當代市場調(diào)研 （原第12版）指導手冊 Instructor Manual ch14 更強大的統(tǒng)計方法

McDaniel&Gates–MarketingResearch,12thEdition Instructor’sManual

Copyright?2021JohnWiley&Sons,Inc. 14-

CHAPTER14

MorePowerfulStatisticalMethods

LEARNINGOBJECTIVES

1.Learnthebivariateanalysisofassociation.

2.Understandbivariateregressionanalysis.

3.Definemultivariatedataanalysis.

4.Gaininsightsintomultivariatesoftware.

5.Describemultiplediscriminantanalysis.

6.Understandclusteranalysis.

7.Understandfactoranalysis.

KEYTERMS

Bivariateregressionanalysis

Causation

Classificationmatrix

Clusteranalysis

Coefficientofdetermination

Collinearity

Conjointanalysis

Correlationanalysis

Dependentvariable

Discriminantcoefficient

Discriminantscore

Dummyvariables

Errorsumofsquares

Factor

Factoranalysis

Factorloading

Independentvariable

K-meansclusteranalysis

Metricscale

Multiplediscriminantanalysis

Multipleregressionanalysis

Multivariateanalysis

Neuralnetworks

Nominalorcategorical

Pearson’sproduct-momentcorrelation

Regressioncoefficients

Scalingofcoefficients

Scatterdiagram

Sumofsquaresduetoregression

Utilities

CHAPTERSUMMARY

Thischapterexaminestheelementsofbivariateanalysisofassociation.Bivariatetechniquesarestatisticalmethodsofanalyzingtherelationshipbetweentwovariables,independentvariablesanddependentvariables.Thebivariateregressionispresented.Ifthedataappearstobelinearwhenplottedonascatterplot,regressioncanbeused.Anexampleisthenusedtoexplorethebasicstepsinbivariateregression.

CorrelationAnalysisisalsodiscussed.Correlationisthemeasurementofthedegreetowhichchangesinonevariableareassociatedwithchangesinanother.ThePearson’sproductmomentcorrelationisusedifmetricdataareinvolved.

Thischapteralsoexaminesseveralmethodsofmultivariatedataanalysis.Thesetechniquesarecomplex,requiringcomputerstodothemathematics.Multipleregressionanalysisistheappropriatemultivariatetechniqueiftheresearcher’sgoalistoexaminetotherelationshipbetweentwoormoremetricpredictorvariablesandonmetricdependentvariables.Discriminantanalysisissimilartomultipleregressionanalysis,exceptthedependentvariableisnominalorcategoricalinnature.ClusterAnalysisisusedtoidentifyobjectsorpeoplethataresimilarinregardtocertainvariablesormeasurements.Factoranalysisisusedtosimplifydataandtosummarizemeasuresintoconcepts.Conjointanalysishelpstodeterminewhatfeaturesanewproductorserviceshouldhaveandhowitshouldbepriced.

QUESTIONSFORREVIEWANDCRITICALTHINKING

Giveexamplesofthreemarketingproblemsforwhichregressionanalysiswouldbeappropriate.

Studentanswersmayvary.Giventhetypeofanalysis,anydescriptionwouldnotethatthevariableshadtobeatleastintervalscale(metric)forPearson’sCorrelation,andmostlikelythecaseinbivariateregression,exceptintheeventoftheuseofdummyvariables.Also,thestudentwouldneedtobeclearaboutwhichvariablewasbeingpredicted(dependentvariable)andwhatvariableorvariables(independentvariables)werebeingusedtomaketheprediction.

Whatpurposedoesascatterdiagramserve?

Ascatterdiagramisusedtoplotdataobservationstodetermineiftherelationshipappearstobelinear,curvilinear,ornon-existent.Thisallowstheresearchertodeterminewhetherusinglinearmeasuresofassociation(e.g.Pearson’sproduct-momentcorrelation)wouldbeappropriate.

Explainthemeaningofthecoefficientofdetermination.Whatdoesthiscoefficienttelltheresearcheraboutthenatureoftherelationshipbetweenthedependentandindependentvariables?

Thecoefficientofdeterminationtellsthemarketingresearcherhowmuchvariationinthedependentvariablecanbeexplainedbyvariationintheindependentvariable.Itisameasureofthestrengthoftherelationship.Ifthecoefficientofdeterminationislow,theindependentvariabledoesnothavesignificantexplanatorypowerinpredictingchangesinthedependentvariable.

Distinguishbetweenmultiplediscriminantanalysisandclusteranalysis.Giveexamplesofsituationsinwhicheachmightbeused.

Multiplediscriminantanalysisanalyzestherelationshipsbetweenasetofmetricindependentvariablesandanominalorcategoricaldependentvariable.Itcantestahypothesizedrelationshipanditdescribeshowtheindependentvariablesdiscriminatebetweenthegroupsofthedependentvariable.Clusteranalysisisastatisticaltoolusedforclusteringpeopleorobjectsbasedonaparticularcriteriaorvariableinthestudy.Forexample,wemighthave15differentmeasuresofbenefitsandwanttoclusterpeopleintobenefitgroupsformarketsegmentation.Thesameconceptcouldbeusedforpersonalvalues,attitudestowardratinghealthyalternatives,orthetypesofrestaurantsfrequented.Multiplediscriminantanalysiscouldbeusedforsegmentingusersfromnonusers,lightfromheavyusers,patronsfromnon-patrons,andahostofotherdependentcategoricalvariables.Anumberofindependentvariablesetssuchasbenefits,attributes,knowledge,andpreferencescouldbeusedtopredictgroupmembership.

Whatpurposedoesmultipleregressionanalysisserve?Giveanexampleofhowitmightbeusedinmarketingresearch.Howisthestrengthofmultipleregressionmeasuresofassociationdetermined?

Multipleregressionanalysisisusedtoexaminetherelationshipbetweentwoormoremetricpredictorvariablesandonemetricdependentvariable.Itcanalsobeusedtogeneratepredictionsforthedependentvariable,givenacombinationofvaluesfortheindependentvariables.Multipleregressionanalysishasmanyapplicationsinmarketingresearch.Onegeneralapplicationrelatestodeterminingtheeffectsofvariousmarketingvariablesonsalesormarketshare.TheCoefficientofDeterminationorR2providesameasureofthepercentageofvariationinthedependentvariableexplainedbyvariationintheindependentvariable(s).

Whatisadummyvariable?Giveanexampleusingadummyvariable.

Theterm“dummyvariable”describesthecodingofnominallyscaledindependentvariablessothattheycanbeusedinregressionanalysis.Anexampleofadummyvariableforameasureoflocationofbirthwouldbe

0=bornintheUnitedStates,1=bornoutsidetheUnitedStates

Itiscriticalthatthestudentunderstandsthatdummyvariablesmustbecodedas0/1andwhenmorethantwocategoriesareused,thatmorethanonedummyvariablemustbeused.Forexample,tocodefreshman,sophomore,junior,orseniorwouldrequirethree(n-1)dummyvariables.

Asalesmanagerexaminedagedata,educationlevel,apersonalityfactorthatindicatedlevelofintrovertedness/extrovertedness,andlevelofsalesattainedbythecompany’s120-personsalesforce.Thetechniqueusedwasmultipleregressionanalysis.Afteranalyzingthedata,thesalesmanagersaid,“Itisapparenttomethatthehigherthelevelofeducationandthegreaterthedegreeofextrovertednessasalespersonhas,thehigherwillbeanindividual’slevelofsales.Inotherwords,agoodeducationandbeingextrovertedcauseapersontosellmore.”Wouldyouagreeordisagreewiththesalesmanager’sconclusions?Why?

Themanagershouldconsiderwhetherageandeducationarecorrelated(collinearity),asoldersalespersonsmayhavegreatereducationandthusthe“educationeffect”mayreallybean“age/experienceeffect.”Itisalsoplausiblethattheextrovertedsalespersonsarealsoolder,asgreaterexperiencegenerallyleadstogreaterconfidenceandperformance.Anotherimportantissueiswhetherthemanagerhascorrectlyspecifiedtheregressionmodel.Itislikelythatmanyotherfactorsthathaveasignificantimpactonsalespersonperformancehavenotbeenincludedinthemodel.Anyofthesecouldbeconfoundedwiththeincludedvariables.Perhapsmostimportantly,themanagershouldkeepinmindthatcausationcanneverbeprovenwithstatisticalevidencealone.

Thefactorsproducedandtheresultsofthefactorloadingsfromfactoranalysisaremathematicalconstructs.Itisthetaskoftheresearchertomakesenseoutofthesefactors.ThefollowingtablelistsfourfactorsproducedfromastudyofcableTVviewers.Whatlabelwouldyouputoneachofthesefourfactors?Why?

Factor1Varietyofprogrammingorrepetitiveprogramming.Allofthequestionsdealwiththemoviechannelsplayingthesamemoviesoverandoveragain.

Factor2Emotionalprogrammingor“Tear-jerking”programming.Alloftheitemsareaboutprogrammingthatelicitsanemotionalresponse.

Factor3Religiousprogramming.Thesequestionsmeasuretheviewers’opinionsofreligiousprograms.

Factor4Homeentertainment.Theitemsmeasuretheviewers’preferencesforviewingmoviesathome.

ThefollowingtableisadiscriminantanalysisthatexaminesresponsestovariousattitudinalquestionsfromcableTVusers,formercableTVusers,andpeoplewhohaveneverusedcableTV.Lookingatthevariousdiscriminantweights,whatcanyousayabouteachofthethreegroups?

For“users,”themostdiscriminatingvariablesareA19(easygoingonrepairs)andA18(norepairservice).For“formers,”themostdiscriminatingvariablesareA4andA18(burnedoutonrepeatsandnorepairservice,respectively).Forthe“nevers,”themostdiscriminatingvariablesareA7andA19(breakdowncomplainerandeasygoingonrepairs,respectively).Theseresultssuggestthatconcernsabout,orreactionsto,servicefailure(breakdowns/repairs)arethemostpredictiveofwhetheraconsumerisauser,aformeruser,orneverauser.

Thefollowingtableshowsregressioncoefficientsfortwodependentvariables.ThefirstdependentvariableiswillingnesstospendmoneyforcableTV.Theindependentvariablesareresponsestoattitudinalstatements.TheseconddependentvariableisstateddesirenevertoallowcableTVintheirhomes.Byexaminingtheregressioncoefficients,whatcanyousayaboutpersonswillingtospendmoneyforcableTVandthosewhowillnotallowcableTVintheirhomes?

Thosewhoarewillingtospendmoneyforcabletelevisionenjoywatchingmoviesandcomedy,andtheyarelikelytodosolateatnight.Theymaybesomewhatlonely(“forlorn”)andmayhaveagreaterneedforexternalsourcesof“stimulation,”suchasmightbeofferedbytelevisionprogramming.Theyarealsodissatisfiedwiththeservicelevelandwishthecablestationsofferedmorevariety.Thosewhowillnotallowcableintheirhomesdonotenjoywatchingsportingevents,objecttosexontelevision,anddonotfeelaneedformanychoicesintheirtelevisionprogramming.

Explainwhatpredictiveanalyticsencompasses.Provideexamplesofsomemarketingproblemstowhichyoumightapplypredictiveanalytics.

Predictiveanalyticsdescribesawidearrayoftoolsandtechniquesthatareusedtoextractandanalyzeinformationfromdatasets.Exampleswillvarybutmightincludesuchthingsas…

Predictingthelikelihoodofpurchasingbabyfurniture

Predictingwhomightvoteforaparticularcandidateorvoterinitiative

Predictingthepotentialmarketshareforanewproduct

Describethestepsinthepredictiveanalyticsprocess.

AcquiringaDataSet.Beforeapplyingpredictiveanalytics,anorganizationmustassembleatargetdatasetrelevanttotheproblemofinterest.

Pre-processing.Onceassembled,thedatasetmustbecleanedinaprocesswhereobservationsthatcontainexcessivenoise,errorsandmissingdataareeditedorexcluded.

Modeling.Thisistheprocessofbuildingarelationshipbetweenthedataandwhatistobepredicted.Techniquesincludeclustering,classification,andestimatingandpredicting.

ValidatingResults.Afinalstepofknowledgediscoveryfromthetargetdataandmodelingistoattempttoverifythepatternsproducedbythepredictivemodelingalgorithmsinawiderdataset.

ApplyingtheResults.Oncethemodelsandcalculationsareinplaceandhavebeenvalidated,theyareappliedtoexistingandfuturecustomerrecordstoimprovetheefficiencyandeffectivenessofmarketingefforts.

WORKINGTHENET

Takealookat.Reviewthediscussionofwhyanalyticsareimportantandusefulinthesportsworld.Lookatsomeoftheapplicationstheydiscuss.DownloadtheAgileSportsAnalyticsGuide.Reviewitandgetafeelforwhatcanbedone.

Responseswillvary.

Goto/resources/multivariate-types-methods.htmlforsomeeasytodigestandcomprehensiveinformationonmultivariateanalysiswithapplications.

Responseswillvary.

REAL-LIFERESEARCH

Case14.1–SatisfactionResearchforPizzaPronto

KeyPoints

Ar