Developmental Science. 2017; 00; e12527. wileyonlinelibrary.com/journal/desc  | 1 of 12https://doi.org/10.1111/desc.12527

© 2017 John Wiley & Sons Ltd

Received:30June2015  |  Accepted:6October2016DOI:10.1111/desc.12527

P A P E R

Children’s selective trust decisions: rational competence and limiting performance factors

Jonas Hermes1,2 | Tanya Behne1,2 | Anna Elisa Bich1 | Christa Thielert1 |  Hannes Rakoczy1,2

1DepartmentofDevelopmentalPsychology,InstituteofPsychology,UniversityofGöttingen,Germany2LeibnizScienceCampusPrimateCognition,Göttingen,Germany

CorrespondenceJonasHermes,UniversityofGöttingen,InstituteofPsychology,DepartmentofDevelopmentalPsychology,Waldweg26, D-37073Göttingen,Germany.Email:jonas.hermes@psych.uni-goettingen.de

Funding informationDeutscheForschungsgemeinschaft,Grant/AwardNumber:Ra-2155/1-1

AbstractRecentresearchhasamplydocumentedthatevenpreschoolerslearnselectivelyfromothers,preferring,forexample,reliableoverunreliableandcompetentoverincompe-tentmodels.Itremainsunclear,however,whatthecognitivefoundationsofsuchse-lective learningare, inparticular,whether itbuildsonrational inferencesoron lesssophisticatedprocesses.Thecurrentstudy,therefore,wasdesignedtotestdirectlythepossibilitythatchildrenareinprinciplecapableofselectivelearningbasedonra-tionalinference,yetreverttosimplerstrategiessuchasglobalimpressionformationundercertaincircumstances.Preschoolers(N = 75)wereshownpairsofmodelsthateitherdifferedintheirdegreeofcompetencewithinonedomain(strongvs.weakorknowledgeablevs.ignorant)orwerebothhighlycompetent,butindifferentdomains(e.g.,strongvs.knowledgeablemodel).Inthetesttrials,childrenchosebetweenthemodelsforstrength-orknowledge-relatedtasks.Theresultssuggestthat,infact,chil-drenarecapableofrationalinference-basedselectivetrust:whenbothmodelswerehighlycompetent,childrenpreferredthemodelwiththecompetencemostpredictiveandrelevantforagiventask.However,whenchoosingbetweentwomodelsthatdif-feredincompetenceononedimension,childrenrevertedtohalo-stylewidegenerali-zationsandpreferredthecompetentmodels forbothrelevantand irrelevanttasks.Thesefindingssuggestthattherationalstrategiesforselectivelearning,thatchildrenmasterinprinciple,cangetmaskedbyvariousperformancefactors.

RESEARCH HIGHLIGHTS

• Youngchildren’sselectivetrustdependsontheirtraitknowledge.• Preschoolers generalize widely to unconnected domains whenchoosingbetweenmodelsthatdifferinoverallcompetence.

• Preschoolerschooserationallybetweentwomodelsthatarehighlycompetent,yetindifferentdomains,butonlyiftheyhavecorrectlyidentifiedthemodels’traits.

• Youngchildrenarecapableofrationalselectivetrustdecisionsbutfallbackonglobalimpressionformationwhenthisprovidesdefinitesolutions.

1  | INTRODUCTION

When acquiring knowledge about the world, young children learnmuch,ifnotmost,throughthetestimonyofothers.Agrowingbodyof recent research has shown that even preschoolers are selectivein whom they learn from rather than trusting anyone indifferently(Clément,2010;Harris,2007;Mills,2013;Robinson&Einav,2014).For example, they prefer to learn from previously accurate label-ers (Koenig,Clement,&Harris, 2004), fromnice (Landrum,Mills,&Johnston, 2013) or attractivemodels (Bascandziev&Harris, 2014),frommodelswhoarefamiliar(Corriveau&Harris,2009)orspeakwith

2 of 12  |     HERMES Et al.

thesameaccentastheparticipant(Kinzler,Corriveau,&Harris,2011)andfrommodelswhoexpresscertainty(Sabbagh&Baldwin,2001).

1.1 | Cognitive foundations of selective social learning

Butwhatcognitivefoundationsissuchselectivesociallearningbasedupon?Does it reflect adult-like rational inductive learning? That is,do children infer from a model’s past behavior certain competen-cies/traits (e.g., thatsomeonewhoprovidedaccurate information isgenerally knowledgeable) thatmake themodel a suitable sourceofinformation for certain future tasks? Alternatively, children’s selec-tivetrustmightnotbebasedonsuchsophisticatedreasoningabouttraitsbutratheronlessdifferentiatedglobalimpressionformation:Ifmodelsdifferwith regard toacertaincompetenceor trait, childrenmightperceivemodelsas somehowpositiveornegative inaglobalsenseandchoosethemorepositiveoneforawiderangeof futureproblems– includingthoseunrelatedtotheoriginalcompetenceortraitdisplayed.

The empirical findings so far cannot resolve this question giventhesomewhatmixedfindingsandthedifferentstructureofproblemschildrenwere confrontedwith. In one set of studies, childrenwereaskedtochoosebetweentwomodelswithsimilardegreesofcompe-tenceyetindifferentdomains.Thesestudiessuggestthatchildrencandifferentiatebetweenmultipleepistemic cues in rationalways: thatis, they infer specific model characteristics and competencies fromobservedbehaviorandconsider these in theirmodelchoice in lightofthetaskrequirements.Forexample,whenpresentedwithexpertsfromdifferentdomainsofcompetence(e.g., toy labelervs. toyfixer)children competently attributed knowledge to these experts whenseekinginformationintherespectivedomains(Kushnir,Vredenburgh,&Schneider,2013;Lutz&Keil,2002;Sobel&Corriveau,2010).Andchildrenunderstandthatadultsandchildrenmaybeexpertsindiffer-entfields,trustingpeersoveradultsfortoyquestionsandadultsoverpeersforfoodquestions(VanderBorght&Jaswal,2009).

Inasecondsetofstudieschildrenwerepresentedwithtwomod-els thatdiffered in thedegreeof competenceon thevery samedi-mension(e.g.,accuratevs. inaccuratelabeling). Inmanysuchstudieschildrenpreferredthemorecompetentmodelalsofortaskstotallyun-relatedtotheparticulardomainofcompetencethemodeldisplayed:For example, they expected a knowledgeablemodel to be nice too(Brosseau-Liard&Birch,2010),anicemodeltobesmartandathletic(Cain,Heyman,&Walker,1997),astrongmodeltobeniceandsmart(Fusaro,Corriveau,&Harris,2011)oranattractivemodeltobeknowl-edgeable (Bascandziev & Harris, 2014). However, children did notalwaysshowsuchapatternofpreferences:First,theinferenceschil-drendrawfromsocialtoepistemicdomainsdecreaseonceepistemiccues,thataremoreclearlypredictiveforepistemicproblems,aremadeavailable:olderpreschoolers,forexample,trustanaccurate-unfamiliarover an inaccurate-familiar informant (Corriveau & Harris, 2009), aforeign-accentedandaccurateoveranative-speakingandinaccuratespeaker(Corriveau,Kinzler,&Harris,2013)orareliabledissenteroveran unreliable majority (Bernard, Proust, & Clément, 2015). Second,

children’spreferenceforamorecompetentmodeloutsideherdomainof competenceseems tobe restricted toavoidanceof incompetentmodels,butdoesnottendtooccurwhenchildrenchoosebetweenaneutralagentandanexpertmodel(Koenig&Jaswal,2011).

Takentogether,children’scognitivestrategiestendtodifferasafunction of the epistemic situations they find themselves in:Whentheyareconfrontedwithtwoexpertswithsimilardegreesofcompe-tenceyetindifferentdomains,theyshowrationalinferencesbasedonthe ascriptionof specific competencies. In contrast,when choosingbetweentwomodelsthatdifferinoverallcompetenceononedimen-sionthepictureislessclear.

1.2 | Rational competence masked by performance factors?

Howcanwemakesenseofthesefindings?Onepossibilityisthatchil-drenarecapable,inprinciple,ofrationalselectivelearningandtrust,but that this competence is masked in some situations by certainperformancefactors.Onesuchlimitingperformancefactormightbethe structure and demands of the task. Childrenmight use simplerstrategiessuchasglobalimpressionformationaslongastheseyielduniquedeterminateanswersforagiventask;andtheymightrecruittheirmorerationalstrategiesonlyintasksforwhichthesimplerstrat-egiesfailtoyielduniqueanswers.Whenmodelsdifferinoverallcom-petenceinanydomain,childrenmightthustendtopreferthemorecompetentmodeleven for tasksoutsideof theactual scopeofhercompetence. Incontrast,whenevermodelsaresimilarlycompetent,yetindifferentdomains,aglobalstrategywouldbeinconclusiveandnotleadtoadeterminatesolution.Insuchcaseschildrenmightthusmakeuseoftheirrationalcompetenceandrecruitmodelsinaccord-ancewiththeirperceivedtraits.Totestforthispossibilitydirectly,onewouldhavetomanipulatethetaskstructureandmodelcontrastsinawithin-subjectdesign,andassesschildren’sperformanceintaskswiththetwostructures,respectively–amethodologicalapproachthat,toourknowledge,nostudyhasrealizedsofar.

Besidestaskstructure,whatotherlimitingfactorsmaymaskchil-dren’scapacityforrationalinferencesinselectivetrust?Anothercan-didatethatcomestomindis(lackof)backgroundknowledge(Sobel&Kushnir,2013).Imagineachildisconfrontedwithmodelsdifferinginacertainbehaviorthatisinfactdiagnosticforacertaintrait(e.g.,liftingobjectsasan indicatorofstrength).Now,thechildcanonlyseethebehaviorasevidentialbasisforascriptionofthetraitinquestion,andthusasabasis forpredictingfuturebehaviorexpressiveof thetraitifshehassomebackgroundknowledgeaboutthedomainandsomemasteryof the relevant trait concept.Just as someonewithout anybackgroundknowledgeaboutfootballcannotseeaneventas‘offside’,a childwithout any background knowledge about physical strengthandthuswithoutthetraitconcept‘strong’cannotseeeventsoflifting,etc.asexpressionsofanagent’sstrength.Intheabsenceoftherele-vantbackgroundknowledgeandtraitconcepts,all thatchildrencando,itseems,istoengageinglobalimpressionformation.

Evidencecompatiblewiththisideacomesfromthosestudiesre-viewedabovethatshowthatchildrendifferentiatebetweenexperts

     |  3 of 12HERMES Et al.

from different domains – since such differentiation plausibly wasbasedonconceptualknowledgeaboutthespecifickindsofexpertise(Kushniretal.,2013;Lutz&Keil,2002;VanderBorght&Jaswal,2009).However, this evidence is rather indirect, since background traitknowledgewas not independently assessed in these studies.Moredirectpreliminarysupportcomesfromarecentstudythatdidassesschildren’sselectivelearningandtheirtraitknowledgeindependentlyand showed thatonly thosechildrenwith the requisite trait knowl-edgemaderationalmodelchoices(Hermes,Behne,&Rakoczy,2015).Convergingevidenceforadirectrelationbetweenconceptualknowl-edgeandselectivemodelchoicecomesfromanotherrecentstudyinwhichchildren’sownunderstandingoffalsebelief,forexample,pre-dictedhowmuchtheyendorsedinformationfromaprotagonistwithsophisticated false belief understanding over that of a protagonistwithnosuchunderstanding(VanReet,Green,&Sobel,2015).

Inadditiontothestructureofthetasksandtheavailabilityofback-groundtraitknowledge,twomorefactorsmightpotentiallyinterferewithchildren’srationalcompetenceforselectivelearning:pragmaticconfusion and executive function. Concerning pragmatic confusion,inthestudiesreviewedabove, inwhichchildrenseemtogeneralizewidelytounrelateddimensions,theformatofthetestquestionsmayhavemisledthem:Whenchildrenareaskedtochoosebetweenmod-elsforataskthatrequiresacertaintrait,andnoinformationconcern-ingthattraitisprovidedaboutthemodels(e.g.,aknowledgequestionaftera strengthdemonstration), themost rational answerwouldbeindifference(‘Idonotknow’,‘both’, ‘none’).Astandardforcedchoicequestion thatprovidesalternatives (e.g., ‘Whoknows…modelAormodelB?’)butexcludesthemostrationalone(‘neither’/’both’)isprag-maticallymisleadingandmightmaskchildren’s competenceand se-ducethemintogivinganswersbasedontheirglobalimpressions(sinceweknowfromalargebodyofresearchthatyoungchildrenarehighlypronetosuggestivecues;Ceci&Bruck,1993).Ifthiswerethecase,itshouldbeeasierforchildrentoexpresstheirrational inferencesifquestionswerelesssuggestive.

Concerningexecutivefunction(EF),weknowthatEFisrelatedtorationalselectivetrustdecisions.Youngchildrenhavearobustbiastotrusttestimonyintheabsenceofcompetinginformationeveniftheinformantwashighlyunreliablepreviously(Jaswal,Croft,Setia,&Cole,2010).However,iftheydohavecompetingfirst-handexperience(e.g.,aboutthelocationofahiddenobject),childrencanovercomethisun-criticalbiastotrust,andthedegreetowhichtheydosostronglycor-relateswiththeirEF(Jaswal&Pérez-Edgar,2014;Jaswaletal.,2014).Thus,preschoolers’useofrationalstrategiesofselectivetrust,moregenerally,maybecloselyrelatedtotheirdevelopingEF:AdvancedEFmighthelptoinhibitglobalreasoningasaprimaryresponsewhentwomodelsdifferinoverallcompetence,ortoidentifytherelevanttraitforagiventaskwhenconfrontedwithtwomodelswithsimilardegreesofcompetenceindifferentdomains.

1.3 | Rationale of the present study

In sum, then, the present empirical picture regarding young chil-dren’sselectivetrustismixed,withsomestudiessuggestingthatthis

capacity is based on sophisticated reasoning about traits whereasother studies tend to suggest that it is based on global impressionformation.How canwemake sense of this? Themain rationale ofthepresentstudywastotestwhetherthegeneralideaofarationalcompetence account (that children’s selective trust is basedon thecognitivecompetenceforrationalinductivelearning,maskedincer-tainsituationsbycertainlimitingperformancefactors)canexplainthedivergingpatternsofperformanceindifferentselectivelearningtasks.More specifically,we tested the influenceof potential limitingper-formancefactors,focusingespeciallyondifferenttaskstructuresandbackgroundtraitknowledge.

Totestwhetherchildren’srationalselectivemodelrecruitmentislimitedbythetaskstructure,wesystematicallyvariedandcontrasted(within-subject)tasksinvolvingtwotypesofmodelcontrasts:

First, in the degree-of-competence contrasts, children were con-frontedwithtwomodelsthatdifferedincompetenceinonedimen-sion(strengthorknowledge)andwereaskedtodecidebetweenthemodels for strength-related tasks and knowledge-related tasks. Ifavailability/lack of trait knowledge is the only limiting performancefactor, childrenwith the corresponding trait knowledge shouldpre-feracompetentoveranincompetentmodelonlyfortaskswithinthescope of the model’s competence (e.g., strong model for strengthtasks) but choose indifferently for tasks outside the scope of hercompetence (knowledge-related tasks after strength manipulation).However,if–inadditiontotraitknowledge–taskstructureisalimit-ingfactor(suchthatchildrenreverttoglobalimpressionformationaslongasityieldsuniqueanswers),childrenshouldgeneralizewidelyinthedegree-of-competencecontrasts(e.g.,preferringthestrongmodelforbothknowledge-andstrength-relatedtasks) irrespectiveoftheirtraitknowledge.

Secondly,inthedomain-of-competence contrast,childrenencoun-tered two models that were both highly competent, but in differ-entdomains (strongvs. accuratemodel). In this contrastweexpectchildrentochoosemodelsrationally inaccordancewiththemodels’competencies, as long as they do not lack the corresponding traitknowledge.Inthiscasethetaskstructureshouldnotlimitrationalrea-soningsince thepredominantstrategiesbasedonglobal impressionformationareinconclusivewhenbothmodelsshowsimilardegreesofcompetence.Thewithin-subjectdesignallowedustotestnotonlyforchildren’sperformanceineachmodelcontrast,butalsofortherelationofperformanceacrossthecontrasttypes.Inparticular,wewereabletotestwhether–aspredictedbyarationalcompetenceaccountthatconsiders taskstructureacrucialperformancefactor–childrenwhorationallychosemodelsinthedomain-of-competencecontrastwerestillsubjecttowidegeneralizationsbasedonglobalimpressionforma-tioninthedegree-of-competencecontrasts.

Totestfortheeffectofbackgroundtraitknowledge,wedirectlymea-suredchildren’straitknowledgeusingexplicitquestions(asinHermesetal.,2015).Ifonlythosechildrenwhoanswerthetraitquestionscor-rectlyrecruitmodelsrationally,thiswouldspeakforageneralrationalcompetenceaccount,withlackoftraitknowledgeasalimitingfactor.

Furthermore, in addition to task structure and trait knowledge,weexploredthepragmaticsofsuggestivequestioningandexecutive

4 of 12  |     HERMES Et al.

functionastwoadditional,potentiallylimitingperformancefactorsonchildren’srationalselectivetrustdecisions.

2  | METHOD

2.1 | Participants

Seventy-five children were included in the final sample; 37 chil-drenwere 4-year-olds (48–58months,M = 53.2months,SD=3.39,17 girls) and 38 children were 5-years-olds (60–71months,M=66.3months,SD=3.48,19girls).Wechosethisagerangebasedonrecentresearchshowingthatchildrenfrom4to5yearsperformcompetentlyinselectiverecruitmentwhenconfrontedwithtwomod-els with equivalent competence, yet in different domains (Hermesetal.,2015;Li,Heyman,Xu,&Lee,2014).Eightadditionalchildrentookpartbutwerenotincludedinthefinalsample,becausetheydidnotcompletethetestsession(n = 3)orfailedtoanswerthecompre-hensionquestionscorrectly(n = 5,fordetailsseebelow).

Children were recruited from a database of families who hadagreed to participate. They were from mixed socioeconomic back-groundsandsomechildrenhadacquiredmorethanonelanguage.Aswehadnodataabouttheirproficiencyinthelanguageusedfortest-ing,avocabularyscreeningwasconducted(vocabularytestfromtheKaufmanAssessment Battery for Children; see Kaufman, Kaufman,Melchers, & Preuss, 1994). Since 13 children scored significantlybelowaverage(twostandarddeviationsbelowthemeanfornormallydeveloping48–51-month-olds),allstatisticalanalyseswerealsorunseparatelywiththesechildrenexcluded.Theresultsofthesecontrolanalysesreplicatedtheresultsforthewholesample.

2.2 | Design and outline

Theselectivetruststudyfolloweda2×2design,examiningtheef-fectof question format (standardquestionvs.reducedsuggestibility)and model contrast (degree-of-competence or domain-of-competence). Thefactorquestion formatwasvariedbetweensubjects:halfof thechildrenwereaskedthestandardclosedquestions,andfortheotherhalfthesuggestivenessofthequestionswasreducedbymakingtheanswer‘boththesame’anequallyavailableoption.

The factormodel contrastwasvariedwithin subjects: In thede-gree-of-competence contrasts, all children saw two pairs of modelsthatdifferedinoverallcompetenceinthedomainsstrength(success-ful vs. unsuccessful lifting) and knowledge (expert labelingvs. inac-curate labeling) respectively, whereas in the domain-of-competence contrast, children were confronted with the two highly competentmodels (strong and accurate). The degree-of-competence contrastswereshownfirsttoensurethatchildrenunderstoodthatthedegreeofcompetenceinstrengthandaccuracymightvarybetweenmodelsandwouldnotassumethatallthesepuppetswerehighlycompetentinbothdimensionswhentheysawthedomain-of-competencecon-trast.Afterthedemonstrationsofeachofthethreemodelcontrasts,childrenwereasked to select amodelwhen theyneeded toobtainthelabelsfornovelobjects(knowledgetasks)andwhentheyfaceda

problemthat requiredphysicalstrength (strengthtasks). Inaddition,traitquestionswereaskedasameasureofbackgroundtraitknowl-edgeandforanalysisthesamplewillbedivided intothosewhoan-sweredtraitquestionscorrectlyandthosewhoansweredincorrectly.

2.3 | Material and procedure

Childrenweretestedinquietroomsintheirdaycarecenters,withtestsessionslastingapproximately30minutes.Thetestsessionsstartedwith thevocabulary screening followedbyapracticegame to trainquestion formats according to condition. The main selective truststudyconsistedoftwodegree-of-competencecontrasts,followedbyatestblockeach,andonedomain-of-competencecontrast,followedbyathirdtestblock.Afterthischildrenansweredtraitquestionsandpreferencequestionsaboutthemodels,andattheendtheydidtheexecutivefunctiontest.

2.3.1 | Question format training

The question formats used in the selective trust studywere intro-ducedinashortanimalquiz:Childrenwereshownthreepairsofani-mals(onepairatatime,e.g.,acowandapig)andaskedtoattributecharacteristics to the animals.The characteristics, thequestion for-matusedandtheansweroptionsprovideddifferedbetweenthetwoquestionformatconditions: Inthe ‘reducedsuggestibility’condition,childrenwereasked toattribute characteristics that applied toonlyoneoftheanimals(e.g.,whoispink?)andadditionallyacharacteristicthatappliedtobothanimals(e.g.,wholivesonafarm?).Childrenwereprovidedwithananswercardthatdepictedthreeansweroptions,(1)justanimalA,(2)justanimalB,or(3)animalAandB;andthequestionformatalsohighlightedthis (e.g.,Whoispink,onlythepig,onlythecow,orboth?).Childrenwereaskedtoanswerbypointingtothean-swercardandweregivenfeedbackontheirresponse.Inthe‘standardquestions’condition,childrenwereshownthesamepairsofanimalsbutwereonlyaskedtoattributecharacteristics thatappliedtoonlyoneoftheanimalswithstandardquestions(e.g.,Whoispink,onlythepigoronlythecow?).Theyansweredwithoutanswercardsandweregivenfeedbackontheirresponse.Fortherestofthetestsessions,allquestions in the ‘reduced suggestibility’ condition inwhich childrenwereaskedtodecidebetweenmodelsendedwith‘…modelA,modelB,orboth?’andchildrenweregivenanswercardsthatdepictedthethreeoptions.Inthe‘standardquestions’condition,allthefollowingquestionsendedwith‘…modelAormodelB?’andchildrenansweredwithoutanswercards.

2.3.2 | Main selective trust study

Childrenwereshownpairsofmodelsonanotebookpc.First,theysawmodelcontrastswithanoveralldiscrepancyincompetence(degree-of-competence contrasts) in the domains strength (strong vs.weakmodel)andlabelingaccuracy(accuratevs.inaccuratemodel).Theninthedomain-of-competencecontrastthetwohighlycompetentmod-els(strongandaccurate)werecombined.Eachdemonstrationphase

     |  5 of 12HERMES Et al.

was followed by a test blockwith four trials (two knowledge trialsandtwostrengthtrials).SeeFigure1forasummaryoftheprocedure.

Degree- of- competence contrastsChildrenparticipatedindegree-of-competencecontrastsinboththeaccuracy and the strength domain, with the order of presentationcounterbalancedacrosssubjects.Foreachdomain,apairofmodelswas introduced in a demonstrationphase, followedby a test blockwithfourtrials(twoknowledgetrialsandtwostrengthtrials).

Accuracy demonstration phase (4 trials).Childrenwereintroducedtotwopuppetmodels(Ms.YellowandMs.Blue;degreesofcompetencecounterbalanced). In each demonstration trial, the modelswere pre-sentedwithapictureofaknownobjectinfrontofthem(e.g.,ajacket,seeSupporting InformationA).Childrenweregiven the samepictureandwereaskedwhatitdepicted.Theynormallyprovidedthecommonlabel(e.g.,jacket).Theexperimenterdouble-checkedthelabelonalistandprovidedanexpertlabel(e.g.,skijacket).Thenchildrenwereshownavideoinwhichonemodelprovidedthesameexpertlabelaspreviouslyreadfromthelistwhereastheothermodelprovidedawronglabel(e.g.,pullover).Thustheaccuratemodelwaspresentedasmoreknowledge-ablethanthechildherselfandtheinaccuratemodelaslessknowledge-able.Expertlabelsalwaysincludedthecommonlabels(e.g.,skijacket) so thatchildrenwouldnotmistakethemasinaccurate.Astheinaccuratemodelwassupposedtobeperceivedasinaccurateratherthanasbizarreordeceptive,thewronglabelswerealwayspartofthesamecategory(e.g., pullover for jacket). After each trial, the experimenter repeatedwhatthemodelshadsaidandafterthefourtrialschildrenwereaskedacomprehensionquestion:Whowasgoodatlabelingthesethings…?(Seeaboveatquestion format training forhowquestionsdifferedbetweenconditionshereandforallfollowingquestions.)Ifthechilddidnotan-swercorrectly,anadditionaltrialwaspresentedfollowedbyarepetitionofthecomprehensionquestion.Ifchildrenstilldidnotanswercorrectly,theywereexcludedfromthefinalsample(fordetailsseebelow).

Strength demonstration phase (4 trials).Childrenwereshownanotherpairofpuppetmodels(Ms.GreenandMs.Red;degreesofcompetencecounterbalanced).Atthestartofeachdemonstrationtrial,onemodelhadalightobject(e.g.,acup)lyinginfrontofherandtheothermodel

aheavyobject (e.g., apotato sack, seeSupporting InformationA fordetails).Onvideochildren then sawonepuppet always failing to liftthelightobjectssaying‘Ican’tmanagethis!’andtheotherpuppetal-ways succeeding in lifting theheavyobjects saying ‘I’mgoodat this’(ordercounterbalancedacrosstrials).Theexperimenterrepeatedwhathadhappenedaftereachtrial(‘Ms…wasabletolift…andMs…wasnot able to lift…’).Theobjects tobe liftedwere chosen so that theweakpuppetwasperceivedasweakerthanthechildherselfandthestrongpuppetasstronger.Afterfourtrials,weaskedacomprehensionquestion:‘Whowasgoodatliftingthesethings…?’.Ifchildrendidnotanswercorrectly,theywereshownonemoredemonstrationtrialandaskedthecomprehensionquestionagain.Childrenwhofailedtoanswertherepeatedcomprehensionquestionscorrectly,eitherfollowingtheaccuracydemonstration(n = 3)orthestrengthdemonstration(n = 1) or afterbothdemonstrations(n = 1)wereexcludedfromthefinalsample.

Test blocks in the degree-of-competence contrasts.Aftereachofthestrengthand theaccuracydemonstrationblocks,childrenwerepre-sentedwitha testblock.Each testblock consistedof two strengthtrialsandtwoknowledgetrials (ordercounterbalanced).Atthestartofeachtrialthemodelshadanobjectlyinginfrontofthemandtheexperimenter said: ‘Look what they have! I brought the same foryou.’Childrenweregiventhesameobjectstoacton(seeSupportingInformationBforobjectsandexacttestquestions).Intheknowledgetrialstheobjectswereunknowntothechildrenandthechildrenwereaskedasatestquestion:‘Whoknowswhatthisiscalled…?’Ifchildrenclaimedtoknowtheobject,thelabeltheyprovidedwasdoubtedbytheexperimenterandiftheyinsisted,aspareobjectwasintroduced.Inthestrengthtrials,theobjectsrequiredphysicalstrengthtoactonthem (e.g., to lift a brick).After children actedoneachobject, theywereaskedwhichmodelwouldbegoodatperformingtheaction.Foreachtask,weacceptedpointingtothepuppet(ortheanswercard,ifapplicable)orsayingthecolorofthepuppetsasanswers.

Domain- of- competence contrastForthedomain-of-competencecontrast,thetwocompetentmodels,i.e.,thestrongandtheaccurateonefromthedegree-of-competencecontrastswerepresented.

F IGURE  1 Procedureofthemainselectivetruststudy.Notethatinthedegree-of-competencecontrasts,theorderoftheaccuracyandstrengthphasewascounterbalancedbetweensubjects

6 of 12  |     HERMES Et al.

Demonstration reminder (4 trials). Children witnessed the strongand the accurate model in four demonstration trials that generallyfollowed the same procedure as in the degree-of-competence con-trasts.Intwotrialsthestrongpuppetsucceededinliftingaheavyob-jectwhiletheaccuratepuppetdidnothing.Intheothertwotrialstheaccuratepuppetprovidedanexpertlabelforaknownobjectwhilethestrongpuppetsaidnothing(seeSupportingInformationAforobjects).Theexperimenteraskedtworeminderquestions:‘Ms…,wasshegoodornotsogoodatliftingtheseobjects?’and‘Ms…,wasshegoodornotsogoodatlabelingtheseobjects?’Childrenwhodidnotanswerbothquestionscorrectlywereshowntwomoredemonstrationtrials (onepuppetperforminganactioneach)andaskedthereminderquestionsagain.Seventeenchildrenstilldidnotanswerbothquestionscorrectly.Sincethesechildrencorrectly identifiedthestrongandtheaccuratemodelinthedegree-of-competencecontrasts,theywerenotexcludedfrom analysis. However, all results for the domain-of-competencecontrastwere replicatedwith the smaller subsample (with childrenexcludedwhoerred in the reminderquestions),withoneexceptiondescribed below.

Test block in the domain-of-competence contrast.Thefollowingtestblockwasthesameasdescribedaboveforthedegree-of-competencecontrasts.Thatis,childrenagainparticipatedintwostrengthtrialsandtwoknowledgetrialsandwereaskedthesamequestions(fordetailsseeaboveandSupportingInformationB).

Trait questions and preference questionsAttheendoftheselectivetruststudy,childrensawthestrongmodelandtheaccuratemodelonthescreenandwereaskedtwotraitques-tions(‘Whoisstrong…?’and‘Whoissmart…?’)andtwopreferencequestions(‘Whoisnice…?’and‘Whowouldsharehersweets…?’)

2.3.3 | Executive functions task

At the end of the test session, children performed the Head-Toe-Knee-Shoulder task (HTKS) (McClelland etal., 2014; Ponitz,McClelland,Matthews,&Morrison,2009;Ponitzetal.,2008)totesttheircapacitytoinhibitapredominantresponse.Thatis,onhearingarequesttoperformaparticularaction(e.g.,‘Touchyourhead’)chil-drenwererequiredtoperformanoppositeaction(e.g.,totouchtheirfoot).Childrenperformedon30trialsinthreestageswhichincreasedincognitivecomplexity(addingadditionalrulesinstage2andchang-ingrulesinstage3;seeMcClellandetal.,2014).

2.4 | Coding procedure

Two coders watched half of the videos each. For the degree-of-competence contrasts, it was coded whether children chosethe competent model (accurate or strong) or the incompetentmodelorexpressed indifference.For thedomain-of-competencecontrast and the trait and preference questions, it was codedwhether children chose the strongmodel or the accuratemodelor expressed indifference (see Supporting InformationC for de-scriptives). An indifference response was scored when children

answered ‘both’ (86%of all cases), ‘none’ (1%), or ‘I don’t know’(13%) (see Supporting InformationD for details). In each trial intheHTKStaskchildrenreceivedtwopointsiftheyspontaneouslyrespondedcorrectly,onepointiftheyinitiallydidawrongaction,butthencorrectedthemselvesandzeropointsiftheyperformedawrongactionanddidnotcorrectit.Asumscorewascalculatedacrossall(upto30)trials(range:0–60points).

2.4.1 | Inter- rater reliability

Oneadditionalcoderwhowasblindtothehypothesescodedasub-set of 20%of the data (n = 17participants). Agreement for all testquestions and comprehension questions in themain selective truststudywasperfect(Ks=1).Thereweredeviationsintwodatapointsin the reminder question resulting inK = .82** for the accuracy re-minderquestionsandK = .62forthesecondstrengthreminderques-tionforthosechildrenwhoansweredthefirstincorrectly.Inter-rateragreementconcerningtheHTKSsumscorewashigh,asindicatedbyaone-wayrandomintraclasscorrelation(ICC=.93,F(14,15)=13.72,p < .01).

3  | RESULTS

3.1 | Focal analysis: patterns of selective model choice in the degree- of- competence contrasts and in the domain- of- competence contrast

We were interested in children’s model choice and patterns ofgeneralizations across domainswhen given a choice between twomodelsthatdifferedinoverallcompetence(degree-of-competencecontrasts)andwhengivenachoicebetweentwomodelswithhighcompetenceyet indifferentdomains (domain-of-competencecon-trast).Toexplorethese,weaggregatedchildren’sresponsesacrossthetwotrialsofeachtask.Specifically,inthedegree-of-competencecontrasts we aggregated children’s responses to four preference scores,oneeachforthestrengthtasksandtheknowledgetasksafterboththestrengthdemonstrationandtheknowledgedemonstration:foreachofthesewedeterminedthenumberoftrials(0–2)inwhichthechildchosethecompetentmodel(strongorknowledgeable)andthenumberoftrials(0–2)inwhichshechosetheincompetentmodel(weakorunknowledgeable),andcomputedthepreferencescorebysubtractingthelatterfromtheformer.Thus,thepreferencescorescouldrangefrom−2to2,withhighvalues indicatingapreferenceforthemorecompetentmodel.Inthedomain-of-competencecon-trast,children’sresponsesacrossthetwostrengthtrialsandthetwoknowledgetrialswereaggregatedtotwoexpert scoresinthefollow-ingway:Foreachtypeofquestion,wedeterminedthenumberoftrials(0–2)inwhichthechildchosethemodelwiththerelevantex-pertise (strongmodel for strength trials, knowledgeablemodel forknowledgetrials),thenumberoftrials(0–2)inwhichshechosethemodelwiththeirrelevantexpertise(e.g.,strongmodelforknowledgetrials)andcomputedtheexpertscorebysubtractingthelatterformthe former.Thus theexpert scorescould range from−2 to2with

     |  7 of 12HERMES Et al.

highvaluesindicatingapreferenceforthemodelwiththerelevantexpertise.1

Figure2 depicts these preference scores and expert scores foreachtestblockineachgroupoftasks.Inthedegree-of-competencecontrastschildrenpreferredthemorecompetentmodelssignificantlyabove chance for all tasks (see Figure2a). This indicates that chil-drengeneralizedwidely,selectingthemorecompetentmodelsforalltasks, includingthose irrelevanttothecompetenciesdisplayed (e.g.,for knowledge tasks after strength demonstration). In the domain-of-competencecontrast,childrenpreferredthemodelswiththerel-evantexpertise(i.e.,thestrongmodelforthestrengthtasksandtheknowledgemodelfortheknowledgetasks)significantlyabovechance(seeFigure2b).Thusinthedomain-of-competencecontrast,childrenrationallychosethemodelsinaccordancewiththeirrespectivecom-petenciesandtherequirementsofthegiventasks.2

3.1.1 | Aggregation across strength and knowledge tasks

We explored whether the degree of children’s preference for thestrong model differed from that for the knowledgeable model. Inthe degree-of-competence contrasts, there was no difference inchildren’sdegreeofpreferenceforthecompetentmodel(strongorknowledgeable),whenthedomainofdemonstrationandtaskmatched(i.e.,preferencescoresforstrengthtaskafterstrengthdemonstrationand for knowledge task after knowledge demonstration,Wilcoxon

signed-ranktest,Z=1.81,p > .05),andwhenthedomainofdemon-strationand taskdidnotmatch (i.e., preference score for strengthtask after knowledge demonstration and for knowledge task afterstrengthdemonstration,Z = 0.78,p > .05).Similarly, in thedomain-of-competencecontrasttheexpertscoreofthestrengthtaskswasnot different from that of the knowledge tasks (Z = 0.12, p > .05). Thus, children’smodel preferences did not differwhen theywerereasoningaboutknowledgeandstrength.

Forfurtherstatisticalanalyseswe,thus,aggregatedthedataacrossstrengthtasksandknowledgetaskstoscoresindicatingwhetherchil-dren chose the relevantmodels in the relevant and irrelevant tasksinthefollowingway:Inthedegree-of-competencecontrastsweag-gregatedchildren’spreferencescoresbywhetherornotthespecificcompetencedemonstratedbythemodelswasrelevanttoagiventask.Tothisend,wecreatedarelevant tasks scorebyaddingthepreferencescoreofthestrengthtrialsafterstrengthdemonstrationandtheonefor theknowledgetrialsafterknowledgedemonstration,yieldinganaggregate score ranging from−4 to4,withhighvalues indicatingapreferenceformodelsthatwerecompetent(ratherthanincompetent)indomainsrelevanttothetasksathand.Andwecreatedanirrelevant tasks scorebyaddingthepreferencescoreofthestrengthtrialsafterknowledgedemonstrationandtheoneoftheknowledgetrialsafterthestrengthdemonstration,yieldinganaggregatescorerangingfrom−4to4,withhighvaluesindicatingapreferenceformodelsthatwerecompetent(ratherthanincompetent)indomainsirrelevanttothetaskat hand. In the domain-of-competence contrastwe aggregated the

F IGURE  2 Children’smodelpreferenceinthestrengthtasksandtheknowledgetasks.Figure2adepictschildren’spreferencescoresinthestrengthandtheknowledgetasksinthedegree-of-competencecontrasts.Positivevaluesindicateapreferenceforthecompetent(i.e.,strongorknowledgeablemodel,accordingtodemonstration)overtheincompetent(i.e.,weakorunknowledgeable)modelsandnegativevaluesindicateapreferencefortheincompetentoverthecompetentmodels.Figure2bdepictschildren’sexpertscoresinthestrengthandtheknowledgetasksinthedomain-of-competencecontrast.Positivevaluesindicateapreferenceforthemodelwiththerelevantexpertise(e.g.,strongmodelforthestrengthtasks)overthemodelwiththeirrelevantexpertise(e.g.,knowledgeablemodelforthestrengthtasks)andnegativevaluesindicatethereversepreference.Thelineat0reflectschancelevel.Asterisksshowdeviationfromchancelevel(Wilcoxonsigned-ranktests,**p < .0017 [Bonferroni-correctedα =.01for6tests]).Errorbarsshowstandarderrors

8 of 12  |     HERMES Et al.

expert score for the strength tasks and the one for the knowledgetasks toacombined relevant model score ranging from−4 to4,withhighvaluesindicatingpreferencesformodelswhosecompetencewasrelevant(ratherthanirrelevant)foragiventask.

3.1.2 | The role of task relevance in the degree- of- competence contrasts

Inthedegree-of-competencecontrastschildrenpreferredthecom-petent over the incompetent models for all tasks including thoseirrelevant to thespecificcompetenceofamodel (seeFigure2).Toassess whether the degree of children’s preference for the morecompetentmodelwashigherfortasksrelevanttothespecificcom-petencethemodelhaddemonstratedthanforthoseirrelevanttothespecificmodel competence,we compared the relevant tasks scoreand the irrelevant tasks score.Childrendid choose the competentmodelmoreoftenintasksrelevanttohercompetencethanintasksirrelevant to her competence (Wilcoxon signed-rank test,Z = 2.99,p < .01, r = .35). This shows that although children did show halo-patterns(choosingthecompetentmodelmoreoftenthantheincom-petentone,evenwhenthecompetencedisplayedwasirrelevanttothe task), therewas still somedifferentiationsuch that theychoseagivenmodellessoftenwhenhercompetencewasirrelevantthanwhenitwasrelevant.

3.2 | Additional analyses

To explore whether children’s patterns of selective model choicewererelatedtotheirage,theirtraitknowledge,thequestionformattheywereexposedto,ortheirdegreeofexecutivefunction,weana-lyzedhowthese factorspredicted the relevant tasksscoreand theirrelevant tasks score in the degree-of-competence contrasts andthe combined relevant model score in the domain-of-competencecontrast.Traitknowledge,agegroup,andquestionformatconditionwereincludedascategoricalpredictorsinordinalregressionsoneachofthesethreescores.DuetomissingdataintheHTKStask,3 we as-sessedtheeffectofchildren’sexecutivefunctionseparately incor-relationanalyses.

3.2.1 | Trait knowledge

We testedwhether children’s trait knowledge, as indicated in theirexplicit answers to the trait questions at the end of the selectivetrust study, affected their patterns of model choice. Preliminaryanalyses showed thatchildrencorrectly identified the ‘smart’modeland the ‘strong’model at abovechance levels.Whenasked ‘who issmart?’more children chose the accuratemodel rather than choos-ing the strong model (χ2(1)=20.25, p < .01). Similarly, when asked‘whoisstrong?’,morechildrenselectedthestrongmodelratherthanchoosing the accuratemodel (χ2(1)=37.88,p < .01) (see SupportingInformationCfordetails).Forordinalregressionanalyses,weaggre-gatedthedataoverbothtraitquestionsandcomparedthesub-sampleofchildrenwhoansweredbothtraitquestionscorrectly(n=46)with

thesub-sampleofchildrenwhoansweredatleastonetraitquestionincorrectly(n = 29).Figure3showschildren’smodelchoiceseparatelyforthesetwosub-samples.

Forthedegree-of-competencecontrasts,traitknowledgedidnotpredicttherelevanttasksscoreorthe irrelevanttasksscore.AscanbeseeninFigure3a,allchildrenincludingthosewhomademistakesansweringthetraitquestionandthosewhoansweredthetraitques-tionscorrectlypreferredthecompetentovertheincompetentmodelsabovechanceinalltasks.Thustraitknowledgedidnothelpchildrenrestrict their generalizations to relevant tasks. In contrast, for thedomain-of-competence contrast, children’s trait understanding didpredicttheircombinedrelevantmodelscore:Thosechildrenwhoat-tributedbothtraitscorrectlywere9.75(95%CI,3.54to26.87)timesmorelikelytoselectthemodelwiththerelevantexpertiseforagiventask than those childrenwho did not attribute both traits correctly(Waldχ2(1)=19.37,p < .001).As canbe seen inFigure3b, childrenwhomadeatleastonemistakeinattributingthetraitschosebetweenthemodelsatchancelevel.Incontrast,thosechildrenwhoattributedboth traits correctly selectively chose the model with the relevantcompetenceforagiventask(i.e.,thestrongmodelforstrengthtasksandtheaccuratemodelforknowledgetasks).

3.2.2 | Age group

Children’s age group did not significantly predict the relevant tasksscoreortheirrelevanttasksscoreinthedegree-of-competencecon-trasts. In the domain-of-competence contrast, children’s age groupdidpredictthecombinedrelevantmodelscore:5-year-oldswere2.54(95%CI,1.05to6.13)timesmorelikelythan4-year-oldstoselectthemodelwiththerelevantexpertiseforagiventask(Waldχ2(1)=4.31,p < .05).However,thecombinedrelevantmodelscorewassignificantlyabove chance level for both 4-year-olds (combined relevant modelscore=0.89;Wilcoxonsigned-ranktest,Z = 2.61,p < .01,r = .43)and5-year-olds(combinedrelevantmodelscore=2.32;Wilcoxonsigned-rank test, Z = 4.68, p < .001, r = .76). Thus, both 4-year-olds and5-year-oldschosethemodelwiththerelevantexpertiseforagiventaskabovechance,but5-year-oldsdidsoevenmorefrequentlythan4-year-olds.

3.2.3 | Question format

In a preliminary analysiswe compared the sumof indifference an-swerschildrenprovidedacrossthe12testtrialsasafunctionofthequestionformat(fordetailsseeSupportingInformationD).Childrenin theReducedSuggestibility conditionprovidedmore indifferenceanswers (M = 3.62) than those in theStandardQuestionscondition(M = 0.74;Mann-Whitneytest,U(38,37)=290,p < .001,r = .53).Yetintheordinalregressionanalyses,thequestionformatchildrenhadencountereddidnotpredicttheirmodelchoiceineitherthedegree-of-competence contrasts or the domain-of-competence contrast.Therefore,althoughthosechildrenwhowereaskedlesssuggestivelyprovided more indifference answers than those who were askedstandard questions, this difference in frequencywas unsystematic.

     |  9 of 12HERMES Et al.

ChildrenintheReducedSuggestibilityconditiondidnotexpressin-difference particularly often in caseswhere indifferencewas justi-fied(whenchoosingmodelsfortasksforwhichtheirdemonstratedcompetencewas irrelevant; irrelevant tasksscore in thedegree-of-competencecontrasts).

To assess potential masking effects of suggestive questions inmoredetail,theresponsesofchildrenwithtraitknowledgeareofspe-cial interest:Would thesechildrenrefrain fromwidegeneralizationswhenasked lesssuggestively?Thecrucialsubsampletotestthis in-cludesonlythosechildreninthereducedsuggestibilityconditionwhoansweredbothtraitquestionscorrectly (n = 21).Plannedanalysisofthis subsample in thedegree-of-competencecontrasts showed thateven these children drewwide generalizations, preferring themorecompetentmodel above chance even for tasks outside themodel’sspecificcompetence(irrelevanttasksscore=2.24;Wilcoxonsigned-ranktest,Z = 3.57,p < .001,r = .78).

3.2.4 | Executive functions

Inordertotestwhetherexecutivefunctionwasspecificallyrelatedtorationalratherthansimpler,impression-basedprocesses,wecorrelatedthe HTKS scores with children’s decisions in both model contrasts:First,forthosetasksinwhichchildrensawmodelsthatdifferedinover-allcompetence(degree-of-competencecontrasts),theHTKSscoresdid

notcorrelatewiththe irrelevanttasksscore (rs=−.06,p = .65)ortherelevanttasksscore(rs=−.03,p = .81).Hencetherewasnoevidencethat higher inhibitory control predicted more restrained generaliza-tionsorhigher ratesof rational choice in thedegree-of-competencecontrasts.Second, for thedomain-of-competencecontrast, thecom-binedrelevantmodelscorecorrelatedpositivelywiththeHTKSscore(rs=.27, p < .05).4 Thus, childrenwith advanced executive functionsshowedhigherdegreesofrationalchoiceinthedomain-of-competencecontrastthandidthosechildrenlowinexecutivefunctions.

3.2.5 | Preference questions

Children’smodelchoicesinthepreferencequestionswerecomparedto chance using χ2 tests (see Supporting InformationC for details).Childrenhadnosignificantpreferenceforthestrongortheaccuratemodelwhenasked‘whowillsharehertoys…?’(χ2(1)=.82,p = .37)or‘who isnice…?’ (χ2(1)=0.5,p = .48).Thus thepreferencequestionsindicatenogeneralpreferenceforoneofthemodels.

4  | DISCUSSION

Theaimofthepresentstudywastoexplorethecognitivefoundationsofchildren’sselectivelearning.Inparticular,wetestedthegeneralidea

F IGURE  3 Children’smodelpreferencesseparatedbytraitunderstanding:Forthedegree-of-competencecontrasts,Figure3adepictschildren’srelevanttasksscores(positivevaluesindicatingchildren’spreferenceforthemorecompetentmodelintasksrelevanttothemodel’scompetence)andirrelevanttasksscores(positivevaluesindicatingchildren’spreferenceforthemorecompetentmodelintasksirrelevanttothemodel’scompetence)separatedbytheirtraitunderstanding.Figure3bdepictschildren’scombinedrelevantmodelscore(positivevaluesindicatingapreferenceforthemodelwiththeexpertiserelevanttothetask,overthemodelwiththeexpertiseirrelevanttothetask).Thelineat0reflectschancelevel.Asterisksshowdeviationfromchancelevel(Wilcoxonsigned-ranktests,**p < .0017[Bonferroni-correctedα =.01for6tests]).Errorbarsshowstandarderrors

10 of 12  |     HERMES Et al.

putforwardbyrationalcompetenceaccountsthatchildren’sselectivelearninggenerallybuildsonacapacityforrationalinductiveinference(Sobel&Kushnir,2013) thatmaygetmaskedundercertaincircum-stancesby limitingperformance factors or taskdemands.Themainfocuswasonthesystematicvariationoftwosuchpotential limitingperformance factors.Thefirstwas the availability/ lack of requisitebackgroundknowledge,hereconceptualizedas traitknowledgeandassessed through direct trait questions.The idea is that only thosechildrenwithsufficienttraitknowledgeshouldbeabletorecruitmod-elsrationallyandrefrainfromglobalimpressionformation.Asasecondfactorthatpotentiallylimitschildren’suseoftheirrationalstrategies,wesystematicallyvariedthestructureoftasksandcomparedtwodif-ferentkindsofmodelcontraststhatdifferinthedegreetowhichtheyencouragesimplerstrategiessuchasglobal impressionformation.Inthe degree-of-competence contrasts, inwhich twomodels differedin overall competence in one domain, global impression formationwouldhaveyieldeduniquedeterminate answers. In thedomain-of-competencecontrastwheretwomodelsshowedthesamedegreeofcompetenceyetindifferentdomains,however,globalimpressionfor-mationwouldnothaveyieldedauniquedeterminateanswer.

Themainfindingswerethefollowing.Firstofall,traitknowledgedidplayakeyrole:When,inthedomain-of-competencecontrast,childrenwereconfrontedwithastrongandaknowledgeablemodel(whodifferedindomain,butnotinoveralldegreeofcompetence),onlythosechildrenwhocorrectlyidentifiedthesmartandthestrongmodelintheiranswerstotraitquestionsshowedrationalpatternsofmodelchoice:theypre-ferredthestrongmodelforthestrengthtasksandtheknowledgeablemodelfortheknowledgetasks.Incontrast,thosechildrenwhodidnotcorrectlyattributetraitschosemodelsatchance.Thesefindingsclearlyspeakinfavorofthegeneralideaofrationalcompetenceaccounts.

Thesecondmainfindingwasthattraitknowledge,althoughpresum-ablynecessary,wasnotsufficientforrationalinductiveanswerpatterns.Rather, taskstructureplayedanadditionalkey role.Whenpresentedwithmodelsthatdifferedincompetencewithinonedomain,children(regardless of their correct or incorrect trait attributions) recruitedthemore competentmodel for all tasks abovechance– irrespectiveofwhetherdomainsofcompetenceandtherequirementsoftasksdidordidnotmatch.Notethatthesewidegeneralizationsdonotreflectageneralhaloeffectsincethesamechildrenwhogeneralizedwidelywhenmodelsdifferedinoverallcompetencechoserationallywhentwomodelswerebothhighlycompetent,butindifferentdomains.Neitherdo thesewidegeneralizations reflect children’s general reluctance toprovideindifferenceanswers,aschildrenfrequentlyexpressedindiffer-enceintheReducedSuggestibilityCondition,inparticularinresponseto thepreferencequestions.Thesefindings are in linewithpreviousstudiesinwhichchildrenshowedwidegeneralizationswhentherewasadiscrepancy inoverallcompetencebetweenmodels (Bascandziev&Harris,2014;Brosseau-Liard&Birch,2010;Fusaroetal.,2011).5

Inmoreexploratoryfashion,twofurtherpotentiallimitingfactorsinchildren’searlyperformancewere investigated, thepragmaticsofsuggestivequestioning and executive function.Concerning the for-mer,childrenintheReducedSuggestibilityConditionprovidedmoreindifferenceanswersthanthoseintheStandardQuestionsCondition,

yetthisdifferencewasunsystematicwithregardtochildren’saggre-gated model preferences:When models differed in overall compe-tence, children did notmake lesswide generalizationswhen askedless suggestively – showing that thewide generalizations betweendomainsfoundinmanyselectivetruststudiesarenotjustanartifactofthequestionformat.Finally,executivefunctionwasexpectedtoberelatedtorational inductiveanswerpatterns (Jaswaletal.,2014). Inthepresentstudy,however,suchrelationswerenotverystrong:Highlevelsofexecutivefunctiondidgoalongwithmorerationalchoiceincaseswhere the twomodelswerebothcompetent,but indifferentdomains(domain-of-competencecontrast),yetitdidnotpredictlesswidegeneralizationswhenmodelsdifferedincompetenceinthesamedomain(degree-of-competencecontrasts).

Allinall,thus,thepresentfindingsspeakinfavorofgeneralratio-nalcompetenceaccountsofselectivelearning:Childrenarecapable,in principle, of rational inductive inferences for selective learning,yetthiscapacityissometimesmaskedbyperformancefactors.Moregenerally,thepresentresultssuggestthatyoungchildrenhavebothrationalandglobalprocessesattheirdisposal,yetusethemindif-ferent typesof tasks.So,whatexactlymight therelationbetweenthe two kinds of processes be?One interesting, yet at this stageclearly speculativepossibility, tobeexplored in future research, isthatsomekindofdual-processtheorymightbestdescribetherela-tionofsimplerprocesses,suchasglobalimpressionformation,andmorerationalprocesses,suchastrait-basedinductiveinferences,intherealmofselective learning(seealsoHeyes,2016).Despitetheconsiderabledifferences indetails invariousdual-process theoriesin cognitive, social and developmental psychology, a common as-sumptionunderlyingmostoftheseaccountsisthefollowing(Evans,2008):TypeIprocessesarefast,operaterelativelyinflexibly,basedonheuristicandassociative(ratherthaninferential)processes,andare implicit, unconscious and independent from general cognitiveresources.TypeIIprocesses,incontrast,arerelativelyslow,operateincontrolledandflexibleways,areexplicitandconscious,andde-pendentupongeneral cognitive resources.Thepresentdata seemcompatiblewithsuchadual-processpictureinseveralrespects:firstofall, childrenmayhaveusedsimplerheuristicprocessesasade-fault in tasks inwhich theyyielded determinate answers (degree-of-competencecontrasts),yet intasks inwhichsuchprocessesdidnot provide definite solutions (domain-of-competence contrast)children’smore rationalType II-like processes intervened. Second,children’s rationalperformance in thedomain-of-competencecon-trast(butnottheirreasoningbasedonglobalimpressionformationinthedegree-of-competencecontrasts)showedsomesignaturesofTypeIIprocesses:children’srationalanswersdependedupontheirbackground knowledge (indicated in the answers to explicit traitquestions),weremorefrequentinolderchildrenandshowedsomerelationtoexecutivefunction.Butofcourse,thepresentdatatakenby themselves are compatible with rather than anything close toproperevidenceforadual-processaccountofchildren’searlyselec-tivelearning.Nonetheless,suchtheoriesmaypresentaninterestingandproductiveframeworkforfutureresearchonthecognitivefoun-dationsanddevelopmentofselectivelearning.

     |  11 of 12HERMES Et al.

NOTES1Sincethesesumscoreswerenotnormallydistributedweusednonparamet-ricstatisticalanalyses.

2Thesepatternsofresultsforthedegree-of-competencecontrastsandthedomain-of-competencecontrastwerereplicatedwithtrial-wiseχ2 analyses forallsingletesttrials(withanexceptioninonetrialonly,fordetailsseeSupportingInformationC).

3FivechildrendidnotcompletetheHTKStaskandfiveadditionalchildrenwereexcludedfromthisanalysisbecausetheyhadproblemsunderstandingtheinstructionsandreceivedavalueofzero.Ifthelatterfivechildrenareincluded,theresultsdonotchange.

4Notethat thiscorrelationfailedtoreachsignificancewhenchildrenwhomademistakesinthereminderquestionswereexcludedfromanalysis(re-mainingsample:n = 59),rs=.25,p = .065

5 InFusaroetal. (2011),childrenshowedlocaldifferencesintheirpatternsof generalization: theywere presentedwith either a pair ofmodels thatdifferedinstrength(strong/weak)orinaccuracyinlabelingobjects(accu-rate/inaccurate). Insubsequenttesttrials,childrenpreferredtheaccuratemodelinlabelingtasksbutnotstrengthtasks,whereastheypreferredthestrongpuppetinbothlabelingandstrengthtasks.Thecurrentstudydidnotreplicatesuch localdifferences ingeneralizationsbetweendomains.Onepossiblereason isthat inFusaroetal. (2011)therewasaconfoundsuchthatthestrengthmanipulationwasmorepowerfulthantheknowledgema-nipulation:theweakpuppetwasnotonlyweakbutalsoinaccurate,falselyannouncing ‘I will lift this’ whereas in the knowledge condition, modelsdifferedonlyinaccuracy.Inthecurrentstudy,thecontrastsbetweenthetwomodels inthestrengthandtheknowledgedemonstrationswerede-confoundedandmorecarefullymatched,withthecompetentmodelscoringaboveandtheincompetentscoringbelowthechild’slevelofcompetenceonjustonedimension.

ACKNOWLEDGEMENTS

Thanks toMarlenKaufmannandKonstanzeSchirmer forhelpwithrecruitingchildrenandorganizingthestudy.Wealsoliketothankallthechildrenwhoparticipatedaswellastheirparentsandpreschoolteachers.ThisworkwassupportedbytheGermanScienceFoundation(DFG),grantRa-2155/1-1.

REFERENCES

Bascandziev, I., & Harris, P.L. (2014). In beauty we trust: Children pre-fer information from more attractive informants. British Journal of Developmental Psychology,32,94–99.

Bernard,S.,Proust,J.,&Clément,F.(2015).Four-tosix-year-oldchildren’ssensitivitytoreliabilityversusconsensusintheendorsementofobjectlabels. Child Development,86,1112–1124.

Brosseau-Liard,P.E.,&Birch,S.A.J. (2010). ‘Ibetyouknowmoreandarenicer too!’:Whatchildren infer fromothers’ accuracy.Developmental Science,13,772–778.

Cain, K.M., Heyman, G.D., & Walker, M.E. (1997). Preschoolers’ abilityto make dispositional predictionswithin and across domains. Social Development,6,53–75.

Ceci,S.J.,&Bruck,M.(1993).Suggestibilityofthechildwitness:Ahistoricalreviewandsynthesis.Psychological Bulletin,113,403–439.

Clément,F.(2010).Totrustornottotrust?Children’ssocialepistemology.Review of Philosophy and Psychology,1,531–549.

Corriveau, K., & Harris, P.L. (2009). Choosing your informant:Weighingfamiliarity and recent accuracy. Developmental Science, 12, 426–437.

Corriveau,K.H.,Kinzler,K.D.,&Harris,P.L.(2013).Accuracytrumpsaccentinchildren’sendorsementofobject labels.Developmental Psychology,49,470–479.

Evans, J.S.B.T. (2008).Dual-processing accounts of reasoning, judgment,andsocialcognition.Annual Review of Psychology,59,255–278.

Fusaro,M.,Corriveau,K.H.,&Harris,P.L.(2011).Thegood,thestrong,andtheaccurate:Preschoolers’evaluationsofinformantattributes.Journal of Experimental Child Psychology,110,561–574.

Harris,P.L.(2007).Trust.Developmental Science,10,135–138.Hermes,J.,Behne,T.,&Rakoczy,H.(2015).Theroleoftraitreasoninginyoung

children’sselectivetrust.Develpomental Psychology,51,1574–1587.Heyes, C. (2016).Who knows?metacognitive social learning strategies.

Trends in Cognitive Sciences,20,204–213.Jaswal,V.K., Croft,A.C., Setia,A.R., & Cole, C.A. (2010).Young children

have a specific, highly robust bias to trust testimony. Psychological Science,21,1541–1547.

Jaswal,V.K.,&Pérez-Edgar,K.(2014).Resolvingconflictsbetweenobser-vationandtestimony:Theroleof inhibitorycontrol. InE.J.Robinson&S.Einav(Eds.),Trust and skepticism: Children’s selective learning from testimony(pp.110–122).NewYork:PsychologyPress.

Jaswal,V.K.,Pérez-Edgar,K.,Kondrad,R.L.,Palmquist,C.M.,Cole,C.A.,&Cole,C.E.(2014).Can’tstopbelieving:Inhibitorycontrolandresistancetomisleadingtestimony.Developmental Science,17,965–976.

Kaufman, A.S., Kaufman, N.L., Melchers, P., & Preuss, U. (1994). K-ABC: Kaufman-Assessment Battery for Children. Amsterdam: Swets &Zeitlinger.

Kinzler,K.D.,Corriveau,K.H.,&Harris,P.L.(2011).Children’sselectivetrustinnative-accentedspeakers.Developmental Science,14,106–111.

Koenig, M.A., Clement, F., & Harris, P.L. (2004). Trust in testimony:Children’suseoftrueandfalsestatements.Psychological Science,15, 694–698.

Koenig,M.A.,&Jaswal,V.K.(2011).Characterizingchildren’sexpectationsabout expertise and incompetence: Halo or pitchfork effects? Child Development,82,1634–1647.

Kushnir,T.,Vredenburgh,C.,&Schneider,L.A. (2013). ‘Whocanhelpmefix this toy?’ The distinction between causal knowledge and wordknowledge guides preschoolers’ selective requests for information.Developmental Psychology,49,446–453.

Landrum,A.R.,Mills,C.M.,&Johnston,A.M.(2013).Whendochildrentrusttheexpert?Benevolenceinformationinfluenceschildren’strustmorethanexpertise.Developmental Science,16,622–638.

Li,Q.-G.,Heyman,G.D.,Xu,F.,&Lee,K. (2014).Youngchildren’suseofhonesty as a basis for selective trust. Journal of Experimental Child Psychology,117,59–72.

Lutz,D.J.,&Keil,F.C.(2002).Earlyunderstandingofthedivisionofcogni-tivelabor.Child Development,73,1073–1084.

McClelland, M.M., Cameron, C.E., Duncan, R., Bowles, R.P., Acock, A.C.,Miao,A., & Pratt,M.E. (2014). Predictors of early growth in academicachievement:Thehead-toes-knees-shoulderstask.Frontiers in Psychology,5,599.

Mills,C.M. (2013).Knowingwhentodoubt:Developingacriticalstancewhenlearningfromothers.Developmental Psychology,49,404–418.

Ponitz,C.E.C.,McClelland,M.M.,Jewkes,A.M.,Connor,C.M.,Farris,C.L.,&Morrison,F.J.(2008).Touchyourtoes!Developingadirectmeasureof behavioral regulation in early childhood. Early Childhood Research Quarterly,23,141–158.

Ponitz, C.C.,McClelland,M.M.,Matthews, J.S., &Morrison, F.J. (2009).Astructuredobservationofbehavioralself-regulationanditscontributiontokindergartenoutcomes.Developmental Psychology,45,605–619.

Robinson,E.J.,&Einav,S.(Eds.)(2014).Trust and skepticism: Children’s selec-tive learning from testimony.NewYork:PsychologyPress.

Sabbagh,M.A.,&Baldwin,D.A.(2001).Learningwordsfromknowledge-able versus ignorant speakers: Links between preschoolers’ theoryof mind and semantic development. Child Development, 72, 1054– 1070.

12 of 12  |     HERMES Et al.

Sobel,D.M.,&Corriveau,K.H.(2010).Childrenmonitorindividuals’exper-tiseforwordlearning.Child Development,81,669–679.

Sobel,D.M.,&Kushnir,T.(2013).Knowledgematters:Howchildreneval-uate the reliability of testimony as a process of rational inference.Psychological Review,120,779–797.

VanderBorght,M.,&Jaswal,V.K. (2009).Whoknowsbest?Preschoolerssometimes prefer child informants over adult informants. Infant and Child Development,18,61–71.

VanReet, J.,Green,K.F.,& Sobel,D.M. (2015). Preschoolers’ theory-of-mindknowledgeinfluenceswhomtheytrustaboutothers’theoriesofmind. Journal of Cognition and Development,16,471–491.

SUPPORTING INFORMATION

AdditionalSupportingInformationmaybefoundonlineinthesupport-inginformationtabforthisarticle.

How to cite this article:HermesJ,BehneT,BichAE,ThielertC,andRakoczyH.Children’sselectivetrustdecisions:rationalcompetenceandlimitingperformancefactors.Dev Sci. 2017; 00:e12527.https://doi.org/10.1111/desc.12527