomes
in general and in particular in undermining women’s math interests and performance (Kiefer
& Sekaquaptewa, 2007a).
The Present Research
Major educational and career decisions are made during the school years. We thus
examined whether implicit math stereotypes can already be observed among children and
adolescents, and to what extent these stereotypes affect long-term math-related outcomes. In
Study 1, we tested whether implicit math-gender stereotypes can be found already in
elementary school children. Investigating implicit cognitions in children, Baron and Banaji
(2006) identified implicit preferences for Whites over Blacks already in 6-year-old children
after they had developed a sufficiently elaborated concept of ethnicity (also see Dunham,
Baron, & Banaji, 2006). Similarly, stereotype consciousness develops between the ages of 6
and 10 (McKown & Weinstein, 2003), and implicit gender stereotypes have been detected in
children, using a Stroop-interference task (Most, Sorber, & Cunningham, 2007). Elementary
school children have also become acquainted with the core school subjects math and language
as evidenced, for example, by ever more differentiated academic self-concepts (cf. Marsh,
1989). The onset of explicit gender stereotypes regarding math in girls has been found around
the age of 9 (Muzzatti & Agnoli, 2007). Thus, it is possible that implicit gender stereotyping
is found already in this age group. Conversely, in early adolescence, girls have not always
endorsed math-gender stereotypes, but at the same time, their math performance suffered after
stereotypes were activated (Muzzatti & Agnoli, 2007). Moreover, gender differences in math
performance favoring males emerge in high school (Hyde, Fennema, & Lamon, 1990), and
Implicit math-gender stereotypes 9
girls’ self-concepts and interests become more gender-specific at that age (Hannover, 1991;
Hyde, Fennema, Ryan et al., 1990). Taken together, including 4th, 7th, and 9th graders in our
sample promised to provide particularly instructive insights.
Our second aim was to examine gender differences in implicit stereotyping. As no
gender differences have been reported for math with adults (Nosek et al., 2002a; 2002b), but
for physics with 11th graders (Kessels et al., 2006), we considered it an open question whether
gender differences in implicit math-gender stereotypes would be obtained. To be able to
compare implicit and explicit gender stereotypes, questionnaire measures of stereotypes were
also included. Third, we wanted to examine whether elementary school children already show
an implicit identification with the verbal or the mathematical domain, that is, a differentiated
implicit self-concept. If so, we, fourth, wanted to know whether girls show a stronger implicit
identification with language (or the respective school subject) versus math than boys.
Moreover, in order to replicate established findings, explicit self-concepts, enrolment
intentions, and school grades in math and German were assessed. Gender differences in math
self-concepts and enrolment intentions were expected to favor boys whereas the opposite
should be the case for language/German. Girls should outperform boys in German grades, but
math grades should not necessarily favor boys (Hannover, 1991; Kimball, 1989; Marsh,
1989). Study 2 served as a conceptual replication of Study 1, acknowledging the possibly
gender-biased nature of the typical computer-based assessment procedure employed in Study
1: We thus used a paper-pencil measure. Moreover, the concept German was replaced by
language, avoiding additional connotations of German (e.g., nationality).
As a fifth research question, associations of implicit math-gender stereotypes with
other math-related cognitions and performance were investigated in a joint analysis of the
data from Studies 1 and 2 in order to enhance statistical power. For girls, we expected
stronger implicit math-gender stereotypes to be related to (i) a stronger explicit and implicit
identification with language relative to math; (ii) a stronger enrolment preference for language
as compared to math; and (iii) better German as compared to math grades. For boys, stronger
implicit math-gender stereotypes should be related to a stronger math orientation in self-
concepts, enrolment preferences, and performance. We tested whether implicit stereotypes
Implicit math-gender stereotypes 10
revealed incremental validity predicting the outcome variables above and beyond the
prediction by explicit stereotypes.
Study 1
Implicit math-gender stereotypes and math self-concepts were investigated with IATs
in a cross-sectional sample of 4th, 7th, and 9th graders. We examined at what age implicit math-
gender stereotyping and identification with either academic domain can be observed as well
as possible gender differences in these cognitions. Before completing implicit measures
assessing stereotypes and self-concepts, implicit gender identity was tested. The gender
identity IAT served to make sure that already 4th graders were able to complete the simple
IATs we used. This IAT should differentiate clearly between boys and girls in all three age
groups because an understanding of the concept “gender” is fully established when entering
school (cf. Bussey & Bandura, 1999). Following the implicit measures, participants
responded to questionnaires addressing their explicit stereotypes, self-concepts, enrolment
intentions, and school grades in math and German.
Method
Participants
The initial sample was comprised of N = 147 participants attending several elementary
schools, intermediate-track, and college-bound high-track high schools in different areas,
rural and urban, of western Germany.ii Permissions to conduct the study were granted by
school principals and parents. Students participated in the study during regular school hours.
Seven participants (4th graders: 3; 7th graders: 1; 9th graders: 3) with error rates exceeding 30%
in one or more critical IAT tasks were removed from analysis because excessive numbers of
errors undermine the validity of reaction-time differences (cf. Greenwald et al., 1998).
Altogether, 59 4th graders (mean age = 9.4 years, SD = .51; 32 girls, 27 boys), 39 7th graders
(mean age = 12.8 years, SD = .66; 22 girls, 17 boys) and 42 9th graders (mean age = 15.0
years, SD = .87; 21 girls, 21 boys) were included in the analysis. Of the 7th and 9th graders, 35
attended the intermediate school track, 46 attended high-track schools. All data are collapsed
across school tracks because that variable had no effects on implicit or explicit measures.
Implicit math-gender stereotypes 11
Materials
Implicit measures. An IAT was used to measure the association of domain (math vs.
language) with gender (male vs. female). Stimuli belonging to the four concepts math,
language, male, and female (e.g., computation, poem, boy, and girl) were presented in a
random order on the computer screen, and participants classified them as fast as possible
using only two response keys. This math-gender stereotype IAT comprised two critical tasks.
In the stereotype-congruent task (see the left part of Figure 1), the concepts math and male
were visible in the top left corner of the computer screen whereas the concepts language and
female were visible in the top right corner. Stimuli belonging to these four concepts were
shown one at a time. The stereotype-congruent task required one response for stimuli
belonging to the concepts math or male (e.g., pressing the key “x” with the left hand) and the
other response for stimuli belonging to the concepts language or female (e.g., pressing the key
“m” with the right hand; henceforth, the math-male task). In the stereotype-incongruent task,
the position of the concepts math and language on the computer screen were reversed. Now,
one response was required for stimuli belonging to language or male and the other response
for stimuli belonging to math or female (see the right part of Figure 1; henceforth, the math-
female task). Participants associating math with male and language with female should be
faster in the stereotype-congruent than in the stereotype-incongruent task. Larger latency
differences with faster responses in the math-male task (i.e.: larger IAT effects) are supposed
to indicate stronger stereotypic associations.
In addition to the math-gender stereotype IAT, a gender identity IAT and a math self-
concept IAT were used. In order to hold measures constant across age groups, the IATs were
constructed to be simple enough even for 4th graders. We used only two stimuli per concept;
such IATs have shown good measurement qualities while being less affected by
interindividual differences in cognitive ability and additional connotations (McFarland &
Crouch, 2002; Nosek, Greenwald, & Banaji, 2005; Steffens, Kirschbaum, & Glados, 2008).
Concepts and stimuli can be obtained from Table 1. German was chosen as a concept because
students use German as a common term for their school subject.
Response keys were Y (located where the Z is on Qwerty keyboards) for left and N for
Implicit math-gender stereotypes 12
right responses. False reactions were indicated by a flashing “F!”. Each IAT started with two
practice tasks of 8 trials each where each dimension was familiarized separately (e.g., boys vs.
girls first and then I vs. other, in the gender identity IAT). The 3rd and 5th IAT task were
critical. Each critical task started with two practice trials followed by 48 trials to be analyzed.
In the gender identity IAT, one critical task required the same reactions for stimuli that
belonged to the categories boys or I and the other for girls or other (i.e., the congruent task for
boys). The other critical task required the same reactions for stimuli that belonged to the
categories girls or I and the other for boys or other (i.e., the congruent task for girls). Stimuli
of all four categories of a given IAT were presented in a random order (e.g, me, others, boys,
girls). In the math-self concept IAT, one critical task required the same reactions for stimuli
that belonged to the categories I or math and the other reaction for other or German. The
other critical task required the same reactions for stimuli that belonged to the categories I or
German and the other reaction for other or math. As described above, in the congruent task of
the math-gender stereotype IAT, the categories boys or math as well as girls or German were
paired, in the incongruent task, girls or math and boys or German were paired.
Critical tasks of the first IAT included eight additional practice trials that were
removed from analyses. The 4th IAT task of each IAT served for practicing the classification
that was reversed between the 3rd and the 5th task (e.g., girls vs. boys in the gender identity
IAT) and was extended to 24 trials in order to minimize so-called task order effects (cf. Nosek
et al., 2005). All IAT effects were computed similar to IAT D effects (Greenwald, Nosek, &
Banaji, 2003): in each IAT, the difference between each participant’s average reaction times
in the two critical tasks was divided by the participant’s overall standard deviation of the
response latencies in these tasks; error reaction times were included in the analyses; diverting
from the typical computation of IAT D effects, as in Steffens et al. (2008), no values were
recoded or error penalty used. Whereas these effects were used in statistical analyses, figures
show millisecond differences between the critical tasks.
For reliability estimation, separate IAT effects were computed for trials with odd
versus even position numbers for each IAT. In the gender identity IAT, Pearson correlations
between IATeven effects and IATodd effects were r = .69 for 4th graders and r = .84 for 7th and
Implicit math-gender stereotypes 13
9th graders, showing satisfactory reliability. The gender stereotype IAT revealed a similar
reliability with correlations reaching r = .80 for 4th graders and r = .84 for 7th and 9th graders.
In the math self-concept IAT, correlations were only r = .52 for 4th graders and r = .65 for 7th
and 9th graders.
Explicit ability self-concepts in math and German. To measure ability self-concepts in
math (German), participants rated their agreement to the three statements “I like math
(German)”, “I am good at math (German)”, and “I learn things quickly in math (German)” (cf.
Marsh, 1989). All explicit ratings were made on 5-point scales, with higher values indicating
stronger agreement. Both the math and the German ability self-concept scale showed good
internal consistency, αs = .84 (with all estimates per grade exceeding α = .72).
Enrolment intentions. In 4th grade, enrolment intentions were measured by a single
item (“In high school I am going to choose many math (German) classes”). Participants
attending the 7th or 9th grade were asked to indicate their agreement to the item “I would like
to drop my math (German) classes”. Additionally, 7th and 9th graders from high-track schools
responded to the statement “I can imagine taking advanced math (German) classes for A-
levels”.
School grades. Children in 4th grade were asked to indicate their most recent class test
and report grades in math, dictation, and composition and, additionally, their most recent
report grade in reading. Class test grades of 4th graders in dictation and composition were
averaged to form a German class test index, while the report grades in dictation, composition,
and reading were combined as a German report grade index. Adolescents in 7th or 9th grade
were asked about their latest class test and report grades in math and German. For all
participants, the report grade and class test grade indices were averaged separately for
German and math as they were highly correlated, r = .57 for German, and r = .73 for math.
Explicit gender stereotypes. First, participants responded to four statements referring
to the giftedness of boys and girls in math or German (e.g., “Boys are often talented for doing
German”) that were averaged for a given target. The differences between giftedness ratings
for girls and boys were averaged, resulting in a gender stereotype score comparable to the
IAT effect. Another index was formed from two further items capturing comparative gender
Implicit math-gender stereotypes 14
stereotypes about math and German (e.g., “Math (German) is rather a typical subject for …”),
using boys and girls as anchor points (cf. Nosek et al., 2002b). Finally, these two indices, r =
.48, were averaged.
Procedure
Participants were tested in groups of up to four by female experimenters. IATs were
administered on portable Macintosh computers. Explicit measures were assessed by paper-
pencil questionnaires. To 4th graders, IAT instructions were explained orally and
questionnaire items were read out. All participants started with the IATs, and the order of
IATs was constant for all participants to un-confound person-specific and situational variance,
which is crucial for detecting correlations (e.g., Banse, Seise, & Zerbes, 2001). The gender
identity IAT was completed first, followed by the math self-concept IAT and the gender
stereotype IAT. Task orders within IATs were counterbalanced.iii After the IATs, explicit
measures were applied in the order described above. Finally, participants were debriefed and
rewarded with small gifts. Each session took about 25 minutes.
Analysis
Main dependent variables were IAT effects in the math-gender stereotype, the math
self-concept, and the gender identity IAT. Gender and grade level (4th, 7th, and 9th) were
treated as independent variables. In addition, IAT task order was considered as a control
factor, resulting in a 2 × 3 × 2 between-subjects design. If large gender differences in IAT
effects with an effect size of f = .50 exist, they can be detected with an α = .05 and a sample
size of N = 40 within each grade with a statistical power of 1 – β = .87.
Results
Unless indicated differently, statistical tests in both studies were conducted with
α
=
.05, and individual p-values are not reported for statistically significant effects. The indicator
of the effect size, R
2
p
, is numerically identical to partial η2 and is an estimate of the proportion
of explained variance after partialling out other factors in the design (Cohen, 1977).
Implicit math-gender stereotypes 15
Implicit Gender Identity
Positive difference scores indicate an association self-girls. Applying a known-groups
approach to test whether IATs worked in all age groups, all participants were expected to
show an association self–own gender. Indeed, the 2 × 3 × 2 ANOVA showed a main effect of
gender, F (1,128) = 91.6, R
2
p
= .42, in addition to an unexpected interaction gender × grade, F
(2,128) = 5.45, R
2
p
= .08. Still, simple main effects of gender within grades revealed significant
gender differences in grade 4 (Mgirls = 101 ms, Mboys = -59 ms), F (1,128) = 12.01, R
2
p
= .09, in
grade 7 (Mgirls = 135 ms, Mboys = -172 ms), F (1,128) = 36.79, R
2
p
= .22, and in grade 9 (Mgirls =
125 ms, Mboys = -113 ms), F (128) = 45.74, R
2
p
= .26. Thus, the gender IAT worked in all age
groups, but the size of the effect was somewhat smaller in grade 4. The only other effect
found (all other Fs < 1.68) was a main effect of task order, F (1,128) = 16.39, R
2
p
= .11,
indicating the typical effect that IAT effects were biased in the direction of the task done first
(Greenwald et al., 1998).
Gender Stereotypes
Implicit math-gender stereotype. Larger IAT effects indicate stronger stereotypic
associations math-boys and German-girls. Girls in grade 4 and 9, but not in grade 7, seem to
reveal much stronger stereotypic associations than boys of their age (see Figure 2). This
gender difference appears to be largest in grade 9. The 2 × 3 × 2 ANOVA revealed a large
overall implicit math-gender stereotype, F (1,128) = 25.17, R
2
p
= .16, in addition to a main
effect of task order, F (1,128) = 22.75, R
2
p
= .15 (all other Fs < 3.28). When we tentatively
examined the observed gender difference in grade 9, the simple main effect of gender within
grade was significant, F (1,128) = 4.86, R
2
p
= .04. Six one-sample t-tests against 0 with an
adjusted
α
= .008 to avoid an overall increase of α level (Bortz, 1999) did not reveal
significant stereotypic associations in boys of any age group, all |t|s < 1.72. Implicit gender
stereotypes were not statistically significant in female 7th graders, t (21) = 2.07, R
2
p
= .17 (p =
.025), whereas girls in grade 4 and 9 showed significant stereotyping, t (31) = 4.15, R
2
p
= .36,
and t (20) = 2.96, R
2
p
= .30. Thus, the implicit math-gender stereotyping that we found was
driven by girls’ implicit stereotypes.
Explicit math-gender stereotypes. In contrast to implicit stereotypes, explicit gender
Implicit math-gender stereotypes 16
stereotypes were comparable for boys and girls of all grades. A 2 (gender) × 3 (grade)
ANOVA with the averaged stereotype index as dependent variable revealed no effects (all Fs
< 1). Six one-sample t-tests against 0 (
α
= .008) suggested that girls and boys in all grades
held gender stereotypes, all ts ≥ 2.61. Means are shown in Table 2.
Math Self-Concepts
Implicit ability self-concept. Larger IAT effects indicate stronger associations self-
German and other-math. According to the lower part of Figure 2, these associations seem to
be prevailing among girls in all grades, whereas boys, on average, showed less pronounced
associations of self with either domain. Gender differences seemed to be largest in grade 9.
The 2 × 3 × 2 ANOVA revealed the expected main effect of gender, F (1,128) = 17.16, R
2
p
=
.12, and a main effect of task order, F (1,128) = 53.39, R
2
p
= .29. Girls showed stronger
associations self-German than boys. Additionally, we found a gender × grade interaction, F
(2,128) = 3.11, R
2
p
= .05 (all other Fs < 1.81). Simple main effects of gender within grades
revealed a statistically significant gender difference only in grade 9, F (1,128) = 17.43, R
2
p
=
.12. Six one-sample t-tests against 0 (
α
= .008) showed that boys, on average, did not show
significant associations of self with either math or German, all |ts| < 1.61. Girls showed an
association self-German in each grade: t (31) = 2.92, R
2
p
= .22, in grade 4; t (21) = 2.76, R
2
p
=
.27 in grade 7 (one-tailed p = .006); and t (20) = 3.65, R
2
p
= .40, in grade 9.
Explicit ability self-concepts. Our findings replicate previous ones regarding gender
differences in ability self-concepts (Marsh, 1989). A 2 × 3 ANOVA on the math self-concept
revealed more positive math self-concepts in boys than in girls, F (1,133) = 11.03, R
2
p
= .08
(Table 2). Additionally, older participants showed less favorable math ability ratings than
younger participants, F (2,133) = 7.74, R
2
p
= .10 (replicating Muzzatti & Agnoli, 2007) (all
other Fs < 1). In a parallel ANOVA for the German self-concept scale, girls showed more
favorable German self-concepts than boys, F (1,132) = 3.76, R
2
p
= .03 (all other Fs < 2.72).
Math-related Outcomes
Enrolment intentions. Enrolment intentions for math classes showed gender
differences paralleling those observed for ability self-concepts (cf. Table 2). In detail, the 2 ×
Implicit math-gender stereotypes 17
2 ANOVA addressing the intention to drop math courses revealed that girls had stronger
intentions to drop math than boys, F (1,76) = 4.06, R
2
p
= .05 (all other Fs < 1.96). No other
gender differences were found for enrolment intentions: The parallel ANOVA examining the
intention to drop German courses yielded only a main effect of grade, F (1,76) = 4.52, R
2
p
=
.06 (all other Fs < 1), with 9th graders being less prone to drop German. For high-track
students who were additionally asked about taking advanced math and German courses, no
gender differences were found, either (all Fs < 1.39). Finally, there were no gender effects on
future enrolment intentions in high school regarding German or math courses in 4th graders
(all Fs < 2.03).
School grades. Our expectations of better German grades in girls, but no gender
differences in math grades were confirmed (cf. Table 2). The 2 × 3 ANOVA showed better
German grades for girls than for boys, F (1,131) = 8.45, R
2
p
= .06. Additionally, a main effect
of grade level suggested that the youngest children had better grades than older children, F
(2,131) = 9.80, R
2
p
= .13 (all other Fs < 1). With math grades, a parallel ANOVA did not show
a gender difference (F < 1), but a main effect of grade level, F (2,131) = 13.16, R
2
p
= .17, with
4th graders, again, receiving the best grades (all other Fs < 1.70).
Discussion
The expected implicit gender identities found in all participant groups confirmed that
implicit associations can be measured in children as young as 9 years old using our IATs that
comprised a limited range of simple words as stimuli. This encouraged us to rely on IATs in
the pursuit of our main research questions. In sum, girls revealed implicit math-gender
stereotypes in grade 4 and 9, while implicit stereotypes in female 7th graders failed to reach
the rigorous level of significance after correction for multiple testing. Boys did not show
implicit math-gender stereotypes in any grade, and significant gender differences were found
in grade 9. Explicit stereotype measures did not mirror these gender differences. Moreover,
girls showed implicit ability self-concepts favoring German over math already in grade 4, and
a gender difference with girls leaning stronger towards German than boys could be observed
in grade 9. Boys did not show a differentiated ability self-concept favoring math or German at
any age. Explicit ability self-concepts, enrolment intentions, and grades replicated typical
Implicit math-gender stereotypes 18
findings on gender differences.
Our finding of implicit gender stereotypes already in female 4th graders, and in girls
attending higher grades, but not in boys, is surprising, as previous studies investigating
implicit math-gender stereotypes in adults did not find such gender differences (Nosek et al.,
2002a; 2002b). One reason for these stronger implicit stereotypes in girls could be a
particularly strong stereotype activation during our study. The experimenters observed that
some boys were enthusiastic about doing a computerized task whereas some girls made timid
remarks about computers. Whereas neither boys nor girls had any problems accomplishing
the computerized tasks, gender stereotypes may have become salient for girls, resulting in a
stronger IAT effect in the gender stereotype IAT. This stereotype activation in girls may have
been facilitated by the activation of gender identity by the first IAT. To rule out this activation
explanation, Study 2 served as a replication in a more gender-neutral setting. Moreover, a
larger sample was included in the second study to test whether the lack of significant implicit
gender stereotypes in female 7th graders in Study 1 was due to low statistical power.
In the math versus German self-concept IAT, girls showed an implicit affinity to
German already in grade 4. The gender difference was significant in grade 9 with girls leaning
towards German versus math and boys showing no general tendency in ability self-concepts
towards either academic domain. This pattern of results corresponds to findings reported by
Nosek et al. (2002b) who employed math vs. arts self-concept IATs in adults. Gender
differences in implicit math self-concepts thus seem to develop during adolescence and can
still be found in adulthood. Our results are in line with the finding that girls refrain from math
particularly when reaching puberty, with their self-concepts and interests becoming more
gender-specific (Hannover, 1991; Hyde, Fennema, Ryan et al., 1990).
Study 2
The major goal of Study 2 was to replicate the findings of Study 1 in a more gender-
neutral setting, avoiding both computerized tasks and a gender-related first task. We
examined whether girls would again reveal stronger implicit math gender stereotypes than
boys. A successful replication would discard a situation-based explanation of this gender
difference. We confined our sample to 7th and 9th graders as the gender difference observed in
Implicit math-gender stereotypes 19
Study 1 seemed to be more pronounced in adolescents than in elementary school children, as
these are the age groups where gender differences in math performance are first observed, and
as no statistically significant implicit math-gender stereotyping effect could be demonstrated
for girls from 7th grade in Study 1. A practical advantage of this restriction is that the same
explicit questions could be used for all participants. We expected that again, girls’ ability self-
concepts would lean towards language versus math more than boys’.
Major modifications of the procedure included (i) the use of a paper-pencil IAT
(Teachman, Gapinski, Brownell, Rawlins, & Jeyaram, 2003); (ii) a gender-neutral practice
IAT; and (iii) oral instructions by the experimenters stressing that no ability tests would be
conducted in order to avoid possible concerns about math tests in girls. As a minor change,
the concept German (that might activate representations of nationality) was replaced by
language. Finally, participants did not indicate their gender until the end of the study to avoid
subtle gender priming, and the order of the math-gender stereotype IAT and the math self-
concept IAT was counterbalanced.
Method
Participants
Data of N = 430 participants attending the 7th or 9th grade of several high-track schools
in western Germany were collected. Permissions to conduct the study were granted by school
principals and parents. The adolescents participated in the study during regular school hours.
Participants were excluded from analyses if they had higher error rates than 30% in at least
one IAT sheet (see below) or if they had completely finished at least one such IAT sheet
(because those participants’ IAT effects could be underestimated). Consequently, data of 17
7th graders and 17 9th graders were eliminated. IAT effects of the gender stereotype and the
math self-concept IAT were then checked for outliers, and one additional participant with a
math self-concept IAT effect 3 SD below the mean was excluded from analyses. Altogether,
data of 186 7th graders (mean age = 13.01 years, SD = 0.46; 102 girls, 85 boys) and 209 9th
graders (mean age = 15.05 years, SD = 0.45; 119 girls, 90 boys) were included in the
analyses.
Implicit math-gender stereotypes 20
Materials
Implicit measures. Each IAT consisted of four sheets, two for each critical task. A
sheet contained two columns of 35 items each, and concepts were printed in bold on the top of
a column. In each column, stimuli appeared in a different random order. Participants practiced
the procedure with the concepts mushrooms, trees, small, and big (see below and Table 1 for
stimuli). They were given 30 seconds to classify as many items as possible on an IAT sheet
without skipping items or correcting mistakes. For example, in the stereotype-congruent task
of the gender stereotype IAT, participants ticked the left side for stimuli belonging to boys or
math, the right side for stimuli belonging to girls or language. Participants were asked to
make small ticks instead of crosses.
To compute IAT effects, we first determined the number of correctly classified items
on each IAT sheet. Second, two difference scores were computed for each IAT based on the
first and second sheets of an IAT task, respectively. Third, each difference score was divided
by the constituent with the higher value in order to control for participants’ individual speed.iv
Correlations of these two IAT effects were used for reliability estimation, and the final IAT
effect was computed by averaging these two values. For the gender stereotype IAT, the IAT
effects were correlated with a satisfactory r = .69. In the math self-concept IAT, IAT effects
were correlated significantly, r = .35, but yielded unsatisfactory reliability; this is taken into
account when interpreting the findings.
Explicit measures. Explicit measures capturing ability self-concepts, enrolment
intentions, school grades, and gender stereotypes were identical to those employed in Study 1.
Internal consistencies were α = .81 for the math self-concept scale and α = .79 for the
German self-concept scale. Regarding school grades, the latest class test and report grades, r
= .62 for math and r = .53 for German, were averaged. Intentions to drop math or German
courses and intentions to choose the subject as an advanced course, r = -.62 for math and r = -
.51 for German, were combined, and so were the two explicit stereotypes indices, r = .60.
Additionally, participants responded to two items asking them to estimate to what extend
most other people hold gender stereotypes regarding German and math. Assessments of
perceived stereotypes have been suggested to provide subtle measures of own stereotypes
Implicit math-gender stereotypes 21
(Bohner & Wänke, 2002). For these items, boys and girls were used as anchor points (“How
would most people judge math (German)? Math (German) is a typical subject for…”). Again,
a difference score was computed with higher values indicating stronger stereotyping.
Procedure
After giving their informed consent, all students in a classroom participated
simultaneously in the study. Female experimenters provided oral instructions and handed out
booklets containing IATs and questionnaires. First, the practice IAT was completed. All
participants started with the trees-big/mushrooms-small task, followed by trees-
small/mushrooms-big. The math-gender stereotype IAT and the math self-concept IAT
followed in counterbalanced order. A distractor task was used after the first critical IAT in
order to prevent carry-over effects (a 2-minute visual search task). Task order within each
IAT was counterbalanced as in Study 1. Following the IATs, participants completed the
explicit measures in the order described above. Finally, participants were thanked and
debriefed. Assessments took about 30 minutes.
Analysis
Main dependent variables were IAT effects in the math-gender stereotype IAT and the
math self-concept IAT. Analyses of variance were conducted with grade level (7th vs. 9th) and
gender as independent variables. Including the control factors task order and IAT order
resulted in a 2 × 2 × 2 × 2 between-subjects design. As large gender differences were not
found throughout in Study 1, sample size was increased so that medium-sized gender
differences in IAT scores (effect size: f = .25) could be detected with α = .05 and a sample
size of N = 170 within each grade with a power of 1 –
β
= .90.
Results
Practice IAT
As expected, boys and girls of both grades showed strong associations of trees-
big/mushrooms-small in the practice IAT. All four one-sample t-tests against 0 with an
adjusted α = .0125 were significant, all ts > 16.08, R
2
p = .72. This shows that the paper-pencil
Implicit math-gender stereotypes 22
IATs can reveal adolescents’ implicit associations.
Gender Stereotypes
Implicit gender stereotypes. As can be seen in Figure 3, we replicated our Study 1
findings. Again, girls showed stronger implicit math-gender stereotypes than boys. In fact,
only girls reacted faster in the math-boys task than in the math-girls task. Moreover, implicit
stereotypes seem to be stronger in female 9th graders than in 7th graders. The 2 × 2 × 2 × 2
ANOVA confirmed this gender difference in implicit stereotyping, F (1,380) = 13.89, R
2
p =
.04. Simple main effects tests confirmed that both girls in grade 7, F (1,380) = 5.68, R
2
p = .02,
and in grade 9, F (1,380) = 8.39, R
2
p = .02, showed stronger implicit math-gender stereotypes
than boys. Additionally, participants starting with the math-boys task showed larger IAT
effects, F (1,380) = 28.70, R
2
p = .07, replicating the typical task-order effect. The only other
statistically significant effect was an interaction of task order and IAT order, F (1,380) = 6.24,
R
2
p = .02, that was not of interest (all other Fs < 3.69). Four one-sample t-tests against 0 with α
= .0125 showed implicit math-gender stereotyping for female 9th graders, t (118) = 5.61, R
2
p =
.21, and female 7th graders only, t (101) = 2.43 (p = .01, one-tailed), R
2
p = .06.
Explicit gender stereotypes. A 2 × 2 ANOVA yielded only a main effect of grade
level, with stronger stereotypes observed in grade 9 as compared to grade 7, F (1,392) =
14.46, R
2
p = .04 (all other Fs < 2.17). Diverging from the findings on implicit stereotypes, four
one-sample t-tests against 0 (α = .0125) yielded significant explicit stereotyping for boys and
girls in each grade, all ts > 6.48 (see Table 2).
Participants estimated other people’s stereotypes regarding math and German. The 2 ×
2 ANOVA yielded both a main effect of grade level, F (1,392) = 36.91, R
2
p = .09, and a main
effect of gender, F (1,392) = 9.02, R
2
p = .02 (all other Fs < 1). Both 9th graders as compared to
7th graders and girls as compared to boys perceived stronger stereotypes in their environment.
Again, four one-sample t-tests (α = .0125) revealed significant stereotyping in all participant
groups, all ts > 7.69, with Ms = 1.44 (SD = 172) and 1.81 (1.39) for boys and girls in grade 7
and 2.28 (1.48) and 2.81 (1.40) in grade 9, respectively (possible range: -4 to 4).
Implicit math-gender stereotypes 23
Math Self-Concepts
Implicit ability self-concept. Implicit ability self-concepts generally leaning towards
math or language could not be detected in any participant group (in all four one-sample t-
tests, all ts < 2.10, and in the 2 × 2 × 2 × 2 ANOVA, all Fs < 2.01), perhaps due to the low
internal consistency of the measure.
Explicit ability self-concepts. As expected, a 2 × 2 ANOVA with math self-concept as
dependent variable yielded a main effect of gender, with boys reporting higher math self-
concepts than girls, F (1,392) = 22.33, R
2
p = .05 (all other Fs < 1.28). In a parallel ANOVA
addressing German self-concept, girls showed higher German ability self-concepts than boys,
F (1,392) = 4.58, R
2
p = .01 (all other Fs < 1.92). The effect was statistically significant, but
small (see Table 2).
Outcome Measures
Enrolment intentions. As expected (cf. Table 2), the 2 × 2 ANOVA including the math
enrolment index as dependent variable pointed to stronger enrolment intentions in boys than
in girls, F (1,392) = 28.18, R
2
p = .07 (all other Fs < 1), whereas the same ANOVA concerning
the German enrolment intention index showed stronger enrolment intentions in girls than in
boys, F (1,392) = 10.77, R
2
p = .03 (all other Fs < 1).
School grades. Similarly, a 2 × 2 ANOVA on German grades indicated that girls
earned better German grades than boys, F (1,389) = 18.62, R
2
p = .05 (all other Fs < 3.38). No
gender differences were observed concerning math grades (all Fs < 1.25, cf. Table 2).
Discussion
In Study 2, female 7th and 9th graders showed stronger implicit math-gender
stereotypes than their male classmates. Boys attending grade 7 or 9 did not reveal average
implicit gender stereotypes whereas girls’ responses pointed to considerable stereotyping.
Thus, as it was replicated under more gender-neutral conditions, the stronger math-gender
stereotyping that we found in girls than in boys in Study 1 seems to be a robust finding that is
not confined to specific activation conditions. With regard to the implicit math self-concept,
effects could not be detected for boys or girls. This could be due the lower sensitivity of that
Implicit math-gender stereotypes 24
particular paper-pencil IAT procedure. Girls attending grade 7 showed significant implicit
gender stereotypes in Study 2, whereas that effect had been non-significant in Study 1. As the
effect is numerically smaller in Study 2 than in Study 1, the failure to find it in the first study
clearly appears to be due to a lack of statistical power.
As in Study 1, both boys and girls reported math-gender stereotypes, however, with
older participants stating stronger stereotypes. Interestingly, girls perceived stronger
stereotyping in their environment than boys. Self-concept measures and school grades again
revealed well-documented gender differences.
Relations between implicit math-gender stereotypes and math-related outcomes
In a final set of analyses, we tested associations between implicit math-gender
stereotypes and math-related outcomes. For girls, stronger implicit math-gender stereotypes
should be related to a stronger implicit and explicit identification with language (vs. math), to
stronger enrolment intentions for language (vs. math) classes, and to better German (vs. math)
grades. For boys, stronger implicit stereotyping should be related to a stronger math
orientation (cf. Nosek et al., 2002b). In addition, we examined whether implicit math-gender
stereotypes show incremental validity in predicting math outcomes when included in
regression analyses together with explicit stereotypes.
Data preparation. In order to maximize statistical power, data sets of both studies
were combined.v Within each of the two data sets, z-values of IAT effects in the gender
stereotype and math self-concept IAT were calculated separately for participant groups
starting with either task order. In a second step, data sets were merged. With respect to ability
self-concepts, an index was computed for explicit self-concepts by subtracting the math self-
concept from the German self-concept.vi Concerning achievement and enrolment preferences,
similar difference scores were computed, with higher scores indicating preferences for
German over math.
Correlations. Correlations are shown in Table 3. For girls as well as boys, implicit
math-gender stereotypes and implicit ability self-concepts were related to other measures in
expected ways. For instance, implicit stereotypes went along with better German (vs. math)
grades, and a stronger enrolment preference for German (vs. math) classes in girls, whereas
Implicit math-gender stereotypes 25
the reverse was true for boys. The correlation between implicit stereotypes and implicit ability
self-concepts was significantly stronger in girls than boys.
Regression analyses. Next, we investigated whether implicit gender stereotypes
predict unique variance in outcomes. We carried out four separate hierarchical regressions
with implicit math self-concepts, explicit math self-concepts, achievement, and enrolment
preferences as criteria. Our predictors were participant gender, explicit stereotypes, the
interaction gender × explicit stereotypes, implicit stereotypes, and the interaction gender ×
implicit stereotypes. The interaction terms were of crucial interest as math-gender stereotypes
were expected to have opposite effects on the math orientations of boys and girls.
The pattern of findings was identical for all criteria: Both interaction terms contributed
to the prediction of outcomes in each case (all overall Fs > 7.29, all adj. R2 > .05): for implicit
math self-concepts,
β
gender × explicit stereotypes = -.15, t = -3.43, and
β
gender × implicit stereotypes = -.11, t =
-2.60; for explicit math self-concepts,
β
gender × explicit stereotypes = -.26, t = -6.34,
β
gender × implicit stereotypes
= -.08, t = -1.83, p < .07; for achievement:
β
gender × explicit stereotypes = -.19, t = -4.40, and
β
gender × implicit
stereotypes = -.16, t = -3.69; for enrolment preferences:
β
gender × explicit stereotypes = -.27, t = -6.66,
β
gender ×
implicit stereotypes = -.10, t = -2.55). Altogether, implicit stereotypes interacted with gender in
predicting unique variance in implicit math self-concepts, achievement, and enrolment
intentions, and in a weaker way in predicting explicit math self-concepts.vii
To disentangle these interaction effects, separate analyses were conducted for boys
and girls. First, the explicit gender stereotype was entered as a predictor; second, the implicit
gender stereotype was added. For girls, implicit gender stereotypes showed unique predictive
power with regard to the implicit and explicit math ability self-concept, school grades, and
enrolment preference (see Table 4). As these findings show, our paper-pencil IATs apparently
measured individual differences in implicit math self-concepts, despite missing mean effects
on the group level. For boys, only explicit gender stereotypes predicted explicit math self-
concepts and enrolment preferences, but explicit and implicit stereotypes both predicted their
school grades.viii In sum, implicit gender stereotypes showed incremental validity beyond
explicit stereotypes in predicting all outcome variables only for girls.
Implicit math-gender stereotypes 26
General Discussion
We investigated implicit math-gender stereotypes and implicit math self-concepts, and
their associations with math-related outcomes. Implicit math-gender stereotypes were
observed among girls as early as 9 years of age. Female adolescents also showed implicit
stereotypes that clearly exceeded those of their male classmates who did not reveal any
stereotypic associations. At the same time, there were no gender differences in explicit gender
stereotypes. Moreover, girls aged 9 already showed implicit ability self-concepts leaning
towards German and away from math (Study 1) while boys on average did not show any
implicit affinity to math or language/German at any age. Gender differences in implicit math
self-concepts were significant in grade 9. Beyond explicit stereotypes implicit math-gender
stereotypes predicted implicit and explicit math self-concepts, enrolment preferences, and
school grades for girls, but only school grades for boys. These findings add to the growing
body of evidence supporting the predictive validity of implicit measures.
Implicit stereotypes were accompanied by explicit stereotypes, and both were present
in younger and older girls. The finding that adolescent girls in the current study reported equal
or even stronger explicit math-gender stereotyping than younger girls diverges from the
previous findings on Italian students’ math-gender stereotypes that had declined by 8th grade
(Muzzatti & Agnoli, 2007). It thus appears that implicit and explicit gender stereotypes
regarding school subjects form rather early, similar to implicit ethnic biases (Baron & Banaji,
2006) and explicit ethnic stereotypes (McKown & Weinstein, 2003). Whether explicit
stereotypes remain stable or decline apparently depends on the cultural context: As Muzzatti
and Agnoli reported, Italian girls aged 14-19 achieve better math grades than boys, and this is
not the case in Germany. On average, it seems that socially desirable responding was not an
issue in our sample—both girls and boys admitted to hold gender stereotypes concerning
academic performance. Still, girls’ implicit math-gender stereotypes demonstrated
incremental validity over explicit stereotypes. This indicates that either, different girls
responded in a socially desirable way to different degrees (corrupting the explicit measures),
or girls lacked the awareness of their implicit math-gender stereotypes that would be
necessary to accurately report them, or both.
Implicit math-gender stereotypes 27
Our finding that implicit gender stereotypes can be found in girls already at the age of
9 is consistent with other research. For example, elementary school girls associated high
spelling skills with girls (Heyman & Legare, 2004). Likewise, girls aged 6-10 years rated
women as less interested and less competent in math than men (Steele, 2003). At first sight,
other findings from that study appear to contradict the implicit gender stereotypes we found:
Girls did not report any negative math stereotypes regarding girls (Steele, 2003). The
restriction of the math-gender stereotype to adults was also observed when assessed by a more
indirect measure where girls had to specify the gender of a mathematically talented person
they heard about in a short story. Girls thought of an adult mathematician most often as a
man, but supposed the mathematically talented child most frequently to be a girl (see also
Ambady et al., 2001). How can these findings be reconciled with the implicit gender
stereotypes we found? First, our implicit measure could have revealed stereotypes the girls
were not aware of or willing to tell. Moreover, the implicit measure we used (the IAT) does
not make a clear distinction between math stereotypes regarding girls vs. women,
respectively, but merely measures associations with the basic category female. Thus, implicit
math-gender stereotypes in younger—and also older—girls may partly reflect their
knowledge of females participating less in math-intensive fields. Finally, as both math and
language stereotypes contribute to the implicit gender stereotype found in the IAT, the
implicit stereotype could be based on strong language-girls associations even in the absence
of pronounced math-boys associations.
In our study, elementary school boys did not show implicit math-gender stereotyping.
This result is consistent with other findings. For example, boys attending grade 1-8 did not
report any math-gender stereotypes (Ambady et al., 2001; Steele, 2003). A more indirect
measure yielded matching evidence. In the gender specification task by Steele (2003), boys
aged 6-10 assumed that both adults and children excelling in math or spelling would be male.
Thus, elementary school boys revealed a strong in-group bias rather than gender stereotyping
in line with cultural representations.
As we further found, adolescent boys showed weaker implicit math-gender stereotypes
than adolescent girls, and male 7th and 9th graders did not reveal significant stereotypic
Implicit math-gender stereotypes 28
associations. This is a somewhat surprising finding, as these boys have most likely been
exposed to gender stereotypes during their socialization, and they also reported explicit math-
gender stereotypes (e.g., Jacobs & Eccles, 1992). These results diverge from the early
development of implicit bias demonstrated by Baron and Banaji (2006). However, it should
be noted that implicit biases (i.e., affective associations—of social groups with positive or
negative valence) clearly differ from implicit stereotypes (i.e., semantic associations—of
social groups with traits or domains), and much more research is needed to delineate the
development of implicit biases and stereotypes with regard to different social groups.
Moreover, the gender differences in implicit stereotyping in our adolescent sample do
not correspond to the absence of such gender differences in the adult samples as studied by
Nosek et al. (2002a, 2002b). Note, however, that the studies differed in the concepts
investigated. Whereas Nosek et al. chose arts or liberal arts as an academic domain, we used
German or language. We cannot exclude the possibility that the concept liberal arts has a
more deeply engraved female connotation than language, which, in turn, would result in
stronger implicit gender stereotypes when math is contrasted with liberal arts than with
language. Similar to the weaker implicit math-gender stereotypes that we found, Kessels and
colleagues (2006) reported weaker implicit physics-gender stereotypes in adolescent boys
than girls. One possibility to explain our finding is to assume that boys have self-serving
associations of both math and language with their own gender (cf. Rudman, Greenwald, &
McGhee, 2001, for basic research on self-serving implicit gender stereotypes). This would be
in line with Steele’s (2003) finding reported above that boys imagine someone excelling in
either task is male. Boys’ self-serving associations would lead to fast reaction times both in
the boys-math and the boys-language task of our IAT, leading to small IAT effects. This
explanation was corroborated in recent studies disentangling implicit stereotypes regarding
math and language with a variant of the IAT (Nosek & Banaji, 2001) that allows separately
assessing math gender stereotypes and language gender stereotypes: Boys associated both
math and language with their gender group (Steffens & Jelenec, 2009). Thus, it appears that
the lack of implicit gender stereotypes we found in boys is another instance of their ingroup
bias regarding both domains.
Implicit math-gender stereotypes 29
Whereas Nosek et al. (2002b) detected strong relations between implicit math-gender
stereotyping and performance for men (r = .51), this relation was generally small for boys in
the current studies. Moreover, we did not find associations between implicit gender
stereotypes and math self-concepts for boys. At first glance, the lack of stereotypic
associations among boys could suggest that the implicit measures were not valid in the male
subsample. However, we found significant linkages of implicit stereotypes and both school
grades and enrolment intentions, and boys’ implicit math-gender stereotypes were also related
with their explicit gender stereotypes and their perceptions of other people’s stereotypes.
These findings support the validity of the gender stereotype IATs also among male students.
It thus seems that gender differences in implicit math versus language gender stereotyping are
a valid finding.
In Study 1, girls aged 9 years did not only show implicit gender stereotypes, but also
an early implicit affinity to German rather than math. Boys of all grades as a group did not
show an implicit affinity to any academic domain. In our computerized math self-concept
IAT, gender differences were significant in grade 9 with girls revealing both stronger implicit
gender stereotypes and a stronger affinity to German versus math than boys. Due to
unsatisfactory measurement properties of the paper-pencil math self-concept IAT in Study 2,
the lack of relations between it and the outcome variables cannot be interpreted. The implicit
gender stereotypes we found with computerized procedures in Study 1 were replicated in
Study 2 with paper-pencil measures. This rules out our previous concern that strong implicit
gender stereotypes in girls are only found if activated by the study procedure. In sum, we
therefore recommend the use of the more sensitive computerized techniques to assess implicit
cognitions rather than paper-pencil formats. Low internal consistencies have often been
reported for implicit measures used in social-cognition research (e.g., Bosson, Swann, &
Pennebaker, 2000; De Houwer, 2003; Nosek & Banaji, 2001), and it has previously been
concluded that these measures appear sensitive enough to detect group-based differences and
correlations, but that they are unsuited as of yet for individual diagnosis (e.g., Steffens &
Buchner, 2003).
It is unclear why implicit math-gender stereotypes consistently predicted unique
Implicit math-gender stereotypes 30
variance in implicit and explicit math self-concept, enrolment preferences, and school grades
only for girls. Work on behavior prediction with implicit measures has identified several
crucial moderators (for a review, see Friese, Hofmann, & Schmitt, 2008). In addition to the
habitualness of behavior (e.g., Conner et al., 2007), according to Friese and colleagues’
analysis, individual-difference variables identified as moderators are: Availability of
cognitive-control resources, mood, and the degree to which individuals rely on either
controlled or automatic processes. As to the latter point, for example, after implicit attitudes
were artificially changed via a transparent conditioning-like procedure, people relied less on
their implicit attitudes in their choice behavior (Ebert, Steffens, von Stülpnagel, & Jelenec,
2009). Another recently identified moderator is mindfulness (Levesque & Brown, 2007).
Applied to the current findings, we do not believe that girls’ working-memory capacity,
mood, or self-control (e.g., Dislich, Zinkernagel, Ortner, & Schmitt, in press) is generally
lower than boys’. This is corroborated by the fact that explicit measures similarly contributed
to explaining girls’ and boys’ outcome variables, which is unexpected from such an
individual-difference analysis. We interpret our finding that implicit math-gender stereotypes
are consistently related only to girls’ math-related outcomes as a hint that implicit gender
stereotypes restrict girls’, but not boys’, school-related choices. Holding the stereotype that
math is a boys’ subject may contribute to girls’ (habitual) decisions not to study as hard for
math, to their conviction that they do not (need not? should not?) excel in math, and finally,
their decisions to drop out of math. Whereas boys also know of ability-related gender
stereotypes, they may still feel free to excel wherever they are good at. Drawing from findings
in a very different context, one could also speculate that girls are more sensitized to math-
stereotype cues than boys (cf. Wiers, Van Woerden, Smulders, & De Jong, 2002).
The links we found between implicit gender stereotypes, domain-specific ability self-
concepts, and achievement-related choices and performance corroborate the Expectancy-
Value model and the notion that these couplings can be guided by spontaneous, automatic, or
implicit processes (Eccles, 2005). Girls’ implicit gender stereotypes could be a reason why
stronger links between domain-specific achievement, explicit self-concept, and interest have
been found in boys as compared to girls (Denissen, Zarrett, & Eccles, 2007).
Implicit math-gender stereotypes 31
A limitation of our study is that we were unable to determine at what age implicit
math-gender stereotyping is first established. We found these stereotypes already in 9-year-
old girls, the youngest age group observed, and they were already related to outcome
measures this early (cf. Footnote 7). These findings surprised us, as we had rather expected
that implicit stereotypes and their relations to behavior increase with age. It should be noted
that the onset question is difficult to answer: There cannot be an empirical proof that implicit
math-gender stereotyping does not yet exist in a specific age group—finding no evidence of it
can always be due to a lack of sensitivity of the implicit measure used. We considered it
unlikely that implicit math-gender stereotypes are found in still younger girls because those
have shown explicit counterstereotypes in previous studies that would manifest in an implicit
association of girls with math (e.g., Muzzatti & Agnoli, 2007). However, in the light of the
present findings, testing this in future studies would be very interesting with regard to the
development of implicit and explicit stereotypes. Such research could use Baron and Banaji’s
(2006) child IAT and would need to establish that this method is sensitive enough to detect
math-gender stereotypes. If implicit math-gender stereotypes were detected in girls before
explicit ones, this early “internalization” could explain the power these stereotypes appear to
have over girls. We can conclude from our study that math-gender stereotypes can be
demonstrated at a very early age, at the same time when they are first explicitly endorsed, and
long before gender differences in math performance are observed.
An additional limitation of our study is that the correlational nature of our data does
not allow conclusions regarding causal relations. We assume that implicit gender stereotypes
shape self-concept variables or performance, and at the same time, personal ability
estimations or achievement also influence stereotypic associations about gender. Only studies
employing longitudinal designs can clarify this issue (cf. Kiefer & Sekaquaptewa, 2007b).
A limitation of implicit measures as used here is that the associations they measure are
generally ambiguous (cf. Fiedler, Messner, & Bluemke, 2006). Moreover, one might ask
whether differences in the size of IAT effects across age groups could be due to
methodological features of IATs. For instance, in Study 1 implicit gender identities (as
reflected in associations I-girl versus I-boy) appeared less pronounced in 4th grade than in 7th
Implicit math-gender stereotypes 32
or 9th grade. This could indicate that children’s cognitive development facilitates larger IAT
effects at a later age (cf. Hummert, Garstka, O'Brien, Greenwald, & Mellott, 2002, for a
discussion with regard to old age). For instance, correlations of IAT effects with fluid
intelligence have been demonstrated, indicating that more intelligent adults may show larger
IAT effects (von Stülpnagel & Steffens, in press). The fact that neither math-gender
stereotypes nor math-self concepts appeared smaller in 4th than 7th or 9th grade speaks against a
methodological explanation. Additionally, our finding that implicit gender identity appears
more pronounced or chronically accessible during puberty is also in line with other research
(Hannover, 1991; Hill & Lynch, 1983; Hyde, Fennema, Ryan et al., 1990). Still, further
systematic investigations and comparisons of implicit measures are needed to assess their
limitations and virtues (cf. Steffens & Jonas, in press).
Conclusion
Implicit math-gender stereotypes could be observed in 9-year-old girls and turned out
to be strong in female adolescents. Girls’ implicit gender stereotypes uniquely predicted math
self-concepts, achievement, and enrolment preferences across age groups. Together with an
early implicit affinity to German/language vs. math, these findings suggest that implicit
processes exert their influence on girls already at an early age and possibly diminish girls’
commitment to math-intensive fields. Educational policies should counteract implicitly
operating biases already in young girls (e.g., Kessels et al., 2006). It could well be that
outstanding female role models such as Ruth Lawrence as well as female math professionals
with a more attainable level of success play a crucial role in changing career-related
cognitions and decisions in female students (e.g., Dasgupta & Asgari, 2004).
Future psychological research is necessary to identify first, what determines children’s
math-gender stereotypes. As individual differences in these stereotypes are relevant for
behavior and intentions, their roots should be discovered. For instance, stereotypes could stem
from parents’ interests (Dotterer, McHale, & Crouter, 2009), lack of role models, peers’
stereotypes and abilities, kindergarten and school teachers’ interactions (e.g., Curby, Rimm-
Kaufman, & Ponitz, 2009) as well as teachers’ expectations (Hinnant, O'Brien, & Ghazarian,
2009), particularly gender-specific aspects of interactions and expectations. The roots of
Implicit math-gender stereotypes 33
implicit stereotypes could well differ from those determining explicit stereotypes (cf. Rudman
et al., 2007).
Second, research should identify effective intervention strategies that address females’
refraining from math and science. Even in the presence of implicit stereotypes, it is possible
that eliciting positive math-related emotions in girls (cf. Daniels et al., 2009) helps them to
overcome stereotypic career choices. Moreover, short-term interventions such as “imagining
stereotypes away” (Blair, Ma, & Lenton, 2001) could show positive long-term effects.
Finally, it should be tested whether the exposure to female math and physics university
students that is part of many universities’ “girls’ days” effectively counteracts implicit gender
stereotypes or whether more extensive interventions are needed, and with younger
adolescents.
Implicit math-gender stereotypes 34
References
Akrami, N., & Ekehammar, B. (2005). The association between implicit and explicit
prejudice: The moderating role of motivation to control prejudiced reactions.
Scandinavian Journal of Psychology, 46, 361-366.
Ambady, N., Shih, M., Kim, A., & Pittinsky, T. L. (2001). Stereotype susceptibility in
children: Effects of identity activation on quantitative performance. Psychological
Science, 12, 385-390.
Asendorpf, J. B., Banse, R., & Mücke, D. (2002). Double dissociation between implicit and
explicit personality self-concept: The case of shy behavior. Journal of Personality and
Social Psychology, 83, 380-393.
Banse, R., Seise, J., & Zerbes, N. (2001). Implicit attitudes towards homosexuality:
Reliability, validity, and controllability of the IAT. Zeitschrift für Experimentelle
Psychologie, 48, 145-160.
Baron, A. S., & Banaji, M. R. (2006). The development of implicit attitudes: Evidence of race
evaluations from ages 6,10, and adulthood. Psychological Science, 17, 53-58.
BBC News Archive. (1985). Retrieved 29.09.2007, from
http://news.bbc.co.uk/onthisday/hi/dates/stories/july/4/newsid_2492000/2492853.stm
Bhanot, R., & Jovanovic, J. (2005). Do parents' academic gender stereotypes influence
whether they intrude on their children's homework? Sex Roles, 52, 597-607.
Blair, I. V., Ma, J. E., & Lenton, A. P. (2001). Imagining stereotypes away: The moderation
of implicit stereotypes through mental imagery. Journal of Personality and Social
Psychology, 81, 828-841.
Bohner, G., & Wänke, M. (2002). Attitudes and attitude change. Hove, UK: Psychology
Press.
Bortz, J. (1999). Statistik für Sozialwissenschaftler. Berlin: Springer.
Bosson, J. K., Swann, W. B., Jr., & Pennebaker, J. W. (2000). Stalking the perfect measure of
implicit self-esteem: The blind men and the elephant revisited? Journal of Personality
and Social Psychology, 79, 631-643.
Bussey, K., & Bandura, A. (1999). Social cognitive theory of gender development and
differentiation. Psychological Review, 106, 676-713.
Cohen, J. (1977). Statistical power analysis for the behavioral sciences (revised ed.).
Hillsdale, NJ: Erlbaum.
Conner, M. T., Perugini, M., O'Gorman, R., Ayres, K., & Prestwich, A. (2007). Relations
between implicit and explicit measures of attitudes and measures of behavior:
Evidence of moderation by individual difference variables. Personality and Social
Psychology Bulletin, 33, 1727-1740.
Cunningham, W. A., & Zelazo, P. D. (2007). Attitudes and evaluations: A social cognitive
neuroscience perspective. Trends in Cognitive Sciences, 11, 97-104.
Curby, T. W., Rimm-Kaufman, S. E., & Ponitz, C. C. (2009). Teacher-child interactions and
children's achievement trajectories across kindergarten and first grade. Journal of
Educational Psychology, 101, 912-925.
Daniels, L. M., Stupnisky, R. H., Pekrun, R., Haynes, T. L., Perry, R. P., & Newall, N. E.
(2009). A longitudinal analysis of achievement goals: From affective antecedents to
emotional effects and achievement outcomes. Journal of Educational Psychology,
101, 948-963.
Dasgupta, N., & Asgari, S. (2004). Seeing is believing: Exposure to counterstereotypic
women leaders and its effect on the malleability of automatic gender stereotyping.
Journal of Experimental Social Psychology, 40, 642-658.
Implicit math-gender stereotypes 35
Davies, P. G., Spencer, S. J., Quinn, D. M., & Gerhardstein, R. (2002). Consuming images:
How television commercials that elicit stereotype threat can restrain women
academically and professionally. Personality and Social Psychology Bulletin, 28,
1615-1628.
De Houwer, J. (2003). The extrinsic affective Simon task. Experimental Psychology, 50, 77-
85.
Denissen, J. J. A., Zarrett, N. R., & Eccles, J. S. (2007). I like to do it, I'm able, and I know I
am: Longitudinal couplings between domain-specific achievement, self-concept, and
interest. Child Development, 78, 430-447.
Devos, T., Blanco, K., Rico, F., & Dunn, R. (2008). The role of parenthood and college
education in the self-concept of college students: Explicit and implicit assessments of
gendered aspirations. Sex Roles, 59, 214-228.
Dislich, F. X. R., Zinkernagel, A., Ortner, T. M., & Schmitt, M. (in press). Convergence of
direct, indirect, and objective risk taking measures in the domain of gambling: The
moderating role of impulsiveness and self-control. Zeitschrift für Psychologie/Journal
of Psychology.
Dotterer, A. M., McHale, S. M., & Crouter, A. C. (2009). The development and correlates of
academic interests from childhood through adolescence. Journal of Educational
Psychology, 101, 509-519.
Dunham, Y., Baron, A. S., & Banaji, M. R. (2006). From American city to Japanese village:
A cross-cultural investigation of implicit race attitudes. Child Development, 77, 1268-
1281.
Ebert, I. D., Steffens, M. C., von Stülpnagel, R., & Jelenec, P. (2009). How to like yourself
better, or chocolate, less: Changing implicit attitudes with one IAT task. Journal of
Experimental Social Psychology, 45, 1098-1104.
Eccles, J. S. (1994). Understanding women's educational and occupational choices.
Psychology of Women Quarterly, 18, 585-609.
Eccles, J. S. (2005). Studying gender and ethnic differences in participation in math, physical
science, and information technology. In J. E. Jacobs & S. D. Simpkins (Eds.), Leaks in
the pipeliine to math, science, and technology careers (pp. 7-14). San Francisco:
Jossey-Bass.
Fazio, R. H. (1990). Multiple processes by which attitudes guide behavior: The MODE model
as an integrative framework. In M. P. Zanna (Ed.), Advances in experimental social
psychology, Vol. 23 (pp. 75-109). Orlando, FL, USA: Academic Press.
Fiedler, K., Messner, C., & Bluemke, M. (2006). Unresolved problems with the "I", the "A"
and the "T": Logical and psychometric critique of the Implicit Association Test (IAT).
European Review of Social Psychology, 17, 74-147.
Friese, M., Hofmann, W., & Schmitt, M. (2008). When and why do implicit measures predict
behaviour? Empirical evidence for the moderating role of opportunity, motivation, and
process reliance. European Review of Social Psychology, 19, 285-338.
Greenwald, A. G., & Banaji, M. R. (1995). Implicit social cognition: Attitudes, self-esteem,
and stereotypes. Psychological Review, 102, 4-27.
Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring individual
differences in implicit cognition: The Implicit Association Test. Journal of
Personality and Social Psychology, 74, 1464-1480.
Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003). Understanding and using the
Implicit Association Test: I. An improved scoring algorithm. Journal of Personality
and Social Psychology, 85, 197-216.
Greenwald, A. G., Poehlman, A. T., Uhlmann, E. L., & Banaji, M. R. (2009). Understanding
and using the Implicit Association Test: III. Meta-analysis of predictive validity.
Journal of Personality and Social Psychology, 97, 17-41.
Implicit math-gender stereotypes 36
Hannover, B. (1991). Zur Unterrepräsentanz von Mädchen in Naturwissenschaften und
Technik: Psychologische Prädiktoren der Fach- und Berufswahl (Underrepresentation
of girls in the natural sciences and technology: Psychological predictors of subject
choice and for choice of occupation). Zeitschrift für Pädagogische Psychologie/
German Journal of Educational Psychology, 5, 169-186.
Heyman, G. D., & Legare, C. H. (2004). Children's beliefs about gender differences in the
academic and social domains. Sex Roles, 50, 227-236.
Hill, J. P., & Lynch, M. E. (1983). The intensification of gender-related expectations during
adolescence. In J. Brooks-Gunn & A. Peterson (Eds.), Girls at puberty (pp. 201-228).
New York: Plenum Press.
Hinnant, J. B., O'Brien, M., & Ghazarian, S. R. (2009). The longitudinal relations of teacher
expectations to achievement in the early school years. Journal of Educational
Psychology, 101, 662-670.
Hofmann, W., Gawronski, B., Gschwender, T., Le, H., & Schmitt, M. (2005). A meta-
analysis on the correlation between the Implicit Association Test and explicit self-
report measures. Personality and Social Psychology Bulletin, 31.
Hummert, M. L., Garstka, T. A., O'Brien, L. T., Greenwald, A. G., & Mellott, D. S. (2002).
Using the implicit association test to measure age differences in implicit social
cognitions. Psychology and Aging, 17, 482-495.
Hyde, J. S., Fennema, E., & Lamon, S. J. (1990). Gender differences in mathematics
performance: A meta-analysis. Psychological Bulletin, 107, 139-155.
Hyde, J. S., Fennema, E., Ryan, M., & Frost, L. A. (1990). Gender comparisons of
mathematics attitudes and affect: A meta-analysis. Psychology of Women Quarterly,
14, 299-324.
Hyde, J. S., Lindberg, S. M., Linn, M. C., Ellis, A. B., & Williams, C. C. (2008). Gender
similarities characterize math performance. Science, 321, 494-495.
Jacobs, J. E., & Eccles, J. S. (1992). The impact of mothers' gender-role stereotypic beliefs on
mothers' and children's ability perceptions. Journal of Personality and Social
Psychology, 63, 932-944.
Jacoby, L. L. (1991). A process dissociation framework: Separating automatic from
intentional uses of memory. Journal of Memory and Language, 30, 513-541.
Kawakami, K., Phills, C. E., Steele, J. R., & Dovidio, J. F. (2007). (Close) distance makes the
heart grow fonder: Improving implicit racial attitudes and interracial interactions
through approach behaviors. Journal of Personality and Social Psychology, 92, 957-
971.
Keller, J., & Dauenheimer, D. (2003). Stereotype threat in the classroom: Dejection mediates
the disrupting threat effect on women's math performance. Personality and Social
Psychology Bulletin, 29, 371-381.
Kessels, U., Rau, M., & Hannover, B. (2006). What goes well with physics? Measuring and
altering the image of science. British Journal of Educational Psychology, 76, 761-780.
Kiefer, A. K., & Sekaquaptewa, D. (2007a). Implicit stereotypes and women's math
performance: How implicit gender-math stereotypes influence women's susceptibility
to stereotype threat. Journal of Experimental Social Psychology, 43, 825-833.
Kiefer, A. K., & Sekaquaptewa, D. (2007b). Implicit stereotypes, gender identification, and
math-related outcomes: A prospective study of female college students. Psychological
Science, 18, 13-18.
Kimball, M. M. (1989). A new perspective on women's math achievement. Psychological
Bulletin, 105, 198-214.
Köller, O., Daniels, Z., Schnabel, K., & Baumert, J. (2000). Kurswahlen von Mädchen und
Jungen im Fach Mathematik: Die Rolle des fachspezifischen Selbstkonzepts und
Interesses (Course selection of girls and boys in mathematics: The role of academic
Implicit math-gender stereotypes 37
self-concept and interest). Zeitschrift für Pädagogische Psychologie/ German Journal
of Educational Psychology, 14, 26-37.
Lane, K. A., Banaji, M. R., Nosek, B. A., & Greenwald, A. G. (2007). Understanding and
using the implicit association test: IV: What we know (so far) about the method. In B.
S. Wittenbrink, N.S. (Ed.), Implicit measures of attitudes: Procedures and
controversies (pp. 59-102). New York.
Levesque, C., & Brown, K. W. (2007). Mindfulness as a moderator of the effect of implicit
motivational self-concept on day-to-day behavioral motivation. Motivation and
Emotion, 31, 284-299.
Marsh, H. W. (1989). Age and sex effects in multiple dimensions of self-concept:
Preadolescence to early adulthood. Journal of Educational Psychology, 81, 417-430.
Marsh, H. W., & Yeung, A. S. (1997). Causal effects of academic self-concept on academic
achievement: Structural equation models of longitudinal data. Journal of Educational
Psychology, 89, 41-54.
McFarland, S. G., & Crouch, Z. (2002). A cognitive skill confound on the Implicit
Association Test. Social Cognition, 20, 483-510.
McKown, C., & Weinstein, R. S. (2003). The development and consequences of stereotype
consciousness in middle childhood. Child Development, 74, 498-515.
Most, S. B., Sorber, A. V., & Cunningham, J. G. (2007). Auditory Stroop reveals implicit
gender associations in adults and children. Journal of Experimental Social
Psychology, 43, 287.
Muzzatti, B., & Agnoli, F. (2007). Gender and mathematics: Attitudes and stereotype threat
susceptibility in Italian children. Developmental Psychology, 43, 747-759.
National Science Foundation. (2006). Women, minorities, and persons with disabilities in
science and engineering. Retrieved 14.02.2007, from
http://www.nsf.gov/statistics/wmpd/sex.htm
Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Verbal reports on
mental processes. Psychological Review, 84, 231-259.
Nosek, B. A. (2005). Moderators of the relationship between implicit and explicit evaluation.
Journal of Experimental Psychology: General, 134, 565-584.
Nosek, B. A., Banaji, M., & Greenwald, A. G. (2002a). Harvesting implicit group attitudes
and beliefs from a demonstration web site. Group Dynamics, 6, 101-115.
Nosek, B. A., & Banaji, M. R. (2001). The Go/No-go Association Task. Social Cognition, 19,
625-666.
Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002b). Math = male, me = female,
therefore math not= me. Journal of Personality and Social Psychology, 83, 44-59.
Nosek, B. A., Greenwald, A. G., & Banaji, M. R. (2005). Understanding and using the
Implicit Association Test: II. Method variables and construct validity. Personality &
Social Psychology Bulletin, 31, 166-180.
Nosek, B. A., & Lane, K. A. (1999). Analyzing paper-and-pencil IAT data: Unpublished
manuscript, Yale University.
Ramm, M., & Bargel, T. (2005). Frauen im Studium. Langzeitstudie 1983 - 2004 (Female
students. Long-term study 1983 - 2004). Bonn, Berlin: Bundesministerium für Bildung
und Forschung (BMBF)/ Federal Ministry of Education and Research.
Ranganath, K. A., Smith, C. T., & Nosek, B. A. (2008). Distinguishing automatic and
controlled components of attitudes from direct and indirect measurement methods.
Journal of Experimental Social Psychology, 44, 386-396.
Rudman, L. A., Greenwald, A. G., & McGhee, D. E. (2001). Implicit self-concept and
evaluative implicit gender stereotypes: Self and ingroup share desirable traits.
Personality and Social Psychology Bulletin, 27, 1164-1178.
Implicit math-gender stereotypes 38
Rudman, L. A., Phelan, J. E., & Heppen, J. B. (2007). Developmental sources of implicit
attitudes. Personality and Social Psychology Bulletin, 33, 1700-1713.
Shih, M., Pittinsky, T. L., & Ambady, N. (1999). Stereotype susceptibility: Identity salience
and shifts in quantitative performance. Psychological Science, 10, 80-83.
Spencer, S. J., Steele, C. M., & Quinn, D. (1999). Stereotype threat and women's math
performance. Journal of Experimental Social Psychology, 35, 4-28.
Steele, J. (2003). Children's gender stereotypes about math: The role of stereotype
stratification. Journal of Applied Social Psychology, 33, 2587-2606.
Steele, J. R., & Ambady, N. (2006). “Math is Hard!” The effect of gender priming on
women's attitudes. Journal of Experimental Social Psychology, 42, 428-436.
Steffens, M. C. (2004). Is the Implicit Association Test immune to faking? Experimental
Psychology, 51, 165-179.
Steffens, M. C., & Buchner, A. (2003). Implicit Association Test: Separating
transsituationally stable and variable components of attitudes toward gay men.
Experimental Psychology, 50, 33-48.
Steffens, M. C., & Jelenec, P. (2009). Separating implicit gender stereotypes regarding math
and language: Implicit stereotypes are self-serving for boys and men, but not for girls
and women. Manuscript in revision.
Steffens, M. C., & Jonas, K. J. (in press). Editorial: Implicit attitude measures. Zeitschrift für
Psychologie/Journal of Psychology (Special Issue: Implicit attitude measures).
Steffens, M. C., Kirschbaum, M., & Glados, P. (2008). Avoiding stimulus confounds in
Implicit Association Tests by using the concepts as stimuli. British Journal of Social
Psychology, 47, 217-243.
Steffens, M. C., & Schulze-König, S. (2006). Predicting spontaneous Big Five behavior with
Implicit Association Tests. European Journal of Psychological Assessment, 22, 13-20.
Strack, F. (1994). Response processes in social judgment. In R. S. Wyer, Jr. & T. K. Srull
(Eds.), Handbook of social cognition (2nd ed.) (pp. 287-322). Hillsdale, NJ, USA:
Lawrence Erlbaum Associates, Inc.
Strack, F., & Deutsch, R. (2004). Reflective and impulsive determinants of social behavior.
Personality and Social Psychology Review, 8, 220-247.
Taasoobshirazi, G., & Carr, M. (2009). A structural equation model of expertise in college
physics. Journal of Educational Psychology, 101, 630-643.
Teachman, B. A., & Brownell, K. (2001). Implicit anti-fat bias among health professionals: Is
anyone immune? International Journal of Obesity, 25, 1-7.
Teachman, B. A., Gapinski, K. D., Brownell, K. D., Rawlins, M., & Jeyaram, S. (2003).
Demonstrations of implicit anti-fat bias: The impact of providing causal information
and evoking empathy. Health Psychology, 22, 68-78.
U.S. Department of Education. (2000). Educational equity for girls and women NCES 2000-
030. Washington, D.C.: U.S. Government Printing Office.
von Stülpnagel, R., & Steffens, M. C. (in press). Prejudiced or just smart? Intelligence as a
confounding factor in the IAT effect. Zeitschrift für Psychologie/Journal of
Psychology.
Wiers, R. W., Van Woerden, N., Smulders, F. T. Y., & De Jong, P. J. (2002). Implicit and
explicit alcohol-related cognitions in heavy and light drinkers. Journal of Abnormal
Psychology, 111, 648-658.
Wilson, T. D., Lindsey, S., & Schooler, T. Y. (2000). A model of dual attitudes.
Psychological Review, 107, 101-126.
Implicit math-gender stereotypes 39
Footnotes
i For readers unfamiliar with IATs, we recommend visiting a demonstration website
for ease of understanding: https://implicit.harvard.edu/
ii In most federal states of Germany, students finish elementary school after 4th grade
and attend either “Gymnasium” (high track) preparing students for university, “Realschule”
(intermediate track), or “Hauptschule” (low track). The sample of Study 2 was comprised of
high-track students only. Information on the ethnic composition of the sample was not
collected; the vast majority of the sample indicated that German was their and their parents’
native language.
iii In detail, girls beginning with girls-self (vs. boys-other) in the gender identity IAT
started with German-self (vs. math-others) in the math self-concept IAT and with German-
girls (vs. math-boys) in the gender stereotype IAT and vice versa. Similarly, boys who
completed the boys-self task first then started with the math-self task and the math-boys task.
iv We refrained from using the scoring algorithm for paper-pencil IATs proposed by
Lane and colleagues (Nosek & Lane, 1999; Teachman & Brownell, 2001; Teachman et al.,
2003) as that scoring algorithm seems to result in more outliers than the variant described
above.
v Separate analyses yielded comparable patterns, but not all effects were statistically
significant due to lower statistical power.
vi We also computed implicit-explicit correlations of math-gender stereotypes and
math self-concepts. For boys and girls taken together, implicit gender stereotypes were
correlated with explicit gender stereotypes (r = .18, N = 535) and with perceived stereotypes
(r = .27, N = 396; Study 2 only). The size of these correlations is in the expected order of
magnitude (Hofmann, Gawronski, Gschwender, Le, & Schmitt, 2005; Lane et al., 2007).
Separate correlations for boys and girls can be obtained from Table 3. For boys and girls
taken together, implicit and explicit math self-concepts were correlated significantly, but low
(r = .12, N = 534).
Implicit math-gender stereotypes 40
vii Preliminary analyses included grade level as an additional predictor and the
interactions of (dummy coded) grade levels with gender and stereotypes. In no case did the
inclusion of the interactions significantly change the variance explained by the model.
Similarly, separate analyses for 4th graders yielded beta weights for the interaction of gender
with implicit stereotypes of at least the same size as for the overall group. Thus, it appears
that the predictive power of implicit stereotypes does not increase across our age groups.
viii Contrary to what one might expect, implicit math self-concepts showed only small
associations with achievement and enrolment preferences. This could be due to the
comparably poor measurement properties of the paper-pencil math self-concept IAT. Implicit
math self-concepts were not related to school grades and barely related to enrolment
preferences for boys, r = .19, and girls, r = .15. Hierarchical regressions conducted separately
for boys and girls with enrolment preferences as criterion and explicit and implicit math self-
concepts as predictors revealed unique predictive power for implicit math self-concepts only
for boys,
β
= .13, t = 2.78, not for girls. As implicit math self-concepts were only weakly
related to the math outcome variables, we could not test a mediation assumption with implicit
math self-concepts mediating the influences of implicit math-gender stereotypes on the
outcome variables.
Table 1. Translations of Concepts and Stimuli used in Study 1 and 2.
IAT
Gender Identity
(Study 1 only)
Concepts
Boys
Girls
I
Other
Stimuli
Boys
Son
Girls
Daughter
I
Me
Other
Foreign
Math Self-Concept
Concepts
Math
German
(Language)
I
Other
Stimuli
Math
Computation
German
Language
I
Me
Other
Foreign
(them)
Gender Stereotype
Concepts
Boys
Girls
Math
German
(Language)
Stimuli
Boys
Son
Girls
Daughter
Math
Computation
German
Language
Note. Concepts and stimuli in parentheses were used in Study 2. The first IAT in
Study 2 consisted of the concepts trees (stimuli: trees, maple) vs. mushrooms (stimuli:
mushrooms, toadstool) and small (stimuli: small, tiny) vs. big (stimuli: big, huge).
Implicit math-gender stereotypes 42
Table 2. Mean Explicit Math-Gender Stereotypes, Math and German Self-Concepts, Grades,
and Enrolment Intentions by Gender and Grade (Study 1 and 2).
Math-gender
stereotypes
Math
self-
concept
German
self-
concept
Math
grade
German
grade
Math
enrolment
intention
German
enrolment
intention
Study 1:
Grade 4
Boys
0.92
(0.94)
4.19
(0.86)
3.32
(1.02)
2.52
(.80)
2.79
(.95)
3.70
(1.14)
2.85
(1.17)
Girls
0.71
(0.85)
3.53
(0.95)
3.61
(1.00)
2.32
(.75)
2.40
(.60)
3.31
(1.20)
3.28
(1.14)
Grade7
Boys
0.87
(1.02)
3.61
(0.69)
2.82
(1.04)
3.03
(1.11)
3.44
(.77)
4.06
(1.39)
3.00
(1.58)
Girls
0.76
(1.04)
2.92
(1.18)
3.21
(1.06)
3.33
(1.30)
3.12
(.77)
3.00
(1.41)
3.00
(1.78)
Grade 9
Boys
0.80
(1.40)
3.33
(0.86)
3.02
(0.62)
3.52
(.91)
3.31
(.75)
3.67
(1.35)
3.62
(1.24)
Girls
0.88
(0.90)
3.08
(0.89)
3.32
(0.95)
3.07
(.75)
2.83
(.86)
3.68
(1.36)
3.81
(1.40)
Overall
Boys
.87
(1.11)
3.76 a
(.89)
3.09 a
(.93)
2.98
(1.01)
3.13 a
(.88)
3.84 a
(1.37)
3.24
(1.42)
Girls
.77
(.91)
3.23 b
(1.03)
3.41 b
(1.01)
2.83
(1.03)
2.73 b
(.78)
3.24 b
(1.39)
3.40
(1.61)
Study 2:
Grade 7
Boys
0.95
(1.04)
3.57
(0.98)
3.15
(0.92)
2.80
(1.01)
2.98
(.83)
3.42
(1.26)
2.82
(1.18)
Girls
0.63
(0.98)
2.99
(0.97)
3.22
(0.79)
2.97
(.97)
2.78
(.67)
2.66
(1.31)
3.09
(1.03)
Grade 9
Boys
1.21
(1.31)
3.37
(1.15)
2.95
(0.96)
3.00
(1.06)
3.27
(.88)
3.36
(1.37)
2.73
(1.27)
Girls
1.22
(1.08)
2.96
(1.00)
3.26
(0.87)
3.00
(1.00)
2.77
(.79)
2.67
(1.42)
3.25
(1.27)
Overall
Boys
1.08
(1.19)
3.47 a
(1.07)
3.05 a
(.94)
2.91
(1.04)
3.13 a
(.87)
3.39 a
(1.31)
2.77 a
(1.22)
Girls
.95
(1.07)
2.98 b
(.99)
3.25 b
(.83)
2.98
(.98)
2.77 b
(.74)
2.67 b
(1.37)
3.18 b
(1.17)
Notes. Higher values in the math and German self-concept scales indicate higher
ability ratings with possible values between 1 and 5. Higher values in stereotypes represent
stronger math-gender stereotypes with possible values between -4 and 4. Lower values in
Implicit math-gender stereotypes 43
grades correspond to better school grades in Germany (range: 1 to 6). Overall scores for
enrolment intentions in Study 1 refer to grade 7 and 9 only. Standard deviations are in
parentheses.
a, b Overall scores for girls and boys differ significantly.
Implicit math-gender stereotypes 44
Table 3. Correlations between Implicit and Explicit Measures, Separately for Boys and
Girls. Data of Studies 1 and 2 are Combined.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(1) Implicit
stereotypes
(2) Implicit
math self-concept
(3) Explicit
stereotypes
(4) Perceived
stereotypes
(5) Explicit
math self-concept
(6) Grade
differences
(7) Enrolment
preferences
(-.03)
.24
.33
(-.11)
-.17
-.13
.25a
-.11
-.16
(.09)
(.03)
.17
.14
.21
.37
-.29
-.19
-.33
.17
(.04)
.33
(-.01)
(-.06)
(-.15)
.15
.11
.28
(.11)
.62
.73
.22
(.08)
.24
(.09)
.65
.50
.19
.14
.27
(.10)
.74
.60
Notes. Values above the diagonal refer to girls, values beneath the diagonal to boys.
Correlations are controlling for source (Study 1 vs. 2) and task order within IATs. Controlling
for grade does not change any correlation more than .02. Perceived stereotypes are from
Study 2 only. Expected correlations between stereotypes and other variables are positive for
girls and negative for boys. Coefficients in parentheses indicate non-significant correlations.
a The relation is significantly stronger for girls than boys.
Implicit math-gender stereotypes 45
Table 4. Beta Weights From Hierarchical Regressions Predicting Math-Related
Outcomes, Separately for Boys and Girls (Study 1 and 2 combined).
Dependent Variables
Step 1
Step 2
Explicit
gender
stereotypes
Adj.
R2
Explicit
gender
stereotypes
Implicit
gender
stereotypes
Adj.
R2
Girls
Implicit math self-concept
Explicit math self-concept
School grades
Enrolment preferences
.21
.28
.24
.27
.05
.08
.06
.07
.18
.26
.22
.25
.22
.11
.19
.15
.09
.08
.09
.09
Boys
Implicit math self-concept
Explicit math self-concept
School grades
Enrolment preferences
(-.11)
-.29
-.19
-.33
(.01)
.08
.04
.11
(-.11)
--
-.28
-.16
-.32
(.00)
(-.04)
-.13
(-.06)
(.01)
.08
.05
.11
Note. Non-significant predictors (p < .05) in parentheses.
Implicit math-gender stereotypes 46
Figure Caption
Figure 1. Stereotype-congruent (left part) and stereotype-incongruent (right part) task
of a math-gender stereotype Implicit Association Test (IAT). Stimuli in the middle are
presented one at a time.
Figure 2. Response latency differences (in ms) in IATs, separately for gender and
school grades in Study 1. Error bars reflect standard errors. Stars on bars indicate effects
significantly different from 0, stars between bars, significant differences.
Figure 3. Differences of numbers of correctly classified items in the two tasks of the
math-gender stereotype IAT, separately for gender and school grades in Study 2. Error bars
reflect standard errors. Stars on bars indicate effects significantly different from 0, stars
between bars, significant differences.
Implicit math-gender stereotypes 47
Figure 1
Implicit math-gender stereotypes 48
Figure 2
Implicit math-gender stereotypes 49
Figure 3
A preview of this full-text is provided by American Psychological Association.
Content available from Journal of Educational Psychology
This content is subject to copyright. Terms and conditions apply.
