Genetic and Environmental Bases of Childhood Antisocial - cnpru

17 Pages·2007·0.38 MB·English

JournalofAbnormalPsychology Copyright2007bytheAmericanPsychologicalAssociation
2007,Vol.116,No.2,219–235 0021-843X/07/$12.00 DOI:10.1037/0021-843X.116.2.219
Genetic and Environmental Bases of Childhood Antisocial Behavior: A
Multi-Informant Twin Study
Laura A. Baker Kristen C. Jacobson
UniversityofSouthernCalifornia UniversityofChicago
Adrian Raine, Dora Isabel Lozano, Serena Bezdjian
UniversityofSouthernCalifornia
Genetic and environmental influences on childhood antisocial and aggressive behavior (ASB) during
childhoodwereexaminedin9-to10-year-oldtwins,usingamulti-informantapproach.Thesample(605
families of twins or triplets) was socioeconomically and ethnically diverse, representative of the
culturallydiverseurbanpopulationinSouthernCalifornia.MeasuresofASBincludedsymptomcounts
forconductdisorder,ratingsofaggression,delinquency,andpsychopathictraitsobtainedthroughchild
self-reports,teacher,andcaregiverratings.MultivariateanalysisrevealedacommonASBfactoracross
informants that was strongly heritable (heritability was .96), highlighting the importance of a broad,
general measure obtained from multiple sources as a plausible construct for future investigations of
specific genetic mechanisms in ASB. The best fitting multivariate model required informant-specific
genetic,environmental,andratereffectsforvariationinobservedASBmeasures.Theresultssuggestthat
parent, children, and teachers have only a partly “shared view” and that the additional factors that
influencethe“rater-specific”viewofthechild’santisocialbehaviorvaryfordifferentinformants.This
isthefirststudytodemonstratestrongheritableeffectsonASBinethnicallyandeconomicallydiverse
samples.
Keywords:antisocialbehavior,aggression,genes,environment
Why do some children grow up to be prosocial, law-abiding early adoption studies in both Scandinavia and the United States
individuals,whereasothersengageinpatternsofdisruptive,defi- haveprovidedtheintriguingfindingthatnotonlydoestheriskfor
ant,anddelinquentbehavior,evenfallingintothecriminaljustice adult criminal offending run in families but familial similarity is
systemwellbeforereachingadulthood?Aplethoraofstudieshave due primarily to shared genetic risk (Bohman, 1978; Cadoret,
investigatedtheetiologyofsuchindividualdifferences,withabun- 1978;Hutchings&Mednick,1971;Loehlin,Willerman,&Horn,
dantevidencedemonstratingtheimportanceofbothsocialcircum- 1985; Sigvardsson, Cloninger, Bohman, & Von Knorring, 1982).
stancesandbiologicalriskfactorsinantisocialbehavioracrossthe Geneticpredispositionshavealsobeenshowntoplayasignificant
life span (Baker, 1999; Raine, 1993, 2002; Raine, Brennan, Far- roleinthenormalvariationinadultaggressivebehavior,perhaps
rington, & Mednick, 1997; Stoff, Breiling, & Maser, 1997). especially in more impulsive forms (Coccaro, Bergeman,
Among these risk factors, genetic and environmental influences Kavoussi, & Seroczynski, 1997). In contrast, studies that have
havebeenofconsiderableinterestandarelikelytoplayakeyrole included adolescents and younger children vary widely in their
inourunderstandingofaggressionandotherantisocialbehaviors estimates of the relative importance of genes and environment,
and,thus,ourabilitytoavertthem. withheritabilityestimates(h2)indicatingthatgeneticeffectscould
Infact,geneticandenvironmentalinfluencesinaggressiveand
explainaslittleasnilorupwardofthreefourthsofthevariancein
antisocialbehavior(ASB)havebeenstudiedextensively.Several
ASB(seeRhee&Waldman,2002,forthemostrecentreview).
Usingmeta-analysisofkeybehavioralgeneticstudiesinASB,
Rhee and Waldman (2002) found that, combining results across
studies,thereweresignificanteffectsofadditivegeneticinfluence
LauraA.Baker,AdrianRaine,DoraIsabelLozano,andSerenaBezd-
jian,DepartmentofPsychology,UniversityofSouthernCalifornia;Kristen (a2 (cid:1) .32), of nonadditive genetic influences (d2 (cid:1) .09), and of
C.Jacobson,DepartmentofPsychiatry,UniversityofChicago. shared(e2 (cid:1).16)andnonsharedenvironment(e2 (cid:1).43).These
s ns
This study was supported by National Institute of Mental Health genetic and environmental effects were found to differ, however,
(NIMH) Grant NIMH R01 MH58354 to Laura A. Baker and NIMH accordingtothedefinitionandmethodofassessingASB,aswell
IndependentScientistAwardK02MH01114-08toKristenJacobson.We asbytheageatwhichASBwasstudied.Thenonadditivegenetic
wishtothanktheUniversityofSouthernCaliforniatwinprojectstaff,for
effects appear most strongly for studies of criminal convictions
assistanceindatacollectionandscoring,andthetwinsandtheirfamilies,
comparedwithallotherdefinitionsofASB.Sharedenvironmental
fortheirparticipationinthisresearch.
effects were stronger for parental reports of ASB compared with
CorrespondenceconcerningthisarticleshouldbeaddressedtoLauraA.
Baker,DepartmentofPsychology,UniversityofSouthernCalifornia,Los self-reports and with official records, and these shared environ-
Angeles,CA90089-1061.E-mail:[email protected] mentaleffectsappeartodiminishfromchildhoodtoadulthood.
219
220 BAKER,JACOBSON,RAINE,LOZANO,ANDBEZDJIAN
It is also noteworthy, however, that age and method of assess- ferent types of antisocial behavior may be a useful method in
ment are confounded across studies—investigations of younger moleculargeneticresearch.
childrentendtorelyonparentorteacherreports,whereasstudies
ofolderadolescentsandadultsaremoreapttouseofficialrecords Informant Variation
or self-report measures of ASB. Thus, the larger effect of shared
environment during childhood may be due to greater reliance on Another important aspect to consider when comparing results
parentalorteacherratings.Giventhesemethodologicalconfounds across studies is the source of the information about ASB. It is
acrossstudies,itisimpossibletoknowthestrengthofgeneticand well-knownthatdifferentinformantsproducedifferentreportsofa
environmental influences on individual differences in childhood child’sbehavior.Correlationsbetweenratersofthesamechildare
ASB in particular. Additional studies are required to resolve the typicallyabout.60betweenmotherandfatherratings,.28between
effectsofgenesandenvironmentinASBinchildren. parent and teacher ratings, and .22 between the parent and child
ratings (Achenbach, McConaughy, & Howell, 1987). Largely,
each rater provides a unique perspective on the child’s behavior.
Defining Antisocial Behavior Children would seem to be the most knowledgeable source to
reportontheirownbehavior(particularlycovertactions)aswellas
Definitions of ASB vary widely across studies and include
their motivations, although their cognitive development, truthful-
violations of rules and social norms (e.g., lawbreaking), various
ness,andsocialdesirabilityfactorsmaylimittheaccuracyoftheir
formsofaggression(e.g.,self-defenseorotherreactiveformsand
reports.Parentsmaybemoreabletoobjectivelyreportonachild’s
proactive behaviors such as bullying), and serious patterns of
externalizing behaviors, although they may be unaware of covert
disruptive and aggressive behavior such as those observed in
actionsorunwillingtoreportthemtoresearchers.Althoughteach-
clinical disorders like conduct disorder and oppositional defiant
ers’ reports may also have the advantage of greater objectivity,
disorderinchildrenorantisocialpersonalitydisorderinadults.The
teachers may have limited knowledge of the child’s antisocial
variability found in the definitions of these key concepts is also
behavior,particularlyasitmayoccuroutsideofclassroomorother
found in the methods of measuring ASB; some studies are based
school settings. Although researchers sometimes combine ratings
on official records such as police arrests, court convictions, or
acrossreportersinanattempttoincreasescalereliability,different
schoolrecords,whereasothersrelyonbehavioralratingsprovided etiologies may exist for scales derived from different informants
by parents or teachers or on self-reports about the participant’s (Bartels et al., 2003, 2004; Saudino & Cherny, 2001). Thus, the
ownASB.Eachassessmentmethodhasitsadvantagesanddisad- bestwaytomodelinformationfrommultipleinformantsistouse
vantages with no one definition or method of assessment being a multivariate, factor-based approach that allows for both differ-
clearlysuperior. encesandcorrelationsacrossinformantssimultaneously(Kraemer
Nevertheless, in spite of the wide variations in definitions of etal.,2003).
ASB, as well as the possibility that the relative importance of There are at least three advantages to using a factor-based
geneticandenvironmentalfactorsmayvaryfordifferentmeasures approach when dealing with multiple informants in twin studies.
(e.g., Eley, Lichtenstein, & Moffitt, 2003; Mednick, Gabrielli, & First, such a model allows for the possibility that there may be
Hutchings, 1984; also see Rhee & Waldman, 2002, for review), differentgeneticandenvironmentaletiologiesdependinguponthe
there is also considerable evidence for a general externalizing perspectiveoftherater.Second,itallowsonetoexplicitlymodel
dimensionofproblembehaviorunderlyingthesevariousbehaviors andtestforthesignificanceofcertaintypesofraterbias.Finally,
and tendencies. Similar to the problem/behavior syndrome de- because the underlying common factor will represent (by defini-
scribed earlier by Jessor and Jessor (1977), a broad latent factor tion)a“sharedview”ofantisocialbehavioracrossinformants,the
hasbeenpurportedtobeacommonlinkamongantisocialbehav- heritability of the common factor may be higher than the herita-
ior, substance dependence, and disinhibited personality traits bilitiesobtainedthroughanyoneinformant.Ifthisisthecase,then
(Kruegeretal.,2002).Theexternalizingdimensionhasbeenfound combininginformationfromdifferenttypesofreportersmayyield
to be more continuous than categorical, with shades of gray de- stronger genetic signals in molecular genetic studies. Previous
scribing a range of deviant behaviors across individuals (Markon studies of ASB in preadolescent children have relied heavily on
& Krueger, 2005; Young, Stallings, Corley, Krauter, & Hewitt, either parent or teacher reports, although a few studies have
2000). Moreover, this common externalizing factor has been obtained data from multiple reporters, most commonly from the
showntohaveastrongheritabilityamongadolescents(h2(cid:1).80), mother and the father (e.g., Bartels et al., 2003, 2004; Neale &
accountingformuchofthecovariationamongvariousaspectsof Stevenson,1989;Hewitt,Silbert,Neale,Eaves,&Erickson,1992)
antisocial behavior and disinhibition (Krueger et al., 2002). orfromparent(s)andteachers(e.g.,Hudziaketal.,2003;Martin,
Among adults, there is also evidence for separate genetic factors Scourfield, & McGuffin, 2002; Vierikko, Pulkkinen, Kaprio, &
forinternalizingversusexternalizingdimensionsofpsychopathol- Rose, 2004) and occasionally from parents and children (e.g.,
ogy (Kendler, Prescott, Myers, & Neale, 2003). This general Simonoffetal.,1995).Weareunaware,however,ofanypublished
externalizing factor found across many studies may reflect an studiesofexternalizingdisorderthathaveusedreportsfromcare-
overalltendencytoactinanunconstrainedmanner,agenetically givers,teachers,andchildrensimultaneously.
basedcharacteristicthatmanifestsitselfinvariouswaysdepending
ontheenvironment(Krueger,2002).Thehigherheritabilityfound
Sex Differences
for this externalizing factor compared with heritabilities obtained
from studies that have focused on only one type of antisocial Afinalquestiontoconsideriswhethertherearesexdifferences
behavior suggests that using a composite measure based on dif- intherelativeimportanceofgeneticandenvironmentalfactorsfor
CHILDHOODASB 221
antisocial behavior. In spite of the fact that males are far more single viewpoint. Fourth, our sample consisted of both male and
likely than females to engage in antisocial, aggressive, and crim- femaletwins,includingopposite-sexpairs,allowingustoexamine
inalbehavior,therearenoapparentdifferencesbetweenthesexes potentialsexdifferencesintheetiologyofasharedviewofASB.
in the relative importance of genetic factors (i.e., heritability) in Finally, it should be noted that although the present results are
explaining individual differences in antisocial behavior among cross-sectional, they are part of a larger, ongoing longitudinal
adults. Heritability of liability toward nonviolent criminality ap- study. Therefore, in future analyses, we will be able to compare
pears equivalent for men and women, in studies of both twins and contrast our results as participants move from the brink of
(Cloninger&Gottesman,1987)andadoptees(Baker,Mack,Mof- adolescenceintoadolescenceandyoungadulthood.
fitt, & Mednick, 1989), although the average genetic predisposi-
tionsdoappeargreaterforcriminalwomencomparedwithcrim-
Method
inal men (Baker et al., 1989; Sigvardsson et al., 1982). A few
studies of childhood and adolescent ASB have examined sex Overview of the USC Twin Study of Risk Factors for
differences in genetic and environmental etiology, although the Antisocial Behavior
resultsarenotconsistent.Somestudieshavefoundgeneticeffects
to be of greater importance in boys and common environment TheUSCTwinStudyofRiskFactorsforAntisocialBehavioris
moreimportantingirlsusingparentalratings(Silbergetal.,1994), a longitudinal study of the interplay of genetic, environmental,
whereasothershavefoundtheoppositeresultusingretrospective social, and biological factors on the development of antisocial
reportsforadolescents(Jacobson,Prescott,&Kendler,2002),and behavior across adolescence. The first wave of assessment oc-
still others have not found sex-specific etiologies (Eley, Lichten- curredduring2001to2004,whenthetwinswere9to10yearsold,
stein, & Stevenson, 1999). Aggregating across studies in their with a 2-year follow-up assessment in the laboratory when twins
meta-analysis, Rhee and Waldman (2002) found that the relative wereages11to12.Twoadditionalfollow-upassessmentswillbe
importanceofgeneticandenvironmentalfactorsinASBdoesnot conductedwhenthetwinsareages14to15(thirdwave)and16to
differformalesandfemales,althoughitshouldbenotedthattheir 17yearsold(fourthwave).Thepresentanalysesarebasedondata
analysesdidnotinvestigatetheextenttowhichsexdifferencesin fromthefirstwave.Comprehensiveassessmentofeachchildwas
etiologymightvaryacrossdevelopmentormethodofassessment made, including cognitive, behavioral, psychosocial, and psycho-
(i.e.,rater).Overall,thequestionaboutdifferentetiologiesofASB physiologicalmeasuresbasedonindividualtestingandinterviews
formalesandfemalesremainsopen. ofthechildandprimarycaregiverduringthelaboratoryvisit,with
additional teacher surveys completed and returned by mail. A
detailed description of the study, including a summary of the
The University of Southern California (USC) Twin Study
measures, can be found in Baker, Barton, Lozano, Raine, and
of Risk Factors for ASB
Fowler(2006).
Thisisoneofthefirstprospectivetwinstudiesofpreadolescent
children to focus on aggressive and antisocial behavior using a Participant Recruitment
multitrait, multi-informant approach. In this article we present
results for the comprehensive phenotypic assessments of aggres- ThetwinsandtheirfamilieswhoarepartoftheUSCStudyof
siveandantisocialbehaviorconductedduringthefirstwaveofthe Risk Factors for Antisocial Behavior were recruited from the
study,whiletheparticipantsareatthebrinkofadolescence(ages larger Southern California Twin Register, which contains over
9and10yearsold),andusemultivariategeneticfactormodelsto 1,400totalpairsofschool-agetwinsbornbetween1990and1995.
examinetheextenttowhichgeneticandenvironmentalinfluences ParticipantsintheTwinRegisterarevolunteers,andfamilieswere
accountforagreementanddisagreementacrossraters.Thisstudy ascertained primarily through local schools, both public and pri-
expands on previous research in the following important ways. vate, in Los Angeles and the surrounding communities—see
First, it examined the relative influence of genetic and environ- Baker, Barton, and Raine (2002) for a detailed description of the
mental factors on antisocial behavior using an ethnically and recruitmentprocessandTwinRegisterfromwhichthetwinswere
socioeconomicallydiversesample.Theethnicandsocioeconomic sampled.Familiesidentifiedashavingtwinsinthetargetagerange
variabilityofthesamplemayallowforgreatergeneralizabilityof weresentlettersbrieflydescribingthestudyandinvitingthemto
resultstothediversepopulationsinurbanareas,whereantisocial, participate.
aggressive, and violent behaviors present serious threats to the Study participation required that the twins be (a) proficient in
communityatlarge.Second,itusedmultipleindicesofantisocial Englishand(b)9or10yearsoldatfirstassessment(seeBakeret
behavior. Rather than relying on univariate comparisons of heri- al., 2006). In addition, either English or Spanish proficiency was
tability estimates for various types and severities of antisocial required for the twins’ primary caregiver. Of the 1,400 families
behavior, the use of a composite measure based on all of the who joined the USC Twin Register and were in the target age
differentindicesmayyieldastrongergeneticsignalthananyone range, approximately 860 families were contacted by phone to
index of antisocial behavior alone. Third, the study relied on explainthestudyingreaterdetailandtoscheduleatestingsession.
reports of antisocial behavior from multiple informants. This al- The sample of 605 tested families thus constituted a 70% partic-
lowedusto(a)examinewhethertherearesignificantdifferences ipation rate of those families whom we were able to contact.
acrossraters;(b)testformallytheextenttowhichraterbiasmay Approximately 30 families (3% of the total eligible sample) did
influenceresults;and(c)combineinformationfromdifferentrat- notqualifybecauseoflimitedEnglishproficiencyinthechildren.
ersinamultivariatemodel,allowingforthepresenceofa“shared” Theremainingfamilieswereeitherneverscheduled,cancelled,or
view of antisocial behavior that may be more reliable than any didnotshowupfortheirtestingsession.
222 BAKER,JACOBSON,RAINE,LOZANO,ANDBEZDJIAN
Procedure and p (cid:1) .14, respectively), indicating that our results for teacher
reportsareunlikelytobebiasedbydifferentialresponsepatterns.
Laboratoryvisitprotocol. Testingandinterviewsofthechild
and caregiver were made during a 6- to 8-hr visit to the USC
Sample Characteristics
laboratories.ThedetailsoftheprotocolcanbefoundinBakeret
al. (2006). Briefly, the visit included behavioral interviews, neu- Participants in the present study consisted of 605 families of
rocognitivetesting,socialriskfactorassessment,andpsychophys- twins (n (cid:1) 596 pairs) or triplets (n (cid:1) 9 sets) and their primary
iologicalrecordingofthetwins.Caregiverswerealsointerviewed caregiverswhoparticipatedinthefirstwaveofassessmentinthe
about their twins’ behavior, as well as their own behavior and USC Study of Risk Factors for ASB. To avoid problems of
relationshiptoeachtwin.Cheekswabsampleswerealsocollected additionalfamilialinterdependencyassociatedwiththesmallnum-
fromtheparticipatingfamiliesinordertoextractDNAandtestfor beroftripletpairs,asinglepairconsistingof2ofthe3tripletswas
zygosity. randomly selected for these analyses. The sample was composed
Participating families were compensated for their visit to USC ofbothmaleandfemaleMZandDZpairs,includingbothsame-
and provided with additional incentives for keeping scheduled and opposite-sex DZ twins. Among the 1,219 child participants,
appointmentsinatimelyfashion(totalpaymentswereupto$125). there was approximately equal gender distribution with 48.7%
Familieswerealsoprovidedwithgroupsummariesofstudyresults boys(n(cid:1)594)and51.3%girls(n(cid:1)625);the605caregiverswere
and individual reports of their twins’ zygosity and each child’s primarilyfemale(94.2%).
cognitivetestingresults. Caregiverparticipantswereprimarilybiologicalmothersofthe
Given the sensitive nature of the information provided by the twinsandtriplets(91.4%;n(cid:1)553),althoughotherrelativeswere
twinsandtheircaregivers(includingillegalbehaviors),aCertifi- also interviewed, including biological fathers (n (cid:1) 35; 5.8%),
cate of Confidentiality was obtained for this study from the Na- stepparents (n (cid:1) 2; 0.3%), adoptive parents (n (cid:1) 4; 0.7%),
tionalInstituteofMentalHealthtohelpprotecttheprivacyofthe grandparents (n (cid:1) 7; 1.2%), or other relatives (n (cid:1) 4; 0.6%). At
participants. All participants were assured that the information thetimeoffirst-waveassessment,nearlytwothirdsofthechildren
theyprovidedwouldbecodednumericallyandnotlinkedtotheir werelivingwithbothbiologicalparents,whowereeithermarried
names and that their individual information would not be shared orlivingtogetherbutunmarried(55.5%and7.2%oftotalsample
withanyoneoutsidetheresearchteam.Thelaboratoryprocedures offamilies,respectively).Amongtheremainingfamiliesinwhich
andallaspectsofthestudywerereviewedbytheUSCInstitutional the biological parents were not living together (because of sepa-
ration,divorce,deathoftheparent,orneverhavingbeenmarried),
ReviewBoardandwerecompliantwithfederalregulationsatthe
themajorityofthesewerenotmarriedorlivingwithapartnerat
time.
thetimeoffirst-wavetesting—only6.2%ofthetotalsamplewas
Assessments were conducted by rigorously trained examiners
remarried to another partner. Thus, the majority of the children
(see Baker et al., 2006, for details). All child interviews were
livedintwo-parenthouseholds,although114twinortripletpairs
conducted in English; caregiver interviews were conducted in
either English (n (cid:1) 492; 81.3%) or Spanish (n (cid:1) 113; 18.7%), (18.8%)didliveinasingle-parenthouseholdwithnootheradult
inthehome.Theremainderofthechildren(12.2%)residedwitha
dependingonthelanguagepreferenceoftheparticipant.Lessthan
single parent as well as one or more other adults (mostly grand-
halfoftheHispaniccaregivers(44.0%)preferredtobeinterviewed
parents).
inSpanish.AllcaregiversurveysweretranslatedintoSpanishand
The child’s ethnicity was determined by the ethnicity of their
back-translatedintoEnglishbyprofessionaltranslators.
two biological parents as reported by the primary caregiver. As
Teachersurveys. Thetwins’teacherswereaskedtocomplete
such, the twin–triplet sample was 26.6% Caucasian (n (cid:1) 161
surveys about each child’s school behaviors and to return their
pairs), 14.3% Black (n (cid:1) 86 pairs), 37.5% Hispanic (n (cid:1) 227
surveypacketstoUSCinprepaid,addressedenvelopes.Teachers pairs),4.5%Asian(n(cid:1)27pairs),16.7%Mixed(n(cid:1)101pairs),
werenotpaidfortheirparticipation.Excludingpairs(n(cid:1)15)who and0.3%otherethnicities(n(cid:1)2pairs).AmongtheMixedgroup,
were either homeschooled or for whom parents felt the teachers mostchildren(57.4%;n(cid:1)58pairs)hadoneHispanicparent,and
did not know their children well enough to rate their child, there thusnearlyhalfofthesample(47.1%;n(cid:1)281twinortripletsets)
wasa60%individualreturnrateforteachersurveys.Althoughwe
wasofatleastpartialHispanicdescent.Thisethnicdistributionis
did not receive teacher surveys for all twins, we did have infor-
comparable to that in the general Los Angeles population (http://
mation on whether twins were in the same class at school for all
www.census.gov/main/www/cen2000.html) and therefore pro-
but18twinpairs.Amongtheentiresample,31.4%oftwinswere videsadiversecommunitysamplerepresentativeofalargeurban
inthesameclassroom.Amongthe269pairsforwhombothtwins area.
had teacher reports (see the Missing data section for details), Medianfamilyincomewasinthe$40,000to$54,000range(the
41.4% were in the same classroom at school and were therefore midpointofwhichis$45,500),whichiscomparabletothemedian
ratedbythesameteacher.Thissuggeststhatteachersweresome- income in Southern California (including Los Angeles, Orange,
whatmorewillingtoreturnsurveysifbothtwinswereinthesame Riverside, Ventura, and San Bernardino counties) between 2000
classatschool.Female–femaletwinpairswereslightlymorelikely and 2002 (average Mdn (cid:1) $43,042; http://www.census.gov/cgi-
to be placed in the same classroom than male–male twin pairs bin/saipe/saipe.cgi) and the state of California between 2001 and
(34.8% vs. 31.5%), and monozygotic (MZ) twins were slightly 2003 (average Mdn (cid:1) $48,979; http://www.census.gov/hhes/
more likely than dizygotic (DZ) twins to be placed in the same income/income03/statemhi.html).Educationlevels,measuredona
class (36.0% vs. 29.0%). However, chi-square analysis revealed 6-point scale, ranged from 1 (less than high school) to 6 (post-
that neither of these effects was statistically significant (p (cid:1) .46 graduate degree). Maternal and paternal education levels were
CHILDHOODASB 223
significantlycorrelated(r(cid:1).61,p(cid:2).01),andsignificantlyhigher total of three unique informants (caregivers, teachers, and chil-
mean levels of education were reported for mothers (M (cid:1) 3.70, dren). Instruments varied in terms of their mode of assessment,
SD(cid:1)1.58)thanforfathers(M(cid:1)3.53,SD(cid:1)1.63),t(552)(cid:1)3.43, with some being administered through semistructured interviews
p(cid:2).001.Acompositemeasureofbothparents’educationlevels, (i.e.,theDiagnosticInterviewScheduleforChildren—VersionIV
occupationalstatus,andfamilyincome(Hollingshead,1975)was [DISC–IV])andothersthroughquestionnairesadministeredeither
usedasanindexofsocioeconomicstatus(SES)inthisstudy.The in an interview format (i.e., the Childhood Aggression Question-
distribution of the SES factor was slightly skewed toward higher naire[CAQ]andtheChildPsychopathyScale[CPS])orinpaper-
levels,althoughtherewasconsiderablerangeinSESinthisstudy. and-pencil format (i.e., the Child Behavior Checklist [CBCL]).
Zygosity of the same-sex twin pairs was determined for the Each instrument was given to at least two of the three possible
majority of pairs (398/458 (cid:1) 87%) through DNA microsatellite informants. The following sections provide detailed information
analysis(sevenormoreconcordantandzerodiscordantmarkers(cid:1) abouteachofthefiveinstruments,includinginformationaboutthe
MZ; one or more discordant markers (cid:1) DZ). A Twin Similarity instrumentitself,modeofassessment,informanttype,anduseof
Questionnaire (Lykken, 1978) was used to infer zygosity for the anyrelevantsubscales.
remaining 60 pairs for whom adequate DNA samples or results DISC–IV (Schaffer, Fisher, Lucas, Dulcan, & Schwab-Stone,
were not available. When both questionnaire and DNA results 2000). TheDISC–IVisahighlystructuredinterviewdesignedto
wereavailable,therewas90%agreementbetweenthetwo. assess psychiatric disorders, adapted from the Diagnostic and
The frequencies of the five gender and zygosity groups are StatisticalManualofMentalDisorders(4thed.;DSM–IV;Amer-
presentedinTable1,alongwithmeanageandethnicdistribution. icanPsychiatricAssociation,1994),andsymptomsinchildrenand
The mean ages during first-wave assessment were 9.60 years
adolescents ages 6 to 17 years. The DISC was designed to be
(SD(cid:1)0.60)forthetotalsampleofchildrenand40.14years(SD(cid:1)
administeredbywell-trainedlayinterviewersforepidemiological
6.61)fortheircaregivers.Althoughzygositygroupsdidnotdiffer
research.Ithasayouthaswellasaparallelparentversion,bothof
in mean age of children at first-wave assessment, F(4, 604) (cid:1)
which inquire about the child’s psychiatric symptoms. The Con-
0.70,p(cid:1).59,thereweresignificantdifferencesincurrentageof
duct Disorder module was administered using both youth and
biological mother across groups, F(4, 594) (cid:1) 3.64, p (cid:2) .01—
parent versions in the present study. Although not a focus of the
mothersofDZpairsweresignificantlyoldercomparedwithmoth-
current report, additional modules assessing oppositional defiant
ersofMZpairs.Therewasalsosignificantethnicgroupvariation
disorder, attention-deficit/hyperactivity disorder, major depres-
acrossthesefivezygositygroups,(cid:3)2(16,N(cid:1)605)(cid:1)33.82,p(cid:2)
sion,andgeneralizedanxietyineachchildwerealsoadministered
.01—BlacksandCaucasiansappearedtobemorefrequentlyrep-
in the parent version. Both symptom counts and diagnoses were
resentedintheDZgroups,whereasahigherpercentageofAsians
provided through computerized scoring of the DISC–IV Conduct
andHispanicswereseenintheMZgroups,particularlyamongthe
Disordermodule.
maleparticipants.Thesedifferencesmaystemfromdifferenttwin-
Conductdisorderdiagnoses(forthepastyear)weremadefor16
ning rates across ethnic groups, due in part to differences in
boys (2.7%) and 8 girls (1.3%) based on caregiver reports in the
maternal age and use of assisted reproduction methods when
DISC–IVandfor9boys(1.6%)and2girls(0.3%)basedonchild
conceiving the twins. Although the overall zygosity distribution
self-report. Although symptom counts for conduct disorder were
amongsame-sexpairs(60.2%MZ)wassignificantlygreaterthan
significantlycorrelatedbetweencaregiverandyouthreports(r(cid:1)
theexpected50%(p(cid:2).01),itwasnotasmarkedlyhighasinmost
.31, p (cid:2) .001), it is noteworthy that there was no overlap in
othervolunteersamples,inwhichtwothirdsofsame-sexpairsare
conductdisorderdiagnoses—thatis,nosinglechildreachedcon-
typicallyMZ(Lykken,McGue,&Tellegen,1987).Thissampleof
duct disorder criteria for both child and caregiver reports. Most
childrenandcaregiversappearstobequiterepresentativeofboth
likely,thispatternofresultsisduetotherelativelyyoungageof
themultiplebirthandgeneralpopulationinsouthernCalifornia.
this sample and the fact that this is a population-based (nonclini-
cal) sample. Although diagnosed cases according to one of the
Measures
raters had elevated symptoms reported by the other rater, these
The present study used a total of 18 different measures of individualsfellshortofreceivingacorrespondingdiagnosisfrom
antisocial behavior taken from five different instruments from a theotherrater.Inaddition,thefocusonconductdisorderbehaviors
Table1
SampleCharacteristics
Pairs Childage Momage Child’sethnicity
Zygosity-sex n % M SD M SD %Caucasian %Hispanic %Black %Asian %Mixed-other
MZmale 138 22.9 9.57 0.55 37.56 5.63 19.6 46.4 9.4 8.7 15.9
MZfemale 139 23.0 9.66 0.61 38.54 6.92 21.6 38.1 12.2 4.3 23.7
DZmale 84 13.9 9.64 0.60 39.88 5.74 28.6 31.0 15.5 2.4 22.6
DZfemale 97 16.0 9.55 0.55 39.43 5.29 35.1 32.0 17.5 2.1 13.4
DZmale-female 147 24.3 9.58 0.59 39.90 5.90 31.3 36.1 17.7 3.4 11.6
Total 605 9.60 0.58 38.98 6.03 26.6 37.5 14.2 4.5 17.2
Note. Age of biological mother is shown, to allow comparison across zygosity groups for maternal age at birth of twins. Mean age of all caregiver
participantswas40.14years(SD(cid:1)6.61).MZ(cid:1)monozygotic;DZ(cid:1)dizygotic.
224 BAKER,JACOBSON,RAINE,LOZANO,ANDBEZDJIAN
duringthepastyear(ratherthanlifetimeprevalenceusedinmost weredeterminedtobeatorbelowasecond-gradelevelasdeter-
retrospective studies of twins) may have reduced both the preva- mined by the Woodcock–Johnson Reading Achievement Test
lence of conduct disorder and the agreement among raters. Nev- (Woodcock & Johnson, 1989). Teachers were also given the
ertheless,giventhelowprevalenceofdiagnosableconductdisor- parallel form of the CBCL (the Teacher Report Form) as part of
der at this age, number of conduct disorder symptoms was used themailsurveypacket.
rather than conduct disorder diagnosis. According to caregiver
reports,54.5%ofboysand39.2%ofgirlshadatleastoneconduct
Short-Term Reliability
disorder symptom. The corresponding figures for child reports
were47.8%ofboysand30.3%ofgirls.
Thirty randomly selected families with complete data (both
TheCAQ. Thisinstrumentwasdevelopedtoassessoverall,as
cotwins and their caregiver) completed the entire first-wave as-
wellasvariousformsof,aggression.Threeparallelformsofthis
sessment a second time, approximately 6 months following their
questionnaire were used: (a) child self-report, (b) caregiver’s re-
original laboratory visit. These retest families included exactly
portofchild’sbehavior,and(c)teacher’sreportofchild’sbehav-
50%ofeachgender(n(cid:1)30boys;n(cid:1)30girls)andwereusedto
ior.ThemajorityoftheitemsweretakenfromRaineandDodge’s
evaluate test–retest reliability for all measures used in this study.
Reactive and Proactive Aggression Questionnaire (Raine et al.,
This sample was the basis for testing reliability of the measures
2006),including11reactiveitems(e.g.,“IdamagethingswhenI
across time. Test–retest correlations are presented in Table 2,
am mad”; “I get mad or hit others when they tease me”) and 12
separatelyforboysandgirls,aswellasforthecombinedsample.
proactiveitems(e.g.,“Ithreatenandbullyotherkids”;“Idamage
There was remarkable stability for these measures, although the
orbreakthingsforfun”).Inaddition,5itemswereaddedtoyield
correlationsvariedsomewhatacrossraterandsexofchild.Great-
relationalaggressioninthechildandteacherversions(e.g.,“Itell
est stability was observed for caregiver reports, especially for
stories about people behind their back when I am mad at them”;
ratingsofboys.ThelowestcorrelationsinTable2areforcaregiver
“When I am mad at someone I tell my friends not to play with
ratings of conduct disorder symptoms (.57) and CBCL Delin-
them”).EachoftheitemsintheCAQwasratedona3-pointscale
(0 (cid:1) never, 1 (cid:1) sometimes, 2 (cid:1) often), and responses were quency (r (cid:1) .47) in girls and for girls’ self-reported conduct
disorder symptoms (r (cid:1) .56). Inspecting graphical summaries of
summedwithineachofthesubtypes,foreachofthe3informants,
these correlations, however, revealed 1 outlier—a girl who re-
separately. All three scales showed good internal consistency
ceived low ratings on these measures in the first testing and
(Cronbach’s alpha ranged from .73 to .76 for child self-report,
considerablyhigherratingsinthesecond.Writtencommentsfrom
from.76to.83formotherratings,andfrom.90to.92forteacher
the examiners for this family indicated that this girl had indeed
ratings).
experienced significant behavioral changes in the 6 months in
The CPS (Lynam, 1997). The CPS is composed of 14 sub-
betweenthetwotestingsessions(confirmedbyboththecaregiver
scales(basedon55yesornoitems),whichconsistofassessments
andchildexaminers).Removingthiscasefromthissmallsample
of Glibness, Untruthfulness, Lack of Guilt, Callousness, Impul-
ofretestfamiliesresultedinhigherretestcorrelations(r(cid:5).60for
siveness, Boredom Susceptibility, Manipulation, Poverty of Af-
all three instances). Thus, although there does appear to be con-
fect,ParasiticLifestyle,BehavioralDyscontrol,LackofPlanning,
siderablereliabilityinthesemeasures,wemustbecognizantofthe
Unreliability, Failure to Accept Responsibility, and Grandiosity.
factthatthepotentialfordevelopmentalchangeispossibleatthis
Minorchangesweremadetothewordingofsomeitemsforease
age. Therefore, we suspect that these estimates of “short-term”
ofunderstandingby9-to10-year-oldchildren.Parallelversionsof
reliabilityareactuallyconservativeestimates(i.e.,underestimates)
theCPSwereadministeredtoboththechildandthecaregiverin
ofthetruereliabilities.
interviewform.Thetwoclassicfactorsofpsychopathy(Factor1:
Callous–Unemotional; Factor 2: Impulsive–Irresponsible) were
derived in each of the caregiver and child self-reports, based on
compositesofthe14subscalesintheCPSwithineachrater.Both Table2
Factor 1 and Factor 2 showed reasonable internal consistency in Six-MonthTest-RetestCorrelationsforAntisocialBehavior
caregiver ratings ((cid:4) (cid:1) .71 and .74, respectively), with somewhat Measures
lowervaluesinchildself-report((cid:4)(cid:1).63and.61).
The CBCL (Achenbach, 1991). The CBCL is a caregiver rat- Caregiverreports Childreports
ing scale composed of 112 items concerning a child’s behavior rTime1-Time2 rTime1-Time2
withinthepast12months.Itemsareratedona3-pointscale(0(cid:1)
Scale Boys Girls All Boys Girls All
nottrue,1(cid:1)sometimesorsomewhattrue,2(cid:1)verytrueoroften
true)andareusedtoderiveeightsubscales:Withdrawn,Anxious/ Reactiveaggression .93 .60 .81 .78 .46 .64
Depressed, Social Problems, Thought Problems, Attention Prob- Proactiveaggression .84 .70 .79 .74 .60 .67
Relationalaggression — — — .79 .77 .68
lems, Delinquent Behavior, Somatic Complaints, and Aggressive
CPSFactor1 .85 .77 .81 .54 .62 .61
Behavior (Achenbach, 1991). For the purposes of the present
CPSFactor2 .87 .73 .81 .67 .51 .61
article, however, only the Delinquent Behavior (13 items) and Conductdisordersymptoms .88 .57 .88 .75 .56 .64
Aggressive Behavior (20 items) subscales were used in our anal- CBCLdelinquency .87 .47 .85 — — —
yses. The CBCL was administered during the laboratory visit to CBCLaggression .84 .80 .82 — — —
Firstprincipalcomponent .95 .88 .94 .82 .83 .81
the caregivers in either survey (paper) or interview form. The
CBCLwasadministeredtothecaregiversininterviewformrather Note. Allcorrelationswerehighlysignificant(p(cid:2).01).CPS(cid:1)Child
than in paper form if the subject’s reading comprehension skills PsychopathyScale;CBCL(cid:1)ChildBehaviorChecklist.
CHILDHOODASB 225
Statistical Analyses viduals with missing data on a given measure, within rater, were
assignedamissingvalueforthecompositescale.Asmissingdata
General issues. Descriptive statistics, mean level compari-
on individual measures were relatively rare among children and
sons, phenotypic correlations, and factor analyses were all con-
caregivers, we had valid factor scores for more than 96% of the
ducted using the SPSS (Version 11.5) statistical package. Multi-
sample (N (cid:1) 1,175 for child-based factor scores, and N (cid:1) 1,193
variategeneticanalysisoftheratereffectsmodelswasconducted
for caregiver-based factor scores). Among the 698 teachers who
usingthestructuralequationmodeling(SEM)programMx(Neale,
reportedontheantisocialbehaviorofthechildren,wecouldcreate
Boker,Xie,&Maes,2003).
factorscoresformorethan97%ofthem(N(cid:1)681;55.9%ofthe
Missing data. Missing data for child- or caregiver reports of
totalsampleof1,219individuals).
the different antisocial behavior measures were quite rare. For
One of the reasons for selecting Mx for the multivariate twin
most measures, we had valid data for 1,210 to 1,219 of our total
analysesisthatitusesfullinformationmaximum-likelihoodwhen
sampleof1,219individualchildren.Missingdataweresomewhat
fitting models to the raw data. Thus, all pairs in which at least 1
greaterforchildandcaregiverratingsofconductdisorder;still,we
twinhasnonmissingdataonatleastonemeasurecanbeincluded
had complete child and caregiver data for more than 95% of the
in the analyses, and fit functions are based on the calculation of
sample(seeTable3forindividualsamplesizesforeachmeasure).
twice the negative log-likelihood of all nonmissing observations
As detailed in the methods, the overall teacher response was
(where an observation is defined by measure, not by individual).
approximately 60%; however, valid teacher-report data on the
Forthepresentanalyses,only1ofthe605possiblepairsdidnot
antisocialbehaviormeasureswereobtainedforapproximately700
individual twins (57.4%). Of the 605 individual twin pairs, 269 have any usable data and was excluded from the twin analyses.
pairs(44.5%)hadteacherreportsforbothtwins,andanadditional Nearly all of the 605 pairs (N (cid:1) 591 pairs, 97.8%) had valid
143pairs(23.6%)hadteacherreportsofantisocialbehaviorforat composite scores for both twins based on the caregiver ratings.
leastoneofthetwotwins.Ofthe269pairsforwhomwehadvalid Over 90% (N (cid:1) 559 pairs, 92.4%) of the pairs sample had both
teacher-report data for both twins, 111 pairs (41.4%) were in the caregiver and child-report composite scores for both twins, and
sameclassroomatschoolandwerethusratedbythesameteacher 42.0%ofthesample(N(cid:1)254pairs)hadvaliddataforbothtwins
informant. fromcaregiver,child,andteacherreports.Anadditional130pairs
Missing data were handled in a variety of different ways. For (21.5%ofthesample)hadcompletedatafromcaregiverandchild
phenotypic analyses of mean level differences and correlations reports, and teacher report data for 1 member of the twin pair.
amongindividualsubscalesofASB,alistwisedeletionprocedure Although these latter pairs could not contribute information re-
wasused,astheseanalysesareconductedfordescriptivepurposes gardingcovarianceacrosstwinsforteacherreports,theydidpro-
only. For the creation of the factor-based composite scores, indi- videinformationforthesamplemeansandvarianceoftheteacher
Table3
Aggression,Delinquency,andPsychopathy:MeansandStandardDeviationsbySexandInformant
Childreport Caregiverreport Teacherreport
Scalea Child’ssexb M SD N M SD N M SD N
Reactiveaggression Malec,d,e 1.66 0.34 593 1.71 0.33 594 1.41 0.39 341
Femaled,e 1.61 0.31 623 1.63 0.33 625 1.28 0.30 356
Proactiveaggression Maled,e 1.09 0.17 593 1.11 0.17 594 1.16 0.26 341
Femalec,d,e 1.06 0.12 623 1.08 0.14 625 1.10 0.18 356
Relationalaggression Malee 1.24 0.29 593 — — — 1.28 0.35 339
Femalee 1.20 0.24 623 — — — 1.28 0.35 356
CPSFactor1 Malec 1.30 0.10 589 1.32 0.12 594 — — —
Femalec 1.27 0.08 621 1.30 0.12 625 — — —
CPSFactor2 Malec 1.25 0.15 589 1.27 0.19 594 — — —
Femalec 1.23 0.14 621 1.21 0.16 625 — — —
CBCLaggression Male — — — 6.36 5.34 589 5.76 8.25 339
Femaled — — — 5.27 5.20 624 3.64 6.46 359
CBCLdelinquency Male — — — 1.53 1.86 589 1.28 2.10 339
Femaled — — — 1.08 1.45 624 0.67 1.44 359
Conductdisordersymptoms(DISC-IV) Malec 1.12 1.91 575 1.44 2.02 588 — — —
Femalec 0.64 1.29 602 0.90 1.54 611 — — —
Note. CPS (cid:1) Child Psychopathy Scale; CPS Factor 1 (cid:1) Callous-unemotional; CPS Factor 2 (cid:1) Impulsive-irresponsible; CBCL (cid:1) Child Behavior
Checklist;DISC-IV(cid:1)DiagnosticInterviewScheduleforChildren—VersionIV.
aPossiblerangeofscoresis1–3forReactive,Proactive,andRelationalAggressionand1–2forChildPsychopathyFactor1andFactor2.CBCLscales
reportedarerawscores,indicatingnumberofitemsendorsedwithineachscale,withamaximumpossibleof40forAggression,26forDelinquency,0–13
forConductDisorderSymptoms.
bMeansforboysandgirlsaresignificantlydifferent(p(cid:2).01)withineachraterforeveryscaleexceptteacherratingsofrelationalaggression.
cSignificant(p(cid:2).05)meandifferencesbetweencaregiverandchild.
dSignificant(p(cid:2).05)meandifferencebetweencaregiverandteacher.
eSignificant(p(cid:2).05)meandifferencebetweenchildandteacher.
226 BAKER,JACOBSON,RAINE,LOZANO,ANDBEZDJIAN
reports,aswellasforthewithin-personcorrelationsacrossinfor- are common across raters). As described earlier, all additive ge-
mants. Thus, including all 384 pairs (63.5%) for whom we had neticinfluences(A)correlate1.0acrossMZtwinsand0.5across
valid teacher reports for at least 1 of the 2 twins minimized DZ twins, shared environmental (C) effects correlate 1.0 across
potential sampling bias. Patterns of missingness did not vary twins,regardlessofzygosity,andnonsharedenvironmentalinflu-
significantlybysexorzygosity(resultsofthechi-squareanalyses ences(E)didnotcorrelateacrosstwins.Allthreemodelsallowfor
areavailableuponrequest).Forexample,validteacherreportdata informant-specific nonshared environmental influences (E , E ,
M K
forbothtwinswereavailablefrom40.8%to48.8%ofanygiven and E ), as any given measure is an imperfect estimate of the
T
zygosity group. Complete pairwise data for caregiver and child underlying “true score”; thus, informant-specific nonshared envi-
reports were available for more than 92% of any given zygosity ronmental effects in this model include errors of measure. In
group. contrast, the nonshared environment that influences the common
Geneticmodels. Theratermodelsusedwerebasedonexten- latentfactor(E )representsenvironmentalfactorsthatvaryacross
C
sionsofthetraditionalACEmodelthatistypicallyusedinbehav- twinsinthesamefamily,whicharesystematicallyassociatedwith
ioral genetic studies. These models use information from the ASB(e.g.,differentialparentaltreatmentordifferentpeergroups).
observedtwinvariancesandcovariances(calculatedfromtheraw Figure 1a presents the rater effects model, which allows for
data) to partition the overall variance into additive genetic (A),
additionalwithin-informantcorrelationacrosstwinsforcaregivers
common (or shared) environmental (C), and nonshared environ-
and teachers, due to the fact that the same rater is reporting on
mental (E) influences (Neale & Cardon, 1992).1 In behavioral
behavior for both twins. Individual twins only reported on their
genetic models, additive genetic influences are correlated 1.0
own behavior; therefore, it was not possible to estimate rater
among MZ twin pairs, as MZ twins have identical genotypes. In
effects for child reports. As can be seen in Figure 1a, this model
contrast, DZ twins share, on average, half of their segregating
(also referred to as the correlated errors model; Simonoff et al.,
genes; thus, these models assume a correlation of .5 among DZ
1995) allows for latent variables representing rater effects to
pairs.Theproportionofvariationthatisduetogeneticinfluences
influence variation in caregiver and teacher reports (R and R ,
is called the heritability. Shared environmental factors include M T
respectively). To the extent that ratings are influenced by the
those environmental factors that serve to make individuals in a
qualities of the informant, this would affect the ratings of both
family similar to one another but that may differ across families.
twins in a pair and may lead to overestimations of the twin
Thus,sharedenvironmentalinfluencescanincludesuchfactorsas
correlations.AsFigure1ashows,thecorrelationfortheratereffect
SES, family structure, and shared peer influences, as well as
among caregiver reports was 1.0, because all caregivers in our
broader contextual factors (e.g., school or neighborhood effects).
sample reported on the behavior of both twins. In contrast, the
IntheACEmodel,sharedenvironmentalinfluencesarecorrelated
correlationoftheratereffectforteacherscouldbeeither1.0or0,
1.0 across twin pairs, regardless of zygosity. Nonshared environ-
dependingonwhetherthesameteacherratedbothtwins(acorre-
mental influences are any environmental influences that serve to
lation of 1.0) or whether a different teacher rated each twin (a
makeindividualsdissimilar,includingmeasurementerrors(which
correlationof0).ByusingafeatureofMxthatallowsfortheuse
areassumedtoberandom).Bydefinition,nonsharedenvironmen-
of definition variables as moderators of individual parameters
talinfluencesdonotcorrelateacrosseitherMZorDZpairs.
(Nealeetal.,2003),wewereabletouseadummycodeforeach
Bycombiningdatafromallthreeinformantssimultaneouslyin
twinpairasadefinitionvariablethatrepresentedwhetherthetwins
multivariate genetic models, we are able to differentiate genetic
andenvironmentalfactorsthatinfluenceasharedviewofantiso- were in the same classroom (and thus were rated by the same
cial behavior from genetic and environmental factors that influ- teacher)tomultiplytheparameterfortheteacherratereffect(rT)
ence each informant’s own particular rating. Moreover, we can by either 1.0 (same class) or 0 (different class). If rT (cid:5) 0, this
alsoinvestigatetheextenttowhichratereffectsmayhavebiased wouldpredicthighercorrelationsamongtwinsratedbythesame
estimates of heritability of ASB. Figure 1 shows the three multi- teacher.
variate models used to address this issue. All three models are Figure1bshowsanalternativemultivariatemodelknownasthe
variantsofacommonpathwaysmodel,whichallowedforgenetic measurement model. This model, which is a restricted version of
and environmental influences on observed measures to operate the model presented in Figure 1a, eliminates the rater effects for
through a single underlying phenotype (i.e., A , C , and E ; see caregiverandteacherreports.Thecriticalassumptionofthemea-
C C C
Kendler, Heath, Martin, & Eaves, 1987; McArdle & Goldsmith, surementmodelisthatthelatentvariablerepresentingtheshared
1990,fordetailsoncommonpathwaysmodels).2Inmultiple-rater view is the “true” representation of ASB and that all meaningful
analyses,theunderlyinglatentvariablethatallowsforcorrelations genetic and environmental influences on variation in reports of
across raters reflects a common, or shared, view of the child’s
antisocial behavior. The genetic and environmental factors that
influence this underlying shared view are further unbiased by 1Preliminary univariate analyses within each informant addressed
whethermodelswithnonadditivegeneticvariancefitbetterthanmodels
either rater effects or measurement error, as these latter effects
with common environmental variance (i.e., ACE vs. ADE models) and
influence only the rater-specific views (this is discussed in more
whethertherewasevidenceforsiblinginteractioneffects.Resultsindicated
detaillater).Eachrater’sindividualviewloadsontheunderlying
thattheACEmodelwithoutsiblinginteractioneffectswasthebestmodel
latentfactorthroughthepathsmarked(cid:6)(withsubscriptsM,K,and
foreachinformant(resultsareavailableuponrequest).
Treferringtocaregiver[mother],child[kid],andteacherreports, 2WealsofitlessstringentindependentpathwaysandCholeskymodels
respectively). Genetic and environmental influences that account toourdata.BasedonBayesianinformationcriterionvalues,thecommon
forvariationinthesharedviewofASBaredepictedthroughpaths pathwaysmodelofferedabetterbalanceofgoodnessoffitandparsimony
a , c , and e (whereby the subscript C refers to influences that thaneitheroftheseotherlessrestrictivemodels.
C C C
CHILDHOODASB 227
ASB are operating through the latent phenotype. Any residual (E).Thus,theamountofvarianceaccountedforbyrater-specific
variance on each rater’s individual perception of ASB that is not E should be consistent with estimates of the reliability of each
explainedbythelatentphenotypeisassumedtoberandommea- rating.
surement error that is not systematically related to characteristics Figure 1c shows the third and final model, which is the full
of the rater and is, therefore, modeled as nonshared environment version of the common pathways model for multiple raters. In
addition to allowing for the uncorrelated errors of measurement
andratereffects(i.e.,correlatederrorsofmeasurement)foundin
the aforementioned rater effects and measurement models, this
modelfurtherallowsforspecificgeneticandsharedenvironmental
factors to influence variation in each informant’s own ratings of
the child’s ASB. For simplicity, the model is shown for 1 twin
only; however, the specific A and C influences on each infor-
mant’s report of ASB correlate across twin pairs in the manner
described earlier. As shown, the model allows for genetic influ-
enceonthespecificviewpointsofeachrater(A ,A ,andA ).The
M K T
general assumption is that these genetic factors represent valid
genetic variance that arises because each rater “sees” different
aspects of ASB (but see the Discussion section for alternative
explanations). Similarly, different rater perceptions of ASB can
alsobeinfluencedbysharedenvironmentalfactors(C ,C ,and
M K
C ). In this model, the potential effect of shared environmental
T
influenceoncaregiver’sreportsofASBisconfoundedbypotential
rater effects, both of which would increase correlations of care-
giverratingsacrosstwins,regardlessofzygosity(seeHewittetal.,
1992, for details). This is represented by the dashed lines for the
R andC effects.Becauseofthisconfound,onlyoneparameter
M M
canbeestimatedinthecommonpathwaysmodel,andthisparam-
etermayrepresentsharedenvironmentinfluences,aratereffect,or
somecombinationofboth.Incontrast,becauseonlysometeachers
rateonly1twinperfamily,andothersratebothtwins,theshared
environmental influences on teacher reports can be statistically
differentiatedfrompotentialraterbias.Asexplainedearlier,chil-
dren reported only on their own behaviors; thus, the common
pathwaysmodelcannotestimateratereffectsforchildreports.
The critical difference between this model and the models
presentedinFigure1aand1bisthatthismodeltreatsdifferences
inreportsofASBacrossratersasmeaningful.Inotherwords,this
model assumes that there are systematic causes for disagreement
among parents, teachers, and children that are not solely due to
random errors of measurement and/or perceptual biases. This
model would be consistent with the notion that parents, teachers,
andchildrenprovideauniqueperspectiveonthechild’sbehavior
Figure 1. (a) Rater effects model. (b) Measurement model. (c) Full
commonpathwaysmodel.Observedvariablesarerepresentedbyrectan-
gles; latent variables are represented by circles. A (cid:1) additive genetic
effects;C(cid:1)shared(common)environmentalinfluences;E(cid:1)nonshared
environmentalinfluences;R(cid:1)ratereffects;MZ(cid:1)monozygotic;DZ(cid:1)
dizygotic;Cgvr(cid:1)caregiver;Tchr(cid:1)teacher.Pathcoefficientswitha,c,e,
and r correspond to the effects of these latent factors on the observed
variables.Pathsmarkedwith(cid:6)representthefactorloadingsontheshared
view of antisocial behavior for each individual rater. Factors and corre-
sponding path coefficients that reflect influences on the shared view of
antisocial behavior are subscripted with C. The subscripts M, K, and T
refertofactorsandcorrespondingpathcoefficientsthatarespecifictothe
caregiver(M),child(K),andteacher(T)reports,respectively.AlllatentA,
C,E,andRfactorshaveanassumedvarianceof1.0;thevarianceinthe
factorrepresentingthesharedviewhaslikewisebeenconstrainedtounity.
228 BAKER,JACOBSON,RAINE,LOZANO,ANDBEZDJIAN
andthatnosingleinformantmaynecessarilybeconsideredmore obtained on the CPS; the Aggression and Delinquency subscales
validorreliablethananother. from the CBCL; and conduct disorder symptom counts from the
Model comparisons. One of the advantages of using SEM to DISC–IV.MeandifferencesinASBwereexaminedbetweenboys
estimate genetic and environmental influences on variation and and girls, as well as among different informants. Significant sex
covariation among traits or behaviors is that SEM provides a differences(p(cid:2).01)emergedintheexpecteddirection(boys(cid:5)
framework for evaluating how well the theoretical model (or girls)forallscalesexceptforteacherratingsofrelationalaggres-
models) fits the observed behavior. Traditionally, two statistics sion, which showed no significant sex difference. Antisocial be-
have been used to compare the fit of two nested models: the haviorwasclearlymoreprevalentinboysthaningirlsatthisage.
likelihood-ratio test (LRT) statistic (Neale & Cardon, 1992) and Thesemeanleveldifferenceswereconfirmedinthegeneticanal-
the Akaike information criterion (AIC; Akaike, 1987; Medsker, yses (results are available upon request); thus, means were esti-
Williams,&Holahan,1994).TheLRTisobtainedbycomparing matedseparatelyformalesandfemalesinallofthetwinmodels.
the–2log-likelihood(–2LL)ofacomparisonmodeltothe–2LL Although not shown in the table, it is noteworthy that diagnostic
ofanested(reduced)model.TheLRTstatisticisthedifferencein rates of disorders in this community sample are comparable to
–2LLbetweenthetwomodels,whichisdistributedasachi-square thosereportedinDSM–IV,forbothconductdisorder(n(cid:1)25boys,
statisticwithdegreesoffreedomequaltothedifferenceindegrees 4.2%;n(cid:1)10girls,1.6%receiveddiagnosesfromeitheryouthor
offreedombetweenthetwomodels.TheAICiscalculatedasthe parentinterviews)andoppositionaldefiantdisorder(n(cid:1)70boys,
LRTminustwicethedifferenceindegreesoffreedom;itindexes 11.9%;n(cid:1)49girls,8.1%;seeBakeretal.,2006).Boththelevel
bothgoodnessoffitandparsimony:ThemorenegativetheAIC, andtherangeofASBinthisethnicallydiversecommunitysample
the better the balance between goodness of fit and parsimony. of twins thus appear to be comparable to those in other nontwin
More recently, the Bayesian information criterion (BIC) is also populationsofchildren.
being used to evaluate model fit. The BIC is similar to the AIC, Severalsignificantdifferencesamonginformantsalsoemerged
except that it also adjusts for sample size (for details on the BIC (seeTable3).Caregiversprovidedsignificantlyhigherratingsthan
andacomparisonoffitstatisticsusingsimulateddata,seeMarkon boys’ratingsofthemselvesforfourofthefivescalestheyhadin
&Krueger,2004).Inthisarticle,wepresentallthreefitstatistics, common (reactive aggression, CPS Factors 1 and 2, and conduct
although when there is discrepancy, preference was given to the disorders,butnotproactiveaggression).Forgirls,asimilarpattern
BIC(adjustedforsamplesize),basedontheresultsofindependent of higher ratings by caregivers than self-reports was also evident
simulationstudies(Markon&Krueger,2004). forseveralscales(proactiveaggression,CPSFactor1,andconduct
Evaluationofmodelfitforthemultivariateanalysesisdoneat disordersymptoms),althoughcaregiverratingsofgirlswerelower
twodifferentlevels.First,amodelisfittothedatathatperfectly for CPS Factor 2 and not significant for reactive aggression.
recaptures the observed means, variances, and within- and cross- Caregiversthusdidnotgenerallyratechildrenhigherorlowerthan
twin covariances from the three informants simultaneously. This children rated themselves across the board, although some rater
“saturated”modelprovidesa–2LLstatisticthatcanbeusedasthe differenceswereevidentforbothgenders.Comparisonsofteacher
baselikelihoodfromwhichtheAICandBICstatisticsfromeach andcaregiverratingsofboysalsorevealedsignificantdifferences
theoreticalmodelarecalculated,providingastandardizedestimate forallscalesexceptCBCLAggressionandDelinquency,although
ofAICandBICvaluesforcomparison.Moreover,bycomparing direction of difference again depended on the scale (i.e., teacher
thefitofeachofourACEmodelstothefitofthissaturatedmodel ratingslowerforreactiveaggression,buthigherforproactiveand
usingtheLRT,weobtainan“absolute”estimateofhowwelleach relationalaggression).Thepatternofcaregiver–teacherdifferences
ofourhypothesizedmodelsfitstheobserveddata.Second,wecan wassimilaringirls,wherebyteachersagainprovidedsignificantly
alsocalculateanLRTstatisticbycomparingACEmodelsthatare lower ratings for reactive aggression, and all three CBCL scales,
“nested”withineachother.Wenotethatthemeasurementmodel but higher ratings for proactive aggression. Teacher ratings were
(Figure 1b) is a nested submodel of the rater effects model (Fig- alsosignificantlylowerthanchildself-reportforreactiveaggres-
ure 1a), which is itself a nested submodel of the full common sion in both boys and girls, but higher for proactive aggression.
pathways model (Figure 1c); therefore, LRT statistics can be Although not shown in the table, there were no differences in
calculatedforeachsetofcomparisons.Moreover,thesignificance caregiver or child reports between children with teacher reports
of potential sex differences can also be calculated by obtaining and children without teacher reports (results are available upon
LRT, AIC, and BIC values from a model where A, C, and E request).
parametersareallowedtovarybysexwithamodelthatconstrains
theparameterstobeequalforboysandgirls.
The Unidimensionality of ASB in Childhood
Phenotypic correlations. We next tested whether each of the
Results
indices of antisocial behavior could be considered manifestations
Sex and Informant Differences in Mean Level ASB of a single higher order construct of externalizing behavior. We
examined this through both correlational and principal-
Descriptive statistics (means and standard deviations) for the components analysis of the various ASB measures obtained
variousratingscalesofaggressionanddelinquencyareprovidedin through each rater. Table 4 presents the full correlation matrix
Table 3, separately for caregiver, child, and teacher reports for (18 (cid:7) 18) for boys (above the diagonal) and girls (below the
boys and girls. These include proactive, reactive, and relational diagonal). Moderate to high correlations were found among the
aggression, measured using the CAQ; psychopathy Factor 1 scalesofaggressionanddelinquencywithineachrater,withcor-
(Callous–Unemotional) and Factor 2 (Impulsive–Irresponsible) relations ranging from .47 to .66 among child report measures,