Math–Gender Stereotypes in Elementary School Children

TL;DR: The findings suggest that the math-gender stereotype is acquired early and influences emerging math self-concepts prior to ages at which there are actual differences in math achievement.

Abstract: A total of 247 American children between 6 and 10 years of age (126 girls and 121 boys) completed Implicit Association Tests and explicit self-report measures assessing the association of (a) me with male (gender identity), (b) male with math (math–gender stereotype), and (c) me with math (math self-concept). Two findings emerged. First, as early as second grade, the children demonstrated the American cultural stereotype that math is for boys on both implicit and explicit measures. Second, elementary school boys identified with math more strongly than did girls on both implicit and self-report measures. The findings suggest that the math– gender stereotype is acquired early and influences emerging math self-concepts prior to ages at which there are actual differences in math achievement. Imagine yourself an elementary school teacher. One of your female students fails to complete an arithmetic assignment and offers an excuse that ‘‘Girls don’t do math.’’ What might be a pretext for avoiding homework could also be the outcome of social-cognitive development. Combining cultural stereotypes (‘‘Math is for boys’’) with the knowledge about one’s own gender identity (‘‘I am a girl’’) to influence one’s self-concept (‘‘Math is not for me’’) reflects the tendency to achieve what social psychologists (Heider, 1946) call cognitive balance.

Summary

Introduction

• In children, the interplay among math–gender stereotype, gender identity, and math self-concept has been studied using self-report measures.
• Earlier versions of this article were presented at a meeting of the Society for Research in Child Development (April 2009) and the American Psychological Association (August 2009).
• IAT’s format allows the measurement of preference for one concept (e.g., math) relative to the preference for a second concept (e.g., reading).

Participants

• All children were recruited through private and public elementary schools from the greater Seattle area.
• The same recruitment procedure was used for both private and public schools: Schools mailed the consent forms to the parents, and completed forms were collected by the teachers.
• The authors were unable to obtain dates of birth for the recruited children; however, the mean age ranges for the first five elementary school grades in the Seattle area based on the school data were as follows:.

Procedure

• Each test session began with a 3–5 min description of the study, during which children were familiarized with the test apparatus.
• The children were told that they would be ‘‘asked some questions’’ and then ‘‘play a computer game.’’.
• They were told that they would see and hear words during the game and would have to press a button to ‘‘let the computer know which word it is.’’.
• The procedure started with the administration of the self-report measures followed by the administration of the IATs.

Math–Gender Stereotype Measures

• The self-report math–gender stereotype measure was created for this study and administered as two Likert-scale questions using images from Harter and Pike’s (1984) Pictorial Scale.
• One question requested selecting the boy or girl character as ‘‘liking to do math more.’’.
• Modifications were similar to those in previous child IAT procedures (Dunham et al., 2006; Rutland et al., 2005), including an adapted computer keyboard and presenting of stimuli simultaneously as written and spoken words (see details in the next section for further adaptations used in this study).
• Following these two single discrimination tasks, children completed two combined discrimination tasks in which all four categories were used.
• The implicit data were also reanalyzed separately using two alternative approaches for computing the D measure by adding penalties to error trials (Greenwald et al., 2003): D–600 ms penalty as well as the D–2SD penalty measures.

Additional Measures: Gender Identity and Math Self-Concept

• Two additional self-report measures were created for this study following Harter and Pike’s (1984) two-item Likert-scale format, as described earlier.
• For each question, children were shown two pictures of a child (e.g., ‘‘On the left the authors have a girl.
• Positive values indicated that the child picked the same-sex character who was doing math.
• In one instructional condition, math and me words shared a response key, as did reading and not-me words.
• The gender identity IAT and math self-concept IAT were counterbalanced in the first and third position, with the math–gender stereotype IAT administered in the second position.

Internal Consistency

• For implicit measures, Cronbach’s alpha was calculated from two D measures computed for matched 24-trial subsets of each IAT.
• Cronbach’s alpha coefficients for the math–gender stereotype, gender identity, and math self-concept IATs were a = .74, a = .89, and a = .78, respectively.
• For the self-report measures, Cronbach’s alpha coefficients for gender identity and math self-concept were a = .93 and a = .79, respectively.
• The two items of the self-reported math–gender stereotype scale measured two distinct constructs (gender stereotype toward math vs. gender stereotype toward reading).
• Thus, the expectation was for low internal consistency of the self-reported math–gender stereotype measure, which was the case, a = .03.

Data Reduction

• This was done to reduce noise in the data by excluding participants who would be identified as outliers on the basis of preestablished criteria, consistent with the usual IAT procedures with adults (Greenwald et al., 2003).
• The analyses following data reduction provided increased power compared to analyses of the full sample, but the pattern of significant results and the conclusions drawn from them remained unchanged.

Math Self-Concept

• The definition of math self-concept used in the current study differentiates children’s identification with math from more global beliefs about themselves such as self-esteem (Marsh, Craven, & Debus, 1991; Wigfield, Battle, Keller, & Eccles, 2002).
• Other researchers investigating sex differences in children’s math self-concepts also recognize the value of sharp distinctions between self-concepts and self-esteem (Wigfield et al., 2002).
• Self-report questions that tap an evaluative aspect (‘‘good at’’) when asking questions about the self raise issues of self-esteem rather than a math self-concept; the latter entails an identification with math without regard to evaluations either about math or about me (me = math).
• Older children are more prone to make domain-specific, stable attributions than younger children (Rholes, Newman, & Ruble, 1990; Ruble & Dweck, 1995).
• The authors methods may be useful for uncovering conditions under which children of different ages make specific attributions about themselves and how such self-attributions interact with academic achievement and choices (Blackwell, Trzesniewski, & Dweck, 2007; Dweck, 1999; Heyman, 2008; Ruble & Dweck, 1995).

Developmental Order of Emergence

• The authors next examined the order of emergence of the three separate measures (gender identity, math– gender stereotype, and math self-concept).
• The finding of clear evidence for gender identity on both implicit and self-report measures is useful in showing that, even at the earliest grades examined, children could follow directions for both of these measures.
• Three measures (explicit gender identity, implicit math–gender stereotype, and explicit math self-concept) showed weakly increasing effects (all ts < 1.33); the other three measures, showed weakly decreasing effects (absolute value of ts < 1.74).
• None of the linear trends was significant (all ps > .08).

Discussion

• The authors distinguished between math–gender stereotypes and math self-concepts using both implicit and explicit measures within the same study.
• The findings confirm that their child IAT (and self-report) procedures are effective inasmuch as they provide the expected evidence of gender identity.
• These methods allowed us to uncover two new findings.
• First, the math– gender stereotype previously found to be pervasive in American samples of adults was found in elementary school children on both implicit and self-report measures.
• Second, elementary school girls showed a weaker identification with math than boys on both implicit and self-report measures (math self-concept).

Math–Gender Stereotypes

• The current demonstration of math–gender stereotypes during elementary school years extends previous work on this topic (e.g., Aronson & Good, 2003; Muzzatti & Agnoli, 2007).
• The children were then asked to repeat the story, and the experimenter recorded whether the child used ‘‘he’’ or ‘‘she’’ when referring to the student.
• Using an implicit measure and conceptualizing, the stereotype as an association between math and boy addresses an issue raised in the child stereotype literature (Signorella, Bigler, & Liben, 1993).
• It may also inform the debate between awareness and endorsement.

Age-Related Changes

• As expected, there was robust evidence for the presence of gender identity, indeed significant evidence as early as Grades 1–2 on both the implicit and explicit measures.
• These findings for gender identity are consistent with previous research (see Ruble & Martin, 1998, for a review).
• Moreover, these findings are useful because they establish that, even at the youngest grades the authors tested, the children could understand instructions for both the implicit and self-report measures and provided interpretable data for both.
• Sex differences indicated that the presence of math–gender stereotypes also emerged during Grades 1–2 .
• This speculation is also consistent with previous research suggesting that sex differences in math self-concepts emerge in middle to late elementary school (Herbert & Stipek, 2005; Muzzatti & Agnoli, 2007).

Relation Between Implicit and Explicit Measures

• For each of the three constructs of gender identity, math–gender stereotype, and math self-concept, implicit and explicit measures were positively correlated:.
• The implicit–explicit correlation was strong for the measures of gender identity (r = .64) but relatively weak for the measures of math–gender stereotype (r = .14) and math self-concept (r = .28).
• Moderate or low positive correlations between implicit and self-report measures are often found in socially sensitive domains such as stereotypes (Hofmann et al., 2005), with IAT measures having greater predictive validity than explicit measures (Greenwald et al., 2009).
• The two weak correlations (between implicit and self-report measures of math– gender stereotype and math self-concept) suggest possible differential predictive use of these measures in subsequent child development research.

Relations Among Identity, Stereotypes, and Self-Concepts

• The data and theory of Eccles, Wigfield, Harold, and Blumenfeld (1993) also do not support the second alternative.
• In elementary school, boys and girls score equally well on math achievement tests (Hyde et al., 2008) and girls receive higher math grades (Kimball, 1989).
• The authors have conjectured that gender identity and math– gender stereotypes interact in the formation of children’s math self-concepts.

Conclusions

• In the present research, young girls showed a weaker identification with math than did their male peers.
• Such gender differences in children’s math self-concepts may arise from the early combination of societal influences (cultural stereotypes about gender roles) and intrapersonal cognitive factors (balanced cognitive organization).

Figures

Figure 2. Sex differences for implicit (A) and self-report (B) measures in first- through fifth-grade children. Note. IAT = Implicit Association Test. Asterisks indicate significant differences. Error bars = standard errors.

Figure 1. For the Child Implicit Association Test, items from four categories appear one at a time on a computer and are spoken over the loudspeaker, and children respond by pressing a response button. In one task (A), math words and boy names share a response key, as do reading words and girl names (stereotype congruent). In the other task (B), these assignments are reversed—math is paired with girl (stereotype incongruent). Note. Children with the math–gender stereotype (i.e., boy = math) should respond faster to the task (A) than (B).

Figure 3. Developmental effects for implicit (A) and self-report (B) measures for gender identity, math–gender stereotypes, and math self-concepts of first- through fifth-grade children. Note. IAT = Implicit Association Test. Asterisks indicate significant sex differences. N = 222 for the implicit and N = 220 for the selfreport measures. Error bars = standard errors.

Full text

Math–Gender Stereotypes in Elementary School Children
Dario Cvencek, Andrew N. Meltzoff, and Anthony G. Greenwald
University of Washington
A total of 247 American children between 6 and 10 years of age (126 girls and 121 boys) completed Implicit
Association Tests and explicit self-report measures assessing the association of (a) me with male (gender iden-
tity), (b) male with math (math–gender stereotype), and (c) me with math (math self-concept). Two findings
emerged. First, as early as second grade, the children demonstrated the American cultural stereotype that
math is for boys on both implicit and explicit measures. Second, elementary school boys identified with math
more strongly than did girls on both implicit and self-report measures. The findings suggest that the math–
gender stereotype is acquired early and influences emerging math self-concepts prior to ages at which there
are actual differences in math achievement.
Imagine yourself an elementary school teacher. One
of your female students fails to complete an arith-
metic assignment and offers an excuse that ‘‘Girls
don’t do math.’’ What might be a pretext for avoid-
ing homework could also be the outcome of
social-cognitive development. Combining cultural
stereotypes (‘‘Math is for boys’’) with the knowledge
about one’s own gender identity (‘‘I am a girl’’) to
influence one’s self-concept (‘‘Math is not for me’’)
reflects the tendency to achieve what social psy-
chologists (Heider, 1946) call cognitive balance.
In the foregoing example, ‘‘Girls don’t do math’’
is a widespread cultural stereotype in the United
States: Studies with both adults (Nosek et al., 2009)
and children (Lummis & Stevenson, 1990) show
that people in United States believe that math is
stereotypically a male domain. Given such stereo-
types, a tendency to keep the related concepts of
self, gender,andmath consistent with one another
(what Heider, 1946, called cognitive balance) may
play a role in why a young girl would say—and
possibly believe—that math is not for her (of
course, there will be individual differences).
Social knowledge can be represented as a net-
work of interconnections among concepts (Green-
wald et al., 2002). In the foregoing example, three
aspects of social cognition are involved. The first is
the association between math and boy or girl. If this
takes a societally characteristic form (e.g., math =
boy), it can be called a math–gender stereotype. The
second involves gender identity, defined as the asso-
ciation between me and either boy or girl. The third
is a math self-concept, the association between self
and math.
Past research using implicit measures with
adults has shown that for women, the stronger the
associations of (a) self with female and (b) math with
male, the weaker the association of self with math
(Nosek, Banaji, & Greenwald, 2002). In children, the
interplay among math–gender stereotype, gender
identity, and math self-concept has been studied
using self-report measures. American elementary
school children often reflect the stereotypic pattern
for academic self-concepts: For math, girls rate their
own ability lower than boys (Fredericks & Eccles,
2002) but do not do so for reading or spelling (Her-
bert & Stipek, 2005; Heyman & Legare, 2004). Using
self-report, this pattern is evident as early as the
first grade (Entwistle, Alexander, Pallas, & Cardi-
gan, 1987), even in the absence of differences in
math achievement (Herbert & Stipek, 2005). Girls’
weaker identification with math may derive from
culturally communicated messages about math
being more appropriate for boys than for girls
(Dweck, 2007; Eccles, 2007; Guiso, Monte, Sapienza,
& Zingales, 2008; National Science Foundation,
We thank the schools, their directors, teachers, parents, and
the children who participated. J. Aronson, S. Cheryan, G. Hey-
man, C. Kaiser, the members of the LIFE Science of Learning
Center, and three anonymous referees provided insightful com-
ments on earlier versions of this article. We also thank C. Fisher,
C. Harris, and G. Owen for assistance. This research was sup-
ported by a grant from the National Science Foundation (SBE-
0354453) to the LIFE Science of Learning Center. Earlier versions
of this article were presented at a meeting of the Society for
Research in Child Development (April 2009) and the American
Psychological Association (August 2009).
Correspondence concerning this article should be addressed to
Dario Cvencek, Institute for Learning & Brain Sciences, Univer-
sity of Washington, Box 357988, Seattle, WA 98195. Electronic
mail may be sent to dario1@u.washington.edu.
Child Development, May ⁄ June 2011, Volume 82, Number 3, Pages 766–779
 2011 The Authors
Child Development  2011 Society for Research in Child Development, Inc.
All rights reserved. 0009-3920/2011/8203-0005
DOI: 10.1111/j.1467-8624.2010.01529.x

2003; Steele, 2003). These patterns are important
developmentally, because as Eccles and others have
shown, children have reduced interest in future
academic courses and occupations that are incom-
patible with their academic self-concept (Denissen,
Zarrett, & Eccles, 2007; Frome, Alfeld, Eccles, &
Barber, 2006; Killen, Margie, & Sinno, 2006;
Liben, Bigler, & Krogh, 2001; Malcom et al., 2005;
Newcombe, 2007).
Previous investigations of children’s math–gen-
der stereotype and math self-concept have focused
on self-report measures (for an exception, see Amb-
ady, Shih, Kim, & Pittinsky, 2001). The wording of
self-report measures often involves asking children
how good they think they are at something or how
much they like it, both of which conflate self-con-
cept (nonevaluative association of self) with self-
esteem (evaluative association of self). For example,
a girl who reports that she is good at math may do
so because she thinks that she is good at many
things (high self-esteem). Similarly, a boy who
reports that he likes math may do so because he
believes that liking math is a positive quality and
he sees himself as having many positive qualities
(high self-esteem). If the focus is children’s math
self-concepts, it is more informative to assess how
strongly a child associates self with math (i.e.,
whether the child has a strong math self-concept
or not).
In order to differentiate the constructs more
cleanly, we adapted a test used with adults in
social psychology that does not require self-report,
the Implicit Association Test (IAT; Greenwald,
McGhee, & Schwartz, 1998). We modified it so that
it could be used with elementary school children.
The IAT originated within social psychology but in
recent years has been applied in cognitive psychol-
ogy (Fazio & Olson, 2003), clinical psychology
(Teachman, Gregg, & Woody, 2001), and develop-
mental psychology (Dunham, Baron, & Banaji,
2006; Rutland, Cameron, Milne, & McGeorge, 2005;
Skowronski & Lawrence, 2001). In adults, IAT
measures correlate with actual math performance
and real-world choices and actions (Greenwald,
Poehlman, Uhlmann, & Banaji, 2009).
The IAT is a computerized categorization task
that measures relative strengths of associations
among concepts. IAT’s format allows the measure-
ment of preference for one concept (e.g., math) rela-
tive to the preference for a second concept (e.g.,
reading). The contrasting category is of practical
importance in investigations involving academic
subjects, because academic choices rarely occur
without alternatives. Reading offers itself readily as
a contrasting category for math because: (a) reading
and math education are mandated from the first
grade on, (b) sex differences in self-concepts have
been demonstrated most often for math and read-
ing, and (c) standardized tests across many coun-
tries have reading and math portions.
In addition to investigating implicit math–gender
stereotype, gender identity, and math self-concept
via a child IAT, we also examined explicit (self-
report) counterparts in the same children. One
motivation comes from research suggesting that
stereotypes can be separated into two underlying
processes—one automatic, unconscious, and impli-
cit and the other controlled, conscious, and explicit
(Devine, 1989; see Killen, McGlothlin, & Henning,
2008, for a review of studies using explicit and
implicit measures with children). In adults, positive
but weak correlations are observed between impli-
cit and explicit measures, especially in socially sen-
sitive domains such as stereotypes (Hofmann,
Gawronski, Gschwendner, Le, & Schmitt, 2005).
One of the explanations for this dissociation
involves motivational influences: Implicit measures
are assumed to be less susceptible to social desir-
ability artifacts. It has also been suggested that
early developmental experiences may shape impli-
cit more than explicit cognition (Liben & Bigler,
2002; Rudman, 2004), again suggesting the value of
using both implicit and explicit measures in the
same study with the same children.
The present research draws on Heider’s (1946)
balance theory. Heider’s principles of cognitive bal-
ance were extended by Greenwald et al. (2002) to
explain how cognitive structures involving atti-
tudes, stereotypes, and self-concepts organize
themselves to become mutually consistent, or
balanced. This extended formulation has been con-
firmed in recent studies (e.g., Greenwald, Rudman,
Nosek, & Zayas, 2006; Greenwald et al., 2002).
Balance theory has often been referenced in
research with older adolescents (undergraduate
students) and adults, but it has not been applied in
early child development with the exception of a
study of disadvantaged Hispanic children (ages
5–12) and adults measuring racial identity, race atti-
tude, and self-esteem using self-report and IAT
measures (Dunham, Baron, & Banaji, 2007). Within
Heider’s theoretical framework, interconnections
among concepts are assumed to self-organize in
ways that reflect cognitive consistency or balance.
Thus, a child who strongly associates self with His-
panic, and Hispanic with good, is predicted to have
higher self-esteem, as found by Dunham et al.
(2007).
Gender Stereotypes 767

The three chief aims of our study were to: (a)
design new measures of children’s math–gender
stereotypes and math self-concepts by adapting
adult work from social psychology, (b) assess
children’s math–gender stereotypes and math self-
concepts during elementary school years, and (c)
do so using both implicit and explicit measures
within the same study. We examined three hypoth-
eses: First, the child IAT created for this study will
provide evidence of gender identity, in accordance
with previous research that has established gender
identity using self-report measures in elementary
school and younger children. Second, American
elementary school children will associate math
more strongly with boys than with girls on both
implicit and self-report measures. Third, on both
implicit and self-report measures boys should self-
identify with math more strongly than girls.
Method
Participants
A total of 247 American children (126 girls, 121
boys) from Grades 1–5 were tested. All children
were recruited through private and public elemen-
tary schools from the greater Seattle area. The same
recruitment procedure was used for both private
and public schools: Schools mailed the consent
forms to the parents, and completed forms were
collected by the teachers. None of the children
tested had repeated a grade. We were unable to
obtain dates of birth for the recruited children;
however, the mean age ranges for the first five ele-
mentary school grades in the Seattle area based on
the school data were as follows: The mean age for
children attending Grade 1 was 6.66 years
(SD = 0.33) and the mean age for children in Grade
5 was 10.68 years (SD = 0.37). The sample sizes and
gender breakdown for our test sample were as fol-
lows: Grade 1, n = 50 (24 boys, 26 girls); Grade 2,
n = 49 (24 boys, 25 girls); Grade 3, n = 51 (25 boys,
26 girls); Grade 4, n = 49 (24 boys, 25 girls); and
Grade 5, n = 48 (24 boys, 24 girls). According to the
available school data, children were predominantly
from middle- to upper-class families. According to
parental report (collected independently by the
schools for their annual reports and provided to us
at testing), the children in our sample were 83.3%
White, 9.6% Asian, and 7.1% African American.
After the study was completed, we provided $10
checks for each participating family to school
administrators who distributed them to the partici-
pating families.
Procedure
Children were tested individually in a separate
quiet room outside of his or her classroom while
seated at a desk facing a computer (either a 43-
or 48-cm screen). Each test session began with a
3–5 min description of the study, during which
children were familiarized with the test apparatus.
The children were told that they would be ‘‘asked
some questions’’ and then ‘‘play a computer
game.’’ They were told that they would see and
hear words during the game and would have to
press a button to ‘‘let the computer know which
word it is.’’ The procedure started with the admin-
istration of the self-report measures followed by the
administration of the IATs.
Math–Gender Stereotype Measures
Self-report. The self-report math–gender stereo-
type measure was created for this study and
administered as two Likert-scale questions using
images from Harter and Pike’s (1984) Pictorial
Scale. For each question, children were shown two
pictures of a child and responded by reporting: (a)
which character (boy or girl) they themselves
believed possessed an attribute (e.g., liking math)
to a greater degree, and (b) whether they believed
the character possessed the attribute ‘‘a little’’ or ‘‘a
lot.’’ This was done by their pointing to one of two
circles (1.1 and 2.3 cm in diameter). One question
requested selecting the boy or girl character as ‘‘lik-
ing to do math more.’’ The other question
requested selecting the boy or girl character as ‘‘lik-
ing to read more.’’ All self-report questions were
memorized by the experimenter and said aloud to
the children. The two scores were subtracted from
one another to arrive at the explicit score with
lower and upper bounds of )2 and +2; positive val-
ues indicated that the child picked the same-sex
character as liking to do math more. The Appendix
provides a full list of all names used in self-report
measures. These self-report measures were not
administered to 16 of the 247 subjects because they
had not been developed yet.
Child IAT. We adapted the standard, adult IAT
(Greenwald et al., 1998) for use with children. Mod-
ifications were similar to those in previous child
IAT procedures (Dunham et al., 2006; Rutland
et al., 2005), including an adapted computer key-
board and presenting of stimuli simultaneously as
written and spoken words (see details in the next
section for further adaptations used in this study).
An IAT score (D; Greenwald, Nosek, & Banaji,
768 Cvencek, Meltzoff, and Greenwald

2003) was calculated by comparing the speed with
which children categorize exemplars from four
categories under two instructional conditions that
vary assignments of the four categories to two com-
puter response keys, one operated with the left
hand and the other with the right hand. The mea-
sure is based on the principle that it is easier to give
the same response to items from two categories if
the two categories are mentally associated than if
they are not. Figure 1 provides a pictorial represen-
tation of the child IAT.
During the math–gender stereotype IAT, chil-
dren first practiced sorting girl and boy names.
They responded to girl names (Emily, Jessica,
Sarah, Ashley) by pressing a response button on
the left side of the keyboard (in the position of a
‘‘D’’ key) and to boy names (Michael, Andrew,
David, Jacob) by pressing a response button on the
right side of the keyboard (in the position of a ‘‘K’’
key). After that, children practiced sorting math
words (addition, numbers, graph, math) and read-
ing words (read, books, story, letters) using the
same two response buttons (Greenwald et al.,
1998).
Following these two single discrimination tasks,
children completed two combined discrimination
tasks in which all four categories were used. Dur-
ing the combined tasks, two of the four categories
were mapped onto the same response key. In one
condition, math words and boy names shared one
response key, with reading words and girl names
sharing the other. The second condition switched
the key assignments of the math and reading catego-
ries. All single discrimination tasks consisted of 16
trials and all combined tasks consisted of 24 trials.
Positive scores indicated stronger association of
math with own gender than with opposite gender.
Greenwald et al.’s (2003) scoring algorithm con-
strains the resulting D measure to have bounds of
)2 and +2. The implicit data were also reanalyzed
separately using two alternative approaches for
computing the D measure by adding penalties to
error trials (Greenwald et al., 2003): D–600 ms pen-
alty as well as the D–2SD penalty measures. For all
three IATs, the D–600 and D–2SD were not statisti-
cally significant from the D–as is measure (all
ps > .26). The D–as is measure is therefore used
throughout the text. In addition, to rule out speed
of a response as a confound, we directly compared
boys’ and girls’ response times (RTs) for each of
our three IATs using independent t tests. In one of
the three IATs, girls had slightly faster RTs, and in
the other two boys had slightly faster RTs. How-
ever, none of the t test comparisons was statistically
significant (all ps > .38), suggesting that boys and
girls did not differ significantly in their overall
speed of response on our IAT measures.
The keyboard was furnished with two large pan-
els to replace the computers’ ‘‘D’’ and ‘‘K’’ keys
(see Figure 1). Stickers with left-pointing and right-
pointing arrows on those buttons indicated their
use for left and right responses. To reduce the need
for reading, each stimulus word—spoken in a
female voice—was synchronized with the onset of
Figure 1. For the Child Implicit Association Test, items from four categories appear one at a time on a computer and are spoken over
the loudspeaker, and children respond by pressing a response button. In one task (A), math words and boy names share a response key,
as do reading words and girl names (stereotype congruent). In the other task (B), these assignments are reversed—math is paired with girl
(stereotype incongruent).
Note. Children with the math–gender stereotype (i.e., boy = math) should respond faster to the task (A) than (B).
Gender Stereotypes 769

the written word on the screen. The intertrial inter-
val was 500 ms. All words used as IAT stimuli
were pretested with elementary school children for
familiarity and comprehension. To ensure that
children understood each IAT task, error responses
were followed by a red question mark appearing
on the computer screen. After committing an error
children could not advance to the next trial until
they provided the correct response. As is standard
in IAT procedures, trial latency was recorded to the
correct response. The Appendix provides the list of
all IAT stimuli.
Additional Measures: Gender Identity and Math
Self-Concept
Self-report. Two additional self-report measures
were created for this study following Harter and
Pike’s (1984) two-item Likert-scale format, as
described earlier. The measure of gender identity
consisted of two questions. For each question, chil-
dren were shown two pictures of a child and the
experimenter explained each picture while pointing
to it (e.g., ‘‘On the left we have a boy. His name is
David’’ and ‘‘On the right we have a girl. Her name
is Emily’’). Children were asked to report: (a) which
character they were more like (e.g., ‘‘Are you more
like David or are you more like Emily?’’) and (b)
the degree to which they were like the selected
character (e.g., ‘‘How much like David [Emily] are
you? A little or a lot?’’). The measure was scored so
that positive values indicated that the child picked
the boy character.
The math self-concept measure also consisted of
two questions. For each question, children were
shown two pictures of a child (e.g., ‘‘On the left we
have a girl. Her name is Jessica. Jessica likes math.’’
and ‘‘On the right we have another girl. Her name
is Sarah. Sarah likes to read’’). Children were asked
to report: (a) which character they were more like
(e.g., ‘‘Are you more like Jessica or are you more
like Sarah?’’) and (b) the degree to which they were
like the selected character (e.g., ‘‘How much like
Jessica [Sarah] are you? A little or a lot?’’). Reading
was expected to ‘‘go with’’ female in the sense that
girls were expected to pick the same-sex character
who was reading as more like them than the same-
sex character who was doing math. Positive values
indicated that the child picked the same-sex charac-
ter who was doing math. Selecting a reading char-
acter in one of the two questions and the math
character in the other would result in a value of 0
(indicating that the child, on this measure, had an
equally strong identification with math and read-
ing). For the self-report measures, the order of the
math–gender stereotype, gender identity, and math
self-concept measures was counterbalanced across
children. The order of characters assigned to left
and right sides and the names used for each charac-
ter were also counterbalanced across children.
Order of administering self-report measures did
not influence scores (all ps > .52) and was therefore
not used as a factor in analyses to be reported.
Child IAT. Two additional IAT measures were
administered. For the gender identity IAT, children
classified the words representing me, not-me, boy,
and girl. In one instructional condition, me words
and boy names shared a response key, with not-me
words and girl names sharing the other response
key. In the other instructional condition, two of the
response assignments were reversed, such that me
words and girl names shared one key whereas not-
me words and boy names shared the other key.
Positive scores indicated stronger association of me
with boy than with girl.
For the math self-concept IAT, children classified
the words representing me, not-me, math, and read-
ing. In one instructional condition, math and me
words shared a response key, as did reading and
not-me words. In the other instructional condition,
left versus right assignment of me and not-me words
was reversed. Positive scores indicated stronger
association of me with math relative to reading. For
the implicit measures, there were 16 counterbalanc-
ing conditions. The gender identity IAT and math
self-concept IAT were counterbalanced in the first
and third position, with the math–gender stereo-
type IAT administered in the second position.
Within each IAT, order of the two instructional
conditions was counterbalanced. The spatial orien-
tation of categories assigned to left and right was
counterbalanced across participants and IATs.
Order of administration did not influence scores on
any implicit measures (all ps > .66) and was there-
fore not retained as a factor in analyses to be
reported.
Internal Consistency
For implicit measures, Cronbach’s alpha was cal-
culated from two D measures computed for
matched 24-trial subsets of each IAT. Cronbach’s
alpha coefficients for the math–gender stereotype,
gender identity, and math self-concept IATs were
a = .74, a
= .89, and a = .78, respectively. For the
self-report measures, Cronbach’s alpha coefficients
for gender identity and math self-concept were
a = .93 and a = .79, respectively. The two items of
770 Cvencek, Meltzoff, and Greenwald

Frequently asked questions

Q1. What contributions have the authors mentioned in the paper "Math–gender stereotypes in elementary school children" ?

This paper found that elementary school boys identified with math more strongly than did girls on both implicit and self-report measures, indicating that the gender stereotype is acquired early and influences emerging math self-concepts prior to ages at which there are actual differences in math achievement. 

Q2. What are the future works in "Math–gender stereotypes in elementary school children" ?

Future studies will profit from unifying the concepts and experimental tools from developmental science and social psychology ( Cvencek, Greenwald, & Meltzoff, in press ; Dunham & Olson, 2008 ; Killen et al., 2008 ; Meltzoff, 2007 ; Meltzoff, Kuhl, Movellan, & Sejnowski, 2009 ; Olson & Dweck, 2008 ; Rutland et al., 2005 ) to explore the development of academic identity and how it contributes to children ’ s educational choices, success, and future aspirations. 

Q3. What are the main conclusions from these supplementary analyses?

The main conclusions from these supplementary analyses are that evidence for cognitive balance patterns is (a) clearly present on implicit measures, (b) more apparent on explicit measures than previously reported in studies using explicit measures in adults, and (c) stronger with increasing school grade. 

Q4. What was the main effect of grade on the measures of gender identity?

In addition, the authors used an analysis of variance (ANOVA) to test for changes over grade in the implicit and explicit measures of gender identity, math–gender stereotype, and math self-concept. 

Q5. What were the coefficients for gender identity and math self-concept?

For the self-report measures, Cronbach’s alpha coefficients for gender identity and math self-concept were a = .93 and a = .79, respectively. 

Q6. What is the role of implicit measures in the study?

Future studies could be designed that use child implicit measures in conjunction with self-report measures to explore the development and interrelation between implicit and explicit knowledge of stereotypes, both for more sensitive (racial preferences) or less sensitive (object preferences) domains (Greenwald & Nosek, 2008; Liben & Bigler, 2002; O’Connor, Cvencek, Nasir, Wischnia, & Meltzoff, 2010). 

Q7. What is the correlation between implicit and self-report measures?

Moderate or low positive correlations between implicit and self-report measures are often found in socially sensitive domains such as stereotypes (Hofmann et al., 2005), with IAT measures having greater predictive validity than explicit measures (Greenwald et al., 2009). 

Q8. What is the significance of the correlation between implicit and explicit measures of gender identity and math self-?

The first holds that children who strongly identify with their gender (strong gender identity) are more likely to internalize cultural stereotypes about their gender (math–gender stereotypes), which in turn influences their math self-concepts. 

Q9. What is the definition of math self-concept used in the current study?

The definition of math self-concept used in the current study differentiates children’s identification with math from more global beliefs about themselves such as self-esteem (Marsh, Craven, & Debus, 1991; Wigfield, Battle, Keller, & Eccles, 2002).