Female and male antisocial trajectories: From childhood origins to adult outcomes

44 Pages·2008·0.95 MB·English

DevelopmentandPsychopathology20(2008),673–716
Copyright#2008CambridgeUniversityPress
PrintedintheUnitedStatesofAmerica
DOI:10.1017/S0954579408000333
Female and male antisocial trajectories:
From childhood origins to adult outcomes
CANDICEL.ODGERS,aTERRIEE.MOFFITT,b,c
JONATHANM.BROADBENT,dNIGELDICKSON,dROBERTJ.HANCOX,d
HONALEEHARRINGTON,eRICHIEPOULTON,dMALCOLMR.SEARS,f
W.MURRAYTHOMSON,dANDAVSHALOMCASPIb,c
aUniversityofCalifornia–Irvine;bKing’sCollege,London;cDukeUniversity;dUniversityof
Otago,NewZealand;eUniversityofWisconsin;and fMcMasterUniversity
Abstract
Thisarticlereportsonthechildhoodoriginsandadultoutcomesoffemaleversusmaleantisocialbehaviortrajectories
intheDunedinlongitudinalstudy.Fourantisocialbehaviortrajectorygroupswereidentifiedamongfemalesandmales
usinggeneralgrowthmixturemodelingandincludedlife-coursepersistent(LCP),adolescent-onset,childhood-limited,
andlowtrajectorygroups.Duringchildhood,bothLCPfemalesandmaleswerecharacterizedbysocial,familialand
neurodevelopmentalriskfactors,whereasthoseontheadolescent-onsetpathwaywerenot.Atage32,womenandmen
ontheLCPpathwaywereengaginginseriousviolenceandexperiencingsignificantmentalhealth,physicalhealth,and
economicproblems.Femalesandmalesontheadolescent-onsetpathwaywerealsoexperiencingdifficultiesatage32,
althoughtoalesserextent.AlthoughmoremalesthanfemalesfollowedtheLCPtrajectory,findingssupportsimilarities
acrossgenderwithrespecttodevelopmentaltrajectoriesofantisocialbehaviorandtheirassociatedchildhoodorigins
andadultconsequences.Implicationsfortheory,research,andpracticearediscussed.
This article tests whether the two prototypes andadolescencelimited(AL),canbeidentified
specifiedbyadevelopmentaltaxonomyofanti- withinaprospectivebirthcohortoffemalesand
social behavior, life-course persistent (LCP) males via general growth mixture modeling
(GGMM). Advanced longitudinal methods are
applied to test taxonomic predictions regarding
ThisworkwassupportedbygrantsfromtheUSNational developmental course, childhood origins, and
Institute of Mental Health (Grants MH45070 and adult consequences, with particularattention to
MH49414), UK Medical Research Council (G0100527),
the study of gender differences in antisocial
EconomicandSocialResearchCouncil,WilliamT.Grant
behavior.
Foundation,HealthResearchCouncilofNewZealand,So-
cialSciencesandHumanitiesResearchCouncilofCanada Untilrecently,interestandresearchdirected
(756-2004-0041),andUSNationalInstituteforDentaland at understanding trajectories of antisocial be-
Craniofacial Research (R01 DE-015260-01A1). Candice haviorhasfocusedprimarilyonmales(Lacourse,
L.OdgersisaMichaelSmithFoundationforHealthRe-
Nagin, Tremblay, Vitaro, & Claes, 2003;
searchTrainee.TerrieE.MoffittandAvshalomCaspiare
Schaeffer,Petras,Ialongo,Poduska,&Kellman,
Royal Society Wolfson Research Merit Award holders.
WethanktheDunedinstudymembers,Unitresearchstaff, 2003;Shaw,Lacourse,&Nagin,2005;Tremblay
Alan Taylor, Louise Arsenault, Daniel Nagin, and study etal.,2004;Wiesner&Capaldi,2003).Classic
founderPhilSilva. longitudinal studies in life-course criminology
Address correspondence and reprint requests to:
anddevelopmentalpsychologyhaveestablished
CandiceL.Odgers,DepartmentofPsychologyandSocial
theimportanceofchildhood-onsetconductprob-
Behavior, University of California–Irvine, Irvine, CA
92697-7085;E-mail:[email protected]. lemsandearlyinvolvementinantisocialbehavior
673
674 C.L.Odgersetal.
inpredictingfuturecrimeamongmales(Elliott, research has consistently shown that girls
Huizinga, & Menard, 1989; Farrington, 1989; have lower rates than boys of symptoms of
Farrington,1995;Wolfgang,Thornberry,&Fig- nervous system dysfunction, difficult tempera-
lio,1987).Overthelastdecade,researchgener- ment, hyperactivity, reading failure and learn-
atedbyadevelopmentaltaxonomyofantisocial ing disabilities (Gorman-Smith & Loeber,
behavior has refined our understanding of the 2005;Laheyetal.,2006;Messeretal.,2006).
importanceofchildhood-onsetbehavioralprob- Thus, the consequent processes of cumulative
lemsbydelineating,andtesting,expectationsre- continuity ensue for fewer girls than boys, re-
gardingtheuniquedevelopmentalcourse,child- sulting in a smaller number of girls following
hoodorigins,andadultprognosisformaleson theLCPpathway.TheALpathwayisalsohy-
the LCP versus AL pathways (Moffitt, 2006; pothesized to be open to females. According
Moffitt, Capsi, Harrington, & Milne, 2002). to the theory, girls, like boys, should begin
Theoriginalstatementofthetaxonomyasserted engaging in antisocial behavior soon after pu-
thatthetheoryshouldaccountforthebehavior berty to the extent to which they have access
offemalesaswellasitaccountsforthebehavior to antisocial role models and perceive the
ofmales(Moffitt,1994).Todate,however,the consequencesofantisocialbehaviorasreinforc-
majority of empirical research has focused on ing. Although girls, because of gender-typed
males, and we know relatively little about the socialization, may experience heightened per-
relevance of the developmental taxonomy to ceptions of serious personal risk associated
females. with involvement in antisocial behavior, it is
Inanutshell,theoriginaldevelopmentaltax- expectedthatfemalesshouldengageinALan-
onomy of antisocial behavior proposed that at tisocial behavior in significant numbers and
leasttwoprototypicalsubtypesunderlietheob- should resemble their male AL counterparts
served age by crime distribution (Moffitt, byhavingfewchildhooddeficitsandmanyde-
1993): an LCP pathway that is characterized linquent peers during adolescence (Moffitt,
bysocial,familial,andneurodevelopmentaldef- 2004). In short, the taxonomy predicted fe-
icits,onsetsinearlychildhood,anddistinguishes malesareseldomLCPbutareoftenadolescent
arelativelysmall,yetpersistentandpathological limited, and the childhood correlates of both
subgroupofindividuals,andanALpathwaythat subtypesareassumedtobesimilaracrossgen-
is hypothesized to be more common, relatively der(Moffittetal.,2001).
transient, and near normative. AL involvement Although still in its infancy, the study of
in antisocial behavior is believed to emerge girls’ involvement in antisocial behavior and
alongside puberty as a relatively normative re- aggression has grown significantly over the
sponse to the roleless years between biological last 2 decades (Giordano, Cernkovich, Stoff,
maturation and access to mature privileges and Breiling, & Maser, 1997; Moffitt et al., 2001;
responsibilities, a period of time labeled the Moretti, Odgers, & Jackson, 2004; Odgers &
“maturity gap.” Although those on the LCP Moretti,2002;Pepler,Madsen,Webster,&Le-
pathway are expected to experience multiple vene,2005;Putallaz&Bierman,2004;Serbin,
problems in adulthood, AL individuals, given Peters, McAffer, & Schwartzman, 1991; Un-
the normative nature of their preteen develop- derwood,2003).Asresearchersbegintopiece
ment, are hypothesized to be more successful together information from across diverse sam-
in their transition to adulthood, provided that ples, the question of whether female-specific
theydonotencountersnares,suchassubstance theories of antisocial behavior are required is
dependencyoracriminalrecord. beginning to come into focus. To date, how-
Theoriginalstatementofthetaxonomywas ever, there is no comprehensive theory of the
intended to apply to females as well as males developmentofantisocialbehaviorthatisspe-
(Moffitt,1994;Moffittetal.,2001).Withinthe cifictofemales.Becausethevastmajorityoffe-
developmental taxonomy, much of the gender maleantisocialbehavioronsetsinadolescence,
difference in levels of antisocial behavior is at- someresearchershavearguedthatachildhood-
tributedto gender differences in the individual onsetLCPsubtypemaynotexistforgirlsand,
risk factors for persistent antisocial behavior; instead, only an adolescent-onset subtype is
Femaleandmaleantisocialtrajectories 675
requiredtocharacterizegirls’antisocialbehav- summarized the growing bodyof empirical
ior(Silverthorn&Frick,1999).Althoughnew evidence for antisocial subgroups not orig-
research is still emerging, to date, virtually all inally specified by the taxonomy; these
epidemiological studies testing whether gen- subgroups, labeled, “childhood limited”
der-specific pathways of antisocial behavior (Farrington, Gallagher, Morley, Stledger,
exist have identified a “childhood-onset” or & West, 1988; Moffitt, Caspi, Dickman,
“early-starter”pathwayamongfemales(Bong- Silva,&Stanton,1996;Wiesner&Capaldi,
ers, Koot, van der Ende, & Verhulst, 2004; 2003) and “low-level chronic” (D’Unger,
Broidy et al., 2003; Cote´, Tremblay, Nagin, Land, McCall, & Nagin, 1998; Nagin, Far-
Zoccolillo, & Vitaro,2002;Fergusson &Hor- rington, & Moffitt, 1995) have emerged
wood, 2002; Lahey et al., 2006; Schaeffer across a number of longitudinal studies,
et al., 2006). For example, in an analysis of butdonotyethavewell-articulatedtheories
epidemiological samples from Canada, New regarding etiology, developmental course,
Zealand,andtheUnitedStates,Broidyandcol- and prognosis. Thus, the present study ap-
leagues (2003) identified an early-onset path- plied the latest generation of trajectory-
wayofgirls,basedonteacherreportedaggres- basedmodelingtechniques,GGM(Muthe´n,
sion, in three of the four samples. Similarly, 2004),totestwhethertheoriginalsubtypes
Cote´ and colleagues, in a representative study ofLCPandALantisocialbehavioremerged,
of Canadian girls, identified two early-starter withaneyetowardinvestigatingwhetheran
pathways of antisocial behavior (aggressive additional subgroup(s) wasrequired.More-
plus oppositional behaviors) that went on to over, prior research has not tested whether
have higher rates of conduct disorder (CD) thesamebehaviorstaptheconstructofanti-
than later starters. More recently,both Schaef- social behavior in the same way across de-
fer and colleagues (2006) and Lahey and col- velopment and gender. In other words, it
leagues (2006) reported finding early-starter has been assumed that administering the
subgroupsoffemaleswhodemonstratedchroni- samemeasureacrosstime,andacrossmales
cally high levels of antisocial behavior across and females, provides a common valid
childhood and early adolescence; these early- metric for assessing antisocial behavior. In
starter girls, like their male counterparts, were thecurrentstudy,weempiricallytestthisas-
atanincreasedriskforlaterantisocialoutcomes. sumptionandevaluatewhetherourmeasure
Although initial evidence suggests that an of antisocial behavior is invariant across
early-onset pathway for girls may exist, there both time and gender prior to mapping de-
has not yet been a comprehensive test of the velopmentaltrajectories.
extension of the developmental taxonomy of 2. Childhood origins: Are the childhood ori-
antisocial behavior to females with respect to ginsfortheLCPandALsubgroupsconsis-
developmental course, childhood origins, and tent with expectations from the taxonomy?
adult outcomes. In the present study we tested That is, similar to their male counterparts,
whether predictions stemming from the devel- females on the LCP pathway are expected
opmental taxonomy hold in a birth cohort of to be characterized by social, familial, and
femalesandmalesthathavebeenfollowedpro- neurodevelopmental deficits in childhood.
spectivelyuntilage32.Withaspecificempha- Incontrast,thoseontheALpathwayshould
sisonage32findings,weaddressedthreesets score closer to the low antisocial subgroup
ofquestions: (or average Dunedin child) on childhood
risk factors. Previous work with males in
1. Developmental course: Do the three sub- the Dunedin study has demonstrated that
groups anticipated by the developmental LCPantisocialbehaviorisdifferentiallypre-
taxonomy,namely,LCP,AL,andlowanti- dictedbythesechildhoodriskfactors(Mof-
social pathways, emerge among females fitt& Caspi,2001),with the same riskfac-
and males in a 30-year prospective birth tors documented for the small number of
cohort study? Is an additional subgroup females classified as belonging to the LCP
needed? A recent review (Moffitt, 2006) pathway (Moffitt et al., 2001). However,
676 C.L.Odgersetal.
prior research has relied on classifications informed consent at ages 5, 7, 9, 11, 13, 15,
derived via clinical algorithms. In the pre- 18, 21, 26, and 32 years of age, when 96% of
sentstudy,subtypesaredefinedusingtrajec- the living Study members were assessed in
tory-based statistical models and validated 2003–2005. Cross-national comparisons lend
using a subset of childhood risk indicators confidenceregardingthegeneralizationoffind-
selected from extensive analyses of earlier ings from the Dunedin study to other industri-
wavesoftheDunedinStudy. alizednations(Moffittetal.,2001).
3. Adult outcomes: Do predictions stemming
from the taxonomic theory regarding adult
Measures
outcomes demonstrate predictive validity
among females? Specifically, do women Antisocialconductproblemsweremeasuredat
on the LCP pathway experience the worst ages 7, 9, 11, 13, 15, 18, 21, and 26 years
adult consequences at age 32? Prior work through scoring six keysymptoms of DSM-IV
with Dunedin females has demonstrated CDasbeingpresentorabsentateachage:phys-
that CD during adolescence predicted a ical fighting, bullying others, destroying prop-
wide range of negative outcomes in early erty, telling lies, truancy, and stealing (Ameri-
adulthood (Bardone, Moffitt, Caspi, & canPsychiatricAssociation,1994).Acomposite
Dickinson,1996)andpreviouspublications score,rangingfrom0to6,wasformedateach
report Dunedin males’ outcomes age 26 assessmentagerepresentingthenumberofdif-
(Moffitt et al., 2002) and age 32 (Odgers, ferent types of antisocial behavior the individ-
Caspi,etal.,2007);however,itisnotknown ualhadengagedinduringthepastyear(antiso-
whetheradultoutcomesvaryacrossdistinct cialconductproblems“varietyscore”).Variety
developmental subtypes of antisocial fe- scores are highly correlated with frequency
males or whether poor prognosis in adult- scores(howoftenthechildexhibitedantisocial
hood extends to include poor physical behavior in the past year) and are commonly
health. Thus, this paper differs from prior used in population-based studies. Other DSM-
publications in that it (a) focuses on devel- IV CD symptoms were not used because they
opmental trajectories for females, along didnotcoverthestudy’sagespan(e.g.,running
with male comparisons; (b) formally tests away, staying out late) or had very rare preva-
whether antisocial behavior is being mea- lence (e.g., fire setting, forced gender, animal
sured in the same way across males and cruelty). Each of the six conduct problem
females; and(c) expandsthe rangeof adult symptoms was operationalized through multi-
outcomestotheageof32toincludeassess- ple items collected at each age: the symptom
ments of violence, mental health, physical was considered present if any item in the set
health,andeconomicproblemsamongboth was endorsed by a reporter. Symptoms were
malesandfemales. adapted across the age span to ensure that the
measures were developmentally appropriate.
“Truancy” included items such as skipping
Method
schoolforyoungerstudentsandworkabsentee-
ismforolderemployedStudymembers;“bully-
Participants
ing”includeditemssuchasbullyingotherchil-
ParticipantsaremembersoftheDunedinMulti- dren, threatening violence, and at older ages,
disciplinary Health and Development Study. robbery; “stealing” included items such as
The cohort of 1,037 children (52% male) was stealing from school or home, shoplifting,
constitutedat3yearsofage,wheninvestigators auto theft, burglary, absconding from a rental
enrolled 91%ofconsecutiveeligiblebirthsbe- with unpaid bills or rent, and embezzlement
tweenApril1972andMarch1973inDunedin, from employers. “Fighting” included items
NewZealand.Cohortfamiliesrepresentthefull such fights with otherchildren, fighting in the
range of socioeconomic status (SES) in New street, gang fighting, and assault. “Telling
Zealand’sSouthIslandandareprimarilyWhite. lies”includeditemssuchastellsliestoparents
Follow-up assessments were conducted with and teachers, lying about their age, and
Femaleandmaleantisocialtrajectories 677
providing false information on job or loan containeditemssuchas,“Inourfamily,webe-
applications. “Destroying property” included lieve you don’t ever get anywhere by raising
items such as purposely destroying or dam- your voice” and “Family members sometimes
aging other’s property. The study’s reporting hiteachother.”
sourceswerealsodevelopmentallyappropriate, Inconsistentdisciplinewasmeasuredatages
includingparentandteacherinchildhood;self, 7and 9aspartofaninterviewabouthowpar-
parent, and teacher in adolescence; and self ents dealt with the study child when he or she
alone in adulthood. The prevalence of each misbehaved. Mothers evaluated their own dis-
CD symptom and average scale scores by age cipline, as well as their husband’s discipline
andgenderareincludedinTable1;withtheex- on a 4-point scale (1 ¼ always the same; 4 ¼
ception of age 15, males scored significantly verychangeable).
higherthanfemalesonantisocialconductatevery
assessmentage.Thereliabilityvaluesforthescale
Parentalfeatures
rangedfroma..60betweenages7and15and
a..50betweenages18and26. Mother’s mental health problems were mea-
sured with the Malaise Inventory, a 24-item
questionnairethatwascompletedbytheStudy
Childhoodpredictors
members’ mothers when the study members
Each childhood measure is described briefly, were 7 and 9. The questionnaire (Rodgers,
accompanied bya reference that reports details Pickles, Power, Collishaw,& Maughan, 1999;
ofdatacollection,variableconstruction,reliabil- Rutter, Tizard, & Whitmore, 1970) samples a
ity,andvalidity.Themeasuresinthiscohortare varietyofcommonsymptomsofemotionaldis-
described in Moffitt et al. (2001, 2002) unless turbance,andisheavilyweightedwithitemsre-
otherwisespecified.Allmeasureshaveareliabil- flecting affective stress response (e.g., easily
ityof..70,asassessedbyinternalconsistency, upset, miserable) and somatic symptoms (e.g.,
test–retest,orinterrateranalysis,as appropriate. tiredness,headaches).
The childhood measures listed below were Mother’s IQ was tested using the Science
selected based on extensive analyses of earlier Research Associates (SRA) verbal test (Thur-
wavesofdataintheDunedinStudy. stone & Thurstone, 1973) when the children
were age 3; standardized to population (M ¼
100, SD ¼ 15). Low mother IQ was defined
Familycharacteristicsandcontext
as,85onthestandardizedSRAscore.
SESwasmeasuredasthehighestoffather’sor Parentcriminalconvictionwasmeasuredby
mother’s occupation using a 6-point scale for parental report in 1998, when parents’ ages
New Zealand (Elley & Irving, 1976); 21% of rangedfrom40to75.Ofparents,12%reported
the families were classified as low SES, 63% theyhadbeenconvictedinthecriminalcourts.
asmediumSES,and16%ashighSES.
Maltreatment was measured using staff ob-
Childfactors
servationsofrejectingmother–childinteraction
at age3, parental reports of harshdiscipline at ChildIQwastestedusingtheWechslerIntelli-
ages7and 9,twoor more changesin primary gence Scale for Children—Revised (WISC-R;
caregiver to age 11, and retrospective reports Wechsler, 1974) at ages 7, 9, 11, and 13, and
bystudymembersatage26ofinjuriousphys- thefourvalueswereaveragedtoenhancerelia-
ical abuse or unwanted sexual contact before bility;standardizedtopopulation(M¼100,SD
age 11. Nine percent of boys and girls had ¼15).LowchildIQwasdefinedas,85onthe
twoor more indicatorsof maltreatment (Caspi standardizedWISC-Rscore.
etal.,2002). Undercontrolled temperament was mea-
Family conflict was measured at ages 7 sured through staff ratings after observing the
and 9 with the Moos Family Relations Index childina90-mintestingsessionwithanunfa-
(Moos & Moos, 1981) completed by mothers miliarexamineratage3.Factorandclusteranaly-
of the study members. The conflict subscale sesreducedthese ratingsto three temperament
Table1. Frequencies of antisocial conduct problem items byageand gender
Age
7 9 11 13 15 18 21 26 32
AntisocialCPItems (%) (%) (%) (%) (%) (%) (%) (%) (%)
Fight
Males 59.1 55.5 44.3 62.1 43.0 52.2 37.5 29.1 11.0
Females 47.5 41.9 35.3 47.8 38.0 37.9 29.7 24.4 10.1
Destroy
Males 27.4 24.9 20.8 23.5 25.7 13.0 18.9 9.0 6.9
Females 16.8 9.5 7.2 8.1 16.3 4.8 5.1 1.0 1.3
Lie
Males 45.1 41.1 39.2 29.0 31.4 72.7 46.3 52.2 43.8
Females 35.5 27.3 27.0 25.5 30.0 64.8 25.4 42.6 30.9
6 Steal
7
8 Males 24.4 25.3 25.6 21.3 24.5 23.3 27.0 31.9 21.5
Females 18.5 12.1 15.3 15.0 20.9 11.4 16.2 21.4 10.5
Truant
Males 3.0 5.1 4.6 13.2 29.4 6.7 11.0 12.4 —
Females 2.0 1.5 1.6 7.0 32.8 7.2 8.8 9.6 —
Bully/rob
Males 31.9 32.6 26.0 22.7 17.4 2.5 1.0 0.6 0.0
Females 26.6 20.2 22.3 17.0 14.6 0.7 0.0 0.0 0.0
7 9 11 13 15 18 21 26 32
AntisocialCPScale M(SD) M(SD) M(SD) M(SD) M(SD) M(SD) M(SD) M(SD) M(SD)
Males 1.91(1.60) 1.84(1.71) 1.66(1.68) 1.73(1.61) 1.70(1.55) 1.70(1.27) 1.41(1.34) 1.48(1.31) 0.83(1.00)
Females 1.47(1.48) 1.13(1.34) 1.14(1.37) 1.22(1.61) 1.52(1.55) 1.27(1.03) 0.84(1.00) 1.04(1.05) 0.53(0.70)
Note:CP,conductproblems.Truancywasnotassessedatage32;thedifferencesbetweenantisocialconductproblemscalescoresformalesversusfemalesarestatisticallysignificant(p,.001)at
allages,exceptforage15(p¼.08).
Femaleandmaleantisocialtrajectories 679
types, including the undercontrolled type applies).Itemswereaveragedtocreateanover-
(Caspi&Silva,1995),sincereplicatedinother allpeerdelinquencyscore.
samples (Asendorpf, Borkenau, Ostendorf, &
VanAken,2001;Hart,Atkins,&Fegley,2003;
Age32outcomes
Robins, John, Caspi, Moffitt, & Silva, 1996).
Full psychometric details are provided else- Each age 32 outcome is described briefly, ac-
where(Caspi,2000). companied by a reference that reports details
Heart ratewas measured by nurse examin- of data collection, variable construction, relia-
ers at ages 7, 9, and 11. At each age, an aver- bilityandvalidity.
age heart rate measurewas derived from mea-
sures of resting heart rate taken by a nurse on
Age32violencetowardothers
threeoccasionsduringthecourseofthephysi-
cal examination. The (age standardized) mea- Violencetowardotherswasselectedasthepri-
suresofrestingheartratefromthethreeagepe- mary age 32 index of antisocial behavior as it
riodswereaveragedtoformtheoverallscore. represents the most serious form of antisocial
Readingachievementwasmeasuredatages behaviorand is recognized as one of the most
7, 9, and 11 by the Burt Word Reading Test significantglobalhealthproblems(Krug,Dahl-
(Scottish Council for Research in Education, berg,&Mercy,2002).
1976), a word recognition test having norma- Partnerabuseinthepastyearatage32was
tivestandardsforNewZealandchildren,which measuredinastandardizedinterviewabout13
resemblestheAmericanWide-RangeAchieve- physical abuse acts (e.g., slapping, strangling,
ment Test of reading. The (age standardized) kicking, hitting, beating up, forcing sex, and
readingscoresfromthethreeageperiodswere using a weapon) and 13 controlling abuse
combinedtoformanoverallscore. acts (e.g., damaging clothes, car or pet; stop-
Attention-deficit/hyperactivity disorder (ADHD) ping contact with family or friends; stalking).
was measured using the Diagnostic Interview Dunedin men’s and women’s self-reports
for Children—Child Version (Costello, Edel- have been previously validated against their
brock, Kalas, Kessler,& Klaric, 1982) at ages partners’ reports and found to be reliable and
11,13,and15.Diagnosesweremadeaccording valid (Moffitt et al., 1997). Study members
to the Diagnostic and Statistical Manual of who reported any perpetration of physical
Mental Disorders—Third Edition (DSM-III; abuse within a relationship were classified as
American Psychiatric Association, 1980) and engaging in physical abuse. Study members
confirmed through parent or teacher report in- who engaged in two or more types of control-
cluding age of onset before age 7; 6% of the ling behavior within the past year were classi-
cohort was diagnosed. Dunedin was the first fiedasengagingincontrollingabuse.
cohortstudytouseastandarddiagnosticinter- HittingachildwasassessedduringtheSelf-
viewin1983whenstudymemberswere11years Report Crime Interview. Study members were
old. Therefore, information on ADHD symp- askedoneitemabouthittingorotherwisehurt-
tomsreportedbymothersandteachersonscales ingachildoutofanger(Moffittetal.,2002).
at ages 5 and 7 was used to confirm symptom Self-reportedviolenceinthepastyearatage
onsetbeforetheageof7forDSMdiagnosis. 32wasmeasuredusingtheUSNationalYouth
Peer delinquency was assessed using 10 Survey Self-Report Crime Interview (Elliott
items from the Revised Problem Behavior et al., 1989). Items ascertained simple assault,
Checklist (Quay, 1983; Quay & Peterson, aggravated assault, gang fighting, robbery, ar-
1993) that was completed by the Study mem- son, and forced sex (Moffitt et al., 2002).
bers’ mothers when the study members were Assaultsagainstpartnersandchildrenwereex-
15. The peer delinquency subscale contained cluded to avoid overlap with the abovemen-
10items,suchas“belongstoagang”and“as- tionedmeasures.Studymemberswhoreported
sociates with rougher peers,” measured on a engaging in at least one act of violencewithin
3-point scale (0 ¼ no, does not apply, 1 ¼ the last year were classified as engaging in
yes, applies somewhat, 2 ¼ yes, certainly self-reportedviolenceatage32.
680 C.L.Odgersetal.
Informant-reportedfightinginthepastyear rated the study member’s impairment on three
atage32wasmeasuredbymailingabriefques- symptoms of anxiety (e.g., “has unreasonable
tionnaire to people nominated by the Study fearsorworries,”“worriesalot,”“getsnervous
member as knowing him/her well (informants easily”),andfoursymptomsofdepression(e.g.,
included friends, partners and family mem- “feelsthat no one lovesthem,” “seems lonely,”
bers). Full details of the Dunedin Study infor- “feels depressed, miserable, sad, or unhappy,”
mant rating system are provided elsewhere “talks about suicide”) using a 3-point scale
(Moffitt et al., 2002). Information from infor- (0¼no,doesnotapply,1¼yes,appliessome-
mantswasavailablefor96%ofstudymembers what, 2 ¼ yes, certainly applies). The seven
seen at age 32. Informants were asked to rate symptomswerecombinedtoformaninternaliz-
whether the study member “got into fights” in ing scale; individuals scoring 1 SD above the
the last 12 months (0 ¼ not a problem, 1 ¼ mean were classified as experiencing infor-
bit of a problem, 2 ¼ yes, a problem); study mant-rated internalizing problems (Moffitt
membersratedas1or2wereclassifiedas“get- etal.,2002).
tingintofights”atage32. Informantreportsofsubstanceuseproblems
Official violence convictions between ages were also collected using the informant rating
26and32weremeasuredbysearchingthecom- system referenced above. Informants rated the
puterized New Zealand Police database. Con- studymemberontwoitems(e.g.,“hasalcohol
victionsincluded,butwerenotlimitedto:com- problems,”“hasmarijuanaorotherdrugprob-
monassault,commondomesticassault,assault lems”)usinga3-pointscale(0¼no,doesnot
ofchild,assault with aweapon,rape,indecent apply,1¼yes,appliessomewhat,2¼yes,cer-
assault on female, robbery aggravated with a tainlyapplies). Study members rated as 1or 2
firearm,maleassaultsfemalewithweapon,re- oneitheritemwereratedasexperiencinginfor-
sistingpolice,andarson. mant-rated substance use problems (Moffitt
etal.,2002).
Age32mentalhealth
Age32physicalhealth
Psychiatricdisordersduringthepastyearatage
32 were assessed in private structured inter- Cardiovascular disease (CVD) risk. Because
viewsusingtheDiagnosticInterviewSchedule the cohort is still too young to present clinical
(DIS; Robins, Cottler, Buckholz, & Compton, endpointsofcardiovasculardisease(e.g.,myo-
1995).DiagnosesweremadeaccordingtoDSM- cardialinfarction),wefocusedonmultiplerisk-
IV criteria (American Psychiatric Association, factorclusteringasameasureofcardiovascular
1994). Prevalence rates in the Dunedin cohort riskasrecommendedbychronicdiseaseepide-
are similar to those from American epidemio- miologists (Grundy, Posternak, Greenland,
logical surveys (Kessler, McGonagle, Swartz, Smith,&Fuster,1999;Munoz&Gange,1998).
Blazer,&Nelson,1993;Newmanetal.,1996). Six biomarkers were used: overweight, high
For this report, we examined grouped anxiety blood pressure, elevated total cholesterol, low
disorders(generalizedanxietydisorder,obses- high-density cholesterol, elevated glycated he-
sive–compulsivedisorder,panicdisorder,ago- moglobin, and low maximal oxygen uptake.
raphobia, social phobia, and simple phobia), Study members were “clustered” if they had
major depressive disorder, cannabis depen- atleastthreeoftheaforementionedriskfactors.
dence,dependenceonotherdrugs,alcoholde- Clinical definitions and the construction of
pendence,andposttraumaticstressdisorder. each measure are reported elsewhere (Caspi,
Suicideattemptsinthe5yearsbetweenages Harrington,Milne,Moffitt,&Paulton,2006).
26and32weremeasuredusingaLifeHistory High-sensitivityC-reactiveprotein(hsCRP)
Calendar (Belli, Shay&Stafford, 2001;Caspi isthoughtbeoneofthemostreliablemeasured
etal.,1996). indicators of vascular inflammation (Ridker,
Informant reports of internalizing symptoms Wilson,&Grundy,2004)andhasbeenrecently
andsubstanceusewerecollectedusingtheinfor- endorsedasanadjuncttotraditionalriskfactor
mantratingsystemreferencedabove.Informants screeningforcardiovascularriskbytheCenters
Femaleandmaleantisocialtrajectories 681
for Disease Control and Prevention and the or more sites with (cid:2)4-mm combined attach-
American Heart Association (Pearson et al., mentloss.
2004; Ridker et al., 2004). Individuals with
hsCRP higher than 3.0mg/l were considered Injuries.Studymembersreportedseriousinju-
at high risk (Ridker et al., 2004). Full details riesbetweenages26and32,definedasanyre-
ofthismeasurearereportedelsewhere(Danese, quiringtreatmentfromadoctor,medicalcenter,
Pariante,Caspi,Taylor,&Poulton,2007). or emergency services. We report the per-
Respiratory function was assessed using a centagewhoexperiencedaninjuryand,among
computerizedspirometerandbodyplethysmo- these individuals, the percentage with a non-
graph;technicaldetailsareprovidedelsewhere sport-relatedinjury.
(Tayloretal.,2002).Measurementsofvitalca-
pacity (VC) were repeated to obtain at least
Age32economicproblems
three repeatable values (within 5%) followed
by full-forced expiratory maneuvers to record SES. Study members were asked about their
the forced expiratory volume in 1s (FEV ) currentormostrecentoccupation;homemakers
1 :
The FEV /VC ratio is reported as the primary andthosewhowerenotworking(e.g.,students)
1
lung function measure because it is the most wereproratedbasedontheireducationalstatus
sensitive measure forassessing airway remod- accordingtothecriteriaincludedinthecurrent
eling in a large cohort (Rasmussen et al., New Zealand Socioeconomic Index (Davis,
2002).Studymembersalsoself-reportedsymp- Jenkin, & Coope, 2003). This information
toms of chronic bronchitis: chronic coughing was coded to a 6-point scale for occupations
andphlegm(Searsetal.,2003).Studymembers in New Zealand; 31% of individuals scored in
who reported problems with coughing or the lowest two groups on this scale and were
phlegmatage32wereclassifiedasexperienc- classifiedaslowSES.
ingsymptomsofchronicbronchitis.
Householdincome.Sourcesofincomewereas-
Sexual health. Serologicalevidenceofinfection certained (e.g., wages, self-employment, gam-
at age 32 with herpes simplex virus type 2 bling winnings, interest, rent collected, and
(HSV-2), the most common cause of genital loans)andusedasanaidtocalculatetotalgross
herpes, was obtained using an indirect enzyme past-yearincome.Insteadofpresentingtheraw
linked immunosorbent assay (HerpeSelect 2 datainlocalcurrency,wereportthepercentage
ELISA IgG; Focus Technologies, Chanhassen, of study members falling below the median
MN).HSV-2infectionwasdiagnosedusingacut- splitonhouseholdincome.
offvalueof3.5andanyequivocalresult(between
0.9and3.5)wasresolvedusingHSV-2Western Unemployed. Months of unemployment be-
blot (Ho, Field, Irving, Packham, & Cumming- tween the ages of 26 and 32 (defined as not
ham, 1993). Full details of this measure are re- working, not a student or homemaker, and
portedelsewhere(Eberhart-Phillipsetal.,2001). looking for work) was recorded using a Life
Smokingduringthepastyearatage32was History Calendar (Belli et al., 2001; Caspi
assessed as part of the DIS (Robins et al., et al., 1996). Study members who spent 1 or
1995).Tobaccodependencewasdiagnosedac- more months unemployed between the ages
cordingtoDSM-IVcriteria. of26and32wereclassifiedasunemployed.
Noeducationalqualificationwasdefinedas
Dental health. Examinations were conducted ending secondary education prior to receiving
using calibrated dental examiners in all four qualifications,andnotreturningtoearnqualifi-
quadrants of the mouth; technical procedures cations byage 32. Qualifications are based on
aredescribedelsewhere(Broadbent,Thomson, nationalexamsthatalmostallstudentstakeby
& Poulton, 2006; Thomson, Broadbent, Poul- age16,whichdeterminepromotioninsecond-
ton, & Beck, 2006). We report the number of ary school and technical schools; passing this
untreated decayed surfaces present at age 32 exam also helps secure better employment in
and presence of gum disease defined as two thelabormarket (Kennedy,1981).
682 C.L.Odgersetal.
Informant-rated financial problems were subgroups in light of expectations regarding
collectedusingtheinformantratingsystemde- childhoodoriginsand(b)testthepredictiveva-
scribedabove.Informantsratedthestudymem- lidity of the taxonomy for females and males
berontwoitems(e.g.,“poormoneymanager” basedonage32outcomes.
or“lackedenoughmoneytomakeendsmeet”)
usinga 3-point scale (0¼not aproblem,1¼
IstheAntisocialConductProblemsScale
bit of a problem, 2 ¼ yes, a problem). Study
InvariantAcrossAgeandGender?
members rated as 1 or 2 on either item were
classified as experiencing informant-rated fi-
Method
nancialproblems(Moffittetal.,2002).
Thefirststepingrowthmodelingistoestablish
No money for food orother necessities. Study that the same construct has been measured
members were asked “since you were 26, did across time. A classic example is charting the
you ever find it difficult to meet the cost of rate of a child’s growth in height across age.
foodandothernecessities?”(0¼no,1¼some- In this case, there is a common metric (e.g.,
times,2¼yes).Studymemberswhoreceiveda inches) available to objectively assess how
scoreof1or2onthisitemwereclassifiedasex- muchthechildhasgrown.Whenpsychological
periencing difficulty meeting the cost of food constructs are measured, however, we cannot
andothernecessities. assume that the same “ruler” or metric has
Homeless/taken in between ages 26 and 32 beenusedateachage.Therefore,wemustfirst
was measured using a Life History Calendar. test whether the scale demonstrates MI across
Study members reported periods when they ages. Establishing MI provides evidence for a
were homeless or were taken in by friends or common quantitative metric and helps to en-
relatives because they had no place to live sure that we are not comparing “apples” to
(Wright, Caspi, Moffit, & Silva, 1998). Study “oranges” when mapping development across
members who spent 1 or more months home- time(McArdle,1996).
less or taken in were classified as homeless/ Invarianceacrossageisnottheonlymeasure-
takenin. ment concern when interpreting results from
growth-modeling analyses. It is also possible
thattheAntisocialConductProblemsScaleoper-
MethodsandResults
atesdifferently,ordemonstratesbias,acrossmales
Resultsarepresentedintwoparts.Thefirstpart versus females. Thus, to ensure that our results
applies multiple-group confirmatory factor were comparable across males and females we
analysis(M-GroupCFA)totestwhethertheAnti- alsotestedforMIacrossgender.Priorworkhas
social Conduct Problems Scale demonstrated assumed,butnottestedfor,MIpriortomapping
measurement invariance (MI) across age and developmentaltrajectoriesofantisocialbehavior.
gender; that is, whether the same construct MI of the Antisocial Conduct Problems
was measured (a) over time and (b) across Scale was assessed through M-Group CFA, a
males and females. Technically, MI refers to technique that compares groups of individuals
the invariant operation of items over time or on latent variables underlying item sets (Sor-
acrosssamples;thatis,theextenttowhichitem bom,1974).M-GroupCFAhasbeenextended
content is being perceived and interpreted ex- to apply to both continuous and ordered cate-
actlythesamewayacrosstimeoracrossgroups gorical outcomes (Lubke & Muthe´n, 2004).
(Byrne & Watkins, 2003). The second part of TodemonstrateMI,theregressionrelationsbe-
the results applies GGMM to test whether the tween the observed items and the underlying
antisocial subgroups anticipated by the devel- factorsareconstrainedtobeequalacrossmulti-
opmentaltaxonomyemergedwithinmalesand plegroups(Meredith,1993).WhentestingMI
females.Standardtechniquesfortheanalysisof acrossage,eachassessmentagewastreatedas
between-groupdifferences(e.g.,analysisofvar- a separate group (see Figure 1 where Group 1
iance,logisticregressionwithplannedcontrasts) ¼ age 7, Group 2 ¼ age 9 . . .); when testing
were then applied to (a) externally validate the for MI across gender, females and males were