The Roots of Stereotype Threat: When Automatic Associations Disrupt Girls’Math Performance

Silvia Galdi and Mara CadinuUniversity of Padova

Carlo TomasettoUniversity of Bologna

Although stereotype awareness is a prerequisite for stereotype threat effects (Steele & Aronson, 1995), researchshowed girls’ deficit under stereotype threat before the emergence of math–gender stereotype awareness, andin the absence of stereotype endorsement. In a study including 240 six-year-old children, this paradox wasaddressed by testing whether automatic associations trigger stereotype threat in young girls. Whereas no indi-cators were found that children endorsed the math–gender stereotype, girls, but not boys, showed automaticassociations consistent with the stereotype. Moreover, results showed that girls’ automatic associations variedas a function of a manipulation regarding the stereotype content. Importantly, girls’ math performancedecreased in a stereotype-consistent, relative to a stereotype-inconsistent, condition and automatic associationsmediated the relation between stereotype threat and performance.

Stereotype threat research (Steele & Aronson, 1995)has shown that both adults and children underper-form in difficult tests when a negative domain-relevant in-group stereotype is made salient (seeInzlicht & Schmader, 2012, for a review). Despitethis overwhelming evidence, the specific require-ments for children’s underperformance under ste-reotype threat are still unclear. With the aim ofproviding new insights into research on stereotypethreat, this study investigates automatic associationsas a potential requirement for the emergence of ste-reotype threat underperformance in children.

Automatic associations are those associationsbetween concepts (e.g., me–woman), or betweenconcepts and attributes (e.g., flower–positive) thatcome to mind unintentionally when an individualencounters a relevant object, that are difficult to con-trol once they have been activated, and that are notnecessarily explicitly endorsed (e.g., Gawronski &Bodenhausen, 2006). Such automatic associations areoften contrasted with conscious beliefs, which arethose mental contents that an individual explicitlyendorses as accurate (e.g., Gawronski & Bodenhau-sen, 2006). Most measures of automatic associationsare based on participants’ performance on com-puter-based, speeded categorization tasks (see Gaw-ronski & Payne, 2010, for a review). These measures,commonly referred to as implicit measures, differ

from self-reports, described as explicit measures,which assess conscious beliefs.

This study employed implicit and explicitmeasures jointly to show that automatic associa-tions consistent with a negative in-group stereotypemay lead to stereotype-induced underperformanceat early stages of development, even thoughchildren do not possess the cognitive competenciesfor stereotype awareness yet, and in the absence ofevidence that they endorse the stereotype content.

Prerequisites for Stereotype Threat Susceptibility

According to the stereotype threat model, for ste-reotypes to affect performance, children need to (a)have developed a concept of social categories (cate-gory awareness), (b) be able to identify themselves asmembers of a social category (self-categorization), and(c) know that the in-group category is negativelyrelated to specific domains or attributes (stereotypeawareness; Aronson & Good, 2003). When childrenenter elementary school, they possess the cognitivecompetencies of category awareness and self-categoriza-tion (e.g., Martin & Ruble, 2010) but not stereotypeawareness (e.g., McKown & Strambler, 2009).

Research has shown that, between the ages of 3and 4 years, children become aware of socialcategories, such as gender and race (e.g., Aboud,

Correspondence concerning this article should be addressed toSilvia Galdi, Dipartimento di Psicologia dello Sviluppo e dellaSocializzazione, Universit�a di Padova, Via Venezia 15, Padova35131, Italy. Electronic mail may be sent to silvia.galdi@unipd.it.

© 2013 The AuthorsChild Development © 2013 Society for Research in Child Development, Inc.All rights reserved. 0009-3920/2014/8501-0017DOI: 10.1111/cdev.12128

Child Development, January/February 2014, Volume 85, Number 1, Pages 250–263

1988; Katz & Kofkin, 1997), and are able to identifythemselves as members of these categories (e.g.,Martin & Ruble, 2010; Quintana, 1998). Moreover,by this age children develop personal stereotypicbeliefs about abilities and characteristics of ethnic(Aboud, 1988) and gender groups. For example,they hold personal stereotypic beliefs about genderdifferences in toy preferences, dressing, and aggres-sive versus prosocial behaviors (Martin & Ruble,2010). However, personal stereotypic beliefs,referred to as stereotype endorsement, differ from theknowledge of stereotypes held by others (not neces-sarily personally endorsed), which is defined asstereotype awareness (e.g., Martinot & D�esert, 2007;McKown & Strambler, 2009). Thus, although inmany cases personal beliefs overlap with sociallyshared stereotypes, this does not imply necessarilythat young children are aware of stereotypes.Indeed, the literature on children’s social perspec-tive taking (e.g., Selman, 1980), theory of mind(e.g., Perner & Wimmer, 1985), and person percep-tion (e.g., Rholes & Ruble, 1984) suggests that chil-dren develop the cognitive competencies forstereotype awareness only by middle childhood,and that up to then they do not distinguish theirpersonal stereotypic beliefs from others’ beliefs.Consistently, children have been shown to be awareof common ethnic and gender stereotypes aboutacademic abilities by the ages of 8 and 9 years (e.g.,McKown & Weinstein, 2003; Quintana, 1998).

The evidence that children do not possess stereo-type awareness when they enter elementary schoolhas important consequences. As noted earlier,stereotype awareness is a specific requirement forstereotype threat. Specifically, the stereotype threatmodel posits that people’s negative stereotypeshamper performance only in situations that inducetargets to become concerned about being judged byothers on the basis of relevant stereotypes (Steele,1997). Consistently, research on performance decre-ments induced by ethnic stereotypes confirms thatonly children who are aware of broadly held stereo-types are vulnerable to stereotype threat effects(McKown & Strambler, 2009; McKown & Weinstein,2003). Therefore, in principle we should not expectdeclines in performance under stereotype threatprior to 8–9 years of age, or even later (Aronson &Good, 2003). Nevertheless, Ambady, Shih, Kim, andPittinsky (2001) showed that 5- to 7-year-old AsianAmerican girls underperformed on a math taskwhen their gender identity was made salient, ascompared to children in a control condition.Similarly, Tomasetto, Alparone, and Cadinu (2011)found that gender identity activation hampered

math performance among 5- to 7-year-old Italiangirls, and Neuville and Croizet (2007) obtainedsimilar findings among 7-year-old French girls.

The last three studies raise a theoretical paradox:How can stereotype-induced performance decre-ments be found in girls who do not possess stereo-type awareness yet? One could assume that for anegative stereotype to affect performance in chil-dren who have not developed stereotype awarenessyet, it is sufficient that children hold the personalstereotypic belief that their in-group is less skillfulin the relevant domain. Indeed, given that youngchildren do not distinguish their own stereotypicbeliefs from others’, and rather assume that theirbeliefs are shared by others as well (e.g., Augousti-nos & Rosewarne, 2001), in principle children’spersonal negative beliefs about abilities of theirin-group could be sufficient to trigger stereotypethreat. If this were the case, stereotype endorsementwould be the key to identify the sources of stereo-type threat in young children.

To date, the potential role of children’s endorse-ment of the math–gender stereotype as a prerequi-site for stereotype threat effects does not fit withthe available evidence. For example, Ambady et al.(2001) found that 5- to 7-year-old American girlsbelieve that boys and girls are equal at math (forsimilar results on Italian children, see Tomasettoet al., 2011); on the contrary, 5- to 7-year-old Amer-ican boys state that boys are better at math, thusexhibiting in-group favoritism (e.g., Powlishta,1995; Yee & Brown, 1994). Similarly, other researchshowed no endorsement of the math–gender stereo-type until 8–9 years of age among Italian andFrench children (Martinot & D�esert, 2007; Muzzatti& Agnoli, 2007).

To our knowledge, only one study found stereo-type endorsement as early as 6–7 years of age in aWestern country (i.e., United States; Cvencek,Meltzoff, & Greenwald, 2011). However, this dis-crepancy from the other findings might be due tothe type of measures used. Whereas all the studiesdiscussed elsewhere in the article assessed theendorsement of the gender stereotype regardingmath ability, Cvencek et al. (2011) assessed theendorsement of the stereotype that boys like mathand girls like language. Therefore, it is possible that6- to 7-year-olds already hold the personal stereo-type that boys like math, but still do not endorse thestereotype about girls’ and boys’ math ability.Consistent with this reasoning, Steele (2003) foundthat 6- to 10-year-old children do not endorse thestereotype regarding boys’ and girls’ abilities in math(see also Martinot, Bag�es, & D�esert, 2012), even

Stereotype Threat Before Stereotype Awareness 251

though they endorse the same stereotype about menand women (stereotype stratification). Thus, to date,there is no evidence that young children endorsethe math–gender ability stereotype, at least inWestern countries. Nevertheless, research showsthat it is specifically the ability component of themath–gender stereotype that affects women’s per-formance, whereas other stereotypical beliefs (e.g.,women are less interested in math) do not triggerstereotype threat effects (Thoman, White, Yam-awaki, & Koishi, 2008).

To summarize, young girls show math underper-formance before being aware of the math–genderstereotype, and in the absence of evidence that theyendorse such belief as a personal stereotype. Onepossibility to explain this paradox is that previousfindings on children’s math–gender stereotype havebeen based only on self-reports, which may notcapture all relevant aspects of the stereotype. Thus,in this research, we used implicit and explicit mea-sures jointly to investigate automatic associations asan alternative route for the emergence of stereo-type-induced underperformance in young children.

Insights From Social-Cognitive Research

Implicit measures, such as the well-known Impli-cit Association Test (IAT; Greenwald, McGhee, &Schwartz, 1998), have been developed to assess thestrength of automatic associations between conceptsor concepts and attributes. These procedures arebased on the rationale that individuals should befaster at pairing concepts, or concepts and attributesthat are strongly associated in their cognitive map,as compared to those that are weakly or not at allassociated. Research has shown that men andwomen differ in the strength of their automaticassociations between gender and academic domains(Nosek, Banaji, & Greenwald, 2002) and that auto-matic associations consistent with the math–genderstereotype predict attitudes toward math, domainidentification, and math performance (e.g., Kiefer &Sekaquaptewa, 2007; Nosek & Smyth, 2011).Together, these results show that automatic associa-tions play an important role in people’s academicachievement, interest, and performance, and thatthe combined use of implicit and explicit measuresmay allow researchers to provide useful insightsthat neither implicit measures nor explicit measuresalone would offer.

Theorizing on the meaning of implicit andexplicit measures posits that explicit measuresreflect more recent as well as newly formedrepresentations that are not strong enough yet to

be automatically activated (e.g., Rudman, 2004;Wilson, Lindsey, & Schooler, 2000), whereas impli-cit measures detect only those highly overlearnedassociations between concepts whose activation hasbecome automatic over the course of long-termexperiences. According to this widespread assump-tion, in terms of emergence, conscious beliefs reflectmore recently formed representations, whereasautomatic associations reflect only the older repre-sentations. However, an emerging body of researchsuggests that the opposite may be true as well.

Recent findings showing experimentally inducedchanges in implicit but not explicit measures (e.g.,Gawronski & LeBel, 2008; Olson & Fazio, 2006)suggest that automatic associations may reflectmore recently formed representations as comparedto the corresponding conscious beliefs. These resultsare also consistent with studies using implicit andexplicit measures in predicting future choices ofdecided and undecided individuals (for a review,see Gawronski & Galdi, 2011). These studies dem-onstrate that well-structured automatic associationsabout a specific object are present also in the cogni-tive maps of those individuals who, on explicitmeasures, report being undecided and donot endorse well-defined conscious beliefs yet.These findings come from research on attitudes(i.e., dealing with the affective evaluation of anobject; Eagly & Chaiken, 1993), whereas stereotypesare cognitive beliefs about the link between a cate-gory and an attribute (Hamilton & Trolier, 1986);nonetheless they suggest that in terms of emer-gence, automatic associations may precede consciousbeliefs. Thus, there is reason to believe that implicitmeasures can also detect recently formed automaticassociations between concepts and stereotypicalattributes (e.g., boys–math) that are not reflected onexplicit measures yet.

Insights From Research on Malleability of AutomaticAssociations

Other research has demonstrated that externalcues (e.g., situational stimuli) may increase ordecrease the activation of automatic associations inadults. For example, Dasgupta and Asgari (2004)found that participants exposed to counterstereo-typical women showed lower activation of auto-matic associations consistent with gender–rolestereotypes, as compared to control participants.Similarly, women exposed to gender-stereotypicalwomen in TV commercials showed increased acti-vation of automatic associations consistent withthe traditional female stereotype, and this higher

252 Galdi, Cadinu, and Tomasetto

activation of automatic associations led to worsemath performance (Davies, Spencer, Quinn, & Ger-hardstein, 2002). The last results are consistent withresearch demonstrating that stereotypical automaticassociations affect working memory resourcesneeded to perform complex cognitive tasks. Specifi-cally, Forbes and Schmader (2010) found thatwomen trained via an IAT to associate their genderwith being good at math showed higher workingmemory and increased math performance. Takentogether, these findings suggest that it may be spe-cifically in automatic associations between groupmembership and stereotypical attributes, and in themalleability of such automatic associations, wherewe should look not only to investigate stereotype-induced underperformance but also to reducestereotype threat effects. Unfortunately, to date nostudy has tested whether automatic associations aremalleable in children as well, or whether changesin activation of stereotypical automatic associationsmay reflect changes in performance. Therefore, wehypothesized that situational cues making a nega-tive stereotype salient may automatically activatecorresponding automatic associations. The activa-tion of such stereotypical automatic associationsshould in turn affect the performance of childrenwho are stereotype targets, even in the absence ofevidence for stereotype endorsement. Conversely,cues challenging the negative stereotype shouldreduce the activation of corresponding automaticassociations, and such a reduction should lead toimproved performance.

Aims of This Study

To test the hypotheses mentioned above, weemployed explicit and implicit measures jointly toassess two aspects of the math–gender stereotype,stereotype endorsement and automatic associations,and their differential relation with young children’sstereotype threat vulnerability.

Research with children has already used implicitmeasures to assess automatic associations (e.g.,Baron & Banaji, 2006). Regarding math–gender ste-reotypes, Steffens, Jelenec, and Noack (2010) foundstereotypical automatic associations in 9- to 14-year-old girls and no automatic associations for boys atany age. Conversely, Cvencek et al. (2011) foundstereotype-consistent automatic associations in bothgirls and boys as young as 6–7 years of age. How-ever, neither Steffens et al. nor Cvencek et al.assessed either performance or automatic associa-tions under stereotype threat. Therefore, these stud-ies are not directly relevant to our main hypothesis.

To our knowledge, only Ambady and collabora-tors used an implicit stereotype awareness task(Ambady et al., 2001, p. 387) to assess the math–gender stereotype in a study on stereotype threat inyoung children. The measure was a memory recalltask, in which the experimenter told a story about astudent especially good at math, and recordedwhether the participant, when repeating the story,used the pronoun he or she to identify the protago-nist. However, rather than a measure of stereotypi-cal automatic associations, this task should beconsidered as an indirect measure of children’sstereotypes. Therefore, it is still unknown whetherautomatic associations may be responsible for ste-reotype threat performance deficits in young chil-dren.

In this study, we focused on first-grade childrenfor two reasons. So far, the only study showing thepresence of automatic associations consistent withgender stereotypes about math and language inyoung children (Cvencek et al., 2011) consideredchildren from 6 to 7 years altogether (Grades 1 and2). Second, studies showing stereotype-inducedmath performance decrements in young girls haveaddressed 5- to 7-year-old children (Ambady et al.,2001; Tomasetto et al., 2011) or 7-year-olds (Neuville& Croizet, 2007). Thus, the focus on 6-year-olds(Grade 1) allowed us to investigate stereotypicalautomatic associations as well as stereotype threateffects specifically when children enter elementaryschool, thus becoming acquainted for the first timewith formal teaching of math (at least in the Italianschool system).

The study had four goals. First, using a child-friendly version of the IAT (Child–IAT; Baron &Banaji, 2006) we tested whether gender stereotypesabout math and language are present as automaticassociations in 6-year-olds. Consistent with Cvenceket al. (2011), we expected stereotype-consistent auto-matic associations for both girls and boys.

Second, we investigated the malleability of chil-dren’s automatic associations. It was hypothesizedthat children’s stereotypical automatic associationswould increase or decrease in activation dependingon the stereotype content of a prior manipulationtask. We predicted an increased activation of stereo-typical automatic associations in a stereotype-consistentcondition than in a stereotype-inconsistent condition,respectively, confirming or contradicting the tradi-tional gender stereotype about math, with results inthe middle in a control condition, in which the math–gender stereotype was not salient.

Third, using an explicit measure, we investigatedwhether 6-year-olds endorse the belief that boys are

Stereotype Threat Before Stereotype Awareness 253

better than girls at math. Three reasons led us toassess stereotype endorsement: (a) children firstdevelop stereotype endorsement and then stereo-type awareness (e.g., Aboud, 1988; Martin & Ruble,2010), (b) 6-year-olds do not possess stereotypeawareness yet (e.g., Rholes & Ruble, 1984; Selman,1980), and (c) children’s personal stereotypic beliefscould be sufficient to trigger stereotype threat, spe-cifically because at this age children assume thattheir beliefs are shared by others as well (e.g.,Augoustinos & Rosewarne, 2001). Consistent withprevious research (e.g., Ambady et al., 2001; Muzz-atti & Agnoli, 2007), we expected no evidence ofendorsement of the math–gender stereotype.

Finally, like in past research on stereotype threatincluding 6-year-olds (e.g., Tomasetto et al., 2011),we expected girls’ math performance to decreaseunder stereotype threat, that is, in the stereotype-consistent as compared to the control condition andthe stereotype-inconsistent condition, which shouldlead to the highest performance. Importantly, wepredicted the increased activation of stereotype-consistent automatic associations to mediate thedecrease in girls’ performance under stereotypethreat.

Method

Participants

Two hundred and seventy-six first-grade chil-dren (143 girls) participated in the study. All ofthem were born in Italy and were of typical age fortheir grade (M = 77.60 months, SD = 0.28). Chil-dren attended one of six elementary schools in oneof three small towns in the northeast of Italy. Per-missions to conduct the study were granted byschool principals and parents. Six female experi-menters were involved in the data collection, con-ducted during school hours. All children weretested in the same school period to avoid potentialeffects of variability in mathematics curricula acrossclasses at the moment of data collection.

Procedure

Children were tested individually in a quiet roomof their school, while sitting at a desk facing theexperimenter. Each experimental session started byasking participants to color one of three pictures,depending on the experimental condition (stereo-type consistent, control, and stereotype inconsistent)to which they were randomly assigned. When chil-dren finished coloring, the experimenter asked them

to describe the picture, and then left the picture onone side of the desk. After the manipulation task,children were invited to play a computer game (i.e.,the Child–IAT) using a laptop computer. Partici-pants were told that they would see pictures duringthe game, and that they would press the red (A key)or the green (L key) button of the computer board toindicate which picture they saw. After the Child–IAT, children performed eight math calculations(i.e., math test). The experimenter asked one calcula-tion at a time (e.g., “How much is 5 plus 5?”) andchildren had to respond within 5 s. For each ques-tion, the experimenter noted whether the responsewas correct or not (i.e., incorrect or no response by5 s), without providing any feedback. Finally, chil-dren performed the explicit stereotype-endorsementtask. They were shown a picture of a boy and a girlside by side and were told: “These are a boy and agirl. They are 6-year-olds and they are good atschool. Is the boy better at math, the girl better atmath, or are they the same at math?” Next, theexperimenter recorded the response, and partici-pants were thanked, given a candy bar, and dis-missed.

Thus, the administration of the implicit measurealways followed the experimental manipulation andalways preceded the math test followed by the expli-cit measure. This order of the tasks was chosenbecause we aimed at testing the effects of the experi-mental manipulation on automatic associationsruling out the possibility that counterbalancingimplicit and explicit measures might weaken theeffects of the experimental manipulation on auto-matic associations. Moreover, there is no evidencethat performing the IAT before a self-report inducesreactance or assimilation tendencies in the subse-quent self-report (Nosek, Greenwald, & Banaji, 2005).

Materials

Experimental manipulation. According to thestereotype threat model, it is the salience of a nega-tive domain-relevant in-group stereotype thatimpairs targets’ performance. A negative domain-relevant in-group stereotype can be made salient inthree ways: (a) describing the task that targets willsubsequently perform as diagnostic of the abilityrelated to the negative in-group stereotype, (b) mak-ing targets’ in-group salient, or (c) making the con-tent of the negative stereotype salient (Inzlicht &Schmader, 2012). In this study, the manipulationchoice was to make salient the stereotype content(math–gender stereotype). Thus, participants wererandomly assigned to one of three experimental

254 Galdi, Cadinu, and Tomasetto

conditions by inviting them to color a picture, inwhich: (a) a boy correctly resolves a math calculationon a blackboard, whereas a girl fails to respond(stereotype-consistent condition); or (b) a girlcorrectly resolves the calculation, whereas a boy failsto respond (stereotype-inconsistent condition); or(c) a landscape was depicted (control condition).

Implicit measure: The math(language)-gender Child–IAT. A Child–IAT (Baron & Banaji, 2006) was usedto assess the relative strength of automatic associa-tions between the target category boy and theattribute category mathematics, and the target cate-gory girl and the attribute category language, ascompared to the opposite pairings (i.e., boy–lan-guage and girl–mathematics).

Categories and stimuli for the Child–IAT. Four pic-tures of math-related objects (i.e., numbers) and fourpictures of reading- and writing-related objects (i.e.,letters), pretested with first-grade children for famil-iarity and comprehension, were used as stimuli forthe attribute categories mathematics and language.Four face-pictures of boys and four face-pictures ofgirls represented the target categories boy and girl.

Procedure of the Child–IAT. Children were instruc-ted to respond fast and accurately to three simple-categorization (practice) blocks and two (third andfifth) critical double-categorization blocks. Eachpractice block included 16 trials; each critical blockincluded 32 trials. On each trial, a target or attributepicture appeared on the screen until participantsgave their response. Children categorized each pic-ture by pressing a key on the computer board, theleft-hand response A or the right-hand response Lkey. To simplify the task, the left-hand key (A) wascolored in red, and the right-hand key (L) was col-ored in green. The intertrial interval was 200 ms. Ared cross, which remained on the screen for 200 ms,followed incorrect responses.

In the first block of the task, children were pre-sented with pictures for the target categories girland boy, with each picture randomly presentedtwice. Participants had to indicate whether the pic-ture on the screen was a boy or a girl by pressingthe red key for girl and the green key for boy. Inthe second block, children were presented with pic-tures for the attribute categories language andmathematics, with each picture randomly presentedtwice. Participants had to press the red key whenthe picture was a reading- or writing-related objectand the green key when the picture was a math-related object. In the third double-categorizationblock, both pictures of boys and girls, and picturesof math-related and reading- or writing-relatedobjects (with each picture representing each cate-

gory randomly presented twice) appeared on thescreen. In this case, children pressed the red keywhen they saw either a girl or a reading- or writing-related object, and the green key when they saweither a boy or a math-related object. In the fourthblock, participants categorized again the picturesrepresenting the categories mathematics and lan-guage. However, different from the key assignmentin the second block, participants had to press thered key when they saw a math-related object andthe green key when they saw a reading- or writing-related object. Finally, in the fifth double-categoriza-tion block, children categorized the same pictures ofboys and girls and the same pictures of math-relatedand reading- or writing-related objects. However,participants now pressed the red key when they saweither a girl or a math-related object, and the greenkey when they saw either a boy or a reading- orwriting-related object. Following the procedureemployed in the third block, each picture represent-ing each of the four categories was randomly pre-sented twice. The order of the two critical blocks,third and fifth, was counterbalanced across partici-pants to avoid order effects.

Math test. Three days before the data collection,a math test was developed in collaboration with allteachers of the classes participating in the study, tocreate a set of difficult math calculations taking intoaccount the mathematics achievement of all classes.The math test was a retrieval of numerical factsincluding five additions and three subtractions (i.e.,5 + 5, 8 � 4, 6 + 4, 10 � 5, 2 + 3, 2 + 2, 6 � 3, and4 + 5). Children had to resolve each math calcula-tion in 5 s.

Explicit stereotype endorsement. As in Ambadyet al.’s (2001) study, children were presented with aphotograph of a boy and a girl described as goodat school, and had to choose whether the boy or thegirl is especially good at math or whether they areequal at math. The two photographs as well as thequestions’ wording for the boy or the girl werepresented in counterbalanced order.

Results

Preliminary Analyses

Twenty-three girls and thirteen boys wereexcluded only because of 30% or higher error rates inat least one critical (double-categorization) block ofthe Child–IAT. Preliminary analyses confirmed thatexcluded participants were evenly distributed acrossconditions (i.e., 13 in the stereotype-consistent, 10 inthe control, and 13 in the stereotype-inconsistent

Stereotype Threat Before Stereotype Awareness 255

conditions), and that their responses did not differfrom those of retained children on both the math testand the stereotype endorsement. The final sampleincluded 240 children (120 girls).

Individual scores of automatic associations werecalculated by means of the D-algorithm, designed foranalyzing data with the IAT (Greenwald, Nosek, &Banaji, 2003). The D-algorithm compares the extentto which performance on the incompatible criticalblock (i.e., girl–mathematics and boy–language shar-ing the same response key) is impaired relative toperformance on the compatible critical block (i.e.,girl–language and boy–mathematics sharing thesame response key), taking into account participants’individual response latencies, standard deviations oflatencies, and error rates in each of the two blocks.Scores were calculated so that positive scores reflectstronger boy–mathematics and girl–language automaticassociations. To estimate reliability of the Child–IAT,two Child–IAT scores were calculated: one using thefirst half and the other using the second half of thetwo critical blocks. Internal consistency was satisfac-tory (Cronbach’s a = .84).

A one-way analysis of variance (ANOVA) wasthen conducted on scores of automatic associationsusing conditions (stereotype consistent, control, ste-reotype inconsistent), gender (male, female), andthe order of administration of the critical blocks(1 = third block: girl–language and boy–math, fifth block:girl–math and boy–language; 2 = third block: girl–mathand boy–language, fifth block: girl–language and boy–math) as the independent variables. A main effect fororder of administration was found, F(1, 234) = 6.64,p = .01, gp

2 = .03. However, because order ofadministration produced no significant interactions(all ps > .09), it is not further discussed.

For each participant, correct responses to themath test (i.e., correct math calculation in 5 s) werecoded +1, and incorrect responses (i.e., no responseor incorrect math calculation in 5 s) were coded 0.For each child, responses were then added up in asingle math score. To correct for potential variabil-ity in mathematics proficiency across classes, mathscores were standardized within each class of thesix schools. Internal consistency of the math testbased on all eight calculations was satisfactory(Cronbach’s a = .89).

Separate one-way ANOVAs were conducted onscores of automatic associations and math scores,using condition (stereotype consistent, control, ste-reotype inconsistent), gender (male, female), andthe schools as independent variables. Schools didnot lead to any significant result (all ps > .70), andthus this variable is not further discussed.

Zero-order correlations between all dependentvariables (automatic associations, math performance,and explicit math–gender stereotype endorsement),separately for girls and boys across experimentalconditions, are presented in Table 1.

Automatic Associations and Their Malleability

We predicted stereotype-consistent automaticassociations for both girls and boys in the controlcondition, in which the math–gender stereotypewas not made salient. In addition, we aimed at test-ing for the malleability of children’s automatic asso-ciations. Higher scores of automatic associationswere expected in the stereotype-consistent than inthe stereotype-inconsistent condition, with results inthe middle for the control condition.

A two-way ANOVA was conducted on scores ofautomatic associations, with condition (stereotypeconsistent, control, stereotype inconsistent) andgender (male, female) as the between-participantsvariables. Results showed a significant Condition 9

Gender interaction, F(2, 234) = 6.35, p = .002,gp

2 = .05. To assess the effect of condition withingender, simple-effect analyses were conducted sepa-rately for girls and boys. Table 2 presents the aver-age scores of automatic associations in eachexperimental condition for both boys and girls,together with information about the planned con-trasts between the means.

No effect of condition emerged for boys’automatic associations (p > .30). Moreover, one-sam-ple t tests against 0 showed no effect in any condi-tion (all ps > .20). Conversely, simple-effect analysison girls’ automatic associations revealed a signifi-cant effect of condition, F(2, 117) = 7.75, p < .01,

Table 1Zero-Order Correlations Between Automatic Associations, Math Per-formance, and Explicit Stereotype Endorsement for Girls and Boys

1 2 3

Girls (n = 120)Automatic associations —

Math performance �.267** —

Explicit stereotype endorsement �.115 .142 —

Boys (n = 120)Automatic associations —

Math performance �.045 —

Explicit stereotype endorsement .000 .034 —

Note. Pearson’s r coefficients are reported for automatic associationsand math performance. Spearman’s rs coefficients are reportedfor the categorical variable explicit stereotype endorsement.**p < .01.

256 Galdi, Cadinu, and Tomasetto

gp2 = .06, linear trend, p < .001. Post hoc tests with

Bonferroni correction of the alpha level for multiplecomparisons showed that girls’ scores of automaticassociations were lower in the stereotype-inconsis-tent as compared to the control (p < .02) and thestereotype-consistent (p < .01) conditions. As shownin Table 2, although girls’ automatic associationswere the highest in the stereotype-consistentcondition, scores in the control and the stereotype-consistent conditions did not differ from each other(p > .90). Moreover, one-sample t tests against 0showed that girls’ scores of automatic associationswere different from 0 in the stereotype-consistentand in the control (both ps < .02) but not in thestereotype-inconsistent condition (p = .20).

Importantly, simple-effect analyses on the maineffect of gender within conditions revealed thatboys’ scores were lower than girls’ scores ofautomatic associations in the stereotype-consistent,F(1, 234) = 5.90, p < .03, gp

2 = .02, and control con-dition, F(1, 234) = 5.84, p < .03, gp

2 = .02. Althoughthis difference fell short of significance, boys’ scoresof automatic associations tended to be higher thanthose of girls in the stereotype-inconsistent condi-tion, F(1, 234) = 3.79, p = .053, gp

2 = .01.In sum, although automatic associations consis-

tent with math(language)–gender stereotypes werenot found for boys, girls revealed stereotype-consis-tent automatic associations in both the control andthe stereotype-consistent conditions. Importantly,girls’ automatic associations showed the expectedmalleability, thus differing in activation as a functionof the experimental condition: They were higher in

the stereotype-consistent and the control conditionsthan the stereotype-inconsistent conditions. No effectof the experimental manipulation was found forboys’ automatic associations.

Math Performance

For girls, we expected performance decrementsin the stereotype-consistent condition as comparedto the control and the stereotype-inconsistent condi-tions. Conversely, no effect of condition on boys’math performance was hypothesized.

An ANOVA on math scores, with condition(stereotype consistent, control, stereotype inconsis-tent) and gender (male, female) as the between-participants variables, was conducted. A Condi-tion 9 Gender interaction emerged, F(2, 234) = 4.69,p = .010, gp

2 = .04. Therefore, the effect of conditionwas tested separately for girls and boys. Simple-effect analyses on boys’ math scores (Table 2)showed that boys performed equally well acrossconditions (p > .30). To the opposite, an effect ofcondition was found for girls, F(2, 117) = 3.66,p < .03, gp

2 = .03, linear trend, p < .01: Girls under-performed in the stereotype-consistent as comparedto the stereotype-inconsistent condition (p < .03). Onthe contrary, math scores in the control and both inthe stereotype-consistent and in the stereotype-inconsistent conditions were not different from eachother (both ps > .20). Thus, whereas the salience ofthe negative in-group stereotype led girls to performworse on the math test, the removal of the samestereotype led girls to perform better in thestereotype-inconsistent condition as compared tothe stereotype-consistent condition.

Explicit Stereotype Endorsement

To assess stereotype endorsement, participantswere shown a picture of a boy and a girl and wereasked to say whether the boy or the girl is espe-cially good at math, or whether they are equal. Foreach child, stereotype-inconsistent response (i.e., thegirl is better at math) was coded �1, neutral choice(i.e., the boy and the girl are equal) was coded 0, andstereotype-consistent choice (i.e., the boy is better atmath) was coded +1. We expected no evidence thatchildren endorsed the math–gender stereotype.

A logistic regression for ordinal dependent mea-sures was performed on children’s choices at the task(�1 = stereotype-inconsistent, 0 = neutral, +1 = stereo-type-consistent choice), with gender (0 = male,1 = female), condition (�1 = stereotype inconsistent,0 = control, +1 = stereotype consistent), and the product

Table 2Average Scores of Automatic Associations and Math Scores as a Func-tion of Condition (Stereotype Consistent, Control, Stereotype Inconsis-tent) for Girls and Boys

Stereotypeconsistent Control

Stereotypeinconsistent

Girls (n = 120)Automaticassociations

**0.28a(SD = 0.45)

*0.19a(SD = 0.78)

�0.10b(SD = 0.43)

Math test �0.36a(SD = 0.73)

�0.01ab(SD = 0.49)

0.16b(SD = 0.92)

Boys (n = 120)Automaticassociations

0.03a(SD = 0.44)

�0.05a(SD = 0.30)

0.10a(SD = 0.50)

Math test 0.28a(SD = 1.03)

0.05a(SD = 1.02)

�0.03a(SD = .75)

Note. Means within rows not sharing the same subscript are sig-nificantly different from each other at the p < .05 level (Bonfer-roni test).*p < .02. **p < .001.

Stereotype Threat Before Stereotype Awareness 257

term as predictors. A main effect of gender emerged,Wald v2 = 11.49, p < .001, indicating that the proba-bility of a counterstereotypical versus neutral versusstereotypical response was not equal for boys andgirls. Regardless of condition, 57% of boys and 57%of girls favored their own gender and indicated theboy or the girl, respectively, as the most talented formath. Only 12% of boys and 21% of girls respondedthat the boy and the girl were the same. Of theremaining children, 22% of girls identified the boyas being better at math, whereas 31% of boysbelieved that the outstanding math student was thegirl. Thus, consistent with predictions, 6-year-oldsdid not endorse the math–gender stereotype, butsimply manifested gender in-group favoritism.

Mediation Analysis

Because of null findings for boys on both auto-matic associations and math scores, a mediationanalysis was conducted only for girls to testwhether automatic associations mediated the rela-tion between condition and math performance.Given that the predictor (i.e., condition) was a cate-gorical variable with three levels, we created twodummy-coded variables to conduct the mediationanalysis (Hayes & Preacher, 2012) with the stereo-type-consistent condition as the reference group.Specifically, Contrast 1 tested the effect of the ste-reotype-consistent (coded 0) versus stereotype-inconsistent (coded 1) condition, with the controlcondition coded 0. Contrast 2 tested for the residualdifference between the stereotype-consistent (coded0) and the control (coded 1) conditions, with thestereotype-inconsistent condition coded 0.

Consistent with the univariate analyses reportedabove, the effect of Contrast 1 on performance wassignificant, b = .55, t(111) = 3.38, p < .01, whereasthe effect of Contrast 2 fell short of significance,b = .34, t(111) = 1.91, p = .06. Similarly, the effect ofContrast 1 on automatic associations was significant,b = �.37, t(111) = 3.56, p < .001, whereas the effectof Contrast 2 was not, p > .40. Moreover, whenautomatic associations and the two contrasts wereentered simultaneously in the model predictingperformance, the effect of automatic associationswas significant, b = �.36, t(111) = 2.32, p < .03, indi-cating that automatic associations negatively affectperformance. Importantly, the effect of Contrast 1(i.e., stereotype-consistent vs. stereotype-inconsistentcondition) was reduced, b = .38, t(110) = �2.06,p < .05, whereas the effect of Contrast 2 (i.e., stereo-type-consistent vs. control condition) remained notsignificant, b = .31, t(111) = 1.77, p = .07. Figure 1

summarizes results for Contrast 1. To test for media-tion, we calculated a bias-corrected 95% confidenceinterval for the indirect effect (.13, SE = .06) using abootstrapping technique (Preacher & Hayes, 2008).As the null hypothesis of no mediation states thatthe indirect effect is 0, the null hypothesis is rejectedwhen the confidence interval does not include 0. Inthe present analysis, the confidence interval (with5,000 resamples) for the estimate of the indirecteffect of Contrast 1 on performance did not include0, 95% CI [0.02, 0.28], thus confirming that auto-matic associations mediated the relation betweencondition (stereotype consistent vs. stereotypeinconsistent) and girls’ performance.

Supplementary Analyses

Although main analyses showed that stereotypeendorsement did not vary as a function ofcondition, supplementary analyses were conductedto test whether the relations among condition,girls’ automatic associations, and girls’ math per-formance were further moderated by stereotypeendorsement. Two-way analyses of variance wereconducted on both scores of automatic associationsand math performance, with condition (stereotypeconsistent, control, stereotype inconsistent) andexplicit stereotype endorsement (stereotype-consis-tent, neutral, stereotype-inconsistent response) asthe between-participants variables. In bothANOVAs, explicit stereotype endorsement yieldedno significant effects: The effect of condition wassignificant on both scores of automatic associa-tions, F(2, 111) = 7.76.12, p < .001, and math per-formance, F(2, 111) = 4.12, p < .02, and was notfurther qualified by explicit stereotype endorse-ment (all ps > .15). However, given that the rela-tively low number of girls who provided a neutralresponse or indicated the other gender group as

Math PerformanceExperimental

Condition

Automatic Associations

.38* (.55**)

-.36**-.37***

Figure 1. Results of mediation analyses testing indirect effects ofexperimental condition (stereotype consistent = 0, stereotype incon-sistent = 1) on math performance via automatic associations ingirls (n = 120).*p < .05. **p < .01. ***p < .001.

258 Galdi, Cadinu, and Tomasetto

better at math resulted in unbalanced cell frequen-cies, this finding could be a consequence of lowstatistical power. To rule out this possibility, werepeated the same analyses by recoding explicitstereotype endorsement into a dichotomous vari-able, comparing girls indicating their gender groupas better at math with those who either indicatedboys as better or provided a neutral response.Results were very similar: The effect of conditionremained significant on both automatic associa-tions, F(2, 114) = 6.83, p < .01, and math perfor-mance, F(2, 114) = 4.59, p < .01, and was notfurther qualified by explicit stereotype endorse-ment (all ps > .09).

We also tested whether automatic associations mayact as a moderator, rather than a mediator, of theeffect of stereotype activation on girls’ math perfor-mance. An analysis of covariance (ANCOVA) wasperformed on math performance with condition(stereotype consistent, control, and stereotype incon-sistent), automatic associations, and the interactionbetween condition and automatic associations aspredictors. The main effect of automatic associationswas significant, F(1, 114) = 5.57, p < .02, whereas theinteraction between condition and automatic associa-tions was not (p > .20), thus confirming that auto-matic associations acted as a mediator of the relationbetween condition and performance.

Finally, we tested whether automatic associationsand stereotype endorsement may interact with eachother in determining girl’s vulnerability to stereo-type threat. We carried out a three-way ANCOVAon math performance with condition (stereotypeconsistent, control, stereotype inconsistent), auto-matic associations, and explicit stereotype endorse-ment response (stereotype consistent, neutral,stereotype inconsistent) as predictors. Again, onlythe main effect of automatic associations was signif-icant, F(1, 114) = 4.01, p < .05, whereas no othermain effect or higher order interaction attained sig-nificance (all ps > .10).

Although boys showed null findings on alldependent variables, the same supplementary anal-yses were conducted for them as well. In all analy-ses, neither main effects nor interactions reachedsignificance (all ps > .08).

Discussion

Although the stereotype threat model identifies ste-reotype awareness as a requirement for stereotypethreat effects, research has shown that gender iden-tity activation affects girls’ math performance before

the emergence of stereotype awareness, and in theabsence of endorsement of the math–gender stereo-type (Ambady et al., 2001; Neuville & Croizet,2007; Tomasetto et al., 2011). The present researchdisentangles this paradox and extends the stereo-type threat model by demonstrating that automaticassociations consistent with a negative in-group ste-reotype represent a key factor that may triggerstereotype threat in young children.

Consistent with research showing that automaticassociations can precede conscious beliefs (e.g., Gaw-ronski & Galdi, 2011), we demonstrated for the firsttime that girls acquire the math–gender stereotype asautomatic associations before its emergence at theexplicit level: Despite the absence of evidence thatchildren endorsed the math–gender stereotype,6-year-old girls, but not boys, even in the controlcondition (i.e., in the absence of any experimentalmanipulation aimed at increasing or decreasing thesalience of the stereotype) showed stereotype-consis-tent automatic associations between boy and mathe-matics, and girl and language. Moreover, consistentwith research showing malle-ability of automaticassociations in adults (e.g., Dasgupta & Asgari,2004), for the first time, malleability of girls’ auto-matic associations was found. After coloring a draw-ing of a boy correctly solving a math problem, girls’stereotypical automatic associations were activated,as compared to the control condition. At the sametime, after coloring a picture of a girl succeeding inmath, stereotypical automatic associations werereduced.

Importantly, the activation of automatic associa-tions in the stereotype-consistent condition ham-pered girls’ math performance as compared to thestereotype-inconsistent condition, in which, con-versely, the reduced activation of automatic associa-tions led girls to the highest performance.Furthermore, highlighting the role of counterstereo-typical information in counteracting stereotypethreat effects, these findings are parallel to researchshowing that providing information about equalgender abilities in math (i.e., stereotype-inconsistentinformation) is an effective strategy to preventwomen’s performance deficits (e.g., Cadinu, Maass,Frigerio, Impagliazzo, & Latinotti, 2003).

An open issue of this study concerns the poten-tial processes underlying girls’ underperformance inthe stereotype-consistent versus -inconsistent condi-tion. Consistent with Forbes and Schmader (2010),one may speculate that activated stereotypical auto-matic associations in the stereotype consistent con-dition may have burdened working memory capacity,thus disrupting subsequent math performance.

Stereotype Threat Before Stereotype Awareness 259

Conversely, the reduced activation in the stereo-type-inconsistent condition may have freed upworking memory resources, thus enhancing subse-quent performance. However, no such conclusionscan be drawn as this study did not test how girls’automatic associations affect performance. There-fore, a direct test of the working memory hypothe-sis should be the goal of future studies.

Another potential process underlying girls’ worseperformance in the stereotype-consistent than thestereotype-inconsistent condition could be a merepriming effect, which would lead individuals tobehave automatically consistent with a cognitivelyactivated image (e.g., Wheeler & Petty, 2001). How-ever, if a priming effect were at work, one shouldexpect underperformance even for boys in responseto stereotype-related stimuli. To the contrary, nei-ther boys’ automatic associations nor performancewere affected by the activation of the math–gender(counter)stereotype. Thus, differently from a prim-ing effect, we argue that in this study it was themembership in a negatively stereotyped group(whose stereotype has been acquired via automaticassociations) that was responsible for girls’ perfor-mance deficit under stereotype threat.

Although girls’ automatic associations increasedand decreased in activation depending on the ste-reotype content of the experimental condition, nochanges emerged at the level of endorsement of ste-reotypical beliefs. Consistent with research showinggender in-group bias in young children, the major-ity of children across conditions indicated their gen-der as superior in math. We argue that this lack ofendorsement of the math–gender stereotype furtherstrengthens the role of automatic associations in theprocess of stereotype acquisition in children.

Relative to the latest point, the present pattern offindings raises the tricky issue concerning thesources and mechanisms underlying the formationof automatic associations. Up to date, highly influen-tial theorizing has posited that automatic associa-tions stem from attitudes and conscious beliefs thathave become overlearned and automatized over time(e.g., Rudman, 2004; Wilson et al., 2000). However,recent research (e.g., Gawronski & Galdi, 2011), aswell as the present results, suggests that also otherunderlying mechanisms may be responsible for theformation of automatic associations. Drawing on thedistinction between associative and propositionallearning (Gawronski & Bodenhausen, 2006), weargue that the automatic formation of stereotype-con-sistent associative links in children’s cognitive mapcould stem from repeated co-occurrences of objectsor stimuli in children’s social environment. For

example, ample evidence shows that children aresensitive to adults’ nonverbal behavior and that suchbehaviors, often occurring in automatic and uncon-scious ways, may represent an important channelthrough which attitudes are transmitted (e.g., Rud-man, 2004; Walden & Ogan, 1988). Thus, eventhough adults may explicitly encourage nonstereo-typical interests in children, they may exhibit differ-ent patterns of nonverbal behaviors: Parents maypurchase more games or manipulative materialsrelated to math and science for their sons than fortheir daughters (Jacobs & Bleeker, 2004), or intrudemore often in their daughters’ than in their sons’math homework to offer unsolicited help (e.g.,Bahnot & Jovanovic, 2005). As a result, such repeatedco-occurrences of objects and events might promotethe automatic formation of stereotype-consistentassociative links in children’s cognitive maps.

In contrast to girls, boys did not reveal stereotypi-cal automatic associations in any conditions.Consistent with research demonstrating that elemen-tary school boys tend to manifest in-group favoritismregarding both math and language (Steele, 2003),these results could simply reflect the fact that boysshowed in-group–serving automatic associationsabout both math and language domains (Steffenset al., 2010), regardless of the stereotype content con-veyed by the experimental condition. Indeed, thepresence of in-group-serving boy–mathematics andboy–language automatic associations would result infast response latencies in both critical blocks of theChild–IAT. Thus, given that the D-algorithm com-pares the extent to which performance on the incom-patible (i.e., girl–math and boy–language sharing thesame response key) is impaired relative to thecompatible critical block (i.e., girl–language andboy–math sharing the same response key), in-group–serving automatic associations could have led to thesmall or null score of automatic associations that wasfound for boys across conditions. However, becauseboth gender stereotypes regarding math andlanguage contribute to the Child–IAT score, and can-not be separated (Nosek et al., 2005), this possibilitycould not be tested in this study. To our knowledge,only one study used an implicit measure to assessmath–gender and language–gender stereotypesseparately (i.e., Go-No-Go association task [GNAT];Nosek & Banaji, 2001) in a sample of 14-year-oldsand university students. Consistent with the in-group–serving explanation, Steffens and Jelenec’s(2011) male participants revealed both boys–mathstereotypical automatic associations and boys–lan-guage counterstereotypical automatic associations.However, as Steffens and Jelenec noted, “the internal

260 Galdi, Cadinu, and Tomasetto

consistency of the GNAT was low . . . and the IATclearly appears more sensitive” (Steffens & Jelenec,2011, p. 332). Thus, a goal of future studies should beto test whether other implicit measures that combinethe measurement quality of the IAT with the separatemeasurement of math and language gender stereo-types, such as the GNAT, can be implemented andadapted for use with young children.

Studies on the development of cognitive compe-tencies offer alternative explanations for boys’ lackof automatic associations. For example, previousresearch has shown earlier knowledge of gendercategories (Zosuls et al., 2009) and earlier achieve-ment of gender constancy (Ruble et al., 2007) bygirls than boys. At the same time, differences ingender categorization abilities between girls andboys could also be the result of socialization pro-cesses. For example, women show stronger auto-matic associations between the self, their gender,and the in-group gender stereotype as compared tomen (Cadinu & Galdi, 2012), and children fromlow-status groups are more likely to be aware ofself-relevant stereotypes than children from high-status groups (McKown & Weinstein, 2003).

Although consistent with Steffens et al. (2010),the result that boys did not manifest stereotypicalautomatic associations is inconsistent with Cvenceket al.’s (2011) findings, showing stereotypical auto-matic associations between gender and academicsubjects in both 6- to 7-year-old boys and girls.However, this discrepancy might reflect cross-cultural variations in the strength of gender stereo-types at the societal level, which have been foundto be associated with either smaller or bigger gen-der differences in the strength of stereotypical auto-matic associations (Nosek et al., 2009), as well as inthe endorsement of gender stereotypes aboutacademic abilities (e.g., Evans, Schweingruber, &Stevenson, 2002). Moreover, in a study involvingChinese, Japanese, and American children, Lummisand Stevenson (1990) found that whereas Americanand Chinese 6-year-olds do not believe that thereare gender differences in math, 6-year-old Japanesechildren tend to believe that boys are better thangirls at math (see also Del Rio & Strasser, 2013).These results show that cultural differences in thestrength of gender stereotypes about academic abili-ties, in schooling, and in out-of-school experiences,may lead to different acquisition ages of theendorsement of these stereotypes, and suggest thatthe same could be true regarding the acquisition ofstereotypical automatic associations.

In conclusion, this study has the potential to con-tribute to the stereotype threat model by suggesting

automatic associations as a novel prerequisite forstereotype-induced underperformance in youngchildren. Importantly, by showing also that chil-dren’s automatic associations are malleable, thesefindings are promising in terms of interventions topromote gender equality in math and sciencebecause they suggest that girls can be protectedfrom the deleterious impact of math–gender stereo-types. Relevant to this claim is a study (Dasgupta &Greenwald, 2001) demonstrating that Caucasianparticipants exposed to images of admired Blackand disliked White exemplars showed lower pro-White automatic associations than participantsexposed to images of admired Whites and dislikedBlacks. Interestingly, such a decrease in pro-Whiteautomatic associations lasted for 24 hr, suggestingthat relatively enduring changes in automatic associ-ations can be obtained. If so, repeatedly presentinggirls with exemplars of successful women in mathand science could promote the reduction in stereo-typical automatic associations, long before girls haveacquired any awareness or endorsement of the ste-reotype favoring males in math. This study suggeststhat this strategy could protect girls’ performance instereotype-threatening situations and potentiallyhelp them to expand their interests toward tradi-tionally male domains.

References

Aboud, F. E. (1988). Children and prejudice. New York:Basil Blackwell.

Ambady, N., Shih, M., Kim, A., & Pittinsky, T. (2001).Stereotype susceptibility in children: Effects of identityactivation on quantitative performance. PsychologicalScience, 12, 385–390. doi:10.1111/1467-9280.00371

Aronson, J., & Good, C. (2003). The development andconsequences of stereotype vulnerability in adolescents.In F. Pajares & T. Urdan (Eds.), Adolescence and educa-tion: Vol. 2. Academic motivation of adolescents (pp. 299–330). Greenwich, CT: Information Age.

Augoustinos, M., & Rosewarne, D. L. (2001). Stereotypeknowledge and prejudice in children. British Journal ofDevelopmental Psychology, 19, 143–156. doi:10.1348/026151001165912

Bahnot, R., & Jovanovic, J. (2005). Do parents’ academicgender stereotypes influence whether they intrude ontheir children’s homework? Sex Roles, 52, 597–607.doi:10.1007/s11199-005-3728-4

Baron, A. S., & Banaji, M. (2006). The development ofimplicit attitudes. Evidence of race evaluations fromages 6 and 10 and adulthood. Psychological Science, 17,53–58. doi:10.1111/j.1467-9280.2005.01664.x

Cadinu, M., & Galdi, S. (2012). The cognitive accessibilityof womanhood: Implicit gender self-categorization

Stereotype Threat Before Stereotype Awareness 261

leads to stronger self-stereotyping by women than men.European Journal of Social Psychology, 42, 546–551. doi:10.1002/ejsp.1881

Cadinu, M., Maass, A., Frigerio, S., Impagliazzo, L., &Latinotti, S. (2003). Stereotype threat: The effect ofexpectancy on performance. European Journal of SocialPsychology, 33, 267–285. doi:10.1002/ejsp.145

Cvencek, D., Meltzoff, A., & Greenwald, A. G. (2011).Math-gender stereotypes in elementary school children.Child Development, 82, 766–779. doi:10.1111/j.1467-8624.2010.01529.x

Dasgupta, N., & Asgari, S. (2004). Seeing is believing:Exposure to counter-stereotypic women leaders and itseffect on the malleability of automatic gender stereo-typing. Journal of Experimental Social Psychology, 40, 642–658. doi:10.1016/j.jesp.2004.02.003

Dasgupta, N., & Greenwald, A. G. (2001). On themalleability of automatic attitudes combating automaticprejudice with images of admired and dislikedindividuals. Journal of Personality and Social Psychology,81, 800–814. doi:10.1037/0022-3514.81.5.800

Davies, P. G., Spencer, S. J., Quinn, D. M., & Gerhard-stein, R. (2002). Consuming images: How televisioncommercials that elicit stereotype threat can restrainwomen academically and professionally. Personality andSocial Psychology Bulletin, 28, 1615–1628. doi:10.1177/014616702237644

Del Rio, M. F., & Strasser, K. (2013). Preschool children’sbeliefs about gender differences in academic skills. SexRoles, 68, 231–238. doi:10.1007/s11199-012-0195-6

Eagly, A. H., & Chaiken, S. (1993). The psychology of atti-tudes. Fort Worth, TX: Harcourt Brace Jovanovich.

Evans, E. M., Schweingruber, H., & Stevenson, H. W.(2002). Gender differences in interest and knowledgeacquisition: The United States, Taiwan, and Japan. SexRoles, 47, 153–167. doi:10.1023/A:1021047122532

Forbes, C., & Schmader, T. (2010). Retraining attitudesand stereotypes to affect motivation and cognitivecapacity under stereotype threat. Journal of Personalityand Social Psychology, 99, 740–754. doi:10.1037/a0020971

Gawronski, B., & Bodenhausen, G. V. (2006). Associativeand propositional processes in evaluation: An integra-tive review of implicit and explicit attitude change. Psy-chological Bulletin, 132, 692–731. doi:10.1037/0033-2909.132.5.692

Gawronski, B., & Galdi, S. (2011). Using implicit mea-sures to read the minds of undecided voters. In M.Cadinu, S. Galdi, & A. Maass (Eds.), Social perception,cognition, and language in honour of Arcuri (pp. 203–216).Padova, Italy: CLEUP.

Gawronski, B., & LeBel, E. P. (2008). Understanding pat-terns of attitude change: When implicit measures showchange, but explicit measures do not. Journal of Experi-mental Social Psychology, 44, 1355–1361. doi:10.1016/j.jesp.2008.04.005

Gawronski, B., & Payne, B. K. (Eds.). (2010). Handbook ofimplicit social cognition: Measurement, theory, and applica-tions. New York: Guilford Press.

Greenwald, A. G., McGhee, D. E., & Schwartz, J. K. L.(1998). Measuring individual differences in implicit cog-nition: The implicit association test. Journal of Personalityand Social Psychology, 74, 1464–1480. doi:10.1037/0022-3514.74.6.1464

Greenwald, A. G., Nosek, B. A., & Banaji, M. R. (2003).Understanding and using the implicit association test: I.An improved scoring algorithm. Journal of Personalityand Social Psychology, 85, 187–216. doi:10.1037/0022-3514.85.2.197

Hamilton, D. L., & Trolier, T. K. (1986). Stereotypes andstereotyping: An overview of the cognitive approach. InJ. Dovidio & S. Gaertner (Eds.), Prejudice, discrimination,and racism (pp. 127–163). San Diego, CA: AcademicPress.

Hayes, A. F., & Preacher, K. J. (2012). Statistical mediationanalysis with a multicategorical independent variable. Man-uscript under review. Retrieved November 10, 2012,from http://www.afhayes.com/public/hp2012.pdf

Inzlicht, M., & Schmader, T. (Eds.). (2012). Stereotypethreat: Theory, process, and application. New York: OxfordUniversity Press.

Jacobs, J. E., & Bleeker, M. M. (2004). Girl’s and boy’sdeveloping interests in math and science: Do parentsmatter? New Directions for Child and Adolescent Develop-ment, 106, 5–21. doi:10.1037/0022-0663.96.1.97

Katz, P. A., & Kofkin, J. A. (1997). Race, gender, and youngchildren. In S. S. Luthar, J. A. Burack, D. Cicchetti, & J.Weisz (Eds.), Developmental psychopathology: Perspectiveson adjustment, risk, and disorder (pp. 51–74). New York:Cambridge University Press.

Kiefer, A. K., & Sekaquaptewa, D. (2007). Implicit stereo-types, gender identification, and math-related out-comes: A perspective study of female college students.Psychological Science, 18, 13–18. doi:10.1111/j.1467-9280.2007.01841.x

Lummis, M., & Stevenson, H. W. (1990). Gender differ-ences in beliefs and achievement: A cross-culturalstudy. Developmental Psychology, 26, 254–263. doi:10.1037/0012-1649.26.2.254

Martin, C., & Ruble, D. (2010). Patterns of gender devel-opment. Annual Review of Psychology, 61, 353–381.doi:10.1146/annurev.psych.093008.100511

Martinot, D., Bag�es, C., & D�esert, M. (2012). French chil-dren’s awareness of gender stereotypes about mathe-matics and reading: When girls improve theirreputation in math. Sex Roles, 66, 210–219. doi:10.1007/s11199-011-0032-3

Martinot, D., & D�esert, M. (2007). Awareness of a genderstereotype, personal beliefs and self-perceptions regard-ing math ability: When boys do not surpass girls. SocialPsychology of Education, 10, 455–471. doi:10.1007/s11218-007-9028-9

McKown, C., & Strambler, M. J. (2009). Developmentalantecedents and social and academic consequences ofstereotype-consciousness in middle childhood. ChildDevelopment, 80, 1643–1659. doi:10.1111/j.1467-8624.2009.01359.x

262 Galdi, Cadinu, and Tomasetto

McKown, C., & Weinstein, R. (2003). The developmentand consequences of stereotype consciousness in mid-dle childhood. Child Development, 74, 498–515. doi:10.1111/1467-8624.7402012

Muzzatti, B., & Agnoli, F. (2007). Gender and mathemat-ics: Attitudes and stereotype threat susceptibility inItalian children. Developmental Psychology, 43, 747–759.doi:10.1037/0012-1649.43.3.747

Neuville, E., & Croizet, J. (2007). Can salience of genderidentity impair math performance among 7–8 years oldgirls? The moderating role of task difficulty. EuropeanJournal of Psychology of Education, 22, 307–316. doi:10.1007/BF03173428

Nosek, B. A., & Banaji, M. R. (2001). The go/no-go associ-ation task. Social Cognition, 19, 625–666. doi:10.1521/soco.19.6.625.20886

Nosek, B. A., Banaji, M. R., & Greenwald, A. G. (2002).Harvesting implicit group attitudes and beliefs from ademonstration web site. Group Dynamics, 6, 101–115.doi:10.1037/1089-2699.6.1.101

Nosek, B. A., Greenwald, A. G., & Banaji, M. R. (2005).Understanding and using the Implicit Association Test:II. Method, variables, and construct validity. Personalityand Social Psychology Bulletin, 31, 166–180. doi:10.1177/0146167204271418

Nosek, B. A., & Smyth, F. L. (2011). Implicit social cogni-tions predict sex differences in math engagement andachievement. American Educational Research Journal, 48,1125–1156. doi:10.3102/0002831211410683

Nosek, B. A., Smyth, F., Sriram, N., Lindner, N., Devos,T., Ayala, A., et al. (2009). National differences ingender-science stereotypes predict national sexdifferences in science and math achievement. Proceed-ings of the National Academy of Sciences of the UnitedStates of America, 106, 10593–10597. doi:10.1073/pnas.0809921106

Olson, M. A., & Fazio, R. H. (2006). Reducing automaticallyactivated racial prejudice through implicit evaluativeconditioning. Personality and Social Psychology Bulletin,32, 421–433. doi:10.1177/0146167205284004

Perner, J., & Wimmer, H. (1985). “John thinks that Marythinks that…: Attribution of second-order beliefs by 5- to10-year old children. Journal of Experimental Child Psychol-ogy, 39, 437–471. doi:10.1016/0022-0965(85)90051-7

Powlishta, K. K. (1995). Gender bias in children’s percep-tions of personality traits. Sex Roles, 32, 17–28. doi:10.1037/0012-1649.31.5.781

Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and re-sampling strategies for assessing and comparing indirecteffects in multiple mediator models. Behavior ResearchMethods, 40, 879–891. doi:10.3758/BRM.40.3.879

Quintana, S. M. (1998). Children’s developmental under-standing of ethnicity and race. Applied and PreventivePsychology, 7, 27–45. doi:10.1016/S0962-1849(98)80020-6

Rholes, W. S., & Ruble, D. N. (1984). Children’s under-standing of dispositional characteristics of others. ChildDevelopment, 55, 550–560.

Ruble, D., Taylor, L. J., Cyphers, L., Greulich, F. K., Lurye,L. E., & Shrout, P. (2007). The role of gender constancyin early gender development. Child Development, 78,1121–1136. doi:10.1111/j.1467-8624.2007.01056.x

Rudman, L. A. (2004). Sources of implicit attitudes. Cur-rent Directions in Psychological Science, 13, 79–82. doi:10.1111/j.0963-7214.2004.00279.x

Selman, R. (1980). The growth of interpersonal understand-ing: Developmental and clinical analysis. New York: Aca-demic Press.

Steele, C. M. (1997). A threat in the air: How stereotypesshape intellectual identity and performance. AmericanPsychologist, 52, 613–629. doi:10.1037/0003-066X.52.6.613

Steele, C. M., & Aronson, J. (1995). Stereotype threat andintellectual test performance of African Americans.Journal of personality and Social Psychology, 69, 797–811.doi:10.1037/0022-3514.69.5.797

Steele, J. (2003). Children’s gender stereotypes aboutmath: The role of stereotype stratification. Journal ofApplied Social Psychology, 33, 2587–2606. doi:10.1111/j.1559-1816.2003.tb02782.x

Steffens, M. C., & Jelenec, P. (2011). Separating implicitgender stereotypes regarding math and language:Implicit ability stereotypes are self-serving for boys andman, but not for girls and women. Sex Roles, 64, 324–335. doi:10.1007/s11199-010-9924-x

Steffens, M. C., Jelenec, P., & Noack, P. (2010). On theleaky math pipeline: Comparing implicit math-genderstereotypes and math withdrawal in female and malechildren and adolescents. Journal of Educational Psychol-ogy, 102, 947–963. doi:10.1037/a0019920

Thoman, D., White, P., Yamawaki, N., & Koishi, H.(2008). Variations of gender-math stereotype contentaffect women’s vulnerability to stereotype threat. SexRoles, 58, 702–712. doi:10.1007/s11199-008-9390-x

Tomasetto, C., Alparone, F. R., & Cadinu, M. (2011).Girls’ math performance under stereotype threat: Themoderating role of mothers’ gender stereotypes. Devel-opmental Psychology, 47, 943–949. doi:10.1037/a0024047

Walden, T., & Ogan, T. (1988). The development of socialreferencing. Child Development, 59, 1230–1240.

Wheeler, S. C., & Petty, R. E. (2001). The effects of stereo-type activation on behavior: A review of possible mech-anisms. Psychological Bulletin, 127, 797–826. doi:10.1037/0033-2909.127.6.797

Wilson, T. D., Lindsey, S., & Schooler, T. Y. (2000). Amodel of dual attitudes. Psychological Review, 107, 101–126. doi:10.1037/0033-295X.107.1.101

Yee, M., & Brown, R. (1994). The development of genderdifferentiation in young children. British Journal of SocialPsychology, 33, 183–196. doi:10.1111/j.2044-8309.1994.tb01017.x

Zosuls, K. M., Ruble, D., Tamis-LeMonda, C. S., Shrout,P. E., Bornstein, M. H., & Greulich, F. K. (2009). Theacquisition of gender labels in infancy: Implications forsex-typed play. Developmental Psychology, 45, 688–701.doi:10.1037/a0014053

Stereotype Threat Before Stereotype Awareness 263