Epidemiologyprovidesthescientificbasisformuchofpublichealthpractice,andthecurrentrevolutioninhealthcareanddiseasepreventionindicatesthatthedemandforvaluableresultsfromthisfieldwillcontinue
togrow.Soundepidemiologicresearchrequiresasolidstatisticalbasisfor
bothstudydesignanddataanalysis.Hence,itshouldcomeasnosurprise
thatepidemiologyhasprovidedmodernstatisticswithsomeofitsdeepest
problems.Asknowledgeabouttheunderlyingcausesofdiseasesincreases,
weoftenseethattheyhavemultiplecauses,sothatitisgenerallynotpossibletolimitconclusionstoasinglefactor.Wealsoneedtoconsiderthe
possibilitythatotherfactorsmayconfoundaparticularassociationormodifyitsestimatedeffect.Therefore,amultivariateapproachtodataanalysis
isusuallyanessentialpartofepidemiologicresearch.
Themultivariatemethodsconsideredinthisbookinvolvethesimultaneousanalysisoftheassociationbetweenmultipleattributesofanindividualandtheriskofadisease.Theunderlyingframeworkisoneinwhich
asingleresponseisassociatedwithmultipleregressorvariables.Somereservethetermmultivariatemethodsfortechniquesthatdealwithmorethan
oneresponseinasingleanalysis,whichisnotthemainfocusofthisbook.
ThedefinitionofvariateintheOxfordEnglishDictionarydoesnotdistinguishattributesofanindividualthatareidentifiedwiththeresponsefrom
thosethatpredictthatresponse.Itismultiplesofthelatterthatarethe
primaryfocusinthefollowingchapters,althoughsomeofthelog-linear
modelswouldfallintoeventhenarrowdefinitionofmultivariatemethods.