Psychopathic, Not Psychopath: Taxometric Evidence for the Dimensional
Structure of Psychopathy
John F. Edens
Southern Methodist University
David K. Marcus
Sam Houston State University
Scott O. Lilienfeld
Norman G. Poythress Jr.
University of South Florida
Although psychopathy is frequently regarded as qualitatively distinct from other conditions, relatively
little research has examined whether psychopaths represent a distinct class of individuals. Using a sample
of 876 prison inmates and court-ordered substance abuse patients who were administered the Psychop-
athy Checklist—Revised (R. D. Hare, 2003), the authors examined the latent structure of psychopathy
using several taxometric procedures developed by Meehl and colleagues (P. E. Meehl & L. J. Yonce,
1994; N. G. Waller & P. E. Meehl, 1998). The results across these procedures offer no compelling
support for the contention that psychopathy is a taxonic construct and contradict previous reports that
psychopathy is underpinned by a latent taxon. The authors discuss the theoretical, public policy, and
practice-level implications of these findings.
Keywords: psychopathy, PCL–R, taxometrics, discrete class, taxon
Psychopathic personality (psychopathy) is a constellation of
relatively distinctive personality traits (e.g., callousness, grandios-
ity, pathological lying) that may occur in the context of a criminal
or socially deviant lifestyle. Although overlapping with the Diag-
nostic and Statistical Manual of Mental Disorders diagnosis of
antisocial personality disorder, psychopathy is a separable con-
struct that places a greater emphasis on affective and interpersonal
traits. In fact, in prison settings, most individuals who meet the
diagnostic criteria for antisocial personality disorder do not meet
the standard definition of psychopathy (Hare, 2003; cf. Skilling,
Harris, Rice, & Quinsey, 2002). The widespread theoretical and
applied interest in the construct of psychopathy is evident in the
large number of peer-reviewed journal articles published on the
topic over the past few decades (e.g., several recent special issues
of journals such as Behavioral Sciences and the Law and Criminal
Justice and Behavior have been devoted specifically to the topic of
psychopathy) as well as in the growing use of this disorder to
inform legal decision making across the world (for overviews, see
Edens & Petrila, 2006; Ogloff & Lyon, 1998).
Psychopathy, typically operationalized by the Psychopathy
Checklist—Revised (PCL–R; Hare, 1991, 2003), is associated
with a number of real world criterion measures of considerable
practical significance, such as increased risk for misconduct while
institutionalized (Guy, Edens, Anthony, & Douglas, in press),
community violence (Skeem & Mulvey, 2001), and criminal re-
cidivism following release from prison (Hemphill, Hare, & Wong,
1998). Laboratory studies also have suggested that high scorers on
the PCL–R and other measures of psychopathy differ in theoreti-
cally meaningful ways from other individuals, such as in their
lower resting electrodermal activity, aberrant processing of affec-
tively charged linguistic and nonlinguistic stimuli, and lack of
responsiveness to punishment cues (see Hare, 2003; Lorber, 2004).
Although the clinical and theoretical importance of psychopathy
is well established, there remain a number of unresolved and
intensely debated issues regarding the nature of the construct, such
as the generalizability of psychopathy across gender and ethnic
groups (e.g., Cale & Lilienfeld, 2002; Skeem, Edens, Camp, &
Colwell, 2004), the putative existence of variants or subtypes of
psychopathy (e.g., Brinkley, Newman, Widiger, & Lynam, 2004;
John F. Edens, Department of Psychology, Southern Methodist Univer-
sity; David K. Marcus, Department of Psychology, Sam Houston State
University; Scott O. Lilienfeld, Department of Psychology, Emory Uni-
versity; Norman G. Poythress Jr., Department of Mental Health Law &
Policy, University of South Florida.
Portions of this research were completed while John F. Edens was a
faculty member at Sam Houston State University.
An earlier version of this research was presented at the annual meeting
of the American Psychology-Law Society, Scottsdale, Arizona, March
2004. Funding for this project was provided by National Institute of Mental
Health Grant 1 R01 MH63783-01A1 awarded to Norman G. Poythress Jr.
We thank Jennifer Skeem, Kevin Douglas, Christopher Patrick, and Paul
Frick for their helpful comments on a draft of this article.
We acknowledge and appreciate the assistance and cooperation of the
following agencies in collecting data for this research; however, none of
the opinions or conclusions expressed in this article reflect any official
policy or position of any of these institutions: Drug Abuse Comprehensive
Coordinating Office, Tampa, FL; Florida Department of Corrections; Gate-
way Foundation, Huntsville, TX; Nevada Department of Prisons; Odyssey
House, Salt Lake City, UT; Operation PAR, Pinellas Park, FL; Oregon
Department of Corrections; Texas Department of Criminal Justice—Insti-
tutional Division; Utah Department of Corrections; Volunteers of America,
Portland, OR; and WestCare, Harris Springs, NE.
Correspondence concerning this article should be addressed to John F.
Edens, Department of Psychology, Southern Methodist University, Dallas
TX 75275. E-mail: firstname.lastname@example.org
Journal of Abnormal Psychology
2006, Vol. 115, No. 1, 131–144
Copyright 2006 by the American Psychological Association
Skeem, Poythress, Edens, Lilienfeld, & Cale, 2003), and the nature
of the underlying factor structure of the PCL–R (e.g., Cooke &
Michie, 2001; Cooke, Michie, Hart, & Clark, 2005; Hare, 2003).
Regarding the latter issue, research on psychopathy typically has
focused on two moderately correlated factors initially identified by
Hare and colleagues (Harpur, Hakstian, & Hare, 1988; Harpur,
Hare, & Hakstian, 1989). More recent factor analytic work (dis-
cussed below), however, suggests that either a three-factor (Cooke
& Michie, 2001; Cooke et al., 2005) or four-factor (Hare, 2003)
model may better represent the dimensions tapped by the PCL–R
than the traditional two-factor approach.
Another ongoing controversy that continues to attract research
attention is the extent to which psychopathic personality is dimen-
sional or categorical in nature. That is, to what extent does the
construct of psychopathy identify a fundamentally distinct class of
individuals who differ qualitatively from the rest of society? Al-
though this question is by no means specific to psychopathy (e.g.,
similar questions have been raised in relation to other personality
disorders [Haslam, 2003] as well as various Axis I diagnoses), the
outcome of this debate may be of considerable practical signifi-
cance given the increasing role of the highly charged label of
psychopath in the legal system, where the PCL–R has been used to
justify indeterminate commitment, rebut insanity defenses, and
bolster support for the death penalty in capital murder trials (Cun-
ningham & Reidy, 2002; DeMatteo & Edens, 2005; Edens, in
press; Edens & Petrila, 2006; Edens, Petrila, & Buffington-
Vollum, 2001; Ogloff, & Lyon, 1998).
Although some researchers have argued that psychopathy is a
constellation or configuration of extreme levels of continuously
distributed personality traits (e.g., Benning, Patrick, Blonigen,
Hicks, & Iacono, 2005; Lilienfeld & Fowler, 2006; Miller, Lynam,
Widiger, & Leukefeld, 2001), other researchers have staked out a
position that strongly endorses a taxonic structure. For example,
Harris, Skilling, and Rice (2001) asserted that psychopaths “com-
prise a discrete natural class [italics added] of individuals” (p.
197) and that there are fundamental, qualitative differences be-
tween psychopaths and nonpsychopaths. Moreover, in other con-
texts, this research team has referred to their alternative measure of
the psychopathy construct as the Child and Adolescent Taxon
Scale (Quinsey, Harris, Rice, & Cormier, 1998), ostensibly reflect-
ing a categorically distinct class of individuals who can be iden-
tified at a very early age. Consistent with this argument, Skilling,
Quinsey, and Craig (2001) reported evidence of a low base rate
taxon underlying antisocial behavior in a community sample of
boys. The extent to which these findings identify a latent taxon in
reference to psychopathy is unclear, as most of this research (see
below) has found support for such a taxon in relation to socially
deviant and antisocial behavior rather than the personality traits
associated with psychopathy (e.g., Cleckley, 1988) per se.
The dimension versus taxon question is relevant to a number of
debates related to psychopathy, such as its etiology. For example,
a taxon in some instances can be explained by a specific causal
agent (e.g., Huntington’s disease results from the action of a single
gene). If psychopathy were shown to be dimensional, however, it
would be unlikely to result from a dichotomous causal agent, such
as the presence–absence of frontal lobe damage, a threshold level
of an environmental stressor (e.g., severe child abuse), or a dom-
inant gene (see Meehl & Golden, 1982). From a pragmatic per-
spective, support for the position that psychopathy is taxonic
would render it more difficult to justify research programs that
focus on studying psychopathy in samples in which the base rate
is likely to be very low, such as college students (Lilienfeld, 1998).
In contrast, support for the position that psychopathy is dimen-
sional would provide at least some indirect support for the practice
of conducting psychopathy research in college or community
samples, although it is worth noting that these samples may be
characterized by a paucity of extremely high scorers on psychop-
athy measures. If certain clinically important behaviors associated
with psychopathy (e.g., serious physical aggression, severe illicit
substance use) tend to emerge only among individuals with mark-
edly elevated scores, then psychopathy research on college or
community samples could yield incomplete or misleading conclu-
sions even if psychopathy were found to be dimensional.
Evidence supporting the taxonic structure of psychopathy de-
rives primarily from an influential study conducted by Harris,
Rice, and Quinsey (1994) as well as a follow-up report reexam-
ining many of the same participants (Skilling et al., 2002). Harris
et al. (1994) conducted taxometric analyses (Meehl & Golden,
1982) of the PCL–R using data from a large sample of mentally
disordered offenders detained in Canadian psychiatric facilities.
Harris et al. (1994) asserted unequivocally that their findings
demonstrated that a taxon could be identified using PCL–R scores,
with a base rate in their sample ranging from .44 to .46 across four
sets of analyses. They also claimed that a cut score of 19 or 20
would optimally differentiate between the taxon and complement
classes, although these scores are considerably lower than the
traditional cut score (total score ? 30) for “diagnosing” psychop-
athy using the PCL–R (Hare, 1991, 2003). Although their conclu-
sions applied to psychopathy per se, Harris et al. (1994) found
evidence of a taxon only for PCL–R Factor 2, which is composed
of items assessing a chronically unstable and socially deviant
lifestyle. Their analyses relevant to PCL–R Factor 1, which rep-
resents such affective and interpersonal features as grandiosity,
callousness, and superficial charm, were inconsistent with a tax-
Since the publication of this study, several criticisms have been
raised regarding Harris et al.’s (1994) methodology and statistical
analyses (Lilienfeld, 1998; Marcus, John, & Edens, 2004). For
example, the sample was an atypical group of criminal offenders—
inmates in a maximum-security psychiatric institution—a large
minority of whom had been adjudicated not guilty by reason of
insanity (NGRI). Although the authors conducted ancillary analy-
ses in which patients with psychotic diagnoses were removed,
examination of the reported sample sizes suggests that a significant
number of NGRI acquittees remained in the sample, even though
they had not been diagnosed as psychotic. This fact raises the
possibility that Harris et al. (1994) could have inadvertently un-
covered a taxon for schizotypy (see Lenzenweger & Korfine,
1992, for evidence that schizotypy is taxonic). The content of the
PCL–R items showing evidence of a taxon (e.g., reflecting persons
who lacked realistic goals, were irresponsible, and had childhood
behavior problems) was relatively nonspecific; Cooke & Michie,
1997). The schizotypy hypothesis also is rendered plausible by
findings suggesting that at least some psychopathy measures,
including a preliminary version of the PCL (the precursor of the
PCL–R), may erroneously detect a nontrivial number of patients
with schizophrenia (Howard, Bailey, & Newman, 1984; but see
Hare & Harpur, 1986, for a critique).
EDENS, MARCUS, LILIENFELD, AND POYTHRESS
Additionally, the PCL–R scores used in the taxometric analyses
were based solely on file review data, with no interview informa-
tion to supplement the ratings. Although it is acceptable to score
the PCL–R for clinical and research purposes solely on the basis of
adequate file data, it is unclear whether or how such a strategy
might impact taxometric analyses conducted on these ratings.
Furthermore, each item of the PCL–R was dichotomized instead of
scored using the standard 3-point scoring system—a scoring
method that may enhance the probability of detecting a pseudo-
taxon (Lenzenweger, 2004; Meehl & Yonce, 1996; M. B. Miller,
1996). Moreover, although it is standard practice in taxometric
studies to report the validity of the indicators included in the
analyses (i.e., the difference in scores on the indicators between
members of the putative taxon and the complement in standard
deviation units), neither Harris et al. (1994) nor Skilling et al.
(2001) reported indicator validity values. Finally, the item-level
indicators were more strongly correlated with the PCL–R total
score in the Harris et al. (1994) sample than would be expected on
the basis of the data reported in the 1991 and 2003 manuals. For
example, the correlation between Item 15 and the total score was
.70, whereas this correlation among forensic patients in the PCL–R
manual (Hare, 2003) is only .41. Similarly, Item 9 was more
strongly correlated with the total score in their sample (r ? .63)
than in the manual (r ? .37). In other words, inmates rated as
possessing one attribute of psychopathy were also more likely
(than was typically the case in the normative PCL–R data) to be
seen as having other attributes of psychopathy. Given the sole
reliance on file review, the dichotomous rating scales used, and the
higher-than-expected corrected item–total correlations, it seems
possible, if not likely, that the taxonic findings obtained for Factor
2 are questionable.
Since the Harris et al. (1994) article, additional taxometric
procedures have been developed (Waller & Meehl, 1998), and
researchers have become more sensitive to the dangers of produc-
ing pseudotaxonic solutions that could result from skewed distri-
butions (e.g., Hankin, Fraley, Lahey, & Waldman, 2005; J. Ruscio,
Ruscio, & Keane, 2004). Recent research examining the latent
structure of psychopathy has been much less supportive of an
underlying taxon, although no published studies explicitly have
examined the structure of psychopathy using the PCL–R. Marcus
et al. (2004) conducted taxometric analyses on a sample of jail and
prison inmates who had been administered the Psychopathic Per-
sonality Inventory (PPI; Lilienfeld & Andrews, 1996), which is a
promising self-report measure designed to assess the core person-
ality dimensions of psychopathy. Across several taxometric anal-
yses, there was virtually no evidence supporting a categorical
model. These findings are broadly consistent with the Harris et al.
(1994) findings for Factor 1 of the PCL–R. Inconsistent with
Harris et al.’s (1994) taxonic results for Factor 2, however, are
findings suggesting that the more chronically antisocial aspects of
psychopathy also are dimensional in nature (Bucholz, Hessel-
brock, Heath, Kramer, & Schuckit, 2000; see also Osgood, Mc-
Morris, & Potenza, 2002). Nevertheless, because the analyses of
Bucholz et al. (2000) and Osgood et al. (2002) were not based on
PCL–R Factor 2 data specifically and were performed using sta-
tistical procedures other than taxometrics, their relevance to Harris
et al.’s (1994) psychopathy findings is indirect at best.
The two published taxometric studies that have directly exam-
ined the latent structure of psychopathy have yielded somewhat
inconsistent results. Although the methodologies used by Harris et
al. (1994) and Marcus et al. (2004) differed in a number of ways
(e.g., Harris et al., 1994, used univariate taxometric analyses,
whereas Marcus et al., 2004, used multivariate methods; Harris et
al., 1994, used a maximum-security forensic sample, many of
whom were acquitted on the basis of NGRI, whereas Marcus et al.,
2004, used jail and prison inmates, who were far less likely to be
severely mentally ill), perhaps the most important difference was
that Harris et al. (1994) used the PCL–R to assess psychopathic
traits, whereas Marcus et al. used the PPI. Thus, the question of
whether psychopathy per se is dimensional or categorical or
whether these differing results were artifacts of the instruments
used remains unresolved. Although concerns have been expressed
about an overreliance on the PCL–R as the exclusive measure of
psychopathic traits (e.g., Poythress, Edens, & Lilienfeld, 1998),
there is little question that it is the most widely used and exten-
sively validated measure of this construct (Hare, 2003). Moreover,
given the potential etiological, pragmatic, and public policy im-
plications of the taxometric status of psychopathy, research ad-
dressing this issue specifically in relation to the PCL–R would be
Additionally, Hare (2003) recently has argued for a four-factor
solution to the PCL–R in which the original two factors are split
into narrower facets. More specifically, Factor 1 is composed of
separable interpersonal (e.g., glibness, pathological lying) and
affective (callousness, shallow affect) facets (four items each),
whereas Factor 2 is composed of separable lifestyle (e.g., impul-
sivity, parasitic lifestyle) and antisocial (e.g., juvenile delinquency,
criminal versatility) facets (five items each). The relevance of this
newer model is that it allows taxometric analyses to be conducted
at the scale (as opposed to item) level, which was not possible at
the time Harris et al. (1994) conducted their analyses. Although
debates exist concerning the theoretical relevance of Hare’s (2003)
fourth facet (Cooke & Michie, 2001; Cooke et al., 2005; Hill,
Neumann, & Rogers, 2004), this model does encapsulate the
multidimensional nature of the construct of psychopathy as opera-
tionalized by the PCL–R, which comprises a mixture of emotional,
behavioral, and interpersonal features. In other words, most indi-
viduals who commit crimes are not psychopathic, nor are most
callous individuals psychopathic: The term typically is reserved for
those who evidence psychopathic qualities across these domains. If
psychopathy is a taxon, then these four domains should co-occur in a
nonrandom manner that would reflect this latent structure.
In the present study, like in Harris et al. (1994), we used the
PCL–R to examine the latent structure of psychopathy. However,
we attempted to avoid many of the pitfalls of that earlier study.
First, by including interviews in addition to file reviews, our
administration of the PCL–R was consistent with the method
preferred by the instrument’s developer (Hare, 2003). Second, we
scored the PCL–R using the standard 3-point scoring method and
did not artificially dichotomize the data. Third, we did not recruit
mentally ill offenders, so if a taxon were to be identified, it would
be less likely to be a schizotypy taxon. Fourth, as additional
consistency checks, we incorporated multivariate taxometric pro-
cedures (e.g., maximum eigenvalue, latent mode factor analysis;
described below) that had not been published until Waller and
Meehl (1998). Fifth, we took advantage of the new four-facet
model of the PCL–R to conduct taxometric analyses of indicators
that should be more stable and valid than the individual-item
PCL–R TAXOMETRIC ANALYSES
indicators used in the previous PCL–R taxometric studies. Finally,
given concerns about (a) the sample, (b) the administration of the
PCL–R, (c) the inflated item–total correlations, and (d) the absence
of indicator validity data in the Harris et al. (1994) study, we
attempted to replicate their findings using the same individual-item
indicators that they had used. If Harris et al. (1994) succeeded in
detecting a psychopathy taxon, independent investigators should
be able to replicate this finding using the same indicators.
Participants were male offenders who had completed the PCL–R (Hare,
1991, 2003) as part of a larger study funded by the National Institute of
Mental Health examining personality features and their relation to social
deviance. These individuals were serving sentences in state prisons in
Florida, Nevada, Utah, or Oregon, or they were in court-ordered residential
drug treatment sites in Florida, Texas, Utah, Nevada, or Oregon. Partici-
pation was limited to English-speaking individuals between the ages of 21
and 40 (inclusive) and to European American or African American racial
groups. Participants also had to complete a preliminary IQ screen and
obtain an estimated IQ ? 70. Across all sites, individuals receiving
psychotropic medication for active symptoms of psychosis were excluded
from participation. Because we conducted item-level analyses, study par-
ticipants who were missing any of the individual PCL–R items (10%) were
dropped from this data set, which resulted in a final sample of 876 for the
analyses reported here. Approximately 59% of the sample identified them-
Participants completed an extensive research protocol related to the
primary objectives of the larger project. We describe here only those
measures relevant to participant screening and to the taxometric analyses.
PCL–R (Hare, 1991, 2003).
The PCL–R consists of 20 items reflecting
psychopathic personality traits and behaviors that are scored on the basis of
a semistructured interview and review of institutional file information.
Each item is scored on a 3-point scale, depending on the degree to which
each applies to the individual, with a possible total score ranging from 0 to
40. As noted earlier, the reliability and construct validity of the PCL–R
have been well established (see Hare, 1991, 2003, for an overview).
PCL–R ratings for this study were performed by research assistants who
had received extensive training in the administration and scoring of this
measure prior to the onset of data collection for the larger National Institute
of Mental Health study. Interrater reliability for a total of 51 cases was
ICC1? .88. Cronbach’s alpha was .81 for the total score in this sample.
Notably, unlike the PCL–R results reported by Harris et al. (1994), our
corrected item–total correlations for the PCL–R items of interest were gener-
ally very similar to those reported in the most recent manual (Hare, 2003).
In terms of the cut score historically used to identify psychopathic and
nonpsychopathic groups, 20.4% of the sample scored ? 30 on the PCL–R
total score, which is generally consistent with base rates reported in other
samples of adult offenders. The mean PCL–R score for the sample was
23.16 (SD ? 7.12), which is generally consistent with mean scores from
other research conducted on similar samples (see Hare, 2003, for an
overview of several data sets). There was a small but statistically signifi-
cant difference in the mean total scores across the prison (M ? 24.22;
SD ? 7.08; n ? 405) and drug treatment (M ? 22.25; SD ? 7.03; n ? 471)
groups, t(875) ? 4.11, p ? .01 (Cohen’s d ? 0.28).
As noted earlier, more recent factor analytic work suggests that a
four-factor model (Hare, 2003) may better reflect the dimensions tapped by
the PCL–R than the two-factor model historically associated with this
measure (Hare, 1991). This approach bifurcates the original Factor 1 items
into separable affective and interpersonal facets and parses the remaining
items into two narrower facets reflecting impulsive lifestyle and antisocial
Quick Test (Ammons & Ammons, 1962).
measure of intellectual abilities that can be administered in approximately
10 min. The respondent is shown a card that displays four pictures, and the
test administrator reads aloud words that represent items or concepts that
are represented in only one of the four pictures. Participants then indicate
the picture in which they believe the item or concept is portrayed. Only
participants who obtained an estimated IQ ? 70 were allowed to continue
in the study. The Quick Test is an excellent predictor of IQ scores in the
normal range (Traub & Spruill, 1982) and provides a good estimate of the
Wechsler Adult Intelligence Scale—Revised (Wechsler, 1981) IQ scores in
both genders and in both Caucasians and African Americans (Craig &
Olsen, 1988). It has also been shown to satisfactorily estimate intelligence
scores in offender populations (DeCato & Husband, 1984; Doss, Head,
Blackburn, & Robertson, 1986; Simon, 1995).
The Quick Test is a screening
At each site, potentially eligible participants were randomly selected
from lists of individuals who met basic inclusion criteria (i.e., age, race,
English fluency). Enrollment interviews were conducted in a private room,
and informed consent was obtained using procedures approved by a uni-
versity institutional review board. After informed consent was obtained,
the IQ screening test was administered, followed by the full research
protocol. As the entire protocol took on average 4.5 hr to complete, it was
administered in at least two, and sometimes three, sessions. Except at one
agency that did not permit participant payments, at the end of protocol
administration, $20 was deposited into the agency account of all partici-
pants for the time they contributed to the study.
Data Analytic Strategy
We analyzed PCL–R scores using a series of taxometric procedures
developed by Meehl and his colleagues (Meehl & Yonce, 1994; Waller &
Meehl, 1998) using statistical software programs written by J. Ruscio
(2004). If psychopathy is taxonic, these procedures can detect a qualitative
distinction between psychopaths (taxon) and nonpsychopaths (comple-
ment). Dimensional data will lack this qualitative difference. Rather than
relying on standard null hypothesis testing, the taxometric method applies
various mathematically distinct procedures to several combinations of
indicator variables, the results of which will tend to converge on a taxonic
or dimensional structure. These procedures also produce one or more
estimates of the base rate of the taxon in the sample. If the construct of
interest is taxonic, these base rate estimates may be consistent (a) both
within and across the various taxometric procedures and (b) with estimates
of the base rate in the population being sampled. Wide variation among
base rate estimates may be more consistent with a dimensional solution.1
In this study, we used four taxometric procedures to examine the latent
structure of psychopathy: mean above minus below a cut (MAMBAC;
1However, J. Ruscio (2005) recently conducted a Monte Carlo study
that found that base rate consistency tests were not a reliable method for
inferring latent structure: Under certain circumstances, dimensional data
yielded more consistent base rate estimates than taxonic data. Because base
rate consistency has become somewhat of a convention in the taxometric
literature, we still report these values, but no decisions about latent struc-
ture were made solely (or even primarily) on the basis of base rate
EDENS, MARCUS, LILIENFELD, AND POYTHRESS
Meehl & Yonce, 1994), maximum covariance (MAXCOV; Meehl &
Yonce, 1996) or maximum eigenvalue (MAXEIG; Waller & Meehl, 1998),
and latent mode factor analysis (L-Mode; Waller & Meehl, 1998).
MAMBAC uses two indicator variables, the input indicator x and the
output indicator y. Cuts are made at regular intervals along the input
indicator (50 in the present study); at each cut, the mean score on the output
indicator for those cases above the cut and the mean scores for those cases
below the cut are computed. The difference between these two means is
then graphed on the y-axis. Taxonic graphs will evidence a single peak,
with the location of the peak reflecting the base rate of the taxon. The
further to the left side of the graph the peak is, the greater the base rate. If
the construct of interest is continuous, the graph will appear concave rather
than peaked. A second MAMBAC analysis is performed with the input and
output variables reversed to provide a more extensive test of the latent
structure. In this study, we used a modification of MAMBAC developed by
J. Ruscio (2004) in which multiple indicators are combined to create a
single input variable. Because this multivariate procedure makes use of all
of the data in each MAMBAC analysis, it may be more powerful than the
traditional approach—assuming that all indicator variables demonstrate
MAXEIG is a multivariate extension of Meehl and Yonce’s (1996)
MAXCOV procedure that uses all of the indicators simultaneously. The
sample is first divided into a succession of overlapping windows along the
input indicator. In MAXCOV, the covariance between the two output
indicators for the cases in each slice is then computed and plotted on the
y-axis. In MAXEIG, all of the remaining indicators are factor analyzed, and
the eigenvalue of the first principal factor, which represents the multivar-
iate analogue of the covariance, is plotted for each window. If taxonic, the
eigenvalue should be maximal in the subsample most evenly divided
between members of the taxon and the complement, and the graph should
peak at this cut. The graph should appear concave, flat, or irregular if the
construct is dimensional. Fifty windows with .90 overlap were used for the
analyses in the present study.
Finally, the L-Mode procedure factor analyzes all of the indicators and
graphs the distribution of scores on the first principal factor. If the construct
of interest is taxonic, the graph will be bimodal, whereas a unimodal graph
is consistent with a dimensional interpretation.
We performed two sets of taxometric analyses. When Harris et
al. (1994) conducted their study, the PCL–R was considered to
have only a two-factor structure, which would preclude using
factor scores for most taxometric analyses. However, as noted
earlier, Hare (2003) recently has argued for a four-factor solution
to the PCL–R in which the original two factors essentially are split
into narrower facets, which allow for scale-level analyses. Thus,
we conducted one set of taxometric analyses using factor scores as
the indicators of psychopathic traits.
Our second set of analyses involved item-level data. Harris et al.
(1994) performed their analyses by selecting eight items from the
PCL–R that correlated most highly with the PCL–R total score in
their data set. We therefore used those same eight indicators in an
attempt to replicate their findings. For analyses that involved
individual items, we used MAXCOV (rather than MAXEIG) and
had two variables serve as the outputs and summed the remaining
six items to create the input to most closely follow the approach
taken by Harris et al. (1994).
Skewed indicators can yield results that appear taxonic even
when the latent structure is dimensional (Hankin et al., 2005; J.
Ruscio et al., 2004). Indicators with a negative skew are likely to
produce graphs that indicate a high base rate taxon, whereas
positively skewed indicators can yield a pseudotaxon with an
apparent low base rate. Three of the four PCL–R factors had
negative skews that were more than twice the standard error
(Affective ? ?.22, Lifestyle ? ?.52, Antisocial ? ?.30; SE ?
.08). Similarly, six of the eight items that Harris et al. (1994) used
had substantial negative skew (see Table 1).
Because visual inspection of the taxometric graphs without an
accompanying context could produce misleading results, we also
used J. Ruscio’s (2004) programs to create simulated taxonic and
dimensional data sets that matched the parameters of the research
data, including the skew of the distribution. This procedure allows
for comparisons across graphs produced by the actual and simu-
lated data. Additionally, J. Ruscio’s (2004) programs compute
goodness-of-fit statistics that quantify whether the graph produced
by the research data is closer in appearance to the simulated
dimensional or taxonic results (see J. Ruscio et al., 2004). A
smaller root-mean-square residual (FitRMSR) index indicates better
fit (see J. Ruscio et al., 2004). Finally, because multiple sets of
simulated data can be created, Cohen’s d is computed using the
mean difference between the dimensional and taxonic fit values
across the replications. Dimensional fit values are subtracted from
taxonic values, so that a positive result suggests a dimensional
structure, whereas a negative value suggests a taxonic structure (J.
Ruscio, 2004). Ten sets of simulated dimensional data and 10 sets
of simulated taxonic data were generated for each of the analyses.
Because our sample consisted of both prison inmates and indi-
viduals who were court ordered into residential drug treatment
programs, we also repeated these analyses for each of the two
subgroups in addition to performing taxometric analyses on the
entire sample. The results of these separate analyses were almost
uniformly consistent with those from the entire sample, so in the
Descriptive Statistics for the Indicators
Harris et al. (1994) PCL–R items
aThe standard error for these skew coefficients is .083.
estimates were calculated on the basis of a cut score of 30 on the PCL–R
cut score of 25 on the PCL–R total score.
PCL–R ? Psychopathy Checklist—Revised.
cThese validity estimates were calculated on the basis of a
PCL–R TAXOMETRIC ANALYSES
interest of parsimony and increased statistical power, only the
results from the entire sample are reported below.2
Analysis of Factor-Level Indicators
To check for nuisance covariance (i.e., correlations among the
indicators within either the taxon or the complement; Meehl &
Golden, 1982), we computed correlations among these four factors
for those participants who scored 30 or above (the standard cut
score for diagnosing psychopathy) on the PCL–R and those who
scored below 30. There was little nuisance covariance among these
indicators for those participants who met the cut score for psy-
chopathy (r ? ?.08) and an acceptable level of covariance for
those in the putative complement (r ? .22). The average correla-
tion among the four indicators in the entire sample was .36. Using
this same cut score, we estimated the average indicator validity to
be 1.30, which exceeds the minimally acceptable value of 1.25
recommended by Meehl (1995). Using the slightly less conserva-
tive cut score of 25 suggested by Harris et al. (1994), we found that
the average indicator validity value was 1.40 (see Table 1).
The four MAMBAC curves are presented in Figure 1. As can be
seen, none of these curves evidenced an inverted-U shape that
would be consistent with a taxonic fit for the data. On the contrary,
all four curves have a concave appearance that is typical of a
dimensional latent structure. The average base rate estimate was
.59 (SD ? .09), with a range of .49 to .68.
Figure 2 displays the average of these four MAMBAC graphs
juxtaposed with the graphs for the simulated taxonic and dimen-
sional data sets. The average MAMBAC graph produced by our
research data looked much more like the graph produced by the
simulated dimensional data than like the graph produced by the
simulated taxonic data. In fact, the average MAMBAC curve fit
the simulated dimensional data (FitRMSR? .009) better than the
simulated taxonic data (FitRMSR? .023). Cohen’s d for the 10
replications was 6.10.
Figure 3 presents the four MAXEIG curves, which also show no
evidence of a taxonic structure. Instead, these curves are consistent
with a dimensional structure, especially one that would be pro-
duced by negatively skewed indicators. The average base rate
estimate was .78, with a much narrower range of .74 to .82 (SD ?
.04). Although these are relatively stable, they are far higher than
would be expected given the MAMBAC average base rate. Figure
4 presents the average MAXEIG curve along with the curves from
the simulated data. Once again, the research data were more
consistent with the simulated dimensional curve (FitRMSR? .079)
than with the curve generated by the simulated taxonic data
(FitRMSR? .125). Cohen’s d for the 10 replications was 1.69.3
Finally, the L-Mode graph (see Figure 5) bore little resemblance
to the bimodal curve one would expect in a taxonic structure. In
contrast, the simulated taxonic data produced a clearly bimodal
curve. Averaging the base rate estimates, the left and right modes
yielded a .48 base rate estimate, and the base rate estimate based
on the classification of cases was .49. L-Mode base rate estimates
of about .50 are typical for dimensional constructs (A. M. Ruscio,
Ruscio, & Keane, 2002; Waller & Meehl, 1998), and both esti-
mates differ from the estimates provided by MAMBAC and
MAXEIG. They are also higher than would be expected in this
sample if psychopathy were underpinned by a taxon roughly
consistent with a PCL–R cut score of 30.
Attempted Replication of Harris et al. (1994)
As noted above, for the second set of analyses, we used the same
set of eight indicators examined in the Harris et al. (1994) study
(PCL–R Items 3, 5, 8, 9, 12, 13, 14, and 15). To check for nuisance
covariance (i.e., correlations among the indicators within either the
taxon or the complement), we computed correlations among these
eight items for those above and below the PCL–R cut score of 30
(Meehl & Golden, 1982). There was little nuisance correlation for
2Copies of the set of taxometric graphs produced for each subgroup are
available from John F. Edens.
3Because there are multiple methods for simulating data, we also used
Fraley’s program (see Hankin et al., 2005) to simulate data for a set of
MAXCOV analyses. These graphs were quite similar to those produced by
J. Ruscio’s (2004) program. Once again, the results of the dimensional
simulation were also much closer to the research data than were the results
of the taxonic simulation. Copies of all of these graphs are available from
John F. Edens.
Psychopathy Checklist—Revised facet scores: Interpersonal (Indicator 1),
Affective (Indicator 2), Lifestyle (Indicator 3), and Antisocial (Indicator 4).
The input indicator for each curve (x-axis) was created by combining the
three facet scores that were not the output indicator. The data were sorted
by the scores on this input indicator, and 50 cuts were made along this input
indicator. The y-axis depicts the differences between the mean scores on
the output indicator of individuals falling above and below each cut. To
stabilize the shape of each curve, we replicated these MAMBAC analyses
10 times by randomly shuffling the cases with equal scores on the input
indicator and recalculating the difference scores on the output indicator (J.
Ruscio et al., 2004).
Mean above minus below a cut (MAMBAC) curves for the four
EDENS, MARCUS, LILIENFELD, AND POYTHRESS
those above the cut, with an average correlation among these eight
items of .04. The average correlation below the cut was also
acceptable (r ? .17). However, the overall correlation among these
eight items in our entire sample was not especially high (r ? .22).
More important, these eight items did not evidence acceptable
levels of indicator validity, averaging only 0.75 (see Table 1),
which is considerably smaller than the level recommended by
Meehl (1995). Given our attempt to replicate the findings of Harris
et al. (1994), we conducted taxometric analyses on these items
despite the poor indicator validity estimates. Such poor indica-
tor validities call into question the generalizability of Harris et
al.’s (1994) findings, especially given that our administration
of the PCL–R yielded results that were more consistent with
the standardization sample in terms of corrected item-to-total
The curves resulting from these eight indicators were more
consistent with a dimensional than a taxonic structure. Figure 6
provides the average of the eight MAMBAC curves as well as the
curves for the simulated taxonic and dimensional data. The aver-
age MAMBAC curve for these analyses failed to demonstrate the
inverted-U shape that would be expected for taxonic data and was
much more similar to the dimensional data (FitRMSR? .006) than
to the taxonic data (FitRMSR? .016). Cohen’s d for the 10
replications was 5.53. Furthermore, the average base rate estimate
from these curves was .69 (SD ? .11), which is much higher
than would be expected in this sample and markedly inconsis-
tent with the results of Harris et al. (1994). The estimates
ranged from .58 to .94.
Figure 7 presents the average curve generated for the MAXCOV
analyses. This curve does not display an inverted-U shape that
would be consistent with a taxon. Instead, this curve was more
consistent with the simulated dimensional curve (FitRMSR? .022)
than with the curve generated by the simulated taxonic data
(FitRMSR? .088). Cohen’s d for the 10 replications was 5.49.
Furthermore, at best only 5 of the 28 individual curves appeared as
simulated dimensional data for the four Psychopathy Checklist—Revised facet scores. The graphs for the
simulated taxonic and dimensional data were produced by generating 10 data sets for each latent structure. The
darker lines represent the actual data, and the lighter lines represent one standard deviation above and below the
average for each simulated data set.
Average mean above minus below a cut curves for the research data, simulated taxonic data, and
athy Checklist—Revised facet scores: Interpersonal (Indicator 1), Affec-
tive (Indicator 2), Lifestyle (Indicator 3), and Antisocial (Indicator 4). The
data were sorted along the x-axis by the scores on the input indicator and
then grouped into 50 subsamples using overlapping windows (.90 overlap).
The remaining indicators were factor analyzed, and the eigenvalue of the
first principal factor was plotted on the y-axis. To stabilize the shape of
each curve, we replicated these MAXEIG analyses 10 times by randomly
shuffling the cases with equal scores before making the cut on the input
indicator and recalculating the eigenvalues on the output indicator (J.
Ruscio et al., 2004). Vars ? Variables.
Maximum eigenvalue (MAXEIG) curves for the four Psychop-
PCL–R TAXOMETRIC ANALYSES
if they might be consistent with a taxonic structure.4There was
also no clear pattern to the taxonic curves that might have sug-
gested that particular indicators were especially valid. The average
base rate estimate (.67) was consistent with the average from the
MAMBAC analyses, but these estimates were not at all consistent
with one another, ranging from .24 to .94 (SD ? .23). They were
also inconsistent, on average, with the base rates reported by
Harris et al. (1994).
Finally, unlike the L-Mode graph for the simulated taxonic data,
despite a slight hitch, the L-Mode graph for the research data (see
Figure 8) did not appear to be bimodal. Averaging the base rate
estimates from the left and right modes yielded a .39 base rate
estimate, and the base rate estimate based on the classification of
cases was .45. Although these two estimates were fairly consistent,
they were not at all consistent with the much higher estimates
yielded by MAMBAC and MAXCOV.
Overall, despite using the same indicators as Harris et al. (1994),
the results of these taxometric procedures seem quite inconsistent
with a taxonic structure. Although this finding is to be expected
given the very poor indicator validity of these eight items, the point
of these analyses is that it was not possible to replicate Harris et
al.’s (1994) findings despite using the same indicators and
Across multiple taxometric procedures developed by Meehl and
his colleagues (Meehl & Yonce, 1994; Waller & Meehl, 1998)—
MAMBAC, MAXEIG (or MAXCOV), and L-mode—our analy-
ses failed to offer support for the contention that psychopathy, as
identified by the PCL–R, is underpinned by a latent taxon. None of
the taxometric graphs appeared taxonic (i.e., the MAMBAC and
MAXEIG curves lacked an inverted-U shape, and the L-Mode
curves were unimodal), and the base rate estimates derived from
our analyses generally were inconsistent within and across proce-
dures, further suggesting that a taxonic solution is implausible.
Additionally, although it is difficult to know what base rate one
would have anticipated in our sample if psychopathy were taxonic,
the average base rates reported here seem to be remarkably high,
especially for the MAXCOV analyses. Such results are difficult to
reconcile with the prevailing notion that primary or “Cleckley”
psychopaths (Cleckley, 1988), who are ostensibly assessed by the
PCL–R, represent a narrow subset of offenders nested with the
heterogeneous category of persons with significant histories of
antisocial conduct (Hare, 2003; Lykken, 1995). Finally, our anal-
yses did not reveal taxonicity for the PCL–R antisocial items
identified by Harris et al. (1994) as taxonic.
Harris et al.’s (1994) article declaring psychopathy to be taxonic
has been highly influential (a Web of Science search revealed that
this article has been cited over 140 times). Nevertheless, our
findings, in conjunction with those for a self-report measure of
psychopathy (the PPI) that correlates moderately with the PCL–R
(Marcus et al., 2004), cast serious doubts on their conclusions.
Although the reasons for the striking discrepancy between our
findings and those of Harris et al. (1994) are unclear, their reliance
on file data alone to score the PCL–R, unusually high item-to-total
score correlations, dichotomous approach to scoring the PCL–R,
and inclusion of a large number of offenders who had been
adjudicated NGRI are likely candidates. The lattermost point raises
the possibility that Harris et al. (1994) may have inadvertently
detected a taxon for schizotypy. Schizotypy has been found to be
4Copies of all 8 MAMBAC curves and all 28 MAXCOV curves for
these analyses are available from John F. Edens.
5In addition to eight Harris et al. (1994) items, we also performed a set
of analyses using the eight items that correlated most highly with the
PCL–R total score in our sample (Items 1, 2, 4, 5, 6, 7, 8, and 10). Although
these indicators yielded slightly better validity estimates overall, they still
were unacceptably low (1.02), calling into question the appropriateness of
using individual PCL–R items for taxometric analyses. Similar to our
analyses of the items used by Harris et al. (1994), the analyses were
inconsistent with a taxonic latent structure. Copies of these graphs and
results are available from John F. Edens.
dimensional data for the four Psychopathy Checklist—Revised facet scores. The graphs for the simulated taxonic
and dimensional data were produced by generating 10 data sets for each latent structure. The darker lines
represent the actual data, and the lighter lines represent one standard deviation above and below the average for
each simulated data set.
Average maximum eigenvalue curves for the research data, simulated taxonic data, and simulated
EDENS, MARCUS, LILIENFELD, AND POYTHRESS
categorical in numerous studies using taxometric procedures (e.g.,
Lenzenweger & Korfine, 1992; see Haslam, 2003, for a review),
and, as noted earlier, psychopathy measures may misclassify at
least some individuals with schizophrenia spectrum disorders as
psychopathic (Howard et al., 1984). Although Harris et al. (1994)
found evidence for taxonicity even after excluding all participants
with a Diagnostic and Statistical Manual of Mental Disorders (3rd
ed.; American Psychiatric Association, 1980) diagnosis of psy-
chotic disorder in subsidiary analyses, it is possible that a substan-
tial number of individuals with schizophrenia spectrum disorders
(e.g., schizotypal personality disorder, paranoid personality disor-
der) remained in their sample even following this exclusion (e.g.,
over 100 participants who were adjudicated NGRI were included
in the analyses that excluded psychotic inmates).
Furthermore, evidence suggests that taxometric analyses of rat-
ing scale data can be influenced substantially by the perceptions of
the persons conducting the ratings (Beauchaine & Waters, 2003).
Along these lines, it is possible that the raters in our study differed
in some important respect from the raters who coded the archival
file data in the Harris et al. (1994) study. One possibility is that the
the simulated taxonic and dimensional data sets. Each graph displays the frequency distribution of scores on the
first factor of a factor analysis of the indicator set. The graphs for the simulated taxonic and dimensional data
were produced by generating 10 data sets for each latent structure (dotted lines). The solid lines indicate the
average of these data sets.
Latent mode factor analysis curves for the four Psychopathy Checklist—Revised facet scores and for
data, and simulated dimensional data for the eight Psychopathy Checklist—Revised items used as indicators by
Harris et al. (1994). The input indicator for each curve (x-axis) was created by combining the seven items that
were not the output indicator. The data were sorted by the scores on this input indicator, and 50 cuts were made
along this input indicator. The y-axis depicts the differences between the mean scores on the output indicator of
individuals falling above and below each cut. To stabilize the shape of each curve, we replicated these
MAMBAC analyses 10 times by randomly shuffling the cases with equal scores on the input indicator and
recalculating the difference scores on the output indicator (J. Ruscio et al., 2004). The graphs for the simulated
taxonic and dimensional data were produced by generating 10 data sets for each latent structure. The darker lines
represent the actual data, and the lighter lines represent one standard deviation above and below the average for
each simulated data set.
Average mean above minus below a cut (MAMBAC) curves for the research data, simulated taxonic
PCL–R TAXOMETRIC ANALYSES
interview information obtained from participants in our study led
to more nuanced ratings of the underlying dimensional construct
assessed by the PCL–R. Alternatively, one could argue that such
information might blur an underlying categorical distinction that is
more evident when relying exclusively on file data. The fact that
Harris et al. (1994) obtained item-to-total correlations that were
unusually high (relative to our data and the data reported in the
PCL–R manual) would seem more consistent with the former
interpretation rather than the latter, although this is by no means
conclusive evidence. To our knowledge, no one has experimen-
tally manipulated a priori beliefs of PCL–R examiners to test
systematically whether or how their ratings could affect taxometric
analyses, although this would be a fruitful area of future
Our failure to identify a latent taxon has important implications
for the etiology and assessment of this disorder, particularly as
other investigators also find evidence of a dimensional latent
structure using the PCL–R (see, e.g., Guay, Ruscio, Hare, &
Knight, 2005; Looman & Abracen, 2005). If psychopathy were
dimensional at a latent level, it would imply that researchers
should direct more of their investigative efforts toward etiological
agents that are themselves dimensional, such as fearlessness
simulated dimensional data for the eight Harris et al. (1994) indicators. The data were sorted along the x-axis by
summing six of the indicators (i.e., the input indicator) and then grouped into 50 subsamples using overlapping
windows (.90 overlap). The covariance between the two other indicators was then was plotted on the y-axis. This
procedure produced 28 graphs that were averaged to create this graph. To stabilize the shape of each of the 28
curves, we replicated these MAXCOV analyses 10 times by randomly shuffling the cases with equal scores
before making the cut on the input indicator and recalculating the covariance on the output indicator (J. Ruscio
et al., 2004). The graphs for the simulated taxonic and dimensional data were produced by generating 10 data
sets for each latent structure. The darker lines represent the actual data, and the lighter lines represent one
standard deviation above and below the average for each simulated data set.
Average maximum covariance (MAXCOV) curves for the research data, simulated taxonic data, and
taxonic and dimensional data sets. Each graph displays the frequency distribution of scores on the first factor of
a factor analysis of the indicator set. The graphs for the simulated taxonic and dimensional data were produced
by generating 10 data sets for each latent structure (dotted lines). The solid lines indicate the average of these
Latent mode factor analysis curves for the eight Harris et al. (1994) indicators and for the simulated
EDENS, MARCUS, LILIENFELD, AND POYTHRESS
(Lykken, 1995), role-taking deficiency (Gough, 1960), graded
deficits in response modulation (Patterson & Newman, 1993), or
other continuously distributed source traits (Cattell, 1951) that
could give rise to the manifold surface traits of psychopathy. Our
findings are also potentially consistent with models positing that
psychopathy is a constellation or configuration of extreme scores
on several continuously distributed personality dimensions, such
as low conscientiousness and low agreeableness (Benning et al.,
2005; J. D. Miller et al., 2001), although they do not provide
support for any particular dimensional model of psychopathy, such
as the five-factor model (Lynam & Widiger, 2001). Conversely,
our findings run counter to the position that psychopathy is un-
derpinned by a dichotomous causal agent (Meehl & Golden, 1982)
and offer little support for developmental models (e.g., Quinsey,
Skilling, Lalumiere, & Craig, 2004) that posit the existence of a
qualitatively distinct or discrete natural class of psychopathic
youths. We should note that, although our results are generally
inconsistent with etiological models based on dichotomous causal
factors, they do not address whether the etiology of psychopathy is
similar to or appreciably different from other externalizing
Our results also dovetail with structural equation models sug-
gesting that various phenotypically diverse psychological condi-
tions and characteristics, including constructs closely related to
PCL–R Factor 2, are underpinned by a latent externalizing dimen-
sion (e.g., Krueger et al., 2002; Krueger, Markon, Patrick, &
Iacono, 2005; see also Gorenstein & Newman, 1980). Moreover,
the latent structure models used by Krueger and colleagues address
the continuous versus categorical question in a manner quite
distinct from the taxometric procedures of Meehl (Meehl & Yonce,
1994; Waller & Meehl, 1998), in that they “involve fitting explicit
mathematical models to sample data by use of well-characterized
estimators of population parameters (e.g., maximum likelihood)
and evaluating the fit of these models by use of quantitative indices
of fit” (Krueger et al., 2005, p. 540). Future research applying
these methods to the psychopathic traits assessed by the PCL–R
would be informative because (a) replication of the dimensional
structure of psychopathy via separate statistical procedures would
bolster the present findings, (b) there is a need to integrate taxo-
metrics with other latent-variable techniques (Waldman & Lilien-
feld, 2001), and (c) these latent structure models can address issues
of comorbidity with other forms of externalizing psychopathology.
On the assessment front, our findings call into question the
widespread practice of dichotomizing or trichotomizing total
scores on the PCL–R to establish discrete groups (Lilienfeld,
1994). If psychopathy were dimensional at a latent level, this
practice would yield scientifically arbitrary groupings of individ-
uals who differ along one or more continuous dimensions. More-
over, this practice would result in eliminating potentially useful
assessment information regarding individuals’ standing on these
dimensions. From the perspective of Gangestad and Snyder
(1985), the class of psychopaths may be a phenetic category, that
is, an arbitrarily formed grouping that possesses no intrinsic mean-
ing (e.g., professional basketball player may be a category, but
there is no natural relationship between height and athletic ability).
Of course, our findings do not imply that distinctions between
psychopaths and nonpsychopaths cannot be made for practical
purposes, such as risk management or violence prediction. If other
research teams replicate our findings, however, this would imply
that such distinctions are purely pragmatic and do not “carve
nature at its joints,” because in the case of psychopathy, there are
no discrete joints to carve.
It is perhaps worth noting, however, that even if psychopathy
were shown to be taxonic, this would not necessarily lend support
to the current PCL–R cutoff of 30, as this score may not corre-
spond to the maximally predictive threshold (Meehl & Golden,
1982) for distinguishing the psychopathy taxon from the nonpsy-
chopathy complement group (cf. Gacono, Loving, & Bodholdt,
2001). Moreover, dichotomizing continuous distributions typically
results in a loss in statistical power (Cohen, 1983; McCallum,
Zhang, Preacher, & Rucker, 2002; but see Farrington & Loeber,
2000, for exceptions), unless there are clear-cut violations of
normality in the data. Given that there does not appear to be a clear
breaking point in the observed distributions of psychopathy scores,
such dichotomization is difficult to justify on strictly scientific
grounds, although it may often simplify the presentation of re-
search findings (Farrington & Loeber, 2000).
From a policy perspective, there is ample evidence that the legal
system is keenly interested in the identification of psychopaths,
based on the amount of legislation in North America and Europe
that uses this or related phrases such as dangerous and severe
personality disorders (see Edens & Petrila, 2006, for an overview).
Similarly, references to psychopathy in case law often involve
debates as to whether a defendant is in fact “a psychopath” (i.e.,
PCL–R ? 30). Such views appear to assume a categorical per-
spective, in that the intent is to identify a class of individuals
deemed appropriate for some form of legal sanction because of
their mental disorder or behavioral abnormality (Edens, in press).
To the extent that our results undermine the implicit or explicit
legal presumption that psychopaths are a discrete category of
criminals, they suggest that it is largely arbitrary to draw precise
categorical boundaries between psychopathic and nonpsycho-
pathic offenders. Although decision makers can and do use PCL–R
scores to inform legal decisions that are by definition categorical
(e.g., presence or absence of a behavioral abnormality, indetermi-
nate commitment), there is no clear scientific evidence for a
natural breaking point at which such categories should be defined
regarding psychopathy. That said, our results by no means argue
against the possibility of using particular cut scores that may prove
valuable for any number of pragmatic decisions (e.g., optimal cut
scores to reduce community violence among released offenders),
which may not necessarily correspond to any putative diagnostic
thresholds (see, e.g., Skeem & Mulvey, 2001).
One limitation of our study, like that of Harris et al. (1994),
Skilling et al. (2001), and the vast majority of taxometric studies
across constructs, is mono-operation bias (Shadish, Cook, &
Campbell, 2002). As Meehl and Golden (1982) noted, taxometric
analyses are ideally performed on indicators derived from maxi-
mally independent domains. In contrast, all of our indicators
derived from one well-validated measure, namely, the PCL–R. It is
conceivable that alternative indicators that tap processes that may
be etiologically relevant to psychopathy, such as poor passive-
avoidance learning (Lykken, 1957; Schacter & Latane, 1964), poor
response modulation (Patterson & Newman, 1993), weak fear
potentiated startle (Patrick, Bradley, & Lang, 1993), or diminished
electrodermal classical conditioning to aversive stimuli (Lykken,
1957), could yield evidence for the taxonicity of psychopathy.
Such presumed “endophenotypic markers” (Gottesman & Gould,
PCL–R TAXOMETRIC ANALYSES
2003) should be incorporated in future investigations of the taxo-
nicity of psychopathy, in part because they may reflect certain
processes that are more closely tied etiologically to psychopathy
and in part because they should help to obviate the problem of
method covariance introduced by an exclusive reliance on a single
measure (e.g., the PCL–R, the PPI). Nevertheless, one should not
necessarily assume that biological or laboratory markers are nec-
essarily endophenotypic as opposed to exophenotypic (hence, our
use of the term presumed in the previous sentence), as they could
reflect the action of processes (e.g., attentional deficits) that lie
causally downstream from psychopathy. Moreover, at least some
of these markers appear to be relatively nonspecific; for example,
deficits in passive-avoidance learning have been reported not only
in psychopathic individuals but also in individuals with borderline
personality disorder (Hochhausen, Lorenz, & Newman, 2002).
In summary, our findings contrast sharply with those of Harris
et al. (1994) and fail to offer support for the view that psychopathy
assessed by the PCL–R is underpinned by a discrete taxon. Our
results are consistent with recent calls for closer research linkages
between the often-disconnected domains of personality and psy-
chopathology (e.g., Benning et al., 2005; Harkness & Lilienfeld,
1997; Krueger & Tackett, 2003; Lynam & Widiger, 2001). More-
over, they are encouraging in that they suggest that researchers
ultimately may be able to draw from the large body of research on
the assessment and causes of continuously distributed personality
traits to better inform their understanding of psychopathy among
American Psychiatric Association. (1980). Diagnostic and statistical man-
ual of mental disorders (3rd ed.). Washington, DC: Author.
Ammons, R. B., & Ammons, C. H. (1962). The Quick Test (QT): Provi-
sional manual. Psychological Reports, 11, 111–161.
Beauchaine, T. P., & Waters, E. (2003). Pseudotaxonicity in MAMBAC
and MAXCOV analyses of rating-scale data: Turning continua into
classes by manipulating observer’s expectations. Psychological Meth-
ods, 8, 3–15.
Benning, S. D., Patrick, C. J., Blonigen, D. M., Hicks, B. M., & Iacono,
W. G. (2005). Estimating facets of psychopathy from normal personality
traits: A step toward community epidemiological investigations. Assess-
ment, 12, 3–18.
Brinkley, C., Newman, J., Widiger, T., & Lynam, D. (2004). Two ap-
proaches to parsing the heterogeneity of psychopathy. Clinical Psychol-
ogy: Science & Practice, 11, 69–94.
Bucholz, K. K., Hesselbrock, V. M., Heath, A. C., Kramer, J. R., &
Schuckit, M. A. (2000). A latent class analysis of antisocial personality
disorder symptom data from a multi-centre family study of alcoholism.
Addiction, 95, 553–567.
Cale, E., & Lilienfeld, S. (2002). Sex differences in psychopathy and
antisocial personality disorder: A review and integration. Clinical Psy-
chology Review, 22, 1179–1207.
Cattell, R. B. (1951). A factorization of tests of personality source traits.
British Journal of Psychology, 4, 165–178.
Cleckley, H. (1988). The mask of sanity. St. Louis: Mosby.
Cohen, J. (1983). The cost of dichotomization. Applied Psychological
Measurement, 7, 249–253.
Cooke, D. J., & Michie, C. (1997). An item response theory analysis of the
Hare Psychopathy Checklist—Revised. Psychological Assessment, 9,
Cooke, D. J., & Michie, C. (2001). Refining the construct of psychopathy:
Towards a hierarchical model. Psychological Assessment, 13, 171–188.
Cooke, D. J., Michie, C., Hart, S. D., & Clark, D. (2005). Searching for the
pan-cultural core of psychopathic personality disorder. Personality and
Individual Differences, 39, 283–295.
Craig, R. J., & Olson, R. E. (1988). Relationships between Wechsler Scales
and Quick Test IQs among disability applicants. Professional Psychol-
ogy: Research and Practice, 19, 26–30.
Cunningham, M. D., & Reidy, T. J. (2002). Violence risk assessment at
federal capital sentencing: Individualization, generalization, relevance,
and scientific standards. Criminal Justice and Behavior, 29, 512–537.
DeCato, C. M., & Husband, S. D. (1984). Quick Test and Wechsler Adult
Intelligence Scale—Revised in a prison’s clinical setting. Psychological
Reports, 54, 939–942.
DeMatteo, D., & Edens, J. F. (2005). The role and relevance of the
Psychopathy Checklist—Revised in court: A case law survey of United
States courts (1991–2004). Manuscript submitted for publication.
Doss, G. H., Head, D. W., Blackburn, J. V., & Robertson, J. M. (1986). A
quick measure of mental deficiency among adult offenders. Federal
Probation, 50, 57–59.
Edens, J. F. (in press). Unresolved controversies concerning psychopathy:
Implications for clinical and forensic decision making. Professional
Psychology: Research and Practice.
Edens, J. F., & Petrila, J. (2006). Legal and ethical issues in the assessment
and treatment of psychopathy. In C. Patrick (Ed.), Handbook of psy-
chopathy (pp. 573–588). New York: Guilford Press.
Edens, J. F., Petrila, J., & Buffington-Vollum, J. K. (2001). Psychopathy
and the death penalty: Can the Psychopathy Checklist—Revised identify
offenders who represent “a continuing threat to society?” Journal of
Psychiatry and Law, 29, 433–481.
Farrington, D. P., & Loeber, R. (2000). Some benefits of dichotomization
in psychiatric and criminological research. Criminal Behaviour and
Mental Health, 10, 100–122.
Gacono, C. B., Loving, J. L., & Bodholdt, R. H. (2001). The Rorschach and
psychopathy: Toward a more accurate understanding of the research
findings. Journal of Personality Assessment, 77, 16–38.
Gangestad, S. W., & Snyder, M. (1985). “To carve nature at its joints”: On
the existence of discrete classes in personality. Psychological Review,
Gorenstein, E. E., & Newman, J. P. (1980). Disinhibitory psychopathol-
ogy: A new perspective and a model for research. Psychological Review,
Gottesman, I. I., & Gould, T. D. (2003). The endophenotype concept in
psychiatry: Etymology and strategic intentions. American Journal of
Psychiatry, 160, 636–645.
Gough, H. G. (1960). Theory and method of socialization. Journal of
Consulting and Clinical Psychology, 24, 23–30.
Guay, J., Ruscio, J., Hare, R., & Knight, R. A. (2005). The latent structure
of psychopathy: When more is simply more. Manuscript submitted for
Guy, L. G., Edens, J. F., Anthony, C., & Douglas, K. S. (in press). Does
psychopathy predict institutional misconduct among adults? A meta-
analytic investigation. Journal of Consulting and Clinical Psychology.
Hankin, B. L., Fraley, R. C., Lahey, B. B., & Waldman, I. D. (2005). Is
depression best viewed as a continuum or discrete category? A taxo-
metric analysis of childhood and adolescent depression in a population-
based sample. Journal of Abnormal Psychology, 114, 96–110.
Hare, R. D. (1991). The Psychopathy Checklist—Revised manual. Toronto,
Ontario, Canada: Multi-Health Systems.
Hare, R. D. (2003). The Psychopathy Checklist—Revised manual (2nd ed.).
Toronto, Ontario, Canada: Multi-Health Systems.
Hare, R. D., & Harpur, T. (1986). Weak data, strong conclusions: Some
comments on Howard, Bailey, and Newman’s use of the Psychopathy
Checklist. Personality and Individual Differences, 7, 147–151.
Harkness, A. R., & Lilienfeld, S. O. (1997). Individual differences science
EDENS, MARCUS, LILIENFELD, AND POYTHRESS
for treatment planning: Personality traits. Psychological Assessment, 9,
Harpur, T. J., Hakstian, A. R., & Hare, R. D. (1988). Factor structure of the
Psychopathy Checklist. Journal of Consulting and Clinical Psychology,
Harpur, T. J., Hare, R. D., & Hakstian, A. R. (1989). Two-factor concep-
tualization of psychopathy: Construct validity and assessment implica-
tions. Psychological Assessment, 1, 6–17.
Harris, G. T., Rice, M. E., & Quinsey, V. L. (1994). Psychopathy as a
taxon: Evidence that psychopaths are a discrete class. Journal of Con-
sulting and Clinical Psychology, 62, 387–397.
Harris, G. T., Skilling, T., & Rice, M. E. (2001). The construct of psy-
chopathy. Crime and Justice, 28, 197–264.
Haslam, N. (2003). The dimensional view of personality disorders: A review
of the taxometric evidence. Clinical Psychology Review, 23, 75–93.
Hemphill, J. F., Hare, R. D., & Wong, S. (1998). Psychopathy and
recidivism: A review. Legal and Criminological Psychology, 3, 139–170.
Hill, C., Neumann, C., & Rogers, R. (2004). Confirmatory factor analysis
of the Psychopathy Checklist: Screening version in offenders with Axis
I disorders. Psychological Assessment, 16, 90–95.
Hochhausen, N., Lorenz, A., & Newman, J. (2002). Specifying the impul-
sivity of female inmates with borderline personality disorder. Journal of
Abnormal Psychology, 111, 495–501.
Howard, R. C., Bailey, R., & Newman, A. (1984). A preliminary study of
Hare’s “Research Scale for the Assessment of Psychopathy” in
mentally-abnormal offenders. Personality and Individual Differences, 5,
Krueger, R. F., Hicks, B. M., Patrick, C. J., Carlson, S. R., Iacono, W. G.,
& McGue, M. (2002). Etiologic connections among substance depen-
dence, antisocial behavior, and personality: Modeling the externalizing
spectrum. Journal of Abnormal Psychology, 111, 411–424.
Krueger, R. F., Markon, K. E., Patrick, C. J., & Iacono, W. G. (2005).
Externalizing psychopathology in adulthood: A dimensional-spectrum
conceptualization and its implications for DSM–V. Journal of Abnormal
Psychology, 114, 537–550.
Krueger, R. F., & Tackett, J. L. (2003). Personality and psychopathology:
Working toward the bigger picture. Journal of Personality Disorders,
Lenzenweger, M. F. (2004). Consideration of the challenges, complica-
tions, and pitfalls of taxometric analysis. Journal of Abnormal Psychol-
ogy, 113, 10–23.
Lenzenweger, M. F., & Korfine, L. (1992). Confirming the latent structure
and base rate of schizotypy: A taxometric analysis. Journal of Abnormal
Psychology, 101, 567–571.
Lilienfeld, S. O. (1994). Conceptual problems in the assessment of psy-
chopathy. Clinical Psychology Review, 14, 17–38.
Lilienfeld, S. O. (1998). Methodological advances and developments in the
assessment of psychopathy. Behaviour Research and Therapy, 36, 99–125.
Lilienfeld, S. O., & Andrews, B. P. (1996). Development and preliminary
validation of a self-report measure of psychopathic personality traits in
noncriminal populations. Journal of Personality Assessment, 66, 488–
Lilienfeld, S. O., & Fowler, K. A. (2006). The self-report assessment of
psychopathy: Pitfalls, problems, and promises. In C. Patrick (Ed.),
Handbook of psychopathy (pp. 107–132). New York: Guilford Press.
Looman, J., & Abracen, J. (2005). Psychopathy as a taxon or a dimension?
Manuscript submitted for publication.
Lorber, M. F. (2004). Psychophysiology of aggression, psychopathy, and
conduct problems: A meta-analysis. Psychological Bulletin, 130, 531–
Lykken, D. T. (1957). A study of anxiety in the sociopathic personality.
Journal of Abnormal and Social Psychology, 55, 6–10.
Lykken, D. T. (1995). The antisocial personalities. Hillsdale, NJ: Erlbaum.
Lynam, D. R., & Widiger, T. A. (2001). Using the five-factor model to
represent the DSM–IV personality disorders: An expert consensus ap-
proach. Journal of Abnormal Psychology, 110, 410–412.
Marcus, D. K., John, S., & Edens, J. F. (2004). A taxometric analysis of
psychopathic personality. Journal of Abnormal Psychology, 113, 626–
McCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, D. D. (2002). On
the practice of dichotomization of quantitative variables. Psychological
Methods, 7, 19–40.
Meehl, P. E. (1995). Bootstraps taxometrics: Solving the classification
problem in psychopathology. American Psychologist, 50, 266–275.
Meehl, P. E., & Golden, R. (1982). Taxometric methods. In P. Kendall &
J. Butcher (Eds.), Handbook of research methods in clinical psychology
(pp. 127–181). New York: Wiley.
Meehl, P. E., & Yonce, L. J. (1994). Taxometric analysis: I. Detecting
taxonicity with two quantitative indicators using means above and below
a sliding cut (MAMBAC procedure). Psychological Reports, 74, 1059–
Meehl, P. E., & Yonce, L. J. (1996). Taxometric analysis: II. Detecting
taxonicity using covariance of two quantitative indicators in successive
intervals of a third indicator (MAXCOV procedure). Psychological
Reports, 78, 1091–1227.
Miller, J. D., Lynam, D. R., Widiger, T. A., & Leukefeld, C. (2001).
Personality disorders as extreme variants of common personality dimen-
sions: Can the five-factor model adequately represent psychopathy?
Journal of Personality, 69, 253–276.
Miller, M. B. (1996). Limitations of Meehl’s MAXCOV-HITMAX proce-
dure. American Psychologist, 51, 554–556.
Ogloff, J. R. P., & Lyon, D. R. (1998). Legal issues associated with the
concept of psychopathy. In D. J. Cooke, A. E. Forth, & R. D. Hare
(Eds.), Psychopathy: Theory, research, and implications for society (pp.
401–422). Mahwah, NJ: Kluwer.
Osgood, D. W., McMorris, B. J., & Potenza, M. T. (2002). Analyzing
multiple-item measures of crime and deviance: I. Item response theory
scaling. Journal of Quantitative Criminology, 18, 267–296.
Patrick, C. J., Bradley, M. M., & Lang, P. J. (1993). Emotion in the
criminal psychopath: Startle reflex modulation. Journal of Abnormal
Psychology, 102, 82–92.
Patterson, C. M., & Newman, J. P. (1993). Reflectivity and learning from
aversive events: Toward a psychological mechanism for the syndromes
of disinhibition. Psychological Review, 100, 716–736.
Poythress, N. G., Edens, J. F., & Lilienfeld, S. O. (1998). Criterion-related
validity of the Psychopathic Personality Inventory in a prison sample.
Psychological Assessment, 10, 426–430.
Quinsey, V. L., Harris, G. T., Rice, M. E., & Cormier, C. A. (1998). Violent
offenders: Appraising and managing risk. Washington, DC: American
Quinsey, V., Skilling, T., Lalumiere, M., & Craig, W. (2004). A taxonomy
of juvenile delinquency and an integrated theoretical perspective. In V.
Quinsey et al. (Eds.), Juvenile delinquency: Understanding the origins of
individual differences (pp. 93–114). Washington, DC: American Psy-
Ruscio, A. M., Ruscio, J., & Keane, T. M. (2002). The latent structure of
posttraumatic stress disorder: A taxometric investigation of reactions to
extreme stress. Journal of Abnormal Psychology, 111, 290–301.
Ruscio, J. (2004). Taxometric program documentation, R language [Com-
puter software and manual]. Retrieved August 10, 2004, from http://
Ruscio, J. (2005). Estimating taxon base rates using the MAXCOV,
MAXEIG, and MAMBAC taxometric procedures: A Monte Carlo study.
Manuscript submitted for publication.
Ruscio, J., Ruscio, A. M., & Keane, T. M. (2004). Using taxometric
analysis to distinguish a small latent taxon from a latent dimension with
positively skewed indicators: The case of involuntary defeat syndrome.
Journal of Abnormal Psychology, 113, 145–154.
PCL–R TAXOMETRIC ANALYSES
Schacter, S., & Latane, B. (1964). Crime, cognition, and the autonomic Download full-text
nervous system. In D. Levine (Ed.), Nebraska Symposium on Motivation
(Vol. 12, pp. 221–275). Lincoln: University of Nebraska Press.
Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and
quasi-experimental designs for generalized causal inference. Boston:
Simon, M. J. (1995). Relationship between the Quick Test and WAIS–R in
low-functioning male criminal defendants. Psychological Reports, 77,
Skeem, J. L., Edens, J. F., Camp, J., & Colwell, L. H. (2004). Are there
racial differences in levels of psychopathy? A meta-analysis. Law and
Human Behavior, 28, 505–527.
Skeem, J. L., & Mulvey, E. P. (2001). Psychopathy and community
violence among civil psychiatric patients: Results from the MacArthur
Violence Risk Assessment Study. Journal of Consulting and Clinical
Psychology, 69, 358–374.
Skeem, J. L., Poythress, N. G., Edens, J. F., Lilienfeld, S. O., & Cale, E.
(2003). Psychopathic personality or personalities? Exploring potential
variants of psychopathy and their implications for risk assessment.
Aggression and Violent Behavior, 8, 513–546.
Skilling, T. A., Harris, G. T., Rice, M. E., & Quinsey, V. L. (2002).
Identifying persistently antisocial offenders using the Hare Psychopathy
Checklist and DSM antisocial personality disorder criteria. Psychologi-
cal Assessment, 14, 27–38.
Skilling, T. A., Quinsey, V. L., & Craig, W. M. (2001). Evidence of a taxon
underlying serious antisocial behavior in boys. Criminal Justice and
Behavior, 28, 450–470.
Traub, G. S., & Spruill, J. (1982). Correlations between the Quick Test and
Wechsler Adult Intelligence Scale—Revised. Psychological Reports,
Waldman, I. D., & Lilienfeld, S. O. (2001). Applications of taxometric
methods to problems of comorbidity: Perspectives and challenges. Clin-
ical Psychology: Science and Practice, 8, 520–527.
Waller, N. G., & Meehl, P. E. (1998). Multivariate taxometric procedures:
Distinguishing types from continua. Newbury Park, CA: Sage.
Wechsler, D. (1981). The Wechsler Adult Intelligence Scale—Revised
manual. San Antonio, TX: Psychological Corporation.
Received March 29, 2005
Revision received July 6, 2005
Accepted September 20, 2005 ?
EDENS, MARCUS, LILIENFELD, AND POYTHRESS