monMethodBiasAssessmentWetriedtominimizetheconcernofcommonmethodbiasbyrequiringtheRosettaNetchampionsateachenterprisetodrilldownwithintheirenterpriseand ndtheexecutive(“keyinformant”)responsiblefortheday-to-dayfunctioningofthepartnershipunderinvestigation(adifferentkeyinformantforeachrela-tionshipifmultiplesurveyswere lledbyacom-pany).Thekeyinformantthenassignedvarioussectionsofthesurveytobecompletedbymanagerswhotheyfeltweremostlikelytoprovideaccurateresponsesforalineofquestioning.So,ineffect,differentrespondentswereassignedto lloutdifferentportionsofourquestionnaire.How-ever,therewasstillsomeconcernaboutthepossi-bilitythatasinglerespondentmayhavecompletedthewholesurvey.Toallaysuchconcern,wecon-ductedtheHarmon’sone-factortest.Anexploratoryfactoranalysisrevealeda ve-factorstructure(Eigenvalue>1)whereallitemsdidnotloadonthe rstfactorandthe rstfactordidnotexplainmostofthevarianceinthedata(～40%oftheoverall76%varianceexplained).Therefore,theconcernaboutcommonmethodbiasinthedatacollecteddoesnotseemtobeanissue.Further,notallthehypothe-sizedpathsweresigni cantinthemodel,andthesigni cantpathsvaryintheirlevelofsigni cance.271Therefore,thecommonmethodbiasseemstobeevenlessofaconcern(Patnayakunietal.2006).4.3.MediationAnalysisOurresearchmodel(Figure1)proposesmediatedimpacts.Therefore,wefollowedtheprocessoutlinedbyPatnayakunietal.(2006)andSubramani(2004)totestformediationeffects.Westartedoutbycompar-ingourresearchmodel,whichproposesadirecteffectandmediatedeffectoftheuseofstandards,againstacompetingmodelthatproposesfullmediation(i.e.,theeffectoftheuseofSEBIsisfullymediatedthroughCIE).Theaimofsuchanalysisistostatisticallytestwhetherthedirecteffectoftheindependentvariables(IVs)explainsadditionalvarianceinthedependentvariable(DV)aboveandbeyondthemediatedeffectsthroughthemediatingvariable(MV).Theproposedfullmodel(directandmediatedeffects—partialmedi-ation)canbecomparedagainstthenestedmodel(fullmediation—mediatedeffectsonly)statisticallyusingPLSresults(Patnayakunietal.2006,Chinetal.2003,Subramani2004).AscanbeseeninTable4,theR2foradaptiveknowledgecreation(DV1 andMA(DV2 inthepartiallymediatedmodels(modelsthatincludedirecteffectsofuseofstandardinterfaces)were0.245and0.287,respectively,comparedwithR2of0.067and0.199inthealternatenested(fullymediated)models.Thisdifferentialeffectinpartialandcompletemedia-tioncanbefurtherinvestigatedbyaproceduresimilartostepwiseregression(Chinetal.2003,Patnayakunietal.2006).Apseudo-FstatisticcanbecalculatedbyusingthedifferenceinR2betweenthefullmodelandthenestedmodel.3Thef2statistic(calculatedbasedonthedifferenceinR2betweenfullandnestedmodel)forthetwodependentvariables(DVofSEBIs(directpaths)1andDV2 ,withrespecttouseis0.06and0.304.Therefore,thepseudo-Fstatisticis2.07(notsigni cantat0.05level)and10.34(signi cantat0.05level).Thisanalysissuggeststhattheaddi-tionaldirectpathfromtheuseofSEBIs(STD)toMAexplainsadditionalvarianceandaddssigni cantlytotheexplanatorypowerofthemodel.However,thereisalackofevidencesupportingadirectrelationship3F=f2 n k 1 ,with1, n k degreesoffreedom.n=samplesize,k=numberofconstructsinthemodel,andf2= R2partial
mediation R2fullmediation / 1 R2partialmediation).
supply chain management
272Malhotra,Gosain,andElSawy:LeveragingSEBIstoEnableAdaptiveSupplyChainPartnershipsInformationSystemsResearch18(3),pp.260–279,©2007INFORMS
Table4
DirectpathTestforMediation—Fullvs.NestedModelAnalysisGraphicmodel
R2fullmodel0.245R2nestedmodel0 199f2value0 06Pseudo-F 1 35 2 07
STD→AKC(0.235 ,t=1 188)
STD→MA(0.518 ,t=3 642)0.2870 070 30410 34
Signi cantatp<0 05level.
betweentheuseofSEBIs(STD)andadaptiveknowl-
edgecreation(AKC).Inotherwords,thereisadirect
effectofuseofSEBIsonMA,andthereisalsoan
indirecteffectmediatedthroughcollaborativeinfor-
mationexchange.Ontheotherhand,theeffectofthe
useofSEBIsonAKCisfullymediatedthroughcol-
laborativeinformationexchange(CIE).
Further,wealsoassessedthemediationeffectof
CIEonthetwoDVs(AKCandMA).Todothis,
weusedthepathcoef cientsandstandarderrorsbe-
tweentheuseofSEBIsandtheMV(CIE)andbetween
CIEandtheDVsobtainedfromthePLS(Patnayakuni
etal.2006,HoyleandKenney1999,Subramani2004).
Themagnitudeofmediationistheproductofpath
coef cientsbetweentheIVandMVandbetweenMV
andDVs.Themagnitudeofthetwomediationeffects
is0.103(STD CIE AKC)and0.087(STD CIE
MA).Thez-statistic4forthetwopathsis2.06and1.98
(signi cantatthep<0 05level).
Thez-scoreiscalculatedusinganapproximationforthestandard
errorofthemediatedpath.Theapproximatedstandarderroris
222222computedusingtheformulasqrt p1s2+p2s1+s1s2 ,wherep1isthe
pathcoef cientofthepathfromIVtoMV,p2isthepathcoef -
cientfromMVtotheDV,ands1,s2arethecorrespondingstandard
deviations.44.4.StructuralModelThe nalresultsofthePLSanalysisareshowninFig-ure2.WeevaluatedthestructuralmodelbasedontheR2valuesandstructuralpathsforeachoftheendogenousconstructs.AscanbeseenfromFigure2,themodelexplainsasubstantialamountofvarianceintheDVs:MA(R2=0 28)andAKC(R2=0 24)betweensupplychainpartners.Thesigni canceofthepathswasobtainedthroughthebootstrappingproce-dureinPLS.Theresults(Figure2)providesupportformostofthehypothesizedpathsintheresearchmodelshowninFigure1.ThereissupportforHypothesis1,asthepathsfromCIEtoMA( =0 33,p<0 05)andAKC( =0 32,p<0 05)arestatisticallysigni cant.TheuseofSEBIshasapositiveeffectonCIE( =0 32,p<0 05),providingsupportofHypothesis2.Hypothe-sis3ispartiallysupported,astheuseofSEBIshasasigni cantdirecteffectonMA( =0 53,p<0 05)butdoesnothaveastatisticallysigni cantdirecteffectonAKC( =0 24).Asdiscussedearlier,theMAshowsthattheeffectoftheuseofSEBIsonmutualadapta-tionbetweensupplychainpartnersispartiallymedi-atedandtheeffectonAKCisfullymediatedthroughCIE(i.e.,thereisnodirecteffect).ThepathsfromthecontrolvariablestotheDVswerenotsigni cant.