FACTORANALYSIS
MTahir
INTRODUCTION
Thefactoranalysistechniquereducesthelarge
numberofvariablesintofewunderlyingfactors
toexplainthevariabilityofthegroup
characteristics.
Theconceptusedinfactoranalysistechniqueis
toinvestigatetherelationshipamongthegroup
ofvariablesandsegregatethemindifferent
factorsonthebasisoftheirrelationship.Thus,
eachfactorconsistsofthosevariableswhichare
relatedamongthemselvesandexplainsome
portionofthegroupvariability
EXPLORATORYVERSUS
CONFIRMATORYFACTORANALYSIS
Inexploratorystudy,wedonotknowanything
aboutthenumberoffactorstobeextracted,the
numberofvariablesincludedineachfactorand
percentageofvariabilityexplainedbythese
extractedfactors.
confirmativefactoranalysisisrequiredtotest
theexistingfactormodel
KEYTERMS
Principalcomponentanalysis(PCA):itisusedto
reducethelargenumberofvariablesintosmaller
numberofprincipalcomponentsthatwillaccount
formostofthevarianceintheobservedvariables
Factorloading:itcanbedefinedasthe
correlationcoefficientbetweenthevariableand
factor
Communality:Itcanbedefinedasthesumofthe
squaredfactorloadingsofavariableinallthe
factors
KEYTERMSCONTD
Eigenvalue:Itmeasuresthevarianceinallthe
variableswhichisaccountedforbythatfactor.
(Criteria>1)
ScreePlot:Itisagraphicalrepresentationofthe
factorsplottedalongXaxisagainsttheir
eigenvalues,ontheYaxis.
STEPSINRUNNINGEXPLORATORY
FACTORANALYSIS
1. Computedescriptivestatisticsforallthe
variables
2. Preparecorrelationmatrix
3. ApplyKMOtesttochecktheadequacyofdata
forrunningfactoranalysis(>0.5)
STEPSCONTD
4.ApplyBartlettstestofsphericity(Criteria=
shouldbe<0.01)
5.Obtaininitial(unrotated)andthenfinalsolution
byusingthevarimaxrotation(Criteria=Factor
loading>0.70)
6.NametheVariablesbasedonresearchers
understanding.
ASSUMPTIONSINFACTOR
ANALYSIS
1.Alltheconstructswhichmeasurethe
conceptshavebeenincludedinthestudy.
2.Sufficientsamplesizehasbeentakenfor
factoranalysis.Normallysamplesizemust
beequalto520timesthenumberof
variablestakeninthestudy.
3.Nooutlierispresentinthedata.
4.Multicollinearityamongthevariables
doesnotexist.
ASSUMPTONSCONTD
5.Homoscedasticitydoesnotexistbetweenthe
variablesbecausefactoranalysisisalinear
functionofmeasuredvariables.
6.Variablesshouldbelinearinnature.Nonlinear
variablesmayalsobeusedaftertransformingit
intolinearvariables.
7.Datausedinthefactoranalysisisbasedon
intervalscaleorratioscale.
COMMANDINSPSS
AnalyzeDimensionReductionFactor
ChecktheoptionUnivariatedescriptiveand
ensurethattheoptionInitialSolutionis
checkedintheStatisticssectionbydefault.
ChecktheoptionCoefficients,Significance
levels,andKMOandBartlettstestof
sphericityinCorrelationMatrixsection.
InExtraction,checkScreeplot.
InRotationcheckVarimaxrotationoption.
OUTPUT
Descriptivestatistics
Correlationmatrix
KMOandBartlettstest
Communalitiesofallthevariables
Totalvarianceexplained
Screeplot
Componentmatrix:unrotatedfactorsolution
Rotatedcomponentmatrix:varimaxrotated
solution
SUMMARYOFCOMMANDS
(i)StartSPSSandpreparedatafilebydefiningthevariablesandtheirproperties
inVariableViewandtypingthedatacolumnwiseinDataView.
(ii)Inthedataview,followthebelowmentionedcommandsequenceforfactor
analysis:
Analyze!DimensionReduction!Factor
(iii)SelectallthevariablesfromleftpaneltotheVariablessectionoftheright
panel.
(iv)ClickthetagDescriptivesandchecktheoptionsUnivariatedescriptives,
InitialSolution,Coefficients,Significancelevels,andKMOand
Bartlettstestofsphericity.PressContinue.
(v)ClickthetagExtractionandthencheckScreeplot.Letotheroptions
remainasitisbydefault.PressContinue.
ClickthetagRotationandthencheckVarimaxrotationoption.Letother
optionremainsasitisbydefault.PressContinue.
(vii)ClickOKforoutput
CONFIRMATORYFACTOR
ANALYSISUSES
Evaluationoftestinstruments
Constructvalidation
Convergentvalidity
Discriminantvalidity
Developmentandtestingofthemeasurement
modelforaSEM.
ADVANTAGESOFCFA
Thefitofthemeasurementmodelcanbe
determinedbeforeestimatingtheSEMmodel.
InSEMmodelsyoucanestablishrelationships
amongvariablesadjustingformeasurement
error.
CFAcanbeusedtoanalyzemeanstructures.
STEPSINCONFIRMATORYFACTOR
ANALYSIS
Step1=Definethefactormodel
Step2=Collectmeasurements
Step3=Obtainthecorrelationmatrix
Step4=Fitthemodeltothedata
Step5=Evaluatemodeladequacy
(InsignificantChisquare,Chisquare
value/df=<5,RMSEA<0.05,GFI>0.90,NFI>0.90,
)
Step6=Comparewithothermodels
