The Relation Between Adaptive Behavior And Intelligence .

34interestjournal of School Psychologyis the diversityof ability;standardof the standardizationscoressample overlap,on the K-ABCMentalwhich includedProcessingall levelsComposite(M 100,SII 15) ranged from 47 to 153.instrumentationAll sub,jccts had been administeredinstruments’standarization.both the K-ABCAs documentedand the Vinclandelsewhereas a part of bothboth instrumentsare well stan-dardized,purposesand there is evidence to support their reliability and validity for a variety of(Kaufman& Kaufman,1983; Sparrow et al., 1984). Test scores on theVinelandand K-ABCwcrc used to create the following variablesfor USC in the presentanalyses.Ability.The Mentalof intellectualKeith & Dunbar.sure verbalProcessingability.and generalSpecifically,and Places,Riddles,or g factor (Keithintellectualability (Keith,Givenwas virtuallythis cvidencc,and Simultaneous1985; KaufmanProcessingformedstandardsubtests.researchof intellectualby Keithscore (Nonverbalent dimension,(e.g.,& Kaufman,AdaptivrpreliminaryBehaviorsupportfactors being similaridenticalFacesto a second-orderand the ample supportforscores as measuresSequential1984),Processingthe K-ABCofstandardSimultaneousand a Verbal ReasoningFaces and Places,for this three-index& Dunbar,and ive(Keith,achieuemmtandis based onTheremainingHood,Ebcrhart,1986: Keith & Dunbar,was measuredThese domainsCompositeto justify(Sparrowfor a second-orderin structureachicvea differ-8r Pottebaum,1984).’by means of three domainnamely, those of the Socialization,overlapK-ABCseem to representof such results is still being debated (cf. Kaufman,behaviordomains.1984).Understanding)in press; Keith,dard scores from the Vincland,sions but with enoughthey measurethat these three verbal tests may provide strong measuresthe interpretationand CommunicationReasoning),1985;may mca-1983), three indexes of intellec-scores on the K-ABCKeith& McLean,Behavior.Xr Dunbar,the rationaleDecodingnamely1985). Of course,1984: KaufmanAgain,suggestingabilityment tests (ReadingAdaptiveMemoryby a\reraging standardArithmeticpreviousVerbalKeith,from the K-ABCProcessingtual ability were used in the present study: the K-ABCscore (called1984;achievement subtestsfactor derivedsubtests1984).Sequentialcan be used as a measureJensen.as much as, or more than,Reasoningand Arithmetic(e.g.,that the K-ABCintelligencea Verbal& Dunbar,\,iewing the K-ABCfrom the K-ABCsome researchers1984) have suggestedabilityarhievcment.CompositeHowever,have been identifiedtheiret al.,adaptivebeingcombined1984,pp.behaviorto the Socialization,40,stan-Daily Living Skills,as separatedimen-into the Vineland’s46).Thus,thereis“factor,” with the first-orderDailyLivingSkills,and Com-‘.4 one rrvirwer correctly noted. usingsubtest compositesas the first step in the analyses; w mcs that the compositesadequatelyreflect the structure(or first-orderfactors) of thr tests M’rbrlic\c that such an assumption,based on thy research cited under Instrumentation,is warrantrdlc,r- both the Vinelandand the K-ARCFurthermore,even fairly substantialdrvi,ttionsin thctorstructurewould likely have only small ef’fecta on the magnitudeof the relation betwwn the secondorder I;tctors (adaptivebehaviorand ability)

Keith et al.municationchildrendomainsof the Vineland.under 6, was not included35The VinelandMotordomain,administeredtoin these analyses.AnalysisConfirmatoryfactor analysisate three hypothesesconfirmatory(LISREL)concerningfactor analysis,are specifiedin advance,tJ”oreskog & Sorbam,how adaptivethe numberbehaviorand compositionand the researcher’shypothesesby how well the model fits the data. In essence,1984) was used to evalu-and intelligenceare related.of factors (calledare supportedIna model)or questionedthe model is the researcher’stheory andthis theory is tested against the data.The basic model to be analyzedmodel in which an adaptiveSkills,Communication,factorand SocializationintellectualabilityReasoning,and Verbal Memorymake explicitin the study is shown in Figurebehaviorfactoris presumedthe assumptioncurvedis assumedscales of the Vineland.to explaincompositesfactorsscores.Thefactors,and it is this value that is manipulatedfirst analysis,adaptiveline (labeled4 or phi)and intellectualsecond analysis,phi is constrainedand intellectualabilityabilitythen comparedto those of the first analysisthirdphi is constrainedanalysis,notion that adaptiveReadersunfamiliarcovariancestructuresthe relationadaptivetest the relationbetweenonly measureapproximatemeasuresthose constructs.bles are then studied,made of those relations.intellecutalVinelandlectualand K-ABCability,“true” relationwith the K-ABCabilitybetweenvariables”behavior.oftest ofa test ofsuch techniquesadequate,Thus,measures,varia-estimatecan bebetween “true” (orbehavior.of “true” adaptiveAlthoughbehaviorthenthe focus of the presentadvantagetheor intel-the unmeasuredand even closer to estimatingbut rather with the constructsAn additionaland byto more closelyoften talk of relationsthe relationand “true” adaptivemultipleof error. It isconstructs,(or factors)iswe canamong such unmeasuredwe examinemeasuresbehavior-less scale-specificthe “true” constructs,or the Vineland,and adaptiveabilitythe constructs.or analysisa betteramountthe underlyingusers of such techniquesmay not provideif they providetheand educationor adaptiveand with an unknownto estimatefor example,factors come close to estimatingit is becausein psychologyAnd when the relationsunmeasured,path analysis,than simply correlatingas intelligencea “purer” (less error-filled),As a result,arethen reflectsshould provideIn essence,“unmeasuredHere,or latent)case the modelour interestimperfectly,among “true” variables.thatIn theconstructsrather than the relation between theAlthoughto createconstructs.the theory that adaptive behaviorand intelligencehowever, to use our measuresusing severaltheIn theThe results of this second analysisfactor analysis,constructs-suchthose constructsbetweenanalyses.to the notionbut relatedwhy such analysesthe underlyingThe paths in the modelability are the fame construct.with a test of intelligence.of those constructs.a generalNonverbalor “cause” the observedthat correspondsin whichwith confirmatorymay questionoften with hypotheticalpossible,to 1.0,factorLivingto see which better “fits” the data. For theand intellectualbetween adaptive g,the correlationare separateconstructs.the Dailyin the three primaryto zero, reflectingare unrelatedexplainrepresentsphi is left free to vary, a modelbehaviorthe Verbalfrom the K-ABC.that the latent1, a simpleto underlieresearchtheis notthese scales measure:of the LISRELapproachover

36Journal of School PsychologyVerbal ReasoningNonverbalReasoningVerbal MemoryDaily Living SkillswCommunicationSocializationFigure1. Confirmatoryfactorother possible approachesmodelof the relationto studyingbetweenlatent variablesadaptivebehaviorand ability.is that several theoreticalmodelscan be tested against each other.’THREEMODELSThe intercorrelationmatrixOF ABILITYof the variablesThe initial model to explain the relationmodel that assumesconstructsstrong,that easoningstill lower loading(.63)used in these analysesbehaviorand intelligencefactor,Verbalhad a lower loadingwas .39, a moderate,The “goodness-of-fit”model proposed(.70)the two factors,providesbut significantstatistics,bchav-factor struc-had the highestSocializationloadingand DailyLivingfactor (.75 and .72, respectivedomain.“true” abilityFinally,theand “true” adaptive(t 2 0), correlation.listed below the model in Figurea generallyadaptive(. 74) and Verbal Memory afor the Communicationor betweenbut relatedof the model showsfor hypothesizedlower loading& Dunbar,and intelligence-aare separate,inspectionsupportSkills had the highest loadings on the adaptive behaviorly), with a slightlyis shown in Table 1.on the correspondingthus offering(cf. KeithBEHAVIORbetween adaptive behaviorfor the six subscalesability factors,For the intellectualNonverbalADAPTIVEshown in FigurP 2. A cursoryloadingsior and intellectualANDgood fit to the data.2, suggest that the’ The adjusted “goodness-of-‘This is a highly simplified explanation ofseveral complex topics. For a guod introduction 10 thenotions of latent variables, path analysis, and analysis of covariance .str ct res, rradcrs canconsult Kerlinger (1986). For more depth, see Bender (1980), Dwyrr (1983), Jtireskog and Siirborn (1979), or Kenny (1979). Th e method of comparing competing models that is used hrrc isdiscussed in Bentler (1980) and in Kerlinger (1986). We do not wish to overstate the advantages ofLISRELor similar techniques; such methods, while powerful, arr not without theI]- drawbat ks,especially their complexity, interpretability,and difficulty of use (Kerlinger, 1986).‘The “goodness-of-fit” statistics provide a measure of how wrll the data fit the proposed model I1thr chi-square value is large compared to thr degrees of freedom, the null hypothesis that thv

Keith et 1.00.47.ll29.161 00.1736.171.00.49.581 00.501.001 001. Verbal Reasoning2. Nonverbal Reasoning3. Vrrbal Memory.4. Daily Living Skills5. Communicationti. Socialization.65.54.15.43.18Verbal ReasoningNonverbalReasoningVerbal Memory-Daily Living Skills-Communication-Socializationx2 83.01df 8Fit Index .876rms ween“true”arethe factorsadaptiveassumedestimatesto beseparatethe degreebutrelatedof the relationbehavior.model fits the data is rejected. An associated probability can also be computed, but it is dependenton sample size, with WC” good models being rejected with large samples such as that used here.Therefore,a “goodness-of-tit” index is calculatrd that increases as the model better lits the data;this adjusted goodness-of-tit index ranges from 0 to 1, a valueof 1 indicating a perfect fit. Finally,the fit of the model can be assessed by comparing the orzginal correlation matrix with the correlations predicted by the model; the average of these differences is the root-mean-squareresidualcorrelation (rmJ). An rms below about .lO is generally considered suggestive of a good fit. For moreintbrmation,see JGreskog and Sijrbom (1984).

38Journal of School Psychologyfit” index for fferencematrix,loadingsbetweenbetween1 was ,876;the root-mean-squarethat the correlationmatrixon the average,and fit statisticsfor Modeladaptivebehaviorand abilityModelModelfactorthat adaptiveIoadinRsthe root-mean-squareincreased,freedom.from eadaptiveis superior1 is considerablythe resultsand intellectualbetterindex reduced to ,843,to160. Chi-squarealso2, with one added degreeat df 1) is significant,behaviorare unrelatedto those shown insuggestingand intellectualabilityand intellectualfor Modelability3, in whichwas constrainedof(as did thein Modelability are separateto the model that holds that the two constructs2 also displaysbehaviorincreasedin ModelabilitysimilarforIn otherbetter fit to the data than does Model 2. Thus,model that posits that adaptive behaviorconstructscorrelationadaptive1 provides a significantly2, the onlythat the correlationand intellectualare quitethe fit for Model1 to 134.36this change in x’ (51.35correlationthat Modelresidualin Modelfrom theto zero (the statisticsthan that for Model 2. For Model 2, the ad,justed goodness-of-fitwhereas(rms) wasof Table 2 to aid comparisons).for this model1, and all are significant.2 beingis here constrainedwords,2 specifiescorrelation2 are shown in Tablethis model and that show in Figure1 are also shown in the last columnTheresidualby this model differedby only ,072.Modelconstructs.predicted1)thebut relatedare unrelated.the correlationto 1 .O.’ Therethe factor loadings in this model (which specifies that adaptive behaviorbetweenare shifts inand intellectualTable 2Factor Loadings (Paths), Constrained Factor Correlations,and Fit Statistics for Models 2 and 3Paths,correlations,and fitModel2Model3Mo lAbility t0.Wrbal ReasoningNonverbalReasoningWrbal MemoryAdaptiveBehavior to:.87.75.6283.72.64.87.7463Daly Living .7OAbilitywith AdaptiveBehavior’0331 .o.75.39Fit statisticsChi-squaw?‘fAd,justed fit ,072IVU Y:Statisticsfor Model 1 are included for comparison‘This value is lixrd at 0 in Model 2, at 1 0 in Model 3.‘Modrt 3 is equivalentto a model in which one general factor is presumedto underlie all abilityand adaptivebehaviormeasures,and the results of an analysis of such a one-factormodel wouldbc identical to those reported here.1

Keith et al.39Verbal ReasoningDaily Living Skills*CommunicationSocializationdf 0Fit Index ,876rms ,072Figure 3. A simple causal model: The infldence of intellectualability are the same construct),also tohighly significant(chi-square3. Again,betteridenticalindexreduced153. Furthermore,over Modelcantlyboth factors reflectingin the statisticsto ,521the increasestructs.constructadaptiveresidualare1; thecorrelation1 to Model 3 isthe superiorityof Modelmodel appearsbehaviorThereto Modelfrom Modelat df l), indicating 296.90than the model specifyingand intellectualUnfortunately,and adaptiveadaptivebehavior1to be signifi-and intellectualabilityasability?behaviorFigureet al.,abilitymodels testingin press)?modelnamed general development.assumesdeviationincreasethe models in Figuresare virtuallymodel shown in Figurewe cannot,of the hebehavioradaptivebe-4 shows a hierarchicaland adaptivebehaviorunmeasuredare notvariable,hereinfluences(path .59).plausible,3 and 4, the factor loadingsand both are almostidenticalthe three modelsand intelligence.to determinehut competing,and lit statisticsfor theto the results of theare statisticallyat least with the present data, further determinedata will be requiredsuch as3 and 4.to adaptivein intellectualof some other,2. In other words,betweenaffects adaptivethe correlation3 and 4 must remainbetween adaptive behaviorlongitudinalabilityIs,intellectualThis model suggests that general developmentAs can be seen in Figurestwo modelsthatthat intelligencethe productbetweenare shown in Figuresdeviation.both ability (path .66) and adaptive behaviorUnfortunately,con-that “causesD both adaptive behaviorthese possibilitiesby .39 of a standardbut are ratherandor identicalOr is the correlationand the path of .39 from intellectualin which it is assumedcausally relatedadap-behaviorof some other variable,simplythat for each standardguishable;relationmodel in which intellectualhavior should increasemodels.rather than unrelatedthe product1978, p. 168)this modelbetweenthat adaptiveIn essence,is causal,suggestsof the relationsuggest3 shows a simple causalbehavior.variablesnatureresultsdo not reveal the true nature of the correlation.(Keith(Lambert,Two plausibleMODELSdue to some sort of causaland intellectualgeneral developmentTheconstructsthe analysesthe correlationPLAUSIBLEthe underlyingability.are related but separatefor example,related,and the averagein chi-squarethe separate-but-relatedresults go far in e behaviorabilitya more verbal content.for this modelTWO ADDITIONALTheseability on adaptive behavior.indistin-the true natureWhile the two constructsarewhich of the two models shown

40Journalof SchoolPsychologyVerbal Reasoning)Nonverbal ReasoningVerbal MemoryX' 83.01df 7Fit Index 350rms .072Figurehere4. A hierarchicalability.model to explainthe relationof this relation.Finally,is the better ain,helpandintelligence(or o explicatea thirdintelligencefurtherthepossibilityof somein detailanalyses,adaptiveis atis,thatthebothset of unmeasuredthosebetweenexists:Thatby Thepurposethe relationdesignedof this researchbetweento separatesuggestbutnotsuggestconstructsthe purposeIactormodel.as measuresof intelligence.Thesehypothesizedbehavioroverandthe cleartheand elsare separatefacetssuperiorityabilityof the naturethe threeor differentconstructs,analysesbehavior1 also d intellectualbutof themodels.Thus,relatedsameof the modelcon-specify-thesebc consideredofbeingresultsas separate.constructs.AlthoughModelbehaviorwhetherof thesethat adaptivebut related,was to compareadaptiveThatis, theseThus,whichresearch,analysesof behaviorvalidityK-ABCcompositeof Dailyto providewithandfindingsthreeLiving(c.g.,do-as measuresSkills,of adaptiveKeithforabilityVinclandcompositesa good measurepreviousfit statisticsand intellectualof theand the threeweightedseemtheof the adaptivebehavioran equallyand Communicationresults,of thisthe compositionSocializabehavior.& Dunbar.

41Keith et eragep.1 l),Therelatedthisin youngeryears,N above;for provideda intelligenceis theaddition,ingotherfurtherof erEvenandwith“true”constructsbeadaptiveis, the latentto explainespeciallyadaptivebetweenset of commonof the samealsoand K-ABCmodelsthe correlationor somemeasuresinPic-research,to determineor whethersubjectsrhe Peabodyrelatedabilitytwo plausibleYet.38.abilitybutor“true”models.and “true”presented.andon the PPVT-R.betweendid the competingunrelatedThealso hildren.( - 1 SD) or lowermodelto bechildren:school-agethewas significantly“true”for school-ageTo testfrommodelconstructsbetween1978).to thosea theoreticalseparateitcloselyages 2 l/2 throughsimilarwe a1so uitechildren,butcorrelationrelated,of theintellectual((i Morrow,forsampleswerewill bc shown85“true ” n daptiveare causallyresultsimilarthebetweenbe morefor low-abilitythethanhavrrelationshouldoverlapare correlated.correlationwill notnotbehavior(PPVT-R);.3 1, andthatwilland adaptiveto the valuein range,wasthen.of thisior andsimilarfit to the dataor constructs)theanalysesthewho scoredrestrictionof rossrelated.levels.relatedwere also tested‘Test-Revisedfor 95 school-agebelowbehavior”significantlythe two typesstandardizationto behypothesizingture Vocabularyarcmentally“significantlyor abilitymodelsof theseabilitywas quitefor theexpectisWhenagesless maturemodelsmodelsthatdifferentas well as school-ageandbehaviorand thusis tresultsthe Vineland/K-ABCthewithand K-ABCTheassessment.in adaptiveapparentlyadaptivein an assessmentreasonablyby “deficitshowever,or otherwiseof the Vinelandmeasuresit alsobetween(or vice versa).changeswe analyzedReasoningin orientation,in a programmightconstructsthatremains,to assumeof bothplacementnot large,behaviorstillpossibility,overlapis alsointelligenceis an importantfoundfor a tionthe inclusionto be accompaniedbehaviorreasonableis primarilycioe providetheusedby the Nonverbalcomposites.to moderatefor onsistcnr;adaptiveheresmalla schoolthe ligence”However,alwaysandTheCompositein importance(Sequential)reported

Adaptive behavior has become an increasingly important component of the assessment of children referred for learning and behavioral problems in educational settings. Yet the construct of adaptive behavior remains ill defined, and fundamental questions about the nature of adaptive behavior remain unanswered.