supply chain management

Malhotra,Gosain,andElSawy:LeveragingSEBIstoEnableAdaptiveSupplyChainPartnerships270

4.AnalysisandResults

4.1.MeasurementValidation

WeusedPLSGraph3.0fordataanalysis.PLShas

anadvantageoverotherstructuralmodeling(SEM)

methodologiesinthatitdoesnotrequiredistributions

benormalorknown(JoreskogandWold1982).Other

structuralestimationtechniqueslikeLISRELassume

multivariatenormaldistributionorWishartdistribu-

tion,butPLStakesanydistributionthatismani-

festthoughmeasurementandcalculatesthebestset

ofpredictiveweightsthroughaseriesofiterations.

AnotheradvantageofusingPLSisthatithasless

stringentsamplesizerequirements.Techniquessuch

asLISRELusechi-squareestimatesfor“goodness-of-

t”indicators.Unfortunately,chi-squareestimatesare

extremelysensitivetosamplesize.The tindicesin

PLSaredescriptivestatisticsandindicateonlythe

amountofvarianceaccountedforinthemodelbythe

speci edrelationships.

Ournextchoicewaswhethertheconstructswould

bemodeledasre ectiveorformative.Inmakingthis

choice,wefollowedtheguidelineslaidoutbyJarvis

etal.(2003).Constructsshouldbemodeledasfor-

mativeunderthefollowingconditions:(1)indicators

areviewedasthede ningcharacteristicofthecon-

struct,(2)changesintheindicatorscauseachange

intheconstruct(andnotviceversa),(3)indicators

donotneedtonecessarilycovary,(4)indicatorsare

notnecessarilyinterchangeable,and(5)indicatorscan

bedrawnfromdifferentnomologicalnetwork(Jarvis

etal.2003,Patnayakunietal.2006).Basedonthese

criteria,alltheconstructsinthisstudyweremod-

eledasformativeconstructs.Speci cally,weused

indexscoresofassociateditemstoestablishamea-

sureforeachformativeconstruct.Wehadtwochoices

tocomputetheindexscore—factorscoresormean

valueofitems.Although,formativeconstructsare

notrequiredtoexhibitinternalconsistency(Jarvis

etal.2003,Raietal.2006,Petteretal.2007),the

itemswerestronglycorrelated.Therefore,wechose

themeanvaluetocomputetheindex,whichwould

naturallycorrelatehighlywithfactorscoresorother

alternateweightingschemesfortheitems(Rozeboom

1979).Moreover,Hairetal.(1987)recommendthe

useofunitmeanscoresforreplicabilityandeaseof

rmationSystemsResearch18(3),pp.260–279,©2007INFORMSSimilarly,breadthofinformationexchange(CIE1),qualityofinformationexchange(CIE2),andprivi-legedinformationexchange(CIE3)weremodeledasformativeconstructs.CIEwasmodeledasaformativeconstructcomprisedofthreeindicators:CIE1,CIE2,andCIE3.Theindexscoresforthesethreeindicatorswerealsoderivedbasedontheunitmeansofassoci-ateditems(seeAppendixAforitems).Webeganourdataanalysisbyassessingthemea-surementpropertiesofconstructs.Weconductedapseudocon rmatoryfactoranalysis(asPLSdoesnotprovidecrossloadingofitemsonconstructsotherthanthosetheyarehypothesizedtoload)followingtheprocedureoutlinedbyKarahannaetal.(1999)andPatnayakunietal.(2006).Ameanfactorscoreforeachconstructwascomputedfromtheitemsthatwerehypothesizedtore ecttheconstruct.Thenalltheitemswerecorrelatedwitheachoftheconstructs.Anindicator’scorrelationwithitshypothesizedcon-structcanbeconstruedas“loading,”whileitscorre-lationwithotherconstructsis“cross-loading.”Eachoftheitemsexhibitsahighercorrelationwithitsownconstructthanotherconstructsprovidingevidencefordiscriminantvalidity(Table2).Tofurthertestfordis-criminantvalidityofourconstructs,weexaminedtheaveragevarianceextracted(AVE)foreachconstructandcompareditwithcorrelationsbetweenconstructs(FornellandLarcker1981).AscanbeseenfromTable2ItemConstructstoOwnConstructCorrelationvs.CorrelationswithOtherConstructItemAKCMASTDCIECNAdaptivecreationknowledgeAKC1AKC200.220.25AKC30.820.200.210.130.09AKC40.0.68.84780.290.350.190.230.320.030.070.160.050.18MutualadaptationMA1MA20.260.330.180.04MA30.4500.130.0.80.91810.360.370.190.010.100.33UsebusinessofstandardinterfaceselectronicSTD1STD20.160.40STD30.200.200.4100.210.210.280.0.85.93840.340.200.410.37CollaborativeexchangeinformationCIE1 CIE2 0.270.13CIE3 0.060.090.080.250.130.140.0600.360.0.91.69830.020.11CooperativenormCN1CN20.140.16CN30.180.180.100.240.060.210.330.420.1200.200.0.87.8689 Indexcomputedasmeanscoresofassociateditems.