Journal of Abnormal Psychology © 2017 American Psychological Association

2018, Vol. 127, No. 1, 51– 65 0021-843X/18/$12.00 http://dx.doi.org/10.1037/abn0000315

What Features of Psychopathy Might Be Central? A Network Analysis of

the Psychopathy Checklist-Revised (PCL-R) in Three Large Samples

Bruno Verschuere, Sophia van Ghesel Grothe, Ashley L. Watts and Scott O. Lilienfeld
and Lourens Waldorp Emory University
University of Amsterdam

John F. Edens Jennifer L. Skeem

Texas A&M University University of California, Berkeley
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Arjen Noordhof
University of Amsterdam

Despite a wealth of research, the core features of psychopathy remain hotly debated. Using network
analysis, an innovative and increasingly popular statistical tool, the authors mapped the network structure
of psychopathy, as operationalized by the Psychopathy Checklist—Revised (PCL-R; Hare, 2003) in two
large U.S. offender samples (nNIMH ⫽ 1559; nWisconsin ⫽ 3954), and 1 large Dutch forensic psychiatric
sample (nTBS ⫽ 1937). Centrality indices were highly stable within each sample, and indicated that
callousness/lack of empathy was the most central PCL-R item in the 2 U.S. samples, which aligns with
classic clinical descriptions and prototypicality studies of psychopathy. The similarities across the U.S.
samples offer some support regarding generalizability, but there were also striking differences between
the U.S. samples and the Dutch sample, wherein the latter callousnesss/lack of empathy was also fairly
central but irresponsibility and parasitic lifestyle were even more central. The findings raise the important
possibility that network-structures do not only reflect the structure of the constructs under study, but also
the sample from which the data derive. The results further raise the possibility of cross-cultural
differences in the phenotypic structure of psychopathy, PCL-R measurement variance, or both. Network
analyses may help elucidate the core characteristics of psychopathological constructs, including psy-
chopathy, as well as provide a new tool for assessing measurement invariance across cultures.

General Scientific Summary

What is psychopathy? Network analyses in three large samples (totaling 7,450 offenders) indicate
that affective-interpersonal features, most notably callousness and lack of empathy, may be central
to psychopathy.

Keywords: psychopathy, Psychopathy Checklist—Revised (PCL-R), network analysis, empathy,

cross-cultural

Research on psychopathy is thriving. Despite the accumula- tures are integral, irrelevant, or peripheral to psychopathy (e.g.,
tion of knowledge on psychopathy, there remains active debate Miller & Lynam, 2012; Lilienfeld et al., 2012; Skeem & Cooke,
on its origins, assessment, and clinical and legal implications 2010). Indeed, an essential question remains: What is psychop-
(DeMatteo et al., 2014; Skeem, Polaschek, Patrick, & Lilien- athy?
feld, 2011). At a more basic level, much disagreement sur- Psychopathy has often been conceptualized in terms of a con-
rounds the mere definition of psychopathy, such as what fea- stellation of traits. Following the structure of the widely used

This article was published Online First November 27, 2017. We are very grateful to Joseph Newman for sharing the data of the
Bruno Verschuere and Sophia van Ghesel Grothe, Department of Clinical Wisconsin sample. We thank the Expertisecentrum Forensische Psychiatrie
Psychology, University of Amsterdam; Lourens Waldorp, Department of Psycho- for making the Dutch data available. We thank Lena Christ and Miguel
logical Methods, University of Amsterdam; Ashley L. Watts and Scott O. Lilien- Beira for the initial analyses.
feld, Department of Psychology, Emory University; John F. Edens, Department of Correspondence concerning this article should be addressed to Bruno
Psychology, Texas A&M University; Jennifer L. Skeem, School of Social Wel- Verschuere, Department of Clinical Psychology, University of Amsterdam,
fare, Goldman School of Public Policy, University of California, Berkeley; Arjen Postbus 15933, 1001 NK Amsterdam, the Netherlands. E-mail: B.J.Vers
Noordhof, Department of Clinical Psychology, University of Amsterdam. [email protected]

51
 52 VERSCHUERE ET AL.

Psychopathy Checklist—Revised (PCL-R; Hare, 2003), these questions the value of impulsivity—at least when viewed or measured
traits can be situated within affective (e.g., fearlessness, shallow as a unitary construct—as a marker for the assessment of psychopa-
affect, callousness/lack of empathy), interpersonal (e.g., detached, thy. The evidence reviewed here suggests that widely held and long-
standing belief that ⬍psychopaths are impulsive⬎ must be reconsid-
manipulative, pathological lying), lifestyle (e.g., irresponsible,
ered. (p. 132)
poor behavioral control), and antisocial domains (e.g., early be-
havioral problems, criminal versatility). Nevertheless, there is no In sum, there is little consensus on the core characteristics of
consensus on which features should be included in the definition of psychopathy (Lilienfeld, Watts, Francis Smith, Berg, & Latzman,
psychopathy, let alone how important they are to this condition 2015).
(see Patrick, Fowles, & Krueger, 2009, for a review). To illustrate, Network analyses, a relatively novel set of statistical techniques,
we highlight three points of discussion.1 First, the question of may help to identify the core characteristics of psychological
whether overt criminal behavior or antisocial behavior more constructs (Borsboom & Cramer, 2013). The network approach
broadly is intrinsic to psychopathy is controversial; this possibility was introduced as an alternative view on covariance between
is suggested by the inclusion of explicit criminality among the phenomena that departs from the largely untested and often faulty
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

PCL-R items. Some authors contend that criminality and antisocial or overly stringent assumptions that underlie factor analytic mod-
This document is copyrighted by the American Psychological Association or one of its allied publishers.

behavior, rather than being intrinsic to psychopathy, are merely els. The first, ontological, assumption is that if certain phenomena
downstream consequences of psychopathy as conceptualized in covary there necessarily exists an underlying latent factor (or class,
terms of affective and interpersonal traits (e.g., Hare & Neumann, or disorder) that explains the covariance. Although a latent factor
2010; Skeem & Cooke, 2010). may exist, the a priori assumption that there must be one is
Second, although many psychopathy researchers consider affec- logically and empirically unfounded. The second, methodological,
tive features to lie at the heart of psychopathy, there is no clear assumption in all factor analytic models (as well as in item re-
consensus on what constitutes this deficient affective experience. sponse theory) is local independence, meaning that phenomena
(i.e., symptoms or items) are not directly related. Hence, their
Some have argued for a broad, pervasive emotional deficit: “Like
covariance ostensibly approaches zero after taking into account
the color-blind person, the psychopath lacks an important element
their common latent factor(s). This assumption is implausible for
of experience—in this case, emotional experience” (Hare, 1993;
many psychopathological syndromes. For instance, depressive
pp. 129; see also Cleckley, 1941/1976). Others have offered alter-
symptoms may involve a negative feedback loop in which symp-
natives to this supposition. For instance, Lykken (1995) posited a
toms like lack of sleep contribute to other depressive symptoms,
narrower fearlessness deficit and argued that psychopathic indi- such as trouble concentrating and anhedonia. This may well also
viduals would not engage in problematic behaviors if they were apply to psychopathy, for instance, early behavior problems may
incapable of experiencing positive affect, whereas McCord and well explain juvenile delinquency. And it is also very conceivable
McCord (1964) contended that the lack of anxiety and emotional that a lack of remorse or guilt explains a failure to accept respon-
coldness in psychopathy arises from profound lovelessness, as- sibility. The third assumption is that correlations between scales or
signing a prominent role to positive emotions. Blair (2005) and syndromes (or with, e.g., genetic markers or brain-structures)
others have argued that psychopathy stems from a broader social- result from a direct relation between latent variables. These cor-
emotional deficit, namely the capacity to experience empathy. relations may result as well or instead from more fine-grained
Research pinpointing specific emotional deficits among psycho- bidirectional interactions at the level of signs and symptoms. Thus,
pathic individuals has been decidedly mixed. An early startle the disadvantage of adhering to factor analytic models is that these
eyeblink study (Patrick, Bradley, & Lang, 1993) found that psy- possibilities are not taken into account.
chopathic individuals showed deficient emotional responding to With network analyses, one plots the features (e.g., symptoms)
negative (e.g., fear-eliciting) stimuli, but normal emotional re- of the construct as nodes and their associations as edges within a
sponding to positive (e.g., erotic) stimuli. Physiological hypore- network. The shortest path length between two nodes is the min-
sponsivity in psychopathy is also most pronounced for aversive imum number of edges required to travel from one node to another
stimuli (Lorber, 2004). In contrast, a meta-analysis of emotion (if B can only be reached from A through C thus A ⬎ C ⬎ B, then
recognition tasks found that psychopathy was associated with the shortest path length between A and B is 2). The stronger the
poorer recognition not only of fear and sadness, but also of association between two nodes, the larger their edge weight (dis-
happiness and surprise (Dawel, O’Kearney, McKone, & Palermo, played in the plot by wider, more saturated edges). The strength of
2012). Owing to inconsistent research findings, a recent review the associations as well as the position of the features in the
contrasting a general emotional blunting with specific emotional network provide information about the importance of the feature to
deficits concluded that the “overall pattern of findings is not the network. A central, well-connected feature is likely to be
clearly consistent with any of the dominant theoretical perspec- particularly important. Importantly, network-analyses go beyond
eye-balling the interitem correlation matrix (or factor loadings) by
tives of emotion processing in psychopathy” (Brook, Brieman, &
providing a growing set of indices that can be used to assess item
Kosson, 2013, p. 979).
Third, although prominent authors view impulsivity as “one of
the hallmarks of psychopathy” (Hare, 2003, p. 139) and “a cardinal 1
In the present study, we can only assess the importance of symptoms
feature of the [psychopathy] construct” (Hart & Dempster, 1997, p. tapped by the PCL-R. Therefore, we do not go into the discussion on features
not directly covered by the PCL-R such as boldness or fearless dominance
212), Cleckley (1941/1976) argued that psychopathic individuals
(Patrick et al., 2009; Vize et al., 2016). Also, we focus on features at the
are not driven by powerful impulses. Furthermore, the review by descriptive level (symptoms) rather than cognitive, physiological, or neural
Poythress and Hall (2011) features.
 CENTRALITY OF PCL-R PSYCHOPATHY FEATURES 53

covariance. The centrality of a feature to the network can be psychiatric sample of violent mentally disordered offenders (n ⫽
assessed in several ways (Costantini et al., 2015).2 Strength re- 1,962; hereafter called the ‘TBS sample); and a North American
flects the overall connectivity with other features by summing the sample of offenders ordered to correctional institutes in Wisconsin
(absolute values of) weights of the feature’s associations to other (n ⫽ 3,963; hereafter called the Wisconsin sample), and we
features. Closeness reflects the distance of a feature to other assessed the similarities by calculating the Spearman rho correla-
features by computing the mean weighted shortest path lengths to tion between the rank orders of the centrality indices between the
all other features. samples.
Network analyses recently have been applied to several mental
disorders, including major depressive disorder (Bringmann, Lem-
Method
mens, Huibers, Borsboom, & Tuerlinckx, 2015), posttraumatic
stress disorder (McNally, Robinaugh, Wu, Wang, Deserno, & This study was approved by the ethical committee of the Psy-
Borsboom, 2015), and substance use disorder (Rhemtulla, Fried, chology Department of the University of Amsterdam (2017-CP-
Aggen, Tuerlinckx, Kendler, & Borsboom, 2016). Applying net- 7957).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

work analysis to DSM–IV symptoms of substance abuse in adult

This document is copyrighted by the American Psychological Association or one of its allied publishers.

twins, for instance, has indicated that using a drug more than
planned was among the most central substance abuse indicators Sample Characteristics
(Rhemtulla et al., 2016). Network analyses similarly seem to be a Exclusion criteria. We excluded data from participants with
promising tool for identifying and clarifying the core features of substantial missing data on the PCL-R (i.e., more than two items
psychopathy and for helping to inform longstanding debates in the missing on either of the two original two factors or more than five
field. items missing in total); see the Samples section. For the included
At the same time, it is still unclear to what extent findings from participants, the proportion of missing values for the PCL-R items
network analyses are stable (within a sample, design, or instru- averaged 0.5% (range: 0% to 4%) for the NIMH sample, 1.50%
ment) and replicable (across samples, designs, or instruments). (range: 0% to 6%) for the TBS sample, and 1.66% (range: 0% to
Network analyses on mental disorders have been based largely on 13%) for the Wisconsin sample. Therefore, although the propor-
single samples (Fried & Cramer, 2017). Moreover, network anal- tion of missing data was overall very low, there were substantial
yses on different instruments, designs or samples have not always missing data for some items in some samples. For the calculation
converged on the same outcome. For example, cross-sectional of the PCL-R total score, data from participants with one to four
analyses of Inventory of Depressive Symptomatology items in missing items were prorated following the guidelines of the PCL-R
3,463 currently depressed individuals indicated that energy loss manual (Table 1; Hare, 2003). For the network analyses, correla-
and sadness were among the most central depression symptoms tions were calculated for all included participants based upon
(Fried, Epskamp, Nesse, Tuerlinckx, & Borsboom, 2016). In con- pairwise complete observations. Analyses on the three subsamples
trast, network analyses on repeated administrations of the Beck with only participants without any missing values produced similar
Depression Inventory in 182 currently depressed individuals dur- findings to those with missing values (see Appendix I; all appen-
ing the course of treatment indicated that past failure and loss of dices are on https://osf.io/4habq/). In case of multiple PCL-R
pleasure were the most central depression symptoms (Bringmann assessments for each participant, only the most recent assessment
et al., 2015). Thus, the stability and replicability of network available was included.
analyses of mental disorders is unclear and also a point of current Samples. The National Institute of Mental Health (NIMH)
controversy (see Forbes, Wright, Markon, & Krueger, 2017; but sample included offenders (n ⫽ 1,661; 82.5% male; Mage ⫽ 31.0
see reply by Borsboom et al., in press). years, SD ⫽ 6.6) ordered to either prison or substance abuse
In the present study, we apply network analyses to the PCL-R, treatment in Florida, Nevada, Oregon, Utah, and Texas. The
which is the most extensively researched measure of psychopathy PCL-R was administered by trained research assistants for re-
(Hare, 2003). Our aim is to begin to identify the core characteris- search purposes. The recruitment strategy favored men (80%).
tics of PCL-R operationalized psychopathy. The construct of psy- Exclusion criteria included (a) estimated IQ below 70, (b) non-
chopathy, like any construct (Cronbach & Meehl, 1955), cannot be English-speaking, and (c) residing in a prison mental health unit or
reduced to a single instrument or ostensible “gold standard” receiving medication for active symptoms of psychosis (see Poyth-
(Skeem & Cooke, 2010)— because that instrument will almost ress et al., 2010, for further details regarding participant recruit-
inevitably omit some essential features of the construct (construct ment). We excluded 102 participants because of excessive missing
underrepresentation) and include others that are peripheral (con- PCL-R items (see the Exclusion Criteria section). The final NIMH
struct overrepresentation). Nevertheless, the PCL-R is a well- sample consisted of 1559 offenders (83.6% male; Mage ⫽ 31.0
validated and reasonably comprehensive measure that represents years, SD ⫽ 6.5).
an excellent starting point for network analyses. To ascertain the The TBS sample (n ⫽ 1,962; 73.0% male; Mage ⫽ 38.8 years,
stability of our PCL-R network analyses within the sample, we SD ⫽ 10.0) consisted of violent mentally disordered offenders
obtained large samples so we could determine whether indicators under mandatory inpatient treatment in the Netherlands (TBS or
of centrality were robust across random subsets of persons within
the same sample (Epskamp, Borsboom, & Fried, 2016). Further, to
2
ascertain the replicability of PCL-R network results, we examined Another often used index is betweenness, which indicates how often a
feature functions as a hub in the network, by counting how often the feature
three samples: a North American sample of offenders ordered to lies on the shortest connection of two other features. Because we found this
either prison or substance abuse treatment (n ⫽ 1661; hereafter index to be of less relevance to the PCL-R network, and there were indications
called the NIMH sample; Poythress et al., 2010); a Dutch forensic that this index was less stable, we focus on strength and closeness.
 54 VERSCHUERE ET AL.

Table 1
PCL-R Items With Their Abbreviation as Used in the Network Plots

Item PCL-R facet

number Abbreviation Item label (Hare, 2003)

1 GLI Glibness/superficial charm Interpersonal

2 SEL Grandiose sense of self-worth Interpersonal
3 STI Need for stimulation Lifestyle
4 LIE Pathological lying Interpersonal
5 CON Conning/manipulative Interpersonal
6 GUI Lack of remorse or guilt Affect
7 AFF Shallow affect Affect
8 EMP Callous/lack of empathy Affect
9 PAR Parasitic lifestyle Lifestyle
10 BEV Poor behavioral controls Antisocial
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

11 SEX Promiscuous sexual behavior —

This document is copyrighted by the American Psychological Association or one of its allied publishers.

12 PRO Early behavioral problems Antisocial

13 GOA Lack of realistic, long-term goals Lifestyle
14 IMP Impulsivity Lifestyle
15 IRR Irresponsibility Lifestyle
16 RES Failure to accept responsibility Affect
17 SHO Many short-term marital —
relations
18 DEL Juvenile delinquency Antisocial
19 REL Revocation of conditional release Antisocial
20 CRI Criminal versatility Antisocial

ter beschikking stelling, which can be translated to disposal to be 22.54 (SD ⫽ 7.47) for the NIMH sample, 23.3 (SD ⫽ 7.0) for the
treated on behalf of the state; Philipse, 2005). TBS is imposed by Wisconsin sample, and 20.9 (SD ⫽ 7.3) for the TBS sample. A
court on high-risk offenders who have committed violent crimes cutoff score of 26 (Europe) or 30 (United States) has been used for
that were determined to result from psychopathology, leading to a clinical psychopathy diagnosis. The proportion of the sample
judgments of diminished responsibility. The duration of the TBS- scoring above the country-specific cutoff (Hare, Clark, Grann, &
order is indeterminate—lasting as long as the offender is consid- Thornton, 2000) was 28.03% (TBS), 21.93% (Wisconsin), and
ered to be high-risk— but averages about 8 years. The PCL-R was 19.69% (NIMH). Cronbach’s alpha was high for the total score in
administered by trained clinicians as part of a mandatory risk all three samples: ␣ ⫽ .82 for the NIMH sample, ␣ ⫽ .81 for
assessment test battery. We obtained PCL-R data from 12 of the 14 Wisconsin sample, and ␣ ⫽ .83 for the TBS sample. In the NIMH
treatment facilities that collect mandatory PCL-R data (http://www sample, the interrater reliability (ICC) for the total PCL-R score
.efp.nl/projecten/ldr-tbs). We excluded 25 participants because of based upon n ⫽ 51 was .88 (Poythress et al., 2010). The ICC of
excessive missing PCL-R items (see “Exclusion criteria”). The PCL-R total scores by diagnostic staff in Dutch TBS clinics based
final TBS sample consisted of n ⫽ 1,937 offenders (73.4% male; upon n ⫽ 16 was found to be .76 (Nentjes, Bernstein, Meijer,
Mage ⫽ 38.8 years, SD ⫽ 10.0). Arntz, & Wiers, 2016). The ICC in the Wisconsin sample was very
The Wisconsin sample (n ⫽ 3,963; 63.3% male; Mage ⫽ 30.3 high: .99 based upon n ⫽ 16 (Baskin-Sommers & Newman, 2014),
years, SD ⫽ 7.0) consisted of offenders from state prisons in the and .96 based upon n ⫽ 101 (Newman et al., 2005).
U.S. state of Wisconsin. The sample has been collected over many Table 1 shows the facet labels corresponding to the PCL-R
years, with the PCL-R being scored by trained research assistants 4-facet structure (Hare, 2003). There is still disagreement regard-
(e.g., Baskin-Sommers & Newman, 2014; Newman, MacCoon, ing the optimal factor structure of the PCL-R (see Footnote 3).
Vaughan, & Sadeh, 2005). During recruitment, participants with
Early factor analytic work on the measure revealed a two-factor
low IQ scores (⬍70) and with significant mental illness (particu-
structure, with Factor 1 encompassing affective-interpersonal
larly psychosis) were excluded. We excluded nine participants
features (selfish, callous, remorseless use of others: Items 1, 2,
because of excessive missing PCL-R items (see the Exclusion
4, 5, 6, 7, 8, and 16) and Factor 2 encompassing chronic
Criteria section). The final Wisconsin sample consisted of n ⫽
antisocial lifestyle (chronically unstable and antisocial lifestyle:
3,954 offenders (63.3% male; Mage ⫽ 30.3 years, SD ⫽ 7.0).
3, 9, 10, 12, 13, 14, 15, 18, and 19; Hare, 1991). Cooke and
Michie (2001) later proposed a three-factor structure that es-
Measures sentially (a) splits Factor 1 into an affective factor (deficient
The PCL-R (Hare, 2003) consists of a checklist of 20 items (see affective experience; Items 6, 7, 8, and 16) and an interpersonal
Table 1). In the TBS sample, the authorized Dutch translation was factor (deceitful interpersonal style; Items 1, 2, 4, and 5), and
used (Vertommen, Verheul, de Ruiter, & Hildebrand, 2002). Based (b) excludes overt criminal items from Factor 2 (now labeled
upon an extensive interview and collateral file information, trained the lifestyle factor; impulsive and irresponsible behavioral
raters score each item as 0 ⫽ absent, 1 ⫽ maybe or partly present, style; Items 3, 9, 13, 14, and 15). Finally, like the three-factor
or 2 ⫽ definitely present. Thus, the PCL-R sum score lies on a structure, the four-factor structure (Hare, 2003), also has iden-
continuum ranging from 0 to 40. The mean PCL-R score was tical affective, interpersonal, and lifestyle factors, but reintro-
 CENTRALITY OF PCL-R PSYCHOPATHY FEATURES 55

duces the items primarily focused on prior criminal activities (2016) recommended that CS (cor ⫽ 0.70) should be at least 0.25 and
into an antisocial factor (Items 10, 12, 18, 19, and 20). preferably greater than 0.50 to interpret centrality differences.
Centrality differences. The network plots and centrality in-
dices indicate the centrality of the PCL-R items. We further used
Statistical Analyses bootnet to conduct bootstrapped difference tests on the centrality
Our network analyses were based upon polychoric correlations, indices to examine whether there are significant differences in
which allow for estimation of two observed ordinal variables that centrality. Although (a) this approach yields numerous compari-
have presumed theoretical normal distributions, as to account for sons and (b) there is presently no definitive solution for multiple
the limited range in PCL-R item responses (response options: 0, 1, testing in network analyses (Epskamp et al., 2016), these differ-
and 2). We elected not to examine the regularized partial correla- ence tests provide additional information on differences in cen-
tion network (glasso). The glasso method downsizes indirect rela- trality of the PCL-R items.
tions among PCL-R items (e.g., the correlation between two Replicability across samples. Rather than examining the cor-
PCL-R items disappears if a third PCL-R item that strongly con- respondence of the absolute centrality positions of the PCL-R
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

items across the samples (Forbes et al., 2017), we examined

nects to those PCL-R items is taken into account), often with the
This document is copyrighted by the American Psychological Association or one of its allied publishers.

whether there was correspondence in their relative centrality po-

aim of finding unique possible casual relations. Our focus was on
sitioning (Borsboom et al., in press), because absolute positioning
centrality, irrespective of whether centrality would be a result of
is a too strict measure of replicability. Whether a particular PCL-R
direct or of indirect relations. Moreover, given the covariance
item is ranked 17th, 18th, or 19th may not matter that much, but if
between PCL-R items, the meaning of each PCL-R item typically
this would be the observed centrality rank across three samples, it
becomes unclear once the variance shared with all other PCL-R
would be clear that the PCL-R item is replicably low in centrality.
items has been controlled (Lynam, Hoyle, & Newman, 2006).
To assess replicability across samples, we calculated the Spearman
Although partialing can rule out spurious connections, it adapts an
rho correlation of rank-ordered centrality between samples.
extreme form of control that may in fact create more inferential
problems than the one it attempts to solve. Indeed, Appendix II
shows that partialing led to several, implausible and unexpected, Results
negative associations between partialed PCL-R items, indicating
that partialing created spurious connections due to common ef- Network Plots
fects.
Figure 1 displays the correlational structure of the PCL-R items
Network plots. The PCL-R network was constructed with the
in the TBS sample (left, Figure 1a), the NIMH sample (middle,
qgraph package (Epskamp, Cramer, Waldorp, Schmittmann, &
Figure 1b), and the Wisconsin sample (right, Figure 1c). The
Borsboom, 2012) using the statistical application R (version 3.2.4;
strength of the relations between PCL-R items translates into the
R Core Team, 2016). In visualization of the network, nodes rep-
thickness of the edges between them and the distance that they are
resent the PCL-R items and the thickness and the color of edges
plotted from each other. Glibness/superficial charm (GLI) and
represents their association strength and valence (red: negative
grandiose sense of self-worth (SEL), for instance, show strong
association, black: positive association), respectively. The layout
interrelations in all three samples. Groups of PCL-R items cluster
of the network was based on Fruchterman and Reingold (1991)’s
together in the densely connected network, except for many short-
algorithm, which places more influential nodes central to the
term marital relationships (SHO), which appears peripheral to the
network and stronger connected nodes closer together in the net-
PCL-R psychopathy network in all three samples.
work.
Centrality indices. In addition to network visualization, we
used the qgraph package to provide two indices of centrality: The Centrality Indices
overall connectivity with other nodes (strength) as well as the Figure 2A–C depicts the closeness and strength of the 20 PCL-R
average shortest path lengths to all other nodes (closeness) (Op- items for each of the three samples. In the NIMH sample, callous/lack
sahl, Agneessens, & Skvoretz, 2010; Epskamp et al., 2016).3 of empathy (EMP) was the most central item, followed by shallow
Within-sample stability. We assessed the stability of the cen- affect (AFF). SHO and revocation of conditional release (REL) were
trality indices. Stability in the network perspective refers to resistance
to change: Is the network robust to the dropping of selected partici- 3
pants from the analyses? Stability across participants indicates that a Our network analyses do not serve nor allow to test the best fit of the
two-, three-, or four-factor model. Descriptively though, a number of
somewhat different sample would not change the rank order of the observations are of interest to the factor analytic work. First, the network
centrality of the nodes. The stability of the centrality order can be plots show that the clustering of items generally fit with the two-, three-, or
determined with subset bootstrapping using the R-package entitled four-factor models. Second, a noteworthy exception is Lack of realistic
bootnet (Epskamp, 2015), providing the centrality of items over a long term goals, which does not cluster with the other items of the factor
it is expected to load on (which holds for both the two-, three-, or
wide range of sampled participants. Moreover, stability can be quan- four-factor model) and appeared rather peripheral to the PCL-R network in
tified by the correlation stability coefficient or conditional stimulus all three samples. Third, while Criminal versatility was not included in the
(CS)-coefficient. Specifically, CS (cor ⫽ 0.70) reflects the proportion original Factor2, it clusters with the other antisocial items in the Wisconsin
of participants that can be dropped to retain with 95% probability that sample and with the antisocial and lifestyle items in the Dutch sample.
Fourth, the clustering of items according the two-, three-, or four-factor
the correlation between the centrality based on the entire sample and analytic solution was least clear in the NIMH sample (where there ap-
that of the bootstrapped subsamples is at least 0.70 (representing a peared a clear affective-interpersonal clustering of items, but such cluster-
very large effect). Based upon simulation studies, Epskamp et al. ing was less apparent for the behavioral-lifestyle items).
 56 VERSCHUERE ET AL.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 1. a– c: Network of the Psychopathy Checklist—Revised (PCL-R; Hare, 2003) based upon polychoric
correlations between PCL-R items for the three samples (1a: Dutch sample; 1b: NIMH sample; 1c: Wisconsin
sample). Nodes represent the PCL-R items: GLI ⫽ glibness/superficial charm; SEL ⫽ grandiose sense of self-worth:
STI ⫽ need for stimulation: LIE ⫽ pathological lying: CON ⫽ conning/manipulative: GUI ⫽ lack of remorse or guilt:
AFF ⫽ shallow affect: EMP ⫽ callous/lack of empathy: PAR ⫽ parasitic lifestyle: BEV ⫽ poor behavioral controls:
SEX ⫽ promiscuous sexual behavior: PRO ⫽ early behavioral problems: GOA ⫽ lack of realistic: long-term goals:
IMP ⫽ impulsivity: IRR ⫽ irresponsibility: RES ⫽ failure to accept responsibility: SHO ⫽ many short-term marital
relations: DEL ⫽ juvenile delinquency: REL ⫽ revocation of conditional release: CRI ⫽ criminal versatility. Positive
correlations plotted in green (online figure) / grey (print figure). See the online article for the color version of this figure.

the least central items in the NIMH sample. In the Wisconsin sample, Within-Sample Stability
EMP was clearly the most central item. SHO, REL, and GOA were
the least central in the Wisconsin sample. In contrast, in the Dutch Figure 3A–C depict the stability of the centrality indices, by
sample, parasitic lifestyle (PAR) and irresponsibility (IRR) were the most plotting centrality over increasingly smaller bootstrapped sub-
central items. promiscuous sexual behavior (SEX), SHO, and—perhaps samples for each of the three samples. Closeness and strength
surprisingly—AFF, were the least central items in the Dutch sample. display high stability in all three samples for a wide range of

Figure 2. Centrality closeness and strengths of the Psychopathy Checklist—Revised items in the three samples
(2a: Dutch sample; 2b: NIMH sample; 2c: Wisconsin sample).
 CENTRALITY OF PCL-R PSYCHOPATHY FEATURES 57
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 3. A–C: Bootstrapped differences in centrality strength and closeness in the three samples (3A: Dutch
sample; 3B: NIMH sample; 3C: Wisconsin sample). See the online article for the color version of this figure.

increasingly smaller bootstrapped subsamples. These impressions the PCL-R network. These observations are also apparent from the
were confirmed by the correlation stability coefficients. CS (cor ⫽ strong Spearman rho correlation between the rank orders of the
0.70) for strength was .89 for the Dutch sample, .91 for the NIMH NIMH and the Wisconsin sample for closeness, ␳ ⫽ .64, and
sample, and .92 for the Wisconsin sample. CS (cor ⫽ 0.70) for strength, ␳ ⫽ .60.
closeness was .87 for the Dutch sample, .89 for the NIMH sample, The centrality ranking of the PCL-R items of the U.S. samples
and .92 for the Wisconsin sample. These findings suggest that the differs in important respects from that of the TBS sample. Most
centrality indices were highly stable and well over the recom- notably, PAR and particularly IRR were most central in the TBS
mended value of .50, which allows for interpreting differences in sample whereas they showed modest to low centrality in the U.S.
centrality. samples. The Spearman rho correlation between the rank orders of
the TBS and the Wisconsin sample were moderate (for closeness,
Centrality Differences ␳ ⫽ .38, and strength, ␳ ⫽ .42), and weak for the TBS and the
NIMH sample (for closeness, ␳ ⫽ .11, and strength, ␳ ⫽ .15).
Figure 4A–C displays the results of the bootstrapped difference
tests in centrality for all three samples. Gray boxes indicate that the Discussion
PCL-R items do not significantly differ in centrality; black boxes
indicate that the PCL-R items differ significantly in centrality. The This study was an effort to shed light on an actively debated and
values in the white box on the diagonal depict the centrality values. theoretically important question: What are the most central fea-
We highlight notable significant differences in centrality. For the tures of psychopathy, at least as operationalized by the most
NIMH sample, the difference tests confirm that EMP and AFF widely used measure of this construct? Using network analyses on
were significantly more central than most of the other PCL-R three large samples, we examined the centrality of PCL-R features
items, and that REL and SHO were significantly less central than of psychopathy, as well as their stability within each sample, and
most of the other PCL-R items. For the Wisconsin sample, the the replicability across samples. We found a densely connected
difference tests show that CAL was significantly more central than network, with EMP being stably most central to the PCL-R net-
all other PCL-R items. They also confirm that SHO, REL, and work in the U.S. offender samples; SHO was stably peripheral to
GOA were significantly less central than most other PCL-R items the PCL-R network in all three samples. The network results did
in the Wisconsin sample. For the Dutch sample, the difference not generalize to the Dutch forensic psychiatric sample, wherein
tests confirm the high centrality of IRR, as it was significantly EMP was also fairly central but IRR and PAR were even more
more central than nearly all other PCL-R items. They also confirm central.
the low centrality of SHO, AFF, and SEX in that sample, as they
were significantly less central than many other PCL-R items. The Centrality of Callous/Lack of Empathy to PCL-R
Psychopathy
Replicability Across Samples
EMP was moderately central in the Dutch forensic psychiatric
The centrality rankings of the PCL-R items of the NIMH and the sample, and exhibited the highest centrality in the two U.S. of-
Wisconsin sample show important similarities. EMP was the most fenders samples. The importance of a lack of empathy fits with
central item in both samples, for both closeness and for strength, early clinical descriptions of psychopathy. For instance, Gough
and largely irrespective of the number of persons taken into the (1948) regarded a deficiency in the ability to mentally “take the
analyses. In both samples, SHO and REL were peripheral to place” of other individuals as the core feature of psychopathy, and
 58 VERSCHUERE ET AL.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 4. A–C: Bootstrapped differences in centrality closeness and strength for the three samples (4A: Dutch
sample; 4B: NIMH sample; 4C: Wisconsin sample). Gray boxes indicate nonsignificant differences, black boxes
indicate significant differences. Centrality closeness and strength values are plotted on the diagonal.
 This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Figure 4 (continued)
CENTRALITY OF PCL-R PSYCHOPATHY FEATURES
59
 This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

60

Figure 4 (continued)
VERSCHUERE ET AL.
 CENTRALITY OF PCL-R PSYCHOPATHY FEATURES 61

he premised his influential Socialization scale of the California interpersonal features, and callousness and deficient empathy, in
Psychological Inventory on this model. Later, McCord and Mc- particular, as central components of psychopathy.
Cord (1964), focusing more on affective than on cognitive empa-
thy, viewed a lack of social emotions—specifically lovelessness Affect, Criminality, and Impulsivity: Core
(inability to form deep attachments) and guiltlessness (absence of
Characteristics of Psychopathy?
remorse)—as the essence of psychopathy.
Our finding also dovetails with the work by Frick and col- Whatever their theoretical differences, virtually all psychopathy
leagues (e.g., Frick, Ray, Thornton, & Kahn, 2014), who view researchers agree that psychopathy is characterized by abnormal
callous-unemotional (CU) traits, including lack of empathy and emotional processing, but there is no consensus on the nature of
guilt, to be of critical importance in delineating youth psychopa- that affective deficiency (Brook et al., 2013). Our findings speak to
thy, as well with proponents of the triarchic model of psychopathy, the issue of whether the affective deficit is general or largely
who accord a central role to meanness, coldheartedness, or cognate specific. The PCL-R contains several items that relate to abnormal
constructs in the conceptualization of psychopathy (Berg, Hecht, affect, some of which suggest specific deficits (lack of remorse or
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Latzman, & Lilienfeld, 2015; Patrick et al., 2009). Likewise, guilt, EMP) and others of which reflect a more global deficit
This document is copyrighted by the American Psychological Association or one of its allied publishers.

DSM–5 (American Psychiatric Association, 2013) included a spec- (AFF). AFF was marked by high centrality only in the NIMH
ifier to define a subgroup of youth with conduct disorder who sample, and it showed low centrality in the two other samples. In
display limited prosocial emotions, including weak empathy and contrast, lack of remorse or guilt showed modest centrality in all
guilt. Furthermore, the central role of EMP accords with Blair’s three samples, and EMP showed modest to high centrality in all
(1995) Violence Inhibition Model, which proposes that a reduced three samples. Awaiting a more comprehensive analysis across the
emotional reaction to distress in others (i.e., less victim empathy) full spectrum of positive and negative emotions, our findings
predisposes to psychopathy-like traits. provisionally speak to a specific rather than a global affective
Interestingly, the results of these network analyses converge deficit in psychopathy, broadly corroborating psychometric find-
with other methods of evaluating the centrality of various features ings that psychopathy is characterized not by poverty in affect
of psychopathy. A number of recent studies have asked mental more broadly (cf. Cleckley, 1941/1976) but by deficits in social
health experts as well as laypersons in North America and Europe detachment more specifically (Lilienfeld & Andrews, 1996; Pat-
to judge items from the Comprehensive Assessment of Psycho- rick et al., 2009).
pathic Personality (CAPP; Cooke, Hart, Logan, & Michie, 2012) The question of whether antisocial and criminal behavior are
on prototypicality for psychopathy (Flórez et al., 2015; Hoff, central to psychopathy remains heavily contested (see Skeem &
Rypdal, Mykletun, & Cooke, 2012; Kreis, Cooke, Michie, Hoff, & Cooke, 2010; Hare & Neumann, 2010, for diverging viewpoints).
Logan, 2012; Sörman et al., 2014). The CAPP assesses 33 features, Although limited to the PCL-R, our network analyses sheds some
including features that closely correspond with PCL-R items (e.g., light on this issue. We found that juvenile delinquency and revo-
lacks emotional depth and lacks remorse), but also features that cation of conditional release (both indices of criminal behavior)
were not (fully) captured by the PCL-R (e.g., aggressive or lacks were relatively peripheral to the network. In contrast, criminal
anxiety). Kreis et al. (2012) found that 30 out of the 33 CAPP versatility, early behavioral problems and poor behavioral controls
items were rated at least moderately prototypical, and that 25 out (the latter two of which mix indices of criminal and antisocial
of the 33 CAPP items were rated as highly prototypical. So behavior) were moderately central. None of the items, however,
although most CAPP items were deemed prototypical for psychop- was highly central to PCL-R psychopathy, calling into question the
athy and their average prototypicality rating were very close to one assertion that nonspecific antisocial behavior is pivotal to the
another, the most protoypical items were lacks remorse, unem- conceptualization and operationalization of psychopathy (see also
pathic, self-centered, manipulative, and lacks emotional depth. Lykken, 1995).
These items—particularly lacks remorse and unempathic— Our findings may shed light on the importance of impulsivity
broadly corroborate the high centrality of EMP in the NIMH and for conceptualizing PCL-R defined psychopathy. Across samples,
Wisconsin sample, and its close connection with lack of remorse or impulsivity did not appear among the most central symptoms. This
guilt. finding fits with the conclusion of Poythress and Hall (2011) that
Moreover, the network findings also align with the findings the “blunt assertion that ‘psychopaths are impulsive’ is no longer
obtained with the elemental view on psychopathy from a basic trait defensible” (p. 120). At the same time, these authors argued that
perspective (Miller & Lynam, 2015). Studies drawing on the five there may be links between psychopathy and impulsivity if both
factor model of personality, using both expert ratings and the concepts are further refined. Impulsivity is a very heterogeneous
relations between psychopathy instruments and five factor person- concept, leading Whiteside and Lynam (2001, p. 677) to distin-
ality measures, have shown that psychopathy can be characterized guish among (lack of) premeditation (“the tendency to delay action
by facets of low agreeableness (e.g., low altruism), low conscien- in favor of careful thinking and planning”), urgency (“the tendency
tiousness (e.g., low self-discipline), high neuroticism (e.g., high to commit rash or regrettable actions as a result of intense negative
hostility), low (e.g., low warmth) and high (e.g., high excitement affect”), sensation seeking (“the tendency to seek excitement and
seeking) extraversion (e.g., Miller, Lynam, Widiger, & Leukefeld, adventure”), and (lack of) perseverance (“one’s ability to remain
2001; Lynam & Widiger, 2001; for a recent review, see Lynam & with a task until completion and avoid boredom”). Notably, these
Miller, 2015). Interestingly, the five-factor model of psychopathy differing subdimensions or pathways of impulsivity bear substan-
suggests that low agreeableness (encompassing EMP) is a central tially different implications for different forms of psychopathology
element of psychopathy. Such findings, which derive from very (Berg, Latzman, Bliwise, & Lilienfeld, 2015). Future research
different research designs, support the generalizability of affective- should examine whether these more refined subcomponents of
 62 VERSCHUERE ET AL.

impulsivity play a more prominent role in psychopathy, or in some the phenotypic expression of psychopathy in the U.S., and IRR and
forms of (e.g., secondary) psychopathy (Karpman, 1948). Anestis, PAR being central to the phenotypic expression of psychopathy in
Anestis, and Joiner (2009), for instance, found preliminary evi- the Netherlands. This possibility, however, would be at odds with
dence that secondary psychopathy was positively related primarily the fact that also Dutch clinicians view affective-interpersonal
to the tendency to act rashly when distressed (negative urgency). traits, most particularly a callous lack of empathy, as being
prototypical to psychopathy (Verschuere & te Kaat, 2017). Alter-
Network Analyses: Stability Within Samples and natively, the psychopathy measurement could differ between the
countries. At least some of the PCL-R items seem to have country-
Replicability Across Samples
specific content. For instance, it is quite uncommon in the Neth-
Although our findings were highly stable within each sample, erlands to engage in marriage or registered partnership more than
our study highlights the importance of examining replicability of twice before the age of 30 (required for a positive score on SHO)
network analyses across samples (Fried & Cramer, 2017), as we and consequently the Dutch sample had a moderately lower mean
found sizable and potentially important differences between the score for SHO compared with the U.S. samples, with the vast
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

U.S. samples and the Dutch sample. EMP was the most central majority (78%) scoring 0 (see Appendix IV). To the extent that the
This document is copyrighted by the American Psychological Association or one of its allied publishers.

item in the two U.S. samples, whereas IRR and PAR were most cross-country differences are replicable, they might require us to
central in the Dutch sample. We can consider at least three possible reconsider differences in the scoring or thresholds of these items
reasons for this divergence. across countries. Moreover, replicable differences would reopen
First, in the U.S. samples, the PCL-R was assessed for research the debate of whether the PCL-R as currently implemented in
purposes by well-trained research assistants, whereas the PCL-R in Dutch forensic psychiatric settings captures core psychopathic
the TBS sample was assessed by clinicians for the purpose of risk traits, or rather a deviant, antisocial lifestyle characterized by IRR
assessment. Although the reliability of the PCL-R, particularly and PAR (see, e.g., Skeem & Mulvey, 2001; Spreen, Ter Horst,
Factor 1 items, in North America and Europe tends to be lower in Lutjehuis, & Brand, 2008).
field than in research settings (Jeandarme et al., 2017; Miller,
Kimonis, Otto, Kline, & Wasserman, 2012), there were no marked
Limitations
differences in the reported interrater reliability statistics across
samples for the total PCL-R score in the present study—with the Our study was marked by several limitations. First, although our
caveat that interrater reliability estimates were based on small choice of the PCL-R can be justified given that it is the most
subsamples. Furthermore, self-selection in the U.S. sample is not extensively validated measure of psychopathy, arguably the most
evident from a comparison of the average PCL-R score of our U.S. important limitation of the present study is that it was restricted to
samples that is very similar to that obtained in previous U.S. a single instrument. This limitation raises the risk of mono-
samples (range of average score was 20 –24 across seven measurement bias (see, e.g., Skeem & Cooke, 2010). It is therefore
U.S. prison and forensic psychiatric samples; Hare et al., 2000). important not to commit the “error of reification” by assuming that
Finally, although the potential consequences of the PCL-R assess- our findings on the PCL-R necessarily bear on all operationaliza-
ment differed greatly between the U.S. samples (zero) and the tions of the construct of psychopathy. As noted later, for example,
Dutch sample (impact on the nature and duration of the mandatory the PCL-R does not explicitly assess fearlessness and other fea-
treatment, providing incentives to downplay psychopathic tenden- tures that some (e.g., Lykken, 1995) regard as central to psychop-
cies), it is important to bear in mind that the PCL-R assessment is athy. Also, the typical PCL-R assessment may bring about local
based not only on an interview, but also on extensive file review. dependencies between PCL-R features, creating artificial correla-
Second, the U.S. samples consisted mostly of non-mentally ill tions. For instance, the same (e.g., unempathic) behavior, such as
prisoners, whereas the Dutch sample consisted of forensic psychi- an especially callous crime, may bear on the ratings of several
atric patients. To illustrate, the prevalence of psychotic disorders in PCL-R items (Cooke & Michie, 2001). Likewise, the centrality of
the Dutch sample has been estimated to be as high as 39% EMP could be a result of a negative halo-effect (Thorndike,
(Nieuwenhuizen et al., 2011). In the U.S. prison samples, a prev- 1920)—that is, a cognitive bias in which the rating of one (unde-
alence of 3–7% psychotic disorders can be expected (Fazel & sirable) feature influences the judgment of other features (e.g., one
Danesh, 2002), and given the applied exclusion criteria, that prev- believes that an unattractive person is also unintelligent. To the
alence is expected to be even lower in the U.S. samples of the extent that clinicians consider EMP to be crucial to psychopathy
current study. To examine whether major psychopathology could (as indicated by prototypicality ratings), the rating of some other
explain the U.S.-Dutch differences, we extracted a subsample of PCL-R items may be influenced by their rating of this feature.
the Dutch sample, excluding those participants with indications of More broadly, if clinicians form a global negative impression of an
current or past severe psychopathology. PAR and IRR were still interviewee largely on the basis of his or her callousness and lack
most central in this subsample (n ⫽ 357; see Appendix III), of empathy (“This interviewee does not seem to be a nice person”),
rendering it less likely that severe psychopathology can explain the this impression could inadvertently shape their ratings of other
differences between the samples. PCL-R items. It will be crucial to extend our observations to other
Third, there are geographic differences between the U.S. (NIMH well-validated psychopathy instruments (e.g., self-report mea-
and Wisconsin) and the Dutch (TBS) samples. The question of sures; Lilienfeld & Fowler, 2006), and to the combinations of
whether PCL-R psychopathy is similar in the Netherlands and the instruments (e.g., combine PCL-R items with CAPP items; Cooke
U.S. entails two possibilities (Skeem, Edens, Camp, & Colwell, et al., 2012).
2004). The psychopathy construct itself could differ, such that Second, the use of other measures will help to clarify the
genetic and sociocultural factors give rise to EMP being central to importance of psychopathy features that are poorly represented in
 CENTRALITY OF PCL-R PSYCHOPATHY FEATURES 63

the PCL-R. Although some authors have placed great emphasis on Baskin-Sommers, A. R., & Newman, J. P. (2014). Psychopathic and
fearlessness and/or boldness in the conceptualization of psychop- externalizing offenders display dissociable dysfunctions when respond-
athy (Lykken, 1995; Patrick et al., 2009), others have argued that ing to facial affect. Personality Disorders, 5, 369 –379. http://dx.doi.org/
these features should be de-emphasized or even “be dropped from 10.1037/per0000077
psychopathy” altogether (Vize, Lynam, Lamkin, Miller, & Pardini, Berg, J. M., Hecht, L. K., Latzman, R. D., & Lilienfeld, S. O. (2015).
2016; p. 1). Although there is some evidence that aspects of Examining the correlates of the coldheartedness factor of the Psycho-
pathic Personality Inventory—Revised. Psychological Assessment, 27,
boldness are viewed by clinicians and researchers as prototypical
1494 –1499. http://dx.doi.org/10.1037/pas0000129
features of psychopathy (Berg, Lilienfeld, & Sellbom, 2017; Sör-
Berg, J. M., Latzman, R. D., Bliwise, N. G., & Lilienfeld, S. O. (2015).
man et al., 2016), network analysis on psychopathy instruments Parsing the heterogeneity of impulsivity: A meta-analytic review of the
that measure boldness, such as the Psychopathic Personality In- behavioral implications of the UPPS for psychopathology. Psychologi-
ventory—Revised (Lilienfeld & Widows, 2005), may shed further cal Assessment, 27, 1129 –1146. http://dx.doi.org/10.1037/pas0000111
light on the relevance of this trait to the psychopathy construct. Berg, J. M., Lilienfeld, S. O., & Sellbom, M. (2017). The role of boldness
Third, a deeper conceptual question is whether it is appropriate in psychopathy: A study of academic and clinical perceptions. Person-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

to apply network analyses to the concept of psychopathy and to ality Disorders. Advance online publication. http://dx.doi.org/10.1037/
This document is copyrighted by the American Psychological Association or one of its allied publishers.

personality disorders more broadly. Borsboom (2017) argued that per0000247

network theory may be more likely to serve as an explanatory Blair, R. J. (1995). A cognitive developmental approach to mortality:
model for some disorders (e.g., posttraumatic stress disorder Investigating the psychopath. Cognition, 57, 1–29. http://dx.doi.org/10
[PTSD]) than for other, particularly slowly developing, disorders; .1016/0010-0277(95)00676-P
the latter conditions may well include personality disorders, such Blair, R. J. R. (2005). Responding to the emotions of others: Dissociating
as psychopathy. Note, however, that the concern applies to net- forms of empathy through the study of typical and psychiatric popula-
tions. Consciousness and Cognition: An International Journal, 14, 698 –
work theory, rather than to network analyses. According to the
718. http://dx.doi.org/10.1016/j.concog.2005.06.004
network theory of mental disorders (Borsboom, 2017) no common
Borsboom, D. (2017). A network theory of mental disorders. World Psy-
cause explains covariance of psychiatric signs and symptoms;
chiatry, 16, 5–13. http://dx.doi.org/10.1002/wps.20375
signs and symptoms instead cause each other (for hybrid models Borsboom, D., & Cramer, A. O. (2013). Network analysis: An integrative
that combine network and latent variables models, see Epskamp, approach to the structure of psychopathology. Annual Review of Clinical
Rhemtulla, & Borsboom, 2017). For PTSD, for instance, anger and Psychology, 9, 91–121. http://dx.doi.org/10.1146/annurev-clinpsy-
hypervigilance might lead to poor sleep, which in turn makes it 050212-185608
difficult to concentrate at work. Network analyses help to elucidate Borsboom, D., Fried, E., Epskamp, E., Waldorp, L. J., van Borkulo, C. D.,
such possible causal effects. In the current study, we used network Van der Maas, H. L. J., & Cramer, A. O. J. (in press). Replicability of
analyses to reveal which features are most central to PCL-R psychopathology networks: The right question but the wrong answer. A
defined psychopathy. Whether network theory, which typically comment on “Evidence that psychopathology symptom networks have
speaks to rapidly developing feature-feature relations (e.g., poor limited replicability” by Forbes, Wright, Markon, and Krueger. Journal
sleep causing concentration difficulties), also applies to feature- of Abnormal Psychology.
feature relations that are likely to unfold more slowly over time Bringmann, L. F., Lemmens, L. H., Huibers, M. J., Borsboom, D., &
(e.g., EMP causing early behavioral problems) remains to be Tuerlinckx, F. (2015). Revealing the dynamic network structure of the
Beck Depression Inventory-II. Psychological Medicine, 45, 747–757.
investigated. Longitudinal designs with repeated measurements of
http://dx.doi.org/10.1017/S0033291714001809
psychopathy symptoms allow for testing this possibility.
Brook, M., Brieman, C. L., & Kosson, D. S. (2013). Emotion processing in
Psychopathy Checklist-assessed psychopathy: A review of the literature.
Conclusion Clinical Psychology Review, 33, 979 –995. http://dx.doi.org/10.1016/j
.cpr.2013.07.008
Applying network analysis to the PCL-R revealed several im-
Cleckley, H. (1941/1976). The mask of sanity. St. Loius, MO: Mosby.
portant findings, including the identification of callous/lack of Cooke, D. J., Hart, S. D., Logan, C., & Michie, C. (2012). Explicating the
empathy as a central feature of psychopathy in all three samples, construct of psychopathy: Development and validation of a conceptual
especially the American sample. At the same time, our findings are model, the Comprehensive Assessment of Psychopathic Personality
restricted to one instrument—the PCL-R—and point to noteworthy (CAPP). The International Journal of Forensic Mental Health, 11,
differences between the U.S. and the Dutch samples, especially in 242–252. http://dx.doi.org/10.1080/14999013.2012.746759
the importance of irresponsibility and PAR in the latter. Extending Cooke, D. J., & Michie, C. (2001). Refining the construct of psychopathy:
network analyses to different measures, samples, and cultures Towards a hierarchical model. Psychological Assessment, 13, 171–188.
should shed further light on the core characteristics of psychopathy http://dx.doi.org/10.1037/1040-3590.13.2.171
and perhaps ultimately on the unresolved question of what psy- Costantini, G., Epskamp, S., Borsboom, D., Perugini, M., Mõttus, R.,
chopathy is. Waldorp, L. J., & Cramer, A. (2015). State of the aRt personality
research: A tutorial on network analysis of personality data in R. Journal
of Research in Personality, 54, 13–29. http://dx.doi.org/10.1016/j.jrp
References .2014.07.003
American Psychiatric Association. (2013). Diagnostic and statistical man- Cronbach, L. J., & Meehl, P. E. (1955). Construct validity in psychological
ual of mental disorders, 5th ed. (DSM–5). Washington, DC: Author. tests. Psychological Bulletin, 52, 281–302. http://dx.doi.org/10.1037/
Anestis, M. D., Anestis, J. C., & Joiner, T. E. (2009). Affective consider- h0040957
ations in antisocial behavior: An examination of negative urgency in Dawel, A., O’Kearney, R., McKone, E., & Palermo, R. (2012). Not just
primary and secondary psychopathy. Personality and Individual Differ- fear and sadness: Meta-analytic evidence of pervasive emotion recog-
ences, 47, 668 – 670. http://dx.doi.org/10.1016/j.paid.2009.05.013 nition deficits for facial and vocal expressions in psychopathy. Neuro-
 64 VERSCHUERE ET AL.

science and Biobehavioral Reviews, 36, 2288 –2304. http://dx.doi.org/ Hart, S. D., & Dempster, R. J. (1997). Impulsivity and psychopathy. In
10.1016/j.neubiorev.2012.08.006 C. D. Webster & M. A. Jackson (Eds.), Impulsivity: Theory, assessment,
DeMatteo, D., Edens, J. F., Galloway, M., Cox, J., Smith, S. T., & Formon, and treatment (pp. 212–232). New York, NY: Guilford Press.
D. (2014). The role and reliability of the Psychopathy Checklist-Revised Hoff, H. A., Rypdal, K., Mykletun, A., & Cooke, D. J. (2012). A proto-
in U.S. sexually violent predator evaluations: A case law survey. Law typicality validation of the Comprehensive Assessment of Psychopathic
and Human Behavior, 38, 248 –255. http://dx.doi.org/10.1037/ Personality model (CAPP). Journal of Personality Disorders, 26, 414 –
lhb0000059 427. http://dx.doi.org/10.1521/pedi.2012.26.3.414
Epskamp, S. (2015). bootnet: Bootstrap methods for various network estimation Jeandarme, I., Edens, J. F., Habets, P., Bruckers, L., Oei, K., & Bogaerts,
routines. Retrieved from https://cran.r-project.org/web/packages/bootnet/ S. (2017). PCL-R field validity in prison and hospital settings. Law and
bootnet.pdf Human Behavior, 41, 29 – 43. http://dx.doi.org/10.1037/lhb0000222
Epskamp, S., Borsboom, D., & Fried, E. (2016). Estimating psychological Karpman, B. (1948). The myth of the psychopathic personality. The
networks and their stability: A tutorial paper. arXiv preprint arXiv:1604. American Journal of Psychiatry, 104, 523–534. http://dx.doi.org/10
08462. .1176/ajp.104.9.523
Epskamp, S., Cramer, A., Waldorp, L., Schmittmann, V., & Borsboom, D. Kreis, M. K., Cooke, D. J., Michie, C., Hoff, H. A., & Logan, C. (2012).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

(2012). qgraph: Network visualizations of relationships in psychometric The Comprehensive Assessment of Psychopathic Personality (CAPP):
This document is copyrighted by the American Psychological Association or one of its allied publishers.

data. Journal of Statistical Software, 48, 1–18. http://dx.doi.org/10 Content validation using prototypical analysis. Journal of Personality
.18637/jss.v048.i04 Disorders, 26, 402– 413. http://dx.doi.org/10.1521/pedi.2012.26.3.402
Epskamp, S., Rhemtulla, M., & Borsboom, D. (2017). Generalized network Lilienfeld, S. O., & Andrews, B. P. (1996). Development and preliminary
psychometrics: Combining network and latent variable models. Psy- validation of a self-report measure of psychopathic personality traits in
chometrika. Advance online publication. http://dx.doi.org/10.1007/ noncriminal populations. Journal of Personality Assessment, 66, 488 –
s11336-017-9557-x 524. http://dx.doi.org/10.1207/s15327752jpa6603_3
Fazel, S., & Danesh, J. (2002). Serious mental disorder in 23000 prisoners: Lilienfeld, S. O., & Fowler, K. A. (2006). The Self-Report Assessment of
A systematic review of 62 surveys. The Lancet, 359, 545–550. http://dx Psychopathy: Problems, pitfalls, and promises. In C. J. Patrick (Ed.),
.doi.org/10.1016/S0140-6736(02)07740-1 Handbook of psychopathy (pp. 107–133). New York, NY: Guilford
Flórez, G., Casas, A., Kreis, M. K., Forti, L., Martínez, J., Fernández, J., Press.
. . . Cooke, D. J. (2015). A prototypicality validation of the Compre- Lilienfeld, S. O., Patrick, C. J., Benning, S. D., Berg, J., Sellbom, M., &
hensive Assessment of Psychopathic Personality (CAPP) model Spanish Edens, J. F. (2012). The role of fearless dominance in psychopathy:
version. Journal of Personality Disorders, 29, 707–718. http://dx.doi Confusions, controversies, and clarifications. Personality Disorders:
.org/10.1521/pedi_2014_28_167 Theory, Research, and Treatment, 3, 327–340. http://dx.doi.org/10
Forbes, M. K., Wright, A. G. C., Markon, K. E., & Krueger, R. F. (2017). .1037/a0026987
Evidence that psychopathology symptom networks do not replicate. Lilienfeld, S. O., Watts, A. L., Francis Smith, S., Berg, J. M., & Latzman,
Journal of Abnormal Psychology. R. D. (2015). Psychopathy deconstructed and reconstructed: Identifying
Frick, P. J., Ray, J. V., Thornton, L. C., & Kahn, R. E. (2014). Can and assembling the personality building blocks of Cleckley’s chimera.
callous-unemotional traits enhance the understanding, diagnosis, and Journal of Personality, 83, 593– 610. http://dx.doi.org/10.1111/jopy
treatment of serious conduct problems in children and adolescents? A .12118
comprehensive review. Psychological Bulletin, 140, 1–57. http://dx.doi Lilienfeld, S. O., & Widows, M. R. (2005). PPI-R: Psychopathic Person-
.org/10.1037/a0033076 ality Inventory Revised: Professional manual. Lutz, FL: Psychological
Fried, E. I., & Cramer, A. O. J. (2017). Moving forward: Challenges and Assessment Resources, Incorporated.
directions for psychopathological network theory and methodology. Lorber, M. F. (2004). Psychophysiology of aggression, psychopathy, and
Perspectives on Psychological Science. Advance online publication. conduct problems: A meta-analysis. Psychological Bulletin, 130, 531–
http://dx.doi.org/10.1177/1745691617705892 552. http://dx.doi.org/10.1037/0033-2909.130.4.531
Fried, E. I., Epskamp, S., Nesse, R. M., Tuerlinckx, F., & Borsboom, D. Lykken, D. T. (1995). The antisocial personalities. Hillsdale, NJ: Erlbaum.
(2016). What are “good” depression symptoms? Comparing the central- Lynam, D. R., Hoyle, R. H., & Newman, J. P. (2006). The perils of
ity of DSM and non-DSM symptoms of depression in a network analysis. partialling: Cautionary tales from aggression and psychopathy. Assess-
Journal of Affective Disorders, 189, 314 –320. http://dx.doi.org/10.1016/ ment, 13, 328 –341. http://dx.doi.org/10.1177/1073191106290562
j.jad.2015.09.005 Lynam, D. R., & Miller, J. D. (2015). Psychopathy from a Basic Trait
Fruchterman, T. M., & Reingold, E. M. (1991). Graph drawing by force- Perspective: The Utility of a Five-Factor Model Approach. Journal of
directed placement. Software, Practice & Experience, 21, 1129 –1164. Personality, 83, 611– 626. http://dx.doi.org/10.1111/jopy.12132
http://dx.doi.org/10.1002/spe.4380211102 Lynam, D. R., & Widiger, T. A. (2001). Using the five-factor model to
Gough, H. G. (1948). A sociological theory of psychopathy. American represent the DSM–IV personality disorders: An expert consensus ap-
Journal of Sociology, 53, 359 –366. http://dx.doi.org/10.1086/220203 proach. Journal of Abnormal Psychology, 110, 401– 412. http://dx.doi
Hare, R. D. (1991). The Hare Psychopathy Checklist—Revised. Toronto, .org/10.1037/0021-843X.110.3.401
Canada: Multi-Health Systems. McCord, W., & McCord, J. (1964). The psychopath: An essay on the
Hare, R. D. (1993). Without conscience: The disturbing world of the criminal mind. Princeton, NJ: Van Nostrand.
psychopaths among us. New York, NY: Simon & Schuster. McNally, R. J., Robinaugh, D. J., Wu, G. W., Wang, L., Deserno, M., &
Hare, R. D. (2003). The Hare Psychopathy Checklist—Revised (2nd ed.). Borsboom, D. (2015). Mental disorders as causal systems: A network
Toronto, Canada: Multi-Health Systems. approach to posttraumatic stress disorder. Clinical Psychological Sci-
Hare, R. D., Clark, D., Grann, M., & Thornton, D. (2000). Psychopathy and ence, 3, 1–14. http://dx.doi.org/10.1177/2167702614553230
the predictive validity of the PCL-R: An international perspective. Miller, C. S., Kimonis, E. R., Otto, R. K., Kline, S. M., & Wasserman,
Behavioral Sciences & the Law, 18, 623– 645. http://dx.doi.org/10.1002/ A. L. (2012). Reliability of risk assessment measures used in sexually
1099-0798(200010)18:5⬍623::AID-BSL409⬎3.0.CO;2-W violent predator proceedings. Psychological Assessment, 24, 944 –953.
Hare, R. D., & Neumann, C. S. (2010). The role of antisociality in the http://dx.doi.org/10.1037/a0028411
psychopathy construct: Comment on Skeem and Cooke (2010). Psycho- Miller, J. D., & Lynam, D. R. (2012). An examination of the Psychopathic
logical Assessment, 22, 446 – 454. http://dx.doi.org/10.1037/a0013635 Personality Inventory’s nomological network: A meta-analytic review.
 CENTRALITY OF PCL-R PSYCHOPATHY FEATURES 65

Personality Disorders: Theory, Research, and Treatment, 3, 305–326. debate. Psychological Assessment, 22, 433– 445. http://dx.doi.org/10
http://dx.doi.org/10.1037/a0024567 .1037/a0008512
Miller, J. D., & Lynam, D. R. (2015). Understanding psychopathy using Skeem, J. L., Edens, J. F., Camp J., & Colwell, L. H. (2004). Are there
the basic elements of personality. Social and Personality Psychology ethnic differences in levels of psychopathy? A meta-analysis. Law &
Compass, 9, 223–237. http://dx.doi.org/10.1111/spc3.12170 Human Behavior, 28, 505–527. http://dx.doi.org/10.1023/B:LAHU
Miller, J. D., Lynam, D. R., Widiger, T. A., & Leukefeld, C. (2001). .0000046431.93095.d8
Personality disorders as extreme variants of common personality dimen- Skeem, J. L., & Mulvey, E. P. (2001). Psychopathy and community
sions: Can the five-factor model adequately represent psychopathy? violence among civil psychiatric patients: Results from the MacArthur
Journal of Personality, 69, 253–276. http://dx.doi.org/10.1111/1467- Violence Risk Assessment Study. Journal of Consulting and Clinical
6494.00144 Psychology, 69, 358 –374. http://dx.doi.org/10.1037/0022-006X.69.3
Nentjes, L., Bernstein, D. P., Meijer, E., Arntz, A., & Wiers, R. W. (2016). .358
The Mask of Sanity: Facial Expressive, Self-Reported, and Physiologi- Skeem, J. L., Polaschek, D. L., Patrick, C. J., & Lilienfeld, S. O. (2011).
cal Consequences of Emotion Regulation in Psychopathic Offenders. Psychopathic personality: Bridging the gap between scientific evidence
Journal of Personality Disorders, 30, 828–S8. http://dx.doi.org/10.1521/ and public policy. Psychological Science in the Public Interest, 12,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

pedi_2016_30_235 95–162. http://dx.doi.org/10.1177/1529100611426706

Newman, J. P., MacCoon, D. G., Vaughan, L. J., & Sadeh, N. (2005).
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Sörman, K., Edens, J. F., Smith, S. T., Clark, J. W., Kristiansson, M., &
Validating a distinction between primary and secondary psychopathy Svensson, O. (2016). Boldness and its relation to psychopathic person-
with measures of Gray’s BIS and BAS constructs. Journal of Abnormal ality: Prototypicality analyses among forensic mental health, criminal
Psychology, 114, 319 –323. http://dx.doi.org/10.1037/0021-843X.114.2 justice, and layperson raters. Law and Human Behavior, 40, 337–349.
.319 http://dx.doi.org/10.1037/lhb0000176
Nieuwenhuizen, C., van Bogaerts, S., Ruijter, E. A. W., Bongers, I. L., Sörman, K., Edens, J. F., Smith, S. T., Svensson, O., Howner, K., Kris-
Coppens, M., & Meijers, R. A. A. C. (2011). TBS-behandeling gepro-
tiansson, M., & Fischer, H. (2014). Forensic mental health professionals’
fileerd. Een gestructureerde casussenanalyse. Den Haag, the Nether-
perceptions of psychopathy: A prototypicality analysis of the Compre-
lands: WODC.
hensive Assessment of Psychopathic Personality in Sweden. Law and
Opsahl, T., Agneessens, F., & Skvoretz, J. (2010). Node centrality in
Human Behavior, 38, 405– 417. http://dx.doi.org/10.1037/lhb0000072
weighted networks: Generalizing degree and shortest paths. Social Net-
Spreen, M., Ter Horst, P., Lutjehuis, B., & Brand, E. (2008). De kwaliteit
works, 32, 245–251. http://dx.doi.org/10.1016/j.socnet.2010.03.006
en de rol van de PCL-R in het Nederlands tbs-systeem [The quality and
Patrick, C. J., Bradley, M. M., & Lang, P. J. (1993). Emotion in the
the role of the PCL-R in the Dutch TBS-system]. Developmental Psy-
criminal psychopath: Startle reflex modulation. Journal of Abnormal
chology, 43, 78 –98.
Psychology, 102, 82–92. http://dx.doi.org/10.1037/0021-843X.102.1.82
Thorndike, E. L. (1920). A constant error in psychological ratings. Journal
Patrick, C. J., Fowles, D. C., & Krueger, R. F. (2009). Triarchic concep-
of Applied Psychology, 4, 25–29. http://dx.doi.org/10.1037/h0071663
tualization of psychopathy: Developmental origins of disinhibition,
Verschuere, B., & te Kaat, L. (2017, May). What are the core features of
boldness, and meanness. Development and Psychopathology, 21, 913–
938. http://dx.doi.org/10.1017/S0954579409000492 psychopathy? A prototypical analysis using the Psychopathy Checklist-
Philipse, M. G. W. (2005). Predicting criminal recidivism, empirical Revised (PCL-R). Paper presented at the Annual Conference of the
studies and clinical practice in forensic psychiatry. Enschede, the Neth- European Association of Psychology and Law, Mechelen, Belgium.
erlands: Febodruk. Vertommen, H., Verheul, R., De Ruiter, C., & Hildebrand, M. (2002).
Poythress, N. G., & Hall, J. R. (2011). Psychopathy and impulsivity Hare’s Psychopathie checklist: Handleiding [Dutch version of the Hare
reconsidered. Aggression and Violent Behavior, 16, 120 –134. http://dx Psychopathy Checklist-Revised-Manual]. Lisse, the Netherlands: Swets
.doi.org/10.1016/j.avb.2011.02.003 Test Publisher.
Poythress, N. G., Lilienfeld, S. O., Skeem, J. L., Douglas, K. S., Edens, Vize, C. E., Lynam, D. R., Lamkin, J., Miller, J. D., & Pardini, D. (2016).
J. F., Epstein, M., & Patrick, C. J. (2010). Using the PCL-R to help Identifying essential features of juvenile psychopathy in the prediction
estimate the validity of two self-report measures of psychopathy with of later antisocial behavior: Is there an additive, synergistic, or curvi-
offenders. Assessment, 17, 206 –219. http://dx.doi.org/10.1177/ linear role for fearless dominance? Clinical Psychological Science, 4,
1073191109351715 572–590. http://dx.doi.org/10.1177/2167702615622384
R Core Team. (2016). R: A language and environment for statistical Whiteside, S. P., & Lynam, D. R. (2001). The five factor model and
computing. Vienna, Austria: R Foundation for Statistical Computing. impulsivity: Using a structural model of personality to understand im-
Retrieved from https://www.R-project.org/ pulsivity. Personality and Individual Differences, 30, 669 – 689. http://
Rhemtulla, M., Fried, E. I., Aggen, S. H., Tuerlinckx, F., Kendler, K. S., & dx.doi.org/10.1016/S0191-8869(00)00064-7
Borsboom, D. (2016). Network analysis of substance abuse and depen-
dence symptoms. Drug and Alcohol Dependence, 161, 230 –237. http://
dx.doi.org/10.1016/j.drugalcdep.2016.02.005 Received April 11, 2017
Skeem, J. L., & Cooke, D. J. (2010). Is criminal behavior a central Revision received July 28, 2017
component of psychopathy? Conceptual directions for resolving the Accepted August 4, 2017 䡲