Developmental Trajectories of Co-Occurring Depressive, Eating, Antisocial, and Substance Abuse ...

15 Pages·2006·0.48 MB·English

JournalofAbnormalPsychology Copyright2006bytheAmericanPsychologicalAssociation
2006,Vol.115,No.3,524–538 0021-843X/06/$12.00 DOI:10.1037/0021-843X.115.3.524
Developmental Trajectories of Co-Occurring Depressive, Eating,
Antisocial, and Substance Abuse Problems in Female Adolescents
Jeffrey R. Measelle Eric Stice
UniversityofOregon UniversityofTexasatAustin
Jennifer M. Hogansen
UniversityofOregon
Growth trajectories of co-occurring symptomatology were examined in a community sample of 493
femaleadolescentswhowerefollowedannuallyfromearlytolateadolescence.Onaverage,depression,
eating disorder, and substance abuse symptoms increased over time, whereas antisocial behavior
decreased. Increases in each symptom domain were associated with relative increases in all other
domains. Initial depressive and antisocial behavior symptoms predicted future increases in the other;
substanceabuseandantisocialbehaviorsymptomsalsoshowedprospectivereciprocalrelations.Initial
depressionpredictedincreasesineatingdisorderandsubstanceabusesymptoms.Initialeatingdisorder
symptomspredictedincreasesinsubstanceabuseproblems.Finally,theresultssuggestthatthedevel-
opmentalcovariationbetweendepressiveandeatingdisordersymptomsandbetweenantisocialbehavior
andsubstanceabusesymptomswasaccountedforbydistinctbutrelated2nd-ordergrowthparameters.
Keywords:comorbidity,developmentaltrajectories,femalepsychopathology
Recentepidemiologicalinvestigationsofgender-relatedaspects course of co-occurring domains of psychopathology. Relatively
of psychopathology have helped to elucidate a number of worri- little is known about the temporal relations among symptom do-
someaspectsaboutfemalepsychopathologyduringthe2nddecade mainsoraboutindividualdifferencesinmentalhealthtrajectories
of life. For female adolescents, adolescence is characterized by comprising co-occurring problems. In addition, few studies have
increased levels of symptomatology in a number of domains, prospectively examined more than two co-occurring conditions,
particularlydepression,eatingpathology,andsubstanceusesymp- particularly those that span syndromal domains, which seem im-
toms(Langerbucher&Chung,1995;Lewinsohn,Pettit,Joiner,& portantgivenevidencethatthepresenceofathirdsyndromemight
Seeley, 2003; Stice, Killen, Hayward, & Taylor, 1998); external- significantlyalterdevelopmentalprocesses(Harrington,Rutter,&
izing problems are also common among teenage girls (Broidy et Frombonne,1996).
al., 2003). Further, adolescence is marked by high rates of co- The overarching goal of this study was to examine the devel-
occurringpsychopathologyforgirls(Kessleretal.,1994).Indeed, opmentaltrajectoriesofthesymptomsofpsychiatricdisordersthat
rates of female comorbidity typically surpass the rates of single oftenco-occurovera5-yearspanofadolescence.Specifically,we
disorders(Angold,Costello,&Erkanli,1999;Kessleretal.,1994) investigated the course and covariation among four domains of
andtendtoincreasewithage(Tolan&Henry,1996). self-reported symptomatology in a community sample of female
Althoughprogresshasbeenmadeinunderstandingpsychiatric adolescents:depressive,eatingdisorder,antisocial,andsubstance
comorbidity (Angold et al., 1999; Caron & Rutter, 1991), signif- abusesymptoms.Theaimswereto(a)characterizegirls’symptom
icant gaps in knowledge remain regarding the developmental trajectories during adolescence, (b) examine the temporal associ-
ationsamongco-occurringsyndromes,and(c)determinethenum-
ber of higher order factors needed to adequately describe the
relations among the first-order growth factors. To address these
JeffreyR.MeaselleandJenniferM.Hogansen,DepartmentofPsychol-
aims,weusedlatentgrowthcurvemodels(LGMs).Giventhatour
ogy,UniversityofOregon;EricStice,DepartmentofPsychology,Univer-
sampleofgirlsrangedfrom13to15yearsofageatthefirstoffour
sityofTexasatAustin.
ThisstudywassupportedbyNationalInstitutesofHealthcareeraward annual assessments, we used an age-based analytic approach
MH01708 and research grant MH/DK61957. We thank Daniel Bauer, (Mehta & West, 2000) in which we fit growth curves in each
TerryDuncan,andHelenaKraemerfortheirvaluablecontributionstothe symptomdomainthatspanneda5-yearperiodofadolescencefrom
analysesandinterpretationofthesedata.Wealsothanktheprojectresearch age13to18.
assistants, Sarah Kate Bearman, Emily Burton, Melissa Fisher, Lisa
Groesz,NatalieMcKee,andKatyWhitenton;ournumerousundergraduate Single and Co-Occurring Syndromes Over Time
volunteers; the Austin Independent School District; and the participants
whomadethisstudypossible. Depression
CorrespondenceconcerningthisarticleshouldbeaddressedtoJeffrey
R. Measelle, Department of Psychology, University of Oregon, Eugene, Depressionisthemostprevalentpsychiatricproblemforfemale
OR97401.E-mail:[email protected] adolescents, with nearly 20% experiencing clinically significant
524
FEMALECOMORBIDITYTRAJECTORIES 525
depressive symptomatology during the teenage years (Kessler, Antisocial Behavior
Avenevoli,&Merikangas,2001;Lewinsohnetal.,2003).Interms
of the developmental course, studies suggest that a significant Although preadolescent conduct problems occur more fre-
proportion of female adolescents will show increasing levels of quently in boys than in girls (Moffitt & Caspi, 2001), the preva-
depressivesymptomsovertime(Garber,Keiley,&Martin,2002; lence of conduct problems is similar in male and female adoles-
Lewinsohn et al., 2003). Although there is some variation in the cents(Hinshaw&Lee,2003;Moffitt,Caspi,Harrington,&Milne,
symptomsofdepressionthatoccurovertime(Angst,Merikangas, 2002). The developmental trajectories of girls’ conduct problems
& Preisig, 1997), severity levels have been shown to increase havereceivedlittleattention.Oneexceptionfoundthatgirls(and
duringadolescenceandearlyadulthood(Lewinsohnetal.,2003). boys) showed declining levels of externalizing symptoms during
Intermsofcomorbidity,girlswithdepressionfrequentlyexpe- adolescence (Bongers, Koot, van der Ende, & Verhulst, 2003).
rience elevated levels of anxiety, conduct, substance abuse, and However, there is evidence that girls’ (and boys’) conduct prob-
eating disorder symptomatology (Angold et al., 1999; Kessler et lems tend to increase over time when they co-occur with depres-
al., 2001; Rohde, Lewinsohn, & Seeley, 1991; Stice, Presnell, & sion (Beyers & Loeber, 2003; Silberg et al., 2003). Indeed, the
Bearman, 2001). Research investigating the temporal relations developmental course of antisocial behavior across adolescence
betweensymptomsofdepressionandco-occurringproblemssug- appearstobesignificantlyintertwinedwithco-occurringformsof
gests that depressive symptoms tend to follow the onset of other psychopathology, particularly depression and substance abuse
mood-related problems (Avenevoli, Stolar, Li, Dierker, & Ries symptoms (Angold et al., 1999; Loeber, Stouthamer-Loeber, &
Merikangas, 2001) as well as conduct problems (Rohde et al., White, 1999). Among girls, however, little is known about the
1991;Silberg,Rutter,D’Onofrio,&Eaves,2003).Thesefindings temporal relations among antisocial behavior and co-occurring
have led to speculation that depressive comorbidity is either a symptomatology(Hinshaw&Lee,2003).
nonspecificreflectionofgeneralpsychopathologyorareactionto
the failures in adaptation created by earlier behavior problems Substance Abuse Problems
(Capaldi & Stoolmiller, 1999). However, there is also evidence
that elevated depressive symptoms predict future onset of other Substance abuse also typically emerges during adolescence
disturbances, including eating pathology (Stice, Burton, & Shaw, (O’Malley,Johnston,Bachman,&Schulenberg,2000).Although
2004; Stice, Presnell, & Spangler, 2002). Because prospective there are some gender differences in the substances used (e.g.,
comorbidityresearchissparse,itisnotclearhowtheassociations male adolescents use and abuse alcohol more than do female
between depressive symptoms and other symptom areas change adolescents, whereas female adolescents favor stimulants; Chas-
over time or how the reciprocal influences among co-occurring sin,Ritter,Trim,&King,2003),epidemiologicalstudiessuggest
symptomdomainsbecomeestablished. that substance abuse symptoms increase steadily across adoles-
cence in female adolescents (Johnson, Cohen, Kotler, Kasen, &
Brook,2002).
Eating Problems
Longitudinal studies indicate that the developmental trajectory
Eatingpathologyisanothercommonpsychiatricproblemfaced ofsubstanceabusesymptomsdependslargelyontheageofonset.
primarily by girls (Lewinsohn, Striegel-Moore, & Seeley, 2000; Earlyonsetofsubstanceabuse(e.g.,beforeage15)isassociated
Wilson, Becker, & Heffernan, 2003), and middle to late adoles- with a stable and escalating course of abuse for girls (Chassin,
cence represents the period of greatest risk for onset of these Pitts,&Prost,2002;Nagin&Tremblay,2001).Investigationsof
problems(Lewinsohnetal.,2000;Stice,Killen,etal.,1998).The substance abuse trajectories that consider other forms of psycho-
trajectories of eating problems appear highly variable and are pathology find that female substance abuse development is inter-
defined by considerable instability (Stice et al., 2002). Many woven with antisocial, depressive, and eating disorder symptom-
female adolescents will recover after initial elevations in eating atology (Angold et al., 1999). Prospective studies indicate that
disordersymptomatology,whereasotherswillmanifestpatternsof substance abuse symptoms typically follow onset of other distur-
relapse and recovery (Fairburn, Cooper, Doll, Norman, & bances,especiallyantisocialsymptoms(Brook,Cohen,&Brook,
O’Connor, 2000). Much remains to be understood about the pro- 1998).Rohde,Lewinsohn,andSeeley(1996),however,foundthat
spective nature of eating pathology, particularly eating disorder alcoholabuseprecededfuturedepressioninfemaleadolescents.In
symptomsthatco-occurwithothersymptomdomains. general, a lack of prospective work on female adolescents leaves
Eatingproblemsshowhighratesofco-occurrencewithsubstance open the question of temporal sequencing with certain substance
abuse (Dansky, Brewerton, & Kilpatrick, 2000; Strober, Freeman, abusecomorbidities.
Bower,&Rigali,1996)anddepressivesymptoms(Lewinsohnetal.,
2000;Sticeetal.,2001).However,understandingtheprocessesthat Overview of Study
giverisetothesecomorbiditiesaswellasclarityabouttheirtemporal
interplayremainstentative.Forexample,eatingdisorderanddepres- The goal of this study was to examine the codevelopment of
sivesymptomatologyappeartocovaryovertime(Sticeetal.,2001); depression,eatingdisorder,antisocial,andsubstanceabusesymp-
however,theoriginandthecourseofthiscovariationareunclear.As tomsinacommunitysampleoffemaleadolescentswhovariedin
withanypairofsyndromes,eatingdisorderanddepressivesymptom- agefrom13to15yearsatthefirstoffourannualassessments.We
atology may covary as a function of shared risk, a direct causal usedLGMstoaddressthesequestions,asitcanprovideunivariate
association (or reciprocal causality), or a common underlying syn- and multivariate estimates of stability and time-related change
dromeofpsychopathology(e.g.,similarcognitivebiases;Fergusson, (T. E. Duncan, Duncan, Strycker, Li, & Alpert, 1999; Willett &
Lynskey,&Horwood,1996;Zoccolillo,1992). Sayer,1994).Ouruseofanage-basedanalyticapproach(Mehta&
526 MEASELLE,STICE,ANDHOGANSEN
West,2000)enabledustogenerategrowthcurvesineachsymp- domains of psychopathology (Krueger, 1999; Vollebergh et al.,
tomdomainthatspanneda5-yearperiodofadolescencefromage 2001). By extension, more than one set of higher order growth
13 to 18. We focused on female adolescents because these data factors may be needed to represent patterns of codevelopment
weredrawnfromalongitudinalstudyoftheriskfactorsforeating among depression, eating disorder, antisocial, and substance use
pathology,whichpredominantlyaffectsgirls.Wefeltthiswasan symptomatology. Our a priori expectation was that two higher
appropriate sample because girls often experience symptomatic orderfactorswouldbenecessary,onetoaccountforthecodevel-
increases in depression (Hankin et al., 1998), substance abuse opment of depressive and eating disorder symptomatology and
(Langerbucher&Chung,1995),andeatingproblems(Stice,Bar- another to account for the codevelopment of antisocial and sub-
rera, & Chassin, 1998) during adolescence. Antisocial behavior stanceabusesymptoms.Theoriesofaffectdisturbanceandmood
mayalsobemorecommoninfemaleadolescentsthanwaspreviously regulation often link depressive symptomatology and eating pa-
thought(Cote,Zoccolillo,Tremblay,Nagin,&Vitaro,2001). thology (Stice et al., 2004), whereas models of behavioral disin-
Our first objective was to characterize the developmental tra- hibition and poor impulse control often link antisocial behavior
jectory of each domain as well as to evaluate the amount of and substance abuse problems (Pennington, 2002). Support for a
sample- and individual-level variability in the initial levels and two-factorhigherordermodelmighthelptoelucidatesimilarities
rates of change in each symptom domain. Although there have and differences among symptom domains. An improved under-
been longitudinal investigations of the trajectories of psychopa- standing of the latent structure underlying co-occurring symptom
thologyinsamplesofmaleadolescents,littleisknownaboutthe domainsmighteventuallyhelpidentifysharedversusspecificrisk
trajectories of female adolescents’ symptomatology. Of interest factors (O’Connor, McGuire, Reiss, Hetherington, & Plomin,
waswhethergirls’symptomtrajectorieschangedinalinear(con- 1998;Silbergetal.,2003).
stant)ornonlinearfashionacrossa5-yearspanofadolescence.
Oursecondobjectivewastoexaminethetemporalassociations Method
among symptom dimensions to better understand whether initial
elevations in one symptom domain predicted future increases in Participants
other symptom domains. The use of a multivariate LGM (T. E.
Participants were 496 female adolescents who were assessed annually
Duncanetal.,1999)thatexaminedtheassociationsamonggrowth
overa5-yearperiod(Time1–Time5[T1–T5]).Becauseantisocialbehav-
parameters enabled us to test processes that might influence the
iorwasnotassessedatT1,onlydatafromT2,T3,T4,andT5wereused
expressionofsymptomco-occurrenceovertime.Initiallevelsand inthepresentreport.Thisstudyhadalowattritionrateandlittlemissing
growth of two syndromes may be related with neither predicting data.Oftheoriginal496femaleadolescentswhostartedthestudyatT1,
futurechangesintheother(unrelatedcoexistence),initiallevelsof 493hadusabledataatT2–T5,andofthese,theamountofmissingdata
one syndrome may predict future changes in the other (unidirec- rangedfrom1.4%to9.7%,dependingonthevariable.1Thetimebetween
tionaleffect),orinitiallevelsofbothsyndromesmaypredictfuture assessmentswasapproximately1year(M(cid:1)376days,SD(cid:1)6days),with
changesintheother(reciprocaleffects;Caron&Rutter,1991).It 80%ofthesamplecompletingeachassessmentwithin(cid:2)15daysoftheir
prior assessment and the remaining girls within (cid:2)30 days of their prior
isimportanttoinvestigatesuchalternativeexplanationsofsymp-
assessment.
tom co-occurrence because each has distinct etiological, preven-
AtT2,therewasageheterogeneityinthesampleasgirlsrangedinage
tion,andtreatmentimplications.Forexample,ifsubstanceabuse
from12to15years(M(cid:1)14.48,SD(cid:1)0.67).Toconductanage-based
is a risk factor for depression, but not vice versa, prevention
analysis(followingproceduresoutlinedbyMehta&West,2000),wefirst
programswouldneedtotargettheformertoeffectivelyreducerisk roundedparticipants’agetothenearestwholeyear.Fivegirlswerewithin
forbothconditions. 2 months of 151⁄2 years; given the instability associated with estimating
Our third objective was to determine the number of second- trajectory means for just five 16-year-olds, the age of these girls was
orderfactorsneededtoaccountforthevarianceinandcovariation rounded to 15. This resulted in a sample comprising 123, 243, and 127
amongfirst-orderinterceptandgrowthfactors.Thegeneralnotion thirteen-,fourteen-,andfifteen-year-olds,respectively,atT2.Participants
ofasecond-orderfactormodelisonefamiliartoresearchersusing werefromfourpublic(82%)andfourprivate(18%)schoolsinametro-
politanareaofthesouthwesternUnitedStates.Thesampleincluded2%
obliqueconfirmatoryfactoranalysis,inwhichthequestionarises
Asian/Pacific Islanders, 7% African Americans, 68% Caucasians, 18%
as to the source of the obliqueness among first-order factors
Latinas,1%NativeAmericans,and4%whospecified“other”or“mixed”
(Hancock, Kuo, & Lawrence, 2001). One explanation of comor-
racialheritage,whichwasrepresentativeoftheethniccompositionofthe
bidity is that it constitutes an undifferentiated accumulation of
schoolsfromwhichwesampled(2%Asian/PacificIslanders,8%African
distress(Krueger,1999;Lilienfeld,2003).Ifasinglehigherorder Americans, 65% Caucasians, 21% Hispanics, 4% “other” or “mixed”).
growthfactorcouldadequatelycharacterizethetemporalcovaria- Averageeducationalattainmentofparents(29%highschoolgraduateor
tionamonggirls’depressive,eatingdisorder,antisocial,andsub-
stanceabusesymptoms,thismightbeinterpretedassupportfora
core or common factor explanation of symptom co-occurrence. 1Mplus draws on the theory of Little and Rubin (1987; expectation/
Evidenceofacorefactormightalsosuggestthatstrongdiagnostic maximizationalgorithm)toallowfortheinclusionofrespondentswhose
data appear to be missing at random. Through the use of maximum-
lines separating common syndromes might be developmentally
likelihoodestimation,theactualdataaresortedintomissingandnonmiss-
premature during adolescence. Alternatively, a core factor might
ingpatterns.Mplusthenestimatesacovariancematrixfromtheavailable
suggestthepresenceofcommonrisk,suchassharedgeneticrisk
raw data and a second coverage matrix in which missing data are held
(Silbergetal.,2003)oracoretypeoftemperamentalvulnerability
constant to correspond with the maximum-likelihood estimation of that
(Krueger,Caspi,Moffitt,&Silva,1998). portionofthenonmissingmodel(Muthe´n&Muthe´n,2001).Thus,while
Studies with adults suggest that more than one higher order not imputing new data, Mplus estimates latent models by using the
factor is needed to account for the covariation among common maximum-likelihoodestimationofthecoveragematrix.
FEMALECOMORBIDITYTRAJECTORIES 527
less,23%somecollege,33%collegegraduate,15%graduatedegree)was .88)and1-weektest–retestreliability((cid:1)(cid:1)1.00)inthepresentstudy(Stice
alsosimilartocensusdataforcomparablyagedadults(34%highschool etal.,2004).
graduateorless,25%somecollege,26%collegegraduate,15%graduate Antisocial behavior. Girls’ antisocial behavior symptoms were as-
degree). sessed with 13 items from the Externalizing Syndrome of the Child
Behavior Checklist (CBCL), 5 of which came from the CBCL’s Delin-
quency subscale and 8 of which came from the CBCL’s Aggression
Procedure
subscale (Achenbach & Edelbrock, 1983). In the present study, the re-
sponsescalewasexpandedtoa5-pointformat,rangingfrom1(never)to
Thestudywasdescribedtoparentsandparticipantsasaninvestigation
5(always),toincreasevariance.Severityratingsforall13symptomswere
of adolescent mental and physical health. An active parental consent
averagedtoformacontinuouslymeasuredantisocialbehaviorcomposite
procedurewasusedtorecruitparticipants,inwhichaninformed-consent ((cid:2)(cid:1).91atT2).Thissymptomcompositeevidencedinternalconsistency
letterandastampedself-addressedreturnenvelopeweresenttoparentsof ((cid:2)(cid:1).88),1-yeartest–retestreliability(r(cid:1).62),andpredictivevalidityin
eligiblegirls(asecondmailingwassenttononrespondersafter2weeks).
apriorstudy(Stice,Barrera,&Chassin,1998).
Adolescent assent was secured immediately before data collection took
Substanceabuse. ItemsadaptedfromStice,Barrera,&Chassin(1998)
place. This procedure resulted in an average participation rate of 56%.
assessed DSM–IV substance abuse symptoms over the past year. These
Althoughlowerthanwehopedfor,thisparticipationratewassimilarto
itemswerespecificallydevelopedtoassesssubstanceabuseinadolescents.
thatofotherschool-recruitedsamplesthatusedactiveconsentprocedures
Items were averaged to create a substance abuse symptom composite at
andinvolvedstructuredinterviews(e.g.,61%;Lewinsohn,Hops,Roberts, each time point ((cid:2)(cid:1) .86 at T2), ranging from 0 (never) to 2 (twice or
& Seeley, 1993). Furthermore, two pieces of evidence suggest that our
more). Prior work with this measure indicated that these items possess
samplewasrepresentative.First,asjustdescribed,theethniccomposition adequate internal consistency ((cid:2)(cid:1) .85) and convergent validity (Stice,
and parental education of the sample were comparable to the ethnic Barrera,Chassin,1998;Sticeetal.,2001).Pilottesting(N(cid:1)62)revealed
composition of the schools from which we sampled. Second, the 1-year
a1-monthtest–retestcoefficientof.78forthesubstanceabusesymptom
prevalenceratesofmajordepression(4.2%),bulimianervosa(0.4%),and
composite.Moregenerally,self-reportsofsubstanceabuseappeartobethe
substanceabuse(8.9%;Sticeetal.,2001)weresimilartotheprevalence
most valid measure of substance abuse (Winters, Stinchfield, Henly, &
ratesfromotherepidemiologicalstudies(Lewinsohnetal.,1993).
Schwartz,1991).
Girlscompletedaquestionnaire,participatedinastructuredinterview,
andhadtheirweightandheightmeasuredbyfemaleresearchassistantsat
allassessments.Femaleassessorswithabachelor’s,master’s,ordoctoral Statistical Analyses
degreeinpsychologyconductedallinterviews.Assessorsattended24hrof
training,inwhichtheylearnedinterviewskills,revieweddiagnosticcriteria
WeusedMplus(Version2.13,Muthe´n&Muthe´n,2001)tofitLGMs
forrelevantdisordersintheDiagnosticandStatisticalManualofMental
using maximum-likelihood estimation procedures. For our purposes,
Disorders(4thedition;DSM–IV;AmericanPsychiatricAssociation,1994),
Mpluswasanappropriateoptionasitallowedfortheestimationofgrowth
observedsimulatedinterviews,androle-playedinterviews.Assessorshad
parametersinthepresenceofageheterogeneitybyusingthemissing-data
todemonstrateaninterrateragreementfordiagnoses((cid:1)(cid:3).80)withexperts
approachinwhichtheoutcomeateachdistinctageinthedataisreadinas
using tape-recorded interviews before collecting data. Interviewers were
aseparatevariable,andeachindividualisassumedtohavemissingdataat
recordedperiodicallythroughoutthestudytoensurethatassessorscontin-
agesforwhichshedidnotprovidedata(Mehta&West,2000;Muthe´n&
uedtodemonstrateacceptableinterrateragreement((cid:1)(cid:3).80).Assessments
Muthe´n, 2001). A common LGM is then applied using these individual
tookplaceduringregularschoolhoursorimmediatelyafterschoolonthe
growthcurves.FollowingproceduresoutlinedbyMehtaandWest(2000),
schoolcampusoratparticipants’houses.Girlsreceivedagiftcertificateor
weestimatedthegrowthparametersinourLGMswithagescaledtoage
cashforcompletingeachassessment.
13,theyoungestvalueatourfirstwaveofdata.
We took an empirical approach to testing the error variance in each
Measures LGM.Becauseamodelwithhomoscedasticvariance(invariantacrosstime
points) is nested within a model with heteroscedastic variance (varying
Depressivesymptoms. AnadaptedversionoftheScheduleforAffec- acrosstimepoints),wecomparedthefitofeachtoseewhichwasmore
tive Disorders and Schizophrenia for School-Age Children (K-SADS; tenable.Ourinitialexpectationwasthatdifferenterrorvarianceswouldbe
Puig-Antich,1982),astructuredpsychiatricinterview,wasusedtoassess neededbecauseofage-relateddifferencesinmeasurementerroraswellas
diagnosticcriteriaforDSM–IVmajordepression.Severityratingsforeach the possible influences of actual age-specific events on girls’ reporting,
symptomwereaveragedtoformacontinuousdepressivesymptomcom- whichmayhaveagreaterimpactatsomeagesthanothers.Additionally,
posite((cid:2)(cid:1).85atT2),rangingfrom1(never)to4(always).TheK-SADS wetestedtoseewhetherallowingforcovaryingerrors,specificallyauto-
generallyhasgoodtest–retestreliability((cid:1)(cid:1).63–1.00),interraterreliabil- correlations among adjacent observed indicators, would be necessary. If
ity((cid:1)(cid:1).73–1.00),andinternalconsistency((cid:2)(cid:1).68–.84)anddiscrimi- sucheffectsweresignificant,theymightbesaidtorepresenttheeffectsof
natesbetweendepressedandnondepressedindividuals(Lewinsohnetal., symptomatology at one point in time on the next, independent of the
1993).TheK-SADSdepressiondiagnoseshaveshownexcellentinterrater developmentaleffectsrepresentedbythelatentgrowthfactors.
agreement((cid:1)(cid:1)1.00)and1-weektest–retestreliability((cid:1)(cid:1)1.00)inthe We assessed model fit with multiple criteria as outlined by Hu and
presentstudy(Sticeetal.,2004). Bentler(1999):thecomparativefitindex(CFI(cid:3).95),theTucker-Lewis
Eating pathology. The Eating Disorder Examination (EDE; Fairburn index (TLI (cid:3) .95), and the root-mean-square error of approximation
& Cooper, 1993), a structured psychiatric interview, was used to assess (RMSEA(cid:4).06)andits90%confidenceinterval(CI).Althoughreported,
diagnostic criteria for DSM–IV bulimia nervosa, anorexia nervosa, and nonsignificantchi-squarevalueswouldbeunlikelygiventhesamplesize
bingeeatingdisorder.Diagnosticitemswereaveragedtoformanoverall and are thus of little value in evaluating model fit (T. E. Duncan et al.,
eatingsymptomcompositeateachtimepoint((cid:2)(cid:1).96atT2),rangingfrom 1999).Tocomparethefitofcompetinghigherordermodels,weexamined
0(never)to2(always).TheEDEgenerallyhasgoodinternalconsistency improvementsinthechi-squarecoefficientsbyusinganestedchi-square
((cid:2)(cid:1) .76–.90) and interrater reliability ((cid:1)(cid:1) .70–.99) and discriminates test(Stoolmiller,1998),Akaike’s(1987)informationcriterion,andBoz-
betweenindividualswitheatingdisordersandcontrols(Fairburn&Coo- dogan’s (1987) consistent version of this statistic, all of which are
per,1993;Williamson,Anderson,Jackman,&Jackson,1995).TheEDE parsimony-based indices intended for model comparison, not fit
eatingdisorderdiagnoseshaveshownexcellentinterrateragreement((cid:1)(cid:1) evaluation.
528 MEASELLE,STICE,ANDHOGANSEN
To test the form of growth in each symptom domain, we initially Table1
hypothesizedthattherewouldbesignificantlinearchangefromage13to DescriptiveStatisticsandIntercorrelationsforFemale
18.Toidentifythelineargrowthmodel,theslopeloadingswerefixedat0 Adolescents’Depressive,EatingDisorder,AntisocialBehavior,
(age13),1(age14),2(age15),3(age16),4(age17),and5(age18)to andSubstanceAbuseSymptoms
reflectaconstantrateofchangebetweeneachage.However,individuals
may not change in a linear fashion alone, and change (accelerations or Symptomandage M SD Mean
decelerations) in each symptom domain may occur at different rates at
differentpointsofdevelopment(Anderson,1993).Thus,totesttheade- Depression
quacyofthelinearhypothesis,wealsotestedunspecifiedmodelswithin Age13 1.30 0.30
eachsymptomdomain(Anderson,1993;S.C.Duncan&Duncan,1996; Age14 1.35 0.35
McArdle & Anderson, 1990). These unspecified models allowed us to Age15 1.39 0.40
Age16 1.40 0.41
evaluatenonlinearitiesbyfreelyestimatingportionsofeachdevelopmental
Age17 1.41 0.42
trajectory(seeMcArdle&Hamagami,1991).
Age18 1.41 0.44
Oncewecouldreasonablycharacterizetheshapeofeachdevelopmental Stability .49
trajectory—beitlinearornonlinear—weexaminedtheassociationsamong Intercorrelation .33
symptom domains. To do so, we tested an associative LGM in which Eatingdisorder
multivariaterelationshipsbetweensymptomgrowthparameterswereeval- Age13 0.41 0.37
uated. A reparameterization of this model enabled us to test prospective Age14 0.55 0.41
associations among symptom domains. To achieve the latter, the slope Age15 0.59 0.37
factorforeachsymptomdomainwasregressedonitsownandthethree Age16 0.61 0.36
Age17 0.63 0.32
otherinterceptfactors.
Age18 0.64 0.41
Finally, we estimated competing higher order LGMs to examine the
Stability .51
degreetowhichtherelationsamongtheprimary,first-ordergrowthfactors Intercorrelation .29
couldbedescribedbyoneormorehigherorderlatentgrowthconstructs. Antisocialbehavior
Thehigherordermodelfollowsastructurethatissimilartothefirst-order Age13 1.62 0.54
associativeLGM;however,thecovariancesamongthefirst-orderfactors Age14 1.64 0.54
arethoughttobeexplainedbythehigherorderfactors(S.C.Duncan& Age15 1.63 0.56
Duncan,1996).Itisimportanttonotethatevenifthehigherordermodel Age16 1.61 0.53
isabletoaccountforallofthecovariationamongthefirst-orderfactors,the Age17 1.56 0.48
Age18 1.50 0.44
goodness-of-fit indices cannot improve over those of the corresponding
Stability .58
first-ordermodel.Nonetheless,ifthefitindicesforthesecond-ordermodel
Intercorrelation .34
approachthoseofthecorrespondingfirst-ordermodel,thenthehierarchical Substanceabuse
model can be considered appropriate (S. C. Duncan & Duncan, 1996; Age13 0.07 0.22
Marsh,1985). Age14 0.07 0.25
OurfirsthigherorderLGMreflectedthehypothesisthatasinglepairof Age15 0.13 0.28
growthfactors(interceptandslopeestimates)wouldbestaccountforthe Age16 0.14 0.30
developmental associations among all four symptom domains. A second Age17 0.12 0.29
competingmodelreflectedouraprioriexpectationthatseparatepairsof Age18 0.12 0.28
Stability .50
growthparameterswouldbeneededtomodelthevarianceinandcovaria-
Intercorrelation .28
tion among the first-order growth factors of each symptom domain. In
particular,weexpectedthatseparateinternalizingandexternalizingdevel- Note. Correlations greater than .15 are significant at p (cid:4) .01. Stability
opmental growth factors would best account for the codevelopment of mean(cid:1)average(Fisherr-to-ztortransformation)ofcorrelationsbetween
depressionandeatingpathologyontheonehandandantisocialbehavior symptomscoresthatareadjacentinyears.Intercorrelationmean(cid:1)average
andsubstanceabuseontheotherhand. (Fisherr-to-ztortransformation)intercorrelationbetweendomains.
Results
Preliminary Analyses Girls’reportswithineachsymptomdomaindemonstratedmoder-
ate but significant levels of 1-year stability. The coefficients in
The means and standard deviations for each symptom domain
Table1alsoshowthatthereweresignificantlevelsofassociation
arepresentedinTable1foralloftheagescoveredintheage-based
analysis. Girls’ depression, eating disorder symptoms, and sub-
stanceabusescoresincreasedonaverage,whereasgirls’antisocial 2Wealsoexaminedtheextenttowhichmean-levelchangesmaskedclin-
behavior decreased on average. Although the observed mean icallymeaningfulchange.Fortheseanalysesweuseddichotomousscoresthat
changesfromage13to18forgirls’depressive(Cohen’sd(cid:1).29), classifiedgirlsaseitherbeloworabovesubthresholdorthresholdlevelsof
eatingdisorder(Cohen’sd(cid:1).59),antisocial(Cohen’sd(cid:1)(cid:5).24), symptomatology.Ofthe435girlswhowerenotclinicallydepressedatT2,
andsubstanceabuse(Cohen’sd(cid:1).24)symptomsrepresentsmall 14%showedonsetofsubthresholdorthresholdmajordepressionbyT5.Ofthe
480girlswithoutself-reportedeatingproblemsatT2,9%showedonsetof
to medium effect sizes (Cohen, 1988), information about sample
subthresholdorthresholdeatingpathologybyT5.Ofthe466nonantisocial
averages is likely to mask evidence of meaningful individual
girlsatT2,11%showedonsetofdiagnosticallyrelevantantisocialbehaviorby
variability in terms of initial levels (intercept) and symptom tra-
T5. Alternatively, despite the downward mean trajectory, 36% of the girls
jectories.Thestatisticalsignificanceofthisvariabilityisaddressed
reporting clinical elevations in antisocial problems at T2 were still above
belowintheunivariateLGManalyses.2 thresholdbyT5.Ofthe461nonsubstance-abusinggirlsatT2,8%showed
Table1alsopresentsmeanstabilitydataaswellasanestimate onset of substance use by T5. These data highlight the changing nature of
oftheaveragelevelofintercorrelationamongsymptomdomains. manygirls’clinicalstatus,withameaningfulnumberofgirlsworseningovertime.
FEMALECOMORBIDITYTRAJECTORIES 529
between domains, with higher levels in one domain associated antisocial model represented a significantly better fit over the
withhigherlevelsineachoftheothersymptomdomains. linearmodel(11.60–2.23(cid:1)9.37,df(cid:1)4,p(cid:4).05).
The linear model for girls’ substance abuse symptoms fit the
datareasonablywell,(cid:3)2(13,N(cid:1)487)(cid:1)15.78,p(cid:1).05;CFI(cid:1)
Univariate Symptom Trajectories .99;TLI(cid:1).98;RMSEA(cid:1).04(90%CI(cid:1).01,.07).Theunspec-
ifiedmodelofgirls’substanceabusesymptomatologyalsofitthe
Our first goal was to characterize the developmental trajectory
datawell,(cid:3)2(9,N(cid:1)487)(cid:1)5.24,p(cid:1).26;CFI(cid:1).99;TLI(cid:1).99;
ofeachsymptomdomain.Withineachdomain,wetestedalinear
RMSEA(cid:1).02(90%CI(cid:1).00,.07)andrepresentedasignificantly
growth model first followed by a nonlinear growth model. To
betterfitovertheinitiallinearmodel(15.78–5.24(cid:1)10.54,df(cid:1)
identify our linear model, each slope factor was scaled to reflect
4,p(cid:4).05).
constantchangebetweeneachage(0,1,2,3,4,5).Totestforthe
Insum,theseresultsindicatethatitisreasonabletocharacterize
possibility of nonlinear change, we relaxed the constraints on
girls’reportsoftheirdepressiveandeatingdisordersymptomatol-
linear growth by freeing all but the first and last loadings on the
ogy as growing in a linear fashion across the 5-year period from
slopefactor.Foridentificationpurposes,atleasttwoloadingsmust
age13to18.Incontrast,itismoreappropriatetocharacterizethe
befixedwhentestingnonlineargrowth(S.C.Duncan&Duncan,
development of girls’ antisocial and substance abuse symptom-
1996; McArdle & Anderson, 1990; Meredith & Tisak, 1990;
atologyasnonlinear,withantisocialbehaviordecreasingovertime
Stoolmiller,1998).Whenthereareenoughpointsintimetofreely
andsubstanceabusesymptomsincreasingovertime.Foreachof
estimatefactorloadingsbeyondthetworequiredforidentification
ourbestfittingmodels,wepresentinterceptandslopemeansand
purposes,theslopefactornowrepresentsindividualdifferencesin
variances, residual variances, and factor intercorrelations in
bothgeneraltrend(upanddown)andnonlinearshapeand,thus,is
Table2.
better labeled as a slope/shape factor (S. C. Duncan & Duncan,
The intercept means were significantly different from zero in
1996;Stoolmiller,1998).Additionally,becausethelinearmodelis
each symptom domain. In absolute terms, however, the intercept
a nested case of the less restrictive unspecified model, a nested
means were indicative of mild levels of symptomatology for the
chi-square difference test can be used to compare the relative
sampleatage13.Additionally,thevariancearoundeachmeanwas
adequacyofeachmodel’scapacitytoaccountforgirls’symptom
also significant, indicating that there was substantial variation in
development(T.E.Duncan,Duncan,&Stoolmiller,1994).
girls’ initial status in each domain. Girls’ depressive, eating dis-
The results of our preliminary error testing procedures indi-
order,andsubstanceabusesymptomsincreasedsignificantlyover
cated that allowing the residual variances in each symptom
time, as evidenced by the positive slope means, whereas girls’
domaintovaryfromagetoagewouldyieldsignificantlybetter
antisocialbehaviordecreasedsignificantlyovertime.Thevariance
fitting models than models in which the residual variance was
around each of the slope means was also significant, indicating
constrainedtobeinvariantacrossages.Additionally,therewas
considerablevariabilityingirls’symptomtrajectoriesovertime.
only inconsistent evidence of the need to allow for covarying
Correlationsbetweensymptominterceptandslopefactorswere
error terms. As such, in the LGM results that follow, models
significant and negative. With increasing means, as was the case
comprised both heteroscedastic error variance and unrelated
withgirls’depressive,eatingdisorder,andsubstanceabusesymp-
error terms.
toms, the negative correlation between intercept and slope indi-
The overall fit indices for depression suggested that the linear
modelfitthedatareasonablywell,(cid:3)2(13,N(cid:1)487)(cid:1)32.20,p(cid:1) cated that girls reporting higher initial levels were more likely to
.001;CFI(cid:1).97;TLI(cid:1).97;RMSEA(cid:1).04(90%CI(cid:1).03,.07). showslowerratesofgrowthovertimethanthosewithlowerinitial
Incontrast,theunspecifiedmodeldidnotfitthedataaswell,(cid:3)2(9, levels.Withdecreasingmeans,aswasthecasewithgirls’antiso-
N(cid:1)487)(cid:1)24.82,p(cid:1).00;CFI(cid:1).94;TLI(cid:1).93;RMSEA(cid:1).06 cial symptom trajectory, a negative correlation between intercept
(90% CI (cid:1) .03, .10). By freeing three slope parameters, the and slope indicated that girls reporting higher initial levels were
chi-square statistic was reduced by just 7.38 (df (cid:1) 4, p (cid:3) .05), more likely to decrease at faster rates than girls reporting lower
initiallevels.3
indicating that the unspecified model did not offer a statistically
Finally,foreachofourbestfittingunivariatemodels,weplotted
betterfitthanthelinearmodel.
theestimatedslopefactorloadings.Therelativesizeofeachfactor
The linear model for adolescents’ eating disorder symptoms
providedagoodfitforthedata,(cid:3)2(13,N(cid:1)487)(cid:1)22.84,p(cid:1).03;
CFI(cid:1).98;TLI(cid:1).98;RMSEA(cid:1).04(90%CI(cid:1).01,.07).The 3Severalfactorsmightexplainthenegativeintercept–slopecorrelations
unspecifiedmodelofgirls’eatingsymptomatologyalsoprovided observed here. First, the correlation between the intercept and slope de-
atenablefit,(cid:3)2(9,N(cid:1)487)(cid:1)17.28,p(cid:1).001;CFI(cid:1).99;TLI(cid:1) pends on the time at which initial status is estimated. In growth curve
.98;RMSEA(cid:1).05(90%CI(cid:1).01,.07).However,thechi-square analyses,adifferentcorrelationbetweeninitialstatusandchangecanbe
difference test was not significant (22.84–17.28 (cid:1) 5.56, df (cid:1) 4, obtained depending on the time of initial status (cf. S. C. Duncan &
p (cid:3) .05), indicating that the unspecified model did not offer a Duncan, 1996; Rogosa, 1988). Second, regression to the mean could
superiorfit. produce these negative correlations, in which extremely high (and low)
interceptvaluesthatarepartiallyafunctionofmeasurementerrorarelikely
Thelinearmodelforgirls’antisocialbehaviorprovidedagood
fit for the data, (cid:3)2(13, N (cid:1) 485) (cid:1) 11.60, p (cid:1) .17; CFI (cid:1) .99; tobeclosertothesamplemeanatrepeatassessment(seeMarsh,Craven,
Hinkley, & Debus, 2003, for a similar suggestion). Third, the negative
TLI (cid:1) .99; RMSEA (cid:1) .02 (90% CI (cid:1) .00, .06). However, the
intercept–slope correlations may be a product of actual developmental
unspecified model of girls’ antisocial behavior appeared to im- deceleration.Girlsatthehigherlevelsmayhaveinitiatedtheirsymptom-
provethefit,(cid:3)2(9,N(cid:1)485)(cid:1)2.23,p(cid:1).52;CFI(cid:1)1.00;TLI(cid:1) atic behavior at earlier ages and were either remitting or escalating at
1.00;RMSEA(cid:1).00(90%CI(cid:1).00,.06).Indeed,theunspecified slowerrates.
530 MEASELLE,STICE,ANDHOGANSEN
Table2
ParameterEstimates,StandardErrors,andCriticalRatiosforUnivariateGrowthModels
Linearmodelresults Nonlinearmodelresults
Depression Eatingdisorder Antisocialbehavior Substanceabuse
Critical Critical Critical Critical
Parameter Est. SE ratio Est. SE ratio Est. SE ratio Est. SE ratio
Factormean
INTatage13 1.32 .02 67.91 0.50 .02 20.94 1.73 .03 51.21 0.12 .01 9.01
Slope 0.03 .01 4.93 0.04 .01 5.32 (cid:5)0.04 .01 (cid:5)5.15 0.01 .00 2.49
Factorvariance
INTatage13 0.10 .01 7.72 0.17 .02 8.99 0.34 .05 7.55 0.04 .01 2.05
Slope 0.01 .00 5.97 0.01 .00 5.44 0.01 .00 3.05 0.02 .00 2.05
Factorcorrelation:INTwithslope (cid:5)0.41 .09 (cid:5)4.29 (cid:5)0.61 .15 (cid:5)3.86 (cid:5)0.55 .13 (cid:5)4.19 (cid:5)0.38 .09 (cid:5)3.85
Residualvariance
Age13 0.01 .01 1.27 0.06 .01 5.25 0.09 .04 2.31 0.03 .01 3.31
Age14 0.05 .01 8.42 0.06 .01 7.33 0.09 .03 3.95 0.04 .01 7.25
Age15 0.07 .01 11.71 0.07 .01 11.26 0.15 .02 9.48 0.05 .00 10.82
Age16 0.06 .01 10.38 0.05 .01 11.01 0.17 .01 11.47 0.05 .01 9.77
Age17 0.06 .01 7.68 0.05 .01 7.74 0.08 .02 4.00 0.03 .01 4.86
Age18 0.09 .02 4.65 0.08 .02 5.13 0.04 .03 1.31 0.01 .02 1.57
Note. Critical ratios of 1.96 indicate estimates are significant at p (cid:4) .05. Est. (cid:1) unstandardized model estimate, including factor mean, variance, and
correlation; SE (cid:1) standard error of each estimated parameter; Critical ratio (cid:1) estimate divided by the standard error; z statistic was used to establish
statisticalsignificanceofparameter;INT(cid:1)intercept.
loading reflects the pattern of developmental change across the opmental factors from the different symptom domains. These
5-year span. The graphs for depression, eating pathology, antiso- correlationsarepresentedinTable3.
cial behavior, and substance abuse are presented in Figure 1. All four intercepts were significantly associated with one an-
Because the factor loadings for the slope for antisocial behavior other;girlswithhigherinitiallevelsinonedomaintendedtohave
were inverted because of our positive scaling of the slope factor, higher initial levels in the other domains. All four slope factors
we reestimated the model by using a negatively scaled slope for were also significantly and positively correlated, indicating, with
antisocial behavior. These graphs suggest that girls’ depression one exception, that girls showing growth in one domain tended
and eating disorder symptoms grew at fairly constant rates be- alsotoshowgrowthinotherdomains.Theexceptionwastheslope
tweenage13andage18.Thegraphforgirls’antisocialbehavior correlationsinvolvinggirls’antisocialtrajectories;here,apositive
suggests that the trajectory was relatively flat across the first 3 correlation signified that greater increases in depression, eating
years but then declined more rapidly after age 15. The graph for
disorder, and substance abuse symptoms corresponded with less
substance abuse symptoms indicates fairly rapid escalations in
rapiddecreasesinantisocialbehavioronaverage.
self-reportedsubstanceabusebetweenage13andage16,followed
Nextwereparameterizedtheassociativemodeltotestwhether
by less rapid growth and even decelerations in the level of sub-
initiallevelsineachsymptomdomainpredictedfuturegrowthin
stanceabusearoundage18.
other symptom domains. To test the unique predictive effects of
eachsymptomdomainatbaseline,weregressedeachslopefactor
Temporal Relations Between Symptom Domains
onallfourintercepts.Asexpected,giventhesignificantintercept–
Toexaminethedevelopmentalrelationsamongthefoursymp- slopecorrelationswithinsymptomdomains,initialsymptomlev-
tomdomains,wetestedanassociativeLGMinwhichthegrowth elswerepredictiveofsignificantchange((cid:4)(cid:1)(cid:5).57to(cid:5).33,ps(cid:4)
factorsforallfoursymptomdomainswereintercorrelated.Addi- .001) in the same domain. Consistent with the associations be-
tionally, when specifying the antisocial behavior and substance tween intercepts and slopes within a given symptom domain,
abuse symptom portions of the associative model, we used the higher baseline levels predicted less rapid symptom change over
actual slope/shape scaling values generated by the unspecified time.Thedirectionofeffectwasmarkedlydifferentwhenpredict-
univariate models as start values for their respective slope/shape ing across domains while controlling for the effects of all other
factors.4 The associative model fit the data well, (cid:3)2(200, N (cid:1)
483)(cid:1)639.59,p(cid:4).001;CFI(cid:1).95;TLI(cid:1).96;RMSEA(cid:1).05
(90%CI(cid:1).03,.08).Parameterestimatesfortheassociativemodel 4We used the actual slope estimates from the unspecified univariate
modelsasstartvaluesinourassociativemodelbecausewefeltthathaving
are presented in Table 3. Although the results of the associative
thesamedegreesoffreedomforeachsymptomtrajectorywouldallowfor
model are not exactly identical to the univariate LGM results in
a more interpretable test of the temporal associations among symptom
terms of factor loadings, factor means and variances, and factor
domains.Thiswasnotarequirement,however,sowealsocomputedthe
correlations,theytendtobeverysimilar(S.C.Duncan&Duncan, associative model (and second-order models below) using the same un-
1996). Because this was the case with our data as well, in this specifiedstructuremodeledintheunivariatesection.Theresultsfromboth
sectionwefocusprimarilyonthecorrelationsbetweenthedevel- approacheswereessentiallyidenticalintermsofabsoluteandrelativefit.
FEMALECOMORBIDITYTRAJECTORIES 531
symptom domains. Initial depression level predicted increases in
eating pathology ((cid:4)(cid:1) .37, p (cid:4) .05) and substance abuse symp-
toms ((cid:4) (cid:1) .40, p (cid:4) .01) and slower decreases in antisocial
behavior ((cid:4)(cid:1) .51, p (cid:4) .01). Initial eating pathology level pre-
dictedincreasesinsubstanceabusesymptoms((cid:4)(cid:1).36,p(cid:4).05).
Initiallevelsofantisocialsymptomspredictedincreasesindepres-
sive ((cid:4)(cid:1) .39, p (cid:4) .05) and substance abuse ((cid:4)(cid:1) .56, p (cid:4) .01)
symptoms. Finally, initial substance abuse level predicted slower
decelerations in antisocial behavior ((cid:4)(cid:1) .45, p (cid:4) .01). Thus,
Table3
ParameterEstimates,StandardErrors,andCriticalRatiosfor
theAssociativeGrowthModel
Critical
Parameter Est. SE ratio
Factormean
Depressionintercept 1.32 .02 69.80
Eatingdisorderintercept 0.51 .03 17.60
Antisocialbehaviorintercept 1.71 .04 42.82
Substanceabuseintercept 0.11 .02 5.87
Depressionslope 0.03 .01 4.67
Eatingdisorderslope 0.04 .01 5.28
Antisocialbehaviorslope (cid:5)0.03 .01 (cid:5)3.36
Substanceabuseslope 0.01 .01 2.34
Factorvariance
Depressionintercept 0.10 .01 7.81
Eatingdisorderintercept 0.17 .02 9.11
Antisocialbehaviorintercept 0.37 .04 9.29
Substanceabuseintercept 0.08 .01 8.51
Depressionslope 0.01 .00 6.53
Eatingdisorderslope 0.01 .01 5.40
Antisocialbehaviorslope 0.01 .00 4.56
Substanceabuseslope 0.02 .00 3.04
Intercept/slopecorrelationwithinsymptom
domain
Depression (cid:5).42 .08 (cid:5)5.22
Eatingdisorder (cid:5).61 .15 (cid:5)4.07
Antisocialbehavior (cid:5).54 .12 (cid:5)4.23
Substanceabuse (cid:5).39 .09 (cid:5)4.37
Interceptcorrelationbetweendomains
DepressionWITH
Eatingdisorder .67 .06 10.61
Antisocialbehavior .59 .06 8.52
Substanceabuse .47 .07 6.75
EatingdisorderWITH
Antisocialbehavior .49 .05 9.18
Substanceabuse .28 .06 4.54
AntisocialbehaviorWITH
Substanceabuse .67 .07 9.09
Slopecorrelationbetweendomains
DepressionWITH
Eatingdisorder .52 .07 7.41
Antisocialbehavior .41 .10 4.83
Substanceabuse .39 .09 4.39
EatingdisorderWITH
Antisocialbehavior .37 .08 5.02
Substanceabuse .38 .08 4.75
AntisocialbehaviorWITH
Substanceabuse .47 .11 4.32
Figure 1. Estimated slope factor loadings from the univariate latent
growth models for depression, eating disorder, antisocial, and substance Note.Criticalratiosof1.96indicateestimatesaresignificantatp(cid:4).05.
abusesymptoms. Est. (cid:1) unstandardized model estimate, including factor mean, variance,
andcorrelation;SE(cid:1)standarderrorofeachestimatedparameter;Critical
ratio (cid:1) estimate divided by the standard error; z statistic was used to
establish statistical significance of parameter; WITH (cid:1) correlation be-
tweenfactors(e.g.,depressioninterceptandeatingdisorderslope).
532 MEASELLE,STICE,ANDHOGANSEN
several of the symptom domains acted as unique risk factors for Table4
growth in other domains, and patterns of unidirectional and bidi- ParameterEstimates,StandardErrors,andCriticalRatiosfor
rectionalprospectiveeffectsemerged. theTwo-FactorHigherOrderFactor-of-CurvesModel
Critical
Hierarchical Structure of Co-Occurring Trajectories
Parameter Est. SE ratio
Ourfinalgoalwastoinvestigatethehierarchicalfactorstructure Factorloading
of the associative model. Initially we fit a single pair of higher InternalizingINT
order growth parameters to the associative model by using the Depression 1.00 — —
Eatingdisorder 0.68 0.07 9.71
“factor-of-curve” modeling procedure (McArdle, 1988). In the
Internalizingslope
factor-of-curvemodel,oneexamineswhetherahigherorderfactor
Depression 1.00 — —
adequatelydescribesrelationshipsamonglowerordergrowthfac- Eatingdisorder 0.67 0.09 7.44
tors (T. E. Duncan et al., 1999; McArdle, 1988). As such, the ExternalizingINT
factor-of-curve model is a second-order extension of the associa- Antisocialbehavior 1.00 — —
Substanceabuse 1.60 0.15 10.67
tivemodel,inwhichthesecond-orderinterceptandslopefactors
Externalizingslope
are estimated from the first-order univariate latent factors. Thus, Antisocialbehavior 1.00 — —
eachfirst-orderportionoftheassociativemodeldescribesindivid- Substanceabuse 1.36 0.22 6.21
ual differences within each univariate symptom domain, and the Factormean
InternalizingINT 1.32 0.02 65.62
second-order common factor model describes the individual dif-
Internalizingslope 0.03 0.01 3.22
ferences among the first-order LGMs. To specify the factor-of-
ExternalizingINT 1.72 0.04 43.75
curve model, the covariances among first-order latent growth Externalizingslope (cid:5)0.03 0.01 (cid:5)3.41
curves’variancesarefixedtoavalueofzero,andfactorloadings Factorvariance
between the first- and second-order factors are restricted to be InternalizingINT 0.13 0.02 6.71
Internalizingslope 0.01 0.00 5.13
equalovertimeforeachsymptomdomain;thisimposesaform
ExternalizingINT 0.43 0.06 6.91
of factorial invariance that ensures similar scaling among mul- Externalizingslope 0.02 0.00 4.80
tiplesecond-orderfactorscores(T.E.Duncanetal.,1999;also Factorintracorrelation
see McArdle, 1988, for a more formal discussion of the math- Internalizing (cid:5).67 .06 (cid:5)10.87
Externalizing (cid:5).59 .11 (cid:5)5.36
ematical representation of this model). We used girls’ depres-
Factorintercorrelation
sion as the reference scaling for the single higher order factor
INTwithINT .57 .05 9.40
model. Slopewithslope .41 .09 6.11
Thesinglehigherordermodelfitthedatapoorly,(cid:3)2(221,N(cid:1) Factorintercept
485)(cid:1)906.25,p(cid:4).001;CFI(cid:1).88;TLI(cid:1).89;RMSEA(cid:1).11 DepressionINT 0.00 — —
(90%CI(cid:1).08,.13).Assuch,thedatadidnotsupporttheideathat DEaetpinregssdiiosnorsdleorpeINT (cid:5)00..0403 0—.10 (cid:5)—4.53
girls’ trajectories in each of the four symptom domains could be Eatingdisorderslope 0.03 0.01 3.17
explainedbyacommonsetofgrowthfactorsacrosstheseyearsof AntisocialbehaviorINT 0.00 — —
adolescence. Antisocialbehaviorslope 0.00 — —
SubstanceabuseINT (cid:5)2.18 0.26 (cid:5)8.38
We then tested our a priori two-factor higher order model in
Substanceabuseslope 0.34 0.03 11.91
which we estimated separate intercept and slope factors on the
Residualvariance
basisofwhatdepressionandeatingpathologyhadincommonand DepressionINT (cid:5)0.01 0.01 (cid:5)1.49
whatantisocialbehaviorandsubstanceabusehadincommon.We Depressionslope 0.00 0.00 4.17
used girls’ depression and antisocial behavior as the reference EatingdisorderINT 0.04 0.01 6.60
Eatingdisorderslope 0.00 0.00 0.03
scaling for what we labeled the internalizing and externalizing
AntisocialbehaviorINT 0.01 0.02 1.14
second-order factors, respectively. The two-factor model fit the Antisocialbehaviorslope (cid:5)0.00 0.00 (cid:5)1.47
data reasonably well, (cid:3)2(214, N (cid:1) 485) (cid:1) 682.29, p (cid:4) .001; SubstanceabuseINT 0.56 0.08 6.83
CFI(cid:1).93;TLI(cid:1).93;RMSEA(cid:1).06(90%CI(cid:1).04,.07). Substanceabuseslope 0.04 0.01 5.59
Factorloadings,interceptandslopemeansandvariances,factor
Note. Criticalratiosof1.96indicateestimatesaresignificantatp(cid:4).05.
correlations,factorintercepts,andresidualvariancesarepresented
Internalizing factor comprises depression and eating disorder symptoms.
inTable4.Parameterestimatesforthetwo-factormodelindicated Externalizing factor comprises antisocial behavior and substance abuse
that both intercept factors were significantly different from zero, symptoms.Dashesindicatedataarenonapplicablebecauseestimateswere
thatthemeansofbothslopefactorswerealsosignificant,andthat fixed to scale the higher order factors. Est. (cid:1) unstandardized model
estimate,includingfactormean,variance,andcorrelation;SE(cid:1)standard
theamountofindividualvariationaroundbothinterceptandboth errorofeachestimatedparameter;Criticalratio(cid:1)estimatedividedbythe
slope factors was significant. In addition, all common factor-of- standard error; z statistic was used to establish statistical significance of
curve loadings were significant. The higher order internalizing parameter;INT(cid:1)intercept.
interceptfactoraccountedforapproximately75%and63%ofthe
variation in the first-order intercepts for depression and eating factorsfordepressionandeatingdisordersymptoms,respectively,
disorder symptoms, respectively, whereas the externalizing inter- and65%and59%ofthevariationinthefirst-orderslopefactors
ceptfactoraccountedfor79%and95%ofthefirst-orderintercepts forantisocialproblemsandsubstanceabusewereaccountedforby
forantisocialbehaviorandsubstanceabuse,respectively.Approx- the internalizing and externalizing higher order slope factors,
imately 73% and 93% of the variation in the first-order slope respectively.
FEMALECOMORBIDITYTRAJECTORIES 533
Finally, the internalizing and externalizing intercept factors Discussion
werepositivelyandsignificantlycorrelated(r(cid:1).57,p(cid:4).05),as
Single-Syndrome Trajectories
were the two slope factors (r (cid:1) .41, p (cid:4) .01). Thus, girls with
higherinitialinternalizingsymptomscoreshadelevatedexternal- The first aim of this study was to generate descriptive data on
izingsymptomscoresaswell.Inlightofthenegativeexternalizing depressive,eatingdisorder,antisocial,andsubstanceabusesymp-
slope, growth in internalizing was significantly associated with tomtrajectoriesinacommunitysampleoffemaleadolescents.Our
slowerdeclinesinexternalizingproblems. results suggest that between the ages of 13 and 18, girls’ self-
On the whole, it might appear odd that the same set of higher reported symptoms of depression and eating disorder symptom
ordergrowthfactorscouldaccountforthefirst-ordervariationin trajectoriesshowedevidenceoflineargrowthaswellassignificant
girls’ antisocial and substance abuse symptoms given that the levels of variability around the aggregate trends. The rate of
respectiveaggregatetrendswerechanginginoppositedirections. growthingirls’depressivesymptomswassignificantalbeitmod-
est,increasingonaverageaboutaquarterofastandarddeviation
However, it should be remembered that there was significant
between age 13 and age 18. These results dovetail with other
variationintheslopesforsubstanceabuseandantisocialbehavior,
studies that have reported sample-level increases in depressive
with some girls likely showing increases and some showing de-
symptomsinfemaleadolescents(Coleetal.,2002;Hankinetal.,
creases in both symptom domains. In an attempt to interpret this
1998).
finding,weusedtheestimatedfactorloadingsateachagetoplot
Girls’eatingdisordersymptomtrajectoriesalsoshowedsignif-
eachgirl’santisocialbehaviorandsubstanceabusetrajectories;we
icant sample-level growth as well as significant individual vari-
thenexaminedthefrequencywithwhichgirls’symptomdevelop-
abilityintherateofgrowth.Likethegrowthindepressivesymp-
mentinthesetwodomainsmovedinsimilarordifferentdirections.
toms,thegrowthingirls’eatingdisordersymptomstendedtobe
We found that 31% of the sample had downward antisocial be-
fairly constant over time. However, in contrast to the depressive
havior trajectories and upward substance abuse trajectories (no
symptom trajectory, the rate of growth in girls’ eating disorder
caseoftheoppositewasfound),44%ofthesamplehadsynchro-
symptomatologywasmorepronounced,increasingmorethanhalf
nous antisocial behavior and substance abuse trajectories (e.g.,
astandarddeviationbetweenages13and18.Toourknowledge,
both increased or decreased), and 25% had one stable trajectory
this study and prior investigations of this sample (Stice et al.,
and a second trajectory that moved either upward or downward.
2004) may be the first to report normative increases in eating
Thus,bymovingfromanomotheticfocusongroup-leveltrendsto
disordersymptomsforfemaleadolescents.
an ideographic focus on individuals’ patterns of codevelopment,
Girls’ substance abuse trajectory also increased significantly
wecanseethattherewasreasonablesynchronyinhowindividual
albeitmodestlyduringthisperiod(e.g.,approximatelyaquarterof
girls’antisocialbehaviorandsubstanceabusesymptomschanged
astandarddeviationbetweenages13and18).Again,theamount
fromage13to18,despitecontradictorysample-levelestimates.It
of variability around the sample-level growth trajectory was sig-
isoursensethatthisdevelopmentalsynchronywascapturedbythe
nificant, indicating considerable individual differences in sub-
second-order slope factor for girls’ antisocial behavior and sub-
stanceabuseprogression.Incontrasttogirls’depressiveandeating
stanceabusesymptoms.
disordersymptomtrajectories,growthinthisdomaintendedtobe
To evaluate further the acceptability of this internalizing–
somewhat uneven or nonlinear. Specifically, the shape of the
externalizing higher order LGM, we tested a series of alternate
substanceabusetrajectorysuggestedthatgirlsreportedmorerapid
higherorderLGMs.Forexample,depression,eatingdisorders,and
increasesinsubstanceabuseproblemsbetweentheagesof13and
substance abuse were grouped to reflect the idea that girls with
16 but then appeared to begin to decelerate between 16 and 18
mood problems might rely on two different types of mood regu- years.Thispatternofincreaseingirls’substanceabusesymptoms
lation strategies. In this particular model, only the first-order is consistent with other investigations and is likely linked to less
growth factors were needed for antisocial behavior. None of our restrictivelivingsituationsaswellassomeoftherisksassociated
alternate higher order models provided an acceptable accounting with greater tolerance of substance use (Chassin & Ritter, 2001;
ofthedata.5 Chen & Kandel, 1995; Hankin et al., 1998; Kandel, Huang, &
We also tested a second multivariate approach proposed by Davies,2001;O’Malleyetal.,2000).Wearelesscertainaboutthe
McArdle(1988),the“curve-of-factor”model(seealsoT.E.Dun- factorsthatmightaccountforgirls’apparentslowingofsubstance
can et al., 1999). In contrast to the factor-of-curve model tested abusebehaviorbetweenages16and18.Perhaps,thisdeclinecan
above, the curve-of-factor model fits a growth curve to factor be linked to girls’ preparation for their transition to adult roles
scoresrepresentingwhatdifferentsymptomdomainshaveincom- (e.g.,college,marriage,employment),whichhavebeenshownto
mon at each time point. The observed variables for all symptom reduce substance abuse symptoms in young adults (Collins &
domains at each age are factor analyzed to produce factor scores Shirley,2001).
(e.g.,afactorscorethatcompriseswhatiscommontodepression,
eating disorder, antisocial, and substance abuse symptoms at age
13),whicharethenusedformodelinghigherordergrowthcurves 5Additionally,wewereconcernedthatoursecond-orderresultsmight
be due to method artifact, namely that the internalizing factor simply
(T.E.Duncanetal.,1999;Hancocketal.,2001;McArdle,1988).
represented interview data whereas the externalizing factor represented
The curve-of-factor alternative to our internalizing and external-
questionnaire data. Fortunately, we replicated our second-order results
izing factor-of-curve model provided a poor accounting of the
whenwereplacedtheK-SADdepressionsymptomcompositewithitems
data,(cid:3)2(202,N(cid:1)485)(cid:1)956.58,p(cid:4).001;CFI(cid:1).84;TLI(cid:1).85;
fromBussandPlomin’s(1984)EmotionalityScale,whichgirlscompleted
RMSEA (cid:1) .12 (90% CI (cid:1) .11, .13); consequently, we did not inquestionnaireformatallassessmentstoreportontheirnegativeaffect
exploreothercurve-of-factoralternatives. andmoodtraits.