A taxometric analysis of the latent structure of psychopathy: Evidence for dimensionality

Text-only Preview

Latent structure of psychopathy


A taxometric analysis of the latent structure of psychopathy: Evidence for dimensionality

Jean-Pierre Guay, Ph.D.
Université of Montréal
Institut Philippe-Pinel de Montréal
John Ruscio, Ph.D.
Elizabethtown College
Raymond A. Knight, Ph.D.
Brandeis University
Robert D. Hare, Ph.D.
University of British Columbia
Published in Journal of Abnormal Psychology

Copyright 2007 by the American Psychological Association
This article may not exactly replicate the final version published in the APA journal. It is not the
copy of record.

Latent structure of psychopathy
The taxonomic status of psychopathy is controversial. Whereas some studies have found
evidence that psychopathy, at least its antisocial component, is distributed as a taxon, others have
found that both major components of psychopathy —callousness/unemotionality and
impulsity/antisocial behavior —appear to distribute as dimensions and show little evidence of
taxonicity. In the present study, recent advances in taxometric analysis were added to Meehl’s
multiple consistency tests strategy for assessing taxonicity (Meehl, 1995), and they were applied
to Psychopathy Checklist-Revised (PCL-R) ratings of 4865 offenders sampled from multiple
forensic settings. The results indicated that both the individual components of psychopathy and
their interface are distributed dimensionally. Both the implications of these results for research
in psychopathy and the integration of these findings with previous taxometric studies of
psychopathy are discussed.

Key words: Psychopathy, taxometrics, latent structure, antisocial behavior

Latent structure of psychopathy
A taxometric analysis of the latent structure of psychopathy: Evidence for dimensionality
Psychopathy is a personality disorder that comprises a distinct cluster of emotional,
interpersonal, and behavioural characteristics (e.g., emotional detachment, callousness,
irresponsibility, impulsivity) and that is characterized by a disregard for the societal rules and the
rights of others (Hare, 1996). Its association with violence (Porter & Woodworth, 2006) and its
usefulness as a risk factor in predicting criminal recidivism (Douglas, Vincent, & Edens, 2006)
have increased its prominence in the last decade in both criminology and psychopathology. The
origins of our current conceptualization of the construct can be traced to Cleckley’s (1976)
classic description of the syndrome. He delineated its characteristics without, however,
addressing the issue of whether it represents the coalescence of extreme manifestations on a
number of dimensional traits or constitutes a taxon, i.e., a distinct, nonarbitrary entity or class.
The first formalized assessment tool for measuring psychopathy was Hare’s (1980)
operationalization of the construct in the Psychopathy Checklist (PCL) and its revision, the PCL-
R (Hare, 1991, 2003). Although Hare conceptualized the PCL as a way of indicating how closely
an individual approximated the “prototypic psychopath” and proposed a PCL-R cutting score of
30 (out of 40) to consider a person sufficiently close to the psychopath prototype, he also
recognized that viable arguments could be made for using the PCL-R to obtain dimensional
The theoretical case for hypothesizing that psychopathy may be distributed as a taxon
revolves around evidence for specific genetic, neurobiological, cognitive, and affective
covariates of the construct that are consistent with the hypothesis that a specific etiology or
specific etiologies may account for it. Probably the most widely accepted theoretical
understanding of taxonicity and the strongest examples of taxa have focused on entities specified

Latent structure of psychopathy
by the conjunction of a distinct pathology and etiology (Meehl, 1973; 1992). Although the
establishment of such covariates does not in itself confirm either a syndrome’s specific etiology
or its taxonicity, and dimensional models can be generated to handle such correlative findings,
the absence of such covariates would certainly argue against proposing a taxonic distribution.
Two (Harpur, Hakstian, & Hare, 1988; Harpur, Hare, & Hakstian, 1989), three (Cooke &
Michie, 2001), and four factor (Hare, 2003; Hare & Neumann, 2006) models of the PCL have all
identified two overarching components involving impulsivity-antisocial behavior and affective-
interpersonal features. Because these two components have consistently yielded distinct patterns
of correlations with characteristics relevant to psychopathy, different models of specific
underlying processes have been proposed for each (e.g., Bloningen, Hicks, Krueger, Patrick, &
Iacono, 2005; Fowles & Dindo, 2006; Patrick & Zempolich, 1998). Such models raise the
possibility that either or both factors may be distributed as taxa. A brief survey of some of the
correlative evidence in the genetic, neurobiological, cognitive, and affective domains provides an
adequate justification to support a taxonomic investigation.
There is considerable evidence from adoption and twin genetic studies, including those of
twins reared apart, that indicate that genetic factors both play a significant role in the likelihood
that a person will commit a criminal act (Gottesman & Goldsmith, 1994; Grove, Eckert, Heston,
Bouchard, et al., 1990; Mednick, Gabrielli, & Hutchings, 1984) and also increase the probability
that an individual will be diagnosed as Antisocial Personality Disorder (APD) during his lifetime
(Cadoret, Yates, Troughton, Woodworth, & Stewart, 1995; Ge, Conger, Cadoret, Neiderhiser,
Yates, et al., 1996; Lyons et al., 1995). Although behavioral genetic research suggests that the
Factor 2 features may have higher heritability than the Factor 1 (Edelbrock, Rende, Plomin, &
Thompson, 1995; Krueger, 2000; Mason & Frick, 1994; Depue, 1996, but see Livesley, 1998,

Latent structure of psychopathy
for contrary evidence), there are also data supporting the importance of independent genetic
influences in the manifestation of Factor 1 (Taylor, Loney, Bobadilla, Iacono, McGue, 2003;
Patrick, 2003).
A number of neurobiological deficits or anomalies have been identified in both
psychopathic criminals and “successful” psychopaths, who have largely avoided extensive
contact with the criminal justice system or extended incarceration. For instance, both specific
neurological structural features in the amygdala, orbito-frontal cortex, and hippocampus (e.g.,
Blair, 2004; Mitchell, Colledge, Leonard, & Blair, 2002; Raine, 2001; Raine et al., 2004;
Tiihonen et al., 2000) and functional anomalies in the amygdala/hippocampal formation,
parahippocampal gyrus, ventral striatum, anterior and posterior cingulated gyri, and fronto-
temporal cortex (e.g., Kiehl, Hare, Liddle, & McDonald, 1999; Kiehl, Smith, Mendrek, Forster,
Hare, & Liddle, 2004; Kiehl, Smith, Hare, Mendrek, Forster, Brink, & Liddle, 2001; Müller et
al., 2003; Völlm et al., 2004) have been linked to the syndrome or its components. Some
speculations have been proposed about a comprehensive and integrated mapping of these deficits
onto psychopathy (Blair, Mitchell, & Blair, 2005) and onto the two major PCL-R factors that
assess it (Patrick, 2003). An understanding of the latent distribution of the PCL-R and its factors
could substantially advance the search for such an integration.
Psychopaths have been found to be deficient on a number of cognitive (e.g., Hervé,
Hayes & Hare, 2001; Morgan & Lilienfeld, 2000; Newman & Lorenz, 2003) and affective tasks
(e.g., Blair, 2001; Blair, Mitchell, Richell, Kelly, Leonard, Newman & Scott, 2002; Patrick,
2001; Williamson, Harpur, & Hare, 1991) that map onto the structural features noted above.
Psychopaths’ difficulty in shifting a dominant behavior when contingencies have been reversed
(Newman & Lorenz, 2003) and their difficulties with working memory and other aspects of

Latent structure of psychopathy
executive functioning (Morgan & Lilienfeld, 2000; Séguin, 2004) implicate deficits in their
orbito-frontal cortex. In contrast, psychopaths’ impairments in passive avoidance learning
(Newman & Kosson, 1986), dysfunctional response to another’s sadness or fear (Blair, 1995),
reduced augmentation of the startle reflex by threat primes (Levenston, Patrick, Bradley, &
Lang, 2000), impaired aversive conditioning (Raine, Venables, & Williams, 1996), and deficient
processing of fearful expressions (Blair, Colledge, Murray, & Mitchell, 2001) are perhaps better
accounted for at the neural level by deficiencies in the functions of the amygdala (Blair, 2004) or
in the integration of frontal-limbic processes (Hare, 2003; Müller et al., 2003). Although
speculative models to integrate these deficits have been proposed (e.g., Blair et al., 2005; Fowles
& Dindo, 2006), unpacking the complexity of the development of the underlying core processes
(Séguin, 2004) and mapping onto the specific behavioral patterns in psychopathy have remained
Such correlations with specific genetic, neurobiological, cognitive, and affective
processes suggest the possibility that specific neurobiological deficiencies may be necessary
antecedents of psychopathy. Although dimensional models may be proposed to account for such
causes (e.g. Benning, Patrick, Blonigen, Hicks, & Iacono, 2005; Miller, Lynam, Widiger, &
Leukefeld, 2001), a taxonic distribution is a viable theoretical possibility that must be
investigated. Resolution of this issue of the latent structure of psychopathy is not only important
for developing theoretical models of the etiology and course of psychopathy, but it also has
critical implications for determining optimal investigative strategies and for specifying the ideal
psychometric qualities of scales constructed both for clinical and dispositional decision-making
(Krueger, 1999; Meehl, 1992; Ruscio & Ruscio, 2004a). Considering the prominence of PCL-R
ratings in risk assessment instruments (e.g., Quinsey, Harris, Rice, & Cormier, 1998), and its

Latent structure of psychopathy
widespread use in legal dispositional decisions (Edens & Petrila, 2006), such scaling issues have
substantial practical import. Consequently, a number of studies have addressed the problem of
psychopathy’s latent distribution (Guay & Knight, 2003; Harris, Rice, & Quinsey, 1994; Marcus,
John, & Edens, 2004; Skilling, Quinsey, & Craig, 2001; Vasey, Kotov, Frick, & Loney, 2005),
but with mixed results. All of these prior studies unfortunately have suffered from sampling and
methodological weaknesses that have limited their ability to provide definitive answers to this
The two studies that found no evidence for taxonic latent structures both analyzed self-
report data. Guay and Knight (2003) conducted a taxometric investigation of the components of
psychopathy using the Multidimensional Assessment of Sex and Aggression (the MASA; Knight
& Cerce, 1999). The MASA is a self-report inventory that covers multiple domains (childhood
experiences, family and social relationships, school and work experiences, alcohol and drug use,
and sexual and aggressive behavior and fantasies) and that was developed to supplement archival
records. Participants were 330 sex offenders, 155 generic criminals, and 93 community controls,
who had been tested on paper-and-pencil and computerized versions of the MASA. A total of
eleven scales measuring conning and superficial charm, emotional detachment and behavioral
problems, and impulse control were generated using factor analysis in combination with a Rasch
model. The various taxometric techniques (MAMBAC, MAXCOV, MAXEIG) that were
assessed all generated results consistent with a dimensional structure, with no evidence of
taxonicity. Marcus, John, and Edens (2004) evaluated a sample of 309 incarcerated offenders
(51.6% African American, 37.5% Caucasian, 8% Hispanic, 2.9% “other,” and 91.3% male) using
the Psychopathic Personality Inventory (PPI: Lilienfeld & Andrews, 1996). The authors
performed MAMBAC, MAXEIG and L-Mode analyses, and their results showed no evidence of

Latent structure of psychopathy
taxonicity. Both studies may have compromised their ability to identify a taxon, because of the
potential for increased nuisance covariation in self-report data that is produced by response styles
and biases. Such method variance could have artifactually increased the correlation of scales
within the taxon and the complement and thereby reduced the potential to identify a taxon.
Harris, Rice, and Quinsey (1994) analyzed data collected from 653 mentally disordered
participants from a maximum-security institution, who had been adjudicated not guilty by reason
of insanity. Although the authors argued that their results provided evidence for a taxon, several
methodological ambiguities of their study undermine their conclusion. First, their sample
comprised a select, potentially biased groups of offenders (Marcus et al., 2004). Second, with the
exception of their application of Meehl’s taxometric analyses, most of the statistical procedures
used by Harris et al. (1994) lacked empirical support as methods for distinguishing between
taxonic and dimensional structures. The validity of their iterative methods approach has not been
tested by any Monte Carlo studies, and the distributions of Bayesian probabilities can readily be
U-shaped even when the latent structure is dimensional, especially when a large number of items
are used as in Harris et al.’ study. Third, their exclusive use of file reviews introduced severe
limitations. The data for accurately rating Factor 1 items in archival files is often missing (Hare,
2003). Moreover, insufficient file information can lead raters to score items for which data are
missing or inadequate from information on related items (Alpert, Shaw, Pouget, & Lim, 2002),
thereby decreasing item covariation within the putative taxon, increasing item correlation across
the taxon and complement, and increasing the potential to find a pseudo-taxon. File reports often
focus on salient, egregious Factor 2 antisocial behaviors that may make ratings more vulnerable
to raters’ a priori taxonic (Beauchaine & Waters, 2003) or item contingency (Bolt, Hare, Vitale,
& Newman, 2004) biases, and thereby may increase the probability of pseudo-taxonicity. In this

Latent structure of psychopathy
regard, it is interesting that Harris et al., (1994) only found evidence for a taxon in their Factor 2
analyses. Fourth, the authors used an admixed sample in which they employed some matching
procedures. The alleged taxon might have been an artifact of their sampling strategy. This
possibility could have been avoided by re-running their analyses within samples, or they could
have presented a contingency table of sample membership and taxon/complement assignment.
Fifth, Harris et al. (1994) did not have access to the simulation programs used in the present
study, which generate taxonic and dimensional comparison data to analyze as an interpretive aid
Without this, they did not notice the limited range and low values of the covariances that their
MAXCOV analyses generated. Consequently, they may not have scaled the ordinate of their
graphs appropriately for the interpretation of the curves calculated from their data. A more
extended ordinate makes their apparently peaked curve look flat and dimensional.
Later, Skilling, Quinsey, and Craig (2001) replicated Harris et al.’s (1994) results, using a
similar methodology, but this time analyzing a sample of 1,111 boys. At first sight, their results
appear to support the taxonic structure of psychopathy, but once again methodological problems
plague this study. First, the authors unnecessarily dichotomized their items for MAXCOV, which
weakens this procedure (Ruscio, 2000). Second, the authors used the goodness of fit index (GFI)
to determine the nature of the latent structure. In the two studies that have examined the GFI
systematically (Cleland, Rothschild, & Haslam, 2000; Haslam & Cleland, 2002), it has been
shown to discriminate poorly between taxonic and dimensional structure. Moreover, because
both taxonic and dimensional structures can yield GFIs well above .90, no universally applicable
threshold has emerged even for data that GFI handles well. Examining the consistency of base
rate estimates has intuitive appeal and is widely recommended and practiced, but nobody has
ever actually established that taxonic structure does in fact yield more consistent estimates across

Latent structure of psychopathy
a realistic range of data parameters. A recent factorial Monte Carlo study (Ruscio, 2006) found
that MAMBAC, MAXCOV, and MAXEIG analyses seldom yielded lower SDs for taxonic
structure than for dimensional structure. More often than not, they were basically the same.
Ambiguous MAMBAC and MAXCOV results may be produced by a positive skew of indicators
or by low endorsement rates in the case of binary indicators. Rising MAMBAC curves and
apparent (but low) peaks toward the right side of MAXCOV curves, highly consistent base rate
estimates, and a high GFI are all as consistent with a latent dimension whose indicators are
positively skewed (low endorsement), as they are with a small taxon.
Skilling, Harris, Rice, and Quinsey (2002) used the same participants as the 1994 study to
investigate taxonic structure of APD, PCL-R, and the Child and Adolescent Taxon Scale
(CATS), a derived instrument based on 8 items associated with the taxon in the Harris et al.
(1994) paper. The authors performed MAMBAC and MAXCOV analyses and concluded that
there was evidence supporting a taxon both for APD and for CATS. They failed to report the
amplitude of the mean differences in MAMBAC and the covariance scores in MAXCOV. Other
problems identified in the Harris et al. (1994) paper, such as possible rater file-review and graph
comparison biases, absence of taxon base rate estimates, the potentially misleading admixed
sample, and limited ordinate values also apply to the Skilling et al. (2002) study.
Recenltly, Vasey, Kotov, Frick, and Loney (2005) studied a sample of 386 children and
adolescents to test the latent structure of psychopathy. The authors used two versions (Parent and
Student) of the Antisocial Process Screening Device (APSD; Frick & Hare, 2001) to assess
psychopathic characteristics. Along with evidence of a taxon for broad antisocial behavior, the
authors claimed that they had found evidence for a psychopathy taxon. Specifically, using
MAXEIG on the five subscales of youth self-report and parent APSD, the results produced