Is antisocial personality disorder continuous or categorical? A

12 Pages·2006·0.2 MB·English

PsychologicalMedicine,2006,36,1571–1581. f2006CambridgeUniversityPress
doi:10.1017/S0033291706008245 PrintedintheUnitedKingdom
Is antisocial personality disorder continuous or
categorical? A taxometric analysis
DAVID K. MARCUS1*, SCOTT O. LILIENFELD2, JOHN F. EDENS3
AND NORMAN G. POYTHRESS4
1DepartmentofPsychology,UniversityofSouthernMississippi,Hattiesburg,MS,USA;2Departmentof
Psychology,EmoryUniversity,Atlanta,GA,USA;3DepartmentofPsychology,SouthernMethodist
University,Dallas,TX,USA;4DepartmentofMentalHealthLawandPolicy,UniversityofSouthFlorida,
Tampa,FL,USA
ABSTRACT
Background. Although the DSM-IV-TR is organized into discrete disorders, the question of
whetheragivendisorderpossessesadimensionaloracategoricallatentstructureisanempiricalone
that can be examined using taxometric methods. The objective of this study was to ascertain the
latentstructureofantisocialpersonalitydisorder(ASPD).
Method. Participants were 1146 male offenders incarcerated in state prisons (n=569), or court-
orderedtoresidentialdrugtreatment(n=577).ParticipantswereinterviewedusingtheStructured
ClinicalInterviewforDSM-IVAxisIIPersonalityDisorders(SCID-II)toassessASPDsymptoms;
theyalsocompletedthePersonalityDiagnosticQuestionnaire-4(PDQ-4)ASPDscale.Taxometric
analyses were performed to examine whether ASPD is underpinned by a discrete category or a
dimensionalconstruct.
Results. Multipletaxometricproceduresusingtwodifferentsetsofindicatorsprovidednoevidence
that ASPD has a taxonic latent structure. Instead, the results were far more consistent with the
propositionthatASPDexistsonacontinuum,regardlessofwhetheritisassessedusingastructured
intervieworaself-reportmeasure.
Conclusions. Evidence that ASPD is dimensional suggests that it is best studied using continuous
measures and that dichotomizing individuals into ASPD versus non-ASPD groups will typically
resultindecreasedstatisticalpower.Thefindingsarealsoconsistentwithamultifactorialetiology
for ASPD and with recent attempts to conceptualize ASPD within the framework of extant
dimensionalmodelsofpersonality.
INTRODUCTION hypertension)? The authors of the Diagnostic
and Statistical Manual of Mental Disorders
A fundamental question in the classification of
(4th edn, text revision) (DSM-IV-TR; APA,
psychiatricdisordersconcernstheirlatentstruc-
2000) sidestepped this issue by stating that
ture. Do these disorders identify discrete con-
‘there is no assumption that each category of
ditions (such as influenza) or dimensions that
mental disorder is a completely discrete entity
vary quantitatively but not qualitatively from
withabsoluteboundariesdividingitfromother
non-pathological conditions (such as essential
mental disorders or from no mental disorder’
(p. xxxi). Nonetheless, for largely pragmatic
* Addressforcorrespondence:DrD.K.Marcus,Departmentof
purposes the DSM-IV-TR is organized into
Psychology, University of Southern Mississippi, 118 College
Dr.#5025,Hattiesburg,MS39406-0001,USA. discretedisorders.
(Email:[email protected]) For much of the history of psychiatric nos-
An earlier version of this research was presented at the 2005 ology, the question of latent structure has been
meetingoftheAmericanPsychology-LawSociety,SanDiego,CA,
USA a source of debate with few data. However,
1571
1572 D.K.Marcusetal.
taxometricprocedures developed by Meehl and personalitydisorder,whichappearstobecloser
colleagues(Meehl&Yonce,1994,1996;Waller to psychopathy in its emphasis on affective and
&Meehl,1998)allowresearcherstotestwhether interpersonal deficits) is likely to be a hetero-
a diagnosis has a taxonic (categorical) or a geneous category that includes psychopathic
dimensional (continuous) latent structure. Al- individuals, dissocial individuals influenced by
thoughthisresearchisstillrelativelypreliminary, deviantsubgroupnorms,andavarietyofother
studies suggest, for example, that schizotypy criminal individuals (Lilienfeld, 1998; Lykken,
(Lenzenweger & Korfine, 1992) may be categ- 1995), making it unlikely that the broad diag-
orical,whereasmanyformsofdepression(Slade nosisofASPDcouldbeadiscretecategory.
& Andrews, 2004) are likely to be dimensional. Taxometric studies by Harris and his associ-
In all likelihood some DSM-IV-TR disorders ates (Harris etal.1994;Skilling etal.2002) ap-
will ultimately prove to have a categorical pearedtoprovideevidenceforanASPDtaxon;
structure,whereasotherswillbedimensional. both studies used similar measures and over-
Theorists have often assumed that many, if lapping samples. However, these studies had
not most, of the personality disorders have a significant methodological limitations that in-
dimensional latent structure. Attempts to con- creased the risk of ‘pseudo-taxonic’ results, in-
ceptualize personality disorders within the ru- cluding(a)thefailuretoincludeinterviewdata,
bric of the five-factor model of personality are which is the preferred method of scoring of the
consistentwiththisassumption(Widiger,2005). Psychopathy Checklist-Revised (PCL-R), (b)
Surprisingly, a recent review of taxometric dichotomizing PCL-R scores instead of using
studiesofthreepersonalitydisordersconcluded the standard three-point scoring system, (c)
that the research favors a categorical structure item-total correlations that were considerably
for schizotypal, antisocial and borderline per- higher than those reported in the PCL-R
sonalitydisorders(Haslam,2003). manual (raising the possibility that the PCL-R
However,recenttaxometricresearchindicates may have been scored with an expectation of a
that borderline personality disorder is more psychopathy taxon), and (d) the use of large
likely to have a dimensional latent structure numbers of mentally ill inmates, who may
(Rothschild et al. 2003). Similarly, the picture have inadvertently yielded a schizotypy taxon
maybemorecomplicatedforantisocialperson- (Lilienfeld,1998;Edensetal.2006).
ality disorder (ASPD), particularly because of Additionally, although much of the evidence
the ambiguity regarding the conceptualizations foran ASPD taxoncomes from theanalysisby
of ASPD and the related construct of psy- Harris et al. of items drawn from Factor II of
chopathy. Although the accompanying text of the PCL-R, which is the factor that assesses
DSM-IV-TR asserts that ASPD is essentially antisocial behavior, several subsequent taxo-
synonymous with psychopathy (p. 702), nu- metricsstudiesfailedtofindevidenceforataxon
merous studies indicate that measures of these forpsychopathyorfortheconstructassessedby
two conditions are only moderately correlated PCL-R Factor II (Marcus et al. 2004; Edens et
and exhibit distinctive personality and psycho- al. 2006; Guay et al. in press). Finally, studies
pathological correlates (Harpur et al. 1989; that have used latent class and latent trait
Lilienfeld, 1998; Hare, 2003). Whereas ASPD, models have found that ASPD in particular
as defined by the DSM-IV-TR, is characterized (Bucholzetal.2000)andexternalizingdisorders
by alifetime historyofantisocial behavior (evi- in general (Krueger et al. 2005) are best con-
dence of childhood conduct disorder, repeated ceptualizedasexistingalongacontinuum.
unlawful behavior), the core features of psy- The present study examined the latent struc-
chopathyareasetofpersonalitytraits,including ture of ASPD using (a) a larger and more
callousness, egocentricity and remorselessness, appropriate sample than had been used in
with only remorselessness listed as a diagnostic previous studies, (b) two methods of assessing
criterion for ASPD. Moreover, the majority ASPD,namelyawell-validatedandwidelyused
of prison inmates with ASPD do not meet semi-structured interview and a self-report
Psychopathy Checklist (PCL) criteria for psy- measure, and (c) simulation methods that per-
chopathy (e.g. Hart & Hare, 1989). Thus, mit researchers to distinguish between taxonic
ASPD (unlike the ICD diagnosis of dissocial structures and pseudo-taxonic results that can
TaxometricanalysisofASPD 1573
result from skewed measures (Ruscio et al. in reliability (mean k=0.72; Maffei et al. 1997;
press).Theseimprovedmethodsshouldprovide Dreessen & Arntz, 1998). In this study, the
more definitive conclusions about the latent inter-rater reliabilities (k) were 0.74 (n=50) for
structureofASPD. ASPD diagnoses and ICC was=0.86 (n=46)
1
fordimensionalASPDsymptomcounts.
METHOD
PersonalityDiagnosticQuestionnaire-4(PDQ-4)
Participants ASPDscale(Hyler,1994)
Participants were 1146 male offenders who
The PDQ-4 ASPD scale is a self-report ques-
were interviewed using the Structured Clinical tionnaire consisting of 22 true–false items, one
Interview for DSM-IV Axis II Personality
foreachASPDchildhoodandadultcriterionin
Disorders(SCID-II;Firstetal.1997)aspartof
DSM-IV. The k coefficient for diagnostic con-
alargerNIMH-fundedstudy.Asubsetofthese cordance between the PDQ-4 and SCID-II
participants(n=876)wasusedintheEdensetal. ASPDscaleswas0.49andthereceiveroperating
(2006) taxometric study of PCL-R-defined psy-
characteristics (ROC) analyses suggested that
chopathy. The participants were incarcerated this scale could serve as a useful screening
in state prisons in Florida, Nevada, Utah or
measure for ASPD in prison samples (Davison
Oregon(n=569),orcourt-orderedtoresidential et al. 2001). The PDQ-4 ASPD scale provided
drugtreatmentinFlorida,Texas,Utah,Nevada
an alternative (and lower threshold) indicator
orOregon(n=577).Approximately58%ofthe of ASPD in addition to the SCID-II. In the
sampleidentifiedthemselvesasCaucasian,34% .
present sample, PDQ-4 total scores (a=085)
asAfricanAmerican,and8%asHispanic.The correlated 0.55 (p<0.001) with total SCID-II
. .
meanagewas3030years(S.D.=660).Potential ASPDscores.
participants were selected randomly from lists
of individuals who met basic inclusion criteria: Dataanalyses
Euro-AmericanorAfrican-American,agerange Thedatawereanalyzedusingasetoftaxometric
21–40 (inclusive), estimated IQo70 on an IQ procedures. For taxonic conditions, these pro-
screen, spoke English, and were not receiving cedures detect a qualitative distinction between
psychotropicmedicationforactivesymptomsof taxon members and those without the disorder
psychosis. The entire protocol took on average (complement). Dimensional data lack this
.
45 hours to complete and was administered in qualitative difference. The taxometric method
two or three sessions. Participants received applies mathematically distinct procedures to
US$20,exceptatoneagencythatdidnotpermit several combinations of indicator variables (i.e.
payments. After complete description of the measures of distinct aspects of the condition).
study, written informed consent was obtained These procedures produce a set of graphs and
fromallparticipants. the shape of the graphs should converge on
a taxonic or dimensional structure. Each pro-
Measures
cedure also produces one or more estimates
Participants completed an extensive research of the base rate of the taxon in the sample.
protocol related to the objectives of the larger Althoughconsistentbaserateestimatesbetween
project. We describe only those measures and within procedures were thought to be in-
relevanttothetaxometricanalyses. dicative of a taxonic latent structure, recent
simulation studies by Ruscio (unpublished ob-
SCID-II
servations) suggest that constructs with dimen-
The ASPD module of the SCID-II (First et al. sionallatentstructurescanalsoyieldconsistent
1997) is a widely used semi-structured psychi- baserateestimates.
atricinterviewthatassessestheDSM-IVcriteria We used four taxometric procedures to
for ASPD, yielding both dimensional and cat- analyze the data: Mean Above Minus Below A
egorical scores. The interviewers for this study Cut (MAMBAC; Meehl & Yonce, 1994),
were clinical psychology graduate students MAXimum COVariance (MAXCOV; Meehl
trained by a senior investigator. The ASPD & Yonce, 1996) or MAXimum EIGenvalue
module exhibits adequate levels of inter-rater (MAXEIG;Waller&Meehl,1998),andLatent
1574 D.K.Marcusetal.
Mode factor analysis (L-Mode; Waller & dimensional structure) to 0.5 (ambiguous–
Meehl, 1998). In MAMBAC, MAXCOV and equally consistent with either structure) to 1
MAXEIG,onemeasureservesastheinputindi- (consistentwithataxonicstructure).
catorandisplottedonthexaxis.InMAMBAC,
cuts are made along an input indicator and
RESULTS
the mean scores on the output indicator
(another measure of the construct) for those Overall, 58% (661) of the participants met the
cases above the cut and for those cases below DSM-IV-TR criteria for ASPD. Taxometric
the cut are computed. The difference between proceduresaremostsensitivewhenthebaserate
thesetwomeansisplottedontheyaxis.Taxonic oftheconstructinquestion isabout50%,so if
graphs will produce a single peak, with the ASPD is taxonic and the DSM-IV-TR diag-
location of the peak reflecting the base rate nostic cut-off approximates to the true cutting
of the taxon. Dimensional graphs will typically score for this taxon, then the present sample
appear concave rather than peaked. For the should be ideal for identifying this putative
MAXCOV analyses, the graph was segmented taxon.Becauseoursampleincludedparticipants
intoasuccessionofoverlappingwindowsalong from both prisons and residential drug treat-
the input indicator. The covariances between ment programs, it ran the risk of yielding
the other two indicators for each window were a ‘pseudo-taxon’ (e.g. instead of identifying a
plotted on the y axis. MAXEIG, a multivariate unique psychiatric disorder, the analyses could
extension of MAXCOV, uses eigenvalues in- be identifying prison inmates). Fortunately,
stead of covariances. If taxonic, the covariance finding that almost identical percentages of
(or eigenvalue) should be maximal in the win- prison inmates (57.5%)anddrugprogrampar-
dow most evenly divided between members of ticipants (57.9%) met the DSM-IV-TR criteria
thetaxonandcomplementandthegraphpeaks forASPDmakesitunlikelythatthetaxometric
at this cut. The graph will appear concave, flat analyseswillproduceapseudo-taxon.Asafur-
orirregulariftheconstructisdimensional.Fifty ther precaution, we also reran the taxometric
.
windows with 090 overlap were used for the analyses separately for the two samples. These
analysesinthepresentstudy.InL-Mode,allof analyses were consistent with the results from
the indicators are factor analyzed and the dis- the entire sample, so only the results from the
tributionofscoresonthefirstprincipalfactoris entiresamplearereportedbelow.
plotted.Iftheconstructistaxonic,thegraphwill There was a moderately high correlation
be bimodal, whereas a unimodal graph is more (r=0.63) between the SCID-II and the PCL-R
consistentwithadimensionalinterpretation(for total in the present sample, suggesting that
a detailed description of these procedures see although ASPD and psychopathy are related
Waller&Meehl,1998;Ruscioetal.2006). constructs,therelationwasfarfromunitygiven
Alltaxometricanalyseswereperformedusing that they only share 40% of their variance. To
software developed by Ruscio (2006). Ruscio’s put this relationship in context, it is worth not-
programs produce simulated taxonic and di- ing that this value is similar to the correlation
mensional datasets that match the parameters between anxiety and depression (Watson et al.
of these data (e.g. indicator correlations, skew, 1995), two other related but separable con-
kurtosis).Tentaxonicand10dimensionalsimu- structs. Therefore, this taxometric analysis of
lations were created for each set of indicators. ASPD is non-redundant with our earlier analy-
To aid interpretation of the results, graphs of sisofPCL-R-assessedpsychopathy(Edensetal.
these simulated datasets can be compared with 2006).
graphsoftheactualdata.Thesesimulationsare The taxometric analyses were performed on
especially useful because skewed (or otherwise threesetsofindicators.First,wefactoranalyzed
non-normal) data can yield ambiguous graphs. all 22 items from the SCID-II. Because more
In Monte Carlo studies this simulation method than half of the SCID items pertain to child-
has outperformed other methods for inter- hood conduct disorder and accounted for three
preting taxometric results (Ruscio et al. in of the four factors that were extracted, we also
press). Ruscio’s programs calculate a curve selectedindividualitemsfromtheSCID-IIfora
fit index that ranges from 0 (consistent with a second set of analyses. These two strategies for
TaxometricanalysisofASPD 1575
Table1. Fourfactor-basedindicatorsfromtheSCID-IIAntisocialPersonalityDisorderItems
Loadingon
Item Factor1 Factor2 Factor3 Factor4
1.AdultAntisocial
UnlawfulBehavior 0.59 0.09 x0.04 x0.03
Deceitfulness 0.66 0.13 0.02 x0.09
Impulsivity 0.51 0.25 x0.22 0.15
Recklessness 0.76 x0.15 0.03 x0.07
Irresponsibility 0.73 x0.01 x0.12 0.02
LacksRemorse 0.50 x0.25 0.27 x0.09
2.ChildhoodNon-assaultiveCrimes
BreakingandEntering x0.02 0.56 0.20 0.03
Lying/Conning 0.21 0.47 0.02 0.06
Theft/Forgery 0.10 0.49 x0.05 0.23
RunningAway x0.05 0.70 x0.03 x0.10
Stayoutlate x0.07 0.70 0.10 x0.13
Truancy x0.04 0.64 0.07 x0.20
3.PhysicalViolence
ChildhoodBullying x0.03 x0.01 0.51 0.29
ChildhoodFights x0.01 0.09 0.69 0.05
ChildhoodWeaponUse x0.01 0.05 0.69 0.04
ChildhoodCrueltytoPeople x0.05 x0.02 0.68 x0.01
ChildhoodRobbed/Mugged x0.04 0.19 0.63 x0.10
AdultFights 0.39 x0.05 0.44 0.03
4.ChildhoodSevereAntisocialBehaviors
CrueltytoAnimals x0.01 x0.27 0.11 0.74
FireSetting x0.10 0.06 x0.11 0.76
Vandalism 0.02 0.23 0.13 0.44
n=1146.Loadingsofo0.30areshowninbold.
selectingindicatorscomplementedoneanother. screetest,weobtainedafour-factorsolutionthat
The factor scores have the advantage of pro- accountedfor43%ofthevariance(seeTable1).
viding more psychometrically sound indicators Factor 1 (23.1% of the variance; eigenvalue=
and a wider range of scores, whereas the indi- 5.07, a=0.68) consisted of six adult anti-
vidual item indicators placed a greater weight social items. Factor 2 (8.5% of the variance;
on adult antisocial symptoms. Finally, because eigenvalue=1.88,a=0.70)consistedofsixchild-
the SCID-II is scored by the interviewer and hood non-violent criminal behaviors. Factor 3
some evidence suggests that rater expectations (6.2% of the variance; eigenvalue=1.36, a=
concerning the latent structure of a construct 0.75) consisted of six items describing bullying
may influence the results of a taxometric or physical assault. Factor 4 (4.9% of the vari-
study (Beauchaine & Waters, 2003), a final set ance; eigenvalue=1.09, a=0.47) consisted of
of taxometric analyses used participants’ self- three destructive childhood behaviors. Taxo-
reported responses to the PDQ-4. Consistency metric analyses require the use of valid in-
in the resultsfrom these three sets of indicators dicators that are capable of discriminating
will increase confidence that the analyses re- between members of a presumptive taxon and
vealedthelatentstructureofASPD. complement group. Based on the subsequent
MAMBAC, MAXEIG and L-Mode analyses,
Analysisoffactor-levelSCIDindicators
theaverage degreesofseparation forthese four
Because indicators that are based on more indicators were 1.59, 1.27 and 1.52 standard
thanoneortwoitemsaregenerallypreferredfor deviationunits,respectively,whichexceededthe
taxometric analyses (Beauchaine, 2003; Cole, recommended minimum of 1.25 (Meehl, 1995).
2004), we created factor scales by using a In addition, when the sample was divided into
promax rotation to factor analyze all 22 adult participants who met the DSM diagnostic cri-
andchildASPDitemsfromtheSCID-II.Using teria and those who did not, the average indi-
the Kaiser criterion (i.e. eigenvalues >1) and catorvaliditywas1.33.Finally,therewasnota
1576 D.K.Marcusetal.
Simulated taxonic data Simulated dimensional data
0·14 0·14
0·12 0·12
e e
c c
n n
e e
differ 0·10 differ 0·10
n n
a a
e e
M M
0·08 0·08
0·06 0·06
0 200 400 600 800 1000 0 200 400 600 800 1000
Input (cases) Input (cases)
FIG.1. Average MAMBAC (Mean Above MinusBelow A Cut) curves for the research data, simulated taxonic data, and
simulateddimensionaldataforthefourSCID-IIantisocialpersonalitydisorder(ASPD)factorscores.Foreachcurve,50cutswere
madealongtheinputindicator.Thegraphsforthesimulatedtaxonicanddimensionaldatawereproducedbygenerating10
datasetsforeachlatentstructure.Thedarklinerepresentstheactualdataandthelighterlinesrepresentonestandarddeviation
aboveandbelowtheaverageforeachsimulateddataset.
problem with nuisance covariance (high cor- taxonicstructure.Thesecurvesyieldedbaserate
relations among the members of the putative estimates ranging from 0.28 to 0.60 (mean=
. .
taxon and complement groups) because the av- 042,S.D.=016).TheaverageoftheseMAXEIG
eragecorrelationamongthesefourindicatorsin curvesjuxtaposedwiththegraphsforthesimu-
theentiresamplewas0.39,whereastheaverage lated taxonic and dimensional datasets are pre-
correlation among those who met the ASPD sented in Fig. 2. Although the graph of the
. .
criteria(018)andthose whodidnot(013)was simulated taxonic data has a clear peak, the
considerablylower. MAXEIGgraphoftheactualdataappearsflat,
None of the four MAMBAC curves (each as does the graph of the simulated dimensional
.
factor score served as the output indicator for dataset.The curvefitindex(017)demonstrates
one graph with the remaining three factors that the actual data were far more consistent
scores summed to create the input indicator) withadimensionalstructure.
exhibited the inverted-U shape consistent with The L-Mode curve for the actual data was
a taxonic structure. These curves yielded unimodal and was similar to the simulated
base rate estimates ranging from 0.32 to 0.55 dimensional data (Fig. 3). By contrast, the
. .
(mean=042, S.D.=010). The average of these simulated taxonic data was clearly bimodal.
four MAMBAC curves juxtaposed with the Averaging the base rate estimates from the left
. . .
graph for the simulated taxonic and dimen- (013) and right modes (10) yielded a 057 base
sional datasets are presented in Fig. 1. In these rate estimate, and the base rate estimate based
.
graphs, the actual data appeared much more on the classification of cases was 052. L-Mode
.
like the simulated dimensional data than the baserateestimatesofabout050aretypicalfor
simulatedtaxonicdata;thecurvefitindex(0.12) dimensionalconstructs(Waller&Meehl,1998).
was also highly indicative of a dimensional Overall, the results from the factor scores were
structure. consistentwithadimensionallatentstructure.
NoneofthefourindividualMAXEIGcurves Because (a) Factor 4 had considerably lower
displayed the clear single peak indicative of a internalconsistencythanthethreeotherfactors,
TaxometricanalysisofASPD 1577
Simulated taxonic data Simulated dimensional data
0·8 0·8
0·6 0·6
e e
u u
al al
v v
n n
ge 0·4 ge 0·4
Ei Ei
0·2 0·2
0·0 0·0
–1·0 –0·5 0·0 0·5 1·0 1·5 –1·0 –0·5 0·0 0·5 1·0 1·5
50 Windows 50 Windows
FIG.2. AverageMAXEIG(MAXimumEIGenvalue)curvesfortheresearchdata,simulatedtaxonicdata,andsimulateddi-
mensionaldataforthefourSCID-IIantisocialpersonalitydisorder(ASPD)factorscores.Foreachcurve,thedataweresorted
alongthexaxisbythescoresontheinputindicatorandthengroupedinto50subsamplesusingoverlappingwindows(0.90
overlap).Thegraphsforthesimulatedtaxonicanddimensionaldatawereproducedbygenerating10datasetsforeachlatent
structure.Thedarklinerepresentstheactualdataandthelighterlinesrepresentonestandarddeviationaboveandbelowthe
averageforeachsimulateddataset.
(b) it was the only factor to yield validity conduct disorder symptoms. This scoring
indicators less than 1.25, and (c) less valid resulted in eight potential indicators. However,
indicators could yield inaccurate dimensional SCID-II item 1 (fails to conform to social
results,allofthetaxometricanalyseswerererun norms) was dropped because it was highly
with Factors 1–3 only. These three indicators skewed (present for 90% of the participants)
displayed strong validity (average validity 1.43, and item 7 (lacks remorse) was eliminated be-
.
minimum 132, when the sample is divided into causeithadthelowest item-totalcorrelation of
those who do and do not meet the criteria for the items and in subsequent analyses appeared
ASPD), yet the results of these analyses were to have to the lowest item validity coefficient.
alsounambiguouslydimensional.Alternatively, (Analyses including item 7 also yielded di-
indicators were also created by forcing the mensional curves.) Based on the subsequent
factor analysis into a three-factor solution. The MAMBAC, MAXEIG and L-Mode analyses,
results of these analyses were also clearly di- the average degrees of separation for these six
mensional.(Copiesofalloftheseresultsandall indicators were 1.32, 1.37 and 1.22 standard
subsequent taxometric analyses that are not deviation units, respectively, which on average
provided in this paper are available from the were slightly greater than the recommended
firstauthor.) minimumof1.25(Meehl,1995).
NoneofthesixMAMBACcurves(eachitem
Analysisofindividual-itemSCIDindicators served as the output indicator for one graph
The SCID-II interview for ASPD consists of with the remaining five items summed to create
sevenitemsthatassessadultantisocialbehavior the input indicator) exhibited the single clear
and 15 items that assess childhood/adolescent peak consistent with a taxonic structure. These
conductdisorder.These15itemswereaveraged curves yielded base rate estimates ranging
into a single indicator of childhood/adolescent from0.52to0.67(mean=0.61,S.D.=0.05).The
1578 D.K.Marcusetal.
Research data Simulated taxonic data Simulated dimensional data
0·4 0·4 0·4
0·3 0·3 0·3
y y y
c c c
n n n
e e e
u u u
q q q
e e e
e fr 0·2 e fr 0·2 e fr 0·2
v v v
ati ati ati
el el el
R R R
0·1 0·1 0·1
0·0 0·0 0·0
–2 –1 0 1 2 3 –2 –1 0 1 2 3 –2 –1 0 1 2 3
Factor scores Factor scores Factor scores
FIG.3. Latentmode(L-Mode)factoranalysiscurvesforthefourSCID-IIantisocialpersonalitydisorder(ASPD)factorscores,
andforthesimulatedtaxonicanddimensionaldatasets.Eachgraphdisplaysthefrequencydistributionofscoresonthefirstfactor
ofafactoranalysisoftheindicatorset.Thegraphsforthesimulatedtaxonicanddimensionaldatawereproducedbygenerating10
datasetsforeachlatentstructure(brokenlines).Thesolidlinesindicatetheaverageofthesedatasets.
averageofthesesixMAMBACcurvesappeared Averaging the base rate estimates from the left
more similar to the simulated dimensional data (0.00)andrightmodes(0.76)yieldeda0.38base
thantothesimulatedtaxonicdata;thecurvefit rate estimate, and the base rate estimate based
. .
index(016)wasalsoindicativeofadimensional ontheclassificationofcaseswas048.
structure. Becausethecompositechildhoodconductdis-
MAXCOV was conducted instead of orderitemwastheonlyindicatorwhosevalidity
.
MAXEIG for the six item-level indicators dipped below 125 based on case assignments
because each pair of items served as output from the MAMBAC (1.15) and MAXCOV
indicators, with the remaining four items (1.11) analyses, all of these taxometric analyses
summedtocreatetheinputindicator,providing were repeated with only the five adult items.
a sufficient range of scores on the x axis (in These analyses were also clearly dimensional.
MAXEIG each input indicator would be a Overall, the results from the individual items
single item with a range of 0 to 2). The six in- were far more consistent with a dimensional
dicators yielded 15 MAXCOV curves. Again, structure.
none of these individual curves displayed the
single peak that would support a taxonic struc- PDQ-4
ture. These curves yielded base rate estimates Wealsoconstructedfourindicatorsusingitems
. . .
that ranged from 005 to 092 (mean=067, from the self-report PDQ-4. These four in-
S.D.=0.26). Although the graph of the simu- dicators included the same items as the four
lated taxonic data had a clear peak, the actual factor scales from the SCID: Adult Antisocial
data did not and the curve fit index (0.19) was (a=0.65), Childhood Non-assaultive Criminal
moreconsistentwithadimensionalstructure. Behavior (a=0.73), Physical Violence (a=
Unlike the L-Mode curve for the simulated 0.78), and Childhood Severe Antisocial Be-
taxonic data, which was clearly bimodal, the havior (a=0.56). Based on the subsequent
L-Mode curves for the actual data and for the MAMBAC, MAXEIG and L-Mode analyses,
simulated dimensional data were unimodal. the average degrees of separation for these six
TaxometricanalysisofASPD 1579
. . .
indicators were 162, 139 and 160 standard of childhood conduct problems’ (p. 81). How-
deviation units, respectively. There was little ever,morerecentresearchusingtaxometricand
problem with nuisance covariance because the latentclassmethodsconvergesonadimensional
averagecorrelationamongthesefourindicators latent structure for the DSM-IV-TR diagnosis
intheentiresamplewas0.42,whereastheaver- of ASPD (present study; Bucholz et al. 2000)
age correlation among those whose PDQ-IV and for Factor II of the PCL-R, which assesses
met the ASPD criteria (0.13) and those whose antisocial behavior (Edens et al. 2006; Guay et
didnotmeetit(0.21)wereconsiderablylower. al. in press). Additionally, using latent class
All three taxometric methods yielded graphs analysis, Krueger et al. (2005) demonstrated
that were highly consistent with a dimensional thatthebroaddomainofexternalizingdisorders
latent structure (i.e. the MAMBAC curves appears to lie on a continuum. Therefore, if
wereconcave,theMAXEIGcurveslackedclear both ASPD and psychopathy are dimensional,
peaks, and the L-Mode curve was unimodal). neither should have priority in a categorical
. .
The base rate estimates were 045 (S.D.=008) diagnostictaxonomy.
forMAMBAC,0.38(S.D.=0.18)forMAXEIG, There has been growing interest in replacing
.
055 when averaging the base rate estimates or supplementing the current categorical ap-
. .
fromtheleft(009)andrightmodes(100)from proachtodiagnosingpersonalitydisorderswith
L-Mode, and 0.53 from the L-Mode classi- a dimensional or prototype approach (e.g.
fication of cases. The fit index for the average Shedler & Westen, 2004; Clark, 2005; Widiger,
.
MAMBACcurvewas 009and thefitindexfor 2005). The results of this study are consistent
the average MAXEIG curve was 0.15, both with the notion of conceptualizing ASPD as
of which are far more consistent with a latent a continuous condition more akin to most
dimensional structure. These analyses were forms of Type II diabetes than to dichotomous
repeated twice more using indicators derived pathologies such as Type I diabetes. For such
from factor analyzing the PDQ (both those dimensional disorders, the relevant issue for
fromathree-factorandthosefromafour-factor researchers and practitioners is not one of es-
solution). These analyses were also clearly tablishing the ideal set of criteria for taxon
dimensional. membership, but rather one of determining at
what point or under what circumstances devi-
ationsalongthiscontinuummeritclinicalatten-
DISCUSSION
tion (Lilienfeld & Marino, 1995), or result in a
Using multiple taxometric procedures with two ‘failure to achieve adaptive solutions to univer-
methodologically different sets of indicators, sallifetasks’(Livesley&Jang,2005,p.265).
our analyses provided no evidence that ASPD IfASPD proves to have a dimensionallatent
has a taxonic latent structure, at least within structure and the evidence continues todemon-
correctional and substance abuse populations. strate that schizophrenia spectrum conditions
Pending replication in other settings (e.g. psy- (e.g. schizotypal personality disorder) are
chiatric and community populations), our find- taxonic, this combination of findings suggests
ingsarefarmoreconsistentwiththeproposition that revising Axis II of the DSM will not be an
that ASPD exists on a continuum, whether easy task. If the current edition is problematic
assessed using a structured interview or a self- because the diagnoses of personality disorders
reportmeasure. reflect arbitrary cuts along one or more con-
Haslam (2003) suggested that the extant tinuous dimensions, replacing these categories
taxometricresearchlentsupporttotheinclusion withdimensionswouldobscuretheexistence of
ofASPD,andnotpsychopathy,asadiagnostic genuine taxa. Ultimately, an evidence-based
category in DSM-IV-R, noting that ‘in view of diagnostic system for personality disorders
theoft-remarkedfailureofrecentDSMeditions may well need to incorporate a ‘hybrid’ model
ofAPD criteriatoencompass psychopathysat- that accommodates both dimensions and taxa.
isfactorily, it is interesting that the evidence for Needlesstosay,considerablymoreresearchwill
thistaxon isstrongest precisely inthose aspects be necessary to determine which personality
of the psychopathy construct that DSM em- disorders are categorical and which are dimen-
bodies,namelyantisocialbehaviorwithahistory sional.
1580 D.K.Marcusetal.
These results also bolster recent attempts acknowledge and appreciate the assistance
to conceptualize ASPD within the framework and cooperation of the following agencies in
of extant dimensional models of personality, collectingdataforthisresearch;however,none
including temperament (Clark, 2005) or trait of the opinions or conclusions expressed in this
frameworks that posit three (Cloninger, 1987; article reflects any official policy or position of
Tellegen & Waller, in press) or five (Costa & any of these institutions: Drug Abuse Compre-
Widiger, 2001) higher-order dimensions under- hensiveCoordinatingOffice(DACCO),Tampa,
lying both normal and abnormal personality. FL; Florida Department of Corrections; Gate-
For example, within the five-factor model, way Foundation, Huntsville, TX; Nevada De-
ASPD features are associated primarily with partmentofPrisons;OdysseyHouse,SaltLake
low levels of agreeableness and conscientious- City, UT; Operation PAR, Pinellas Park, FL;
nessandhighlevelsofcertainfacetsofneuroti- Oregon Department of Corrections; Texas
cism,suchashostilityandimpulsivity(Milleret Department of Criminal Justice-Institutional
al. 2003). It is unlikely that these trait models Division; Utah Department of Corrections;
will be sufficient for describing and explaining Volunteers of America, Portland, OR; West-
ASPD, given that many individuals with this care, NV. We thank Jennifer Skeem and Kevin
configuration of traits do not exhibit this con- Douglas for their helpful comments on an
dition. Presumably, these trait dispositions earlierdraftofthispaper.
manifest themselves in antisocial behavior only
among individuals exposed to certain environ-
DECLARATIONOFINTEREST
mentalfactors(Lykken,1995).
Arecentcriticismofthetaxometricliterature None.
is that the findings from different studies of the
samedisorderhaveoftenbeeninconsistent(e.g.
REFERENCES
Krueger et al. 2005; Widiger & Samuel, 2005),
making it possible to dismiss the results of APA(2000).DiagnosticandStatisticalManualofMentalDisorders
thisandtheHarrisetal.studyasmerelyanother (4th edn, text revision) (DSM-IV-TR). American Psychiatric
Association:Washington,DC.
set of contradictory findings. Although the Beauchaine, T.P. (2003). Taxometrics and developmental psycho-
present study is not without limitations (e.g. pathology.DevelopmentalPsychopathology15,501–527.
Beauchaine, T.P. & Waters, E. (2003). Pseudotaxonicity in
exclusionofwomen),thereareamplereasonsto
MAMBACandMAXCOVanalysesofrating-scaledata:turning
placeconfidenceinthepresentfindings.First,all continua into classes by manipulating observer’s expectations.
ofthepublishedresearchsuggestingataxonfor PsychologicalMethods8,3–15.
Bucholz,K.K.,Hesselbrock,V.M.,Heath,A.C.,Kramer,J.R.&
ASPD or psychopathy has been generated by a
Schuckit,M.A.(2000).Alatentclassanalysisofantisocialper-
singleresearchteam,whereasdimensional find- sonalitydisordersymptomdatafromamulti-centrefamilystudy
ings have now been published by at least three ofalcoholism.Addiction95,553–567.
Clark,L.A.(2005).Temperamentasaunifyingbasisforpersonality
independent groups of researchers. Second, by and psychopathology. Journal of Abnormal Psychology 114,
including general population inmates and in- 505–521.
Cloninger,C.R.(1987).Asystematicmethodforclinicaldescription
dividuals ordered into residential drug treat-
and classification of personality variants. Archives of General
ment, and excluding participants with active Psychiatry44,573–588.
psychotic conditions, we minimized the risk of Cole, D.A. (2004). Taxometrics in psychopathology research: an
introduction to some of the procedures and related method-
detectingataxonforschizotypy.Finally,thefact
ologicalissues.JournalofAbnormalPsychology113,3–9.
thatwedetectedconsistentevidenceofadimen- Costa,P.T.&Widiger,T.A.(eds)(2001).PersonalityDisordersand
sional structure using multiple sets of construct the Five Factor Model of Personality (2nd edn). American
PsychologicalAssociation:Washington,DC.
valid indicators for ASPD renders it unlikely Davison,S.,Leese,M.&Taylorm,P.J.(2001).Examinationofthe
that the present results can be attributed to in- screeningpropertiesofthePersonalityDiagnosticQuestionnaire
adequatemeasurementofthiscondition. 4+ (PDQ-4+) in a prison population. Journal of Personality
Disorders15,180–194.
Dreessen,L.&Arntz,A.(1998).Short-intervaltest–retestinterrater
reliability of the Structured Clinical Interview for DSM-III-R
ACKNOWLEDGMENTS Personality Disorders (SCID-II) in outpatients. Journal of
PersonalityDisorders12,138–148.
This research was supported by grant no. Edens, J.F., Marcus, D.K., Lilienfeld, S.O. & Poythress, N.G.
RO1-63783-01A1 from the National Insti- (2006).Psychopathic,notpsychopath:taxometricevidenceforthe
dimensionalstructureofpsychopathy.JournalofAbnormalPsy-
tute of Mental Health to Dr Poythress. We chology115,131–144.