In Wozniak,RobertH. (Ed), Fischer,Kurt W. (Ed), et al. (1993).Developmentin context:Acting and thinkin2 in specificenvironments.TheJeanPiaget The Dynam cs of

Hillsdale, NJ,USA: LawrenceEribaum Competence: How ContextContributes Directly to Skill

Kurt W. Fischer Daniel !H BullockHarvard University Boston University

Elaine J. Rotenberg Pamela RayaDenver, Colorado Harvard University

In psychologicaldevelopment,personand context collaborateto produce

actionandthought.Althoughthisstatementor somethingclosetoit is nowgenerallyacceptedby manydevelopmentalscholars,thefull implicationsof

the contributionsof context are still not widely appreciated.Contextdoesnot merely influence behavior.It is literally part of the behavior,partici-pating with the personto producean action or thought.

Psychologicalsystemsin generalarisefrom the collaborationof personandcontext.Most theoriesof psychologicalsystemstreatthe personas thesourceof thesystemsandrelegatethecontextto aminorrole. An especiallyclearcaseof this mistakeis the conceptof áompetence,which characterizesa person’sbestknowledge—theupperlimit of whatheor shecansay or do.A personis typically said topossessa certaincompetence,independentofits use in any context.

In this chapterwe show thatcompetenceis an emergentcharacteristicofa person-in-a-context,not of thepersonalone.Competencearisesfrom thecollaborationbetweenpersonandcontext,with competencechangingwhencontext changes.People are especiallyimportant in this collaboration,molding the contextto supportparticularkindsof actionsandthoughtsinthose they interact with. The effects of this sort of social support aredramatic,producingsharpshiftsin competencelevel in individual children.Competencerises abruptly with the provision of support and dropsdramaticallywhenthe supportis removed.

Theoriesof mind havegenerally sufferedfrom the fundamentalmistakeof focusingexplanationprimarily on eitherthe organismor the environ-mentastheprimarysourceof knowledgeor intelligence(Fischer& Bullock,

93

M FISCHER ST AL 4. DYNAMICS OF COMPETENCE 95

1984). Theoriesof competencehavebeenfundamentallyflawed by theirlocus on the organismand their failure to recognizethe contributionsof:ontext to competence.We suggesta different approachthat grounds:ompetencein the conceptof skiff, starting with the assumptionthat allbehaviorarisesfrom collaborationof personandcontext.Thedynamicsof:hangesin competenceareexplainedby analysisof developmentallevelsofskills as well as a neuralnetwork model.

THE FAILURE OF COMPETENCE THEORIES

Cognitive scientistsandpsychometriciansoftenspeakof a person’scompe-tence,or ability, asif the personpossesseda fixed capacityanalogousto theamount of liquid that canbeplaced in a glass.Whatevercontextthechildis in, the competenceremainsthesame,accordingto this view. Variationsin performanceacrosscontextandageprovidea seriousproblem for suchtheories.

The most extreme versions of such competencetheorieshave proveduntenable.Chomsky(1965) treatedthechild’s languagecompetenceasfixedfrom earlyinfancy throughthe operationof aninnatelanguageacquisitiondevice. To explain the vast developmentalchangesthat researchershavedocumentedin languageandcognition, he and hisstudentsimputebiolog-ical constraintsto the child that somehowinterferewith the Chomskiancompetence,preventingit from becomingfully evident in behavioruntillateryears.Building on thisanalysis,neonativistshaverepeatedlysearchedfor someearlybehaviorthat relatesto a “competence”and then neglectedto analyzehow the purportedcompetencedevelopsor how it is affectedbycontext(Fischer& Bidell, 1991).

Another exampleof an extremecompetencetheory is Piaget’s hypothe-sized epistemicsubject, defined as a knower uninfluenced by context,analogousto a moving object in a perfect vacuum, where there is noresistancefrom other objects, events, or energy fields. Although Piaget(1936/1952)wasone of the earlyvoicescalling for an approachintegratingorganismic and environmentalinfluences,he built a theory that focusedprimarily on the child and neglectedthe environmentand the bothersomedevelopmentaldecalagethat it produced(Beilin, 1971; Broughton, 1981;Piaget, 1971). In the lastyears of his life, however,Piaget recognizedtheproblemswith his earlier view and outlined a different view giving a moreimportantrole to context (Piaget1981.-1983/1987).

As developmentalists,most of us havetakenapathsimilar to Piaget’s,rejecting extreme competencetheories. In their place, competence/performancemodelshavebeenproposed,providingmore moderatechar-acterizationsof competence(e.g., Flavell & Wohlwill, 1969; Klahr &

Wallace, 1976; Overton& Newman, 1982; Pascual-Leone,1970). For agiven domainandage, thechild is consideredto havea fixed competence,asreflectedin hisor herhigheststageof performance.Variation belowthishigheststageoccurscommonlyandis attributed to factorslike effort andtaskdifficulty thatimpededemonstrationof thetruecompetence.As in theextremecompetencetheories,the person’scompetenceis fixed at any onetime, like the capacityof a glass to hold water. But unlike in the extremetheories,asetof processesarespecifiedby which thecompetenceeventuatesin performance—waysthat the personactivatesand utilizes the compe-tence.Justas theglasscanbe half empty,peoplemayonly useaportionoftheir competenceat any moment.When all the performancefactors arecontrolled,peoplewill show their true competence,the realupperlimit ontheir performance.

Like their predecessors,thesetheories fail becausethey segregatetheorganism from the environment, locating most organismic factors incompetenceand most environmentalones in performance.This funda-mental error not only fails to recognizethe collaborationof personwithcontext,but it alsoinsulatesthe theoryfrom test,The performancefactorsin the theoryinterferebothwith the expressionof competenceandwith thetesting of the theory of competence.When findings do not support aprediction, they are interpreted as reflecting some performancefactorrather than requiring a revision of the theory of competence.Like thePtolemaicview that the starsandplanetscircle the earth, the competencetheory is savedby post hoc epicycles in the performancecomponenttomaintain the perfectspheresof competence.The framework proposedinthis chaptereliminatesthe segregationof organismfrom environmentandmakescompetenceadirectly observablecharacteristicof individual people-in-context.

TAKING CONTEXT SERIOUSLY: THE ECOLOGY OF MIND

In recent years, there have been many calls for giving environmentorcontext a more active role in explaining cognition and development.Bronfenbrenner(1979),Neisser(1976),J. J.Gibson(1979), andothershavecalled for an ecologicalapproach.For cognitivedevelopment,theworks ofVygotskyandGibsonhavebeenespeciallypowerful in leadinginvestigatorsto analyzethecontributionof environment.Vygotsky(1978)focusedon thesocial environment—howother peoplecontributeto children’s cognitivedevelopmentandhow children internalizethese social influences.Gibson(1979)emphasizedthat the perceptualinputs for peoplein specificenviron-ments,calledaffordances,are richlystructuredandthat peoplecandetectandusethem without theneedfor complexinternal,mental construction.

96 FISCHER ET AL4. DYNAMICS OF COMPETENCE 97

Many voices argue currently that context is actually a part of people’saction,perception,thought,and knowledge(e.g., Cole & Scribner, 1974;Magnusson,1988; Rogoff & Lave, 1984; chaptersby Bronfenbrenner,Meacham,Reed,Rogoff, Wozniak,this volume).

Theseviews requirea radicalrestructuringof developmentaltheories.Itis not enough to recognizethe importance of experiencein cognitivefunctioning. Of course, people need to experiencea specific context tomasterskills in it or to detect affordancesin it. A mechanicwho hasmasteredthe repair of a Toyota four.’cylinder enginewill typically havedifficulty when first facedwith the fancy engineof a Porsche.Likewise, apersonwho hasgrown up in Peoriawill often havedifficulty making senseof a myth told by African hunter—gatherers.Examples like the automechanicand the person from Peoria are frequently cited to supportcontextualism,but they are not convincingto skepticsbecausetheyare tooobvious,showingonly aglobal effect of experience.Virtually anyframe-work that allows for the effects of experiencewill predicteffectslike these,including competence/performancetheories.

What is neededinsteadis analysisof the dynamiceffectsof context onskill. For competencein development,this analysis predictsthat contextaffectsthedevelopmentallevelorstageofa person’scompetenceevenwhenthe effectsof experienceand domain are controlledfor. We describeresearchthat shows powerful effects in which for a narrowly specifieddomain,a person’sdevelopmentallevel variesdramaticallyas a functionofcontextualsupport.Thiseffect is sopowerfulthat aperson’scompetenceorability can no longer be treatedas a fixed characteristicof the personindependentof context.

SKILL: COLLABORATION BETWEEN PERSONAND CONTEXT

The conceptof skill is a good startingpoint for the integrationof personwith context (Bruner, 1982; Fischer, 1980), In ordinaryEnglish usage,itimplies both personand context simultaneously.Peoplehavea skill forriding a bicycle, a skill for listening to their friends, a skill for repairingToyotaengines,askill for doing analysisof variance.A personcannothavea skill independentof a context. Skill requiresa collaboration betweenpersonand context.

This conceptionmeansthat skills vary not only betweenpeoplebut alsoacrosscontextsfor a given person(Fischer& Farrar, 1987). When a manborrowssomeoneelse’s bicycleandrides it or rideshis familiar bicycle onanunfamiliarkind of terrain (say,acrossagrassyfield insteadof onaroador sidewalk), he must adapthis skill to the contextof the new bicycle or

terrain. Hecannotimmediatelyride skillfully by usingtheskill he possessesfrom before.Heinitially rides awkwardly,working to adapttheold skill tothe new bicycle or terrain. Similarly, when a womanattemptsto performanalysisof variancewith a differencecomputerprogramor whenshetriesto analyzethe datain astudywith an unfamiliardesign,shehasto work toadapther skill. It cantakedaysor weeksof hardwork to generalizetheskillto the new context.

Noticethat the skill conceptincludesthepersonaswell as the context.Itis as much a mistaketo leave out the personas to leaveout the context(Fischer& Bullock, 1984). Skills are characteristicsof persons-in-contexts.

Theconceptof skill providesa foundationfor building a theory of howpersonandenvironmentcollaborateto producecompetence.Skill replacesthe organismicdefinitionof competencewith theradicalideathatcapacitiesliterally arisefrom the collaborationof personwith context.A major goalof theory andresearchthen becomesfinding principles that specify howperson and context collaborate to producecompetences.Empirically,competenceis definedmost simplyas an upperlimit on the developmentallevel of behavior.Our researchshowsthat behaviorshowsnot one upperlimit but different limits as a functionof context.

HOW CONTEXTUAL SUPPORT DIRECTLY AFFECTSDEVELOPMENTAL LEVEL

A persondoesnot haveasingle developmentallevel, evenwhenassessmentis limited to a specific domain. Level varies systematicallyboth acrosspeopleandacrosscontextswithin the domain.For example,within a fewminutes a 7-year-old child will demonstrate,in Piagetianterminology,concreteoperationalthinkingashis or herbestperformancein onecontextandthenpreoperationalthinkingas hisor her bestperformancein aslightlydifferent context.

Thedomainsin ourresearchwerehighlyspecific.In oneseriesof studies,we assessedindividual children acting out and telling pretendstoriesinwhich they maderealisticdolls actnice and/ormeanwith eachother.Allassessmentcontextsinvolvedthesamesetting,toys,andcontents,thesameexperimenter,andsimilar procedures.Another domaininvolvedindividualchildrensortingblocksinto boxesforming classificationmatricesbasedoncolor, shape,andsize, againwith eachcontext involving the samesetting,toys, contents, experimenter,and procedures.Yet anotherdomain (de-scribedin a later section)involved adolescentsand adults explaininghowthey madedecisionsaboutcomplexknowledgedilemmas.

Within eachdomain,thecontextsvariedprimarily in termsof thedegreeandtypeof social supportthat the”experimenterprovidedfor the task. In

98 FISCHER ET AL. 4, DYNAMICS OF COMPETENCE 99

low supportcontexts,heor shesimply askeda child to actout somemeanandnicestoriesor to sort someblocksinto boxes.In highsupportcontexts,he or sheprovidedexplicit support for aparticularbehavior—forexample,modeling a specific story or away of sortingblocks.

The understandingof meanandnicesocialinteractionswasmeasuredonthe multistepdevelopmentalsequenceshownin Table 4.1, which capturesdevelopmentbetweenapproximately2 and 15 years of age. For example,for Step3, thestory involved one-dimensionalsocial influenceor reciproc-ity: One doll acted mean(or nice) to a second,andthe secondone actedmean(or nice) in return becauseof the first one’s meanness(or niceness).The steps in Table 4.1 were specified in terms of the cognitive-developmentallevels andtransformationsof skill theory,which wasusedtopredictthe sequence(Fischer,Hand,Watson,Van Parys,& Tucker,1984;Hand, 1982; Rotenberg,1988).The sequencewas testedvia thestatisticsofscalogramanalysis, and in several studiesit formed a virtually perfectOuttmanscale.

Despitethenarrownessof the domain,theindividual child’s competencevaried dramaticallywith assessmentcontext. Competencewas definedasthe upperlimit on the child’s performance,his or her higheststep. In onetypeof context, as shownin Fig. 4.1, a typical7-year-oldproducedstoriesat the upperlimit of Step3, as well as at lower steps.In anothertype ofcontext,that samechild produceda story at the upper limit of Step 6, aswell as at lower steps.For both types of contexts, the datafit the basicempiricalcriterion for competence:Behavioroftenvariedbelowthehigheststep (3 and6, respectively),but it did not exceedthat step.

We havereplicatedthis phenomenonacrossa scoreof studiesof storiesand classificationinvolving hundredsof middle-classU.S. girls and boysbetween3 and 18 yearsof age(Elmendorf,in press;Fischer& Elmendorf,1986; Fischeret al., 1984; Fischer,Shaver,& Carnochan,1990;Lamborn&Fischer, 1988; Rose,1990; Woo, 1990).The storieshaveincludednot onlythe domain of nice and meaninteractions,but also various other social

domains, including social roles (such as doctor—patient,mother—father-—child, boy—girl, and child—adult), attributions about aggression,andperspectivetaking. The phenomenonhas also replicated for several non-story tasks,including classificationof blocks.

In theclassificationresearch,adevelopmentalscalefor classificationthatwas generallysimilar to that in Table 4.1 wasusedwith childrenbetweenIand7 yearsof age(Fischer& Bidell, 1991; Fischer& Roberts,1991). Whenchildren were repeatedlytestedin low and high support contextsover a2-monthperiod,theyshoweddifferentcompetences(upperlimits), with thelow supportcontextconsistentlyevokinga competenceseveralstepslowerthanthe highsupportcontext.

In summary, the researchshowed that in diverse domains, childrendemonstratedtwo very different competences,which we havecalled theirfunctionalandoptimal levels.Thesetwo competencesweretied to differentkinds of social-contextualsupport: Low support contexts allowing rela-tively spontaneous behavior produced functional-level competence,whereashigh supportcontextspriming morecomplexbehaviorproducedoptimal-level competence.The interval betweenthe two levels is called achild’s developmentalrange(Lamborn& Fischer,1988).

Spontaneous Contexts and Functional Level

In severalkinds of spontaneouscontexts,children showedthe sameupperlimit—their functional level. For the meanandnice story tasks,individualchildren acted out or told storiesspontaneouslyin two different contextsafter theyhadseenanadult act out a seriesof storiesaboutmeanandniceinteractions. In onecontext, calledfreeplay, the adult askedthe child tomakeup somestoriesof herown while the adultwent awayto do somethingelse for several minutes.In the other context,called beststory, the adultreturnedandaskedthe child to show the beststory shecould.

Children’supperlimit wasthe samein both spontaneouscontexts.Duringfree play, theyproducedseveralstoriesthat rangedfrom the upperlimitdown to lower stepsin the sequence.The 7-year-oldin Fig. 4.1 showedatleastonestoryatStep3 andseveralotherstoriesatSteps 1 and/or2. In thebest-storycontext, children produceda single story, and it was almostalwaysat thehigheststepshownin freeplay. For the7-year-oldin Fig. 4.1,the best.storywas at Step 3.

To testwhetherthe children wereindeedshowing atrue upperlimit, weintroducedseveralproceduresthat could reasonablybe expectedto inducehigherperformance.The best-storycontextitself was onesuchcheck, andit supportedthe competencehypothesis:When askedto give the beststorytheycould, 80Q/o—100%of the childrenproducedthesamehigheststepasinfreeplay, andmostof theremainingchildren were within one stepof that

Level StepRpl 1

2Rp2 3 FUNCTIONAL LEVEL (low support)

45

Rp3 6 OPTIMAL LEVEL (highsupport)7

Rp4/A1 8A2 9

FIG. 4.1 Developmental rangeof a 7-year-old.

-s

TABLE 4.1A Developmental Sequence tor Understanding Mean and Nice Social Interactions

Level Step Skill Examples

Rpl: Single representations I

Rp2: Representationalmappings

Rp3: Representationalsystems

Active agent:A personperformsat leastonebehaviorfitting asocial-interactioncategoryof meanor nice.

2 Behavioralcategory: A personperformsat leasttwobehaviorsfitting aninteractioncategoryof meanornice.

3 One-dimensionalsocial influence:Themeanbehav-ionsof onepersonproducereciprocalmeanbehaviorsin asecondperson.Thesamecontingencycan occurfor nicebehaviors.

4 One-dimensionalsocial influencewith threecharac-ters behavingin similar ways:SameasStep3, butwith threepeopleinteractingreciprocallyin amean•way(ora niceway).

5* One-dimensionalsocialinfluencewith threecharac-ters behavingin oppositeways. Thenice behaviorsofonepersonandthemeanbehaviorsof a secondpersonproducereciprocalnice andmeanbehaviorsin thethird person.

6 Two-dimensionalsocial influence:Two peoplein-teractin ways fitting oppositecategories,suchthatthefirst oneacts both nice andmean,and thesecondonerespondswith reciprocalbehaviorsin thesamecategories.

Child pretendsthat one doll hits anotherdoll(“mean”) or gives anotherdoll candy(“nice”).

Child hasonedoll actnice to anotherdoll, giving itcandyandsaying,~‘tlike you.” Theseconddoll canbe passive.Child hasonedoll saymeanthingsandhit anotherdoll, who respondsby hitting andstatingdislike forthefirst one, The secondone’sbehavioris clearlyproducedby thefirst one’s behavior.With threedolls, child hasoneteasetheothers,whileasecondonehits the others.Thethird doll rejectsboth of thefirst two becausetheyaremean.

With threedolls, child hasoneact friendly to others,while asecondonehitsothers.Thethird dollrespondsnicely to thefirst doll andmeanlyto thesecond.

Child hasonedoll initiate friendshipwith aseconddoll but in ameanway. Thesecondone,confusedaboutthediscrepancy,declinesthefriendshipbe-causeof themeanness.Thefirst then apologizesandmakesanotherfriendly gesture,which thesecondonerespondsto accordingly.

C

Rp4/A1: Single

abstractions

Al: Abstractmappings

7 Two-dimensionalsocialinfluencewith threecharac-ters: SameasStep6 but with threepeopleinteractingreciprocallyaccordthgto oppositecategories.

8 Single abstractionintegrating oppositebehaviors:Two instancesof interactionsinvolving oppositebe-haviors takeplaceasin Step 6, andthe relationsbe-tween thetwo interactionsareexplainedin termsofsomegeneralabstraction,suchasthat intentionsmattermoreactions.

9 Relationof two abstractionsintegratingoppositebe-haviors:Two instancesof interactionsinvolving op-positebehaviorsareexplainedin termsof therela-tion of two abstractions,such asintentionandresponsibility:Peoplewho havea deceitful intentioncanbeforgiven if they takeresponsibilityin awaythatundoesthedeceit.

With threedolls, child hasonedoll act friendly to asecondone, while a third initiates play in a meanway. Theseconddoll actsfriendly to thefirst oneandrejectsthethird, pointingout thelatter’smean-ness.Thethird thenapologizesfor beingmean,whilethefirst one doessomethingnewthat is mean.Theseconddoll acceptsthethird one’sapologyandre-jectsthefirst one,pointing out thechangein his orherbehavior.With threecharacters,child hasoneact friendly to asecond,while athird initiates playin a meanway.Thesecondcharacterrespondsto eachaccordingly,but then learnsthat theniceonehadmeanintentionswhile themeanonehadniceintentions.The secondcharacterthen changeshis or her behaviorto eachtomatchtheir intentionsandexplainsthathe or shecaresmoreaboutpeople’s intentionsthantheir ac-tions.With threedolls, child hastwo of them act niceonthesurfaceto athird, both with theintention of de-ceiving him or her into doing their homework.Whenthedeceitis discoveredby the third character,thefirst onetakesresponsibilityfor thedeceit by admit.ting the intentionandre-establishinghis or herhon-esty.But thesecondonedoesnotshowsuchrespon-sibility. Thethird characresforgivesthefirst one,butnot the second,becauseheor shecaresaboutpeopletakingresponsibilityfor their deceitfulintentionandundoingthedeceit.

*Step S is transitionalbetweenLevelsRp2 andRp3. Apparentlyit canbe masteredat Level Rp2,but it is mucheasierto do at Level Rp3.

‘2 FISCHER ET AL 4. DYNAMICS OF COMPETENCE 103

nit. Whentherewas adifferencebetweenthetwo contexts,beststory wasually one step lower than freeplay. This differenceis predictablefromeasurernenterror, becausethe child had only one chanceto show thegheststep in the best-storycontextbut multiple chancesin the free-playntext.Otherchecksof the commonlimit in the two contextsincludedpractIceith the storiesandinstruction. Underboth circumstances,the phenom-LOfl replicated,with the samehigheststep continuingto obtainfor botheeplayandbeststory.When practiceor instructionproducedachangeintnctional level in free play, it typically producedthe samechangein bestory. Overall, however,the effects of practiceand instruction were onlyodest.Averagefunctionallevel improvedat mostonestepwhen childreneregiven both instructionandpractice(Rotenberg,1988).Taken alone, theseresultswould seemto supporta simple competenceew, becausefree-playand best-storycontextsproducedthe same uppermit evenwith repeatedassessment.However, the results for the secondvpe of assessmentcontext were dramatically different: The children’sompetenceincreasedsubstantiallyin high supportcontexts.

upportive Contexts and Optimal Leve’

~hetype of contextthat evokeda higherlevel involved immediatecontex-ual support for performance:An adult presentedkey information to thehild aboutastory,andthenthechild actedout or told astorybasedon thatnformation. Across two different contexts providing such contextualupport, children showedthe sameoptimal level. In one, called elicitedrn/tat/on, the adult actedout and explainedthe story in detail and thentsked the child to makeup a similar story. In the other, called memoryPrompt, the adult remindedthe child of the gist and key elementsof an~arlierstory and thenaskedthe child to show that story. Both of these:ontextspromptedkey elementsin the story.

Children showedthe sameupper limit in both contexts—theiroptimalevel. In both elicited-imitationand memory-promptcontexts, they cor-rectly producedall storiespresentedup to the upper limit and failed allstoriesbeyondthe limit. For the 7-year-oldin Fig. 4.1, the limit was Step6—a jump of threestepsabove the functional level. This level is calledoptimal becauseit is hypothesizedto reflect the best performancethatchildren can produceon their own. We havearguedelsewherethat it alsoshowsstagelikediscontinuitiesin development,whereasthefunctional levelshows nonstagelikecontinuouschange(Fischer & Pipp, 1984). But thatissueis not essentialto this argument.

Children showed consistent optimal levels across repeatedtrials. Instudieswhere children practicedthe stories repeatedly,therewas only a

modestincreasein step,averagingatmostonestepin Table4.1 (Rotenberg,1988). Individual differenceswereconsistentfrom trial to trial. Instructionalsocauseda small increasein performancebeyondthat of practice,anditwas similarly reliable. The occurrenceof optimal level was thus a stable,replicablephenomenonin individual children.

Figure4.2 showsresultsof oneof thestudiesof practiceandinstruction,in which children performedunderfour contexts—twosupportivecontexts(elicited imitationandmemoryprompt)andtwo spontaneouscontexts(freeplayandbeststory; Rotenberg,1988).Eight 7-year-olds,instructedin howto recall the gist of the story to help their performance,were testedon thestoriesfor Steps3, 5, 6, and7 in Table 4.1. They were assessedthreetimeswith the elicited-imitation, free-play, and best-storyconditionsand oncewith the memory-promptcondition. In the latter, memorypromptsweregiven for each story, with the prompt providing the gist of the story,includingafew keyactionsandobjects.In Fig. 4.2 theupperlimit is shownfor the third elicited-imitationassessment,the secondand third free-playandbest-storyassessments,andthe single memory-promptassessment.

The higheststepselicited by the two supportivecontextswere virtuallyidentical,andthoseelicitedby thetwo spontaneouscontextswerelower andvirtually identical. In elicited imitation, the storieswere at Step6. In freeplay andbeststory, theydroppedprecipitouslyto Step3. Then,thememorypromptsweregiven, andthestoriesagainroseto Step6, whichwasidentical

2

:iS

4

2

1.

3 2 2 I 3 3

ELICIMrt FREEPLAY EESTSTORY PEDMfl FtQEPLAY ,,srs,ogy

CONDn1ONS AND TRiALS

FIG. 4.2 Reliability of highest step for understanding nice and mean under high andlow support conditions with repeated assessments (Note: Elicited imitation and promptconditions provided high support. Vree play and best story conditions provided lowsupport. Numbers on x-axis indicate the repetition of the condition that is graphed).

104 FISCHER ET AL 4. DYNAMICS OF COMPETENCE 105

to the resultsfor elicited imitation. Free-playandbest-storycontextswererepeated,andthe storiesagain felt to Step3. Every child showedthe samegeneral patternof change,with some variation in the individual child’soptimal andfunctional levels.

Clearly, the optimal- andfunctional-levelresultsarehighly replicable.Inthesestudies children showeddistinct competencesfor spontaneousandsupportivecontexts.Their competencesarosefrom the dynamicinterplayof personwith context, changingin a matter of minutes as the contextchanged.A child participatesin onecontext,producingaspecificlevel, andthen the child changesto participate in a different context, producing adifferentspecific level. Returnto the first contextproducesa return to theinitial level, andsoforth. This kind of effect canberepeatedtrial after trialfor each individual child.

DEVELOPMENTAL RANGE—WHERECOMPETENCES GROW

We proposethat functional and optimal levels define the developmentalrangeof adomain for achild — from skills thatthe child canproduceeasilyon his or her own to skills that the child can produceonly with strongcontextualsupport(Bidell & Fischer,1991). It is primarilywithin this rangethat short-termgrowth in skill occursandthat p!actice, instruction,andcontextualvariation havetheir effects.

Researchersor educatorswantingto assessa child’s understandingneedto think in termsof arange,not apointon ascale.And they needto alwaysconsidercontextas anintegralpartof anycompetencetheyassess.Contextincludesnotonly thedimensionof socialsupportbut alsoissuesof domain.Children’s functionalandoptimal levelsvarysubstantiallyacrossdomains.Failureto considerrangeand contextleadsto majorerrorsof assessment.

The developmentalrangeis relatedto Vygotsky’s (1978)conceptof thezone of proximal developmentas well as the associatedconcept ofscaffolding(Bruner, 1982; Wood, 1980). Like developmentalrange,theseVygotskianconceptsemphasizethat the child’s actionsvary over a rangecloselytied to developmentandare strongly affectedby the behaviorsofotherpeople.

Thereis at least oneimportant differencebetweendevelopmentalrangeandzoneof proximal development,however.In mostof the studiesof thezone,the adultactuallyintervenesin thetaskandperformspartof it for thechild. In ourresearch,on thecontrary,the adultdoesnot directly intervenein the performanceof the task. It is no surprisethat a child and an adulttogethercan performa taskbetterthan a child alone. It is moresurprisingthat the mereprovision of social contextualsupport strongly affects thechild’s soloperformance.In thesupportivecontext,an adultpromptsaskill

in the child andthendoes nothingmore: The child has no directaid fromthe adult. But evenwith the adult not doing any of the task,the child andthe supportivecontext collaborateto produceoptimal performance.Thechild truly demonstratesacompetencetoacton hisorherownwith support.

In addition, theresearchresultsshowwhatappearsto beacontradictionof theVygotskiananalysis.Thezoneof proximaldevelopmentinvolvesthechild’s gradualinternalization of interactionsbetweentwo people,oneofwhom is an adult or an accomplishedpeer (Vygotsky, 1978). As the childbecomesadult, he or she becomesable to control the structuresindividu-ally, without scaffolding. Thus, the zone gradually decreasesor evendisappearswith age.On the otherhand,the developmentalrangedoesnotshrink with agebutgrowslarger,asshownin Fig. 4.3 for the meanandnicestories(Hand, 1982). In infancyandearlychildhood,children’sfunctionallevel seemsto be closeto their optimal level, at least for familiar domains(Fischer & Hogan, 1989; Watson& Fischer, 1980). Startingat about3½years,the gap betweenfunctional and optimal level becomesstrong, andthereafterit seemsto grow everlargerwith age.

CONTEXTUAL SUPPORT OF HIGHER REASONING:REFLECTIVE JUDGMENT

The developmentalrangedoes not end with childhood but extendsintoadulthood.A study of the developmentalrangefor reflectivejudgment

3

AGE IN YEARS

FIG, 4.3 Highest step for understanding nice and mean under three conditions as afunction of age.

...,- EUCY~EDIMI-rATIONFREEPLAY

6

BESTSWaY

5.

4

2

— .E.

3 4-1/2 6 7-1/2 9 12

06 FISCHER ET AL. 4. DYNAMICS OF COMPETENCE 107

hows that developmentalrangegrowslargerat least throughthe late20s,.s people construct high-level abstract reasoningabout the basesofnowledge.Optimal andfunctionallevels occurat theseagestoo, andthelistancebetweenthem grows with age.

Reflectivejudgmentis reasoningaboutthe basesfor knowing, especiallyvhen dealingwith conflicting argumentsabouta complexissue.Kitchenertnd King (1981) devisedan interview for assessingreflectivejudgmentbytskingpeopleto dealwith dilemmasin whichatleasttwo opposingopinionswe statedaboutan issue.For example,oneof thedilemmasdealswith theiealth effects of chemical additivesto food: Do theseadditivespromote~iealthor causedisease?

Longitudinal researchhas shown that people’s judgments developthrough the sevenstagesshown in Table 4.2. Children start out with theview that knowing is a concretestatebasedon direct experience.At themiddle stages,they cometo understandthat knowledgedependson one’sviewpoint andis thereforeuncertain.In the later stages,theymovebeyondthe focus on uncertaintyandconsiderthe justificationandevidencefor aconclusionand the processof inquiry by which it was reached.Kitchenerand Fischer (1990) presenteda skill analysis of the stagesof reflectivejudgment.

To assessoptimal andfunctionallevels of reflectivejudgment,we askedstudentsto reasonaboutknowledgedilemmasin two contexts(Kitchener,Lynch, Fischer,& Wood, in press),The low supportcontextwas the tradi-tional ReflectiveJudgmentInterview(RJI), in whichthepersonis presentedwith aseriesof dilemmasandfor eachdilemmais askedto stateapositionSand to explain the basesfor it. The high support context was a newassessmentinterview, the PrototypicReflective JudgmentInterview (Kit-chener& Fischer,1990). Peoplewere presentedwith the samedilemmasasin the RJI, but contextualsupport was provided by presentationof aprototypicanswerfor each stageof eachdilemma. (Theseprototypeswerebasedon answersgiven by peoplein earlier studiesusingthe RJI.)After astudentreadoneof theprototypes,he or shewas askedto explain it in hisor her own words.

Subjectswere 104studentsbetween14 and28 yearsof age,halfmale andhalf female. Studentswere testedindividually in two sessions,with eachsessionincluding first the spontaneouscontext (RJI) and then the sup-portive context (PrototypicReflectiveJudgmentInterview). After the firstsession,studentswerealsogiven aseriesof questionsandguidelinesto helpthem think about drawing conclusionsabout complex issues before thesecondsession;thesematerialsdid not includedirect statementsabouttheactualdilemmas.

The resultsshoweda clearseparationof functional and optimal levels,with studentsperformingapproximatelya stagehigher in the supportive

TABLE 4.2Stages of Development of Reflective Judgment

Skill Level Stage of Reflective Judgement

Level Rpl: StageI:Single Singlecategoryfor knowing: To know meansto observedirectlyrepresentations without evaluation.Level Rp2: Stage2:Representational Two categoriesfor knowing: Peoplecan be right aboutwhat theymappings know, or they can be wrong.Level Rp3: Stage3:Representational Threecategoriesfor knowing: Peoplecanbe right aboutwhat theysystems know, or they canbe wrong,or knowledgemaybe incompleteor

temporarilyunavailable. The status of knowledgemaydiffer in dif-ferent areas.

Level Rp4/Al: Stage4:Systemsof Knowledgeis uncertain:The fact that knowledgeis unknownin sev-representational eralinstancesleadsto understandingknowledgeas an abstractpro-systems,which are cessthat is uncertain.single abstractionsLevel A2: Stage5:Abstract mappings Knowledgeis relativeto a contextor viewpoint; it is subjectto inter-

pretation.Thus it is uncertain in science, history, philosophy,etc.Conclusionsmustbe justified,

Level AS: Stage6:Abstract systems Knowledgeis uncertainandsubjectto interpretation,but it is pos-

sible to abstractsome justified conclusionsacrossdomains or view-points. Knowledge is an outcome of theseprocesses.

Level A4: Stage 7:Systems of abstract Knowledge occurs probabilistically via inquiry, which unifiesconceptssystems, which are of knowledge. Knowledge can be reached with various degrees of cer-principles tainty depending on justifications and evidence,

Note. Descriptions are adapted from Kitchener and Fischer (1990) and Kitchener and King(1981).

contextthanin the spontaneousone.Theseresultsheldover bothsessions.That is, producinghigher stageresponsesin the supportivecontextin thefirst sessionandhaving2 weeksto think aboutthe dilemmasdid not reducethe differencebetweenoptimal andfunctional levels, althoughtherewas asmalloverall increasein level betweensessions.

Consistentwith the previousfinding that developmentalrangeincreasedwith agein childhood,thedistancebetweenfunctionalandoptimal levelsofreflectivejudgmentgrewwith ageduringadolescenceandadulthoodtoo. Intheteenageyears,themeandifferencewas about.6 stages,but by the late20s it had grown to twice as much, 1.2 stages.The increasingsize of thedevelopmentalrangewith agethusseemsto extendwell beyondthe yearsofchildhood into at least the yearsof earlyadulthood.

The reflective-judgmentresults thus illustrate the generality of the

108 FISCHER ET AL 4. DYNAMICS OF COMPETENCE 109

developmentalrangeacrossdomainsand ages.Peopleshow onedevelop-mentallevel—onecompetence—whentheyactinaspontaneouscontextandamuchhigherdevelopmentallevel — adifferentcompetence— whentheyactin a socially supportivecontext.The differenceis remarkablyrobust.Notonly doesit occuracrossmanydomains,but practiceandsimpleinstructiondo not eliminateit.

DYNAMICS OF COMPETENCE IN A NEURAL NETWORKIN CONTEXT

The robustnessandgeneralityof the developmentalrangesuggestthat it isa basiccharacteristicof humancognition, apropertyof the way the humannervoussystem operatesin context. To begin to understandthe neuralfoundationsfor developmentalrange,we lookedto modern.neuralnetworktheory, especiallymodelsof parallel, distributed networksthat involve acollaborationbetweentop-downprocessesin thenetwork itself andbottom-up processesfrom input to the network,Adaptiveresonancetheory (ART)hasthesepropertiesandhasbeenusedwith successto modelmanycognitiveprocesses(Bullock, Carpenter,& Orossberg,1991; Grossberg,1980). ARTnetworkshave the capacityto learn and to function in many ways likeintelligent organisms,As parallel distributednetworks, they use a set ofinput to makegeneralizationsabouttheform of thatinput, oftenproducingsurprisesnot built into the original network. Building on theseproperties,they achievegreatpower through specifying particular, diversenetworkarchitectureslike thoseof humanneuralsystems.

DevelopmentalRange in Neural Networks

Within adaptiveresonancetheory,neural networks haveexactly the dy-namicpropertieswe anticipated:They generatenot merelyonecompetencebut a rangeof competencesaffectedpowerfully by input from the contextin which the network is functioning. Especially relevant to this develop-mentalrangeis the role of contextualinput in stimulatingcomplexneuralactivity thatis sustainablethroughshort-termmemoryprocesses.

The networks contain short-term memory componentsthat can beactivatedwithout beingdirectly encodedinto long-term memory. Indeed,complexneuralsystemsexhibitingshort-termmemory, long-termmemory,anddifferentiationof thetwo wouldseeminevitably to produceapropertylike developmentalrange.When the sho.rt-term memorycomponentsareactivatedby context, theyallow the consequencesof the transientcontex-tual input to persistwithout any input from long-term memory compo-nents.For a significant interval after the contextualinput, the network

exhibitsthis competence,but theability is fragile becauseit dependson theinducedshort-termmemory components,which are not subject to long-term memoryencodingatthe currentlevel of network maturity.Therefore,after intervening activities push the system into some other state, thenetwork cannotautonomouslyre-enterthe stateoriginally inducedby thecontext. The competenceassqciatedwith having enteredthe contextuallyinduced state is real, but the state itself cannot be regeneratedby thenetwork alonewithoutappropriatecontextualinput.

In general,whenbottom—upinput (like that coming from context)andtop—downinput (like that coming from individual goalsor plans) showanappropriatematch,theyproduceresonancein the circuit. When the matchis absentbecauseof the absenceof oneor theotherinput, thenetwork canstill function, but it functionsdifferently, in a less complexway. Thus asingle network can show oneorganizationwhen it is functioningwithoutmatchinginputs and a more complexorganizationwhenit is functioningwithbothcontextualinputandmatchingtop-downinput, suchasthatfromshort-termmemory.

The structureof this network is illustratedin a highly schematicway inFig 4.4. Theoutput processinvolving networksitesR1 and R2 producesasimpleactivity stateat anearlystageof developmentwhenit is activatedbysignalsalongpathwaysS~and~2• As developmentproceeds,this processisreorganizedby hierarchicalinputs from sitesC5 and C2 to sites R1 and R2alongpathways53 andS4. WhensitesC5 andC2areactivatedevenbriefly,theycanmaintaintheir activestateby virtueof theexcitatoryfeedbackloop

FIG. 4.4 A neural network thatshows the phenomena of develop-mental ranje.Key: Circles mark neural networksites, and arrows mark activationpathways.C and R designate hierarchally or-ganized sites, with C providinginput to It.The + sign indicates excitatoryfeedbackloops, whichsustainshort-term memory for a site.S designates pathways of activa.tion from one neural site toanother.I marks input from the context.M designates memory pathways.The diagram shows only the partof the network central to the text.The open circle at the top andactivation pathways 5, and ~2

show links to other parts of thenetwork. /

II

SI

110 FISCHER El AL 4. DYNAMICS OF COMPETENCE 111

(shown with the + sign to indicate its excitatorynature).That is, whencontextual inputs I~and 12 activate C, and C2, the pattern can bemaintainedin reverberatoryshort-termmemory and continueto enablereorganizedprocessingthrough sites R, andR2-- optimal level behavior.

However,whenthe systemis resetby, for example,a changein context,theloopwill be interrupted,andthe reorganizedprocessingwill cease.Nowall that the systemcansustainis the simpleractivity producedby signalsS,andS2—functionallevel behavior.To overcomethis functional-levellimit,thesystemmustbeableto generatethe morecomplexactivity throughsitesC, andC2 on its own without contextualinputs I, and12. Thisdevelopmentoccurswhenthe long-termmemorypathwaysM, andM2 to sitesC, andC2become functional. As further developmentand learning bring theselong-term memory pathways into operation, a child can permanentlyencodethe patterninducedat C1 andC2 in long-termmemory.As a result,the child becomesableto endogenouslyregeneratethe induced stateat alater time in the absenceof immediatecontextualsupport.Now, what waspreviouslyoptimal level becomesfunctionallevel: Behaviorthat previouslydependedon contextualsupport can now be producedspontaneously.

The partial independenceof short- and long-termmemory processesisonly onepart of the generalcomplexity and diversity of animal nervoussystems.A central nervoussystem(CNS) is madeof manycomponentsincompositestructure,including diverseneuralcircuits anddiverseinputs tothosecircuits. The componentsare not only parallel and distributedbutoften distinct in structure. Advanced brains are built up from manyseparatelocal circuits that operatein partial independenceof other localcircuits. Although all regions of the CNS are ultimately linked, partialindependenceis assuredby variability in linkagestrengths,occurrenceofboth cooperative(mutually excitatory) and competitive (mutually inhibi-tory) interactions, radical differencesamongnetworks in sensitivitiestoinputsof varioustypes,andhighly diverseinputs from theenvironmentandthe body. In addition, partial independenceis also assuredby develop-mental delays in effective interactionbetweenmany componentnetworks(Fischer & Rose,in press;Thatcher, 1991, in press),as in the delay indevelopmentof theM componentsin Fig. 4.4. Givenall the complexityandindependenceof components,it is inevitable that among the system’smultiple abtivity states, some will be dependenton specific contexts. Inaddition,someof thesecontext-dependentstateswill eventuallydevelopsothat a child cangeneratethem autonomously.

Advantagesof Multiple Levels of Competence

In time, children typically developthe capacity to evoke the complexactivity at sitesC, andC2 on their ownthroughlong-termmemorysitesM,

and M2. That is, after somekey experiences,a child becomescapableofautonomouslygeneratingthe developmentallyadvancedperformanceear-lier exhibited only transiently—acapacitythat Bullock, Carpenter,andGrossberg(1991) calledautonomoussupercessionof (endogenous)control.This capacityhas majoradvantages,of course,but therealsoseemto begood reasonsthat its developmentis delayed.

The advantagesof autonomouscontrol areclear.It is advantageoustobeableto re-entera statethat generatesan adaptivebehaviorwithout strongdependenceon exogenousinput suchas socialcontextualsupport.Insteadof a lengthy processof searchfor the supportivecontext to producethebehavior,the organismcandirectly generatethe desirablebehavior.Suchre-entryto desirablestates is a themein both Piagetiantheory, with itsemphasison circular reactionsandthe regenerationof sensorimotorstates(Piaget, 1936/1952;seealso Kaufmann, 1980), and conditioningtheory,with its emphasis on the regenerationof positively reinforcing states(Skinner, 1969). In fact, thereare many different kinds of autonomoussupercessionof control within andacrossspecies(Bullock, 1981;Bullock etal., 199!; Fischer& Bullock, 1984). Becauseof the emphasison autono-mous control, neural network theories of perceptualand motor skilllearninghavefocusedon showinghow moreendogenouslyactivatedinputpathwaysto component networkscan gracefullysupercedemore exoge-nously activatedinput pathways,as illustrated with the supercessionofcontrol by long-term memory in Fig. 4.4. (Note that in neural networkmodelsthe supercessioneffect is gradedratherthanall-or-none.)

Despiteall theseobvious advantages,children do not developautono-mouscontrol quickly in all domains. Instead,they developit slowly andhierarchically, with vast arenasof behavior requiring social contextualsupport for years beforechildrengain autonomouscontrol overthem. Ofcourse, it is these delays that produce the developmentalrange, thedifferencebetweenoptimal and functional levels. This aspectof develop-mentalrangehasnot beena major subjectof researchin neural networktheoryor in otherpartsof cognitive science.

The key questionsare:Whatdoesan organismhaveto gainby delayingthetimethata level of supercessionof control matures,andhow is the delayachieved?Of these two aspectsof the problem, how the delay could beachievedis easierto answer.Thefull functioning of neuralconnectionscanbe readily delayed by many processes,including slow myelination ofpathways.For example, if learnedsupercessionof control depends,as inFig. 4.4, on the correlatedactivation of sites C, and C2 with long-termmemorypathwaysM, and M2, which project to C1 and C2 from remotebrainregions,thensupercessioncanbepreventedby delayedmyelinationofpathwaysM, and M2prior to that time. Without myelin, signalswill betransmittedslowly andwith greatattenuationalongM, andM2. Theresult

112 FISCHER El AL. 4. DYNAMICS OF COMPETENCE 113

will be negligible long-term memory encodingof C, and C2 activations,even if the cells that give rise to pathwaysM, and M2 show functionalconnectionswithin the local circuit to C, and C2. In fact, myelinationis aslow developmentalprocessin humanbeings(Yakovlev & Lecours,1967),andthereare manyotherprocessesaswell thatdelaythe full functioningofneuralconnections(Thatcher,in press).

Questionsabout the adaptivevalue of such delays haveseldom beenasked by developmentalists.They seem to assumethat developmentaldelaysareexplainedin termsof intrinsic maturationalfactors,suchas theinherentdependenceof upperlevels of a hierarchyon prior developmentoflower levels.We havearguedsincethe early l980sthat timing of develop-mental transitionsarisesfrom aset of dynamicallyinteractingfactors,notmerelyfrom intrinsic maturation(Fischer,1980; Fischer& Bullock, 1981).

The advantagesof delaying autonomoussupercessionof control, wehypothesize,center on the relation betweenlevels of organizationin ahierarchy.Whenahigher level assumescontrol, thereis truncationof thesearchprocessat the lower level—thatis, reductionin the scopeof searchfor adaptivecombinationsfor generalizationat the lower level. Conse-quently,delayingsupercessionof control by level n + 1 prolongsthesearchfor new adaptivecombinationsat level n.

Delayingthecontrol of level n + I not only allows timeto find adaptivecombinationsat level n, but it reducesthe risk of finding inadequatecombinationsthere. The form that activity takes in a neural networkdependsvery much on the input it experiences—itssamplingbase. Withinsufficientexperienceat level n, poorgeneralizationscanbeformedthere.Delays in supercessionof control to a higher level will avoid powerfulgeneralizationsdrawn from insufficient sampling.

A self-organizinghierarchy can produce compactrepresentationsto.gether with great generativepower, but this potential requires that itsgeneralizationsbe well suitedto its task environments.The effectivenessofits generalizationsare directly relatedto thethoroughnessof its samplingoftask environments.Indeed, Elman (1992) showed this limitation in aparallel,distributednetworkthat learnedtospeakbasedon experience.Forthe networkto learn adequategeneralizationsabout lower levels of speechproduction,it requiredextensiveexperienceat a lower level beforemovingto ahigher level. Whenthenetworkwas not requiredto functionat a lowerlevel first, it missedimportantgeneralizations.Thereis a selectivebenefittoprolonginglower level samplingwell beyondthe minimum that is strictlynecessaryfor constructionof skills to begin at the next level.

Evenwhile thereis a disadvantageto truncatinglower level samplingtooearly, thereis also an advantageto being able to activate higher levelgeneralizationsthat havebeensuccessful.The developmental-rangephe-nomenonprovidesa way of havingboth advantagesat the sametime by

separatingthe two levels. A child can sustainhigher level generalizationswhen the context induces them, but the child can simultaneouslydelayhigher level control in order to haveextensiveopportunity for learningimportantgeneralizationsat the lower level. This kind of processhasbeenoutlinednot only for developmentbut also for multiple memorysystemsinprimates(Levine& Prueitt, 1989;Mishkin, Malamut,& Bachevalier,1984)and for alternativesubstratesfor learningin neuralnetworks (Grossberg,1978).

The coexistenceof lower level and higher level functioning in thedevelopmentalrange essentially separatessampling and generalizationprocesses.The collapsingof theseseparatedprocesseswould poseseriousproblemsfor a developingorganismdependenton learning. Someneuralnetworktheories,suchas backpropagationmodels(McClelland& Rumel-hart, 1986), havearchitecturesthat virtually collapsesamplingandgener-alization,andas a result the networksmust learnvery slowly in order toprevent premature,poor quality generalizations(Bullock & Grossberg,1990; Grossberg,1987; seealsoPrince& Pinker, 1988).

The separationof competencesevident in the phenomenaof develop-mentalrangethus makesensein termsof howneuralnetworksfunctionandin termsof the demandsof adaptationto a complexenvironment.

SUMMARY AND CONCLUSIONS: THE DYNAMICS OFCOMPETENCE

Acrossdomainsandages,contextcontributesdirectly to competence.Thatis, skill level is a characteristicnot only of a personbut alsoof a context.Peopledo nothavecompetencesindependentof context.

The phenomenonof developmentalrange shows one way that thisperson—environmentcollaborationworks. Immediatecontext contributesdirectly to skilil, affecting the developmentallevel of a child’s behavior.Byevoking specificskill components,context inducesa particularskill. Thiseffect is powerful, with performancevarying from momentto momentupand down a developmentalscaleas a function of degreeof contextualsupport for high-level functioning. When the supportchanges,the child’slevel changes.

Traditional conceptionsof competenceand performancefail becausethey treatcompetenceas a fixed characteristicof the child, analogousto abottle with a fixed capacity. Performancefactors are seenas somehowinterferingwith this capacity.The conceptof skill overcomestheselimita-tions by providing a dynamic framework for analyzing variations inbehaviorwith context.

Our research shows that children do indeed have stable levels of