The Dimensionality of the 12-Item General Health ... - Frontiers

The Dimensionality of the 12-Item General Health Questionnaire (GHQ-12): Comparisons of Factor Structures and Invariance Across Samples and Time. DownloadArticle DownloadPDF ReadCube EPUB XML(NLM) totalviews ViewArticleImpact SHAREON ClaudioBarbaranelli SapienzaUniversityofRome,Italy KOSTASA.PAPAGEORGIOU Queen'sUniversityBelfast,UnitedKingdom DONATELLARITAPETRETTO UniversityofCagliari,Italy Theeditorandreviewer'saffiliationsarethelatestprovidedontheirLoopresearchprofilesandmaynotreflecttheirsituationatthetimeofreview. Abstract Introduction MaterialsandMethods Results Discussion Conclusion DataAvailabilityStatement EthicsStatement AuthorContributions Funding ConflictofInterest References Exportcitation EndNote ReferenceManager SimpleTEXTfile BibTex Checkforupdates Peoplealsolookedat ORIGINALRESEARCHarticle Front.Psychol.,11June2020Sec.QuantitativePsychologyandMeasurement https://doi.org/10.3389/fpsyg.2020.01300 TheDimensionalityofthe12-ItemGeneralHealthQuestionnaire(GHQ-12):ComparisonsofFactorStructuresandInvarianceAcrossSamplesandTime SigurdW.Hystad*andBjørnHelgeJohnsen DepartmentofPsychosocialScience,UniversityofBergen,Bergen,Norway Becauseofitsbrevity,the12-itemGeneralHealthQuestionnaire(GHQ-12)hasbecomeoneofthemostpopularandusedmeasurefordetectingpsychologicaldistress.Originallyintendedasaunidimensionalmeasure,themajorityofsubsequentfactor-analyticstudieshavefailedtosupportGHQ-12asaunitaryconstructandhaveinsteadproposedaplethoraofmultidimensionalstructures.Inthisstudy,wefurtherexaminedthefactorstructureintwodifferentmilitarysamples,oneconsistingofcrewmembersfromfourdifferentfrigatesdeployedinanti-piracyoperationsandStandingNATOMaritimeGroupdeployments(N=591)andoneconsistingofcrewmemberfromthreedifferentminehunters/sweepersservinginStandingNATOMineCounter-MeasuresGroupdeployments(N=196).Resultsfromconfirmatoryfactoranalyses(CFA)performedinthefirstsamplesupportedabifactormodel,consistingofageneralfactorrepresentingcommunalityamongallitemsandtwospecificfactorsreflectingcommonvarianceduetowordingeffects(negativelyandpositivelyphraseditems).Amulti-groupCFAfurtherconfirmedthisstructuretobeinvariantacrossoursecondsample.Structuralequationmodelingalsoshowedthatthegeneralfactorwasstronglyassociatedwithsymptomsofinsomniaandmentalhealth,whereasthespecificfactorswereeithernon-significantlyorconsiderablyweakerassociatedwiththecriterionvariables.Overall,ourresultsarecongruentwiththenotionthatthemultidimensionalitydemonstratedinmanypreviousinvestigationsismostlikelyanexpressionofmethod-specificvariancecausedbyitemwording.Theexplaineduniquevarianceassociatedwiththesespecificfactorswasfurtherrelativelysmall.IgnoringthemultidimensionalityandtreatingGHQ-12asaunitaryconstructwillthereforemostlikelyintroduceminimalbiastomostpracticalapplications. Introduction Increasedfocusonmentalhealthproblemsanditsimpactonthepopulationhaveresultedindevelopmentofscreeningprogramsfordifferentsub-groupsatriskfordevelopingseverepsychopathologies.ThediversityinprogramsrangefromscreeningformentalhealthinwomenwithriskoftransferringHIVtotheirchildren(Iheanachoetal.,2015)tomentalhealthevaluationofelderliesinKoreancommunities(Juetal.,2017).Mentalhealthscreeninginthemilitarydomainisparalleltocommunityscreeningprogramsandhaslonghistoricalroots(Wrightetal.,2002).Atpresent,mostnationsimplementprogramsbeforeandafterdeploymentstooperationalareas(Ronaetal.,2005),andsomenationshaveanadditionalmid-operationevaluation(Sandenetal.,2014).Themainaimsofthesescreeningproceduresareearlycaseidentificationinordertoimplementadequateinterventionsaswellasdefiningspecificstressorsforthepersonnelinvolvedandforthetypesofmissionsinwhichtheyareinvolved.Inordertobeabletoprocesslargeamountofdatainbriefperiodsoftime,shortself-reportinventoriesarepreferable.Shorterinventoriesdonotonlyreducetheassessmenttimeandrelatedcostsbutcanalsoimproveparticipationratesandreduceparticipationfatigue,andassuchleadtobetterdataquality.However,manyresearchershavealsopointedouttheinevitabletrade-offbetweenthesepragmaticreasonsandthepsychometricqualitythatislostwhenusingshorterorabbreviatedscales(Smithetal.,2000;Credéetal.,2012).Forinstance,Credéetal.(2012)reportedlowerinternalconsistencyestimatesandlowerpredictivepoweracrossarangeofoutcomesforshortversuslongscales. Oneshortquestionnaireoftenusedinbothcommunityandmilitaryscreeningisthe12-itemversionoftheGeneralHealthQuestionnaire(GHQ-12;GoldbergandWilliams,1988).TheGHQ-12derivesfromtheoriginal60-itemversion,andadditionallyexistsin30-,28-,and20-itemsversions(GoldbergandWilliams,1988).TheadvantageofGHQ-12isthatitisshort,caneasilybescored“clinically”(symptomspresentorabsent)aswellaslevelsofsymptomspresent(Likert-typescoring).Thescalewasoriginallydesignedasascreenforriskforcommonmentaldisorders(BöhnkeandCroudace,2016),buthasalsobeenusedasameasureofgeneralsymptomload(Johnsenetal.,1998),Positivementalhealth(Huetal.,2007)andminorpsychologicalproblems(Nordmoetal.,2020).Theinstrumentisfrequentlyusedinscreeningofcivilianpopulationsindifferentcultures(Iheanachoetal.,2015;Endsleyetal.,2017;Juetal.,2017;Tseliouetal.,2018).Thefrequentuseofthescaleindifferentculturesandthedifferentinterpretationsofwhatthescalemeasures,motivatesforapsychometricanalysisinordertoclarifythevalidityoftheinstrument. DimensionalityofGHQ TheGHQ-12wasintendedasaunidimensionalmeasureofpsychologicaldistress(GoldbergandWilliams,1988).Becauseofitsbrevity,theGHQ-12hasbecomeoneofthemostusedinstrumentsfordetectingpsychologicaldistressinnon-clinicalsamples(Hankins,2008;Tomásetal.,2017).Theinstrumenthasbeentranslatedintomanydifferentlanguages,includingSpanish(Cuéllar-Floresetal.,2014),Portuguese(Tomásetal.,2017),German(Romppeletal.,2013),French(Salama-Younesetal.,2009),Italian(Politietal.,1994),Dutch(Corneliusetal.,2013),Norwegian(Nordmoetal.,2020),Farsi(Namjooetal.,2017),Japanese(Suzukietal.,2011),Thai(Gelayeetal.,2015),andChinese(Ye,2009).AlthoughmostresearchtodatehasusedtheGHQ-12tocomputeaglobaldistressscore,thestructureanddimensionalityofthemeasureisstillamatterofdebate.Infact,themostcommonfindingfromthemanystudiesthathaveexploredthedimensionalityofGHQ-12isthefailuretofindsupportforasingle-factorstructure. Themostcommonalternativesemergingfromexploratoryanalysesseemtobeeitheratwo-factororathree-factorstructure.Inoneearlystudy,Politietal.(1994)identifiedtwodifferentfactorsthattheylabeled“Generaldysphoria”and“Socialdysfunction,”whichtheyalsofoundtohavedifferingdiscriminatorypower.Similartwo-factorstructureshavebeenfoundinseverallaterandmorerecentstudies,althoughthenomenclatureandqualitativemeaningdesignatedtothedifferentfactorshavevariedsomewhatacrossstudies.Forexample,Centofantietal.(2019)consideredthesecondfactortobeanexpressionof“Generalfunctioning”ratherthan“Socialdysfunction,”whileGlozahandPevalin(2015)consideredthefirstfactortoreflect“Socialanxiety”ratherthan“Generaldysphoria.”Othershaveusedlabelssuchas“Anxiety/Depression,”“Distress”and“Lossofpositiveemotions”(DoiandMinowa,2003;Sarkováetal.,2006;Suzukietal.,2011;Gaoetal.,2012).Addingtotheconfusionarestudiesthatpresentqualitativelysimilarfactors,butwhichoftencontainnoticeablydifferentfactorloadingpatterns(e.g.,Montazerietal.,2003;Cuéllar-Floresetal.,2014;Gelayeetal.,2015). Alternativemodelswiththreefactorsalsoexistintheliterature(Picardietal.,2001;DoiandMinowa,2003;Padrónetal.,2012;Gelayeetal.,2015;Guan,2017),ofwhichthemodelproposedbyGraetz(1991)havegainedthemostattentionandhavelaterbeenreplicatedinconfirmatoryanalyses(FrenchandTait,2004;ShevlinandAdamson,2005;AbubakarandFischer,2012).Thismodeldistinguishesbetween“Anxiety,”“Socialdysfunction”and“Lossofconfidence.”ThesocialdysfunctionfactorinGraetz’smodelmirrorsthenamesakefactorinPolitietal.(1994)model,whereastheanxietyandlossofconfidencefactorsisabreakdownofthegeneraldysphoriafactorintotwodifferentfactors. Anumberoftwo-andthree-factormodelshaveroutinelyalsobeentestedwithinaconfirmatoryfactor-analyticalframework,withsupportfoundforbothtypesofmodels.Severalstudieshavefoundatwo-dimensionalrepresentationtofittheobserveddatabest,butthesearenotalwayscomparableastheydifferinrespecttothelatentfactorcontent,theparameterizationofthemodeloreventhenumberofitemsincludedintheanalysis.Forexample,studieshavefoundsupportfortwofactorsbasedonareduced7-item(WongandO’Driscoll,2016),8-item(Kalliathetal.,2004;IpandMartin,2006),or10-item(Salama-Younesetal.,2009)versionoftheGHQ.Othershaveincludedcorrelationsbetweentheuniquevarianceofitemswithoutprovidinganylogicalortheoreticaljustificationsfortheseadditions(e.g.,Namjooetal.,2017).Confirmatorythree-factormodels,incontrast,haveforthemostpartfollowedthemodelproposedbyGraetz(1991).Forexample,FrenchandTait(2004),ShevlinandAdamson(2005),andAbubakarandFischer(2012)allfoundthethree-factormodeltobethebestfittingmodelamongthosetested(whichalsoincludedatwo-dimensionalmodel). Amajorproblemwithboththedominatingtwo-dimensionalmodelbyPolitietal.(1994)andthethree-dimensionalmodelbyGraetz(1991)istheseparationofnegativelyandpositivelyphraseditemsintoseparatefactors.TheGHQ-12consistsofanequalnumberofpositivelyandnegativelyphraseditems,anditiswellknownthatwhenpsychologicalratingscalescontainamixofnegativelyandpositivelyphraseditems,factoranalysesoftheseitemsoftenrevealapparentlydistinctfactorsreflectingthewordingoftheitems(Marsh,1996).Thisisindeedthecasewithboththetwo-factorandthree-factormodels.InPolitietal.(1994)two-factorstructure,allpositivelywordeditemsloadedononefactorandallnegativelyphraseditemsloadedontheotherfactor.Theonlyexceptionwasitem12(“Beenfeelingreasonablyhappy”),whichloadedaboutequallyonbothfactors.Similarly,inGraetz’sthree-factormodel,onefactorcontainsallthepositivelyphraseditems,whilethenegativelyphraseditemsaredividedintotwoseparatefactors.Incaseslikethese,thequestionariseswhetherthesefactorsaresubstantivelymeaningfulfactorsorartifactsofresponsestylesassociatedwiththepositivelyandnegativelyphraseditems. Inresponsetothischallenge,laterstudieshaveexplicitlytriedtomodelwordingeffectsinconfirmatoryfactormodels.Hankins(2008)comparedatwo-andthree-dimensionalmodelwithaunidimensionalmodelthatincorporatedwordingeffectsbyallowingcorrelatederrortermsonthenegativelyphraseditems.Resultsfromthiscomparisondemonstratedthattheunidimensionalmodelwithwordingeffectsprovidedabetterfitthanboththetwo-dimensionalandthree-dimensionalmodel.Whilecorrelatederrorsareclearlyindicativeofsystematicerrorvariance,theydonotnecessarilypointtoasingle,commonmethodfactorasanexplanation,asseveraldifferentlatentfactorsmaycausethesecorrelations.However,Ye(2009)tookasimilarapproachandmodeledaspecificmethodfactorassociatedwiththenegativeitemsinadditiontoageneralfactorrepresentinggeneraldistress.Yefoundthatthismodelprovidedagoodfittothedata,althoughbothatwo-andthree-dimensionalmodelfittedequallywell. Studieshaveextendedthislogicbyincludingtwoseparatespecificfactors,oneforthenegativelyphraseditemsandoneforthepositivelyphraseditems.Thissortofmodelisoftenreferredtoasabifactormodel,andareusedinsituationswhenthecovarianceamongasetofitemscanbeaccountedforbyasingle,unidimensionalfactorthatrepresentsthecommunalityamongalltheitems,inadditiontodomain-specificfactorsthatreflectadditionalcommonvarianceamongsubsetsoftheitems(Reiseetal.,2007).Bifactormodelinghassomeadvantagesovertraditionalfactormodels,becauseitallowsustoexamineifameasureisessentiallyunidimensionaloriftheitemsaremultidimensionalandwhethersubscalescoresprovideadditionalreliableinformationbeyondthetotalscore(Reiseetal.,2007,2013a).Forexample,inadditiontotraditionalfitstatistics,bifactormodelsalsoofferstheopportunitytoevaluatethepercentageofcommonvariancethatcanbeattributedtothegeneralfactorinthemodel(Reiseetal.,2013b). BothTomásetal.(2017)andCentofantietal.(2019)includedbifactormodelsamongthedifferentfactorstructuresthattheytested,withsomewhatmixedresults.Centofantietal.(2019)foundthatthebifactorwasthebestfittingmodelandreportedanomegahierarchical(ωh)valueof0.81.Omegahierarchicalisanexpressionofthetotalamountofobservedscorevariancethatisattributabletothegeneralfactorinabifactormodel,andωh=0.81thussupportsthepresenceofastronggeneraldistressfactor.Tomásetal.(2017),ontheotherhand,foundthatabifactormodeldidnotimprovethefitoverathree-factormodelbasedonGraetz(1991). AimsoftheCurrentStudy Theimportanceofvalidandeasytousetoolsforscreeningmilitarypersonnelcannotbeoverestimated.Forinstance,thescaleofUnitedStatesmilitarydeploymentsisextremelyhighwith7.5milliontroopsdeployedsince9/11(McCarthy,2018).AsanexampleofaEuropeannation,theUnitedKingdomhasdeployedalmost300,000troopstoAfghanistanandIraqalone(MinistryofDefence,2015).Norwayhasoverthelastdecadesincreasedtheirparticipantsininternationaloperationsanditisestimatedthatabout100,000soldiershavebeendeployedto40countriessinceworldwartwo(NorwegianArmedForces,n.d.). ThewidespreaduseoftheGHQ-12bothincivilianandmilitarysettingscombinedwiththeongoinguncertaintyregardingitsfactorstructure,motivatedustoscrutinizefurtherthepsychometricpropertiesofthemeasure.Despitethemanydifferentmodelsthathavebeenproposedandtestedintheliterature,thereisstillnoconsensusregardingthemostappropriatedimensionaldescriptionoftheGHQ-12.BecausethefactorstructureoftheGHQ-12toalargedegreeseemstovaryfromstudytostudyandsampletosample,itisalsoimportanttoexaminewhetherthefactormodelidentifiedasthebeststructurecanbereplicatedindifferentsamplesorovertimeinthesamesample.TestingformeasurementinvarianceoftheGHQ-12allowsustoexaminewhethertheitemsoftheoveralldistressfactororanysub-factorsareinterpretedthesameacrosssamplesandmeasurementpoints. IftheGHQ-12measuresqualitativelydifferentconstructsratherthanageneralandunidimensionalmentalhealthfactor,thenweshouldexpectthedifferentfactorsalsotohavedistinctnomologicalnetworks.Furthermore,ifthesubscalesaretoofferanyutility,thenamultidimensionalmodelshouldofferuniquepredictivevaliditybeyondageneralGHQ-12factor,inadditiontoprovidingastatisticallybetterfitinaconfirmatoryfactoranalysis.ShevlinandAdamson(2005),forexample,questionedtheutilityofthethree-factormodeltheyfoundtobethebestrepresentationofthedata,becausethethreefactorsprovidedlittleinformationbeyondthatofageneralfactor.Inthepresentpaper,weplantoexaminetheassociationsbetweenGHQandsymptomsofinsomniaandmentalhealth. Theaimsofthisarticlewerethereforeto(a)testandcomparethedifferentmodelsthathavebeenproposedinpreviousstudies;(b)assesswhetherthemodelidentifiedasthebestfittingmodelwasinvariantacrosssamplesandacrosstime;and(c)explorewhetherthedifferentlatentfactorsunderlyingtheGHQ-12(ifany)havedistinctnomologicalnetworksorpredictivevaliditybeyondageneralfactor. MaterialsandMethods Samples Atotalof591crewmembersfromfourdifferentfrigatesservingintheRoyalNorwegianNavycomprisedourfirstsample.Thefrigatesweredeployedininternationalanti-piracyoperationsandstandingNATOmaritimedeploymentsatvariousperiodsduring2013–2017.Thissampleservedasourprincipalsamplethatweusedtotestandcomparethevariousfactormodels,aswellasexplorethepredictiveutilityandtesttheinvarianceofthebestfittingmodelacrosstwotime-points. Crewmembers(N=196)fromthreedifferentminehunters/sweepersservinginstandingNATOminecountermeasuregroupdeployments(between2014and2017)servedasthesecondsample.ThissamplewasusedtotestifthefactormodelidentifiedasthebestfittingmodelinSample1wasinvariantacrossgroups.Bothsamplesbelongedtovesselssailingininternationaloperations.Normaldeploymentcyclesare6and4monthsforfrigatesandmine-countermeasurevessels,respectively. Theabovesampleswerechosenbecausealthoughbothmilitary,therearealsosomekeydifferencesbetweenthem.Comparedwithminehunters,frigatesarelargervesselsandareusuallymannedbyacrewofabout120sailors.Minehuntersareconsiderablysmaller,withacrewofabout35.Thecrewonboardfrigatesareonaverageolderandcomprisesrelativelymoreofficersandenlistedpersonnel.Themanagementstructureonboardalsodiffers.Duetotheirsize,frigatesarecharacterizedbyastrongerhierarchicalstructure,whichfurtherentailsthattheleadershipislessdirectandtoalargerdegreeexecutedthroughdepartmentheads.Minehunters,onthecontrary,arecharacterizedbylesssocialdistanceandamoredirectleadershipstructure.Frigatesfurtherhavethecapacityforlongertimesatseawithoutreplenishmentandhavemorevariedoperationalcapacities,includinggreen-water(coastal)andblue-water(openocean)operations.Minehunters,ontheotherhand,arelessself-sufficientandgenerallyspendlesstimeatsea,andtheiroperationalcapacitiesaremoreweatherdependentandrestrictedtocoastalwaters. Procedure ThedatacollectionwaspartofthestandardprocedureforpsychologicalevaluationintheRoyalNorwegianNavy(seeSandenetal.,2014,foranoverview).Theproceduresincludepre-deploymentscreeningaswellasmid-andpost-deploymentevaluation.Thepost-deploymentscreeningwasconductedwhiletransitingbacktotheNorwegianhomebase.Inthecurrentpaper,weusedatafromthepre-andpost-deploymentscreenings. Measures GeneralHealthQuestionnaire-12(GHQ-12) TheGHQ-12consistsof12statementstowhichrespondentsindicateagreementonafour-pointscale(0=Notatall;3=Morethanusual;GoldbergandWilliams,1988).AllitemsareavailableinTable2. BergenInsomniaScale(BIS) TheBISconsistsofsixitemsmeasuringdifferentaspectsofinsomnia(e.g.,sleeponset,earlymorningwakeninganddaytimeimpairment),constructedbasedontheinclusioncriteriaforinsomniaintheDiagnosticandStatisticalManualofMentalDisorders(Pallesenetal.,2008).Foreachitem,participantsindicatehowmanydaysperweekduringthelastmonththeyexperiencedproblemswiththatparticularaspectofsleep.Eachitemisratedonan8-pointscale,rangingfrom0to7daysperweek.Theitemscanbecombinedtocreateasingleinsomniascore.Anexampleitemis:“Duringthepastmonth,howmanydaysaweekhasittakenyoumorethan30mintofallasleepafterthelightwasswitchedoff?”(sleeponset). HopkinsSymptomChecklist-25(HSCL-25) TheHSCL-25isa25-itemscreeningtooldesignedtodetectsymptomsofanxietyanddepression(Derogatisetal.,1974).Respondentsareaskedtoindicatetowhatdegree(1=notatall;4=verymuch)eachofthe25symptomshavebeentroublingorconcerningthemduringthelast2weeks.Exampleitemsare“Suddenlyscaredfornoreason”and“Spellsofterrororpanic.”Allitemscanbecombinedtoformatotaldistressscore.Alternatively,thefirst10itemscanbeusedtocreateananxietyscoreandthelast15itemscanbeusedtocreateadepressionscore. StatisticalAnalyses Weplannedtoexaminearangeofdifferentfactormodelspreviouslyusedintheliterature,illustratedinFigure1andbrieflydescribedbelow: FIGURE1 Figure1.ThedifferentfactormodelsforGHQ-12testedandcompared. Model1 Aunidimensionalmodelwithasinglefactorexplainingthecovariancebetweenallitems. Model2 Amodelwithtwocorrelatedlatentfactorscontainingonefactorwithallnegativelyphraseditemsandonefactorwithallpositivelyphraseditems.ThismodelwasoriginallyproposedbyAndrichandvanSchoubroeck(1989)andisidenticaltotheGeneraldysphoriaandSocialdysfunctionfactorsproposedbyPolitietal.(1994),exceptforoneitemthatloadedonbothfactorsinthelatterstudy.Forconceptualclarity,wedonotincludethisdoubleloading. Model3 Acorrelatedthree-factormodeloriginallysuggestedbyGraetz(1991).Thethreelatentfactorsinthismodelrepresent“Anhedonia/Socialdysfunction”(allpositivelyphraseditems),“Anxiety/Depression”(fournegativelyphraseditems)and“Lossofconfidence”(twonegativelyphraseditems).Themajordifferencebetweenthismodelandthepreviouscorrelatedtwo-factormodelisthatitdividesthenegativelyphraseditemsintotwodistinctlatentfactors. Model4 Aunidimensionalmodelwithanadditionalorthogonalmethodfactorspecificallyforthenegativeitems(Ye,2009).ThismodelreframesModel2asanartifactualdivisionintodifferentfactorscausedentirelybywordingeffects. Model5 ThismodelextendsModel4toincludeanorthogonalspecificfactorforthepositivelyphraseditemsaswell.Thistypeofmodelisoftenreferredtoasabifactormodel(Reiseetal.,2007)andhaspreviouslybeentestedbyTomásetal.(2017)andCentofantietal.(2019). ModelFitand-Comparison Individualmodelfitwasevaluatedbyexaminingthesizeandstatisticalsignificanceoffactorloadings,aswellasseveralcommonlyusedgoodness-of-fitstatistics.Specifically,weusedthecomparativefitindex(CFI),thestandardizedrootmeansquareresidual(SRMR)andtherootmeansquareerrorofapproximation(RMSEA),togetherwithits90%confidenceinterval(CI).MacCallumetal.(1996)havesuggestedthatvaluesof0.01,0.05,and0.08forRMSEAcorrespondtoexcellent,goodandmediocrefit,whereasavaluelessthan0.08forSRMRisgenerallyconsideredagoodfit(HuandBentler,1999).ForCFI,avaluecloseto0.95indicatesagoodfitbetweenthehypothesizedmodelandtheobserveddata,whereasvaluesintherangeof0.90–0.95areconsideredacceptable(Kline,1998;HuandBentler,1999;McDonaldandHo,2002). Tocomparethecompetingmodels,weusedtwomeasuresofcomparativefit,TheAkaikeinformationcriterion(AIC)andtheBayesianinformationcriterion(BIC),aswellasthelikelihood-ratiotestwhenappropriate.LowervaluesforbothAICandBICindicateabetterfit. PredictiveValidity Thefactormodelidentifiedasthebestfittingmodelwasincludedinafullstructuralmodel(SEM)withinsomnia(BIS)andmentalhealthproblems(HSCL-25)asendogenouslatentvariables.WeformeditemparcelsoftheindicatorsforbothBISandHSCLinordertokeepthecomplexityofthemodeltoaminimum.Whentheinterestliesinthestructuralrelationshipsratherthanthemeasurementparameters,itemsparcelingcanbedefensibleundersomepreconditions(BandalosandFinney,2001).Importantly,itemsshouldbecombinedonlywithinunidimensionaldomains.PreviousfactoranalysesofBIShavesuggestedbothasinglefactorandatwo-factorsolutionwherenocturnalsymptomsanddaytimesymptomsofinsomniaformedseparatefactors(Pallesenetal.,2008).Anexploratoryfactoranalysisinourdatasetreproducedthetwo-factorsolutionwiththefirstthreeitemsloadingontoonefactorandthethreelastitemsloadingontoasecondfactor.Wethereforecreatedtwoparcelsfortheinsomniaitems,onecontainingthefirstthreeitemsandonecontainingthelastthreeitems. AlthoughtheHSCL-25wasoriginallythoughttocaptureseparateanxietyanddepressiondimensions,lateranalyseshavesuggestedavarietyofdifferentfactorstructures(Skogenetal.,2017).Aninitialexploratoryfactoranalysisinourdatasetsuggestedfivefactorswitheigenvaluesgreaterthantheaverageoftheinitialcommunalities(i.e.,ananalogtotheeigenvalue-greater-than-one-ruleusedforprincipalcomponentanalysis;Afifietal.,2012).However,thefirstfactorwasclearlydominant,withaneigenvaluemorethansixtimesgreaterthantheeigenvaluesoftheotherfourfactors.Moreover,theremainingfourfactorshadeigenvaluesthatwereonlymarginallylargerthantheaverageoftheeigenvaluesfromsimulateddata(parallelanalysiswith10,000replications).Wethereforedecidedtoextractasinglefactorandthendroptheitemswithhighuniqueness(>0.70).Theremaining11itemswerecombinedintothreeparcels,twowithfouritemseachandonewiththreeitems. MeasurementInvarianceAcrossSamplesandTime Testingformeasurementinvarianceacrosssamplesentailsseveralsteps,eachwithsuccessivelymorerestrictionsplacedonthemodels.First,weperformedatestofconfiguralequivalence,whereinequalfactorstructuresaretested.Thisisachievedbyspecifyingthesamepatternoffixedandfreefactorloadingsinboththefrigateandtheminehuntersampleinamulti-groupCFA,andaimsatexaminingwhethertheGHQ-12evokesthesamecognitiveframeofreferenceforrespondentsacrosssamples.Thismodelalsoservesasabaselinemodelwithwhichlater,morerestrictedmodelscanbecompared.Second,weperformedatestofmeasurementinvariance,inwhichfactorloadingsforlikeitemsareconstrainedtoequalityacrossthetwosamples.Thisexamineswhethertheassociationsbetweenlikeitemsandtheunderlyingconstructsarethesameacrossgroups,andthuswhethertheconstructindicators(i.e.,theitems)arecalibratedtotheunderlyingconstructinthesamemanner. Allerrorvarianceswereallowedtovaryfreelyacrossthetwosamples,becausetherequirementthaterrorvariancesbeequalbetweengroupsisconsideredexcessivelystringentandoflittlepracticalvalue(ByrneandWatkins,2003).Becausetheobjectiveofthecurrentstudywasnottocomparelatentfactorscoresacrosssamples,wealsodidnotconstraininterceptstoequalityacrosssamples. Theprocessoftestinginvarianceovertimeissimilartotestinginvarianceacrosssamples,exceptthatwenolongerestimateamulti-groupCFA,butinsteadfitasinglemodelinthefrigatessample.Forthetestofconfiguralinvariance,thesamenumberoflatentfactorsarespecifiedatbothtime-points,withthesamepatternoffixedandfreefactorloadingsateachappropriatetime-point.Inaddition,covariancebetweenthecorrespondingfactorsatT1andT2,aswellasbetweenresidualsforlikeitems,areincludedtoallowforthemlikelycorrelatingovertime.Exceptofanyconstraintsneededforidentificationpurposes,nootherconstraintsonthefactorloadingsareincludedatthistime. Asbefore,thetestformeasurementinvarianceinvolvesconstrainingallfactorloadingstobeequivalentacrosstime-points.Forbothinvarianceacrosstimeandsamples,themorerestrictedmeasurementinvariancemodelisnestedinthebaselinemodelthatallowsallparameterstovaryfreelyandcanthereforebestatisticallycomparedusingalikelihood-ratiotest(LRχ2). Results Table1presentsthefitstatisticsforthedifferentplannedmodels.Theunidimensionalmodel(Model1)asoriginallyproposedshowedtheworstfitofallmodelstested(CFI=0.754,SRMR=0.079,RMSEA=0.110,and90%CIforRMSEA=0.101–0.120).Thefitstatisticsimprovedsomewhatwiththetwomultidimensionalalternativeswithoutmethodeffects.However,onlythethree-factormodel(Model3)obtainedacceptablestatisticsonboththeSRMRandtheRMSEA(SRMR=0.57,RMSEA=0.074,and90%CIforRMSEA=0.064–0.085).InModel2,thetwofactorscorrelatedr=0.62,whereasinModel3thefactorcorrelationswere:r(P,N1)=0.69;r(P,N2)=0.50;andr(N1,N2)=0.81.ItshouldbenotedthatthefactorloadingforitemP5(“Beenabletofaceproblems”)wasnon-significantinallmodelssofar.Weneverthelesschosenottore-runourmodelswiththisitemdeletedsothatourmodelsareascomparableaspossibletothemodelspreviouslytestedintheliterature. TABLE1 Table1.Fitstatisticsforthetestedmodelsofthe12-itemGeneralHealthQuestionnaire(GHQ-12),N=562. Themodelwithanartifactualfactorcontainingallthenegativeitems(Model4)didnotfitthedatabetterthanthethree-factormodel(CFI=0.880,SRMR=0.059,RMSEA=0.082,and90%CIforRMSEA=0.071–0.093).Inaddition,boththeAICandtheBICweresmallerforthethree-factormodelthanfortheartifactualmodel.Thebifactormodel(Model5),onthecontrary,obtainedacceptablevaluesonallfitstatistics,andhadthelowestAICandBICvaluesofallmodelstested(seeTable1). ThestandardizedfactorloadingsforthebifactormodelarepresentedinTable2.WorthnoticingfirstisitemP5thatdoesnotloadsignificantlyonanyofthefactors.Thisitemthereforedoesnotseemtobeagodmarkerforeitherthegeneralfactororthespecificfactor.Intotal,thegeneralfactoraccountsforabout55%ofthecommonvarianceinthe12GHQitems(ECV=0.547). TABLE2 Table2.FactorloadingsandvariancecompositionforthebifactormodeloftheGeneralHealthQuestionnaire(GHQ-12). Theomega(ω)valueforthegeneralfactorisanexpressionoftheamountofobservedscorevarianceaccountedforbyalltheconstructsthatunderlieascalescore(Brunneretal.,2012),thatis,thegeneralfactorandthetwospecificfactorsinthisinstance.Thus,ifaunit-weightedtotalscalescoreofthe12GHQitemswascreated,81%ofthevarianceinthisscalescorewouldbeaccountedforbythegeneralfactorandthetwospecificfactorsincombination(ω=0.81).Omegahierarchical(ωh),ontheotherhand,isanexpressionofthetotalamountofobservedscorevariancethatisattributabletojustthegeneralfactor.FromTable2,onecanseethatapproximately60%ofthetotalscorevarianceisaccountedforbythegeneralfactor(ωh=0.598).Bytakingthesquarerootofωh,wecanalsogetanexpressionofthecorrelationbetweentheunit-weightedcompositescoreandthetargetfactor.Thus,theωhof0.598wouldindicateacorrelationof0.77betweenthegeneralfactorandtheobservedtotalscore.Reiseetal.(2013a)havesuggestedthatωhvalues>0.50canbeusefulindeterminingwhetheracompositescoreprovidesunique,reliablevariance.Conversely,valuesbelowthisrenderacompositescorebasedontheindicatorsverydifficulttointerpret,aslessthanhalfoftheobservedvarianceinthecompositescorewouldbeduetotheconstructofinterest(GignacandWatkins,2013). Theomegahierarchicalsubscale(ωhs)istheomegacounterparttoωhapplicabletothespecificfactors.About23%ofthevarianceinthepositivesubscalescoreisaccountedforbythespecificfactor(ωhs=0.225)andabout36%ofthevarianceinthenegativesubscaleisaccountedforbythespecificfactor(ωhs=0.360)aftercontrollingfortheeffectsofthegeneralfactor.Theωhsvaluesforboththespecificfactorsarequitelowrelativetotheirrespectiveomegavalues(ωsinTable2),suggestingthatmuchofthereliablevarianceofthesubscalescorescanbeattributabletothegeneralfactorratherthanwhatisuniqueforthesetwospecificfactors(Rodriguezetal.,2016).Dividingtheωhsvaluebytheωsvaluegivestherelativeomega,whichshowsthatonlyabout34%ofthevarianceinthepositivesubscale(0.668/0.225)and46%ofthevarianceinthenegativesubscale(0.774/0.360)isindependentofthegeneralfactor. PredictiveUtility Totestthepredictiveutilityofthegeneralfactorversusthetwospecificfactors,weperformedstructuralequationmodelingwithBISandHSCLasendogenouslatentvariablespredictedbythegeneralandthetwospecificfactors.PriortoperformingthefullSEManalysis,wefirstperformedaCFAtoverifythemeasurementportionofthemodelsinvolvingthelatentBISandHSCLfactors.Thistwo-factorCFAmodelresultedinagoodfittothedata(CFI=0.998,SRMR=0.016,RMSEA=0.033,and90%CIforRMSEA=0.000–0.076). TheresultsfromthestructuralmodelareillustrateinFigure2.ThegeneralfactorwasstronglyandstatisticallysignificantlyassociatedwithboththeBISandHSCLfactorsintheexpecteddirection.Thatis,higherlevelsonthegeneralGHQ-factorwasassociatedwithhigherlevelsofinsomniaasmeasuredbyBISandmentalhealthsymptomsasmeasuredbyHSCL.Noassociationswerefoundbetweenthepositivesub-factorandeitherBISorHSCL,whereasthenegativesub-factorwaspositivelyandstatisticallysignificantlyassociatedwithmentalhealthsymptoms. FIGURE2 Figure2.StructuralmodelwiththegeneralGHQ-factorandtwospecificsub-factorspredictingsymptomsofinsomniaandmentalhealth(χ2=368.737,p<0.001,CFI=0.92,SRMR=0.048,RMSEA=0.069,90%confidenceintervalforRMSEA=0.061–0.077).Regressionweightsarestandardizedcoefficients.***p<0.001.**p<0.01. ModelInvariance InvarianceAcrossSamples Westartedbytestingforconfiguralinvarianceacrossthefrigateandminehuntersamples.Thisentailsfittingthesamebifactormodelstructurewithallparametersestimatedfreelyacrossthesamplesinamultigroupmodel.Thismodelalsoservesasabaselinemodelwithwhichlater,morerestrictedmodelscanbecompared.Themultigroupbifactormodelfitthedataacceptablywell,χ2=262.380,p<0.001,RMSEA=0.075,90%CIforRMSEA=0.065–0.086,RMSEA=0.060,andCFI=0.911. Next,weconstrainedallfactorloadingstoequalityinatestofmetricinvariance.Themorerestrictedmetricinvariancemodelisnestedinthebaselinemodelthatallowsallparameterstovaryfreelyandcanthereforebestatisticallycomparedusingalikelihood-ratiotest(LRχ2).Theresultofthiscomparisonshowedthatthemodelconstrainingallfactorloadingsfitthedataequallywellasthelessrestrictedbaselinemodel,LRχ2=31.16,df=24,p=0.15. Insum,theseanalysespointtowardtheevidenceofequal-forminvariance,inthatthenumberoffactorsandthepatternoffactor-indicatorrelationshipsareequivalentacrosssamples(i.e.,configuralinvariance).Further,resultsfromthetestofmetricequivalencesuggeststhateachitemcontributedtothelatentfactorstoasimilardegreeacrossthetwosamples. InvarianceAcrossTime ApproximatelyhalfoftheparticipantsfromourfrigatesamplecompletedtheGHQ-12asecondtimeapproximately6monthsafterthefirstadministration(n=276).Toexaminethestabilityofthebifactormodelovertime,wenexttestedformeasurementinvarianceacrossthesetwotime-pointsinthissub-sample.BecauseofmissingvaluesononeormoreoftheGHQitemsateithertime-points,oursamplewasfurtherreducedn=248.Aswiththetestofinvarianceacrosssamples,westartedbyestablishingwhetherthepatternofloadingswassimilaracrosstime(i.e.,configuralinvariance).Toachievethis,wefittedamodelwithsixfactors(twogeneralfactorsandfourspecificfactors)whereallthelatentfactorsloadedontheitemsfortheappropriatetime-point.Weincludedcorrelationsbetweenthecorrespondingfactorstoallowfortheconstructslikelybeingcorrelatedovertime.Thecovariancebetweennon-correspondingfactors(e.g.,thegeneralfactoratT1andspecificfactorsatT2)wasconstrainedtozero.Wealsoincludedcorrelationsbetweentheresidualsforcorrespondingitemsacrossthetime-pointstoallowforsystematicuniquevarianceintheitemsacrosstime. Thefitofthismodelwasacceptablejudgedbytwoofthefitstatistics(RMSEA=0.062andSRMR=0.060),butnotthethird,CFI=0.879.Thegeneralfactorcorrelatedr=0.50overtime,whereasthecorrelationsforthespecificfactorswasr=0.26forthepositivefactorandr=0.79forthenegativefactor.Constrainingallfactorloadingstobeequalacrosstimeinatestofmetricinvarianceresultedinasignificantlyworsefittingmodel,LRχ2=50.39,df=24,p<0.01.Inspectingthefactorloadingsfromtheinitialunconstrainedmodeldidsuggestdiscrepanciesinsomeofthefactorloadingsacrossthetwotime-points.Mostofthesediscrepancieswereminor,exceptforoneloadingonthespecificpositivefactor(itemP2)andoneloadingonthespecificnegativefactor(itemN6).Allowingthesetwoitemstovaryfreelyacrosstime-pointsresultedinanon-significantlikelihoodratio,LRχ2=30.94,df=22,p=0.097. Discussion ThispapertestedaseriesofalternativefactorstructuresforthewidelyusedGHQ-12scale.Amongthefivealternativemodelstested,abifactorstructurewithonegeneralfactorandtwospecificfactorsprovedtobethebestrepresentationofthedatafromastatisticalperspective.Thismodelallowedforfactor-specificresidualvariationsbeyondageneraldistressfactorcommontoall12items.Becausethesefactor-specificvariationsreflectthedifferentphrasingoftheitems,withonefactorcontainingentirelynegativelywordeditemsandonefactorcontainingentirelypositivewordeditems,theyaremostlikelyanexpressionofmethod-specificvariance(Hankins,2008;Ye,2009).OurresultsthereforesuggestthattheGHQ-12isnotstrictlyunidimensional,butratherreflectssomemultidimensionalityduetowordingeffects.ThismultidimensionalitycanposeachallengewhenusingtheGHQ-12tocomputeaglobaldistressscorebyeitheraveragingorsummingallitemsasiscommonlydone,becausethiscompositemayreflecttheinfluenceofdifferentsourcesbeyondthegeneraldistressfactor.Incontrast,itisnotuncommonforfactoranalyticstudiesofpsychologicalmeasurestorevealminorsecondarydimensionsinadditiontoadominantgeneralfactor(Marsh,1996).Inouranalyses,thegeneralfactoraccountedfornearly60%ofthetotalscorevariance(ωh=0.598),whilethevarianceassociatedwithourtwospecificsub-factorswereincontrastrelativelysmall(23and36%forthepositiveandnegativefactors,respectively).Infact,alargerproportionofthevarianceassociatedwiththespecificfactorscouldbeattributedtothegeneralfactorthantowhatwasuniquetothesetwofactors.AsnotedbyRodriguezetal.(2016,p.225),interpretingsuchfactors“asrepresentingtheprecisemeasurementofsomelatentvariablethatisuniqueordifferentfromthegeneralfactor,clearly,ismisguided.”Fromanappliedperspective,wethereforebelievethatthepossiblebiasintroducedtoaglobalcompositescoreduetomultidimensionalityorwordingeffectsmostlikelywillbesmall. ThattheGHQ-12itemsprimarilyreflectageneralfactordespitetheevidenceofsomemultidimensionalityalsoimpliesthatcreatingsub-factororsubscalescoresismostlikelyoflimitedusefulness.ThiswasalsoillustratedinourSEManalysis,wherethegeneralfactorhadstrongassociationswithBISandHSCL,whereasthetwospecificfactorswerenon-significantlyorconsiderablyweakerassociatedwiththecriterionvariables.Otherresearcherswhohaveassessedthepredictivevalidityofsubscalesvis-à-visageneralfactorhavereachedsimilarconclusions.Forinstance,bothGaoetal.(2004)andShevlinandAdamson(2005)foundthethree-factorsolutionbasedonGraetz(1991)tobethebestrepresentationofthedata.However,whenexaminingtheutilityofthethreesubscales,theyappearedtoprovidelittleinformationbeyondthatofageneralfactor.Gaoetal.(2004)alsoconcludethatthereislittleneedtoconsiderthemultidimensionality,butratherthat“fromapragmaticpointofviewweconsideritacceptabletousethisinstrumentasaone-dimensionalmeasure”(p.6). Thefinalfindingfromourstudyisthatthebifactormodelprovedtobefairlyrobustacrossdifferentsamples.Wefoundthestructuretobeinvariantacrosstwodifferentmilitarysamples,onecomprisingcrewmembersfromfrigatesandtheothercomprisingcrewmembersfromminehunters/sweepers.Ourresultsshowthattheparticipantsinthetwosamplesrespondedtotheitemsinasimilarmannerandattributedthesamemeaningtothelatentfactors.Incontrast,thebifactormodelwaslessrobustwhentestedforinvarianceacrosstwotime-pointsinthefrigatesample.Ourbaselinemodelthatspecifiedthesamepatternoffixedandfreeloadingsatthetwotime-pointsdidnotprovideagoodfittothedata,atleastasjudgedbyoneofthefitstatisticsweused(CFI=0.879).Thissuggestthatthecrewmembersonboardthefrigatesdidnotconceptualizetheconstructsinthesamewayatthetwodifferenttime-points.Theprerequisiteforfurthertestingthemetricinvarianceacrosstimewasthereforestrictlyspeakingnotmet. Onereasonforthechangeinconceptualizationoftheitemscouldbeduetoanend-effectofthemissions.End-effectsrepresentachangeinevaluationsandperformanceattheendofatask,andaprerequisiteforsuchaneffectistheknowledgeoftheendpointofthetask.Sincethepost-evaluationofthecrewwasperformedintransitbacktohomebase,allcrewmembershadaknowledgeoftheterminationofthemissionandtheevaluationwasperformedclosetothisendpoint.End-effectshavebeenfoundinseveraldomainsofpsychology(e.g.,Catalano,1973;Lai,2008).UsingGHQasameasureofmentalwell-being,Taylor(2006)foundapositiveend-effectwithanincreasedwell-beingatlatecomparedtodatacollectedearlyintheweek. Limitations OuranalyseswerelimitedtotheLikertscoringsystemoftheGHQ-12.Whilethisisapopularapproach,othermethodshavebeenproposedandusedintheliterature(foranoverview,seeReyetal.,2014).Oneoptionistodichotomizetheitemsbycollapsingthefirsttworesponsecategoriesandthelasttworesponsecategoriesandscoringthemasrespectively“0”and“1”(GHQ-0011).Aslightlydifferentapproachistousetheabovescoringsystemforthepositiveitemsbutcollapsethelastthreeresponsecategories(0-1-1-1)forthenegativeitems.Finally,differentLikert-typeformatshavealsobeenused,suchasasix-point(Kalliathetal.,2004)oraseven-pointscale(Ye,2009).Obviously,ourresultsdonotextendtothesedifferentscoringssystems.Onthecontrary,thereisevidencesuggestingthatthescoringsystemcanaffectthenumberoffactorsaswellastheparticularpatternofitem-factorloadings(Aguadoetal.,2012;Gaoetal.,2012;Reyetal.,2014). Themilitarysamplesusedinthepresentstudymustbeconsideredwhenassessingthegeneralizabilityofourresultstoothersamplesandsettings.Itisconceivablethatmilitarypersonnelmaydifferfromotheroccupationalgroupsand/orthegeneralpopulationinhowtheyperceiveandrespondtotheGHQ-12.Asfarasweknow,thereislimitedresearchthathavecomparedortesteddifferencesinGHQ-12betweenmilitarysamplesandotheroccupationalgroups.OneexceptionisastudybyGouveiaetal.(2012)thatfoundsomeminordifferencesinthefactorstructurebetweenamilitarygroupandtheothergroupsincluded(students,schoolteachersandthegeneralpopulation).However,theauthorsconductednodirectstatisticalcomparisonsofthedifferentsamples. Itmust,however,alsobestressedthattheNorwegianArmedForcesisbasedonmandatorymilitaryserviceforbothmenandwomen,andthusthecrewonboardNorwegiannavalvesselsconsistbothofprofessionalsoldiers,mainlyofficersandnon-commissionedofficers,aswellaslowerrankmandatoryconscripts.Oneargumentbehindconscriptionforbothwomenandmenistosecureabettercross-sectionofthepopulation.Selectionproceduresdoofcourseintroducesomelimitationsregardingwhoisallowedtoserve,aspsychopathologyandsubclinicalandclinicalsymptomsofadjustmentdisordersareexclusioncriteria.InourviewthiscouldbesaidtoworktoouradvantageastheGHQisnotintendedforseverepathologyandsuchcasescouldintroduceunwantednoisetoourdata. Althoughweacknowledgethatthegeneralizabilityofoursampleconstitutesalimitation,wewouldliketostressthattherearealsoadvantagesassociatedwithusingmilitarysamples.Mostimportantly,navalvesselsarerelativelyisolatedunits.Thisentailsthatthepersonnelonboardisexposedtoapproximatelythesamelevelsofisolationfromsignificantothers,thesameenvironmentalinfluences,thesametypesofstressorslikeexercises,andsoon.Comparedwithciviliansamples,ournavalsamplesthusoffergreatercontroloverexternalfactorsthatcanproducesymptomsofpsychologicaldistressandensuresthateveryoneonboardisexposedtoroughlythesametypesandlevels. Conclusion Overall,ourresultsarecongruentwiththesuggestionofHankins(2008)andYe(2009),andservalothersthatitemwordingcanintroduceresponsebiastotheGHQ-12.Asaresult,themultidimensionalitydemonstratedinmanypreviousstudiescanbeanexpressionofmethodeffects,specifically,thedivisionofGHQ-12intopositivelyandnegativelyphraseditems.Assuch,theGHQ-12isnotstrictlyunidimensional,butinadditioncontainsfactor-specificvariationsassociatedwiththeitemswording.However,theexplaineduniquevarianceassociatedwiththesespecificfactorswasrelativelysmall.Theconsequencesofignoringthismultidimensionalityandinsteaduseacompositescorearethereforemostlikelysmallformostpracticalpurposes. DataAvailabilityStatement Therawdatasupportingtheconclusionsofthisarticlewillbemadeavailablebytheauthors,withoutunduereservation,toanyqualifiedresearcher. EthicsStatement Ethicalreviewandapprovalwasnotrequiredforthestudyonhumanparticipantsinaccordancewiththelocallegislationandinstitutionalrequirements.Thepatients/participantsprovidedtheirwritteninformedconsenttoparticipateinthisstudy. AuthorContributions BJorganizedthedatacollection.SHandBJcontributedtothetheorydevelopment,designandwritingofthemanuscript.SHprimarilyconductedthestatisticalanalyses. Funding ThisresearchwassupportedbytheUniversityofBergen. ConflictofInterest Theauthorsdeclarethattheresearchwasconductedintheabsenceofanycommercialorfinancialrelationshipsthatcouldbeconstruedasapotentialconflictofinterest. References Abubakar,A.,andFischer,R.(2012).Thefactorstructureofthe12-itemgeneralhealthquestionnaireinaliterateKenyanpopulation.StressHealth28,248–254.doi:10.1002/smi.1420 PubMedAbstract|CrossRefFullText|GoogleScholar Afifi,A.A.,May,S.,andClark,V.A.(2012).PracticalMultivariateAnalysis.BocaRaton,FL:ChapmanandHall. GoogleScholar Aguado,J.,Campbell,A.,Ascaso,C.,Navarro,P.,Garcia-Esteve,L.,andLuciano,J.V.(2012).Examiningthefactorstructureanddiscriminantvalidityofthe12-itemgeneralhealthquestionnaire(ghq-12)amongSpanishpostpartumwomen.Assessment19,517–525.doi:10.1177/1073191110388146 PubMedAbstract|CrossRefFullText|GoogleScholar Andrich,D.,andvanSchoubroeck,L.(1989).Thegeneralhealthquestionnaire:apsychometricanalysisusinglatenttraittheory.Psychol.Med.19,469–485.doi:10.1017/s0033291700012502 PubMedAbstract|CrossRefFullText|GoogleScholar Bandalos,D.L.,andFinney,S.J.(2001).“Itemparcelingissuesinstructuralequationmodeling,”inNewDevelopmentsandTechniquesinStructuralEquationModeling,edsG.A.MarcoulidesandR.E.Schumacker(Mahwah,NJ:LawrenceErlbaum),269–296. GoogleScholar Böhnke,J.R.,andCroudace,T.J.(2016).Calibratingwell-being,qualityoflifeandcommonmentaldisorderitems:psychometricepidemiologyinpublicmentalhealthresearch.Br.J.Psychiatry209,162–168.doi:10.1192/bjp.bp.115.165530 PubMedAbstract|CrossRefFullText|GoogleScholar Brunner,M.,Nagy,G.,andWilhelm,O.(2012).Atutorialonhierarchicallystructuredconstructs.J.Pers.80,796–846.doi:10.1111/j.1467-6494.2011.00749.x PubMedAbstract|CrossRefFullText|GoogleScholar Byrne,B.M.,andWatkins,D.(2003).Theissueofmeasurementinvariancerevisited.J.CrossCult.Psychol.34,155–175.doi:10.1177/0022022102250225 CrossRefFullText|GoogleScholar Catalano,J.F.(1973).Effectofperceivedproximitytotheendoftaskuponend-spurt.Percept.Mot.Skills36,363–372.doi:10.2466/pms.1973.36.2.363 PubMedAbstract|CrossRefFullText|GoogleScholar Centofanti,S.,Lushington,K.,Wicking,A.,Wicking,P.,Fuller,A.,Janz,P.,etal.(2019).Establishingnormsformentalwell-beinginyoungpeople(7–19years)usingthegeneralhealthquestionnaire-12.Austr.J.Psychol.71,117–126.doi:10.1111/ajpy.12227 CrossRefFullText|GoogleScholar Cornelius,B.L.R.,Groothoff,J.W.,vanderKlink,J.J.L.,andBrouwer,S.(2013).TheperformanceoftheK10,K6andGHQ-12toscreenforpresentstateDSM-IVdisordersamongdisabilityclaimants.BMCPublicHealth13:128.doi:10.1186/1471-2458-13-128 PubMedAbstract|CrossRefFullText|GoogleScholar Credé,M.,Harms,P.,Niehorster,S.,andGaye-Valentine,A.(2012).AnevaluationoftheconsequencesofusingshortmeasuresoftheBigFivepersonalitytraits.J.Pers.Soc.Psychol.102,874–888.doi:10.1037/a0027403 PubMedAbstract|CrossRefFullText|GoogleScholar Cuéllar-Flores,I.,Sánchez-López,M.P.,Limiñana-Gras,R.M.,andColodro-Conde,L.(2014).TheGHQ-12fortheassessmentofpsychologicaldistressoffamilycaregivers.Behav.Med.40,65–70.doi:10.1080/08964289.2013.847815 PubMedAbstract|CrossRefFullText|GoogleScholar Derogatis,L.R.,Lipman,R.S.,Rickels,K.,Uhlenhuth,E.H.,andCovi,L.(1974).Thehopkinssymptomchecklist(HSCL):aself-reportsymptominventory.Behav.Sci.19,1–15.doi:10.1002/bs.3830190102 PubMedAbstract|CrossRefFullText|GoogleScholar Doi,Y.,andMinowa,M.(2003).Factorstructureofthe12-itemgeneralhealthquestionnaireinthejapanesegeneraladultpopulation.Psychiatr.Clin.Neurosci.57,379–383.doi:10.1046/j.1440-1819.2003.01135.x PubMedAbstract|CrossRefFullText|GoogleScholar Endsley,P.,Weobong,B.,andNadkarni,A.(2017).ThepsychometricpropertiesofGHQfordetectingcommonmentaldisorderamongcommunitydwellingmeninGoa,India.AsianJ.Psychiatr.28,106–110.doi:10.1016/j.ajp.2017.03.023 PubMedAbstract|CrossRefFullText|GoogleScholar French,D.J.,andTait,R.J.(2004).MeasurementinvarianceinthegeneralhealthQuestionnaire-12inyoungAustralianadolescents.Eur.ChildAdolesc.Psychiatry13,1–7.doi:10.1007/s00787-004-0345-7 PubMedAbstract|CrossRefFullText|GoogleScholar Gao,F.,Luo,N.,Thumboo,J.,Fones,C.,Li,S.-C.,andCheung,Y.-B.(2004).Doesthe12-itemgeneralhealthquestionnairecontainmultiplefactorsanddoweneedthem?HealthQ.LifeOutcom.2:63.doi:10.1186/1477-7525-2-63 PubMedAbstract|CrossRefFullText|GoogleScholar Gao,W.,Stark,D.,Bennett,M.I.,Siegert,R.J.,Murray,S.,andHigginson,I.J.(2012).Usingthe12-itemgeneralhealthquestionnairetoscreenpsychologicaldistressfromsurvivorshiptoend-of-lifecare:dimensionalityanditemquality.Psychol.Oncol.21,954–961.doi:10.1002/pon.1989 PubMedAbstract|CrossRefFullText|GoogleScholar Gelaye,B.,Tadesse,M.G.,Lohsoonthorn,V.,Lertmeharit,S.,Pensuksan,W.C.,Sanchez,S.E.,etal.(2015).Psychometricpropertiesandfactorstructureofthegeneralhealthquestionnaireasascreeningtoolforanxietyanddepressivesymptomsinamulti-nationalstudyofyoungadults.J.Affect.Disord.187,197–202.doi:10.1016/j.jad.2015.08.045 PubMedAbstract|CrossRefFullText|GoogleScholar Gignac,G.E.,andWatkins,M.W.(2013).Bifactormodellingandtheestimationofmodel-basedreliabilityintheWAIS-IV.Multiv.Behav.Res.48,639–662.doi:10.1080/00273171.2013.804398 PubMedAbstract|CrossRefFullText|GoogleScholar Glozah,F.N.,andPevalin,D.J.(2015).Factorstructureandpsychometricpropertiesofthegeneralhealthquestionnaire(GHQ-12)amongGhanaianadolescents.J.ChildAdolesc.Ment.Health27,53–57.doi:10.2989/17280583.2015.1007867 PubMedAbstract|CrossRefFullText|GoogleScholar Goldberg,D.P.,andWilliams,P.(1988).AUsers’GuideToTheGeneralHealthQuestionnaire.London:GLAssessment. GoogleScholar Gouveia,V.V.,deLima,T.J.S.,Gouveia,R.S.V.,Freires,L.A.,andBarbosa,L.H.G.M.(2012).QuestionáriodeSaúdeGeral(QSG-12):oefeitodeitensnegativosemsuaestruturafactorial[Generalhealthquestionnaire(GHQ-12):theeffectofnegativeitemsinitsfactorialstructure].CadernosSaúdePública28,375–384.doi:10.1590/S0102-311X2012000200016 PubMedAbstract|CrossRefFullText|GoogleScholar Graetz,B.(1991).Multidimensionalpropertiesofthegeneralhealthquestionnaire.Soc.PsychiatryPsychiatr.Epidemiol.26,132–138.doi:10.1007/BF00782952 PubMedAbstract|CrossRefFullText|GoogleScholar Guan,M.(2017).MeasuringtheeffectsofsocioeconomicfactorsonmentalhealthamongmigrantsinurbanChina:amultipleindicatorsmultiplecausesmodel.Intern.J.Ment.HealthSyst.11:10.doi:10.1186/s13033-016-0118-y PubMedAbstract|CrossRefFullText|GoogleScholar Hankins,M.(2008).Thefactorstructureofthetwelveitemgeneralhealthquestionnaire(GHQ-12):theresultofnegativephrasing?Clin.Pract.Epidemiol.Ment.Health4:10.doi:10.1186/1745-0179-4-10 PubMedAbstract|CrossRefFullText|GoogleScholar Hu,L.,andBentler,P.M.(1999).Cutoffcriteriaforfitindexesincovariancestructureanalysis:conventionalcriteriaversusnewalternatives.Struct.Equ.Model.6,1–55.doi:10.1080/10705519909540118 CrossRefFullText|GoogleScholar Hu,Y.,Stewart-Brown,S.,Twigg,L.,andWeich,S.(2007).Canthe12-itemgeneralhealthquestionnairebeusedtomeasurepositivementalhealth?Psychol.Med.37,1005–1013.doi:10.1017/s0033291707009993 PubMedAbstract|CrossRefFullText|GoogleScholar Iheanacho,T.,Obiefune,M.,Ezeanolue,C.O.,Ogedegbe,G.,Nwanyanwu,O.C.,Ehiri,J.E.,etal.(2015).Integratingmentalhealthscreeningintoroutinecommunitymaternalandchildhealthactivity:experiencefrompreventionofmother-to-childHIVtransmission(PTMCT)trialinNigeria.Soc.PsychiatryPsychiatr.Epidemiol.50,489–495.doi:10.1007/s00127-014-0952-7 PubMedAbstract|CrossRefFullText|GoogleScholar Ip,W.Y.,andMartin,C.R.(2006).FactorstructureoftheChineseversionofthe12-itemgeneralhealthquestionnaire(GHQ-12)inpregnancy.J.Reprod.InfantPsychol.24,87–98.doi:10.1080/02646830600643882 CrossRefFullText|GoogleScholar Johnsen,B.H.,Laberg,J.C.,andEid,J.(1998).Copingstrategiesandmentalhealthproblemsinamilitaryunit.Mil.Med.163,599–602.doi:10.1093/milmed/163.9.599 CrossRefFullText|GoogleScholar Ju,H.B.,Jung,D.U.,Kim,S.J.,Kim,H.J.,Park,J.H.,Seo,Y.S.,etal.(2017).Mentalhealthevaluationforelderlyincommunity,pilotstudy(Englishabstract).J.Korea.Geriatr.Psychiatry21,59–66. GoogleScholar Kalliath,T.J.,O’Driscoll,M.P.,andBrough,P.(2004).Aconfirmatoryfactoranalysisofthegeneralhealthquestionnaire-12.StressHealth20,11–20.doi:10.1002/smi.993 CrossRefFullText|GoogleScholar Kline,R.B.(1998).PrinciplesandPracticeOfStructuralEquationModeling.NewYork,NY:Guilford. GoogleScholar Lai,R.K.(2008).IsInventory’sFiscalYearEndEffectCausedBySalesTiming?ATestUsingANaturalExperimentFromGermany.HarvardBusinessSchoolTechnology&OperationsMgt.Unit(ResearchPaperNo.08-086).Amsterdam:Elsevier. GoogleScholar MacCallum,R.C.,Browne,M.W.,andSugawara,H.M.(1996).Poweranalysisanddeterminationofsamplesizeforcovariancestructuremodeling.Psychol.Methods1,130–149.doi:10.1037/1082-989X.1.2.130 CrossRefFullText|GoogleScholar Marsh,H.W.(1996).Positiveandnegativeglobalself-esteem:asubstantivelymeaningfuldistinctionorartifactors?J.Pers.Soc.Psychol.70,810–819.doi:10.1037/0022-3514.70.4.810 PubMedAbstract|CrossRefFullText|GoogleScholar McCarthy,N.(2018).2.77MillionServiceMembersHaveServedon5.4MillionDeploymentsSince9/11.Forbes.Availableonlineat:https://www.forbes.com/sites/niallmccarthy/2018/03/20/2-77-million-service-members-have-served-on-5-4-million-deployments-since-911-infographic/#6de46ad250db(accessedApril20,2020). GoogleScholar McDonald,R.P.,andHo,M.-H.R.(2002).Principlesandpracticeinreportingstructuralequationanalyses.Psychol.Methods7,64–82.doi:10.1037//1082-989X.7.1.64 CrossRefFullText|GoogleScholar MinistryofDefence(2015).DefenceStatistics(Tri-Service).NewDelhi:MinistryofDefence. GoogleScholar Montazeri,A.,Harirchi,A.M.,Shariati,M.,Garmaroudi,G.,Ebadi,M.,andFateh,A.(2003).The12-itemgeneralhealthquestionnaire(GHQ-12):translationandvalidationstudyoftheIranianversion.HealthQual.LifeOutcomes1:66.doi:10.1186/1477-7525-1-66 PubMedAbstract|CrossRefFullText|GoogleScholar Namjoo,S.,Shaghaghi,A.,Sarbaksh,P.,Allahverdipour,H.,andPakpour,A.H.(2017).Psychometricpropertiesofthegeneralhealthquestionnaire(GHQ-12)tobeappliedfortheIranianelderpopulation.AgingMent.Health21,1047–1051.doi:10.1080/13607863.2016.1196337 PubMedAbstract|CrossRefFullText|GoogleScholar Nordmo,M.,Hystad,S.W.,Sanden,S.,andJohnsen,B.H.(2020).Mentalhealthduringnavaldeployment:Theprotectiveroleoffamilysupport.Milit.Med.14:usz436.doi:10.1093/milmed/usz436 PubMedAbstract|CrossRefFullText|GoogleScholar NorwegianArmedForces(n.d.)Veteranhistorier[Veterans’Stories].Availableonlineat:https://forsvaret.no/tjeneste/veteraner/historier(accessedApril20,2020). GoogleScholar Padrón,A.,Galán,I.,Durbán,M.,Gandarillas,A.,andRodríguez-Artalejo,F.(2012).Confirmatoryfactoranalysisofthegeneralhealthquestionnaire(GHQ-12)inSpanishadolescents.Q.LifeRes.21,1291–1298.doi:10.1007/s11136-011-0038-x PubMedAbstract|CrossRefFullText|GoogleScholar Pallesen,S.,Bjorvatn,B.,Nordhus,I.H.,Sivertsen,B.,Hjørnevik,M.,andMorin,C.M.(2008).Anewscaleformeasuringinsomnia:theBergeninsomniascale.Percept.Mot.Skills107,691–706.doi:10.2466/pms.107.3.691-706 PubMedAbstract|CrossRefFullText|GoogleScholar Picardi,A.,Abeni,D.,andPasquini,P.(2001).Assessingpsychologicaldistressinpatientswithskindiseases:reliability,validityandfactorstructureoftheGHQ-12.J.Eur.Acad.Dermatol.Venereol.15,410–417.doi:10.1046/j.1468-3083.2001.00336.x PubMedAbstract|CrossRefFullText|GoogleScholar Politi,P.L.,Piccinelli,M.,andWilkinson,G.(1994).Reliability,validityandfactorstructureofthe12-itemgeneralhealthquestionnaireamongyoungmalesinItaly.ActaPsychiatr.Scand.90,432–437.doi:10.1111/j.1600-0447.1994.tb01620.x PubMedAbstract|CrossRefFullText|GoogleScholar Reise,S.P.,Bonifay,W.E.,andHaviland,M.G.(2013a).Scoringandmodelingpsychologicalmeasuresinthepresenceofmultidimensionality.J.Pers.Assess.95,129–140.doi:10.1080/00223891.2012.725437 PubMedAbstract|CrossRefFullText|GoogleScholar Reise,S.P.,Scheines,R.,Widaman,K.F.,andHaviland,M.G.(2013b).Multidimensionalityandstructuralcoefficientbiasinstructuralequationmodeling:abifactorperspective.Educ.Psychol.Measur.73,5–26.doi:10.1177/0013164412449831 CrossRefFullText|GoogleScholar Reise,S.P.,Morizot,J.,andHays,R.D.(2007).Theroleofthebifactormodelinresolvingdimensionalityissuesinhealthoutcomesmeasures.Q.LifeRes.16,19–31.doi:10.1007/s11136-007-9183-7 PubMedAbstract|CrossRefFullText|GoogleScholar Rey,J.J.,Abad,F.J.,Barrada,J.R.,Garrido,L.E.,andPonsoda,V.(2014).TheimpactofambiguousresponsecategoriesonthefactorstructureoftheGHQ–12.Psychol.Assess.26,1021–1030.doi:10.1037/a0036468 PubMedAbstract|CrossRefFullText|GoogleScholar Rodriguez,A.,Reuise,S.P.,andHaviland,M.G.(2016).Applyingbifactorstatisticalindicesintheevaluationofpsychologicalmeasures.J.Pers.Assess.98,223–237.doi:10.1080/00223891.2015.1089249 PubMedAbstract|CrossRefFullText|GoogleScholar Romppel,M.,Brachler,E.,Roth,M.,andGlaener,H.(2013).Whatisthegeneralhealthquestionnaire-12assessing?dimensionalityandpsychometricpropertiesofthegeneralhealthquestionnaire-12inalargescaleGermanpopulationsample.Compr.Psychiatry54,406–413.doi:10.1016/j.comppsych.2012.10.010 PubMedAbstract|CrossRefFullText|GoogleScholar Rona,R.J.,Hyams,K.C.,andWessely,S.(2005).Screeningforpsychologicalillnessinmilitarypersonnel.J.Am.Med.Assoc.293,1257–1260. GoogleScholar Salama-Younes,M.,Montazeri,A.,Ismaïl,A.,andRoncin,C.(2009).Factorstructureandinternalconsistencyofthe12-itemgeneralhealthquestionnaire(GHQ-12)andthesubjectivevitalityscale(VS),andtherelationshipbetweenthem:astudyfromFrance.HealthQ.LifeOutcom.7:22.doi:10.1186/1477-7525-7-22 PubMedAbstract|CrossRefFullText|GoogleScholar Sanden,S.,Johnsen,B.H.,Eid,J.,Sommerfelt-Pettersen,J.,Koefoed,V.,Størkesen,R.,etal.(2014).Mentalreadinessformaritimeinternationaloperation:proceduresdevelopedbyNorwegiannavy.Int.Marit.Health65,93–97.doi:10.5603/IMH.2014.0020 PubMedAbstract|CrossRefFullText|GoogleScholar Sarková,M.,Nagyová,I.,Katreniaková,Z.,MadarasováGecková,A.,Orosová,O.,Middel,B.,etal.(2006).Psychometricevaluationofthegeneralhealthquiestionnaire-12androsenbergself-esteemscaleinhungarianandslovakearlyadolescents.Stud.Psychol.48,69–79. GoogleScholar Shevlin,M.,andAdamson,G.(2005).AlternativefactormodelsandfactorialinvarianceoftheGHQ-12:alargesampleanalysisusingconfirmatoryfactoranalysis.Psychol.Assess.17,231–236.doi:10.1037/1040-3590.17.2.231 PubMedAbstract|CrossRefFullText|GoogleScholar Skogen,J.C.,Øverland,S.,Smith,O.R.F.,andAarø,L.E.(2017).ThefactorstructureoftheHopkinsSymptomsChecklist(HSCL-25)inastudentpopulation:acautionarytale.Scand.J.PublicHealth45,357–365.doi:10.1177/1403494817700287 PubMedAbstract|CrossRefFullText|GoogleScholar Smith,G.T.,McCarthy,D.M.,andAnderson,K.G.(2000).Onthesinsofshort-formdevelopment.Psychol.Assess.12,102–111.doi:10.1037/1040-3590.12.1.102 PubMedAbstract|CrossRefFullText|GoogleScholar Suzuki,H.,Kaneita,Y.,Osaki,Y.,Minowa,M.,Kanda,H.,Suzuki,K.,etal.(2011).Clarificationofthefactorstructureofthe12-itemgeneralhealthquestionnaireamongJapaneseadolescentsandassociatedsleepstatus.PsychiatryRes.188,138–146.doi:10.1016/j.psychres.2010.10.025 PubMedAbstract|CrossRefFullText|GoogleScholar Taylor,M.P.(2006).TellmewhyIdon’tlikeMondays:investigatingdayoftheweekeffectsonjobsatisfactionandpsychologicalwell-being.J.R.Stat.Soc.169,127–142.doi:10.1111/j.1467-985x.2005.00376.x CrossRefFullText|GoogleScholar Tomás,J.M.,Gutiérrez,M.,andSancho,P.(2017).Factorialvalidityofthegeneralhealthquestionnaire12inanAngolansample.Eur.J.Psychol.Assess.33,116–122.doi:10.1027/1015-5759/a000278 CrossRefFullText|GoogleScholar Tseliou,F.,Donnely,M.,andO’Reilly,D.(2018).Screeningforpsychiatricmorbidityinthepopulation–acomparisonoftheGHQ-12andself-reportedmedicationuse.Intern.J.Popul.DataSci.3:5.doi:10.23889/ijpds.v3i1.414 CrossRefFullText|GoogleScholar Wong,K.C.K.,andO’Driscoll,M.P.(2016).Psychometricpropertiesofthegeneralhealthquestionnaire-12inasampleofHongKongemployees.Psychol.HealthMed.21,975–980.doi:10.1080/13548506.2016.1140901 PubMedAbstract|CrossRefFullText|GoogleScholar Wright,K.M.,Huffman,A.H.,Adler,A.B.,andCastro,C.A.(2002).Psychologicalscreeningprogramoverview.Milit.Med.167,853–861.doi:10.1093/milmed/167.10.853 CrossRefFullText|GoogleScholar Ye,S.(2009).Factorstructureofthegeneralhealthquestionnaire(GHQ-12):theroleofwordingeffects.Pers.Indiv.Differ.46,197–201.doi:10.1016/j.paid.2008.09.027 CrossRefFullText|GoogleScholar Keywords:GeneralHealthQuestionnaire(GHQ-12),confirmatoryfactoranalysis(CFA),bifactormodels,wordingeffects,factorstructure,measurementinvariance,military Citation:HystadSWandJohnsenBH(2020)TheDimensionalityofthe12-ItemGeneralHealthQuestionnaire(GHQ-12):ComparisonsofFactorStructuresandInvarianceAcrossSamplesandTime.Front.Psychol.11:1300.doi:10.3389/fpsyg.2020.01300 Received:04March2020;Accepted:18May2020;Published:11June2020. Editedby: ClaudioBarbaranelli,SapienzaUniversityofRome,Italy Reviewedby: KostasA.Papageorgiou,Queen’sUniversityBelfast,UnitedKingdom DonatellaRitaPetretto,UniversityofCagliari,Italy Copyright©2020HystadandJohnsen.Thisisanopen-accessarticledistributedunderthetermsoftheCreativeCommonsAttributionLicense(CCBY).Theuse,distributionorreproductioninotherforumsispermitted,providedtheoriginalauthor(s)andthecopyrightowner(s)arecreditedandthattheoriginalpublicationinthisjournaliscited,inaccordancewithacceptedacademicpractice.Nouse,distributionorreproductionispermittedwhichdoesnotcomplywiththeseterms. *Correspondence:SigurdW.Hystad,[email protected] Peoplealsolookedat Download