Consumer Decision Making in Online Shopping Environments The Effects of Interactive Decision Aids

InteractiveDecisionAids

GeraldHa¨ubl•ValerieTrifts

FacultyofBusiness,UniversityofAlberta,Edmonton,Alberta,CanadaT6G2R6gerald.haeubl@ualberta.ca•trifts@datanet.ab.ca

Abstract

Despitetheexplosivegrowthofelectroniccommerceandtherapidlyincreasingnumberofconsumerswhouseinteractivemedia(suchastheWorldWideWeb)forprepurchaseinfor-mationsearchandonlineshopping,verylittleisknownabouthowconsumersmakepurchasedecisionsinsuchset-tings.Auniquecharacteristicofonlineshoppingenviron-mentsisthattheyallowvendorstocreateretailinterfaceswithhighlyinteractivefeatures.Onedesirableformofinter-activityfromaconsumerperspectiveistheimplementationofsophisticatedtoolstoassistshoppersintheirpurchasede-cisionsbycustomizingtheelectronicshoppingenvironmenttotheirindividualpreferences.Theavailabilityofsuchtools,whichwerefertoasinteractivedecisionaidsforconsum-ers,mayleadtoatransformationofthewayinwhichshop-perssearchforproductinformationandmakepurchasede-cisions.Theprimaryobjectiveofthispaperistoinvestigatethenatureoftheeffectsthatinteractivedecisionaidsmayhaveonconsumerdecisionmakinginonlineshoppingenvironments.

Whilemakingpurchasedecisions,consumersareoftenunabletoevaluateallavailablealternativesingreatdepthand,thus,tendtousetwo-stageprocessestoreachtheirde-cisions.Atthefirststage,consumerstypicallyscreenalargesetofavailableproductsandidentifyasubsetofthemostpromisingalternatives.Subsequently,theyevaluatethelat-terinmoredepth,performrelativecomparisonsacrossprod-uctsonimportantattributes,andmakeapurchasedecision.Giventhedifferenttaskstobeperformedinsuchatwo-stageprocess,interactivetoolsthatprovidesupporttoconsumersinthefollowingrespectsareparticularlyvaluable:(1)theinitialscreeningofavailableproductstodeterminewhichonesareworthconsideringfurther,and(2)thein-depthcom-parisonofselectedproductsbeforemakingtheactualpur-chasedecision.Thispaperexaminestheeffectsoftwodeci-sionaids,eachdesignedtoassistconsumersinperformingoneoftheabovetasks,onpurchasedecisionmakinginanonlinestore.

Thefirstinteractivetool,arecommendationagent(RA),al-lowsconsumerstomoreefficientlyscreenthe(potentiallyverylarge)setofalternativesavailableinanonlineshoppingenvironment.Basedonself-explicatedinformationaboutaMarketingScience᭧2000INFORMSVol.19,No.1,Winter2000,pp.4–21

consumer’sownutilityfunction(attributeimportanceweightsandminimumacceptableattributelevels),theRAgeneratesapersonalizedlistofrecommendedalternatives.Theseconddecisionaid,acomparisonmatrix(CM),isde-signedtohelpconsumersmakein-depthcomparisonsamongselectedalternatives.TheCMallowsconsumerstoorganizeattributeinformationaboutmultipleproductsinanalternativesןattributesmatrixandtohavealternativessortedbyanyattribute.

Basedontheoreticalandempiricalworkinmarketing,judgmentanddecisionmaking,psychology,anddecisionsupportsystems,wedevelopasetofhypothesespertainingtotheeffectsofthesetwodecisionaidsonvariousaspectsofconsumerdecisionmaking.Inparticular,wefocusonhowuseoftheRAandCMaffectsconsumers’searchforproductinformation,thesizeandqualityoftheirconsiderationsets,andthequalityoftheirpurchasedecisionsinanonlineshop-pingenvironment.

Acontrolledexperimentusingasimulatedonlinestorewasconductedtotestthehypotheses.Theresultsindicatethatbothinteractivedecisionaidshaveasubstantialimpactonconsumerdecisionmaking.Aspredicted,useoftheRAreducesconsumers’searcheffortforproductinformation,decreasesthesizebutincreasesthequalityoftheirconsid-erationsets,andimprovesthequalityoftheirpurchasede-cisions.UseoftheCMalsoleadstoadecreaseinthesizebutanincreaseinthequalityofconsumers’considerationsets,andhasafavorableeffectonsomeindicatorsofdecisionquality.

Insum,ourfindingssuggestthatinteractivetoolsde-signedtoassistconsumersintheinitialscreeningofavailablealternativesandtofacilitatein-depthcomparisonsamongse-lectedalternativesinanonlineshoppingenvironmentmayhavestrongfavorableeffectsonboththequalityandtheef-ficiencyofpurchasedecisions—shopperscanmakemuchbetterdecisionswhileexpendingsubstantiallylesseffort.Thissuggeststhatinteractivedecisionaidshavethepotentialtodrasticallytransformthewayinwhichconsumerssearchforproductinformationandmakepurchasedecisions.

(DecisionMaking;OnlineShopping;ElectronicCommerce;De-cisionAids;RecommendationAgents;ConsumerBehavior;Infor-mationSearch;ConsiderationSets;InformationProcessing)

0732-2399/00/1901/0004/$05.00

1526-548XelectronicISSN

CONSUMERDECISIONMAKINGINONLINESHOPPINGENVIRONMENTS

Introduction

ThepopularityofinteractivemediasuchastheWorldWideWeb(WWW)hasbeengrowingataveryrapidpace(see,e.g.,GVU1999).Fromamarketingperspec-tive,thishasmanifesteditselfprimarilyintwoways:(1)adrasticincreaseinthenumberofcompaniesthatseektousetheWWWtocommunicatewith(potential)customers,and(2)therapidadoptionoftheWWWbybroadconsumersegmentsforavarietyofpurposes,includingprepurchaseinformationsearchandonlineshopping(Albaetal.1997).Thecombinationofthesetwodevelopmentsprovidesabasisforsubstantialgrowthinthecommercialuseofinteractivemedia.Thefocusofthispaperisononespecifictypeofcommercialuseofinteractivemedia:shoppinginonlineenvironments.Weconceptualizethisbehaviorasashoppingactivityperformedbyaconsumerviaacomputer-basedinterface,wheretheconsumer’scom-puterisconnectedto,andcaninteractwith,aretailer’sdigitalstorefront(implementedonsomecomputer)throughanetwork(e.g.,theWWW).Aconsumercanengageinonlineshoppinginanylocation,butourcon-ceptualizationisbasedontheassumptionsthattheproductsofinterestarenotphysicallypresentatthetimeandthatnoface-to-faceassistanceisavailabletotheshopper.

Auniquecharacteristicofonlineshoppingenviron-mentsisthattheyallowfortheimplementationofveryhighdegreesofinteractivity.Thelatterisamultidi-mensionalconstruct,thekeyfacetsofwhichincludereciprocityintheexchangeofinformation,availabilityofinformationondemand,responsecontingency,cus-tomizationofcontent,andreal-timefeedback(Albaetal.1997,Ariely2000,Zack1993).Inthecontextofcomputer-mediatedcommunication,adistinctionhasbeenmadebetweenpersoninteractivityandmachineinteractivity.Whiletheformerdescribestheabilitytocommunicatewithotherindividuals,thelatterreferstotheabilitytointeractivelyaccessinformationinanonlinedatabase(HoffmanandNovak1996,p.53).Givenourconceptualizationofshoppinginonlineen-vironments,theconceptofmachineinteractivityisofparticularinterest.

Whileithasbeenhypothesizedthatconsumers’shoppingbehaviorinonlinestoresmaybefundamen-tallydifferentfromthatintraditionalretailsettings

MarketingScience/Vol.19,No.1,Winter2000(Albaetal.1997,Wineretal.1997),theorizingaboutthenatureofthesedifferenceshasbeensparse.Weproposethatconsumerbehaviorinanonlineshoppingenvironmentisdeterminedlargelybythedegreeandtypeofmachineinteractivitythatisimplementedinsuchasetting.Specifically,wehypothesizethatthewayinwhichconsumerssearchforproductinforma-tionandmakepurchasedecisionsisafunctionoftheparticularinteractivetoolsavailableinanonlineshop-pingenvironment.Werefertosuchtoolsasinteractivedecisionaidsforconsumers.

Inthispaper,weidentifytwotypesofinteractivedecisionaidsthat,inlightofestablishednotionsaboutpurchasedecisionprocesses,seemparticularlyvalu-abletoconsumers.Basedontheoreticalandempiricalworkinmarketing,judgmentanddecisionmaking,psychology,anddecisionsupportsystems,wedevelopasetofhypothesespertainingtotheeffectsofeachofthesetoolsonconsumers’searchforproductinfor-mation,thesizeandqualityoftheirconsiderationsets,andthequalityoftheirpurchasedecisionsinanonlinestore.Theresultsofacontrolledexperimentindicatethateachoftheinteractivedecisionaidshasasubstan-tialimpactonconsumerdecisionmaking,thusprovid-ingademonstrationofhowtheavailabilityofsuchtoolsmaytransformthewayinwhichindividualssearchforinformationandmakepurchasedecisionsinonlineenvironments.

Thepaperisorganizedasfollows.First,webrieflydiscusstherelevantliteratureonhumandecisionmak-inganddecisionaids.Next,weprovideanoverviewofinteractivedecisionaidsforonlineshoppingandmotivatethechoiceofthetwoparticulartoolsinves-tigatedinourstudy.Wethendevelopasetofhypoth-esespertainingtohowweexpecteachofthesedecisionaidstoaffectdifferentaspectsofconsumerdecisionmakinginonlineshoppingenvironments.Thisisfol-lowedbyadescriptionofthemethodusedtotestthesehypotheses.Wethenreporttheresultsofourempiricalstudy.Thepaperconcludeswithageneraldiscussionofthefindings.

HumanDecisionMakingandDecisionAids

Humansadapttheirdecisionmakingstrategiestospe-cificsituationsandenvironments(see,e.g.,Payne

5

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

1982).Theycanbedescribedas“cognitivemisers”whostrivetoreducetheamountofcognitiveeffortassociatedwithdecisionmaking(Shugan1980).Thenotionthatindividualsaretypicallywillingtosettleforimperfectaccuracyoftheirdecisionsinreturnforareductionineffortiswellsupported(Bettmanetal.1990,JohnsonandPayne1985)andconsistentwiththeideaofboundedrationality(Simon1955).Becauseofthistrade-offbetweeneffortandaccuracy,decisionmakersfrequentlychooseoptionsthataresatisfactorybutwouldbesuboptimalifdecisioncostswerezero.Thisisparticularlycommonwhenalternativesarenu-merousand/ordifficulttocompare,i.e.,whenthecomplexityofthedecisionenvironmentishigh(Payneetal.1993).

Oneformofcopingwithhighlycomplexdecisionenvironmentsistousedecisionsupportsystems.Thelatterarecomputer-basedtechnologiesdesignedtoas-sistanindividual(oragroup)inmakingadecisionorchoosingacourseofactioninanonroutinesituationthatrequiresjudgment(Kasper1996).Decisionsup-portsystemscontainoneormoretools,ordecisionaids,thatperformdistinctinformationprocessingtasksorfunctions(e.g.,searchadatabaseorsortobjectsbysomecriterion).Themotivatingprincipleunderlyingdecisionaidsisthatresource-intensive,butstandar-dizable,informationprocessingtasksareperformedbyacomputer-basedsystem,thusfreeingupsomeofthehumandecisionmaker’sprocessingcapacity.De-termininganadequate“divisionoflabor”betweenhu-manandcomputeriscrucial.Humandecisionmakersaretypicallygoodatselectingvariablesthatarerele-vantinthedecisionprocess,butweakatintegratingandretaininglargeamountsofinformation.Effectivedecisionaidsshouldbedesignedtocapitalizeonthestrengthsandcompensatefortheinherentweaknessesoftheirusers(HochandSchkade1996).

Astandardassumptioninpastresearchondecisionsupportsystems,mostofwhichhasfocusedonman-agerialdecisions(e.g.,PearsonandShim1994),isthatdecisionmakerswhoareprovidedwithdecisionaidsthathaveadequateinformationprocessingcapabilitieswillusethesetoolstoanalyzeproblemsingreaterdepthand,asaresult,makebetterdecisions(cf.HochandSchkade1996).However,behavioraldecisionthe-orysuggeststhatbecausefeedbackoneffortexpendi-turetendstobeimmediatewhilefeedbackonaccuracy

6issubjecttodelayandambiguity,decisionmakersmaybeinclinedtofocusmoreonreducingcognitiveeffortthanonimprovingdecisionaccuracy(EinhornandHogarth1978,KleinmuntzandSchkade1993).Thus,decisionaidsmayleadindividualstomerelyreduceeffortwithoutimprovingthequalityoftheirdecisions.Infact,thereisempiricalevidencethattheuseofde-cisionaidsdoesnotnecessarilyenhancedecisionmak-ingperformance(cf.BenbasatandNault1990),andthatthelattermayevenbereducedasaresult(ToddandBenbasat1992,p.373).Giventhismixedevidence,itcannotbeassumedthataconsumer’suseofinter-activedecisionaidsinanonlineshoppingcontextwillleadtoincreaseddecisionquality.Rather,thisrepre-sentsanopenquestion,whichisaddressedinthispaper.

Inthefollowingsection,wefirstprovideageneraloverviewofinteractivedecisionaidsavailabletocon-sumersforthepurposeofonlineshopping.Basedonestablishednotionsaboutpurchasedecisionmakingandoncharacteristicfeaturesofonlineshoppingen-vironments,twodecisionaidsareselectedforinclu-sioninourempiricalstudy.Thesetwotoolsarethendiscussedindetail.

InteractiveDecisionAidsforOnlineShopping

OverviewofTools

Thetechnologyavailableforimplementingmachineinteractivityinonlineshoppingenvironmentshasthepotentialtoprovideconsumerswithunparalleledop-portunitiestolocateandcompareproductofferings(Albaetal.1997,p.38).Suchcapabilitiesareparticu-larlyvaluablegiventhatonlinestorescannotofferphysicalcontactwithproducts,donotallowface-to-faceinteractionwithasalesperson,andmayofferaverylargenumberofalternativesbecauseoftheirvir-tuallyinfinite“shelfspace,”i.e.,thelackofphysicalconstraintswithrespecttoproductdisplay.

Interactivedecisionaidsthatmaybeofusetocon-sumerswhowishtoshoponlineincludeawidevarietyofsoftwaretools,rangingfromgeneral-purposesearchengines(e.g.,www.infoseek.com,www.lycos.com)tosophisticatedagent-mediatedelectroniccommerce

MarketingScience/Vol.19,No.1,Winter2000

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

systems(e.g.,compare.net,www.jango.com).Acom-monclassificationofinteractiveshoppingagentsisbasedonwhetheratoolisdesignedtohelpaconsumerdetermine(1)whattobuyor(2)whomtobuyfrom.Thesetwotasksmaybereferredtoasproductbrokeringandmerchantbrokering,respectively(seeGuttmanetal.1998).Forthepurposeofthispaper,weconfineourattentiontotheformer.

Amongtoolsforproductbrokering,adistinctioncanbemadebetweendecisionaidsthatoperatewithinaparticularmerchant’sonlinestore(e.g.,www.personalogic.com)andonesthatoperateacrossmerchants(e.g.,www.shopper.com).Theprimaryfocusofthispaperisontheformer.Thedecisionaidsweinvesti-gateareimplementedwithinanonlinestore(seebe-low).However,thisresearchalsopertainstothosecross-merchantdecisionaidsthatallowshoppersdi-rectaccesstoacommonproductdatabase(e.g.,www.jango.com),providedthatthesetoolsdonotdis-criminatebetweenproductsonthebasisofwhichven-dortheyareassociatedwith.

Awell-knownphenomenonregardingdecisionmakingincomplexenvironmentsisthatindividualsareoftenunabletoevaluateallavailablealternativesingreatdepthpriortomakingachoice(Beach1993).Instead,theytendtousetwo-stageprocessestoreachtheirdecisions,wherethedepthofinformationpro-cessingvariesbystage(Payne1982,Payneetal.1988).Inthecontextofpurchasedecisionmaking,atypicaltwo-stageprocessmayunfoldasfollows.First,theconsumerscreensalargesetofrelevantproducts,withoutexamininganyofthemingreatdepth,andidentifiesasubsetthatincludesthemostpromisingalternatives.Subsequently,s/heevaluatesthelatterinmoredepth,performscomparisonsacrossproductsonimportantattributes,andmakesapurchasedecision.Giventhedifferenttaskstobeperformedinthecourseofsuchtwo-stagepurchasedecisionprocesses,inter-activetoolsthatprovidesupporttoconsumersinthefollowingtworespectsseemparticularlyvaluable:(1)theinitialscreeningofavailableproductstodeterminewhichonesareworthconsideringfurther,and(2)thein-depthcomparisonofselectedproductsbeforemak-ingtheactualpurchasedecision.Wefocusontwode-cisionaids,eachdesignedtoassistconsumersinper-formingoneofthesekeytasks.Thetwointeractivetoolsarediscussedinturn.

MarketingScience/Vol.19,No.1,Winter2000RecommendationAgent:AToolforScreeningAlternatives

Weconceptualizearecommendationagent(RA)asaninteractivedecisionaidthatassistsconsumersintheinitialscreeningofthealternativesthatareavailableinanonlinestore.Basedoninformationprovidedbytheshopperregardinghis/herownpreference,anRA“recommends”asetofproductsthatarelikelytobeattractivetothatindividual.Elementaryformsofthistypeofdecisionaidarecurrentlyimplementedonanumberofonlineretailsites(e.g.,www.macys.com,www.netmarket.com).Areal-worldtoolthatcorre-spondsverycloselytoourconceptualizationofRAistheconsumerdecisionguidedevelopedbyPersona-Logic௣(www.personalogic.com).

TheRAusedinthepresentstudygeneratesaper-sonalizedlistofrecommendedalternatives,inwhichalternativesaredescribedbytheirbrandandmodelname.1Thisrecommendationisbasedonthreetypesofparametersprovidedbytheconsumer.First,acon-sumer’sself-explicatedattributeimportanceweightsareusedtocomputeasummaryscoreforeachalter-nativeasthesumoverallproductsof(standardized)attributelevelscalevalueandcorrespondingimpor-tanceweight.2Thisscoredeterminestheorderofal-ternativesintheRA’soutput.Thus,theRAiseffec-tivelyanautomatedimplementationofaweightedadditiveevaluationrule(Payneetal.1993).Second,theRAallowsconsumerstospecifyminimumacceptableattributelevels,andonlyalternativesthatmeetallsuchspecificationsareincludedinthepersonalizedlist.Thiscorrespondstoanautomatedimplementationofacon-junctivedecisionrule(seeWright1975).Finally,theRAallowsshopperstoimposeaquotacut-off(FeinbergandHuber1996),i.e.,tolimitthenumberofproductstobeincludedinthelist.

ComparisonMatrix:AToolforOrganizingProductInformation

Theseconddecisionaidweexamine,acomparisonma-trix(CM),isconceptualizedasaninteractivetoolthatassistsconsumersinmakingin-depthcomparisons

1Fromthere,detailedinformationaboutaproductmayberequestedbyclickingonitsmodelname.

2Thesescoresareimperfect,approximateindicatorsofanalterna-tive’s(unknown)trueutilitytoaconsumer.

7

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

amongthosealternativesthatappearmostpromisingbasedontheinitialscreening.TheCMallowsshopperstoorganizeattributeinformationaboutmultipleprod-ucts.Verybasicformsofthistypeofdecisionaid,usu-allyreferredtoasashoppingcartorbasket,areim-plementedonmanyonlineretailsites(e.g.,www.amazon.com,www.shopping.com).Mostofthesetoolsdonotcurrentlyallowforside-by-sidecom-parisonsofproductsintermsoftheirattributes.How-ever,onereal-worldcomparisonaidthatdoescloselymatchtheabovedefinitionofCMisavailableonCompareNet’ssite(compare.net).

TheCMusedinthepresentstudyisimplementedasaninteractivedisplayformatinwhichproductin-formationispresentedinanalternatives(rows)ןat-tributes(columns)matrix.Itisdesignedtoenableshopperstocompareproductsmoreefficientlyandac-curately.Whileviewingdetailedinformationaboutanalternativeintheonlineshoppingenvironment,acon-sumercanchoosetohavetheproductaddedtohis/herpersonalCM.(Onceincluded,alternativesmayalsobedeletedfromtheCM.)ThedisplayformatisinteractiveinthatashoppercanhaveallproductsintheCMsortedbyanyattribute.Useofthisdecisionaidshouldresultinashiftinemphasisfrommemory-basedtostimulus-basedpurchasedecisionsinthesensethatretainingspecificattributeinformationaboutrelevantalternativesinmemorybecomeslessimportant(seeAlbaetal.1997).

Hypotheses

DependentVariables

WeexpectthatuseoftheRAandtheCMwillhaveanimpactonthreegeneralaspectsofconsumerdecisionmakinginanonlineshoppingenvironment:(1)amountofinformationsearch,(2)considerationsets,and(3)decisionquality.

Amountofsearchforproductinformationisconceptu-alizedasthenumberofalternativesforwhichdetailedinformationisacquired(Moorthyetal.1997).Inourstudy,thiscoincideswiththenumberofpagescon-tainingattributeinformationaboutaparticularprod-uctthatareviewed.Thisisanindicatoroftheeffortanindividualexpendstoscreenavailablealternatives.

8Considerationsetisconceptualizedasthesetofalter-nativesthataconsumerconsidersseriouslyforpur-chase(HauserandWernerfelt1990).3Weuseboththesizeandthequalityofthissetasdependentvariables.Theformerissimplythenumberofproductsconsid-eredseriously,whichcanbeviewedasanindicatorofashopper’srelativeproductuncertaintywhenmakingapurchasedecision.Considerationsetqualityiscon-ceptualizedastheshareofconsideredproductsthatare“non-dominated,”i.e.,notobjectivelyinferiortoanyalternative(seetheMethodsectionfordetails).Decisionqualityismeasuredusingbothobjectiveandsubjectiveindicators.Thisconceptcanbedefinedbybasicprinciplesofcoherence,suchasnotselectingdominatedalternatives(Payneetal.1993).Thus,oneindicatorofobjectivedecisionqualityiswhetherornotaconsumerpurchasesanondominatedalternative.4Oursecondmeasureofobjectivedecisionqualityiswhetherornotashopper,aftermakingapurchasede-cision,changeshis/hermindandswitchestoanotheralternativewhengivenanopportunitytodoso.Switch-ingindicatespoorinitialdecisionquality(seetheMethodsectionfordetails).Finally,subjectivedecisionqualityisconceptualizedastheconsumer’sdegreeofconfidenceinapurchasedecision.

3Analternativeconceptualizationofconsiderationsetistoviewitasadynamicconstructthatevolvesovertimeasproductsarebeingaddedtoanddroppedfromtheset(e.g.,Nedungadi1990).Withinthisdynamicframework,ourconceptualizationofconsiderationsetcorrespondstothefinalconsiderationset,i.e.,thesetofalternativesconsideredatthetimetheactualpurchasedecisionismade.4Measuringthequalityofpurchasedecisionsandconsiderationsetsisaveryambitiousendeavor.Inthiscontext,qualityisconceptual-izedasthedegreeofmatchorfitbetweenheterogeneousconsumerpreferencesanddifferentiatedproducts.Becauseanindividual’spreferencesarenotsubjecttodirectobservation,itisimpossibletoaccuratelymeasuredecisionqualityinuncontrolledreal-worldset-tings.Themeasurementapproachusedinthepresentstudyisbasedontheideaofanobjectivestandardforqualityandrequiresacom-binationofobjectivelydominatedandnondominatedalternatives.Thesetsofavailableproductswereconstructedinsuchawaythat,ir-respectiveofanindividual’sutilityfunction,thepurchaseofpartic-ularalternativesindicates(withcertainty)thats/hemadeapoordecision.Choicesofdominatedalternativesinourcontrolledstudyaretheequivalentofreal-worldpurchasedecisionsthataresubop-timalgivenanindividual’sutilityfunctionatthetimeofpurchaseandthesetofavailableproducts,irrespectiveofwhetherornotanyofthealternativesareobjectivelydominated.

MarketingScience/Vol.19,No.1,Winter2000

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

Wepresentasetofhypothesesabouthowthesesixaspectsofconsumerdecisionmakingareaffectedbyuseofeachofthetwointeractivedecisionaids,RAandCM.Allhypothesesarestatedintermsoftheexpecteddifference,everythingelsebeingequal,5betweenasce-narioinwhichoneofthesetoolsisusedinanonlineshoppingencounterandacasewhereitisnot.Thebasecaseinwhichneithertoolisavailabletoashoppercor-respondstoatypical,“bare-bones”onlinestore.EffectsofUsingtheRecommendationAgentAmountofSearchforProductInformation.Theamountofsearchforproductinformationisdeter-minedbyconsumers’uncertaintyabouttheabsoluteutilityassociatedwithanalternativeandabouttherelativeutilityofalternativesinaset(Moorthyetal.1997,RatchfordandSrinivasan1993).Inanonlineshoppingenvironment,theamountofinformationsearchisalsodependentupontheconsumer’sabilitytoscreeninformationeffectively(Albaetal.1997,Bakos1997).BecausetheRAautomaticallysortsavail-ableproductsbasedoncriteriaprovidedbytheshop-per,thelatterisbetterabletodeterminetherelativeutilityofalternatives,andthisshould,inturn,leadtoareductionintheamountofsearch(Moorthyetal.1997).Thus,wehypothesizethatindividualswhohavethistooltoassistthemintheirshoppingtaskwillviewattributeinformationaboutfewerproductsthanthosewhodonot.

HypothesisH1.Useoftherecommendationagentleadstoareductioninthenumberofalternativesforwhichde-tailedproductinformationisviewed.

ConsiderationSetSize.Modelsofconsiderationsetsizearetypicallybasedonthenotionofatrade-offbetweenthemarginalbenefitsandcostsofconsideringanadditionalalternative(e.g.,HauserandWernerfelt1990,RobertsandLattin1991).Thesemodelsassumethataproduct’sutilityisunknowntotheconsumerpriortoevaluation.However,becausetheRAscreensandranksalternativesbasedonconsumer-specifiedcriteria,itprovidesinformationabouttherelativeutil-ityofavailableproductspriortoprocessingspecific

5Forbrevity,wesuppressthestatement“ceterisparibus”inallourhypotheses.

MarketingScience/Vol.19,No.1,Winter2000productinformation.Asaresult,themarginalbenefitsofincludingadditionalproductsintheconsiderationsetdiminishesmuchmorerapidlythaninasituationwheretheconsumerhasnopriorinformationabouttherelativeutilityofalternatives.Therefore,weexpectthatindividualswhousetheRAwillhavesmallercon-siderationsetsthanthosewhodonot.

HypothesisH2.Useoftherecommendationagentleadstoareductioninthenumberofalternativesconsideredse-riouslyforpurchase.

ConsiderationSetQuality.BecausetheRAusesself-explicatedattributeimportanceweightsandmin-imumacceptableattributelevelstoproduceaperson-alizedlistofrecommendedalternatives,theproductswiththehighestsubjectiveutilitywilltendtoappeartowardsthetopofashopper’slist.Therefore,consum-ersshouldbelesslikelytoconsiderinferioralterna-tivesforpurchase.Inaddition,considerationsetsaremorelikelytobecomposedofproductswithsimilarutilityvaluesthanproductswithdissimilarones(LehmannandPan1994).Therefore,weexpectthattheshareofalternativesincludedintheconsiderationsetthatarenondominatedwillbegreaterwhentheRAisusedthanwhenitisnot.

HypothesisH3.Useoftherecommendationagentleadstoalargershareofnondominatedalternativesinthesetofalternativesconsideredseriouslyforpurchase.

DecisionQuality.TheRAenablesshopperstoscreenproductsusingcomplexdecisionruleswithveryloweffort.Researchondecisionsupportsystemsindicatesthatdecisionaidsdesignedtoscreenlargenumbersofalternativesmayreducedecisionmakers’cognitiveeffort(ToddandBenbasat1994)andimprovedecisionqualitybyenablingindividualstomakecom-plexdecisionswithhighaccuracy(SinghandGinzberg1996).Byapplyingdecisionrulesinanautomatedfashion,suchtoolscanreducetheamountofsuperflu-ousinformationtobeprocessedand,thus,augmenthumaninformationprocessingcapabilities.Inaddi-tion,theabilitytoscreenalternativesinanefficientmannerenhancesthe“quality”oftheinformationthatisprocessed,which,combinedwithreducedinforma-tionquantity,shouldhaveapositiveimpactondeci-sionquality(KellerandStaelin1987,19).Finally,

9

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

WidingandTalarzyk’s(1993)findingssuggestthatelectronicdecisionformatsbasedonweightedaveragescoresforalternativesleadtolessswitchingafterinitialchoice.Thus,wehypothesizethatconsumers’useoftheRAwillhavethefollowingeffectsonthethreein-dicatorsofdecisionquality.

HypothesisH4.Useoftherecommendationagentleadstoanincreasedprobabilityofanondominatedalternativebeingselectedforpurchase.

HypothesisH5.Useoftherecommendationagentleadstoareducedprobabilityofswitchingtoanotheralternative(aftermakingtheinitialpurchasedecision).

HypothesisH6.Useoftherecommendationagentleadstoahigherdegreeofconfidenceinpurchasedecisions.EffectsofUsingtheComparisonMatrix(CM)AmountofSearchforProductInformation.Throughitscapabilityfororganizinginformation,theCMallowsconsumerstomoreefficientlycompareanddeterminetherelativeattractivenessofalternatives.Whensearchingfordetailedproductinformation,shopperswhohaveaccesstotheCMwillanticipatebeingabletosubsequentlyusethistooltomakeac-curateside-by-sidecomparisonsofproductsand,therefore,tendtoinitiallyacquireinformationaboutalargernumberofalternatives.Ifaproductappearsat-tractiveatfirstglance,theconsumercanaddittotheCM,evaluateitindirectcomparisonwithotheralter-natives,andthendecidewhetherornottoretainitinthematrix.BecausetheCMfacilitatesstimulus-based,asopposedtomemory-based,comparisons(seeAlbaetal.1997),itreducesthecombinedmarginalcostofacquiringandprocessingattributeinformationaboutanalternative.Therefore,weexpectthatindividualswhousethistoolwillviewinformationaboutmoreproductsthanthosewhodonot.

HypothesisH7.Useofthecomparisonmatrixleadstoanincreaseinthenumberofalternativesforwhichdetailedproductinformationisviewed.

ConsiderationSetSize.WhileavailabilityoftheCMshouldincreasetheamountofsearch(seeH7),weexpectthatonceconsumersactuallytakeadvantageofthisdecisionaid’scomparison-facilitatingcapabili-ties,theywillbeabletomorequicklyandmore

10accuratelyeliminateunwantedproductsfromtheirconsiderationset.Decisionaidsthathelporganizein-formationhavebeenfoundtoreducethenumberofalternativesconsideredbydecisionmakers(Goslaretal.1986).TheCMimprovesconsumers’abilitytobothdeterminetheirpersonalefficientfrontiersandidentifydominatedalternatives(Wineretal.1997).Asaresult,useoftheCMreducesrelativeproductuncertaintyand,thus,themarginalbenefitofincludinganaddi-tionalproductintheconsiderationset(HauserandWernerfelt1990,RobertsandLattin1991).Thus,wehypothesizethatconsumerswhousetheCMwillse-riouslyconsiderfeweralternativesforpurchasethanthosewhodonot.

HypothesisH8.Useofthecomparisonmatrixleadstoareductioninthenumberofalternativesconsideredseri-ouslyforpurchase.

ConsiderationSetQuality.TheCM’salternativesןattributesformatfacilitatesside-by-sidecompari-sonsofproductsintermsoftheirattributes.Thisdis-playformat,inconjunctionwiththeCM’scapabilityforsortingallselectedalternativesbyanyattribute,reducesthedemandonmemoryandimprovescon-sumers’abilitytoidentifysuboptimalalternatives(seePayneetal.1993,Wineretal.1997).BecausetheCMallowsforefficientdiscriminationbetweenproductswithrespecttotheirsubjectiveoverallutility,itwillrenderconsumerslesslikelytoeithereliminateexcel-lentalternativesfromorretaininferioralternativesintheirconsiderationset.Therefore,weexpectthatcon-siderationsetqualitywillbehigherforshopperswhousetheCMthanforthosewhodonot.

HypothesisH9.Useofthecomparisonmatrixleadstoalargershareofnondominatedalternativesinthesetofal-ternativesconsideredseriouslyforpurchase.

DecisionQuality.Thefindingsofnumerousstud-iessuggestthatthewayinwhichinformationisdis-playedinfluencesdecisionprocessesbyaffectingtheeaseofcarryingoutdifferentprocessingoperations(seeKleinmuntzandSchkade1993).Becausedecisionmakersgenerallytrytoconservecognitiveeffort,theytendtouseprocessingstrategiesthatarefacilitatedbyagivendisplayformat(e.g.,Russo1977).TheCMen-hancesconsumers’abilitytocompareproductsin

MarketingScience/Vol.19,No.1,Winter2000

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

termsoftheirattributes(Albaetal.1997).Asaresult,useofthistoolshouldleadtoashiftinemphasisfrommemory-basedtostimulus-basedchoice.Thelatterhasbeenfoundtoresultinareducedprobabilityofanin-ferioralternativebeingchosen(Muthukrishnan1995,Exp.2).Inaddition,informationdisplayformatsthatreducetaskdifficultyhavebeenfoundtolowerthefrequencyofpreferencereversals(Johnsonetal.1988).Thelattermaybeviewedasindicatorsofsuboptimalchoice.Thus,wehypothesizethatuseoftheCMwillhavethefollowingeffectsonthethreeindicatorsofdecisionquality.

HypothesisH10.Useofthecomparisonmatrixresultsinanincreasedprobabilityofanondominatedalternativebeingselectedforpurchase.

HypothesisH11.Useofthecomparisonmatrixleadstoareducedprobabilityofswitchingtoanotheralternative(af-termakingtheinitialpurchasedecision).

HypothesisH12.Useofthecomparisonmatrixleadstoahigherdegreeofconfidenceinpurchasedecisions.

Method

AcontrolledexperimentwasconductedtotesttheabovehypothesesabouttheeffectsoftheRAandtheCMonthesixdependentmeasuresofinterest.Themaintaskconsistedofshoppingforandmakingapur-chaseofaproductineachoftwocategories—back-packingtentsandcompactstereosystems—inanon-linestore.Inthissection,wediscuss(1)theexperimentaldesignofthestudy,(2)themodelingap-proach,(3)thesampleandincentive,and(4)theex-perimentalprocedure.

ExperimentalDesign

A24full-factorialexperimentaldesignwasused.Themanipulatedfactorsare:RA(yes,no),CM(yes,no),productcategory(backpackingtent,compactstereosystem),andproductcategoryorder(tentfirst,stereofirst).Whileproductcategoryisawithin-subjectsfac-tor,RA,CM,andorderweremanipulatedbetweensubjects.Respondentswererandomlyassignedtooneoftheeightconditionsofthe23between-subjectssubdesign.

Foreachofthetwoproductcategories,atotalof54

MarketingScience/Vol.19,No.1,Winter2000alternativeswereconstructed(9modelsforeachof6brands).Sixactualbrandnameswereusedineachproductcategory.Allmodelnameswerefictitiousbutrepresentativeoftherespectivecategory.Eachalter-nativewasdescribedonsevenattributesinadditiontobrandandmodelname.Fiveoftheseattributeswerevariedsystematicallyacrossalternatives,whiletwoat-tributeswereheldconstant.Thetentattributesthatwerevariedare(numberoflevelsinparentheses):polematerial(3),warranty(3),weight(12),durabilityrating(12),andprice(12).Flyfabricandvestibulewereheldconstantacrossalternatives.Forstereos,thevariedat-tributesare:CDplayertype(3),tunerpresets(3),out-putpower(12),soundqualityrating(12),andprice(12).Cassettedecksandremotecontrolwereheldconstant.6Themeasurementofconsiderationsetqualityandoftwoaspectsofdecisionqualityrequiresalternativesthatareknowntobenondominated.Foreachproductcategory,sixnondominatedalternatives—oneforeachbrand—wereconstructed.Thatis,6ofthe54productsweremutuallynondominated.Theydid,however,dominateallremainingmodels.7Havingonenondom-inatedalternativeforeachbrandguaranteedthat,ir-respectiveofanindividual’srelativepreferenceforbrandnames,oneofthenondominatedproductswasthesinglemostpreferredalternative.Thesixnondom-inatedalternativeswereconstructedbyfirstassigningtothemthebestlevelofthetwothree-levelattributes.Next,forthethreeattributeswith12levels,allsixwereassignedthebestlevelofone,thesecondbestofan-other,andthethirdbestoftheremainingattribute.Allpossiblecombinationsoffirst,second,andthirdbestwereused.Thetwobestlevelsofthe12-levelattributeswerereservedforthenondominatedalternatives.Theremaining48productswereconstructedbymeansof

6Theexactdescriptionsofall54modelsofbackpackingtentsareprovidedinAppendixA.Thecorrespondinginformationforcom-pactstereosystemsisavailablefromtheauthorsuponrequest.7Analternativeisdominatedifthereisatleastoneotheralternativethatissuperioronatleastoneattributewhilenotbeinginferioronanyattribute.Thatis,adominatedalternativeisknowntobewithintheefficientfrontierofanyconsumer.Bycontrast,analternativeisnondominatedifnootheralternativeissuperioronanattributewith-out,atthesametime,beinginferioronatleastoneotherattribute.

11

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

aniterativealgorithmthatapproximatedatargetma-trixofplausibleacross-attributecorrelations,whilead-heringtotherequiredpatternof(non-)dominanceamongalternatives.

ModelingApproach

WeuseGeneralizedEstimatingEquations(GEE)models(Diggleetal.1995,LiangandZeger1986)totestourhypotheses.Thismodelingtechnologygeneralizesclassicallinearmodelsintwoways,bothofwhichareessentialtoourstudy.Webrieflydiscusseachoftheseextensionsinturn.

First,GEEmodelscanaccommodateavarietyofre-sponsedistributionsinadditiontothecommonnor-mal(Gaussian)distribution.Giventhedifferentre-sponsetypesusedinthepresentstudy,thiscapabilityisrequiredforthepropermodelingofourdependentvariables.WithintheGEEframework,therelationshipbetweenadependentvariableandasetofpredictorsisexpressedas

pg(E(Y|x))סb0ם

b(1)

i͚ס1ixi,whereYisthedependentvariableandxס(x1,...,

xp)arethevaluesofasetofpredictorvariablesX1,...,Xp.Theinterceptb0andacoefficientbiforeachpre-dictorareestimated.Thelinkfunctiong,whichmaybeanymonotonicdifferentiablefunction,allowsnonlin-earrelationshipsbetweenpredictorandoutcomevari-ables(McCullaghandNelder19).

Inaddition,GEEmodelsrelaxtheassumptionthatresponseshaveindependentdistributionswithcon-stantvariance.Inparticular,thevarianceofthedepen-dentvariablecanbespecifiedasafunctionofthemeanresponseE(Y|x)viaavariancefunctionVsuchthat

var(Y)סV(E(Y|x))␾,

(2)

where␾isascaleparameter.Thelinkandvariancefunctionsallowawiderangeofnon-normalresponsedistributions,includingbinomial,Poisson,andgamma(ZegerandLiang1986).

ThesecondgeneralizationofclassicallinearmodelsreflectedinGEEmodelsistherelaxationoftheas-sumptionofindependenceamongobservations,whichallowsamoreadequatemodelingofdatathatfollow

12ahierarchicalsamplingpatternorareotherwiseclus-teredbydesign(LiangandZeger1986).Becauseprod-uctcategoryisawithin-subjectfactorinourexperi-ment,theresponsesforeachdependentvariableareclustered(byrespondent)ratherthanindependent.Thus,theabilitytoaccountforsystematicrelationshipsamongmultipleobservationsforanindividualises-sentialtoourstudy.Foreachdependentvariable,a“workingcorrelation”betweenthetworesponsesofanindividualisestimatedasafreeparameter(ZegerandLiang1986).Thisisrequiredforanadequatemod-elingofthedata,althoughtheworkingcorrelationsarenotofsubstantiveinteresthere.

Twoofourdependentmeasures—amountofsearchandconsiderationsetsize—arebasedoncountdatawithnoeffectiveupperlimitandareproperlytreatedasfollowingaPoissondistribution.InourGEEmod-els,thisisimplementedbyspecifyingg(l)סlog(l)asthelinkfunctionandV(l)סlasthevariancefunction,wherelסE(Y|x).Twootherdependentvariables—choiceofanondominatedalternativeandswitching—arebinary.Inaddition,considerationsetqualityismeasuredasafractionbasedonasetofbinaryre-sponses.Abinomialdistributionisadequateforthesethreeresponsevariables.Thisishandledbyusingg(l)סlog(l/(1מl))andV(l)סl(1מl)astheGEEmodel’slinkandvariancefunctions,respectively.Fi-nally,confidenceinpurchasedecisionsismeasuredonanine-pointratingscaleandcanthusbetreatedasstandardGaussian(i.e.,usingaGEEmodelwithiden-titylinkfunctionandconstantvariance).

SampleandIncentive

Eightyundergraduatepsychologystudentspartici-pated(forpartialcoursecredit)inapilotstudyaimedattestingthevalidityoftheexperimentalmanipula-tions.Allmanipulationsweresuccessful,andnotech-nicalproblemswereencounteredwithrepsecttotheelectronicshoppingenvironment.Forthemainstudy,249undergraduatebusinessstudentscompletedtheonlineshoppingtaskforbothproductcategories.8Inadditiontopartialcoursecredit,subjectsinthemainstudyparticipatedinalotterydesignedtoincreasethe

8Thenumberofrespondentspercellinthe23between-subjectssub-designrangedfrom29to33.AllsubjectshadpriorexperienceusingtheWWW,asdidallparticipantsinthepilotstudy.

MarketingScience/Vol.19,No.1,Winter2000

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

validityofthefindingsbymakingtheshoppingtaskmoreconsequential.Priortoenteringtheelectronicshoppingenvironment,subjectswereinformedthattworandomlyselectedwinnersweretoreceiveoneofthetwoproductsthey“purchased”duringtheirshop-pingtrip,plusthedifferencebetween$500andthepriceofthatproductincash.Onetentandonestereosystemweredispensed.Becausethealternativesusedinthestudywereconstructed(seeabove),thetwowin-nersreceivedthereal-worldmodelthatbestmatchedtheirchosenalternative(i.e.,samebrandnameandsimilarattributelevels).Eachprize’scashcomponentwasbasedonthechosenmodel’spriceusedintheshoppingenvironment.

Procedure

Datawerecollectedinauniversitycomputerlabinsessionsof15to20subjects.Uponarrival,participantswereassignedtoapersonalcomputerandinformedthattheywouldbepilot-testinganewonlinestorebyshoppingforaproductineachoftwocategories.Theexperimenterthenhelda10-minutepracticesessionaimedatdemonstratingthefeaturesoftheshoppingenvironment.

Beforeshoppingforthefirstproduct,subjectswereaskedtoratetheirlevelofproductcategoryknowledgeandinterest(usingnine-pointratingscales).Theythenreadadescriptionofthetaskandofthelotteryincen-tive.SubjectsintheNo-RAconditionsweretakentoahierarchicallystructuredwebsitewithallsixbrandslistedatthetoplevelandallmodelsforabrandlistedatthelowerlevel.Subjectsaccesseddetailedinforma-tionaboutaproductbyfirstclickingonabrandnameandthenonamodelname.IntheconditionsinwhichtheRAwasavailable,9subjectsstartedbyprovidingattributeimportanceweightsusinga100-pointcon-stantsumscale,minimum-acceptableattributelevels,andthemaximumnumberofalternativestobein-cludedintheirpersonalizedrecommendationlist.Fromthatlist,theywereabletorequestdetailedin-formationaboutaparticularproduct.SubjectswhousedtheCMwereabletoaddtheattributeinformationdisplayedonaproduct’spagetotheCM,fromwheretheyeventuallymadetheirpurchase.Subjectsinthe

9IntheRAconditions,allsubjectsactuallyusedthisdecisionaid.ThesameistruefortheCMconditions.

MarketingScience/Vol.19,No.1,Winter2000No-CMconditionsmadetheirpurchasefromoneoftheindividualproductpages.Inallconditions,re-spondentswereaskedtoconfirmtheproductoftheirchoicebeforethepurchasewasfinalized.

Aftersubjectsmadetheirfirstpurchase,ameasureofconfidenceintheirpurchasedecision(“Howconfi-dentareyouthattheproductyoujustpurchasedisreallythebestchoiceforyou?”)wasobtainedusinganine-pointratingscale.Next,theywerepresentedwiththelistofalternativesandaskedtoreporttheirconsid-erationset(“Pleaseindicatewhichoftheproductsyouconsideredseriouslybeforemakingyourpurchasede-cision.”).Subjectscouldthenswitchfromthepur-chasedalternativetoeachofsix(five)nondominatedalternatives.10Thisswitchingtaskwaspresentedasaseriesofpairwisecomparisonsinwhichcompletede-scriptionsofbothproductsweredisplayedsidebyside.11Subjectswereencouragedtoswitchwhenevertheysawanalternativetheypreferredovertheirinitialchoice.Theywereinformedthatthelotterywinnerswouldreceivewhateverproducttheyhad“intheirbasket”aftertheswitchingtask.Thesameprocedurewasrepeatedforthesecondproductcategory.Afterthat,subjectscompletedaquestionnairecontainingmanipulationchecks.Finally,theadministratorde-briefedtheparticipantsandconcludedthesession.

Results

ManipulationChecks

Toverifythattheexperimentalmanipulationsweresuccessful,subjectsrespondedtomanipulation-checkquestionsaftercompletionoftheirsecondshoppingtrip.First,theyexpressed(usinganine-pointratingscale)howdifficultitwasforthemtolocatetheprod-uctsthatbestmatchedtheirpersonalpreferences.ThiswasusedtochecktheRAmanipulation.Themeanrat-ingsobtainedfromtheRAandNo-RAconditionsare2.95and4.67,respectively.Thisdifferenceinmeansishighlysignificant(pϽ0.001,x2ס0.178)andinthe

10Thenondominatedproductsusedforthispurposewereidenticaltotheonesusedduringtheshoppingtask.Thenumberofswitchingopportunitiesdependeduponwhetherasubjecthadinitiallychosenadominated(6)ornondominatedalternative(5).

11ThisswitchingtaskissimilartoamethodusedbyWidingandTalarzyk(1993).

13

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

Table1ModelResults:CoefficientEstimatesDependentVariablesAmountofSearch2.158a(0.032)***מ0.574(0.063)***0.022(0.063)מ0.191(0.126)0.109(0.035)**0.284(0.035)***ConsiderationSetSize1.054(0.027)***מ0.094(0.054)*מ0.543(0.054)***0.282(0.109)**0.085(0.042)*מ0.076(0.043)ConsiderationSetQuality0.469(0.098)***2.362(0.199)***0.970(0.195)***0.455(0.391)מ0.487(0.106)***0.383(0.107)***PurchaseofNondominatedAlternative1.686(0.182)***2.086(0.359)***0.497(0.358)1.114(0.717)מ0.569(0.192)xf.**0.212(0.192)ConfidenceinPurchaseDecision6.554(0.086)***0.314(0.173)*מ0.068(0.173)מ0.519(0.346)0.849(0.116)****מ0.291(0.116)*PredictorsInterceptSwitchingמ0.441(0.109)***מ1.728(0.218)***מ0.452(0.218)*מ0.094(0.436)0.013(0.185)מ0.177(0.185)RecommendationAgent(RA)ComparisonMatrix(CM)RAןCMInteractionProductCategoryOrderPositionCellFormat:CoefficientEstimate(Standarderror)͗tvalue͘LevelofSignificanceaLevelofSignificance:*denotessignificanceat0.05level**denotessignificanceat0.01level***denotessignificanceat0.001levelintendeddirection.TheCMmanipulationwascheckedbyaskingsubjectstorate(onanine-pointratingscale)howdifficultitwasforthemtocomparedifferentproducts.ThemeanratingsobtainedfromtheCMandNo-CMconditionsare2.80and4.44,respectively.Thisdifferenceinmeansishighlysignificant(pϽ0.001,x2ס0.176)andintheintendeddirection.Weconcludethatbothofourmanipulationsweresuccessful.HypothesisTestsTotestthehypothesesregardingtheeffectsoftheRAandCM,aGEEmodelwasestimatedforeachofthesixdependentvariables.12Inadditiontoanintercept,thefollowingpredictorvariableswereincludedin12thesemodels:maineffectsforRA,CM,productcate-gory,andorderposition,plusanRAןCMinteractioneffect.TheRAandCMmaineffectsareofprimarysubstantiveinterest.Productcategory,orderposition,andtheRAןCMinteractionareincludedforanad-equaterepresentationofthedataandforexploratorypurposes.ThetwolevelsofRAandCMwerecodedמ0.5(notused)and0.5(used).Productcategorywascodedמ0.5(tent)and0.5(stereo),andorderpositionwascodedמ0.5(firstproductcategory)and0.5(sec-ondproductcategory).13TheRAןCMinteractionAllGEEmodelswereestimatedusingtheS-Plusstatisticalanalysisandprogrammingenvironment(MathSoft,Inc.1998)andaGEEli-braryfunctiondevelopedbyV.J.Carey,DepartmentofBiostatistics,HarvardUniversity.13Inmodelsthatincludeinteractiontermsintheformofproductsof14MarketingScience/Vol.19,No.1,Winter2000

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

wastreatedastheproductoftheRAandCMpredictorvariables.

Table1providesanoverviewofthemodelresults.EachcolumncontainstheGEEcoefficientestimateswithrespecttooneofthesixdependentvariables.Standarderrorsareinparentheses.Thelevelofstatis-ticalsignificanceofacoefficientisindicatedbyaster-isks.14ThetestsofourhypothesesarebasedonthecoefficientsfortheRAandCMmaineffects.Figures1through6containthecellmeans,percentages,orratiosofthedependentvariablesasafunctionofwhetherornottheRAandCMwereused.WediscusstheeffectsoftheRAandtheCMinturn.

TheeffectoftheRAontheamountofsearchforproductinformationishighlysignificant(pϽ0.001)andintheexpecteddirection.AsshowninFigure1,subjectsvieweddetailedproductinformationforsub-stantiallyfeweralternativeswhentheRAwasused(6.58onaverage)thanwhenitwasnot(11.78).ThisprovidesstrongsupportforH1.Asexpected,useoftheRAalsoledtosmallerconsiderationsets.Theav-eragenumberofalternativesconsideredseriouslyforpurchasewas2.78intheRAconditionsandslightlyabove3intheNo-RAconditions(SeeFigure2).ThiseffectissignificantatpϽ0.05andsupportsH2.Whilereducingconsiderationsetsize,useoftheRAresultedinadrasticincreaseinconsiderationsetquality.Theaverageshareofthealternativesconsideredseriouslyforpurchasethatwerenondominatedwas0.85whentheRAwasusedand0.42whenitwasnot(seeFigure3).Thiseffectishighlysignificant(pϽ0.001)andpro-videsstrongsupportforH3.

UseoftheRAhadthefollowingeffectsonthethree

categoricalpredictorvariables,thecoefficientestimateandstatisticaltestforapredictorthatisincludedinaninteractiontermarenotinvarianttothecodingofotherpredictorsincludedinthesamein-teractionterm(seeIrwinandMcClelland2000).Weusestandard-izedmean-centeredcodingforallmaineffects.Asaresultofthemean-centering,allmainandinteractioneffects(i.e.,coefficients)arewithrespecttotheaverageofthetwolevelsofotherfactors.Asaresultofstandardization,allmaineffectsareexpressedintermsofthedifferencebetweenthetwolevelsofafactor.

14BecausewearetestingdirectionalhypothesesregardingtheeffectsoftheRAandCM,thelevelofsignificanceoftheseeffectsisbasedonone-tailedtvalues.Forallothereffects,two-tailedtvaluesareused.

MarketingScience/Vol.19,No.1,Winter2000Figure1

EffectsofRAandCMonAmountofSearchWithinOnlineStore

Figure2EffectsofRAandCMonConsiderationSetSize

Figure3

EffectsofRAandCMonConsiderationSetQuality

15

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

Figure4

EffectsofRAandCMonPurchaseofNondominatedAlter-native

Figure5

EffectsofRAandCMonSwitchingDuringPost-PurchaseSwitchingTask

Figure6EffectsofRAandCMonConfidenceinPurchaseDecision

16measuresofdecisionquality.15Theshareofsubjectswhopurchasedanondominatedalternativewasabout93%whentheRAwasused,andabout65%whenitwasnot(seeFigure4).Thishighlysignificantresult(pϽ0.001)supportsH4.Theproportionofsubjectswhoswitchedtoanotheralternativewhengivenanoppor-tunitytodosoduringthepost-purchaseswitchingtaskwasonlyslightlyabove20%intheRAconditionsbutabout60%amongthosesubjectswhodidnotusetheRA(seeFigure5).Thiseffectisalsohighlysignifi-cant(pϽ0.001)andprovidesstrongsupportforH5.Finally,useoftheRAresultedinasignificant(pϽ0.05)improvementinconsumers’confidenceintheirchoice(meansof6.71vs.6.41)aspredictedbyH6(seeFigure6).ThiseffectoftheRAonthesubjectivemea-sureofdecisionqualityisnoticeablyweakerthanitseffectonthetwoobjectivemeasures.Insum,useoftheRAreducessearcheffortforproductinformation,de-creasesthesizebutincreasesthequalityofconsider-ationsets,andimprovesthequalityofpurchasedecisions.

WenowturntotheeffectsoftheCM.Aspredicted,useoftheCMledtoanincreaseintheaveragenumberofalternativesforwhichsubjectsvieweddetailedproductinformation.However,thiseffectisnotsig-nificant(pϾ0.1),andthusH7isnotsupported.Bycontrast,theCM’simpactonconsiderationsetsizeisverysubstantial.Theaveragenumberofalternativesconsideredseriouslyforpurchasewas2.19whenre-spondentsusedtheCMand3.77whentheydidnot(seeFigure2).Thiseffectishighlysignificant(pϽ0.001)andprovidesstrongsupportforH8.Inaddition,theaverageshareofthealternativesconsideredforpurchasethatwerenondominatedwasabout0.68whentheCMwasusedandabout0.57otherwise(see

15ThepairwiseSpearmancorrelationcoefficientsamongthesethreemeasuresareqסמ0.44(pϽ0.001)betweenpurchaseofnondom-inatedalternativeandswitching,qס0.09(pϽ0.05)betweenpur-chaseofnondominatedalternativeandconfidence,andqסמ0.12(pϽ0.01)betweenswitchingandconfidence.Allthreecorrelationsareintheexpecteddirection.Thefactthatthepairwisecorrelationsbetweeneachofthetwoobjectivemeasuresandthesubjectiveoneareofsmallmagnitudeshowsthatthetwotypesofindicatorsarenotmerelyredundantwitheachotherandthatobtainingobjectivemeasuresofdecisionqualityprovidesvaluableinsightaboveandbeyondwhatsubjectiveindicatorsmayreveal.

MarketingScience/Vol.19,No.1,Winter2000

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

Figure3).ThisfavorableeffectoftheCMonconsid-erationsetqualityisalsohighlysignificant(pϽ0.001),thussupportingH9.

ConsistentwithH10,useoftheCMledtoanin-creaseintheshareofsubjectswhopurchasedanon-dominatedalternative.Thiseffectisonlymarginallysignificant(pϽ0.1).AspredictedbyH11,thepropor-tionofsubjectswhoswitchedtoanotheralternativewhengivenanopportunitytodosoduringthepost-purchaseswitchingtaskwaslowerwhentheCMwasused(38%)thanwhenitwasnot(44%)(seeFigure5).ThispositiveeffectoftheCMondecisionqualityissignificantatpϽ0.05.Finally,wedonotfindanysup-portforH12—theCMdidnotreliablyaffectsubjects’confidenceintheirchoice.Insum,useoftheCMleadstoadecreaseinthesizebutanincreaseinthequalityofconsiderationsets,andittendstohaveafavorableeffectonobjectivedecisionquality.

OtherResults

Inadditiontothetestsofourhypotheses,severalotherresultsareofinterest.First,interactioneffectsbetweenRAandCMwereincludedinthemodelsdiscussedabove.Foroneofthesixdependentvariables,consid-erationsetsize,thisinteractionisstatisticallysignifi-cant(pϽ0.01).Follow-uptestssuggestthatuseoftheRAleadstoareductioninconsiderationsetsizewhentheCMisnotavailable(pϽ0.01)buthasnosucheffectwhentheCMisavailable(pϾ0.7).

Foranadequaterepresentationofthedata,mainef-fectsforproductcategoryandorderpositionwerein-cludedinthemodels.Whilewehavenosubstantiveinterestinthesetwofactors,wenotethateachaffectssomeofthedependentvariables.Productcategoryap-pearstohaveanimpactonfiveoftheoutcomes.How-ever,inclusionofsubjects’knowledgeandinterestwithrespecttothecategoryaspredictorsrenderstheeffectsofproductcategoryonallbuttwodependentvariables(considerationsetquality,purchaseofanon-dominatedalternative)statisticallyinsignificant.Theeffectsoforderpositionmayreflectincreasingfamil-iaritywiththeshoppingenvironmentovertime.

ToexaminewhethertheeffectsoftheRAandCMweremoderatedbyproductcategory,orderposition,knowledge,orinterest,anumberofadditionalGEEmodelswereestimatedforeachofthedependentvari-ables.Themoderatingrelationshipswereexpressed

MarketingScience/Vol.19,No.1,Winter2000andtestedintheformofinteractiontermsbetweenRA,CM,ortheRAןCMinteractionononehandandapotentialmoderatingvariableontheother.EachofthesetermswasincludedinaseparatemodelthatalsocontainedmaineffectsforRA,CM,productcategory,andorderposition,aswellasanRAןCMinteractioneffect.Noneofthesemoderatingeffectswerestatisti-callysignificant(atpס0.05)withrespecttoanyofthesixdependentvariables.Thissuggeststhatthegener-alizabilityofthesubstantivefindingsacrossproductcategoriesissatisfactory.

Discussion

Acharacteristicfeatureofelectronicshoppingenviron-mentsisthelackofphysicalconstraintswithrespecttoproductdisplay.Thevirtuallyinfinite“shelfspace”availableinonlinestoresallowsvendorstoofferanextremelylargenumberofalternativeswithinaprod-uctcategory.Fromaconsumerperspective,havingac-cesstoaverylargenumberofproductsishighlyde-sirable.Atthesametime,however,consumershavelimitedcognitiveresourcesandmaysimplybeunabletoprocessthepotentiallyvastamountsofinformationaboutthesealternatives.Apotentialsolutiontothisdilemmaistoprovideconsumerswithsophisticatedinteractivedecisionaidsdesignedtohelpthemeffec-tivelymanageandcapitalizeontheenormousamountsofproductinformationthatmaybeavailableinelectronicshoppingenvironments.

Theobjectiveofthepresentstudywastoexaminetheeffectsofsuchinteractivedecisionaidsonvariousaspectsofconsumerdecisionmakinginanonlineshoppingcontext.Inparticular,wefocusedontwotoolsthatrepresentobviouschoicesgiventhewell-establishednotionthatconsumersoftenreachpur-chasedecisionsviaatwo-stageprocess.TheRAassistsconsumersintheinitialscreeningofalternatives,andtheCMfacilitatesin-depthcomparisonsofselectedal-ternativesthatareconsideredseriouslyforpurchase.Theresultsofourstudyindicatethatuseofthesetoolshasasubstantialimpactontheamountofsearchforproductinformation,thesizeandqualityofshoppers’considerationsets,andthequalityoftheirpurchasedecisions.

Inanutshell,thetwointeractivedecisionaidsallow

17

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

consumerstomakemuchbetterdecisionswhileexpend-ingsubstantiallylesseffort.Giventhewell-establishednotionthatatrade-offbetweeneffortandaccuracyisinherenttohumandecisionmakingintraditionalen-vironments(Payneetal.1993),itisinterestingthattoolsliketheRAandtheCMcansimultaneouslyin-creasedecisionqualityandreduceeffort.Thefindingsofourstudyshowhowdrasticallyinteractivedecisionaidsimplementedinonlineshoppingenvironmentsmaytransformthewayinwhichconsumerssearchforproductinformationandmakepurchasedecisions.Thepresentstudyexaminestheeffectsofinteractivedecisionaidsonconsumerdecisionmakinginapar-ticularelectronicshoppingsetting.Howwelldoourempiricalresultsgeneralizetoothertypesofonlinere-tailenvironments?WhileboththeRAandtheCMwereoperationalizedasdecisionaidsavailablewithinanonlineretailer’ssite,therelevanceofourfindingsisnotlimitedtointeractivetoolsthatareexclusivetoanindividualmerchant.Inparticular,theresultsapplytoallwithin-andcross-storedecisionaidsthatallowonlineshoppersdirectaccesstoacommondatabaseofproductsandthatdonotdiscriminatebetweenprod-uctsonthebasisofwhichvendortheyareassociatedwith.Totheextentthatsuchdecisionaidsareimple-mentedinthecontextofmulti-retaileronlinemalls,cross-merchantcomparisonschemes,orgroupsofstoresthatallowforunrestrictedcross-storecompari-sons,ourresultsareofrelevance.

Threeimportantboundaryconditionswithrespecttothepresentstudy’sfindingsshouldbemadeexplicit.First,thefocusofthisresearchisonconsumers’goal-directedshoppingbehaviorinanonlineenvironment,ratherthanonexploratorynavigationbehavior.Thus,noconclusionsaboutthelattermaybedrawnbasedontheresultsreportedhere.Second,ourfindingsdonotpertaintosituationsinwhichaconsumerhasnotyetselectedanelectronicstore,onlinemall,orotherentitywithcommonproductofferings.Hierarchicaldecisionprocesses,suchasfirstselectingoneofasetofcompetingonlinemerchantsandsubsequentlyse-lectingaproductfromthatmerchant’sofferings,shouldbeexaminedinfutureresearch.Finally,theRAusedinthepresentstudyisahigh-qualitydecisionaid.Real-worldrecommendationsystemsmaysufferfrom

18substantialimperfections.Forexample,theymayne-glectrelevantattributes,overlooksomeattractiveal-ternativesentirely,orevenbesystematicallybiasedinfavorofasubsetofproducts(e.g.,thoseofacertainbrand).Therefore,wedonotconcludethatrecommen-dationagentswillalwaysandunconditionallyleadtodesirableoutcomesforconsumers.Rather,theeffectsoftheRAfoundinthepresentstudyshouldbeviewedasademonstrationofthepotentialeffectsoftypical,well-functioningrecommendationtools.

Whilethepresentstudyprovidesvaluableinsightsintotheeffectsofinteractivedecisionaidsonconsumerdecisionmakinginonlineshoppingenvironments,furtherresearchwillbeneededtoobtainadeeperun-derstandingoftheseeffects.Inparticular,anexami-nationofpotentialmoderatorswouldbevaluable.Fac-torsthatmightmoderatetheeffectsreportedhereincludethenumberofavailablealternatives,theamountofriskassociatedwithapurchase,andcon-sumers’confidenceintheintegrityoftheinteractivedecisionaids.

Inconclusion,thefindingsofthepresentstudysug-gestthatinteractivedecisionaidsdesignedtoassistconsumersintheinitialscreeningofavailableproductsandtofacilitatein-depthcomparisonsamongselectedalternativesmayhavehighlydesirablepropertiesintermsofconsumerdecisionmaking.Suchtoolsallowshopperstomoreeasilydetectproductsthatareover-pricedorotherwisedominatedbycompetingalterna-tives,thusincreasingmarketefficiency.Moregener-ally,theavailabilityofinteractivedecisionaidsinonlineshoppingenvironmentsshouldenhancetheabilityofindividualstoidentifyproductsthatmatchtheirpersonalpreferencesand,therefore,leadtosub-stantialpositivewelfareeffectsforconsumers.1616ThisresearchwassupportedbyaNovaCorporationFacultyFel-lowshipawardedtothefirstauthorbytheUniversityofAlberta,acontributionfromMacromediaInc.,aswellasgrantsfromTelusCommunicationsInc.andtheUniversityofAlbertathroughitsSo-cialScienceResearchProgram.TheauthorsthankTerryElrodforhisvaluableassistance,V.J.CareyformakingavailablehisS-PlusfunctionforfittingGEEmodels,BarryArdforhisprogrammingassistance,andAdamFinn,JulieIrwin,aswellastheEditor,theAreaEditor,andtwoanonymousMarketingSciencereviewersfortheirinsightfulcomments.Correspondenceshouldbeaddressedtothefirstauthor.

MarketingScience/Vol.19,No.1,Winter2000

AppendixAProductDescriptions:BackpackingTents

PoleMaterial(aluminumtype)regular-strengthregular-strengthhigh-strengthregular-strengthregular-strengthultrahigh-strengthhigh-strength

ultrahigh-strengthhigh-strengthultrahigh-strengthultrahigh-strengthhigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthhigh-strengthregular-strengthhigh-strengthhigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthultrahigh-strengthregular-strengthregular-strengthhigh-strengthhigh-strengthhigh-strength

ultrahigh-strengthultrahigh-strengthregular-strengthregular-strengthhigh-strengthregular-strengthhigh-strength

ultrahigh-strengthultrahigh-strengthhigh-strengthregular-strengthultrahigh-strengthultrahigh-strength

Weight(kilograms)

3.33.53.33.73.33.33.63.33.53.44.04.13.54.23.93.13.53.53.33.93.13.63.33.33.33.34.13..23.34.23.23.33.83.33.63.33.83.43.53.83.33.43.63.33.94.03.33.24.23.43.34.03.8

DurabilityRating(0to100points)

848688868286849191918591829183928686849084878286919191918481829390838291888391848491

Warranty(years)

224334242233433443244333332244444434334224432244444344

Price($)329.99324.99359.99324.99324.99369.99329.99314.99329.99324.99324.99324.99369.99324.99324.99319.99354.99334.99359.99324.99324.99329.99334.99329.99354.99359.99324.99324.99324.99334.99324.99314.99369.99344.99359.99349.99324.99324.99359.99329.99329.99319.99369.99324.99324.99324.99329.99369.99324.99324.99349.99324.99324.99359.99

BrandEurekaEurekaEurekaEurekaEurekaEurekaEurekaEurekaEurekaKeltyKeltyKeltyKeltyKeltyKeltyKeltyKeltyKeltyOutboundOutboundOutboundOutboundOutboundOutboundOutboundOutboundOutboundREIREIREIREIREIREIREIREIREI

SierraDesignsSierraDesignsSierraDesignsSierraDesignsSierraDesignsSierraDesignsSierraDesignsSierraDesignsSierraDesignsQuestQuestQuestQuestQuestQuestQuestQuestQuest

ModelAdventurerChallengerDrifterHuntsmanMountaineerNaturalistOutfitterTravelerWandererGlacierLakeLakeside

MountainSpringsOasis

RagingTideRiverRapidSeabreezeSwiftCurrentWaterfallGalaxy

LunarEclipseMoonscapeNeptuneNorthStarSkylineStargazerSunlightWestwindBearPawCoyoteEagleGrizzly

MountainLionNightOwlRavenRedFoxTimberwolfBacktrailBadlandsBigCountryForestMistLandscape

MountainRangeRockyRidgeSouthRidgeSummitDaydreamFreestyleGreatEscapeJourneyPeacemakerSerenitySolitudeSpirit

Tranquility

FlyFabric2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oznylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon2.3oz.nylon

Vestibuleyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyesyes

Note:Boldfaceindicatesnondominatedalternatives.

MarketingScience/Vol.19,No.1,Winter200019

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

References

Alba,Joseph,JohnLynch,BartonWeitz,ChrisJaniszewski,Richard

Lutz,AlanSawyer,StaceyWood.1997.Interactivehomeshop-ping:consumer,retailer,andmanufacturerincentivestopar-ticipateinelectronicmarketplaces.J.Marketing61(July)38–53.Ariely,Dan.2000.Controllingtheinformationflow:effectsoncon-sumers’decisionmakingandpreferences.J.ConsumerRes.Forthcoming.

Bakos,J.Yannis.1997.Reducingbuyersearchcosts:implicationsfor

electronicmarketplaces.ManagementSci.43(12)1676–1692.Beach,LeeR.1993.Broadeningthedefinitionofdecisionmaking:

theroleofprechoicescreeningofoptions.Psych.Sci.4(4)215–220.

Benbasat,Izak,BarrieR.Nault.1990.Anevaluationofempirical

researchinmanagerialsupportsystems.DecisionSupportSys-tems6(2)203–226.

Bettman,JamesR.,EricJ.Johnson,JohnW.Payne.1990.Acompo-nentialanalysisofcognitiveeffortinchoice.Organ.BehaviorHumanDecisionProcesses45111–139.

Diggle,PeterJ.,Kung-YeeLiang,ScottL.Zeger.1995.Analysisof

LongitudinalData.ClarendonPress,Oxford,UK.

Einhorn,H.,R.Hogarth.1978.Confidenceinjudgment:persistence

oftheillusionofvalidity.Psych.Rev.85395–416.

Feinberg,FredM.,JoelHuber.1996.Atheoryofcutoffformation

underimperfectinformation.ManagementSci.42(1)65–84.Goslar,MartinD.,GaryI.Green,TerryH.Hughes.1986.Decision

supportsystems:anempiricalassessmentfordecisionmaking.DecisionSci.1779–91.

Guttman,RobertH.,AlexandrosG.Moukas,PattieMaes.1998.

Agent-mediatedelectroniccommerce:asurvey.KnowledgeEngrg.Rev.13(2)147–160.

GVU.1999.GVU’sWWWUserSurvey.GeorgiaInstituteofTechnol-ogy:Graphic,Visualization,andUsabilityCenter[http://www.gvu.gatech.edu/usersurveys].

Hauser,JohnR.,BirgerWernerfelt.1990.Anevaluationcostmodel

ofconsiderationsets.J.ConsumerRes.16(March)393–408.Hoch,StephenJ.,DavidA.Schkade.1996.Apsychologicalapproach

todecisionsupportsystems.ManagementSci.42(1)51–.Hoffman,DonnaL.,ThomasP.Novak.1996.Marketinginhyper-mediacomputer-mediatedenvironments:conceptualfounda-tions.J.Marketing60(July)50–68.

Irwin,JulieR.,GaryH.McClelland.2000.Heuristicsformoderated

regressionmodels.J.MarketingRes.forthcoming.

Johnson,EricJ.,JohnW.Payne.1985.Effortandaccuracyinchoice.

ManagementSci.31394–414.

——,——,JamesR.Bettman.1988.Informationdisplaysandpref-erencereversals.Organ.BehaviorHumanDecisionProcess.421–21.

Kasper,GeorgeM.1996.Atheoryofdecisionsupportsystemdesign

forusercalibration.Inform.SystemsRes.7(2)215–232.

Keller,KevinL.,RichardStaelin.1987.Effectsofqualityandquantity

ofinformationondecisioneffectiveness.J.ConsumerRes.14200–213.

——,——.19.Assessingbiasesinmeasuringdecisioneffective-nessandinformationoverload.J.ConsumerRes.15504–508.20Kleinmuntz,DonN.,DavidA.Schkade.1993.Informationdisplays

anddecisionprocesses.Psych.Sci.4221–227.

Lehmann,DonaldR.,YigangPan.1994.Contexteffects,newbrand

entry,andconsiderationsets.J.MarketingRes.31(3)3–374.Liang,Kung-Yee,ScottL.Zeger.1986.Longitudinaldataanalysis

usinggeneralizedlinearmodels.Biometrika73(1)13–22.MathSoft,Inc.1998.S-Plus4.5.MathSoft,Seattle,WA.

McCullagh,P.,J.A.Nelder.19.GeneralizedLinearModels.Second

Edition,ChapmanandHall,London,UK.

Moorthy,Sridhar,BrianT.Ratchford,DebabrataTalukdar.1997.

Consumerinformationsearchrevisited:theoryandempiricalanalysis.J.ConsumerRes.23(March)263–277.

Muthukrishnan,A.V.1995.Decisionambiguityandincumbent

brandadvantage.J.ConsumerRes.22(1)98–109.

Nedungadi,Prakash.1990.Recallandconsumerconsiderationsets:

influencingchoicewithoutalteringbrandevaluations.J.Con-sumerRes.17(December)263–276.

Payne,JohnW.1982.Contingentdecisionbehavior.Psych.Bull.92

382–402.

——,JamesR.Bettman,EricJ.Johnson.1988.Adaptivestrategyse-lectionindecisionmaking.J.Experiment.Psych.:Learning,Mem-ory,andCognition14534–552.

——,——,——.1993.TheAdaptiveDecisionMaker.CambridgeUni-versityPress,Cambridge,UK.

Pearson,J.Michael,J.P.Shim.1994.Anempiricalinvestigationinto

decisionsupportsystemscapabilities:aproposedtaxonomy.Inform.Management2745–57.

Ratchford,BrianT.,NarasimhanSrinivasan.1993.Anempiricalin-vestigationofreturnstosearch.MarketingSci.12(Winter)73–87.

Roberts,JohnH.,JamesM.Lattin.1991.Developmentandtestingof

amodelofconsiderationsetcomposition.J.MarketingRes.28(November)429–40.

Russo,J.Edward.1977.Thevalueofunitpriceinformation.J.Mar-ketingRes.14(May)193–201.

Shugan,StevenM.1980.Thecostofthinking.J.ConsumerRes.7

(September)99–111.

Simon,HerbertA.1955.Abehavioralmodelofrationalchoice.

Quart.J.Econom.6999–118.Singh,Danie`leT.,MichaelJ.Ginzberg.1996.Anempiricalinvesti-gationoftheimpactofprocessmonitoringoncomputer-mediateddecisionmakingperformance.Organ.BehaviorHu-manDecisionProcess.67(2)156–169.

Todd,Peter,IzakBenbasat.1992.Theuseofinformationindecision

making:anexperimentalinvestigationoftheimpactofcomputer-baseddecisionaids.MISQuart.16(September)373–393.

——,——.1994.Theinfluenceofdecisionaidsonchoicestrategies:

anexperimentalanalysisoftheroleofcognitiveeffort.Organ.BehaviorHumanDecisionProcess.6036–74.

Widing,RobertE.II,W.WayneTalarzyk.1993.Electronicinforma-tionsystemsforconsumers:anevaluationofcomputer-assistedformatsinmultipledecisionenvironments.J.MarketingRes.30(May)125–141.

MarketingScience/Vol.19,No.1,Winter2000

¨UBLANDTRIFTSHA

ConsumerDecisionMakinginOnlineShoppingEnvironments

Winer,RussellS.,JohnDeighton,SunilGupta,EricJ.Johnson,

BarbaraMellers,VickiG.Morwitz,ThomasO’Guinn,ArvindRangaswamy,AlanG.Sawyer.1997.Choiceincomputer-mediatedenvironments.MarketingLett.8(3)287–296.

Zeger,ScottL.,Kung-YeeLiang.1986.Longitudinaldataanalysisfor

discreteandcontinuousoutcomes.Biometrics42121–130.

Zack,MichaelH.1993.Interactivityandcommunicationmode

choiceinongoingmanagementgroups.Inform.SystemsRes.4Wright,Peter.1975.Consumerchoicestrategies:simplifyingvs.op-(3)207–239.

timizing.J.MarketingRes.12(February)60–67.

ThispaperwasreceivedMay13,1998,andhasbeenwiththeauthors9monthsfor3revisions;ProcessedbyWilliamBoulding.

MarketingScience/Vol.19,No.1,Winter200021