Psychometric properties of the short form of the Physical Self-Description Questionnaire in a French adolescent sample

ContentslistsavailableatScienceDirect

Body Image

j o u r n al ho me p ag e :w w w . e l s e v i e r . c o m / l o c a te / b o d y i m a g e

Psychometric properties of the short form of the Physical Self-Description Questionnaire in a French adolescent sample

Christophe Maïano

, Alexandre J.S. Morin

, Nicolas Mascret

aCyberpsychologyLaboratory,DepartmentofPsychoeducationandPsychology,UniversitéduQuébecenOutaouais(UQO),Saint-Jérôme,Canada

bInstituteforPositivePsychologyandEducation,AustralianCatholicUniversity,Strathfield,Australia

cAixMarseilleUniversité,CNRS,ISMUMR7287,13288Marseille,France

a r t i c l e i n f o

Articlehistory:

Received11June2014

Receivedinrevisedform21October2014 Accepted22October2014

Keywords:

Sex Age Sportpractice

PhysicalSelf-DescriptionQuestionnaire Convergentvalidity

Measurementinvariance

a b s t r a c t

Recently,Marsh,Martin,andJackson(2010)developedashortformofthePhysicalSelf-Description Questionnaire(PSDQ-S).Theobjectiveofthisstudywastoexaminetheconstructvalidityandreliability ofthePSDQ-SinaFrenchadolescentsample.Thesampleusedinthisstudyincluded587adolescents(247 boys,340girls,Mage=14.62).Confirmatoryfactoranalyses(CFA)providedsupportforthefactorvalidity, reliability,andconvergentvalidityoftheFrenchversionofthePSDQ-S,andthestrictmeasurement invarianceofPSDQ-Sacrosssex,age,bodymassindex,andinvolvementornotinsportpractice.However, thelatentmeansofthePSDQ-Sdidnotprovetobeinvariantacrosssex,bodymassindex,andinvolvement ornotinsportpractice.OurfindingssuggestthattheFrenchversionofthePSDQ-Spresentsacceptable psychometricpropertiesandmaybeconfidentlyusedinresearchorpracticetoassessthephysicalself- conceptionsofFrenchadolescents.

©2014ElsevierLtd.Allrightsreserved.

Introduction

Intheirseminalwork,FoxandCorbin(1989)adaptedShavelson, Hubner,andStanton’s(1976)multidimensionalandhierarchical self-conceptmodeltothephysicalself-conceptarea.Inthismodel, thehigherlevelisoccupiedbytheglobalself-concept(i.e.,over- allpositiveornegativeperceptionand/orassessmentofoneself).

The intermediate “domain” level is occupied by a global con- structrepresentingtheglobalphysicalself-concept(i.e.,positive ornegativeperceptionand/orassessmentofoneselfintheglobal physicalarea).Finally,thelower“subdomain”levelisoccupiedby constructsrepresentingmorespecificcomponentsofphysicalself- conceptions,suchassportcompetence,physicalcondition,physical strength,flexibility,coordination,andphysicalappearance.

Tooperationalizethismodel,Marsh,Richards,Johnson,Roche, and Tremayne (1994) developed and validated the Physical Self-Description Questionnaire (PSDQ). The PSDQ was initially developedforadolescentsandincludedatotalof70itemsassessing

∗Correspondingauthorat:UniversitéduQuébecenOutaouais,CampusdeSaint- Jérôme,DépartmentdePsychoéducationetdePsychologie,5rueSt-Joseph,Saint- Jérôme,QuébecJ7Z0B7,Canada.

E-mailaddress:christophe.maiano@uqo.ca(C.Maïano).

1 Thesetwoauthorscontributedequallytothisarticleandtheirorderwasdeter- minedatrandom:bothshouldbeconsideredfirstauthors.

11dimensions:activity,appearance,bodyfat,endurance,coordina- tion,flexibility,health,sportcompetence,strength,globalphysical self-concept,andglobalself-esteem.Confirmatoryfactoranalyses (CFA)conductedonasampleof710Australianhighschoolstu- dentssupportedthefactorialvalidityandmeasurementinvariance ofthePSDQacrossmaleandfemalestudents,aswellasthecon- vergentanddiscriminantvalidityofthePSDQwithotherphysical self-conceptsinstruments(Marshetal.,1994).Subsequentanal- yses demonstrated thatthesubscales showedsatisfactoryscale scorereliabilitycoefficientsrangingfrom˛=.82to.96acrosssam- ples,andhadsex-basedmean-leveldifferences,withboystending tohavehigherscoresonmostPSDQsubscalescomparedtogirls (Marshet al.,1994).Additionalstudiesconductedamong three Australianadolescentsamplesrevealed:(a)goodconvergentand discriminantvalidityofthePSDQwithexternalcriteria(e.g.,body composition, physical fitness tests; Marsh & Redmayne, 1994;

Marsh,1996b),and;(b)goodtest-reteststabilityofthe11scales (withtest-retestcorrelationsrangingfromr=.70to.89overa3- monthperiod,andbetweenr=.31and.82overa14-monthperiod;

Marsh,1996a).

Although other instruments are currently available for the assessmentofphysicalself-conceptionsamongyoungpersons,the PSDQremainsbyfarthemostcomprehensive(covering11dimen- sionsversus6–7fortheotherinstruments)andthemostwidely validatedinstrumentavailabletodate(forreviewsseeMarsh&

Cheng,2012; Sypsa& Simons,2008).Indeed,the psychometric http://dx.doi.org/10.1016/j.bodyim.2014.10.005

1740-1445/©2014ElsevierLtd.Allrightsreserved.

propertiesofthePSDQ(i.e.,factorvalidityandreliability,conver- gentanddiscriminantvalidity)havebeenextensivelydocumented (forameta-analysisseeSchipke&Freund,2012)across:(a)lan- guageversions(French,Dutch,German,Hebrew,Italian,Spanish, Turkish,etc.),(b)agegroups(children,adolescents,adults,elderly), (c) type of sample (normal, psychiatric, athletes,bodily handi- capped, etc.),and (d)context (research, diagnostic,counseling, etc.).

Amongitsfewdownsides,thePSDQismuchlonger(70items versus12–35 items)thanalternatives,especially when used in conjunctionwithmultipleotherinstrumentsinthecontextofcom- prehensivestudies(Marsh,Martin,etal.,2010).Thislimitationled Marsh,Martin,etal.(2010)todevelopandvalidatea40-itemshort formofthePSDQ(PSDQ-S),balancingbrevityandpsychometric strength.ThePSDQ-Swasdevelopedandvalidatedusinganormat- ivearchivalsample(N=1605)ofAustralianhighschoolstudents (12–18yearsold)andwascross-validatedamongfiveadditional samplesofAustralianstudents(N=708),Australianeliteadolescent athletes(N=349),Spanishadolescents(N=986),Israeliuniversity students(N=395),andAustralianolderadults(N=760).ThePSDQ- ScoversallsubscalesincludedintheoriginalPSDQ,eachofwhichis assessedusingeitherthree(Appearance,bodyfat,endurance,flex- ibility,globalphysicalself-concept,sportcompetence,strength), four(Activity)orfive(Coordination,globalself-esteem,andhealth) items.AseriesofCFAsprovidedsupportforthe(a)factorialvalid- ityand measurement invarianceof thePSDQ-Sacrosssamples, sex-groups,age-groups,andversions(PSDQversusPSDQ-S);(b) convergentanddiscriminantvalidityofthePSDQ-Samongsamples ofAustralianadolescents(N=322)andIsraeliuniversitystudents (N=395);and(c)sex-andage-baseddifferences,showingthatboys andadolescentstendtopresenthigherscoresonmostPSDQsub- scaleswhen comparedwithgirlsand olderadults,respectively.

Subsequentanalysesdemonstratedsatisfactoryscalescorereliabil- ity(˛=.77–.94,acrossthesixsamples)andtest-retestreliability (r=.57–.90over1-yearamong212Australianadolescentsand553 olderadults,respectively)ofthe11subscales.

AdaptationstoOtherLanguages

Todate,severalstudieshaveprovidedtentativesupporttothe psychometricpropertiesof variouslinguisticadaptations ofthe PSDQ-S.For instance, Agarwal,Bhalla, Kaur, and Babbar(2013) adaptedthePSDQ-SforIndianuniversitystudentsandreported acceptable levels of scale score reliability for the various sub- scales (˛ rangingfrom .77 to .97). Likewise, starting from the GermanadaptationofthefullPSDQ(Stiller&Alfermann,2007), JonesandStumbrys(2014)createdashortformusingthesame itemsas Marsh, Martin, et al.(2010), and administered it to a sampleof72universitystudents.Althoughtheseauthorsdonot reportinformationaboutthescalescorereliabilityorfactorvalid- ityof thePSDQ-S,theirresultsrevealedsignificantassociations betweenthesesubscales,psychologicalwell-being,and thefre- quencyofluciddreaming.Inamoreextensivestudy,Papaioannou etal.,(2013;alsoseeDudaetal.,2013)administeredtheglobal self-esteem subscale of the PSDQ-S to a total of 7789 early adolescent soccer players from France, Greece, Norway, Spain, and England. Across countries/languages, theirresults revealed modest-to-acceptablescalescorereliabilityestimatesforthissub- scale(˛rangingfrom.51to.73)andsignificantrelationshipswith subjectivevitalityandmoderatevigorousphysicalactivity.Sim- ilarly,Castonguay,Sabiston,Crocker,andMack(2014)usedtwo subscales(i.e., appearanceand body fat)of the PSDQ-S among a mixedsample of 435English-and French-speaking Canadian adults.Theirresultsrevealthatbothsubscaleshadacceptablescale scorereliability(˛=.86and.92)andweresignificantlyrelatedto

subscalesoftheBodyandAppearance-relatedself-consciousEmo- tionsScale.

Althoughpromising,thesestudiesremainpreliminary,asno studyhasyetsystematicallyexaminedthepsychometricproperties oftheselinguisticadaptations.Thus,apartfrominitialattemptsby Marsh,Martin,etal.(2010)tovalidatethePSDQ-Samongsamples ofSpanishandIsraelistudents,thecross-culturalgeneralizability ofthePSDQ-Samongsamplesofnon-Englishadolescentsremains anopenquestionthatmustbeaddressedbeforethePSDQ-Scan beconfidentlyusedinthecontext ofcross-culturalcomparison studies.Inthisstudy,wefocusontheFrenchversionofthePSDQ-S.

Currently,twoself-reportquestionnairesareavailabletomea- sure the physical self-concept among French-speaking young persons:thelongformofthePSDQ(Guérin,Marsh,&Famose,2004) andtheshortandveryshortformsofthePhysicalSelf-Inventory (PSI-S&PSI-VS;Maïanoetal.,2008;Morin&Maïano,2011a,b).

DespitethenumerousadvantagesofthePSIscales(seeMarsh&

Cheng,2012),theirreducedlengthalsomakesthemlesscompre- hensivethanthePSDQ.Inturn,thelengthofthePSDQmakesit unwieldyinmanyresearchcontextswherethenumberofitems needstobelimitedforpracticalreasons.Consequently,thevali- dationofaFrenchversionofthePSDQ-Swillfillanimportantgap inproviding areasonablyshort,yetcomprehensive,measureof thephysicalself-conceptfortheFrench-speakingresearchcom- munity.However,theusefulnessofsuchaFrenchversiongoeswell beyondtheFrenchcommunityinprovidingavaluableinstrument on which to anchor cross-cultural comparison studies (Tomás, Marsh,González-Romá,Valls,&Nagengast,2014).Inthisregard, theavailabilityofavalidatedFrenchversionisparticularlyimpor- tantgiventhatFrenchistheofficialorco-officiallanguagein29 countriesandterritoriesworldwideandoneofthemostcommonly usedlanguageinNorthAfrica.

MeasurementInvarianceandLatentMeansDifferences Acritical issuein theassessmentof thepsychometricprop- ertiesofanymeasurementinstrumentiswhetheritcanbeused withindividualscomingfromdifferentsegmentsofthepopulation andwhethercomparisonsconductedacrossthesesubpopulations willbemeaningfulorreflectmeasurementbiases.Measurement biasesoccurwhenaninstrumentbehavesdifferentlyacrossdis- tinctsubgroupsfromthepopulationandleadstotheimpossibility ofcomparingscoresobtainedontheinstrumentacrossthesedis- tinct subgroups. In practice,this verificationis conductedvia a sequenceoftestsofmeasurementinvariance(e.g.,Meredith,1993;

Millsap,2011)whereequalityconstraintsareprogressivelyadded todifferentparametersfromameasurementmodel(i.e.,loadings, intercepts,and uniqueness)acrosssubgroups ofparticipantsto systematicallytestwhethertheseconstraintholdinpractice.The non-invarianceofthefactorloadingssuggeststhattheinstrument doesnotmeasurethesameconstructsacrosssubgroups,andpre- cludesanyformofgroup-basedcomparison.Thenon-invariance ofitems’interceptsrather suggeststhatparticipantspresenting thesame true score onthe constructof interest (e.g.,physical self-conceptions)willstilltend toscore higherorlower onthe measurementscalesas aresult oftheirmembershipinspecific subgroups.Evidenceofinvarianceofthefactorloadingsanditem intercept is animportant pre-requisiteto validategroup-based mean-levelcomparisons.Thenon-invarianceoftheitems’unique- nesses finallysuggests thatthe measurement errors differas a functionofgroupmembership,and thusthattheconstructsare assessedwithdifferentlevelsofprecision.

Although Marsh, Martin, et al. (2010) reported evidence of measurementinvarianceacrosssubsamplesofboysandgirls,ado- lescents and older adults, and typical adolescents versus elite athletes,theydidnotexaminethemeasurementinvarianceofthe

PSDQ-Sacrosssamplesofearlyandlateadolescents,adolescents involvedornotinsportpracticeatamorenormativelevel,²andas afunctionofbodymassindex(BMI)levels.Furthermore,itremains necessarytoverifywhetherevidenceofsex-basedmeasurement invariancewillgeneralizetotheFrenchPSDQ-S.Theseobserva- tionsareworrisomeasthePSDQ–andbyextensionthePSDQ-S– isfrequentlyusedtocomparephysicalself-conceptionsasafunc- tionofBMIlevelsandsubgroupsofboysandgirls,earlyandlate adolescents,andyouthinvolvedornotinsportpractice.

Pendingevidenceofmeasurementinvariance,itthenbecomes possible to conduct more advanced tests of the invariance of thelatentvariances,covariances,and meansacrossthesesame subgroupsofparticipants,providingadirecttestofthediscrim- inantvalidityoftheFrenchPSDQ-S.Indeed,ifthepsychometric properties of the original PSDQ and PSDQ-S are maintained inthe Frenchversion,then group-basedmean-level differences obtainedwiththeFrenchPSDQ-Sshouldreplicatethoseobserved in previous investigations of physical self-conceptions among adolescent populations. Generally, previous studies of physical self-conceptionsconductedamong samplesof adolescentshave revealedthat(e.g.,Bowker,2006;Findlay&Bowker,2007;Marsh, 1998;Marsh,Hau,Sung,&Yu,2007;Marsh,Hey,Roche,&Perry, 1997;Schmalz&Davison,2006):(a)boystendtohavehigherlev- elsofphysicalself-conceptionsthangirlsacrossmostdimensions ofthephysicalself-conceptthatwereconsidered;(b)earlyado- lescentstend tohave higher levelsofphysical self-conceptions (acrossmostdimensions)thanmiddle,and lateadolescents;(c) adolescentswiththehighestBMIstendtohavethelowerlevels ofphysicalself-conceptions(exceptforthestrengthsubscaleon whichtheypresentedhigherlevels);and(d)adolescentsathletes tendedtopresenthigherlevelsofphysicalself-conceptions(across mostdimensions)thannonathletes.

Therefore,themainobjectiveofthepresentstudywastoexam- inethepsychometricpropertiesoftheFrenchPSDQ-Samong a sampleofadolescents.Morespecifically,thisstudyexaminedthe:

(a)factorialvalidityandscalescorereliabilityofthePSDQ-Samong asampleofFrenchadolescents;(b)measurementandlatentmean invarianceofthePSDQ-Sacrosssex-groups(boysversusgirls),age- groups(earlyversuslateadolescence),sportpracticeinvolvement (i.e.,involvedversusnotinvolved),andBMI;and(c)convergent validityof thePSDQ-SwiththePSI-S, anditsinvariance across sex-groups,age-groups,andsportpracticeinvolvement.

Method SampleandProcedures

A sample of 587 adolescents (agerange=11–18 years;

Mage=14.62 years, SD=1.80) attending seven middle and high schoolslocatedinSouthernFranceformthesamplewasusedin thisstudy.Ofthose:(a)247wereboys(42.1%)and340weregirls (57.9%);(b) 262(44.6%)wereearlyadolescents(agerange:11–14 years) and 340 (55.4%) were late adolescents (agerange: 15–18 years);and (c)218(37.1%) wereinvolvedinphysicaleducation (PE) and didnot practicesportoutside school and 369(62.9%) were involved in PE and practiced sport outside school. Only adolescents who returned consentforms signed bythemselves andtheirparentswereincludedinthestudy.Alladolescentsthat wereincludedinthisstudycompletedthequestionnairesduring PE classes. This project was approved by the French Advisory CommitteeonInformationProcessinginMaterialResearchinthe

2We use“sport practice” to refer to practicing sports and fitness activi- ties/exercises(fitness,aerobics,steps,weighttraining,running,etc.)forrecreational orcompetitivepurposes.

FieldofHealthandbytheChiefEducationOfficeroftheAcadémie ofAix-Marseille.

Measures

Demographics. Participants completed a questionnaire in whichtheywereaskedtoreporttheirsex,age,height,weight,and involvementinsportpracticeoutsideoftheschoolcontext(i.e.,“Do youpracticeasportoutsideofyourphysicaleducationclassesand oftheschoolsportassociation?”).Self-reportedheightandweight wereusedtocomputeparticipantBMI(weight/height²).

Physical self-conceptions. Multidimensional physical self- conceptionsweremeasuredusingaFrenchversionofthePSDQ-S andofthePSI-S(Maïanoetal.,2008;Morin&Maïano,2011a).The FrenchversionofthePSDQwasdevelopedandvalidatedbyGuérin etal.(2004)amongasampleof1040Frenchadolescents.Intwo studies,theseauthorsprovidedsupportforthe11-factorstructure, reliability(i.e.,scalescorereliabilityandtest-reteststability),and convergentvalidity(with13physicalfitnesscriteria)ofthePSDQ.

ThisFrenchversionincludes70itemsthatareratedona6-point scalerangingfrom1(False)to6(True).Tobuildtheshortversion usedinthecurrentstudy,weretainedfromthese70 itemsthe same40itemsselectedbyMarsh,Martin,etal.(2010)todevelop theEnglishPSDQ-S.

ThePSI-SisaFrenchadaptationofthePSPP(Fox&Corbin,1989), whichhasbeenextensivelyvalidatedandcross-validatedwitha sampleof3047Frenchadolescentsintwostudies,whichsupported thefactorstructure, scalescorereliability, test-retestreliability, andconvergentvalidityofthePSI-S(Maïanoetal.,2008;Morin&

Maïano,2011a).ThePSI-Sincludes18itemsratedona6-pointscale rangingfrom1(Notatall)to6(Entirely)andassessing6dimensions (global self-worth,physicalself-worth,physicalcondition,sport competence,physicalattractiveness,andphysicalstrength)ofthe physicalself-concept.

Analyses

The analyses were conducted with Mplus 7.11 (Muthén &

Muthén,2013).Becauseofsignificantmultivariatenon-normality ofthedata(Mardia’snormalizedestimate=348.20),analyseswere performedusingtherobustmaximumlikelihoodestimator(MLR).

In addition,full-informationestimation wasusedtocorrect for the small amounts of missing data present at the item level (Missing_range=3.24–6.47%;M_missing=4.43%;Enders,2010).Inthe firststageoftheanalyses,theapriori11-factormodelofthePSDQ- SwasexaminedusingCFA(Model1-1inTable1),hypothesizing that:(a)youths’responsestothePSDQ-Swouldbeexplainedby 11correlatedfactors;(b)eachitemwouldhaveanon-zeroload- ingonthefactor itwasdesignedtomeasureandzeroloadings onotherfactors;and (c)uniquenesseswouldnotbecorrelated.

Moreover,becauseone-fourthoftheitemsofthePSDQ-Sarenega- tivelyworded(i.e.,fiveitemsfromthehealthsubscale,threeitems fromthebody fatsubscale,andtwoitemsfromtheglobalself- esteemsubscale),anadditionalmodelwasexaminedinorderto controlforthemethodologicalartifactintroducedbythewording oftheseitems.Indeed,evidenceisrapidlyaccumulatingthatneg- ativelywordeditemstendtobeaccompaniedbymethodological artifactsthatneedtobeexplicitlytakenintoaccountinstatistical models(e.g.,DiStefano&Motl,2006;Marsh,Scalas,&Nagengast, 2010) in order to “purge” the estimated latent factors from

“contaminatingmethodeffects”(Marsh,Scalas,etal.,2010,p.369) due toconstruct-irrelevantvariability (seealsoMorin, Arens,&

Marsh,2014).Thismodel(Model1-2)usedanadditionalortho- gonallatentmethodfactor(LMF)associatedwiththenegatively wordeditems.Itshouldbenoted,however,thatover-and-above

Table1

Goodness-of-fitstatisticsofconfirmatoryfactoranalytic(CFA)models.

Model Description ²(df) CFI TLI RMSEA 90%CI CM R²(df) CFI TLI RMSEA

CFA 1-1.CFA11-factor 1614.43(685)^** .920 .908 .048 .045–.051 – – – – –

1-2.CFA-LMF11-factor 1411.96(675)^** .936 .926 .043 .040–.046 1-1. 131.57(10)^** +.016 +.018 −.005 1-3.CFA-PSDQ-S&PSI-S 2909.64(1449)^** .924 .913 .041 .039–.044 –

CFA:

Age

2-1.Configuralinvariance 2330.95(1350)^** .919 .907 .050 .046–.053 – – – – –

2-2.sinvariant 2340.92(1388)^** .922 .912 .048 .045–.052 2-1. 27.18(38) +.003 +.005 −.002 2-3.s,sinvariant 2357.87(1416)^** .923 .915 .048 .044–.051 2-2. 13.05(28) +.001 +.003 .000 2-4.s,s,ısinvariant 2379.54(1456)^** .924 .919 .046 .043–.050 2-3. 38.46(40) +.001 +.004 −.002 2-5.s,s,ıs,s/ϕsinvariant 2455.98(1523)^** .923 .921 .046 .042–.049 2-4. 77.53(67) −.001 +.002 .000 2-6.s,s,ıs,s/ϕs, sinvariant 2480.78(1535)^** .922 .921 .046 .042–.049 2-5. 26.01(12)^* −.001 .000 .000 CFA:

Sex

3-1.Configuralinvariance 2229.79(1350)^** .922 .910 .047 .044–.051 – – – – –

3-2.sinvariant 2316.66(1388)^** .917 .907 .048 .044–.051 3-1. 86.37(38)^** −.005 −.003 +.001 3-3.s,sinvariant 2402.56(1416)^** .912 .903 .049 .045–.052 3-2. 85.34(28)^** −.005 −.004 +.001 3-4.s,s,ısinvariant 2458.93(1456)^** .911 .905 .048 .045–.052 3-3. 60.78(40)^* −.001 +.002 −.001 3-5.s,s,ıs,s/ϕsinvariant 2662.27(1523)^** .899 .896 .050 .047–.054 3-4. 195.70(67)^** −.012 −.009 +.002 3-6.s,s,ıs,s/ϕs(^HE,^AC)invariant 2626.72(1521)^** .902 .899 .050 .047–.053 3-4. 161.70(65)^** −.009 −.006 +.002 3-7.s,s,ıs,s/ϕs(^HE,^AC), sinvariant 2825.28(1533)^** .885 .883 .054 .050–.057 3-6. 283.68(12)^** −.017 −.016 +.004 CFA:

Sport

4-1.Configuralinvariance 2276.09(1350)^** .919 .907 .048 .045–.052 – – – – –

4-2.sinvariant 2327.70(1388)^** .918 .908 .048 .045–.051 4-1. 52.46(38) −.001 +.001 .000 4-3.s,sinvariant 2369.78(1416)^** .917 .909 .048 .045–.051 4-2. 41.62(28) −.001 +.001 .000 4-4.s,s,ısinvariant 2395.60(1456)^** .918 .912 .047 .044–.050 4-3. 40.88(40) +.001 +.001 −.001 4-5.s,s,ıs,s/ϕsinvariant 2552.44(1523)^** .910 .908 .048 .045–.051 4-4. 157.49(67)^** −.008 −.004 +.001 4-6.s,s,ıs,s/ϕs, sinvariant 2763.97(1535)^** .893 .891 .052 .049–.055 4-5. 226.46(12)^** −.017 −.017 +.004 CFA

including BMI

5-1.MIMICnulleffect 1606.15(715)^** .924 .913 .046 .043–.049 – – – – –

5-2.MIMICsaturated 1409.73(675)^** .938 .924 .043 .040–.046 5-1. 206.55(40)^** +.014 +.011 −.003 5-3.MIMICinvariantintercept 1454.41(704)^** .936 .926 .043 .040–.046 5-2. 43.03(29)^* −.002 +.002 .000 Note.AC=activity;BMI=body-massindex;CFI=comparativefitindex;CM=comparisonmodel;df=degreesoffreedom;HE=health;LMF=latentmethodfactor;RMSEA=root meansquareerrorofapproximation;PSDQ-S=PhysicalSelf-DescriptionQuestionnaire-shortform;PSI-S=PhysicalSelf-Inventory-shortversion;TLI=Tucker–Lewisindex;

90%CI=90%confidenceintervaloftheRMSEA;=loading;=intercept;ı=uniquenesses;=variance;ϕ=covariance; =factormeans;²=chi-square;R²=robust chi-squaredifferencetests(calculatedfromloglikelihoodsforgreaterprecision);=changefrompreviousmodel.

* p<.05.

** p<01.

theimpactofthisLMFonmodelfit,parameterestimatesremained relativelyunchangedbythisinclusioninthepresentstudy.

Themeasurement invarianceand latentmeandifferences of thePSDQ-Swasthenexaminedacrossage-groups(11–14years versus 15–18 years; see models 2-1 to 2-6), sex-groups (girls versusboys;seemodels3-1to3-7),andsportpracticeinvolve- ment (involved versus not involved; see models 4-1 to 4-6).

Theseinvariancetestswereperformedinthefollowingsequence (Meredith, 1993; Millsap, 2011): (a) configural invariance, (b) weak(loadings)invariance,(c)strong(loadings,intercepts)invari- ance,(d)strict(loading,intercepts,uniquenesses)invariance,(e) variance/covariance(loading,intercepts,uniquenesses,variances, covariances)invariance,and(f)latentmeans(loading,intercepts, uniquenesses,variances,covariances,latentmeans)invariance.

Thesample sizeswithin each of theBMI-derived categories (e.g.,underweight,overweight,obese)weretoosmalltoconduct completetestsofmeasurementinvariancesacrossBMIsubgroups.

Thus, we relied on multiple indicators multiple causesmodels (MIMIC; e.g., Marsh, Ellis, Parada, Richards, & Heubeck, 2005;

Marsh,Nagengast,&Morin,2013;Marsh,Tracey,&Craven,2006;

Morin,Marsh,&Nagengast,2013)toinvestigatetheinvarianceof theinterceptsandlatentmeansofthePSDQ-Smeasurementmodel asafunctionofBMIlevelstreatedasacontinuouspredictor(see models5-1to5-3).TheseMIMICinvariancetestsasafunctionof BMIlevelswereperformedinthefollowingsequence(Marshetal., 2013;Morinetal.,2013):(a)MIMICnulleffectmodel(i.e.,thepaths fromBMItothelatentfactorsanditeminterceptsareconstrainedto bezero);(b)MIMICsaturatedmodel(i.e.,thepathsfromBMItothe iteminterceptsarefreed);and(c)MIMICinvariantinterceptmodel (i.e.,thepathsfromBMItothelatentfactorsarefreed,butthepaths fromBMItotheiteminterceptsareconstrainedtobezero).The goodness-of-fitindicesoftheMIMICnulleffectmodelcanbecom- paredwiththoseoftheMIMICsaturatedandinvariantintercept

modelstoexaminewhethertheBMIhasasignificanteffectonthe itemsinterceptorlatentmeanofthePSDQ-S,respectively(Marsh etal.,2013;Morinetal.,2013).Additionally,thecomparisonofthe MIMICsaturatedandinvariantinterceptmodelsindicateswhether theiteminterceptsofthePSDQ-Sareinvariantornotaccordingto theBMIoftheadolescents(Marshetal.,2013;Morinetal.,2013).

Giventheknownoversensitivityofthechi-squaretestofexact fittosamplesizeandminormodelmisspecifications(e.g.,Marsh, Hau,&Grayson,2005),wereliedoncommonfitindicestodescribe thefit ofthe alternativemodels(e.g.,Hu&Bentler, 1999):the comparativefitindex(CFI),theTucker–Lewisindex(TLI),theroot meansquareerrorofapproximation(RMSEA),andthe90%con- fidenceintervaloftheRMSEA.Valuesgreaterthan.90and.95for theCFIandTLIindicate,respectively,acceptableandgoodmodelfit, whilevaluessmallerthan.06and.08fortheRMSEA,respectively, supportgoodandacceptablemodelfit.Becausechi-squarediffer- encetestspresentproblemssimilartothoseofthechi-squareitself, robust²differencetests(R²;Satorra,2000;Satorra&Bentler, 2001)werealsocomplementedbytheexaminationofchangesinfit indices.AccordingtoCheungandRensvold(2002)andChen(2007) aCFIof.01orlessandaRMSEAof.015orlessbetweenamore restrictedmodelandtheprecedingoneintheinvariancesequence suggestthattheinvariancehypothesisshouldnotberejected.Since indicesincorporatingapenaltyforparsimony(TLIandRMSEA)can alsoimproveinmorerestrictedmodels,TLIsarealsoinspected (Marsh,Hau,etal.,2005).Furthermore,althoughtheefficacyofthe proposedguidelinesforthecomparisonofnestedinvariancemod- elshasbeenvalidatedinCFAfortestsofweak,strong,andstrict measurementinvariance(Chen,2007;Cheung&Rensvold,2002), theyappeartobeofquestionableefficacyfortestsoflatentmean invariance(Fan&Sivo,2009).Inaddition,theseindicesstillappear toshowsensitivitytodesignconditionsandmodelcomplexity(e.g., Fan&Sivo,2005,2007).Overall,theseguidelinesshouldnotbe

treatedasgoldenrules,butonlyasroughguidelinesfordescrip- tivemodelevaluationandcomparisonthatshouldalsotakeinto accountparametersestimates,statisticalconformity,andtheoret- icaladequacy(Fan&Sivo,2009;Marshetal.,2005).

Scale score reliability was computed from the CFA stan- dardized parameter estimates, using McDonald’s (1970) ω=(|i|)²/([|i|]²+ıii) where i are the factor loadings andıii,theerrorvariances,whichhastheadvantageoftakinginto accountthestrengthofassociationbetweeneachitems,andthe latentconstructitispurportedtoassess,aswellasitem-specific measurementerror(Sijtsma,2009).Finally,theconvergentvalidity ofthePSDQ-SsubscaleswiththePSI-Ssubscaleswasalsoexam- inedamongtheoverallsampleusinglatentvariablecorrelations between all subscales from both instruments. In addition, the invarianceoftheselatentvariablecorrelationswasalsoexamined acrossage-andsex-groups,aswellassportpracticeinvolvement.

Thesetestsallstartedfromamodelofconfiguralinvarianceofthe CFAmeasurementmodelsofbothinstruments(i.e.,loading,inter- cepts,uniquenesses,andvariance-covariancesofeachinstrument wereinvariant,while thecovariancesbetweenthePSDQ-S and PSI-S werefreelyestimated). Thegoodness-of-fit of this model wasthencomparedtothefitofa modelwherethecovariances betweenthePSDQ-SandPSI-Swereconstrainedtoinvariance.

Results FactorValidityandReliability

Thegoodness-of-fitstatisticsoftheCFAmodelsarepresented inTable1.Theresultsshowedasatisfactorydegreeoffittothe data(i.e.,CFIand/orTLI>.90;RMSEA<.05)forthetwoCFAmod- els(11-factormodelwithoutandwithaLMF).However,changesin goodness-of-fitindicesrevealedthatthe11-factormodelwithLMF (Model1-2)providedasignificantlybetterleveloffittothedata thanthetraditional11-factormodel(Model1-1).Thestandardized parameterestimatesofthismodelarepresentedinTable2,while thelatentfactorscorrelationsandscalescorereliabilityestimates arereportedinTable3.Theseresultsrevealthatallfactorloadings weresubstantialand significant(M=.75),exceptforonenega- tivelywordeditemfromtheglobalself-esteemsubscale(itemGSE 5“NothingIeverdoseemstoturnoutright”–“Danslavie,riendece quejefaisnesembleaboutir”)thatwasassociatedwithasuboptimal loading(=.153).Theresultsfurthershowedthatmostofthelatent factorcorrelationsweresignificant(rrange=−.15to.80;Mr=.41), exceptforthoseinvolvingthehealthsubscale,whichprovednotto besignificantlycorrelatedwiththeotherPSDQ-Ssubscales.Inaddi- tion,theresultsalsoshowedstrongcorrelations(r>.70)between (a)thecoordinationandflexibilitysubscales,(b)thesportcom- petenceandfourothersubscales(i.e.,activity,endurance,global physicalself-concept,andstrengthsubscales),(c)theglobalphysi- calself-conceptandtwoothersubscales(i.e.,globalself-esteemand appearance),and(d)theenduranceandtwoothersubscales(i.e., activityand strength).Finally,scale scorereliability coefficients wereacceptableforallsubscales(ωrange=.74–.91;M_ω=.86).

MultipleGroupMeasurementInvarianceTests

The measurement invariance of this final, retained model was then verified across age groups (Models 2-1 to 2-6), sex (Models3-1to3-7),andsportpracticeinvolvement(Models4-1to 4-6).Theresultsfromthemeasurementinvariancetestsconducted acrossagegroupsandsportpracticeinvolvementwereverysimilar andshowedthat(a)allofthe²andmostoftheR²weresignif- icant;(b)theCFI,TLIandRMSEAindicatedadequatemodelfitatall steps;(c)theRMSEAnevershowedanincreasegreaterthan.015;

and(d)theCFIandTLInevershowedadecreasegreaterthan .01,exceptforthelatentmeaninvarianceacrosssubgroupsformed basedonsportpracticeinvolvement(Model4-6).Theseresultscon- firmedthefullmeasurementinvarianceofthePSDQ-Sacrossage groupsandsportpracticeinvolvement,butalsohighlightedsignif- icantlatentmeandifferencesbetweenadolescentsinvolvedornot insportpractice.Resultsshowedthat,whenthelatentmeansof adolescentsnotinvolvedinsportpracticewerefixedtozero(to serveasthefirstcomparisongroup) foridentificationpurposes, thelatentmeans(expressedinstandarddeviationunits)ofthose involvedinsportpracticeweresignificantly(p≤.05)higheronthe coordination(0.60),strength (0.60),flexibility(0.30),endurance (0.79),activity(1.57),sportcompetence(0.69),appearance(0.21), globalphysicalself-concept(0.31),andglobalself-esteem(0.34) subscales. No significant differences were foundfor the health (0.13)andbodyfat(0.02)subscales.

Additionally,themeasurementinvariancetests(Models3-1to 3-7)conducted acrosssex groupsshowedthat (a)allof the² andmostoftheR²weresignificant;(b)theCFI,TLIandRMSEA indicatedadequatemodelfitatallsteps(exceptforsteps3-5and 3-7);(c)theRMSEAnevershowedanincreasegreaterthan.015;

and(d)theCFIandTLIremainedbelow.01forthefirstthree steps,confirmingthestrict(i.e.,configural,loadings,intercepts,and uniquenesses)measurementinvarianceofthemodel.However,the CFIwasgreaterthan.01atthefifthstep(Model3-5),suggesting alackofinvarianceforthelatentvariance-covariance.Thislackof invariancewasduetoahigherlevelofvariabilityinthehealthand activitysubscalesamonggirlsrelativetoboys.Whentheinvariance constraintswererelaxedforthevariancesofthesefactors(Model 3-6),theresultssupportedthepartialinvarianceofthevariance- covariancematrix.Finally,theCFIandTLIwerealsogreater than.01forthelaststep(Model3-7),providingevidenceoflatent meansdifferencesacrosssex.Probingthesedifferencesrevealed that,whengirls’latentmeanswerefixedtozero(toserveasthefirst comparisongroup)foridentificationpurposes,boys’latentmeans (expressedinstandarddeviationunits)weresignificantly(p≤.05) higheronthecoordination(0.58),strength(1.13),endurance(1.06), health (0.92),activity (0.68),body fat (0.62),sport competence (0.87),appearance(0.45),globalphysicalself-concept(0.68),and globalself-esteem(0.64)subscales.Nosignificantdifferenceswere foundontheflexibility(0.07)subscale.

MIMICModels

TheresultsfromtheMIMICnulleffectmodel(seeModel5-1 inTable1)providedasatisfactoryfittothedata(CFIandTLI>.90;

RMSEA<.05).Additionally,theMIMICsaturatedmodel(seeModel 5-2 in Table 1)provided a satisfactoryfit tothedata (CFI and TLI>.90;RMSEA<.05).ResultsalsoshowedthattheMIMICsatu- ratedmodelprovidedasubstantiallybetterfittothedatathanthe MIMICnulleffectmodel(i.e.,CFIandTLIincreaseof.014and .011,respectively).Thethirdmodel(i.e.,theMIMICinvariantinter- ceptmodel,seeModel5-3inTable1)alsoprovidedasatisfactory fittothedata(CFIandTLI>.90;RMSEA<.05)and alsoprovided abetterfittothedatathanMIMICnulleffectmodel.Finallythe resultsshowedthatthefitoftheMIMICsaturatedmodelwasnot substantiallybetter(i.e.,CFI,TLI,andRMSEAareallunder .01)thanthefitoftheMIMICinvariantinterceptmodel.Thisresult showsthattheeffectsofBMIlevelsontheitemsarefullyexplained bytheeffectsofBMIlevelsonthelatentmeans,thussupporting theinvarianceoftheitemsinterceptsasafunctionofBMIlevels.

DetailedresultsfromtheMIMICinvariantinterceptmodelrevealed thatadolescentswithhigherlevelsofBMItendedtopresenthigher latentmeansonthestrengthsubscale( ˆˇ=.162)andlowerlatent

Table2

Standardizedparametersestimatesfromtheconfirmatoryfactoranalyticmodelwithlatentmethodfactor.

Items CO() ST() FL() EN() GSE() HE() AC() BF() SC() GPSC() AP() LMF() ı

CO1 .682 .535

CO2 .739 .454

CO3 .815 .336

CO4 .827 .316

CO5 .796 .366

ST1 .850 .278

ST2 .850 .278

ST3 .850 .278

FL1 .790 .376

FL2 .884 .219

FL3 .811 .342

EN1 .768 .410

EN2 .880 .225

EN3 .787 .381

GSE1 .727 .472

GSE2^a .327 .505 .638

GSE3 .761 .421

GSE4 .742 .449

GSE5^a .153 .654 .549

HE1^a .617 .371 .481

HE2^a .537 .594 .358

HE3^a .623 .505 .357

HE4^a .389 .556 .539

HE5^a .361 .660 .434

AC1 .734 .461

AC2 .800 .361

AC3 .845 .287

AC4 .817 .333

BF1^a .831 .312 .213

BF2^a .859 .352 .138

BF3^a .774 .210 .357

SC1 .855 .269

SC2 .873 .238

SC3 .896 .197

GPSC1 .860 .261

GPSC2 .827 .315

GPSC3 .871 .242

AP1 .847 .283

AP2 .743 .448

AP3 .894 .201

Note.

aReverseditems.

=loading;ı=uniquenesses;AC=activity;AP=appearance;BF=bodyfat;CO=coordination;EN=endurance;FL=flexibility;GPSC=globalphysicalself-concept;GSE=global self-esteem;HE=health;LMF=latentmethodfactor;SC=sportcompetence;ST=strength.Allloadingsanduniquenessessignificantatp<.01

Table3

Latentfactorcorrelations(top)andconvergentvalidity(bottom)oftheconfirmatoryfactoranalytic(CFA)modelwithlatentmethodfactor.

Factors CO ST FL EN GSE HE AC BF SC GPSC AP ␻

LatentfactorcorrelationsfromtheCFAmodel

CO – .88

ST .669^** – .89

FL .705^** .382^** – .87

EN .643^** .730^** .443^** – .85

GSE .626^** .619^** .405^** .629^** – .74

HE .033 −.021 −.148 −.069 −.035 – .75

AC .592^** .605^** .428^** .755^** .504^** .021 – .88

BF .150^* .109 .069 .248^** .255^** .072 .164^** – .90

SC .700^** .776^** .364^** .802^** .683^** −.030 .708^** .271^** – .91

GPSC .586^** .623^** .353^** .622^** .781^** −.009 .490^** .491^** .710^** – .89 AP .437^** .551^** .353^** .461^** .656^** −.086 .396^** .225^** .440^** .702^** – .87 Latentfactorcorrelationsfromtheconvergentvalidityanalyses

PSI-S-GSW .415^** .419^** .256^** .379^** .745^** .018 .272^** .504^** .469^** .874^** .696^** .70

PSI-S-PSW .627^** .717^** .316^** .686^** .713^** .032 .571^** .384^** .773^** .847^** .581^** .86

PSI-S-PC .511^** .642^** .301^** .922^** .471^** −.027 .570^** .234^** .680^** .500^** .394^** .88

PSI-S-SC .594^** .728^** .315^** .688^** .619^** .014 .593^** .286^** .884^** .647^** .473^** .91

PSI-S-PA .499^** .571^** .369^** .510^** .705^** −.064 .432^** .492^** .556^** .840^** .877^** .47

PSI-S-PS .484^** .938^** .253^** .640^** .514^** −.038 .514^** .086 .665^** .533^** .454^** .83

Note.ω=McDonald(1970)scalescorereliabilitycoefficient;AC=activity;AP=appearance;BF=bodyfat;CO=coordination;EN=endurance;FL=flexibility;GPSC=global physicalself-concept;GSE=globalself-esteem;GSW=globalself-worth;HE=health;PA=physicalattractiveness;PC=physicalcondition;PS=physicalstrength;PSI- S=PhysicalSelf-Inventory-shortversion;PSW=physicalself-worth;SC=sportcompetence;ST=strength.

* p<.05.

** p<.01.

meanonthebodyfat³( ˆˇ=−.456)subscale.However,nosignifi- cantdifferenceswerefoundfortheotherPSDQ-Ssubscales.

ConvergentValidity

TheconvergentvalidityofthePSDQ-Swasassessedinrelation toanothervalidatedmeasureofphysicalself-concept(PSI-S)using latentcorrelations.Thegoodness-of-fitstatisticsoftheCFAmodel includingbothinstrumentsarefullysatisfactory(seeModel1-3in Table1),andthescalescorereliabilitycoefficientsofthePSI-S(see Table3)weresatisfactoryformostsubscales(ωrange=.70to.91, M_ω=.84),withtheexceptionofthephysicalattractivenesssubscale forwhichthescalescorereliabilityprovedunsatisfactory(ω=.47;

fordiscussionofproblemsassociatedwiththisscale,seeMorin&

Maïano,2011a).Thus,thissubscalewasnotanalyzedfurther.

Whenthelatentcovariancesbetweenthesubscalesfromboth instrumentswere constrained tobeinvariantacross subgroups formed on the basis of age (²=108.11; df=66; p<.001;

CFI=−0.002;TLI<0.000;RMSEA<0.000),sex(²=103.12;

df=66;p<.001;CFI=−0.002;TLI<0.000;RMSEA<0.000), and sport practice involvement (²=79.99; df=66; p>.05;

CFI=−0.001; TLI=+0.002 RMSEA<0.000), theresultsfully supported the invariance of these latent covariances across subgroups.Theseresultsthussupporttheequivalenceofthecon- vergentvaliditycorrelationsacrosssubgroups.Forthisreason,we onlyfocusonthelatentcorrelationscalculatedonthebasisofthe completesample.

Thelatentvariablescorrelationsbetweenbothinstrumentsare reportedinTable3.Results showedthattheglobalself-esteem (r=.75),globalphysicalself-concept(r=.85),endurance(r=.92), strength(r=.94),andsportcompetence(r=.88)subscalesofthe PSDQ-SwerestronglycorrelatedtotheirrespectivePSI-Ssubscales.

Furthermore,theglobalphysicalself-conceptsubscaleofthePSDQ- Swasalsostronglysignificantlycorrelatedtotheglobalself-worth (r=.87)subscaleofthePSI-S.

However, the results also revealed some high correlations (r>.70)betweennon-correspondingsubscalesfrombothinstru- ments.Forinstance:(a)thestrengthsubscaleofthePSDQ-Swas highlycorrelatedwiththephysicalself-worth(r=.72)andsport competence (r=.73) subscales of the PSI-S, (b) the sport com- petencesubscale of thePSDQ-Swashighly correlated withthe physicalself-worth (r=.77)subscale ofthePSI-S,(d)theglobal self-esteemsubscaleofthePSDQ-Swashighlycorrelatedwiththe physicalself-worth (r=.71)subscale of thePSI-S,(e)theglobal physicalself-conceptsubscaleofthePSDQ-Swashighlycorrelated withtheglobalself-worth(r=.87)subscaleofthePSI-S.Supporting the convergent validity of the PSDQ-S subscales, it is interest- ing tonote that these highcorrelations are in linewith those observedbetweenmatchingPSDQ-Ssubscales.Forinstance,the correlationbetweenthestrengthandglobalphysicalself-concept subscales of thePSDQ-S (r=.62) was similarto thecorrelation betweenthestrength subscaleof thePSDQ-Sand physicalself- worthscaleofthePSI-S(r=.72).Thesecorrelationsarealsoinline withthosepreviouslyreportedamongmatchingPSI-Ssubscales (e.g.,Morin&Maïano,2011a).Finally,itmustbenotedthatthecor- relationsinvolvingtheglobalself-esteemandtheglobalphysical self-conceptsubscaleswereinlinewiththeproposedhierarchical natureofthephysicalself-concept(e.g.,Fox&Corbin,1989),sug- gestingstronglevelsofassociationsbetweenhierarchicallyorder self-components.

3Itshouldbenotedherethathigherscoresonthebodyfatsubscalereflectper- ceptionsthatone’sbodyisthinner,whereaslowerscoresreflectperceptionsthat one’sbodyisfatter.

Discussion

Thefirstobjectiveofthisstudywastoevaluatethefactorvalidity andreliabilityofthePSDQ-SamongasampleofFrenchadolescents.

Theresultsshowedthattheapriori11-factormodelprovideda satisfactorydegreeoffittothedataandthattheincorporationofa LMFrelatedtothenegativelywordeditemsfurtherincreasedthefit oftheaprioriPSDQ-Smodel.Thesuperiorityofamodelincluding aLMFrelatedtothenegativelywordeditemsconfirmsnumerous previousrecommendationsregardingtheneedtocontrolforitem wordingeffectsinmeasurement models(e.g.,DiStefano&Motl, 2006;Marsh,Scalas,etal.,2010).

The resultsshowedthat this final measurement model pre- sented good scale score reliability coefficients (M_ω=.86) and that mostof the PSDQ-Ssubscales weremoderately correlated (Mr=.41),withtheexceptionoftheHealthsubscale,whichpre- sentednon-significantcorrelationswithallotherPSDQ-Ssubscales.

Theseresultsaremostlyconsistentwithpreviousresultsshowing eithernon-significant,oratleastlower,correlationsinvolvingthe Healthsubscale(e.g.,Agarwaletal.,2013;Marsh,Martin,etal., 2010),whichisalsoconsistentwiththefactthatthisspecificsub- scaletapsintoasomewhatdifferentfacetofphysicalconceptions thantherestofthePSDQ-Ssubscales.Overall,theresultsfromthe factorialvalidityandreliabilityofthePSDQ-Sareconsistentwith thoserecentlyreportedbyMarsh,Martin,etal.(2010).

Theresultsalsoshow thatone negativelyworded item (out of a total of 40 itemsincluded inthe PSDQ-S)fromthe global self-esteem subscale was suboptimal. This result is consistent with recent findings reported by Martin and Whalen (2012) among American athleteswith physical disabilities,which also demonstratedthatthetwonegativelywordeditemsoftheglobal self-esteemscaleofthePSDQ-Sweresuboptimal.Althoughthese authorsusedthisobservationtosupporttheeliminationofthese items,werecommendthatfuturestudiesusingboththeFrenchand originalEnglishversionofthePSDQ-Sshoulddevotespecialatten- tiontotheperformanceofthisitem.Furthermore,itmayalsobe interestingforfutureresearchusingboththeEnglishandFrench versionofthePSDQ-Stoinvestigatetheadditionofapositively wordedversionofthisitem(i.e.,“EverythingIdoseemstoturnout right”–“Danslavie,toutcequejefaissemblebientourner”).This willhelptodeterminewhetherthepsychometricpropertiesofthe currentFrenchversionofthePSDQcanbepreservedorimproved withtheproposedreformulationofthisitem,whileatthesame timeensuringthatthesamesetofitemsisretainedacrosslinguistic versions.

Thesecond objectivewasto examinethemeasurement and latentmeaninvarianceofthefactorstructureofthePSDQ-Sacross varioussubgroups(i.e.,sex,age,BMI,andsportpracticeinvolve- ment)of adolescents.Analyses provided strongsupportfor the strictmeasurementinvariance(i.e.,invarianceofthefactorload- ings, items’ intercepts,and items’uniquenesses)of thePSDQ-S acrossallofthesesubgroups.Theseresultsarenotonlyconsistent withthosefromMarsh,Martin,etal.(2010)forthesexandage subgroups,butalsodemonstrateforthefirsttimetheinvarianceof thePSDQ-Sacrosssubgroupsformedbasedontheirinvolvement, orlackthereof,innormativelevelsofsportpractice.

Additionally, resultsrevealed significant latent mean differ- encesacrosssexandsportpracticesubgroups.Theseresultsare consistentwiththosereported frompreviousstudies thathave investigatedmean-leveldifferencesinphysicalself-conceptions, thussupportingthevalidityoftheFrenchversionofthePSDQ-S (e.g.,Bowker,2006;Findlay&Bowker,2007;Marshetal.,1997;

Marsh,Martin,etal.,2010;Marsh,1998;Schmalz&Davison,2006).

Moreprecisely, theseresultsshowedthatboysandadolescents involvedinsportpracticetendtopresenthigherlevelsonmost (exceptforflexibilityacrosssex,orhealthandbodyfatacrosssport