Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
1
LSM551 Transcripts
Transcript: Course Introduction
Successfulbusinessesdeliverproductsandservicesthatmeetrealcustomerneeds.Thekeytosuchbusinessesisthattheyhaveadeepunderstandingofcustomerpreferences.Thiscourseintroducesyoutoconjointanalysis,whichisatimetestedandversatiletechniquetoincorporatecustomerpreferencesintobusinessdecisions.Eachyear,conjointanalysisisusedinthousandsofbusinesssituationsaroundtheworld.
Thecoursebeginswithadiscussionofalternativeapproachestomeasurecustomerpreferences.Wethentalkabouthowconjointanalysisworks,howitcompareswithothertechniques,theprosandconsofdifferenttypesofconjointanalysismethods,thekindsofdecisionsconjointanalysisishelpfulfor,andlimitationsofthetechnique.Wealsogothroughdetailedexamplesofhowdatagatheredviaconjointanalysisareanalyzedandinterpreted,andhowthesedatainformbusinessdecisions.
Thecoursewillgiveyouacomprehensiveunderstandingofawidelyusedmarketingresearchtechnique.Aftertakingthiscourse,youcanidentifysituationsinwhichconjointanalysisisrelevanttoyourbusiness,aswellaseffectivelycommissionandbuymarketresearchfromvendors.Withtheadventofbigdata,itisnowpossibleformanybusinessestotestthefindingsofsmall-scaleconjointexperimentsbyrunningreal-worldexperiments,fieldexperiments,quickly.Thisdevelopmenthasfurtherincreasedtheapplicabilityofconjointanalysis.
Transcript: Author Welcome IamSachinGupta,aprofessorofmarketingattheSamuelCurtisJohnsonGraduateSchoolofManagementatCornellUniversity.
Thiscourseisaboutmarketresearchapproachesforlearningwhatcustomersreallywant.Thecourseintroducesyoutoconjointanalysis,atime-testedandversatiletechniquetoincorporatecustomerpreferencesintobusinessdecisions.Wediscussprosandconsofconjoint
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
2
relativetootherapproaches.Wealsodelvedeepintohowtoconductconjointanalysis,howtogatherandanalyzedata,andhowtousethedatatomakebetterdecisions.Iwillalsointroduceyoutoindustryexpertswhowilltalkabouthowtheyhavesuccessfullyusedconjointanalysistosolvereal-worldproblemsthattheyfaced.
Transcript: Consumer Preference Matters
Solet'stalkfirstaboutwhyitisimportanttomeasureconsumerpreferences.
Onereasontodothisisthatitallowsthefirmtoofferproductsandservicesthatmeettheneedsofconsumersbetterthancompetitorscanmeetthoseneeds,therebyallowingthefirmtogethighersales,marketshareandprofits.Inparticular,inmanyindustriesfirmshavetoconstantlyinnovateandoffernewproductsandservices.Andsotheyneedtohaveagoodunderstandingofconsumerspreferences.
Thereareseveralreasonswhyfirmsmightfinditadvantageoustoinnovatecontinuously.Onereasoniswhatthiscalledpioneeringadvantage.Theideaofpioneeringadvantageisthatithasbeenfoundthatthefirmthatentersthemarketfirsttheearlyentrytothemarket,hassustainedmarketshareadvantagesrelativetolaterentrancetothemarket.Insomecasesthisadvantageextendstoprofitsaswell.
Asecondreasonforfirmstoinnovateistheconceptofproductlifecyclewhichisthefollowingidea.Ithasbeenfoundthatifyoulookatthesalesofmanyproductcategories,fromthetimewhenthecategorywasfirstintroduced,thenyoufindthatthereisadistinctpatterninthosesalesovertime.Initiallythereisa,afterintroductionthereisaphaseofgrowth,atsomepointthegrowthslowsdownandthemarketreachesmaturity,andthenthemarketgoesintodecline.Sowhenafirmfindsthatitsproductsareeitherinthematurityoratthedeclinephaseoftheproductlifecycle,ithastoworryaboutreplenishingitssalesandprofitsthroughanewproductintroductions.
Athirdreasonisinthecaseofmarketsthathavepatents.Withthepatentthefirmhasprotectionavailabletoitandisabletoreachsalesandmarketshares,butwhenthepatentcomestoanend,thennewproductsenterintothemarketwhichareessentiallycopycatsofthisoriginalproductandthosecopycatserodethemarketshareandprofitsoftheinitialentrance.Anexampleofthisisprescriptiondrugmarketswherefirmsthatinventnewmoleculeshavepatentprotectionforacertainperiodoftime,butwhenthatpatentcomesto
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
3
anend,geneticdrugsenterthemarketatlowerprices,whichthenreducesthemarketshareandprofitsoftheoriginalentrance.
Sofirmshavetofacetwoimportantstrategyquestionswithrespecttonewproductdevelopment.Oneiswhatiscalledthe"goorthenogo"decision.Thefirmhasdevelopedanewproductorserviceandnowneedstodecidewhethertointroducetheproductinthemarket.Andsoitneedstoforecastwhattheperformanceofthenewproductislikelytobe.Asecondtypeofquestionisadesignquestionwhichis:Howcanwedesignorre-designproductsandservicessoastoimprovethelikelihoodofmarketsuccess?
Letmenowtalkaboutacoupleofexamplesofproductdesignquestions.Thefirstoneisinthecontextofafirmthatsaleslaptops,oneofthekeyfeaturesofalaptopistheweightofthelaptopandtheformofinterestandunderstandingconsumerpreferencesaroundtheweightofthelaptop.Forinstancehowmuchisthecustomerwillingtopayforanultralightlaptop,whichisalaptopthatweightstwotofourpounds,asagainstalightlaptopwhichweightsfivetosevenpounds.Ortakethecaseofafirmthatoffersacreditcard,andwantstounderstandhowconsumerstradeofffeaturesofthecreditcard,suchasthebrandnameofthecreditcard,theinterestrate,andthecreditlimit.
Buttherearetwobroadwaysforanalyzingconsumerpreferencesandforforecastingnewproductperformancefromthemarketresearchpointofview.Oneiscalledconcepttestandthesecondiscalledconjointanalysis.Andwe'llspendalotoftimeinthiscourseonconjointanalysis.Letmestartwiththeconcepttest.
Sowhatistheconcepttest?Inaconcepttest,consumersinamarketresearchstudyarepresentedwiththeproductorserviceidea,whichistheconcept.Andtheyaredirectlyaskedfortheirreactiontothisconcept.Thereactionisusuallymeasuredintheformofapurchaseintentonthescale,andthescalemightbeafivepointorasevenpointscale.Aconcepttestisusefulbecauseitprovidesarelativelyquickreadonthemarket'sacceptanceofaproduct.Therearetwodifferentwaystopresenttheconcept.
Onewayiswhatiscalledthecoreideainwhichtheconsumerinthemarketresearchstudyisinformedaboutthekeybenefitthattheproductorservicewouldoffertotheconsumerandthecostatwhichthisbenefitwilloccur.
Thesecondwaytodothisiswhatiscalledthepositioningconcept,whereinadditiontothecoreideatheconsumerisinformedaboutsomeofthemarketingelementssurroundingtheproduct.Suchasthebrandname,thepackaging,theadvertisingmessageandsoon.Inotherwordstheproductconceptisshowninapersuasiveformwhentheconsumers'reactionisobtainedtoit.
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
4
Transcript: Limitations of Concept Tests
Soconcepttestsareveryuseful,buttheyhaveakeylimitationandthatisthatiftheconceptisratedpoorly,wedon'tlearnalotabouthowtoimprovetheconcepttoincreaseconsumeracceptance.Inotherwords,thereisn'talotofdiagnosticinformationthatwegetfromaconcepttest.
Conjointanalysis,whichisthealternative,wasdesignedtoovercomethisfundamentallimitationofconcepttests.Inconjointanalysis,were-orientourselvesnottothinkaboutacceptanceoftheproductconcept,buttogaininsightsintotheunderlyingconsumerpreferencesunderlyingtheconcept.Soconjointanalysisprovidesnumericalmeasuresofconsumers'valuesystems(whichisreallyadescriptionofhowmuchconsumersvaluedifferentattributesandthelevelsofthoseattributes).Sothatoncewehavethesenumericalmeasureswecanpredicttheacceptanceofalternativenewproductconcepts,notjustoneconcept.
Transcript: Attributes and Levels
Manyproductsandservicescanbethoughtofasbundlesofattributes.Anexampleisthecomputerwhichhasmanyattributesandletmetalkaboutthreeoftheseattributes,ifI'mrecognizingthattherearemanyothersandthreeoftheseattributesarememory,sometimescalledRAM,theharddiskortheharddisksize,andthepriceofthecomputer.Now,foreachoftheseattributesorfeatures,wecanthinkaboutlevelsoftheseattributes.Let'stakememoryasanexample.Now,theattributememorycouldbeatlevelsforinstance2gigabytes,4gigabytes,or8gigabytes.Harddisksizecouldbeat500gigabytes,750gigabytes,or1000gigabytes.
Similarly,pricecouldbeatthreedifferentlevels,couldbeatmanylevelsbutthreeofeachcouldbe$800,$1000,or$1200.Nowonceweorientourselvestostartthinkingaboutproductsandservicesasbundlesofattributesandeachattributeaspotentiallybeingatseverallevels,wecanthenthinkaboutproductsasbeingcombinationsofonelevelofeachattribute.So,whatyouseehereisanexampleofacomputerwhichisdefinedonthreeattributes.Ithas2gigabytesofmemory,thereis500gigabytesofharddisksize,anditisatapriceof$1000andso,wehavebeenabletoconstructthehypotheticalcomputerbyjustpickingonelevelfromeachattribute.
So,whatdoesconjointanalysisgiveusonceagain?Itisintendedtogiveuscostumer'svaluesystemswhichisinotherwordstwodifferentmeasures.Oneis,whatistherelativeimportanceofeachattributetothecostumer?Sogoingbacktothecomputerexample,conjointanalysisis
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
5
totellus-isintendedtotellus,whatisthepercentageimportanceofmemoryrelativetoharddiskforagivenconsumer?Ifaconsumerforinstancerunscomputerapplicationsinwhichtheyhavetoloadverylargefileswiththememory,theymayplacehighimportanceonmemoryrelativetoharddisksize.
Thesecondmeasurethatwewantfromconjointanalysisisthevalueprovidedtothecostumerbyeachattributelevel.So,let'sthinkabouttheattributeharddisksize.Wehadthreelevelsofthat;500gigabytes,750gigabytes,and1000gigabytes.Now,wewouldexpectforanyconsumerthatthevalueassociatedwouldgoupaswegofrom500to750to1000gigabytes,butitmaybethecasethatthevaluegoesupsignificantlywhenwegofrom500to750gigabytes,buttheconsumerdoesnotcarealotbetween750and1000gigabyte.Thatistheutilityisaboutthesame.Theconsumerthinksthat750gigabytesisaboutalltheywouldeverneed.So,conjointanalysisisreallyintendedtogiveusthesetwoanswerstothesetwoquestions.Therelativeimportanceoftheattributesandthevalueassociatedwitheachlevelofeachattribute.
Transcript: Ask The Expert: Marco Vriens on Using Conjoint Analysis
Canyoudescribeatimewhenyouusedconjointanalysistosolveabusinessproblem?
I'veactuallydealtwithmanybusinessproblems,butIthoughtaboutitthismorning.ThiswasawhilebackforSonywhenIwasstillbackintheNetherlands,andProfessorDickWittinkwasactuallymythesisadviseratthattime.AndwegotengagedonaprojectwithSony,andtheproductwascarstereos.
AndthebusinessproblemthatthemarketingpeopleatSonyintheNetherlandshadwastwofold.Onewas,theywantedto--theyusuallygotbasicallyabunchofcarstereosshippedfromJapan,andtheyhadtosellthosecarstereosintheNetherlands.SothatwasalittlebitfrustratingforthembecausetheywantedtoreallyofferasetofproductsintheNetherlandsthatwouldappealtotheDutchaudience,andthecarstereosthatcameoverfromJapanwerenotalwaystherightonesfortheDutchmarket.
Thesecondproblemthattheyhadwasthattheywantedtobemorescientificindeterminingwhatthebestandoptimalfeatureswere,thatthecarstereoshouldhaveinthefirstplace.Andthat'showwegottoapplyingconjointtothatparticularbusinessproblem.
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
6
Couldyouexpandonthatexample?
Soifyouthinkaboutproductslikecarstereosandoftentimesproductsthataretechnologyproductswhichhaveoftentimes--thinkaboutcameras,carstereos,computers,etceteraandevenforotherproductsaswell.Oftentimesproductscanbedefinedonalargenumberofattributesandlevels,andsoifyouthinkaboutcarstereo,youcanthinkaboutwhattypeofcontroldoesacarstereohave,isittouchcontrolversusrotarycontrol,whatisthesoundquality,isitaverageorisittopnotchqualityanddoesithaveaCDindexyesorno,doesitgiveyouautomatictrafficupdates,whatisthepricelevel,itcouldbe$50,or$100,$200,whatisthebrandandwhatisthedesignofthecarstereo.
Ifyoulookatallthesefeaturesandalltheseattributes,thereareliterallyhundredsifnotthousandsofpotentialproductsthatyoucancreatefromtheseattributes,that'sasituation,ifyoureallythinkaboutit,thatyoucannotevaluateortestinaconceptvalueapproachbecauseitsjusttoomanyalternatives.Whatyoudowithconjointisyoubasicallyselectinasmartwayasmallnumberofprofilesthatyoucaneitheraskrespondentstoevaluatethatsmallsetofprofiles-theseprofilesthatarehypotheticalproducts.Youcaneitheraskrespondentstoevaluatethoseprofilesonebyoneoritcouldputtheminachartandaskpeoplelikeifyou-ifthesewouldbetheoptionsthatwouldbeavailablewhichwouldyouchoose.
Butbydoingthatbyaskingrespondentstorespondtothatsmallsetofprofilesandthenapplyingmodelingtogetdata,Iamabletoactuallygivemanygoodinformationaboutallthepotentialproductsthattheycanbuildwiththatattributesandlevels.AssoeventhoughIamonlyaskinglet'ssay,evaluationsonlet'ssay16hypotheticalproducts,Iamgoingtobeabletogivemanagementinsideintohowappealingthousandsofpotentialproductscouldbeandthat'sreallythekeybenefitofdoingconjoint.
Whywasconjointpowerfulinthatinstance?
Thereasonwhythiswassopowerfulwas,obviouslynumberone:becauseofthekeybenefit,asmallnumberofresponsesfromthesampleofrespondentsallowsyoutogetinsightintothethousandsofpotentialproductsthatSonycouldproducebasedontheattributesandlevels.That'sthekeybenefitnumberone.
Butspecifictothesituationandtootherbenefits,onewasbecauseweusedcomputer-aideddesigntechniquetocreatethehypotheticalcarstereos,peoplegotaveryrealisticviewonhowthecarstereoswouldactuallylooklike,andthatenabledSonytolookathowthedesignofthecarstereoactuallyinteractedwithsomeofthefeatures,andobviouslynowadaysthereisalotoftalkabouthowimportantthedesignofaproductis.Thismethodologyreallyallowedustocombinegettinginsightintothesofterelementslikedesign,andcombineitwiththeharderelementsofthefeaturesthattheproducthas.Sothatwasverypowerful.
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
7
ThethirdbenefitwasthatthesmallmarketingteamintheNetherlandshadtonegotiatewiththepowerfulmanagementteaminJapan,andunlesstheywerereallyabletocomeupwithsomeharddata,andsomevalidatedinsights,basicallytheyhadnoleverage,butwiththeseinsightstheJapanmanagementteamwasactuallyverysurprisedandveryimpressed,becausetheDutchteamwasabletoactuallycreateproductsthattheJapanmanagementteamhadnoteventhoughtof.
Whatcriticalinformationdidconjointyieldinthatexample?
Wellonceyougatherthisdataandyouestimateyourmodel,nowkeepinmindthatyouhave,wehadhundredsofrespondents.Andasmanypeopleknowinmarketing,noteverybodyhasthesamepreferences,somepeoplelikethisandsomepeoplelikesomethingelse.Andsoitbecomesalittlebitofacomputationalproblem.Butconjointallowsyoutodothat.
Onceyouhaveyourmodelestimated,Icancreateasimulationmodel.Andthatsimulationmodelallowsmetosay,okay,pickacrossthisgroupofconsumers,pickthefour,fiveprofilesandcarstereosthatwouldbemaximally,optimallyappealingtoyourtargetaudience.Sothatisahugebenefit.
Asecondbenefitisalsothatyoucangetsomeinsightintoiftherearemaybesomepotentialdiscretemarketsegmentsthatreallyhaveverydifferenttastesfromeachother,somaybeyouneedtotakethatintoaccount,anditcouldevenbethatifthosesegmentshavedifferentdemographics,thatcouldactuallymeanthatyoumayneedadifferentmarketingcampaignforthat.
Canyoudescribesomecommonpitfallstoavoidinconjointanalysis?
Thereareacoupleofcommonpitfallsandacoupleofmistakesthatpeoplewanttoavoid.Makesurethatyourexperimentaldesigncanactuallyestimatethemodelthatyouneed.NowadaysthereisalotofstandardsoftwareavailablefromSPSStoassortedsoftware.Thatsoftwareisingeneralverygood,butyoustillhavetobecarefulbecausetherearesituationswherethesoftwarecannothandleyourparticularproblemifyoudon'tstillgoaheadandcomeupwithastandarddesign,thatdesignmightnotbethecorrectdesignforyourproblem.
Ifyoudostartmodeling,thinkaboutsubstitutionandcannibalization,andifyoureallythinkthatthoseeffectsaregoingtotakeplaceinyourmarket,youneedtomakesurethatwhenyouanalyzethedatathatyouusetherightmodels,sothatyoucanactuallygetinsightintothosekindsofdynamics.Yeahandjustfollowthebestpracticesasmuchaspossible.Ithinkthisisreallyanareawhereyouwanttohirepeoplewhodothisasadayjobandwhodothisalotsothatyouknowthatthepeoplewhoaregoingtodothisprojectforyouwilldoitwell.
Whataresomebestpracticesthatyou’drecommend?
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
8
Ifyoutalkaboutbestpractices,intheareaofconjoint,thatisactuallyabigquestion.Ididalotofthinkingaboutitthismorning,becauseobviouslypeoplehavebeenwritinghundredsofarticlesandbooksaboutthedifferentmethodologies,andthereareIdon'tknowhowmanyvarieties.SoIwantedtobringitdowninacoupleofsimplerecommendationswithoutgoingintotoomuchdetail,becauseifyoureallywantthefullanswerIthinkyouneedtoreadapaperIthink.Butitboilsdowntothreeareas.
Youhaverecommendationsandbestpracticesintheareaofdatacollection,youhaverecommendationsandbestpracticesintheareaofanalysisandmodeling.Andyouhavebestpracticesintheareaofsimulationsandrecommendations.
Intheareaofdatacollection,therewereacoupleofthings,Ithink,peopleshouldthinkabout.Oneisthatbecauseyouareaskingpeopleabouttheirpreferencesandyouwanttopredictultimatelywhatthey'regoingtodointhemarketplace,Ithinkitisreallyimportantthatyoudesignataskthatisasrealisticandcloseaspossibletowhatpeopledointherealworld.Andthat'soneofthereasonswhyinourSonystudyweusedcomputerdesigningtechniques--toreallycreatethoseprofilesthatreallylookedlikerealcarstereos,becausethecloseryouaretotherealworld,thebetteryou'regoingtopredicttherealworld.
Thesecondbestpracticeisthatyouneedtodoapretesttomakesurethatpeopleunderstandthetaskandthatpeoplecanreallydowhatyou'rewantingthemtodo.Thentryevaluatingsixteenprofiles,thatisnotaneasythingtodo,especiallyifeachprofileisdefinedonbetweenfivetotenattributes,thatcanbeachallengingtask.Andsomeoftheproblemsthatwedealwitharemuchbiggerthanthis.Wegotapproachedbyacarcompanyandtheywantedtotestahundredattributes,soobviouslyyouneedtodo,youneedtodosomethingalittlebitmorespecialized.Butinthosesituations,itbecomesincrediblyimportantthatyouendupwithatask,thatyoureallyknowpeoplecandothis,andpeoplewilldothis.AndI'vebeeninthosesituationsthatIgeneratedataskfortherespondents,andIdidacoupleoftestinterviews,and,loandbehold,halfway,peoplearecheckingout.They'restillgivingyouanswers,butthey'renottheanswersthatyou'relookingfor.Soyouwanttomakesurethattheynotonlycandoit,buttheyalsowilldoit.
Intermsofexperiementaldesign,whichisobviouslyaveryfoundationalpartofcreatingaconjointstudy,thebestpracticeisreallytomakeuseofBayesianmethods.Idon'tknowifProfessorSachinGuptadiscussesBayesianmethodsinhisclass.Butthekeybenefitofthatapproachisthatitreallyallowsyouto,withfewerprofiles,getthesamequalityofinformation.Andthatoftentimesmeansthatyoumayonlyneedhalfoftherespondentsthatyouwouldhaveotherwiseneededinordertogetthesamequalityofinformation,soit'saverypowerful,
verypowerfulapproachtodesigningexperiments.Intheanalysisandmodeling,IwouldsaythecurrentstateoftheartIthinkisoneofhierarchicalbaseestimationmethods.
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
9
Andit'salso,Iwouldsay,usingnon-standardchoicemodels.Foralongtime,thestandardmodelofanalysiswasthelogitmodel,themultinominallogitmodel,andit'saverypowerfulmodel.Butinaconjointchoicecontext,itdoesnotalwaysworkasyouwantitto.Andoneofthereasonsforthatisthatthatmodelisnotreallyagoodmodeltocapturecannibalizationandsubstitutioneffects,andthoseeffectsareveryreal.Andsothinkabout,ifyouareSony,andyouarethinkingaboutdesigningaproductlineforyourcarstereos,thathasmaybe,Idon'tknow,eightortendifferentcarstereos,youwanttoknowthateachindividualproductthatyouputinyourproductlineishopefullystealingmarketsawayfromyourcompetition--butnotfromtheotherproductsinyourproductline,right?Soifyouwanttogetinsightintothosetypesofdynamics,that'swhereyouneed,Iwouldsay,moreadvancedchoicemodels,likethesource-of-volume-probitmodel.NowthatmaygoalittlebittoodeepforanMBAclass,butIjustwanttocallitoutthatifanybodyeverdoesastudyinthisarea,thattheyareawareofthefactthattherearethingsbeyondthemultinominallogitmodel.
NowthethirdareaIthinkisreallyimportant,andthathastodowithsimulationsandrecommendations,andletmejustmaybefocusonrecommendations.Ifyoudoaconjointstudy,theconjointstudy,theconjointmodelwillallowyoutodevelopamarketsimulation,andI'mprettysurethatProfessorGuptawillshowsomeofthesemarketsimulatorsinhisclass.Andthosearepowerfulintstrumentsbecausetheywillgiveyoupredictedmarketshares,andtheywilltellyouhowsuccessfulaproductcanbe.Butyouneedtobealittlebitcautious,andyouneedtobecautiousforanumberofreasons.
Oneis,intheconjointyouassumethatpeopleinthemarketare100%awareofthealternativesandare100%awareofthenewfeaturesofthenewproducts.Andthat'sobviouslynotalwaysthecase.Soyouneedtoadjustyourpredictionsdownward.
Theotherelementis,peopledon'talwaysdowhattheysaythattheywilldoinasurvey.Noweventhoughthisisnotaresearchthing,Ithinkitisstillveryimportant,becauseIhavebeeninsituationswherepeoplewoulddoaconjoint,theywouldgiveadvicetomanagement,andthenloandbehold,lateronintherealworlditturnedoutthattherealsuccessoftheproductwasnotnearlyasgoodaswhatwethoughtitwouldbe.Andsonow,havingsaidthat,Imustsaythatconjointoftentimespredictsreal-worldperformancereallywell,butyoustillneedtogivethatdisclaimerbecauseofthereasonsIjustmentioned.Iftheawarenessisnotwhereitneedstobe,ifpeopleonthesurveyarenotdoingwhattheysaythey'redoing,then,yeahyoujusthavetobecarefulwiththat,youhavetosettheexpectations,Ithink,inaveryrealisticway.Oratleasthavethatconversationwiththepeopleyou'regivingthatadviceto.
Transcript: How Conjoint Works
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
10
So,howdoesconjointanalysiswork?Inconjointanalysis,wedonotaskedconsumerstodirectlytellustheimportanceofattributesorthevaluetheyreceivefromeachleveloftheattribute.
Forinstance,wedon'tgototheconsumerandsay,"Howimportantismemorytoyou?"or"Howmuchdoyouvalue4gigabytesofmemoryrelativeto2gigabytesofmemory?"Insteadinconjointanalysis,weaskconsumerstoperformarealistictask:thatofprovidingoverallevaluationsofproductsandthenweusethoseevaluationstoinfertheconsumer'svaluesystem.
Forinstance,wewillshowaconsumeracomputerasawholeandaskthemtoevaluatethecomputer.Fromthoseevaluations,wewouldinfertheimportanceassociatedwiththeunderlyingattributes.Inprovidingthisevaluation,therespondentsmustsimultaneouslyconsiderboththegoodandthebadcharacteristicsoftheproductinmakingthejudgment.Thatiswemustmaketradeoffsandthentheygaveustheevaluationoftheproduct.Thealternativetoconjointisofcoursetoasktheconsumerdirectlytotellushowimportantistheattributeandhowmuchtheyvaluetheleveloftheattribute.
Now,ithasbeenfoundthatthereareseveralconcernswhenweaskconsumersdirectly.Oneconcernisthatconsumersoftentendtorateallattributesasbeingimportant.Sotheysay,"Memoryisimportanttome,harddiskisimportanttomeandthepriceisimportanttome."Andthat'saproblembecausetheyarenotreallymakingtradeoffsbetweenattributes.
Anotherconcernisthataskingconsumersdirectlyisanunrealistictask.Thisisnothowconsumersgoaboutmakingdecisionsabouttheproduct.Theydon'tlookatoneattributeatatimeandthinkabouthowimportantthatattributeis.Instead,theyareusedtoevaluatingproductsinaholisticway.
Thethirdconcernisthatsometimesconsumersareunwillingtotellustheirtruthfulevaluationofunderlyingattributesandhereisanexampleofthatlastconcern.AstudywasdoneatabusinessschoolwherethegoalwastounderstandhowMBAstudentsthinkabouttheattributesofjoboffers.So,ifyouthinkaboutjoboffers,thejobcouldhaveattributeorfuturesuchasthesalaryassociatedwithit,theregionofthecountryinwhichthejobislocated,thefunctionalareainwhichthejobis;ifit'samarketingjoboranaccountingjoborafinancejob,howmuchtraveldoesthejobinvolved,howmuchopportunitythereistoadvanceinthejobandsoonandsoforth.Andtheresearcherswantedtounderstandthepreferencesofthestudentsforthesedifferentfeaturesofjobs.Theydividedthestudentsintotwogroups
randomly.Thefirstgroupwasaskedtodirectlytelltheresearchershowimportanttheattributeswereandtheywereaskedtorankordereightfeaturesofthejobs.Thesecondgroupwasgivendescriptionsofjoboffers.Eachjobwasdescribedinallthefeaturesoftheattributes
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
11
andthentheresearchersinferredfromthatwhatmustbetheimportanceassociatedwiththefeaturesofthejobs.
So,let'sthinkaboutthedifferenceintheresultsbetweenthesetwogroupsintermsoftheimportanceoffeaturesandlet'sfocusononeparticularfeaturewhichisthesalary.Interestingly,theresearchersfoundinthiscasethatthegroupthatwasaskedtodirectlytelltheresearcherstheimportanceofsalarysaidthatsalaryissixthmostimportant;it'snotreallythatimportant.Theothergroupwhichwasnotaskeddirectly;whentheyinferredtheimportanceofsalary,weinfactfoundthatsalarywasthemostimportantfeature.Andso,thesecondgroupwasreallydoingaconjointtaskandwefindthatwhenweinferredtheimportanceofcertainfeatures,thoseareactuallyquitedifferentfromthedirectlystatedimportanceofthatparticularfeature.
Transcript: Ask The Expert: Johannes Gehrke on Big Data
Whatis"bigdata"?
So,Iamgoingtotalkaboutbigdatainthiscourse.So,whatisthisbigdataallabout?Bigdataisdatathatislarge,complexdifficulttoanalyzeusingtraditionaltools,suchasrelationaldatasystemsandtraditionalstatisticaltools.
Let'slookalittlebitintothehistorybecausethatisactuallyquiteinterestinghowthisallcameabout.So,inthe80's,itwasreallythetimewhenrelationaldatabasesystemstookoff.Relationaldatabasesystemsaredatabasesystemswhereyouhavedatalikeyouthinkaboutwhat'sinadatabasesystemintermsofatableandyourrowsandcolumnsinitanditisreallythebackboneoftoday'sinformationtechnologyinfrastructure.Thismaybewheneveryoumakeabanktransfer,wheneveryoumakeapurchaseatthesupermarket,wheneverIsentyouanitem,youmakeentriesonthesedatabasesandtheyhavenicepropertiesofnotlosinganyoftheinformationthatisinthere.Andtraditionally,theykeptthecurrentstateoftheenterprise,yourcurrentcustomers,yourcurrentsales,yourcurrentinventory,yourcurrentprices,andthisisvaluablebecausethedatabaseservedasastructurallayeronwhichyoucouldbuildotherapplications,addanewbaseeitherinternalorexternalcustomers.
Forexample,enterpriseresourceplanning,humanresourceapplications,billing,payment,procurement,andsoonandthesedatabases,theybecameaveryimportantassetand
thereforeitstartedsprawlingallovertheenterprise.Andtheresultwasthatenterprisesendedupnotonlywithonedatabase,butwithlotsofdatabases,oneforthesalesdepartment,
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
12
anotherforthemarketingdepartment,andthethirdonethathandlessales,inventory,andsoon.
Istherevalueacrossthesedatabases?
So,nowthatwehaveallthesedatabasesacrosstheenterprise,istherevalueacrossthesedatabases?AndIthinkthat'sexactlywhatpeoplestartedtorealize.So,theywantedtocreateanintegratedviewofallofthisdata.
Forexample,havinga360-degreeviewofacustomerthatcreatesalotofvalueandyounowseethecustomerintegratedacrosstheservicedatabaseandthemarketingdatabase.Forexamplenow,whenthecustomercallsandcalls,forexample,aboutaserviceproblem,afterwardsyoucantrytoupsellhim,eitherto,orher,toadifferentlevelofserviceortodifferentversionoftheproduct,orcrosssellhertodifferentproductatall.Andalso,youmayfindjustwithinoneofthesedatabasesyouhavethesamecustomer,threetimes.Forexample,lookatmylastname,youknow,Igetspammailwheremylastnameismisspelledinvariousdifferentvarieties,andthecustomerthatlookssobad,actually,thecustomerappearsthreetimesinthedatabasemaynotbesobadafterall,becausethissumactuallyhavethreeindividualcustomers.Andsothis,eventhoughthesethreedifferentcustomersareslightlydifferent,anddataintegrationcanhelpyoutocleanupyourdataandcorrectyourerrors.Andthereareseveral,alotofvalueintheintegrationofenterprisesthatyouarealsointeractingwith.Forexample,datadirectoryfromthesupplierswherethereareseamlessinstantinformationflowacrossthewholesupplychain.
What'sthehistoricalperspective?
Didn'tthesedatabasesthenalreadygetquitebig?Yes,theydid.Therewasanothercompoundingissue.Manypeoplerealizedthatitisnotonlythecurrentstateofenterprisethatisreallyimportantbutthereishugevalueinlookingatevolutionofthestateoftheenterpriseovertime.
So,especiallyinthe90'sthenthisresultedintothecreationoftheso-calleddatawarehouses,thatdidn'tonlykeepcustomerswiththeircurrentaddresses,butallthepreviousaddressesofthecustomers.So,didn'tonlykeeptherecenttransactionsbutthewholetransactionhistory.
Forexampleagain,forthepriceoftheproduct,itdidn'tonlykeepthecurrentpricebutthewholehistoryofthepricesandhowthepricefluctuatedovertime.Forexamplewithdifferentholidaysorspecialeventsandthisthenallowsyoutogetreallydeepunderstandingthatyoucouldn'tgetjustfromasnatchofthedata.
Let'sjust,forexample,takeacustomer.Acustomermaybetrackedwithloyaltycardstounderstandthelifecycleofthecustomer,forexample,fromcollegetothefirstjob,tothefirst
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
13
babyandthatthenallowedenterprisestomarkettheirwhole-partportfoliovery,verydifferentandveryspecificallytothiscustomeratdifferentpointsintheirlife.Andthesizeofnowthisintegrateddatabasesthatalsokeptthewholehistoryofeverythingthatwasgoingon–thatarereallynowusedfordecisionsupport,notforthecurrentstateoftheenterprise,theythenmadetraditionalstatisticaltoolsinadequate.
Andtherefore,theyresultedintheemergencealso,thisfirstwaveofbigdatawhichwascalled"datamining"orknowledgediscoveryindatabases,andthisisreallytheprocessofdiscoveringpatternsinthedataandthisincludesactuallymanyofthetechniquesthatarecoveredinthiscourse,buttheyarescaleduptoworkovergigabytesorpetabytesofdata.
Asimpleexamplewouldbethatdataminingwouldallowretailertofindoutwhatitemscustomerswouldpurchasetogetherinasupermarketorinwhatsequencetheywerepurchasingbooksfromabookstore.
Whatarethe"3V's"ofbigdata?
So,bigdataisreallytheculminationofallofthesedevelopmentsthatwejusthavetalkedabout,soit'snotonlylargebutithaswhat'softencalledthethreeV's.So,thethreeV'sarevolume,velocityandvariety.So,letmequicklytalkthrougheachofthem.
So,whatdoesvolumereallymean?Imean,ifyoulookattheamountofdataintheworld,thisisgrowingexponentially.SoImeanonewaytothinkaboutthisexponentialgrowthisthatifyouthinkabouttheamountofdataintheworld,90%ofthathasbeengeneratedjustoverthelasttwoyears,andactuallytheamountofdatathatisgeneratedeveryyearisdoublingabouteveryfortymonths.So,thedriversofallofthisareallaroundus.
Forexample,there'sthisubiquityofdata-generatingmachinery.Forexample,sensorsthatcaptureactivitiesinmanufacturingorforexample,serverlocksfromdatacentersthatcanbeanalyzedforintrusionsorforthehelpofserversorfortheanalysisofoutrageousorsoftwareproblems.Butit'salsoreallydrivenbyallofthegadgetsthatwecarryaroundeverydayandthesocialmediathatweareusing.
Forexample,ifyoujustlookatoneextremeintheperiodbetweenNewYear'sEveatNewYorkState2012-2013.Thereare1.1billionphotosuploadedtoFacebook.IfyoulookatTwitter,it'snowprocessinghalfabilliontweetsperday.Allofthisisreallyagoldmineformarketing.Andreallywhatthisalsomeansisthatifyouthinkaboutpetabytesinterms-insteadofterabytesandsotraditionally,theterabytesinthe90'swasoftencalledthe"terrorbyte"becausepeoplehadalotofterrorinfrontaboutit,butnowit'sreallypetabytes.
Forexample,Wal-Martsupposedlygets2.5petabytesofdatafromitscustomertransactionsperhour.Soifyoucouldanalyzetheimpactofcustomerpropertiesonsales,youcould
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
14
traditionallybuildamodel,forexample,usingfivevariablesbutnow,youcandothisusing500andsimilarlythisisallthestoragecapacity.Now,itactuallydoescostyouacoupleofhundredofdollarstopurchaseaharddrivethatcanholdalloftheworld'smusic.
Whatdowemeanby"velocity"and"variety"?
So,wejusttalkedaboutthe"3Vs"andwejustcoveredvolumeandtheothertwo"Vs"arevelocityandvariety.
So,velocityisreallythespeedofdataanditmeansthatwithrealtimeinformation,wecanmakemuchmoreaggressivemarketingdecisionsthanwithtraditionalmethods.
Let'sjusttakethisexamplethatI'mindowntownIthaca,it'saboutlunchtime,andnowwhatcouldhappenisthatIcouldgetacoupononmycellphonefromarestaurantthatisnearbythatmatchesmytaste.ThisismuchmoreeffectivethanthecouponinthenewspaperthatisontheweekendandthatIneededforcuttobringalong.Alsobecauseofitsimmediacy,it'ssortofreallycapturesthecontextinthatI'mcurrentlyin.
Anotherexamplewouldbe,youknow,takeashoppingmall.Theymayhavecamerasinstalledthatmeasuretrafficinandoutofstores.Whatthisdoesisthatitallowsyounowtogetanestimateofsalesofthestoresevenbeforetheyhavepublishedanysalesfiguresanditcouldbeveryprofitableforthefinancialindustry.
So,thelast"V"isvarietyandthisisalsoimportantbecauseifyoulooktraditionally,sothatinrelationaldatabases,youknow,youhadyourrowsandcolumnsandyourcustomersandemployeesanddepartmentsandmanagersandnow,thevalueonbigdatacomesupbylookingonlyatonedatasourcebutatthisunionofallthedatasourcesortheintegrateddatasources,andthisnowcomesfromdatafromdiscussionforumsandblogsandGPSdatafrommobiledevicesorevenvideosandphotospostedintweetsandsocialmediaandaccesstowebsites.Sothereisalotofvalueinintegratingthedatathatallofthiswithhugevarietyandcreatingvaluefromit.
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
15
Transcript: Ask The Expert: Marco Vriens on Big Data and Conjoint Analysis
Howdobigdatarelatetoconjointanalysis?
Sonowadaysthere'salotoftalkaboutbigdata,andobviouslybigdatahasanincrediblepotentialtoinformmarketingpeople.Butforthesetypesofproblemsbigdataisactuallynotverysuitedbecausebigdataonlyprovidewhat'salreadyintheworldandcangiveyou,likehowdopeoplerespondtowhat'salreadythere.Butconjointactuallyallowsyoutogetinsightintothingsthatdon'tevenexistyet.SoIjustwanttopointthatout,becausethat'sthediscussionfortoday,andIthinkthat'soneofthereasonswhyIthinkeverybodywhopursuesadegreeinmarketingshouldknowaboutdiscretechoiceandconjointmethods.
Willbigdataanalysisreplaceconjointanalysis?Bigdata,Ithink,isanemergingfield,thereareacoupleofmarketingapplications,thatIthinkforwhichbigdataareincrediblyuseful.Itmaynotbepartofthiscourse,butthingslikeamarketingmixanalysis,determininghowefficientyourmarketingactivitiesare,thatisanarea,Ithink,bigdatacanhelp.Eveninproductdevelopmentwhereyou'retryingtounderstandlike,okay,whatarethetrendsandwhatattributesandwhatfeaturespeoplelikeandhowdoesthatevolveovertime,thatscenarioIthinkbigdatacanhelp.Keepinmindthatconjointisasnapshot.I'minterviewingpeopletodayandI'mgettinginsightintowhatpeopleliketoday.Butitdoesn'tnecessarilytellmehowdatawillevolveovertime,andsothat'swhybigdatawouldcomeintoplay.Andsotosomedegreeaconjointapproachandbigdataaresupplementaryifanything,butit'snotlikebigdataaregoingtoreplacewhatyouwoulddoinconjoint,butitcouldsupplementit.
Transcript: Steps in Conjoint Analysis ThefirstimportantstepindesigningaconjointstudyisselectingtheattributesandlevelstobeincludedinthestudyandIwanttotalkaboutsomeoftheconsiderationsthatgointothisstep.Thefirstconsiderationisthatit'softenusefultouseconsumer'sinputtoidentifytheimportantattributestobeincludedinthestudyandthiscanbedoneindifferentways,butoftentheuse
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
16
offocusgroupsisusedforidentifyingattributes.Attributesintheconjointstudycanbehardattributes.Forinstance,inacomputerstudy,thememorymaybeahardattribute,buttheycouldbemoodsoftattributessuchasthebrandnameofproductswhichcanbeincludedasanattributeinthestudy.Oneusefulwaytothinkaboutidentifyingattributesofproductistoprofilethecurrentsetofproductsavailableinthemarket.Now,theresearcherneedstofindrealisticwaystocommunicatetheattributesandthelevelsofthoseattributestorespondentsinthestudyandthatcanbedoneindifferentways.Audiovisualaidsareoftenuseful.Forinstance,carcompaniesthatusedconjointanalysistohelpthemdesignnewcarswilluseprototypesofcarsandthenhavetherespondentssitintheprototypetobeabletoexperienceattributeslikelegroomorheadroomandtheseareattributesthatarenoteasilycommunicatedthroughdescriptionbutarecommunicatedbetterthroughexperience.Theuseofreferenceproductisalsoanothermechanismtocommunicateattributesandlevels.Now,itisimportanttokeepinmindthattheattributesandlevelsthatareincludedinaconjointstudymustbeactionlevelbythebusinessthat'sdoingthestudy.Otherwise,therewouldnotbe-itwouldnotbeaveryusefulstudy.It'salsoimportanttokeepinmindthatweshouldhaverealisticlevelsofattributesinastudyandexampleisprice.Ifpriceisanattribute,that'sincludedinthestudy,wedon'twanttoincludelevelsofpricethataresolowthatthefirmcannotactuallydelivertheproductatthoseprices.Ifwedoincludeattributelevelsthatunrealistic,thenwe'llfindthattheinferredimportanceoftheattributewillbeinflated.Letmenextuseanexampletoillustratesomeofthesepoints.Now,whatyouhavehereisdataforaprescriptiondrugmarketinwhichaformofstrengtheneddesignatconjointstudyandtherespondentsinthestudyaregoingtobephysicians.Thephysiciansarequitefamiliarwiththismarketandtherearefourexistingproductsinthismarket;alpha,beta,gamma,anddelta.Ifwelookatthelastrowhere,thelastrowshowsthecurrentmarketsharesoftheproductsandwefindthedeltaisthedominantproductinthismarketwhichhasa70%marketshareandasaresultofthat,mostphysiciansarequitefamiliarwiththeseproductdelta.Now,whenweprofiletheseproductsinthemarket,wecomeupwiththenumberofattributesoftheseproducts.Let'stakeoneoftheseattributesasanexample,durationofsideeffectsofthedrugsisanattributeandinthemarketatpresent,thelevelsoftheseattributeareoneday,twodays,orthreedaysandthesegivestheresearcherssomeindicationofwhatthelevelsinthestudyshouldbe.Wealsonoticedthatanattributelikeclinicalcureratehasbeendescribedbyreferencetoaproductthatphysiciansarefamiliarwiththatisincomparisonwithdelta.So,itcouldbe10%belowtheclinicalcurerateofdeltaoritcouldbe10%abovetheclinicalcurerateofdeltaoritcouldbeequaltotheclinicalcurerateofdelta.So,thismakesitabetterwayto
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
17
communicatethisattributetothephysician.Letmealsopointoutthedrugcostswhichatthemomentinthemarketisattwodifferentlevels,either$44.60or$58.85.Takingthisdata,nowlet'sseetheattributesinlevelsthatwereactuallyusedbythefirminthatconjointstudy.Thesearetheattributelevelsactuallyusedinthedrugstudyandwefindthatthesameattributesthatexistsinthemarkethavebeenused.Attributeshavedifferentnumbersoflevels,sosomeattributesareatfourlevels,forinstance,1and2and3.Thereisanotherattributewhichisat3levels.So,attributescouldbeadifferentnumberoflevels.Letmefocusondrugcost.Wesawbeforethatthecurrentproductsinthemarketwereeitherat$44.60orat$58.85.Inthisconjointstudy,thefirmdecidedtousealevelofdrugcostslightlylowerthanwhatisavailableinthemarketandalsoslightlyhigherthanwhat'savailableinthemarket,buttheydidnotgotoofaroutsidetherangeofcurrentlyavailableproductsinthemarket.Transcript: Collecting Data Thenextstepinthedesignoftheconjointstudyistousetheattributesandlevelsthathavebeenidentifiedtodevelophypotheticalproductsthatwillbeshowntotherespondentandtodecidehowtherespondentwillbeaskedtoevaluatetheseproducts.Now,usingtheattributesandlevelswecancreatethesetofallpossibleproductsthatcanbecreated.Thissetiscalledthefullfactorial.Usuallythefullfactorialistoolarge,inotherwordsithasjusttoomanyproductstoaskthedespondenttoevaluateinamarketresearchsetting.Sowhatwedois,weselectasubsetoftheseproductstoshowtotherespondent,thissubsetiscalledthefractionalfactorial,andwehaveusuallystatisticalsoftwarethatweusetoidentifythesubset.Thesubsetofproductsischosensoastohaveareasonablenumberofproductsforthesurvey,thatisnottoomanyfortherespondenttoevaluateButenoughtobeabletorunstatisticalanalysesoneachrespondentdataseparately.Nowwhatdoweactuallyasktherespondenttodo?Inmetricconjointanalysis,weasktherespondtoeitherratetheproducts,orweasktherespondenttoranktheproducts.Ifthesetofproductstobeshowntotherespondentislarge,thentherankorderingofproductsisdifficultfortherespondentandwemightoftengointhatsituationtoasktherespondenttoratetheproducts.Transcript: Conjoint Surveys
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
18
Next,let'stalkabouthowconjointdataarecollectedandanalyzed.Conjointdataarecollectedtypicallyfromasampleofrespondentsfromthetargetmarketbymarketresearchfirms.Aconjointsurveyhastwopartstypically.Infirstpart,respondentsareaskedtoprovidetheirjudgmentofhypotheticalproducts;ratingsorrankings.Inthesecondpart,wecollectinformationabouttherespondent'sdemographiccharacteristics,thepurchasinghabitsandbehaviorsandsoforth,sothatwecanlinkpreferencesthatwehavecollectedintheconjointstudywiththecharacteristicsoftherespondents.Howareconjointserviceadministered?Theyaretypicallyadministeredremotelyontheweb,butdependingoncircumstancessuchastheaccessibilityoftherespondentsandthecomplexityofthetaskthattheyarebeingaskedtodo,theymaybeadministeredonthephoneorinperson.Whenwecometothestageofanalyzingconjointdata,inmetricconjoint,weanalyzethedataforeachrespondentseparately.Whatthisdoesisitallowsforcompletelyidiosyncraticpreferencesofeachrespondent.Notworespondentsneedtobe-needtohavethesamepreferencesfortheattributesinthecategory.Thegoalofthisanalysisistoobtainthepart-worthsortheutilitiesthatareassociatedwitheachattributelevelandtheimportanceofeachattribute.ThisistypicallydoneusingregressionanalysisandIillustratetheuseofregressionanalysistoanalyzeconjointdatausingthecarpetcleanerexample.Transcript: Case Study: Carpet Cleaners
Soasanexample,wehaveastudyhereinwhichcarpetcleanerproductswerestartedinthemarket.Asaconjointstudy,fiveattributeswereidentifiedusingfocusgroupsandtheseattributesarepackagedesign,brandname,price,aGoodHousekeepingseal,andamoney-backguarantee,andasyoucanseehere,theseattributeshavebeendefinedonacertainnumberoflevels.PackagedesignisonthreelevelswhicharecalledA,B,andC.BrandnameisinthreelevelsaswellwhicharecalledCarbona,Resolve,andWoolite.Pricesatthreelevels:$4.99,$5.39,and$5.79.GoodHousekeepingsealisyesorno.It'spresentornotpresent.Andamoneybackguaranteeagainisyesorno.so,ifyouthinkaboutthetotalnumberofpossiblecarpetcleaners-hypotheticalcarpetcleanersthatcanbecontractedusingthesefiveattributesandthelevels,thatnumberisgivenbytheproductofthenumberoflevelsherewhichis108.
Now,thatofcourseisaverylargenumberandinanystudy,wecan'treallyexpectarespondenttoevaluatethe108productsbecausethatistoomany.So,inthisparticularstudy,18products,asubsortof18productswaschosentoshowtoeachrespondentandtherespondentwasaskedtoevaluateeachofthese18carpetcleaners.Thetaskgiventothemwas
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
19
torankedorderthe18carpetcleanerswith18beingthemostattractable,themostpreferredcarpetcleanerandonebeingtheleastattractivecarpetcleaneranddatawerecollectedfromeachrespondentinthisfashion.
Transcript: Analyzing Data Using Regression So,inanalyzingthedatathathadbeengatheredfromtherespondents,wewanttobeabletodecomposetheoverallpreferenceforacarpetcleanerintothepreferencesforthefiveattributesthatmakeupthecarpetcleaner.Forinstance,considertwocarpetcleanersthatareidenticalinallaspectsexceptforthebrandnameandonecarpetcleanerthatiscalledResolveisrankedhigherthantheonethatiscalledWoolitebyaparticularrespondent.Then,wecaninferthatthisrespondentprobablyprefersthebrandnameResolvetoWoolite.Andnowifyouanalyzeall18judgmentsofrespondenttogether,wewilluseastatisticaltechniqueorregressionandtodemonstratethat,letmeshowyouinanexcelspreadsheetthedataofonerespondent.So,whatyouseehereisthedatafromonerespondentinwhichwehave18carpetcleanersorprofilesthatwasshowntotherespondent.Eachofthesecarpetcleanerisdescribedonthefiveattributes;packagedesign,brandname,price,GoodHousekeepingseal,andmoney-backguarantee.Andwedeceivetherespondent'sevaluationforeachofthesecarpetcleanerswhere18isbestandsowefindthattheseparticularcarpetcleaner,thelastoneisrankedasthemostattractiveandthecarpetcleanernumber10isconsideredtheleastattractive.Now,wetakethisdataandweconvertthemaswehaveseenintodummyvariables.So,wehavethesamedataherebutthefiveattributesofthecarpetcleanershavebeencodedasdummyvariables.Ourgoalnowistotaketherespondent'sevaluationsofthe18carpetcleanersanddecomposethoseevaluationsintothepreferencesforthefiveunderlyingattributesandtodothatasIsaidwe'llusemultipleregressionstoanalyzethedata.Nowinexcelyoucanrunmultipleregressionbygoinguntothedatatabandlookingforthedataanalysistool.Whenyouclickondataanalysis,yougetanumberofpotentialanalysistoolsandwe'llusetheonecalledregression.So,ifyouclickonregressionandclickokay,you'llgetawindowinwhichyouhavetofirstspecifytheinputbyrangewhichisthiscaseistherespondent'sevaluationsforthe18carpetcleaners.So,I'mgoingtotakethese18evaluationsandspecifytheseasbeingtheinputbyrange.NoticethatIhaveincludedthefirstcellwhichisthelabelrespondentevaluations.I'mthengoingtospecifytheinputx-rangewhichisthedummyvariablescorrespondingtothefive
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
20
attributesandthesedummyvariablesarespecifiedhere.Andso,thesedatanowaretheXvariables.Sincewedohavelabelsinthedata,I'mgoingtochecktheboxcalledlabels.I'mgoingtospecifyanoutputrangewhichiswheretheoutputfromtheregressionshouldappearandinordertothat,I'mgoingtochooseanewworksheet.Let'ssaywespecifytheoutputappearhereandthat'ssortofallthatweneedtospecify,ifyouclickokay,wegettheoutputfromtheregression.Now,IhavealreadydonethisregressionbeforeandI'mgoingtoconcentrateinthisareawhereIhaveobtainedtheresultsoftheregression.Whatdoweseehere?Wefindthatthisregressionhad18observations.IthasanRsquareof0.98soithasaverygoodfittothedataandthesearethecoefficientofthedifferentdummyvariableswhichrepresentthevaluesystemofthisconsumerwhichisthenumberswe'retotakeandinterpretnowintermsofutilitiesassociatedwiththeattributesunderlyingthecarpetcleaners.Transcript: Interpreting Results
Soasyourecallwehavecollecteddatafromonerespondentinamarketresearchstudy,wheretherespondentwasshown18hypotheticalcarpetcleaners,eachcarpetcleanerwasdescribedinfiveattributes,andtherespondentwasaskedtoprovideajudgmentintheformofrankorderingofthesecarpetcleaners.Wetookthisdatafromtherespondentandrunaregressionanalysisonthedataandthecoefficientinthatregressionarethepartworthsofthisrespondent.Sowearegoingtousethesepartworthstounderstandthiscustomer'spreferences.Solet'stakealookatthepartworths.
WhatIhaveinthespreadsheetherearethefiveattributesonwhichthecarpetcleanersweredescribed.PackageDesign,BrandName,Price,GoodHousekeepingSeal,andMoneyBackGuarantee.AndforeachattributeIhavethelevelsoftheattributes.
Forinstance,forpackagedesignthethreelevelswereABandC.Inthecolumnlabelutilitieswehavetheestimatedpartworthswhicharenothingbutthecoefficientsfromtheregressionanalysis.
Let'sseewhatthesecoefficientsorpartworthstellusaboutthisrespondent'spreferences.Let'stakealookfirstatthepartworthsforpackagedesign.
WeseethatthisrespondentpreferspackagedesignedBthemostfollowedbypackagedesignCfollowedbypackagedesignA.Notethatthisinformationwasnotavailabletousuntilafterwehadcollectedthisrespondent'sdataandruntheregressionanalysis.Therespondentwasnot
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
21
directlyaskedtostatewhichofthreepackagedesignswasmostpreferred.Similarlylet'slookatthepartworthsforprice.Weseethatasthepricegoesupfrom$4.99to$5.39to$5.79thepartworthorutilitygoesdownwhichmakessense.Therespondentpreferslowerpricestohigherprices.Butweseethatthepriceincreasesresultinutilitydecreasesinaparticularfashion.Whenthepricegoesupbythefirst40centsfrom$4.99to$5.39utilitydropsby2.83units.Whenpricegoesupbythenext40centsfrom$5.39to$5.79utilitydropsbymorethan2.83units,infactitdropsby4.84unitswhichisthedifferencebetween7.67and2.83.Whatwehavejustlearnedisthatpriceincreasesresultinanon-lineardecreaseinutility.Let'sconductthefollowingthoughtexperimenttoseefurtherwhatwelearnedaboutthisrespondentspreferences.Let'sconsideracarpetcleanerwhichisatthepriceoffivedollarsandthirtyninecents,anddoesnothaveamoneybackguaranteeassociatedwithit.Nowlet'saskthefollowingquestion,ifweweretoraisethepricebyfortycents,inotherwordsraisethepricetofiveseventynine.Andgivetherespondentamoneybackguaranteeonthecarpetcleaner.Wouldtherespondentbebetterofforworseoff.Wellwhatweseeisthatwhenthepricegoesupfortycents,theutilitydropsbyfourpointeightfourunits.Whenthemoneybackguaranteeisprovided,utilitygoesupbyfourpointfiveunits.Sincefourpointfiveislessthanfourpointeightfour,weconcludethattothisrespondentthemoneybackguaranteeisnotworthfortycents.Alloftheseareinferencesthatwedrewabouttherespondentspreferenceswithoutdirectlyaskingtherespondent.
Nextlet'slookattheimportancesassociatedwiththedifferentattributes.Tocomputetheimportanceassociatedwithanattribute,welookattherangeofthepartworthswithintheattribute.Forinstance,let'sconsiderpackagedesign.Therangeofthepartworthisthatthelowestpartworthiszero,andthehighestpartworthiseight.Thisgivesusarangeofeight.
Similarlylet'slookatbrandname,thelowestpartworthisminuspointfiveandthehighestpartworthisonepointfive.Thatgivesusarangeoftwo.Whatweseefromthisisthatifthepackagedesignweretobechangedfromitsworstlevel,whichisAinthiscase,toitsbestlevel,whichisBinthiscase.
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
22
Wecanswingtherespondentsutilitybyeightunitswhichiswhyweconsiderthisnumbereighttobeanapproximatemeasureoftheimportanceofthisattribute.Tocomputeameasure,arelativemeasureofimportance,wesumtheserangestogetatotalrangeoftwentythreepointsixseven.Andwethencomputethefractioneightdividedbytwentythreepointsixseven,whichispointthreefour.Similarly,twodividedbytwentythreepointsixsevenispointzeroeight.Thesenumbersinparenthesesareinterpretedasthemeasureofrelativeimportance.Sowhatweseehereisthatpackagedesignisthemostimportantattributetothisrespondedwithanimportanceofapproximatelythirtyfourpercent.Bycontrast,theleastimportantattributeforthisrespondentisthegoodhousekeepingsealwithanimportanceofonlysixpercent.Transcript: Ask The Expert: Johannes Gehrke on Advanced Marketing Research
Howcanbigdatahelpwithdecisionmakinginmarketing?
So,inthiscourse,wetalkedalotaboutdecisionmakinginmarketing.So,howcanbigdatahelpwithit?Well,traditionallydecisionswereveryoftenmadebasedonintuitionorpeople'sopinionsorexperienceandoftenthentheHiPPOmadethedecision(wheretheHiPPOistheHighestPaidPerson'sOpinion).Andthere'ssomegoodexamplewhydatamaytellyouthatthisisnottherightthingtodo.
Forexample,afamousexampleisatAmazon,whereassumeyoujusttookanitemandyouaddittoyourshoppingcartatyourwebsite.Now,mostsitesusedtoshowthecartafterwardswiththegoalthatwhileyouputthisitemuntothecartandthenyouwouldgoaheadandcheckoutandpay.AndatAmazon,redlinkinghadtheideaofactuallyyouwouldshowrecommendationsbasedontheitemsinyourcartsothatyouwouldactuallyputmoreitemsintothecart.Theadvantageofthatwouldbethatwhileyoucould,maybesellmoreitems,youwillhaveamuchbiggeraveragebasketsizeafterall.Thedisadvantagewouldbewhileitmaydistractpeople
fromcheckingout,sotheymayactuallyabandontheshoppingcartwhichwouldreduceconversion,andtherefore,reducedonprofits.Andsoatthatpointintime,theHighestPaidPerson'sOpinionwastostoptheproject.Thisisnotworthwhiledoing,butactuallythanjust,averysimpleexperimentwasrun,itwasreallyhighlysuccessfulandtherestisnowhistory.Youknow,actually,whereveryougoandyoushop,Imeanyouaddsomethingtotheshoppingcart,yournearlyallofdecidedpickwillshowyourecommendations.Andthisactuallyisansample
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
23
wheretheywasaHiPPOofthisandthendataandasimpleexperimentwasusedtoactuallyhelpinthedecisionmaking.
HowcouldyouconductasimpleA/Bexperimentonyourwebsite?
So,inthepreviousclip,wetalkedaboutthisexperiment,whereweactuallyusedaverysmallexperimenttomeasuretheeffectivenessofapossibleaction.So,assumethatyouhaveawebsitenowthroughwhichyousoldyourproducts.Howwouldyouactuallydosuchasanexperiment?Howdoesthiswork?
So,let'sjustlookbackandseehowactuallywhatwouldbeasimplewayofthinkingaboutthisisthatwheneverarequestcomesin,itgoestosomeserverwithintheinfrastructureofthiswebsite,andthenthatservesitoutforasetofservicesintheenterprisetheresultpagethengetsassembledandit'sreturnedtoyou.So,thisisITinfrastructurebehinditthatdoesallofthis.
Now,inwhatiscalledABtesting,youactuallyrandomlysplitthetrafficbetweentwodifferentversionsofthewebsite,forexamplethecurrentwebsiteandthenanotherwebsitewhereyouchangesomethingbutactuallysomethingjustvery,verysmall.
Forexampleyoumayhavechangedthealgorithmthatyouusedtodeterminewhatitemsarerelatedtotheitemthatyou'recurrentlyshowing.Youmayhavechangedtheadsonthepage.Youmaychangetherecommendationsonthepageandsothisistheonlychange,soyoudon'tchangethesitelayoutoranythingelse.Andthenwhatyoualsohaveissomeclearmetricsofwhattomeasure.
Forexample,inyourwebsite,itcouldbeassimpleas,youknow,howmanyitemshaveyousoldandwhatistheprofitmarginonthem.Forexampleingames,itcouldbetheengagement.Youknow,coulditbein-gamepurchases,thetotalminutesplayed,thenumberofadsclicked.Onacarrentalwebsite,itcouldbethetimetocompletionoftherental,theconversionpercentage,thenumberofupgrades,anythingelse.Andnowthatyouhavetraffic,yousplititintotwoparts.OnethatgoestositeAandtheotheronethatgoestotheinternalothersiteBwhereyouarerunningthesmallexperiments.Andnowyougoaheadandmeasure.Andespeciallyifyourproductisdigitalasinthegameexampleaboveorinthewebsite,thenyoucannotonlyshowtwodifferentwaysofpackagingitbutyoucouldshowlotsofdifferentways
ofpackagingitandactuallynowthesearetwodifferentproductsandyouseetheimpactdirectlyoneachofthemonyourcustomers.
Whataresomeofthepitfallswiththeseonlineexperiments?
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
24
So,whataresomeofthepitfallswiththeseonlineexperiments?Sofirstofall,hereinthisnewonlineworld,it'sbettertohavelotsandlotsofideasthatarecheaptoimplementandthenthrowoutthosethatdon'twork.Andtheimportantthingisthatyouactuallywanttotrytomeasureallofthembecauseeveningeneral,youmaybewrongmanytimes,butiftheseareallexperiments,youareonlygoingtoimplementthosethatweresuccessful.
Soeven,forexample,youtakeahundredexperiments,90ofthemwereunsuccessfuland10ofthemweresuccessful.Now,thevalueoftheoverallpropertyoryourwebsiteanddigitalproducthasincreasedsignificantlybecausetheonlythingsthatyouimplementwerethesuccessfulsteps.Thisisreallydonealsowithinbigcompanies.
Forexample,in2009,Googlerunmorethan12,000APtestsandonlyabout10%oftheseweresuccessful.
What'stheimportanceofvalidatingyourideaswithdata?
So,oneofthe,oneoftheotherpitfallsthatyouhavewithonlineexperimentsisthatyouhavetoknowwhatareyourmetrics.Themetricswilldeterminewhetheryourexperimentwassuccessful.
Forexample,ifyousetanexperimenttocrosssell,thenyoushouldreallymeasurewhatthecustomersactuallypurchasesomethinginyourcategoryratherthanwiththeoverallpurchasesgotorwhether,forexample,itboostssalesinexistingcategories.Andthisisalsoimportantfromapeoplemanagementpointofview.So,productmanagersusedtodecideonthenextsetoffeaturestoimplement.Butnow,youhavetogettothemindsetthat,whileyoudon'treallyknowwhatisgoingtoworkornot,butactuallyitismuchbettertoactuallycomeoutwithlotsofdifferentwaysoftryingthingsout,especiallycheaplyandthenmeasure,andtheneachraidoutthisprocess.
So,especiallyverysuccessfulpartsofthecompanythathaveahistoryofsuccessmaybeveryresistanttothisbecausewhilewe'veallstudiedthisway,we'vebeensuccessfulwiththis,but-sotheleastyoucandointhiscaseisatleasttogoaheadandvalidateyourideaswithdata.Sonow,youhaveanideathatyouimplementedandwiththisyounowgoaheadandmeasurewhethertheintroductionwasright.Thismaynotbethefirststepintogettingintothestrictestdata-drivenmindsetwhereactuallyyouknowhowtodrivethedecisionprocesswithdataand
yougoaheadtrylotsofideasandlotsofcheapways,andthenyoumeasureandthenyoumakeyourfinaldecisionbasedonthesemeasurements.
Whatisadatascientist?
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
25
So,oneofthetermsyoualsohearalotwhenpeopletalkaboutbigdataisthenotionofadatascientist.Sowhatisadatascientist?Andyouknow,simplyexplainedit'sthepersonwithinanorganizationwhomakesdiscoveriesindata.Andthisreallymeansgoingthroughallofthedifferentdiscoverysteps.
Firstofall,findingtherightdataindataenterpriseorexternaldatathenintegratingit,youknow,itmaynotbeasniceandrelationalasthetraditionaldataintermsofrowsandcolumns.Theremaybelotsofbaddataformatsorbogusdata,youknow.Youmayhavelotsofconfusingdatafieldslikesynonymsanddifferencesandthengoingthroughthewholeextractionprocessofintegratingit.
So,asimpleexample,let'sassumedyouwantstomeasuresentimentabouttheproductbylookingatTwittertweets.Now,thetoolsavailablethatactuallytaketweets,andthentrytoextractnamesofproductsespeciallygivenyourownproductdatabase,buttheyallneedtobeintegratedbetweentheoverallflowofthedatascienceexperimentsthatyouwanttocreate.Andthenneedtobefedwithyourproductnamesandtherealsomaybeotherprofessionalwebsitesinwhichyourproductisdiscussed.
So,youneedtoagain,takethistextandthentrytoanalyzewhetherthisdiscussionispositiveornegativeorwhatthediscussionsareabout.Andnoneofthiscanbedoneinamanualwaybecauseitdoesn'tscale.Andthenofcourseintheend,thedatascientistbuildsmodelsandlotsofexperiments.So,whatthismeansisthatactually,weneedaskillofsomebodywhohassortofacombinationofstatistician,butalsoasortofahackerandalsosomebodywhohasagoodunderstandingofthebusinesssitetoknowwhattherightquestionsareandalsoinsubstanceandintheend,playtherolesofaconsultanttothedecisionmakers.IthinkthisisabigdifferencebecauseifyoulookattheStatisticsDepartmentinbigcompanies,theywereassignedveryspecific,narrowlydefinedtasksandtheyalsonotethatmuchcommunicationbetweenthepeoplewhomakethestrategicdecisionsandthepeoplewhodockintothedata.Buttoday,datascientists,theyareveryhighvaluepeople.Theyaremuchbetterintegrated.Theycommunicatetheirfindingstoproductmanagers,toexecutives,andpeopleinoperations.Andtheyalsohaveagoodunderstandingofthevalueoftheirfindingsonproductsanddecisionsandcan,therefore,drivealloftheanalysisthattheyaredoingthemselves.
Transcript: Simulations Sonext,wewilltalkabouthowtheestimatedpreferences,orpart-worths,ofrespondentsareusedtosimulatewhatmighthappeninthemarket.Weestimatedthepreferenceofeachrespondentinthesampleseparately,andwestoredtheminadatabase.Wecannowusethese
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
26
part-worthstoanalyzetheimpactofbusinessdecisions,suchasaddinganewproductintothemarket,ordeletinganexistingproductfromthemarket,orchangingthelevelofanattributesuchasthepriceoftheproduct.Thebasicapproachinallthesecasesistosimulatetheoutcomeforeachindividualinthesampleandthenaggregatethepredictedoutcomesacrosstheindividuals.Let'sgobacktothecarpetcleanerexamplethatwehavetalkedaboutpreviously.Inthatcarpetcleanerexample,wehadthepreferencesofonerespondentthatwehadobtainedusingregressionanalysis.Imaginenowthatthemarketconsistsofthreecarpetcleaners,eachofwhichcanbedescribedasabundleofthefiveattributesinthatexample.Thesethreecarpetcleanersmaybeexistingproducts,ortheymaybehypotheticalproductsthatthecompanyisconsideringincludinginthemarket.So,thequestionofinterestis,forthisparticularrespondentwhosepreferenceswehaveobtainedusingconjoint,canwesaysomethingaboutwhichofthethreecarpetcleanersthisindividualwillchoose?Inotherwords,wewouldliketobeabletopredictthisindividual'schoiceusingtheestimatedpreferencesorpart-worths.Let'sgototheExcelfilewherewehavethedata.Nowinthisdatawehavethethreehypotheticalproductsthatareeithercurrentproductsinthemarketornewproductsthatthefirmisconsideringincluding.Let'stakeanexamplewiththefirstproducthere,aspackagedesignB.Ithasbrandname"Resolve."Itisatthepriceof$5.39.ItdoesnothaveaGoodHousekeepingSeal,anditdoesnothaveamoney-backguarantee.Similarly,therearetwootherproductsinthemarket.Nownext,wehavethedummyvariablecodingoftheseattributesoftheproduct,whichissomethingwehaveseenbefore.Recallthatwehadalreadyestimatedthepreferencesofthisonerespondentusingconjointanalysis,andthesearethepartworth'softhisrespondentthatwehaveobtainedpreviously.Now,thequestionofinterestis,foreachofthesethreeproducts,whatisourpredictionoftheutilitythatthisrespondentwillgetfromthatcarpetcleaner.Howdoweobtainthat?Let'stakethefirstcarpetcleanerhere.Theutilitythatthisrespondentwillgetfromthiscarpetcleanerwillbethesumoftheutilityfromeachoftheunderlyingattributes.Forinstance,forpackagedesignB,theutilitywillbe8unitsofutility.ForbrandnameResolve,theutilityis-0.5.Forthepriceof$5.39,theutilityis-2.83.AndnothavingaGoodHousekeepingSealmeansthattheutilityiszero,andnothavingamoney-backguaranteemeansthattheutilityiszero.So,wesumtogether6.5,whichistheintercept,8unitsminus0.5andminus2.83,whichgivesus11.17unitsofutility.So,whatwearesayingisthatfromthisfirstcarpetcleaner,thisrespondentwillget11.17unitsofutility.Similarly,wecancomputetheutilityfromthesecondproductandthethirdproduct,andthoseutilitiesare15.67and9.33units,respectively.Thenextquestionofinterestis,ifthesearetheutilitiestheindividualgetfromthreecarpetcleaners,whichcarpetcleanerwillthisindividualchoose?Theassumptionwemakeisthattheindividualchoosesthecarpetcleanerthatgivesthispersonthehighestutility,whichinthiscaseisthesecondcarpet
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
27
cleaner.Andthereforethepredictedchoiceisthatthisindividualwillpickthesecondcarpetcleanerinthemarketandwillnotpickthefirstandthethirdcarpetcleaners.Wecandothisexerciseforeveryrespondentinourstudyandthensumuptheirchoicestoobtainthetotalnumberofunitsthatwillbesoldforeachofthethreecarpetcleanersinthismarket.Transcript: Choice-based Conjoint
So,sofar,wehavetalkedaboutoneformofconjointwhichiscalledfullprofilemetricconjoint.Thisformofconjointisattractivebecausewecanestimatethepart-worths,orutilities,separatelyforeachrespondent,therebyallowingeveryrespondent'spreferencestobecompletelydifferent.Thismighthave(doeshave)adrawbackandthedrawbackisthatwhatwecanpredictinthismethodispreferences,notchoices.Andthenwehavetoconvertthepreferencestochoicesinsomekindofsomewhatad-hocmanner.
Analternativeapproachistodirectlycollectfromconsumerschoicesinsteadofratingsorrankings.Thisapproachiscalledchoice-basedconjointandisnowaverypopularformofconjoint.Sohowdoeschoice-basedconjointwork?Inchoice-basedconjoint,therespondentispresentedwithachoiceset.Inthischoiceset,therearealternativeproductsandtherespondentisaskedtoindicateadiscretechoice.Whichoneoftheseproductswouldtherespondentchoose?Thetaskisthenrepeatedwithseveralchoicesets,let'ssay12choicesets,andtherefore,wehave12observationsfromeveryindividualwhichareevaluationsoftheproducts.Theanalysisofchoicedatanowhastobedoneusingachoicemodel.Anexampleofachoicemodelisalogitmodelinsteadofusingregressionanalysis.
So,whatisthebigadvantageofchoice-basedconjoint?Thebigadvantageisthatthisisamorerealisticmarketresearchtask.Choicesishowconsumersexpresstheirpreferencesinthemarketandsothisismorerealistic.Thedisadvantageofchoice-basedconjointisthatitisoftendifficulttoestimatepart-worthsorutilitiesseparatelyforeachrespondent.Thedatahavetobepulledacrossrespondentstoestimatetheconjointmodel.
Letmeshowyouanexampleofachoice-basedconjointtaskfortherespondent.Whatyouseehereisataskthatwasgiventorespondentinachoice-basedsettingwheretherespondentwasshownfourluxurySUVs.EachoftheseSUVsisdescribedonseveralattributesandsowehaveeightattributesonwhicheachoftheseSUVshasbeendescribed.Therespondent'staskistoanswerthequestion:Whichvehiclewouldyoubemostlikelytopurchase?AndtheyexpresstheanswerofthisquestionbypickingoneofthefourluxurySUVsthattheyareshown.OncetheypickanSUV,thetaskiscompletedandtheymoveontothesecondtaskinwhichfour
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
28
differentluxurySUVsarenowshownandtheymakeadiscretechoiceagain,andsoon.Andinthisfashion,eightchoicesoftherespondentarepicked.
Transcript: Managerial Uses of Conjoint
Conjointanalysisisusedextensivelybymarketingmanagerstomakeimportantbusinessdecisions.Therearefourbroadareasinwhichconjointanalysisresultsgetused.
Oneisobviouslyproductdesign.Conjointisveryusefulfordetermininghowtodesignproductsandservicesthatappealtoconsumers.Asecondapplicationofconjointisinforecastingthemarketshareofnewproductconceptsthathadbeendesignedbythefirm.Athirdapplicationisinsegmentationandtargeting.Thisapplicationisirrelevantbecauseinconjoint,weareabletoobtainpreferencesforindividualconsumersandtherefore,wecangroupconsumersbasedupontheirpreferencesafterconjointdatahavebeenobtained.
Afourthapplicationispricing.Oneoftheimportantoutputsfromconjointisthatwecanobtainthewillingnesstopayfordifferentattributelevelsinthestudyandthereforethisallowsthefirmtodeterminehowtopricenewproductandservices.Becauseofthesewidespreadapplications,wefindthatconjointhasbeenusedinmanyapplications.Thisincludesconsumerdurablessuchasautomobiles,digitalcameras,smartphones,highdefinitiontelevision,andsoon;consumernon-durables,suchassoapsandshampoosandfoodproducts;anumberofindustrialproducts,suchasaircraft,computersoftware,forklifttrucks;serviceslikecarrentals,hotels,andotherapplications,suchasMBAjobchoices,banneradvertisingintheInternetcontext,andsoon.
Let'sthinknowaboutthekindsofproblemsituationsinwhichconjointismostapplicable.Herearesomeconsiderations.Oneisthattheproductthatwearedealingwithmustberealisticallydecomposableintobasicattributes.Thishappensmosteasilyinproductssuchasconsumerelectronicproducts(digitalcameras,laptops,smartphones,andsoon)butitdoeshappeninanumberofotherproductsituations.
Thesecondconsiderationisthattheproductchoiceshouldbeareasoned,highstakesdecision.Inotherwords,conjointisnotaveryusefultechniquewhentheproductisanimpulsepurchasekindofproduct.Athirdconsiderationisthatfactorialcombinationsofbasicattributelevelsshouldmakesenseandifthat'sthecaseintheproductcategory,thenconjointisaveryapplicabletechnique.Anotherconsiderationisthatproductscanberealisticallydescribedverballyorpictorially.Thisisimportantbecauserespondentsareprovidingtheirevaluationsofproductsbasedonthestimulithattheyseeandthestimulineedstoberealistic.Andthefinal
LSM551:MeasuringCustomerPreferencesSamuelCurtisJohnsonGraduateSchoolofManagement,CornellUniversity
© 2015 eCornell. All rights reserved. All other copyrights, trademarks, trade names, and logos are the sole property of their respective owners.
29
considerationisthatdesirablenewproductalternativescanbesynthesizedfromthebasicattributes.Thesearethesituationsinwhichconjointismostapplicable.