Automatic Prejudice in Childhood and Early Adolescence

Automatic Prejudice in Childhood and Early AdolescenceJuliane Degner

University of AmsterdamDirk Wentura

Saarland University

Four cross-sectional studies are presented that investigated the automatic activation of prejudice inchildren and adolescents (aged 9 years to 15 years). Therefore, 4 different versions of the affectivepriming task were used, with pictures of ingroup and outgroup members being presented as prejudice-related prime stimuli. In all 4 studies, a pattern occurred that suggests a linear developmental increase ofautomatic prejudice with significant effects of outgroup negativity appearing only around the ages of 12to 13 years. Results of younger children, on the contrary, did not indicate any effect of automaticprejudice activation. In contrast, prejudice effects in an Implicit Association Test (IAT) showed highlevels of prejudice independent of age (Study 3). Results of Study 4 suggest that these age differencesare due to age-related differences in spontaneous categorization processes. Introducing a forced-categorization into the affective priming procedure produced a pattern of results equivalent to thatobtained with the IAT. These results suggest that although children are assumed to acquire prejudice atmuch younger ages, automatization of such attitudes might be related to developmental processes in earlyadolescence. We discuss possible theoretical implications of these results for a developmental theory ofprejudice representation and automatization during childhood and adolescence.

Keywords: automatic attitudes, implicit prejudice, childhood, adolescence, development

In the last 2 decades, the field of intergroup research hasexperienced a trend towards new social-cognitive theories andmeasures of prejudice and stereotypes. Considerable progress hasbeen made in understanding the mental representation and cogni-tive processes underlying stereotypes and prejudice (for reviewssee, e.g., Fiske, 2005; Mackie & Smith, 1998). For example,various studies have revealed how and under which conditionsprejudice and stereotypes can be activated, without the perceiversintent, merely by exposure to a relevant stimulus cue in theenvironment. It is a widely accepted assumptionyet little ex-ploredthat such automatic prejudice is rooted in early socializa-tion experiences and learning processes in childhood: A broad andfrequent use of stereotypes and prejudice from childhood onwardspromotes their automatization (e.g., Devine, 1989). This assump-tion is in line with general theories of attitude automatizationwhich is typically conceptualized as a slow-learning process basedon gradual accumulation of trace experiences or well-learned as-sociations (e.g., Fazio, Sanbonmatsu, Powell, & Kardes, 1986;Greenwald, & Banaji, 1995; Rudman, 2004). It is interesting to

note that the underlying developmental processes of prejudiceautomatization have not yet been studied thoroughly (but seeBaron & Banaji, 2006; Rutland, Cameron, Milne, and McGeorge,2005 for notable exceptions), and the field of social cognitionresearch lacks studies focusing on younger participants. The aim ofthe current research is to fill that void and explore automaticprejudice activation in children and adolescents. As the researchwas conducted in Germany and the Netherlands, we focus onattitudes towards Turkish and Moroccan immigrants because theyconstitute the largest ethnic minority groups in these two countries(approx. 3.4% of the German resident population; Woellert,Krohnert, Sippel, & Klingholz, 2009; approx. 4.4% of the Dutchresident population; Centraal Bureau voor de Statistiek Neder-lands, 2009). It has been repeatedly demonstrated that 10% to 20%of German school children and one fourth to one third of Dutchand German adolescents openly express negative attitudes towardsTurkish and Moroccan immigrants (e.g., Boehnke, Hagen, &Hefner, 1998; Frindte, Funke, & Waldzus, 1996; Masson &Verkuyten, 1993; Wagner, van Dick, Pettigrew, & Christ, 2003). Ifearly acquisition and frequent use of prejudice throughout child-hood and adolescence promotes their automatization (Devine,1989), automatic prejudice should either come along with the openexpression of prejudice or follow soon after. Nonetheless, notheory offers clear predictions about at which age, and under whichconditions, automatic prejudice is to be expected.

Developmental Theories of Intergroup PrejudiceDevines (1989) dual process theory suggests that children of an

unspecified young age passively store parental and societal inter-group attitudes without being able to actively examine or disputethese attitudes. She assumed that these early acquired societalattitudes are stably represented in long term memory and form thebasis on which later personal attitudes and beliefs are developed

Juliane Degner, Department of Social Psychology, University of Am-sterdam, the Netherlands; Dirk Wentura, Department of Psychology, Saar-land University, Saarbrucken, Germany.

The research reported in this article was supported by Grant WE2284/3-5 from the Deutsche Forschungsgemeinschaft to Dirk Wentura andJuliane Degner. We thank Jasmin Schneider, Carmen Kuhn, Anna War-necke, and Jellie Sierksma for their assistance in data collection and RahulPremraj and Skyler Hawk for thoughtful comments and stylistic correc-tions.

Correspondence concerning this article should be addressed to JulianeDegner, Department of Social Psychology, University of Amsterdam,Roetersstraat 15, 1018WB Amsterdam, the Netherlands. E-mail:[email protected]

Journal of Personality and Social Psychology 2010 American Psychological Association2010, Vol. 98, No. 3, 356374 0022-3514/10/$12.00 DOI: 10.1037/a0017993

356

when the child is equipped with the necessary (yet unspecified)cognitive abilities. Because the early stored societal attitudes havea longer history of activation and are thus assumed to be repre-sented as strong associative structures, they continue to influencethe older childs and the adults thoughts, feelings, and behaviorstowards outgroups. Devines theory concurs with social develop-mental theories of intergroup prejudice (e.g., Aboud, 1988, 2008;Bigler & Liben, 2006; Nesdale, 1999, 2001) in that they all assumethat social categories and their societal evaluations are acquired ata very early age. Starting at the age of 3 years, children becomeable to distinguish and identify social categories, especially whenthese are perceptually distinguishable (e.g., by skin color), explic-itly labeled (e.g., Blacks, African Americans), salient by propor-tional group size (e.g., minority status), and explicitly and/orimplicitly loaded with social significance (e.g., observed segrega-tion and discrimination). This early awareness is typically accom-panied by strong ingroup preferences that seem to increase untilthe ages of 7 years or 8 years. This early acquisition and use ofsocietal evaluations of social groups and their members could thusbe the basis of prejudice activation and automatization. Further-more, Devine (1989) assumed that at a later age, dissociationsbetween this (passively acquired) societal knowledge and morevariable personal beliefs about social groups occur because ofage-related expansion of cognitive ability and flexibility which inturn allows for the development of egalitarian and nonprejudicedviews. Similarly, Aboud (1988, 2008) assumed in her social cog-nitive developmental theory that prejudice expression declines inolder children (above 7 years) because of developmental shifts inchildrens cognitive abilities which are related to their focus ofattention (from the self to groups to individuals) and dominantinformation processes (from perceptual to affective to cognitive).Thus, the older child is assumed to be increasingly able to under-stand individual rather than group-based characteristics of peopleand, therefore, develops more egalitarian views and reduces dis-criminative behaviors. According these theories, one would as-sume to find automatic prejudice activation at a very early age withan increasing dissociation to deliberate prejudice expression abovethe age of 7 years.

Another social developmental theory, the social identity devel-opment theory (Nesdale, 1999, 2001), challenges these assump-tions. After reviewing existing research in intergroup biases inchildren, Nesdale (1999, 2001, 2001) pointed out that the earlyexpressions of intergroup biases found in prejudice research aredriven by mere ingroup preference (in forced choice researchparadigms) and have little reference to negative evaluations as thebasis of outgroup prejudice. He suggested that although well awareof societal prejudice at an early age (as suggested above), childrenadopt these outgroup prejudice as personal attitudes only later inlife and with higher interindividual variability, related to (a) de-velopmental processes of social identity formation and (b) chil-drens varying perceptions of threat, tensions, or conflicts betweensocial groups. When childrens awareness of their own member-ship in a social group holding negative outgroup attitudes shiftsinto active identification with this ingroup, children approve theseattitudes as their own, depending on their varying perceptions ofintergroup relations. Nesdale (1999, 2001) assumed that the start-ing age for the shift in social identity and intergroup attitudes isaround the age of 7 years. This is in line with research associatinglate childhood and early adolescence as a critical phase of active

identity development that includes increasing exploration and for-mation of ones own social (ethnic) identities (e.g., French, Seid-man, Allen, & Aber, 2006; Newman & Newman, 2002; Phinney &Chavira, 1992). During these years, children and adolescents de-liberately face their racial, ethnic, cultural, national, and sexualidentities and begin to more consciously identify with their in-groups. This identification is accompanied by an internalization ofingroup values, beliefs, and attitudes, as well as acknowledgmentand accentuation of differences between ones own group andother groups. One could thus assume that social categories becomepersonally and socially meaningful only during late childhood andearly adolescence and are from then on more frequently used inrelation to social identity development (Nesdale, 2001). If preju-dice automatization relies on frequent activations and use of out-group categorization and evaluation, the social identity develop-ment perspective would predict a far later onset of automaticprejudice with a much higher interindividual variability. Thus,from current theorizing, opposing predictions about the onset ofautomatic prejudice can be derived with an early onset followingpassive acquisition and nondeliberate use of societal intergroupattitudes in early childhood versus a late onset following activeacquisition and motivated use in late childhood and early adoles-cence.

Hitherto, there are only few studies that explored automaticprejudice in younger participants. Using a child-friendly version ofthe Implicit Association Test (IAT, Greenwald, McGhee, &Schwartz, 1998) in a cross-sectional study, Baron and Banaji(2006) found that 6-year-old children showed the same amplitudeof automatic prejudice as 10-year-olds and young adults aged 19years. Similarly, Rutland and colleagues (2005) found evidence ofimplicit intergroup prejudice on an IAT with children aged from 6years to 8 years and from 10 years to 12 years, as well asadolescents aged 14 years to16 years. Not only were IAT scoresthe same size across age groups but they were also unaffected bychildrens knowledge of social norms or concerns for self-presentation. The finding of automatic prejudice at such early ageclearly supports the first perspective: It might be concluded thatautomatic prejudice is acquired soon after the age at which chil-dren learn the conception of ethnic categories and remains rela-tively stable across development, stored in memory as stablevalence associations. Automatization of negative intergroup atti-tudes would thus not depend on frequent recall or expression ofnegativity toward outgroups and their members (e.g., Banaji,Baron, Dunham, & Olson, 2008).

Assessing Automatic PrejudiceIt is noticeable that the above cited studies relied on the IAT,

currently the most prominent response timebased indirect mea-sure of attitudes in social psychology. The IAT involves repeatedlycategorizing target stimuli according their social category mem-bership (e.g., Black vs. White) and attribute stimuli according theirvalence (i.e., positive vs. negative), whereas the response-categoryassignments are varied (see Greenwald et al., 1998, for a detaileddescription). Without discrediting the valuable results found usingthe IAT, it remains a worthwhile endeavor to expand our knowl-edge about developmental gradients of automatic prejudice-relatedevaluations by using other measures. Related to the aforemen-tioned theoretical considerations, there are several reasons why

357AUTOMATIC PREJUDICE IN CHILDREN AND ADOLESCENTS

IAT results should be complemented by research with other mea-sures of automatic prejudice. First, the IAT is a comparativemeasure and does not allow distinction of ingroup positivity fromoutgroup negativity. The documented stable IAT effect across agegroups might be the result of a stable ingroup positivity, of a stableoutgroup negativity, or of complementary developmental changesin ingroup and outgroup evaluations. For example, if decreasingingroup positivity was accompanied by increasing outgroup neg-ativity, a stable IAT effect would be observed but essential devel-opmental changes would remain hidden. Second, the IAT is acategory-based measure, in that it assesses evaluative reactions togiven category labels (e.g., Blacks and Whites). However, negativeevaluations of a social category do not necessarily corresponddirectly to evaluative reactions toward individual exemplars of thatcategory (see Livingston & Brewer, 2002; Olson & Fazio, 2003).It is conceivable that childrens aforementioned early awareness ofsocietal evaluations of social categories translates into correspond-ing category-based IAT scores when group labels are made salient,which may or may not be related to automatic negative reactionstoward individual exemplars of that category. Also, categoriza-tions in the IAT are forced-choice: The target has to be categorizedinto one of two given groups. Without further research, we do notknow for sure that these categories are actually the ones childrenspontaneously use as a basis of impression formation in daily lifeencounters. Third, there is an ongoing debate about the extent towhich IAT effects are related to factors other than associationsbetween concepts (see De Houwer, Teige-Mocigemba, Spruyt &Moors, 2009, for a recent review). For example, it has beenproposed that effects of stimulus salience, familiarity, and percep-tual fluency might play a substantial role in producing effects inthe IAT (e.g., Brendl, Markman, & Messner, 2001; Chang &Mitchell, 2009; Rothermund & Wentura, 2004). From this criticalperspective, stable IAT effects across different age groups woulddemonstrate a stable difference between ingroup and outgroupmembers regarding their salience, familiarity, and/or perceptualfluency and do not necessarily point to evaluative associations.Finally, there is some evidence that the IAT might be susceptibleto demand effects, such as when adult participants produce desiredIAT scores in relation to novel objects if they think the experi-menter expects them to do so (De Houwer, Beckers, & Moors,2007). This has important implications for the use of the IAT forexploring developmental aspects of attitudes. For example, it can-not be ruled out that children assume that they are asked to shownegative evaluations of outgroup members if that seems to be thesocietal standard. Thus, it cannot be distinguished whether a givenscore in an IAT reflects a childs personal prejudice or her or hisawareness of societal prejudice (see Han, Olson, & Fazio, 2006;Olson & Fazio, 2004; for a related argument).

The aforementioned issues illustrate that complementing re-search with other measures is needed to get a clearer picture ofprejudice acquisition and automatization throughout childhood andadolescence. For our studies, we decided to use another prominentindirect attitude measure, the affective priming task (Fazio et al.,1986; Fazio, Jackson, Dunton, & Williams, 1995). In the affectivepriming task, two stimuli are presented in short succession, thefirst being the prime, the second being the target. The main task ofthe participants is to categorize the targets with regard to theirvalence as positive or negative. Although the valence of the primeis completely irrelevant for the task, it is known that it promotes

the response to a target of same valence, whereas it interferes withthe response to a target of opposite valence. Thus, differences inresponse latencies to positive versus negative targets in relation toattitudinal primes allow inferring the prime valence. The underly-ing mechanism of affective priming effects seems to be rather wellexplained and directly linked to automatic activation of primevalence (De Houwer et al., 2009; Degner, Wentura, & Rother-mund, 2006; Klauer & Musch, 2003; Wentura & Degner, in press).The affective priming task offers a number of virtues for thecurrent research question. First, it is assumed that affective prim-ing effects are primarily determined by the attitudes towards theindividual prime stimuli because participants do not have to con-strue primes in terms of their social category memberships. Thus,the paradigm permits conclusions about the degree to which ex-emplar outgroup members automatically activate evaluations whenthey are presented as primes (Fazio & Olson, 2003; Livingston &Brewer, 2002; also see De Houwer, 2001). Nonetheless, coverstory and task instructions can be varied in a way that instigatescategory-based versus exemplar-based processing of primes (Fazio& Olson, 2003). Second, the design of the affective priming taskallows separate consideration of age differences in ingroup versusoutgroup evaluations, although this should be done with caution(see Scherer & Lambert, 2009; Wentura & Degner, in press).Third, the procedure allows primes to be presented outside ofparticipants conscious perception, and demand effects or anyeffects of intentional modulation can therefore be avoided (at leastwith masked prime presentation; see Degner, 2009). Also, it hasbeen shown that affective priming effects are less susceptible tonormative information or prevalent societal attitudes (Han et al.,2006). Thus, priming effects might be straightforwardly inter-preted as indicators of personal attitudes instead of mere awarenessof societal attitudes. In conclusion, even though affective primingand the IAT both belong to the class of indirect response latencybased (implicit) attitude measures, they do not necessarily tap thesame underlying processes and/or concepts. Although finding con-verging results will certainly be considered as mutual evidence forvalidity, diverging results would suggest a more differentiatedpattern of processes related to automatic social evaluations inchildren and adolescents.

Overview

To complement existing studies on age effects in automaticprejudice, we used the affective priming task in four cross-sectional studies with German and Dutch school children, agedfrom 9 years to 15 years (overall sample size N 838). As thepriming measure is based on a simple evaluative categorizationtask, we expected it to be easily accomplished by children in theseage groups. To our knowledge, no studies have yet used thisparadigm with participants of a younger age. Thus, the generalapplicability of the measure to younger participants had to bedemonstrated. We therefore included a standard priming conditionusing normatively positive and negative primes, which shouldresult in the typical affective priming effects. For the assessment ofprejudice-related evaluations, we included pictures of persons ofTurkish (Study 1 to 3) and Moroccan (Study 4) origin as outgroupprimes and pictures of White German (Study 1 to 3) and Dutchpeople (Study 4) as ingroup primes.

358 DEGNER AND WENTURA

The main focus of the studies was to explore whether automaticprejudice activation in the affective priming task differs accordingto the age of participants. Study 1 and Study 2 complement oneanother: In Study 1, we used briefly presented, but clearly percep-tible, pictures of Turks and Germans as primes. In Study 2,however, we used masked presentation of primes thus precludingconscious identification of the primes. In Study 3, the aim was tomake a direct comparison between the affective priming task andthe IAT. In Study 4, we tested the assumption that age differencesfound in automatic prejudice activation in the affective primingtask are related to childrens use of ethnic categories in personperception.

Study 1

In Study 1, we employed an affective priming task with portraitsof young men of German versus Turkish origin presented asclearly visible primes. To establish whether the affective primingparadigm is generally applicable for the assessment of automaticevaluations in children, we implemented a control condition usingstandard positive and negative stimuli as primes (i.e., emotionalpictures). Thus, the main goals in Study 1 were (a) to ascertain thegeneral applicability of the affective priming paradigm with chil-dren and (b) to explore age differences of automatic prejudice-related evaluations in the paradigm. In addition, we tried to im-plement a between-participants manipulation of prime construalvia different instructions in a prime memorization task, intended toinitiate category-based versus individual-based processing ofprime pictures (see Olson & Fazio, 2003). However, during datacollection, we observed that children ignored or forgot the mem-orization task, which eventually proved to have no influence onpriming effects. We therefore do not discuss this manipulation indetail. However, we return to this issue with Study 4, where weimplemented a successful manipulation of prime construal.

MethodParticipants. Participants were 304 students from Grades 5 to

8 in two public German high schools. The schools were situated intwo midsized towns in Southwestern Germany with comparablepercentages of immigration in the city populations (12.9% and 9.6%,respectively). Approximately 3% to 5% of the school populationswere from a visible ethnic minority, and all teachers were WhiteGerman. The sample is divided into 81 fifth-graders (46 boys, 35girls; median age 11 years, ranging from 10 years to 12 years), 84sixth-graders (45 boys, 39 girls; median age 12 years, ranging from11 years to 13 years), 91 seventh-graders (64 boys, 27 girls; medianage 13 years, ranging from 12 years to 14 years), and 48 eighth-graders (30 boys, 18 girls; median age 14 years, ranging from 12years to 15 years). Prior to the study, we obtained written parentalconsent for participation. As we focus on development of intergroupbias in majority children, the data of 45 further students with variousimmigration backgrounds (German was not their native languageand/or they were not born in Germany) were discarded from analyses.Data of another 7 participants had to be excluded because of individ-ual circumstances impeding data collection (i.e., arm fraction, visualimpairment, attention deficits). Each class received a reward of 50EUR (equivalent to U.S.$60 at the time of data collection) for com-munal use, regardless of the number of participating children. The

study was conducted in the computer classroom of the schools duringthe last 3 weeks of the school year prior to the summer vacation.

Materials.Priming measure. The priming task conformed to a 2 5

within-subjects design, made up by the factorial combination oftarget valence (positive vs. negative) and prime type (positive vs.negative vs. neutral vs. Turkish vs. German). For the prejudice-related primes, portrait pictures of 10 Turkish and 10 Germanyoung men were used. These pictures were selected from a largepool of 200 portrait pictures, after intensive pretesting with uni-versity students (see Degner, Wentura, Gniewosz, & Noack, 2007,for a description of pretests). A further pretest was conducted inwhich 53 children from two fifth-grade (n 27) and seventh-grade classes (n 25) had to rapidly categorize the depicted menas Turkish or German (Cronbachs .83).1 Only pictures withfast responses (1000 ms) and a minimum of 90% correct re-sponses were included as primes in the current experiment. Viathese additional pretests, it was assured that the depicted individ-uals could easily and unambiguously be categorized as Turkishand German on first sight by the younger participants in this study.The Turkish and German prime sets were matched according theirintragroup typicality, their attractiveness, and the emotionality oftheir facial expressions, to avoid the possibility that variables otherthan ethnicity could account for differential priming effects. Allfaces were in frontal view against a white background. They werein black and white, approximately 75 mm high, and 50 mm wide.To obtain a reference priming effect with standardized stimuli,pictures from the International Affective Picture System (IAPS,Lang, Bradley, & Cuthbert, 2005) were included as additional primes,five with positive, negative, and neutral contents. Positive primeswere characterized by a mean valence of M 7.86 (SD 0.31) anda mean arousal value of M 4.95 (SD 0.23). Negative primes hada mean valence of M 2.91 (SD 0.53) and a mean arousal valueof M 5.60 (SD 1.03). Neutral primes had a mean valence of M5.02 (SD 0.11) and a mean arousal value of M 2.76 (SD 0.15).

The target set for the priming task consisted of 50 IAPS pictures,25 with a positive content and 25 with a negative content. Positivetargets were characterized by a mean valence of M 7.64 (SD 0.35) and a mean arousal value of M 4.97 (SD 1.04). Negativetargets had a mean valence of M 3.09 (SD 0.74) and a meanarousal value of M 5.24 (SD 0.86). To assure that positive andnegative pictures chosen from the IAPS were distinctly and unan-imously positive or negative for children of the respective agegroups, a speeded valence classification task had been included inthe aforementioned pretest with children. Only pictures with highagreement (90%) in this task were chosen as stimuli for thecurrent study.

Explicit measures. To assess participants open expression ofprejudices toward Turks living in Germany, items of the BlatantPrejudice Scale (Pettigrew & Meertens, 1995) were reformulatedfor children (e.g., I would not mind if a Turkish person wasappointed as my teacher as long as he is giving good classes; seeDegner et al., 2007). A scale to assess general xenophobic attitudes(e.g., Foreigners take away our jobs; see Balke, El-Menouar, &Rastetter, 2002) was also included. All items were answered on a

1 Cronbachs alpha () refers to the reliability of the picture scores bytaking the raters as items.


6-point Likert scale ranging from 3 (I do not agree at all) to3 (I totally agree). Internal consistencies of both questionnairescales were in the range of expectations, with Cronbachs s .74and .78 for the blatant prejudice scale (in Study 1 and 2) and s .83 and .74 for the antiforeigner scale (in Study 1 and 2, respec-tively). Interscale correlation was r .64, p .001 in Study 1 andr .24, p .001 in Study 2, indicating that the two scalesmeasured only partially overlapping concepts.

Procedure. Participants were tested in groups of 6 to 14students seated separately in a computer classroom of theirschools. They were informed that they would accomplish a seriesof tasks on the computer and fill out a questionnaire. All instruc-tions were given by one of two female experimenters using pre-sentation slides for illustration.

The affective priming task was introduced as a mixed test ofevaluation and memory abilities. Children were introduced to thecomplete task in three steps. In the first part, children were famil-iarized with the target evaluation task. They were informed thatpictures would be presented very briefly on the computer screen,and their task was to quickly categorize the pictures accordingtheir valence by means of a key press (with A negative and 5 onthe number pad positive). Participants completed two practiceblocks of 25 trials each. Each trial started with a fixation crosspresented for 317 ms on the screen. After a blank of 133 ms, thetarget pictures appeared and remained on screen until participantsresponded. To encourage fast responding, the picture remained onthe screen for 1750 ms maximally, after which it was replaced bya feedback zu langsam (too slow). Participants received instan-taneous accuracy feedback after each trial. Furthermore, partici-pants received a summarized feedback on their percentage ofcorrect responses and mean response time at the end of the practicephase. The experimenters explained the feedback and encouragedparticipants to ask any questions about the task.

In the second step, participants were familiarized with the mem-ory task. They watched a sequence of 10 pictures of young Turkishand German men, interspersed with 15 IAPS pictures with posi-tive, negative, and neutral contents. Participants were instructed tomemorize the person pictures. Participants received different in-structions for the memory task, depending on their assigned ex-perimental condition. Participants within a particular session re-ceived the same instructions in order to ensure their unawarenessof the experiments purpose. Participants assignments to the ses-sions were randomly defined. In the categorization condition,children were instructed to memorize the number of Turks andGermans for a later test. In the personalization condition, childrenwere instructed to memorize the individual faces. A memory testfollowed directly after the memorization phase. Participants in thecategorization condition reported the number of Turks and Ger-mans they remembered seeing. Participants in the personalizationcondition saw the earlier person pictures interspersed with newpictures, and pressed an old key when they had seen the picturebefore and a new key when the picture was novel.

In the third, experimental, phase of the affective priming task,participants were informed that the evaluation task and the mem-ory task they had practiced so far would now be combined. Theywould therefore always be presented with two pictures in shortsuccession: the first picture (if it was a person picture) being targetfor the memory task, the second being target of the evaluation task.The experimenters repeated the instructions for both tasks (see

above) and proceeded only when they were sure that all childrenhad fully understood all instructions. The experimental phaseconsisted of four blocks, each containing 50 trials. Each trialstarted with the presentation of the prime stimulus for 317 ms,replaced by a blank screen. The target stimulus followed after 133ms (stimulus onset asynchrony [SOA] 450 ms) and remaineduntil the participants responded or the response deadline hadelapsed.2 The intertrial interval was 1,000 ms. Within a block, eachprime was presented once in each target condition. During thewhole experiment, participants received instantaneous accuracyfeedback and summarized feedback at the end of each block (i.e.,mean response times and percentage of correct responses). Thepriming task took approximately 20 min. After a small break anda filler task (approximately 10 min), the experimenter handed outthe questionnaires and explained the item format and answeringscales with two neutral exemplar items (e.g., I prefer lemonadeover water.). All items of the questionnaire were then read aloudby the experimenter, and participants filled in their responses onindividual sheets. Participants then provided demographic data,followed by a full debriefing.

Results

Analyses are presented separately for questionnaire and primingdata.

Explicit measures. To analyze age effects, scale means wereincluded in a 2 (scale: blatant prejudice vs. xenophobia) 4(grade: 5 vs. 6 vs. 7 vs. 8) analysis of variance with repeatedmeasures on the first factor. A significant main effect of scale wasfound, F(1, 300) 54.67, p .001, p2 .154, the main effect ofgrade missed the conventional level of significance, F(3, 300) 2.47, p .06, p2 .024. Main effects were qualified by asignificant interaction, F(3, 300) 4.24, p .006, p2 .041.General xenophobic attitudes were somewhat higher than blatantprejudice against Turks and did not differ according to age groups,F(3, 300) 1.42, ns (see Table 1). On the contrary, blatantprejudice against Turks depended on participants age, F(3,300) 4.03, p .008, p2 .039; linear trend F(1, 300) 9.71,p .002.3 Older participants expressed lower levels of explicitprejudice, as compared to younger participants. Participants gen-der did not affect prejudice expression (F 1, ns).

Priming effects. Response time (RT) data were trimmed tocorrect for anticipatory responses and momentary inattention. Tri-als with outlying response latencies slower than 1.5 interquartileranges above the third quartile of the individual RT distribution(see Tukey, 1977) or below 300 ms were excluded from analyses(6.8%). Analyses were run on response latencies of trials withcorrect responses only. The mean error rate was 9.54% (SD

2 Because of different settings of computer screens, timing of stimuluspresentation was not exactly the same for all participants. Specifically, theparameters described above applied to the majority of the 198 participants(60 Hz screens). For another 86 participants, the primes were presented for253 ms (75 Hz screens; SOA 470 ms, response deadline 1,400 ms).For another 20 participants, prime presentation lasted for 365 ms (52 Hzscreens; SOA 540 ms, response deadline 2,020 ms). However, timingof stimulus presentation did not affect results.

3 For all trend analyses throughout the article, quadrate and cubic trendswere nonsignificant and are therefore not reported in detail.


8.01); no systematic influences of prime presentation on error rateswere found (Fs 1.40, ps .24).

Priming indices were calculated by subtracting the mean re-sponse latencies in congruent trials from incongruent trials, thusrepresenting the crucial interaction of prime valence and targetevaluation. Accordingly, priming indices for the standard primesrepresent the difference of mean RTs for trials with the samevalence of prime and target (i.e., positivepositive, negativenegative), as compared to mean RTs of trials with primes andtargets of different valence (positivenegative, negativepositive).The expected response facilitation in congruent trials, as comparedto incongruent trials, would thus be indicated by a positive primingscore. Likewise, priming indices were computed for the attitude-relevant primes, taking the Germanpositive and Turkishnegativepairings as congruent combinations. Thus, positive priming scoresrepresent higher prejudice in terms of relative negativity towardsTurks compared to Germans. Priming effects for standard andattitude-related primes were analyzed separately.

In the standard priming condition (i.e., IAPS pictures as primes),we found a significant priming effect (M 13 ms, SD 36),t(303) 6.21, p .001, d 0.36, indicating that participantsreacted faster in trials with a valence-congruent prime-target com-bination as compared to incongruent pairings. Age did not affectthe priming score, F(3, 300) 0.56, ns; significant priming effectsof comparable size were found in all age groups (ts 2.30, ps .03, 0.31 d 0.43). The standard priming effect was affectedneither by participants gender nor by type of memory task (Fs 1, ns). This pattern of results indicates that the priming paradigmyielded the expected effects: Prime valence systematically influ-enced target evaluations, regardless of age or sex of participantsand regardless of memory task. It can therefore be concluded thatthe priming task was sensitive for effects of automatic valenceactivation in this sample of children and adolescents.

For the prejudice-related prime pictures, a significant effect ofgrade occurred, F(3, 300) 3.02, p .03, p2 .03; F(1, 300) 5.47, p .02, for the linear trend. As can be seen in Figure 1,participants in the younger grades did not show significant primingeffects (M5th 2 ms, SD 36, t[80] 0.48, p .63; M6th 4 ms, SD 25, t[83] 1.61, p .11), whereas participants in thehigher grades displayed significant priming effects (M7th 9 ms,SD 30, t[90] 2.93, p .004, d 0.30; M8th 9 ms, SD

20, t[47] 3.32, p .002, d 0.48). Thus, whereas the reactionsof the younger children were not differentially affected by thepresentation of Turkish versus German primes, a systematic prim-ing effect could be found for the adolescents. Planned contrasts(Helmert) were calculated to test for the occurrence of significantdifferences of priming effects between age groups. Therefore, thepriming score of each age group (starting with the eighth grade)was tested for difference from priming scores of the younger agegroups, respectively. This contrast was significant for the seventhgrade, F(1, 300) 4.40, p .04, but not for the sixth and eighthgrades (Fs 1.95, ps .16). These results indicate a develop-mental shift between sixth and seventh grade. The prejudice-related priming effect was affected neither by participants gendernor by type of memory task (Fs 1.36, ns).

To explore the extent to which the age effect upon prejudice-related priming resulted from automatic ingroup preference versusout-group derogation, we computed separate priming indices forTurkish and German primes. Therefore, mean response latencies intrials with positive targets were subtracted from trials with nega-tive targets. Note that the absolute value of these indices cannot bestraightforwardly interpreted as direct indices of positive versusnegative attitudes because they are distorted by main effects of

Table 1Explicit Prejudice Measures in Studies 13

Measure

5th grade 6th grade 7th grade 8th grade 9th grade

M SD M SD M SD M SD M SD

Study 1Blatant prejudice .74 3.17 1.11 3.04 1.04 2.72 0.96 2.66 1.04General xenophobia .83 3.24 1.02 3.46 0.94 3.19 1.04 3.17 0.86

Study 2Blatant prejudice .78 3.07 0.67 2.99 0.51 3.00 0.44 2.98 0.42General xenophobia .74 3.22 0.92 3.23 0.94 3.09 1.04 2.98 0.91Rating Turks .48 3.31 1.14 3.67 1.31 3.56 1.03 3.85 1.07Rating Germans .78 4.73 0.88 4.67 1.17 4.74 0.79 4.76 0.69

Study 3Rating Turks .92 3.26 1.46 3.39 1.43 3.78 1.19 3.09 1.04 3.32 1.15Rating Germans .81 4.48 0.99 4.64 0.78 4.63 0.68 4.74 0.91 4.47 0.90

-10

-5

0

5

10

15

5 6 7 8

Grade

Aff

ectiv

e pr

imin

g ef

fect

(in

ms)

Figure 1. Age effects on the prejudice-related priming effects in Study 1(error bars represent standard errors).


target valence (e.g., reactions to negative targets are typicallyslower as compared to positive targets, e.g., Wentura & Degner, inpress). Still, these difference scores can be used to explore differ-ences in age trends for Turkish and German primes.4 The primingindex for German primes was unrelated to participants age (F 1, ns), indicating that automatic evaluations of German pictureswere similar for participants of all age groups (M 29 ms, SD 50). On the contrary, the priming indices for Turkish pictures weresignificantly related to age, F(3, 300) 4.26, p .006, p2 .041,indicating that evaluative reactions to Turkish primes caused theage effect in priming indices (M5th 36 ms, SD 64; M6th 18ms, SD 39; M7th 10 ms, SD 54; M8th 12 ms, SD 39).

As an estimate of internal consistency of the priming measure,we calculated split-half partial correlations for the Turkish primingindex, controlled for the German priming index for 30 independentrandom splits of the trial sequence. The median Spearman-Brown-adjusted reliability score was r .31, replicating findings on lowinternal consistency of the affective priming paradigm (e.g., Bos-son, Swann & Pennebaker, 2000; De Houwer, 2003; but seeCunningham, Preacher & Banaji, 2001). There were no significantcorrelations of composite or separate priming scores with self-report measures for either blatant prejudice or general xenophobia(.01 rs .06, ns).

Memory performance. As already noted in the introductionto this study and as evident from the null results concerning thisfactor, our manipulation of prime categorization versus personal-ization was supposedly too weak to influence priming processes inchildren. Nevertheless, we briefly summarize basic results of thememory tests for the sake of completeness.

The memory test for participants in the categorization conditionconsisted of an estimation of the number of Turkish and Germanpictures that they had seen during the priming task. Overall,participants tended to slightly underestimate the percentage ofTurkish primes in comparison to German primes (M 46.23%SD 9.22), t(139)4.85, p .001, for the deviance from 50%.Category estimates were not affected by participants age, F(3,136) 1.85, p .14, and were unrelated to affective primingeffects (r .02, ns).

The memory test for the personalization condition asked forparticipants judgments of whether or not various portraits ofTurks and Germans had been presented during the previous prim-ing task. As an indicator of memory performance, the nonpara-metric signal detection sensitivity index A (Pollack, 1970) wascomputed for Turkish and German primes separately. Values weretested for a positive deviation from .50, indicating better-than-chance performance in recognizing the prime pictures. Memoryperformance was not significantly different for German and Turk-ish primes, F(1, 160) 2.38, ns. Not surprisingly, memory per-formance was significantly related to participants age, F(3,160) 7.00, p .001, but no interaction between prime-type(Turkish vs. German) and age occurred, F(3, 160) 1.03, ns.Memory performance of fifth-graders did not differ from chance,(M .53, SD .12), t(42) 1.66, p .11. Sixth- and seventh-graders performed slightly better than chance (M6th .56, SD .14), t(43) 2.61, p .01; (M7th .57, SD .15), t(46) 3.19,p .003. Eighth-graders performance clearly exceeded thechance level (M .67, SD .11), t(29) 8.32, p .001. Again,memory performance scores were not related to affective priming

effects (r .09, ns and r .04, ns; r .04 for their differencescore) and thus did not mediate the age effects on priming scores.

DiscussionIn this study, we showed that an affective priming task is

generally applicable as an indirect measure of childrens andadolescents automatic activations of evaluations. To the best ofour knowledge, this is the first time that an affective priming effectwas shown in participants younger than the typical college studentsamples. The significant standard priming effect found with nor-mative positive and negative primes demonstrates that the tech-nique is sensitive to effects of automatic valence activation in thestudied age groups. Independent of participants age, responses tovalenced stimuli were facilitated or inhibited as a function of thevalence congruence or incongruence of the preceding primes.

The implication of this standard effect is that when pictures ofGermans and Turks are used as primes, their facilitative or inhib-itory effects on the target categorization allow for an estimation ofautomatically activated evaluations of ingroup and outgroup mem-bers. This prejudice-related priming effect was significantly re-lated to participants age. Younger participantsattending fifthand sixth gradeshowed no differences in spontaneous evalua-tions of Turkish and German prime pictures. Older participantsattending seventh and eighth gradeyielded significant primingeffects indicating the presence of automatically activated evalua-tions and prejudice.

The age effects found for automatic prejudice activation stand insharp contrast to open prejudice expression. The overall level ofopen prejudice was rather low, and expression of negativity tendedto decrease among participants as they increased in age. It thusseems that developmental characteristics differ for automatic prej-udice activation and explicit prejudice expression.

It is interesting to note that priming effects were not influencedby the manipulation of prime construal, as introduced by thedifferent memory tasks. The results of the memory test reveal thatoverall memory performance was generally rather poor. A veryplausible explanation (supported by the experimenters observa-tions during data collection) is that children abandoned the mem-ory task while focusing on the target evaluations because theyreceived immediate accuracy feedback only on the latter. Thiswould also explain why we found no effect of categorizationversus personalization in the priming task. We return to this issuewith Study 4. Still, even if widely ignored, the memory tasks mighthave partially shifted participants attention toward the primes.One might even argue that the attention spent on primes cansomehow account for the found age differences in the primingeffects: Whereas young participants might naively not have con-sidered prejudice in this task, older participants might have be-come aware of the prejudice-relatedness of the primes, whichprompted them to activate evaluations of the primes. Although webelieve this to be rather unlikely, it seems to be worth exploringwhether priming effectsand the respective age differencesalso

4 Of course, these analyses have to be interpreted with caution. If, forexample, both indices would have yield a comparable age trend, the moststraightforward interpretation would have been to assume an age trend inthe difference between responses to negative versus positive target stimuli,irrespective of priming effects.


occur in children when they devote no attention to the primes.Thus, to complement the results of Study 1, we conducted a secondstudy using the affective priming task with masked prime presen-tation and a shorter SOA. Masking primes excludes participantsconscious awareness of the priming event (i.e., subliminal prim-ing); thus, demand effects or any sort of intentional activation ofprejudiced evaluations can be precluded. Finding priming effectsunder such conditions would allow for the straightforward inter-pretation that evaluations being indirectly measured with the af-fective priming are truly automatic, in the sense of being uninten-tional and unconscious. Although we found masked effects ofautomatic evaluation in former studies with adults and adolescents(e.g., Degner & Wentura, 2009; Degner et al., 2007; Otten &Wentura, 1999; Wentura, Kulfanek, & Greve, 2005), it is impor-tant to replicate masked standard affective priming effects withchildren.

Study 2

In Study 2, we conducted a modified version of the affectivepriming task, using masked prime presentation. The experimentwas conducted with children of the same age groups as in Study 1.


8 from a public high school that is different from the high schoolsin Study 1 (situated in the same region). Participants were com-prised of 68 fifth-graders (34 boys, 34 girls; median age 10years, ranging from 9 years to 11 years), 53 sixth-graders (26 boys,27 girls; median age 11 years, ranging from 11 years to 12years), 82 seventh-graders (34 boys, 48 girls; median age 12years, ranging from 11 years to 13 years), and 61 eighth-graders(25 boys, 36 girls; median age 13 years, ranging from 12 yearsto 14 years). The data of 13 additional participants with immigra-tion backgrounds were discarded from analyses. Data of 7 furtherparticipants were incomplete due to computer failure, and data ofanother 5 participants had to be excluded because of individualcircumstances that impeded data collection. The study was con-ducted at the beginning of the school year, which explains whystudents of the same grades were approximately 1 year younger inStudy 2 than in Study 1.

Materials and procedure. We used the same questionnairesand priming stimuli as in Study 1. The procedure was similar toStudy 1, with the following exceptions: The affective priming taskwas introduced as a test of evaluation and categorization abilities.That is, there was no memorization task, and the occurrence ofprimes was not revealed to participants. Prime presentation wasreduced to 30 ms. To assure participants unawareness of primepresentation, the prime was embedded in a forward mask, pre-sented for 70 ms, and a backward mask, presented for 10 ms beforereplaced by the target; thus the SOA amounted to 40 ms. Thesubjective impression of this presentation sequence was a briefflicker. Because it is known from masked affective priming studiesthat effects are more robust if participants are forced to respondfaster as needed to maintain the typically high accuracy (see, e.g.,Draine & Greenwald, 1998), we employed a response-deadlinetechnique. That is, the target remained on screen for only 500 ms,and children were instructed to press the correct key as fast as

possible within the span of target presentation. If they did notrespond within this time span, the target was replaced by a bluesign, which signaled that the response was too slow. Such modifica-tions largely remove any variance of interest from latency data.Therefore, in accordance with other studies (e.g., Degner et al., 2007;Draine & Greenwald 1998), the error rate was the dependent variableof interest in Study 2.

After the priming task, additional ratings of liking were obtainedfor all Turkish and German prime pictures. To acquire these ratings,the pictures were presented on screen in a random order, and partic-ipants indicated their liking for each portrayed person on a Likertscale ranging from 1 (do not like at all) to 7 (like very much). Childrenwere then interviewed about prime awareness in a funneled debriefingprocedure. Therefore, children were first informed about the hiddenpresentation of primes and then asked to indicate whether they hadnoticed the priming during the priming task (yes vs. no), whether theyhad perceived the hidden pictures (yes vs. no), and if yes, whetherthey could list the content of the pictures they had recognized. Sub-sequently, a direct test of prime recognition was administered. Par-ticipants worked through 30 priming trials with the standard positiveand negative primes and 30 priming trials with the Turkish andGerman primes. They were instructed to identify the prime picturesand to categorize them as being either positive or negative and eitherTurkish or German. To assure comparable masking without showingthe valence-connoted targets as during the priming task, a neutralIAPS picture was presented as target stimulus in each test trial.Thereafter, questionnaires were completed after a filler task as inStudy 1.

Results

Explicit measures.Questionnaires. No significant effects of age on open prejudice

expression in the questionnaires were found (Fs 1.18, ns; see Table1). Contrary to Study 1, students of all grades expressed similar lowlevels of blatant prejudice towards Turks (M 3.01, SD 0.52) andgeneral xenophobic attitudes (M 3.12, SD 0.96).

Stimulus ratings. The pleasantness ratings of the portrait pic-tures were submitted to a 2 (prime: Turkish vs. German) 4(grade) analysis of variance. We found a significant main effect ofprime origin on the ratings, F(1, 260) 212.39, p .001, p2 .452, indicating that German stimuli (M 4.73, SD .88) wererated significantly higher in pleasantness than were Turkish stimuli(M 3.59, SD 1.14). This main effect was qualified by amarginally significant interaction effect with age, F(3, 260) 2.17, p .09, p2 .024. Whereas ratings of German stimuli wereequally positive in all age groups (F 1, ns), ratings of Turkishstimuli were positively related to age, such that older participantsrated them as being less negative than did younger participants,F(3, 260) 2.61, p .05, p2 .028 (see Table 1). The magnitudeof rating differences between German and Turkish stimuli wassignificantly related to open prejudice expressions in the question-naires, with r .16, p .001 for the blatant scale and r .32, p.001 for the xenophobia scale.

Priming effects. Analyses of priming effects were based onerror rates throughout the experimental phase (200 trials). Meanerror rate was 23.75% (SD 9.60%). The mean response latencywas 484 ms (SD 58). In correspondence to the response deadlinetechnique, no systematic influences of prime presentation on re-


sponse latencies were found (Fs 2, ps .15). Priming indiceswere calculated by subtracting the mean error rates in congruenttrials from incongruent trials, thus representing the crucial inter-action of prime valence and target evaluation (see Study 1). Asignificant priming effect was found in the standard priming con-dition (M 2.82 %, SD 8.52), t(263) 5.37, p .001, d .33.More important, age did not affect the standard priming score, F(3,260) 0.19, ns. Significant priming effects of comparable sizewere found in all age groups (ts 2.20, ps .03; .30 d .34).Thus, prime valence systematically influenced target evaluations,regardless of participants age. It can therefore be concluded thatthe masked priming task was equally sensitive for effects ofautomatic valence activation in children and adolescents, as thepriming procedure used in Study 1.

In the prejudice condition with Turkish and German primes, asignificant effect of participants age occurred, F(3, 260) 2.62,p .05, p2 .029 (linear trend: F(1, 260) 4.44, p .03). Onlypriming effects of eighth-graders differed significantly from zero,t(60) 2.66, p .02, d .31; no significant prejudice-relatedpriming effects were found in any of the younger age groups, (ts1.17, ps .25; see Figure 2). Planned contrasts (Helmert) com-paring each age group with the respective younger age groupshowed only one significant difference between priming effects ofthe eighth-graders and mean priming effects of the younger par-ticipants, F(1, 260) 7.72, p .006, indicating a developmentalshift between seventh and eighth grade. On the contrary, primingeffects of seventh and sixth graders did not differ from the respec-tive younger participants (Fs 1, ns).

As in Study 1, separate priming indices were computed forGerman and Turkish primes, to explore whether the age effect wasdriven by reactions towards ingroup versus outgroup members. Asin Study 1, no significant effect of age was found for the Germanpriming index (F 1, ns; M 0.16, SD 15.64), whereas asignificant effect of age was found for the Turkish priming index(M5th 1.94%, SD 15.65; M6th 2.80%, SD 14.52; M7th 0.91%, SD 15.42; M8th 4.10%, SD 12.57), F(3, 260) 2.66, p .05, p2 .03.

Similar to Study 1, internal consistency of the Turkish primingindex (controlled for the German index, see Study 1 for procedural

details) was low, with a median Spearman-Brown-adjusted split-half correlation of r .31. As previously, priming effects wereunrelated to questionnaire scores (.02 rs .05, ns) and werealso unrelated to explicit stimulus ratings (.06 rs .06, ns).Again, these results indicate a dissociation of automatic valenceactivation and open attitude expression.

Prime awareness. None of the participants reported havingnoticed the hidden primes during the priming task. With the dataof the direct test, we computed the nonparametric signal detectionsensitivity index A for the categorizations of the masked primes.Indices were computed separately for the positive-versus-negativedetection task and the Turk-versus-German detection task. MeanA was M .51 (SD .16) for the positivenegative detection,t(263) 1.09, ns, and M .54 (SD .16) for the TurkishGerman detection, t(263) 3.67, p .001, with both testsconcerning the deviation from chance level .50. Thus, whereas thedetection of standard primes did not differ significantly fromchance, the direct test yielded an above-chance result for therecognition of Turkish and German primes. Detection of standardprimes was unrelated to age, F(3, 260) 0.28, ns, but performancein the TurkishGerman classification test increased with age, F(3,260) 2.83, p .04. Whereas fifth- to seventh-graders showeddetection indices that did not differ significantly from chance level(M5th .50, SD .16, M6th .53, SD .17, M7th .53, SD .17, ts 1.5, ns), eighth-graders showed somewhat better discrim-ination (M8th .58, SD .16, t[60] 4.03, p .001). Thus,although none of the participants reported conscious awareness ofprime presentation during the affective priming task, older partic-ipants were somewhat better able to discriminate German fromTurkish primes when specifically instructed to do so.

We took several measures to ensure that the priming effects ofeighth-graders cannot be attributed to a higher prime sensitivity, asindicated by the direct test. First of all, the prejudice-relatedpriming effect was not related to prime sensitivity in the direct test(r .05, ns), thus prime sensitivity cannot be a mediator of theage effect on priming. Correspondingly, in an analysis of covari-ance using the TurkishGerman prime sensitivity as a covariate,the age effect on priming remained significant, F(3, 260) 3.03,p .03, p2 .034, and no significant effect of prime sensitivityoccurred (F 1, ns). Second, to show that the priming effectfound for eighth-graders did not depend on those children withpartial awareness, we followed the suggestion by Draine andGreenwald (1998) to regress the (prejudice-related) affective prim-ing index on a transformation of A centered around zero (i.e., wesubtracted 0.5 from the original A for each participant). While thecentered A was associated with a nonsignificant regressionweight, t(59) 0.09, ns, the intercept of this regression deviatedsignificantly from zero with a regression weight of B 2.61,t(59) 2.13, p .04. Since both the priming index and the directmeasure have a natural zero point, we can interpret this effect as anabove-zero priming effect in the absence of objective prime aware-ness.

Discussion

Results of this study show that the affective priming task isapplicable for the assessment of spontaneous evaluations in chil-dren and adolescents, even when the primes are presented outsideconscious awareness. The significant standard priming effects with

-4

-2

0

2

4

5 6 7 8

Grade

Aff

ectiv

e pr

imin

g ef

fect

(in

erro

r %)

Figure 2. Age effects on the prejudice-related priming effects in Study 2(error bars represent standard errors). Note that the dependent variable isthe error rate due to use of the response-window procedure.


IAPS-pictures as primes demonstrate that this technique and thenonmasked priming procedure used in Study 1 are similarly sen-sitive to automatic evaluation.

Most important, we replicated the age influence upon theprejudice-related priming effect that had been observed in Study 1.Whereas adolescents attending eighth grade showed significantprejudice-related priming effects, no such effects occurred foryounger children in the fifth to seventh grades. Note that thesepriming effects occurred independently from participants primeawareness.

The age effects on open prejudice expression in the question-naires of Study 1 could not be replicated in Study 2. Participantsof all age groups expressed relatively low levels of open prejudiceagainst Turks or general xenophobia. The ratings of liking forprime portraits showed a comparable age effect, however: Partic-ipants of all age groups expressed levels of liking for the Germanpictures that were significantly higher than for Turkish pictures.This difference in liking for Turkish and German pictures de-creased with higher age. Open prejudice expression was againunrelated to affective priming effects, suggesting a dissociation ofthe developmental course of open prejudice expression and auto-matic activation of prejudice-related evaluations.

The age differences found in the affective priming effect forStudies 1 and 2 stand in contrast to results of previous IAT studies,in which no effects of age on automatic prejudice were found (e.g.,Baron & Banaji, 2006; Dunham, Baron, & Banaji, 2006; Rutlandet al., 2005). We assume that this contrast is related to the differentmeasurement methodsaffective priming versus IATand theway they assess prejudice-related evaluations. However, beforeelaborating on this assumption and its implications for develop-mental theories of prejudice automatization, at least one furtherstudy is needed that directly demonstrates that the age effects ofStudy 1 and 2 are related to the specific measurement procedure ofaffective priming. Thus, in Study 3, we aimed to rule out alterna-tive explanations, based on differences between previous IATstudies and the current studies related to the societal and culturalcontext of the studies. Previous IAT studies have been conductedcontrasting Black people and White people as target groups in theUnited States (Baron & Banaji, 2006) or the United Kingdom(Rutland et al., 2005). These groups differ from Turks in Germanyin many regards, for example in the degree of visibility or percep-tual salience compared to the majority ingroup and in historical aswell as contemporary characteristics of the specific intergrouprelations. If the age effects found in Study 1 and 2 were related tosuch differences, the same age effect should be found with anymeasure of automatic prejudice activation, whether it is an affec-tive priming task or an IAT. If the age effect is related to themeasurement procedure of affective priming; as we assume, itshould not be found with an IAT.

Study 3

In this study, we aimed to replicate the age effects on automaticprejudice activation, found with the affective priming tasks inStudies 1 and 2. Furthermore, we explored whether a similar ageeffect would be found in a prejudice IAT with pictures of Turksand Germans as target categories. Participants in Study 3 thereforecompleted both an affective priming task and an IAT, using the

same set of stimuli for the assessment of automatic prejudice.Furthermore, we expanded the age range to include ninth-graders.


9, drawn from a different public high school (situated in the sameregion). The sample was comprised of 43 fifth-graders (22 boys,21 girls; median age 10 years, ranging from 9 years to 11 years),31 sixth-graders (12 boys, 19 girls; median age 11 years,ranging from 11 years to 12 years), 26 seventh-graders (9 boys, 17girls; median age 12 years, ranging from 11 years to 14 years),16 eighth-graders (7 boys, 9 girls; median age 13 years, rangingfrom 13 years to 15 years) and 25 ninth-graders (13 boys, 12 girls;median age 14 years, ranging from 13 years to 15 years). Thedata of 29 additional participants with immigration backgroundswere discarded from analyses. Data of 5 more participants wereincomplete, due to computer failure. Furthermore, for 14 fifth-grade participants, no IAT were recorded due to computer mal-function. Similar to Study 2, data were collected at the beginningof the school year.

Materials and procedure. The affective priming taskcontained the same priming stimuli as in Study 1 and 2. For theIAT, the same 10 German and 10 Turkish prime pictures repre-sented the target categories, and a selection of 10 positive and 10negative target pictures from the priming task represented theso-called attribute dimension in the IAT.

Children always completed the affective priming task first,followed by the IAT. The affective priming procedure was iden-tical to that of Study 1, with the exceptions that (a) the SOA wasshortened to 317 ms, by immediately replacing the prime stimuluswith the target stimulus, and (b) no additional memory task in-structions were given. The primes were introduced as distracterpictures, and children were explicitly instructed to ignore them andfocus on the target evaluation task.

After the priming task, participants took a short break andsubsequently worked through the IAT. They were first familiarizedwith the classification task for Turkish and German pictures. Theywere told that the pictures that had served as distracters in theprecedent task were now task-relevant and would have to becategorized as depicting either Turkish or German young men. TheTurkishGerman categorization task was practiced in 20 trials.Response key assignment was counterbalanced between groups ofparticipants. As instructions were given verbally, children in oneexperimental session were always included in the same condition.No additional practice phase was administered for the positivenegative categorization task, as it was identical to the evaluationtask children had completed in the priming task. Instead, childrenwere informed that the two tasks were now mixed and that theyhad to categorize the colored pictures as being positive or negativeand the black and white portrait pictures as depicting either aTurkish or a German person. Participants then started immediatelywith the first mixed block of the IAT, in which target and attributestimuli were presented in a random, alternating order. Accordingto category-key assignment, approximately half of the participantsstarted the IAT with a congruent versus incongruent condition. Thefirst block consisted of 50 trials, of which the first 10 trials wereregarded as additional practice trials. Afterwards, participants werefamiliarized with the reversed key-assignment for the Turkish and


German trials and practiced it in another 20 practice trials. Theythen continued with the second mixed block, which again con-sisted of 10 additional practice trials and 40 experimental trials.Participants took another short break after completion of the IATand before the liking ratings were obtained for all Turkish orGerman portraits. Because of time constraints, no further explicitprejudice measures were obtained.

Results

Stimulus ratings. The pleasantness ratings of the portraitpictures of Turks and Germans were submitted to a 2 (primeorigin: Turkish vs. German) 5 (participants age: fifth vs. sixthvs. seventh vs. eighth vs. ninth grade) analysis of variance. Asignificant main effect of prime origin on the ratings was found,F(1, 136) 98.75, p .001, p2 .421, indicating that Germanstimuli (M 4.57, SD .87) were rated significantly higher inpleasantness than Turkish stimuli (M 3.38, SD 1.31). Noother effect gained significance, indicating that this bias wasunaffected by participants age (F 1, ns; see Table 1).

Priming effects. Priming effects were calculated as describedabove, based on trimmed response latencies of correct responsesthroughout the experimental phase (see Study 1). Mean error ratewas 7.41% (SD 5.60%).5

A significant priming effect was found in the standard primingcondition (M 18 ms, SD 29), t(141) 8.25, p .001, d 0.70. As in the previous studies, age did not affect the primingscore, F(4, 136) 1.56, p .19. Significant priming effects ofcomparable size were found in all age groups (ts 2.30, ps .03;0.42 ds 0.96).

A significant effect of participants age was found in the prejudicepriming condition with Turkish and German primes, F(4, 136) 2.50, p .04, p2 .07; F(1, 136) 6.77, p .01, for the lineartrend. In accordance with the previous studies, only priming effects ofeighth-graders (M 12 ms, SD 16), t(15) 3.07, p .008, d 0.74, and ninth-graders (M 11 ms, SD 16), t(24) 3.41, p .002, d 0.68, differed significantly from zero, whereas no signifi-cant prejudice-related priming effects were found for any of theyounger age groups (ts 1.66, ps .11, 0.01 ds 0.32; seeFigure 3). Planned contrasts between age groups (Helmert) were

computed as in the previous studies. In Study 3, however, the contrastindicating a developmental shift between eighth grade and theyounger grades missed significance, F(1, 136) 2.49, p .12.However, the difference between sixth- and fifth-graders was mar-ginally significant, F(1, 136) 2.84, p .09 (other Fs 1.63, ns).

Separate priming indices were computed for German and Turk-ish primes. Again, no significant effect of age was found for theGerman priming index (F 1, ns, M 31 ms, SD 38), whereasa significant effect of age was found for the Turkish priming index,F(4, 136) 3.68, p .007, p2 .098 (M5th 39 ms, SD 51;M6th 21 ms, SD 36; M7th 18 ms, SD 31; M8th 16ms, SD 28; M9th 3 ms, SD 30, respectively). As in theprevious studies, internal consistency of priming effects wasrather low with a median Spearman-Brown-adjusted split-halfcorrelation of r .38 (see Study 1 for procedural details).Separate priming effects, as well as composite scores, wereunrelated to stimulus ratings (.04 rs .07, ns).

IAT scores. For reasons of comparability, RT data in the IATwere trimmed according the same criteria used for the affectivepriming task, excluding 0.8% of the trials. Analyses were con-ducted only for trials with correct responses. The mean error ratewas 7.74% (SD 5.93). IAT scores were computed as the differ-ence between compatible and incompatible blocks (with compat-ibility defined from the perspective of positive ingroup and neg-ative outgroup evaluation). Additional analyses were conductedbased on the IAT scoring algorithm (D) recommended by Green-wald, Nosek, and Banaji (2003) that yielded the same pattern ofresults. Overall, a significant IAT effect was found (M 89 ms,SD 108), t(131) 10.15, p .001, d 0.88; D 0.44, SD 0.37, t(131) 13.45, p .001. The IAT score was submitted toa 2 (block sequence: compatible first vs. incompatible first) 5(participants age: fifth vs. sixth vs. seventh vs. eighth vs. ninthgrade) analysis of variance. This analysis revealed a significantmain effect of block order, F(1, 131) 6.15, p .01, p2 .05,indicating that IAT effects were significantly enlarged when thetask started with the compatible block, (M 107 ms, SD 119;D .53, SD .37), as compared to the incompatible block (M 65 ms, SD 85; D .32, SD .35). Most important, nosignificant main effect of age or interaction effect of age withblock sequence was found (Fs 1, ns, replicating results of otherstudies). Significant IAT scores occurred in all age groups withM5th 90 ms (SD 138), M6th 104 ms (SD 75), M7th 66ms (SD 86), M8th 104 ms (SD 144), M9th 84 ms (SD 91), ts 2.88, ps .01. The Spearman-Brown-adjusted reliabilityscore for the IAT (D score) was r .73. The IAT score wasunrelated to the prejudice-related affective priming score (r .09, ns). However, the IAT score was negatively related toexplicit evaluations of the Turkish stimuli (r.25, p .01), butno significant relation was found between the IAT score and themean ratings of German stimuli (r .004, ns). Thus, the higher theIAT score, the more negative the evaluations of the Turkishstimuli.

5 Contrary to Study 1, a significant priming effect was found in the errorrates for standard positive and negative primes (M 2.34, SD 5.94),t(140) 4.68, p .001, d .39, but not for the prejudice related primes(M 0.68, SD 4.71). Priming effects in error scores were unrelated toage (Fs 1, ns).

-10

-5

0

5

10

15

5 6 7 8 9

Grade

Aff

ectiv

e pr

imin

g ef

fect

(in

ms)

Figure 3. Age effects on the prejudice-related priming effects in Study 3(error bars represent standard errors).


For a direct comparison of age differences on prejudice effectsfound with the different measures, z-transformed IAT and primingscores were submitted to a 2 (measure: IAT vs. priming) 5(participants age) mixed model analysis of variance that revealeda significant interaction when testing for linear trends, F(1, 128)4.74, p .03. Thus, differences between the age effects on thedifferent measures can be interpreted as such.

Discussion

In Study 3, the age effect on prejudice-related priming found inthe previous studies was replicated. Again, significant primingeffects indicating relative negativity towards Turks as comparedwith Germans were only found in children older than 13 years ofage. Most important, no such age effect was found in an IAT withthe same stimuli. Rather, a significant prejudice effect in this taskwas found throughout the age range replicating IAT results foundin other prejudice (e.g., Baron & Banaji, 2006; Rutland et al.,2005). This pattern of results indicates that the age differencesfound in our studies are unlikely to be related to cultural differ-ences or peculiarities of the current samples. Rather, the contrastbetween results of our studies and previous studies on automaticprejudice should be explained in terms of the characteristics of thetwo measures, affective priming and IAT.

One of the most blatant differences between the two measures isthe salience and task-relevance of social categorizations: In theIAT, prejudice assessment is based on forced-choice categoriza-tions of target stimuli, and IAT scores thus represent evaluations ofa given category label. One cannot straightforwardly conclude thatthe categorizations predetermined in the IAT are actually the oneschildren spontaneously use in person perception. Accordingly, onecannot conclude that individual category members are spontane-ously evaluated the same way as the category labels (see DeHouwer, 2001; Olson & Fazio, 2003 for related arguments). In theaffective priming paradigm, on the contrary, category-based eval-uations play a subordinate role, and the measure is expected toassess spontaneous evaluations on an exemplar-based level (Olson& Fazio, 2003; Livingston & Brewer, 2002). We assume that thecontrast between stable category-based prejudice effects in theIAT and reliable age differences on exemplar-based prejudice inthe affective priming task reveals differences in the developmentalprocesses underlying prejudice activation, related to the degree ofspontaneous versus forced category use.

It is well demonstrated that children acquire awareness of racialand ethnic categories and their societal evaluations very early,between the ages of 3 years and 6 years (e.g., Aboud, 1988, 2008).These evaluations of social categories might become stably repre-sented in long-term memory and will be recalled when children areexplicitly prompted to categorize people according their socialgroup memberships, such as in the IAT (and any other forcedchoice measure). This, however, does not necessarily imply that inreal-life encounters children actually (a) do automatically catego-rize people according their ethnicity and (b) do automaticallyactivate the corresponding, category-based evaluations. Childrenmight be able to indicate whether a person is Black or White, orGerman or Turkish, and might also express societal intergroupbiases if explicitly asked to do so. However, such categorizationsand evaluations might not yet play a role in childrens spontaneousperson perceptions. For example, it has been repeatedly demon-

strated that young children are familiar with societal evaluations ofspecific social category labels without being able to correctlyassigning the labels to specific category members (e.g., Bar-Tal &Teichman, 2005; Nesdale, 2001). Thus, the finding of stable IATeffects in response to social category labels (Baron & Banaji,2006; Rutland et al., 2005) does not imply that children automat-ically activate negative evaluations in encounters with outgroupmembers whose category membership is not made salient. Further,the notion of unconditional automatic categorization has also beenquestioned for adults (e.g., Macrae, Quinn, Mason, & Quadflieg,2005). It thus makes sense to assume a developmental change fromearly category-based prejudice expression to later exemplar-basedprejudice activation: Whereas children are early familiarized withsocietal evaluations of social categories that they reproduce ifcategory membership is made salient and/or task relevant (asdocumented by the IAT), they only later acquire the necessaryexpertise for effortlessly applying social categories to individualsand thus show spontaneous activations of category-related evalu-ations only later in life (as documented in the affective primingparadigm). These assumptions overlap with theories on ethnicidentity development (e.g., Quintana, 1998) or theories of preju-dice development (e.g., Nesdale, 1999, 2001) that describe age-related increases in the use of categorical information in personperception and evaluation. Other developmental models, like thesocial cognitive developmental theory by Aboud (1988, 2008),assume a decrease in category reliance in children above the age of7 yearsrelated to their increasing cognitive abilities. The goal inStudy 4 was thus to test (a) whether the assumed age-relatedincrease in the spontaneous use of ethnicity can be found and (b)which role spontaneous versus forced categorizations play for theautomatic activation of intergroup prejudice in the same age range.

Study 4

To test these assumptions, a further study was conducted fol-lowing two complementary strategies. First, we aimed at demon-strating that the affective priming paradigm yields the same resultswith children and adolescents as the IAT if social categorizationsare made similarly task relevant as in the IAT. In Study 1, we hadmade an unsuccessful attempt to test for category based vs. exem-plar based prejudice activation within the affective priming task,following Olson and Fazio (2003). However, this manipulationlinked to an additional prime memorization taskseems to havebeen too subtle to actually alter childrens prime processing be-cause they tended to ignore or forget this additional task. Wetherefore used a stronger manipulation in Study 4 in which socialcategorizations were made task relevant. Children had to indicatethe social category membership of prime persons (Dutch vs.Moroccan) immediately after each target reaction (forced catego-rization condition). With this task, prime salience was kept con-stantly high and prime categorization was task relevant as in atypical IAT. The remaining participants completed a standardaffective priming procedure as described in Study 3 (no categori-zation condition). We expected affective priming effects in theforced categorization condition to be similar to IAT effects ofStudy 3, thus showing a stable prejudice-related priming effectindependent of participants age. In the no categorization condi-tion, the typical age differences found in Studies 13 should bereplicated.


Second, we aimed at testing directly whether age-related differ-ences in childrens spontaneous use of social categories in personperception could be found, implementing a measure of spontane-ous category use, the who-said-what paradigm (Taylor, Fiske,Etcoff, & Ruderman, 1978, see Bennett & Sani, 2003, for use ofthe paradigm with children). In this paradigm, children are showna series of individuals making a number of statements (e.g., In theafternoon, I like to play soccer with my friends.). Each individual isidentified by a photograph. In our adaption, the target individualsorthogonally varied in terms of age (child vs. adult) and ethnicity(White Dutch vs. Moroccan). Most important, neither age nor ethnic-ity was explicitly mentioned throughout the task, and the content ofthe statements was unrelated to social category memberships. Chil-dren were later asked in a surprise recall test to remember thestatements and the persons who have made them. The dependentvariable of interest in this paradigm is the number and distribution ofidentification errors, thus, cases in which statements are assigned tothe wrong person: If within-category confusions occur more fre-quently than between category confusions, it can be inferred that thecategory membership was attended to, was encoded in memory, andwas subsequently used in the memory task. Thus, this paradigmallows indirect assessment of spontaneous use of social categories andtesting of whether age-related increases in spontaneous category usemediates age effects on prejudice-related priming scores. Further-more, by orthogonally manipulating two social categories (i.e., agerand ethnicity) the paradigm allows estimation of age effects in therelative salience of these two categories.6

For the current research, we adjusted the classical procedureaccording to suggestions by Klauer and Wegener (1998). In theclassical paradigm, participants do not have the option to state thatthey do not remember the statement and have to guess the speaker.This forced guessing invites possible confoundsespecially be-cause one can assume that memory performance differs withparticipants age, and younger participants are less likely to re-member a statement than are older children. For example, ifguessing is independent of category membership, poor statementmemory and frequent guessing would level out differences be-tween within-category and between-category errors, and thus thespontaneous use of category information would be underestimatedfor (younger) children with poor statement performance. For thisreason, we adjusted the memory test by interspersing new state-ments. Children indicated first whether they remembered a state-ment as having been presented before (old) or not (new). Only forold recognized statements were they prompted to indicate who hadgiven it (see Klauer & Wegener, 1998).

Study 4 was conducted in the Netherlands, contrasting attitudestowards the White Dutch majority ingroup and the Moroccanimmigrant outgroup. Thus, the study took place in a differentsociocultural and intergroup context and therefore also exploredthe generalizability of results of the first three studies to anintergroup context outside Germany.

MethodParticipants. Participants were 129 students from Grades 5 to 8

of a public Dutch elementary school, situated in a small-sized town ina central Netherlands region with slightly lower percentages of (non-Western) immigrants than in the previous studies (approximately5%). Approximately 2% of the school populations were from a visible

ethnic minority; all teachers were White Dutch. The sample is dividedinto 29 fifth-graders (15 boys, 13 girls, 1 not reported; median age 9 years), 30 sixth-graders (18 boys, 12 girls; median age 10 years),36 seventh-graders (19 boys, 17 girls; median age 11 years), and34 eighth-graders (15 boys, 18 girls, 1 not reported; median age 12 years). Prior to the study, we obtained parental consent for thechilds participation. Incomplete data of 26 further participants(due to computer malfunctions or individual circumstances imped-ing data collection) as well as data of 5 further students withimmigration backgrounds were discarded from analyses. The studywas conducted in the computer classroom of the schools at the endof the school year, prior to the summer vacation. In the Nether-lands, mandatory education starts at the age of 5 years, whichexplains why students of the same grades were 1 year younger,compared to the German students in the previous studies who startschool at the age of 6 years.

Materials.Priming measure. The priming task conformed to a 2 4

within-subjects design, resulting from the factorial combination oftarget valence (positive vs. negative) and prime type (positive vs.negative vs. Moroccan vs. Dutch). A new set of portraits of 20young men of White Dutch origin versus Moroccan origin wasselected from a database of 280 portrait pictures after extensivepretesting with Dutch students. The selected Dutch and Moroccanpictures were matched according to perceived ingroup typicality,age-estimates, attractiveness, and emotionality of facial expres-sion. Contrary to the previous studies, colored pictures were usedwhich slightly increases perceptual salience of group membership.All faces were in frontal view against a grey background and wereapproximately 75 mm high and 50 mm wide. The IAPS picturesused in previous studies were included as standard primes andtargets. Contrary to the previous studies, pictures depicting personswere not used as targets (to avoid that children mistakenly applythe forced categorization task to the target instead of the primes).

Who-said-what paradigm. The who-said-what paradigm con-formed to a 2 2 within-subjects design, resulting from thefactorial combination of target age (child vs. adult) and targetethnicity (Moroccan vs. Dutch). Therefore, portraits of four adultsand four children of White Dutch origin versus Moroccan originwere selected from the same database according the same match-ing criteria (see above). For the memory task, two sets of 24statements were created that describe a persons preferences forfood, activities, or pets (e.g., In the afternoon, I like to play soccerwith my friends.). One set was assigned to be old, to be presentedduring the learning phase; the other was assigned to be new, to bepresented as filler items in the memory test.

Procedure. Participants were tested in groups of 6 to 8 stu-dents seated separately in computer classrooms of their school. Allinstructions were given by one of two White Dutch female exper-

6 Applying this paradigm to children, Bennett and Sani (2003) indeedfound significantly more within- than between-race confusions, with noage differences between 5-, 8-, and 11-year old children, which contradictsour hypothesis. However, the age range in this study does not includeyoung adolescents, and it is difficult to draw conclusions about higher agegroups. Furthermore, this study involved stimulus targets who only dif-fered according to one category (i.e., ethnicity), limiting the generalizabil-ity of results to naturally occurring contexts in which people vary in termsof multiple categorizations (e.g., age, ethnicity, gender).


imenters. The experimental session always started with the who-said-what paradigm. Children were informed that they would getto know a group of eight people who would tell them about theirfavorites and that their task was to form an impression of thesepeople. Presentation followed a fixed random order in which eachstatement was presented together with a picture for 8 seconds.Children were instructed to read the statement and look at theperson. In the fifth and sixth grade, the children were instructed toread the statements aloud together with the experimenter. Afterviewing the eight individuals, each with three statements, childrenwere introduced to the surprise recall task in which they had toremember the statements and the target persons who had madethem. Therefore, the original 24 statements were interspersed with24 new statements. Each statement was presented one by one onthe computer screen, and children first indicated whether they hadpreviously encountered the statement or not via mouse click on anold or a new button. If the statement was classified as old, thepictures of all eight target persons were subsequently presented onscreen (in one fixed, random order), and participants chose theperson that they thought had made the statement previously. If theyclassified a statement as new, the task continued directly with thenext statement. Completion of the who-said-what paradigm lastedabout 5 min.

Before the affective priming procedure started, all childrencompleted two practice blocks in which each prime picture had tobe categorized according social group membership.7 To keep theprocedure identical with the later priming trials, each picture waspreceded by a fixation cross presented for 1,000 ms. The pictureitself was presented for 316 ms and then replaced by a blank.Finally, with a delay of 1,200 ms, a probe question appeared,asking whether the previous picture depicted a Dutch person orwhether it depicted a Moroccan person (i.e., Was dat eenNEDERLANDER? or Was dat een MAROKKAAN?). Chil-dren responded by pressing a right key M for yes and a left keyC for no. Key-response reminders were presented in the lowerleft and right corners of the screen. They received instantaneousperformance feedback, with the messages JUIST! (correct) orFOUT! (incorrect) appearing on screen for

Documents

Automatic Prejudice in Childhood and Early Adolescence