Convergent and Discriminant Validity of Psychopathy Factors Assessed Via Self-Report

Abstract

Psychopathy has been conceptualized as a personality disorder with distinctive interpersonal-affective and behavioral deviance features. The authors examine correlates of the factors of the Psychopathic Personality Inventory (PPI), Self-Report Psychopathy–II (SRP-II) scale, and Antisocial Process Screening Device (APSD) to understand similarities and differences among the constructs embodied in these instruments. PPI Fearless Dominance and SRP-II Factor 1 were negatively related to most personality disorder symptoms and were both predicted by high Dominance and low Neuroticism. In addition, PPI Fearless Dominance correlated positively with antisocial personality features, although SRP-II Factor 1 did not. In contrast, PPI Impulsive Antisociality, SRP-II Factor 2, and both APSD factors correlated with antisocial personality features and symptoms of nearly all personality disorders, and were predicted by low Love. Results suggest ways in which the measurement of the constructs in each instrument may be improved.

Psychopathy is a personality disorder that is distinguished from common criminality and chronic antisociality by the presence of distinctive interpersonal-affective features (Hare, 2003; Harpur, Hare, & Hakstian, 1989; Lilienfeld, 1994; Lykken, 1995; Patrick & Lang, 1999). Recent self-report measures of psychopathy, including the Self-Report Psychopathy scale (SRP; Hare, 1985, 1991) and the Psychopathic Personality Inventory (PPI; Lilienfeld & Andrews, 1996), demonstrate better convergence and coverage of both the interpersonal-affective and antisocial lifestyle features of psychopathy. Factor analytic work on these and other self-report instruments of psychopathy has revealed that many of these instruments have a two-factor structure, each of which parses the interpersonal-affective and behavioral features into separate subscales (Benning, Patrick, Hicks, Blonigen, & Krueger, 2003; Frick, O’Brien, Wootton, & McBurnett, 1994; Hare, 1991). Here, we examined in an undergraduate sample the construct validity of the two-factor structures of the PPI, the SRP-II scale, and a self-report version of the Antisocial Process Screening Device (APSD; Frick & Hare, 2001). We examined the relations of each psychopathy factor to the others, to symptoms of other personality disorders, and to normal-range personality dimensions.

THE NOMOLOGICAL NETWORK SURROUNDING PSYCHOPATHY

Cronbach and Meehl (1955) proposed the concept of the nomological network as a theoretical framework for understanding construct validity. In this framework, a network is set up between unobserved theoretical constructs based on their relations to the observed variables used to measure them. Each theoretical construct is also linked to other constructs in the network, which broadens in coverage and scope as additional variables are related to the initial nodes. In this way, open-ended constructs are continually redefined and clarified through their lawful, probabilistic relationships with other constructs.

There are multiple layers of this nomological network into which each psychological construct is embedded. In the case of psychopathy measures, the first layer of the nomological network involves one measure’s relationships with other measures of psychopathy. If psychopathy measures assess similar constructs, they should be strongly related to each other. A second layer in the nomological network is the relationship of each psychopathy measure with measures of other personality disorders. At this layer, the construct validity of psychopathy measures is demonstrated by their convergent relationships with related personality disorders, such as the dramatic-emotional Cluster B disorders, and divergent relationships with unrelated personality disorders, such as the odd-erratic Cluster A disorders.

A third layer of the nomological network involves the relationships of psychopathy measures with normal-range personality dimensions. At this layer, psychopathy measures exhibit construct validity by their similar relationships with theoretically meaningful personality dimensions. Psychopathy’s profound effects on a person’s interpersonal behavior have been mapped onto the interpersonal circumplex (which organizes personality around two orthogonal axes of agentic Dominance and communal Love; Trapnell & Wiggins, 1990). In this framework, psychopathy seems to be located in the arrogant-calculating octant (Harpur et al., 1989), represented by high Dominance and low Love. Psychopathy has also been described with reference to the Five-Factor Model of personality (FFM), which divides personality into five domains of Neuroticism, Extraversion (which has strong parallels with interpersonal Dominance; Trapnell & Wiggins, 1990), Openness, Agreeableness (which has strong parallels with interpersonal Love; Trapnell & Wiggins, 1990), and Conscientiousness. In the FFM, an average expert-generated prototype of psychopathy consisted of facets marking low Neuroticism, Agreeableness, and Conscientiousness (Miller, Lynam, Widiger, & Leukefeld, 2001).

Early attempts to measure psychopathy through self-report, such as the Pd scale of the Minnesota Multiphasic Personality Inventory–2 (McKinley & Hathaway, 1944) and the So scale of the California Psychological Inventory (Gough, 1960), shared only 7% to 28% of their variance (Hare, 1985), suggesting that they varied substantially in the psychopathy constructs that they measured. These measures also shared only 8% to 14% of their variance with global ratings of “Cleckley psychopathy” (Cleckley, 1941/1988), Psychopathy Checklist (PCL) scores, and antisocial personality disorder (APD) symptoms (Hare, 1985), implying that they showed poor convergence both with other measures of psychopathy and with related personality disorders. Lilienfeld (1994) suggested that one major reason these early self-report psychopathy measures showed such poor construct validity was because they did not share a common nomological network of traits that described this personality disorder.

A RESURGENCE IN THE MEASUREMENT OF PSYCHOPATHY

A new approach to operationalizing the criteria for psychopathy that Cleckley (1941/1988) delineated in the Mask of Sanity created a central node for the nomological network of psychopathy. Hare’s (1980) PCL and its successor, the PCL–Revised (PCL-R; Hare, 1991), are composed of items that discriminated between prisoners who received high ratings on global ratings of Cleckley psychopathy from those who were rated as low on global psychopathy. This checklist provided an anchor for the burgeoning nomological network of psychopathy, and self-report measures designed to facilitate larger-scale assessments of the psychopathy construct embodied in Cleckley’s (1941/1988) criteria emerged shortly afterward. Hare (1985) created the first self-report instrument based on the PCL, the Self-Report Psychopathy Scale (SRP), although its correlation with PCL scores was modest (.28; Hare, 1985). A 60-item revision, the SRP-II, performed substantially better, correlating .54 with PCL-R total scores in a sample of incarcerated males (Hare, 1991) and between .55 and .62 with the PCL: Screening Version, (PCL:SV; Forth, Brown, Hart, & Hare, 1996), the non-forensic version of the PCL-R. SRP-II total scores also relate to Machiavellianism and narcissism, which are constructs long associated with psychopathy (Paulhus & Williams, 2002).

During this time, other self-report measures of psychopathy were developed. One was an alternative measure of the PCL-R that was developed for use in children: the APSD (Frick et al., 1994). Although the APSD was originally designed to be completed by parent and teacher raters of a child, the APSD has also been used as a self-report measure in adolescents to study the developmental precursors to psychopathy. For example, youths who have an earlier onset of behavioral problems also have higher APSD scores (Silverthorn, Frick, & Reynolds, 2001). In contrast, the PPI (Lilienfeld & Andrews, 1996) was developed to provide a comprehensive assessment of the personality characteristics that multiple theorists viewed as salient to psychopathy without explicit reference to the antisocial acts that individuals with psychopathy frequently commit. Total PPI scores correlated .29 with Pd scores and −.59 with So scores, and impressively, they correlated .91 with SRP-II scores (Lilienfeld & Andrews, 1996), suggesting that the nomological network of self-report psychopathy had tightened substantially since Hare’s (1985) critique. Scores on the PPI also correlated .54 with PCL-R total scores (Poythress, Edens, & Lilienfeld, 1998), implying that the PPI shares a position in the nomological network with this respected psychopathy measure as well.

FACTORS OF SELF-REPORT PSYCHOPATHY

The nomological network surrounding psychopathy required significant revision when the PCL-R was revealed to have two factors that indexed separable interpersonal-affective and behavioral deviance facets of psychopathy and demonstrated distinct external correlates (Hare et al., 1990; Harpur et al., 1989). For example, Factor 1 of the PCL-R is more strongly related to dominant, narcissistic personality features (Hart & Hare, 1989), whereas Factor 2 is more strongly related to APD symptoms (Hart & Hare, 1989) and symptoms of substance misuse (Smith & Newman, 1990). To mirror the factor structure of the PCL-R in the SRP-II, Hare (1991) rationally assigned items from this instrument to create separate scales that measured these two psychopathy facets. SRP-II Factor 1 scores are related to low state and trait anxiety scores, low empathetic personal distress, and narcissistic personality features; but SRP-II Factor 2 scores are also consistently correlated positively with narcissistic features (Williams & Paulhus, 2004; Zagon & Jackson, 1994). SRP-II Factor 2 has been associated with Machiavellian personality features and all facets of deviant behavior, whereas SRP-II Factor 1 has not (Williams & Paulhus, 2004). Within the FFM, SRP-II Factor 1 has been associated with high Extraversion and low Neuroticism. In contrast, SRP-II Factor 2 has been related to high Extra-version, Neuroticism, and Openness and low Conscientiousness (Williams & Paulhus, 2004).

Both factors of the SRP-II are also related strongly to PPI total scores (.54 with Factor 1, .71 with Factor 2; Williams & Paulhus, 2004), and recent research indicates that the PPI has a two-factor structure of its own (Benning et al., 2003). The subscales of the PPI aggregate into two uncorrelated higher-order factors: PPI-I, or “Fearless Dominance” (Benning, Patrick, Blonigen, Hicks, & Iacono, 2005), is composed of Social Potency, Stress Immunity, and Fearlessness; PPI-II, or “Impulsive Antisociality” (Benning et al., 2005), is composed of Machiavellian Egocentricity, Impulsive Nonconformity, Blame Externalization, and Carefree Nonplanfulness. Fearless Dominance is related to adult symptoms of antisocial personality disorder, higher socioeconomic status, and high dominance and low trait and anticipatory anxiety (Benning et al., 2003). In contrast, Impulsive Antisociality is related to child and adult symptoms of antisocial personality disorder, substance abuse and dependence symptoms, low socioeconomic status, and high trait negative emotionality and low behavioral constraint (Benning et al., 2003).

The APSD also comprises distinct factors that exhibit discriminant associations with criterion measures (Frick et al., 1994). The APSD Callous-Unemotional traits factor of the APSD bears similarities to Factor 1 of the PCL-R, correlating with sensation seeking and low anxiety in children (Frick et al., 1994). It also correlates with low empathy, a social-cognitive structure oriented toward tangible reward and dominance (Pardini, Lochman, & Frick, 2003), reduced facilitation in processing negative words (Loney, Frick, Clements, Ellis, & Kerlin, 2003), and being adjudicated as a sex offender (Caputo, Frick, & Brodsky, 1999). In contrast, scores on Impulsive-Conduct problems are related to conduct disorder symptoms (Frick et al., 1994), even in girls with adolescent-onset conduct disorder (Silverthorn et al., 2001), and with general behavioral dysregulation (Pardini et al., 2003).

CURRENT STUDY: THE NOMOLOGICAL NET OF PSYCHOPATHY FACTORS INDEXED BY SELF-REPORT

Salekin, Trobst, and Krioukova (2001) employed a multilevel approach to the construct validation of self-report psychopathy measures within multiple layers of the nomological network: those of psychopathy, of personality disorders, and of normal-range personality broadly. In this study, total SRP-II and PPI scores were preferentially related to symptoms of Cluster B personality disorders, particularly borderline and antisocial personality disorders. PPI and SRP-II total scores were both located within the arrogant-calculating octant of the interpersonal circumplex. Furthermore, SRP-II Factor 1 and the subscales that make up PPI Fearless Dominance appeared to cluster around the assured-dominant octant. In contrast, the scales composing PPI Impulsive Antisociality appeared to cluster around the coldhearted octant of the interpersonal circumplex, whereas SRP-II Factor 2 scores resided in the arrogant-calculating octant.

However, Salekin et al. (2001) were concerned primarily with total self-report psychopathy scores and briefly touched only on the interpersonal locations of the SRP-II factors without examining additional aspects of their criterion validity. In addition, Salekin et al. (2001) did not report on the APSD, which we included to explore its structure and construct validity in undergraduates, for whom the APSD was not explicitly designed as a measure of psychopathy. In the current study, data originally presented in Salekin et al. (2001) were reanalyzed and combined with additional data collected in that sample. This procedure allowed us to perform a number of analyses on the factors of three measures of psychopathy—the PPI, the SRP-II, and the APSD—to locate them in a number of layers of the nomological network.

First, the internal replicability of each instrument’s two-factor structure was assessed through the use of exploratory factor analyses and by calculating factor congruence coefficients between the resulting factor structures and published factor loading matrices (cf., Harpur, Hakstian, & Hare, 1988). The published factors of each psychopathy measure were then correlated with one another and subjected to a joint factor analysis to investigate their common structure. The PPI, SRP-II, and APSD factors were also correlated with the three subscales of the Antisocial Features scale of the Personality Assessment Inventory (PAI-ANT; Morey, 1991) scale to examine their relations within the layer of the nomological network pertaining specifically to psychopathy and with self-reported symptoms of the diagnostic and experimental personality disorders from the fourth edition of the Diagnostic Statistical Manual of Mental Disorders (DSM-IV; American Psychiatric Association, 1994) to understand their place in the layer of the nomological network pertaining to other personality disorders. Finally, the projections of PPI, SRP-II, and APSD factors onto the interpersonal circumplex and their relations to Big Five personality dimensions were investigated to place these factors of psychopathy within the nomological network layer of normal-range personality dimensions. Understanding the location of each factor of self-report psychopathy in the nomological network allowed us to refine the factors of these instruments to measure common psychopathy constructs more effectively.

HYPOTHESES

Psychopathy factors

Because of the similarity of the constructs measured by PPI Fearless Dominance and SRP-II Factor 1, along with their apparently shared locations within the interpersonal circumplex, they were expected to correlate highly. However, because PPI Fearless Dominance is unrelated to the deficient affective experience features of psychopathy (Benning et al., 2005)—which, based on its item content, the APSD Callous-Unemotional traits scale may measure—we hypothesized that APSD Callous-Unemotional traits would be divergently valid from PPI Fearless Dominance and SRP-II Factor 1. Because PPI Impulsive Antisociality, SRP-II Factor 2, and APSD Impulsive-Conduct problems are all designed to measure similar constructs, we predicted that they would all be strongly intercorrelated.

Personality disorders

Because total psychopathy scores have been shown to be preferentially related to Cluster B personality disorders in this sample, we expected that each factor would also be preferentially associated with Cluster B personality disorder symptoms. However, we predicted that each factor would be differentially associated with particular Cluster B disorders. We hypothesized that PPI Fearless Dominance and SRP-II Factor 1 would be related most strongly to symptoms of narcissistic personality disorder, whereas PPI Impulsive Antisociality, SRP-II Factor 2, and APSD Impulsive-Conduct problems would be preferentially related to antisocial personality disorder symptoms. We also hypothesized that PPI Fearless Dominance and SRP-II Factor 1 are related to fewer symptoms of anxiety; we also predicted that they would be inversely related to symptoms of Cluster C personality disorders. Because the deficient affective experience factor of psychopathy is related to low Social Closeness (Hall, Benning, & Patrick, 2004), we hypothesized that APSD Callous-Unemotional traits might also be related to the lack of closeness with other people that is described by schizoid personality symptoms.

Interpersonal circumplex and FFM dimensions

From Salekin et al. (2001), we predicted that PPI Fearless Dominance and SRP-II Factor 1 would lie within the assured-dominant octant of the circumplex, that PPI Impulsive Antisociality would lie in the coldhearted octant, and that SRP-II Factor 2 would reside in the arrogant-calculating octant. Because PCL-R Factor 1 resides in the arrogant-calculating octant and PCL-R resides in the coldhearted octant of the interpersonal circumplex (Harpur et al., 1989), we predicted that APSD Callous-Unemotional traits and Impulsive-Conduct problems would occupy the same interpersonal locations, respectively. Based on results from Ross, Benning, Patrick, Thompson, and Thurston (2004) for the PPI, along with the hypothesized strong correlation between PPI Fearless Dominance and SRP-II Factor 1, we expected that these two factors would be predicted by high Dominance, low Neuroticism, and low Love. In contrast, we expected that PPI Impulsive Antisociality, SRP-II Factor 2, and APSD Impulsive-Conduct problems would all be most strongly predicted by low Love, although we also expected that SRP-II Factor 2 would also be predicted by high Dominance, given its interpersonal circumplex location in Salekin et al. (2001). The predictions for APSD Callous-Unemotional traits and normal-range personality traits were unclear, because few normal-range traits have been shown to relate to similar features of psychopathy (Hall et al., 2004).

METHOD

Participants

Participants were 326 psychology students (mean age = 22.0 years, SD = 7.09) who participated to fulfill a course requirement. Participants were 56% female, with the following racial breakdown: 19.6% African American, 2.8% Asian, 23.3% Caucasian, 42.3% Hispanic, 0.3% Native American, 4.6% biracial, and 11% missing racial data.

Measures

PPI (Lilienfeld & Andrews, 1996)

The PPI is a 187-item measure that was developed as a comprehensive measure of the personality characteristics in psychopathy in nonincarcerated samples. Items are coded on a 1 (false) to 4 (true) scale. It has eight subscales that may be organized into a two-factor higher-order structure: Social Potency, Stress Immunity, and Fearlessness comprise Fearless Dominance; and Machiavellian Egocentricity, Impulsive Nonconformity, Blame Externalization, and Carefree Nonplanfulness comprise Impulsive Antisociality (Benning et al., 2003). As measured by coefficient alpha, internal consistencies of individual PPI subscales ranged from .76 (Carefree Nonplanfulness) to .88 (Machiavellian Egocentricity). Because the PPI factors were computed from the sums of z scores, the means for Fearless Dominance and Impulsive Antisociality were 0 (SDs = 2.29 and 2.94, respectively).

SRP-II (Hare, 1991)

The SRP-II is a 60-item experimental self-report measure of psychopathy that was designed to measure the construct of psychopathy embodied in the PCL-R. Each item is scored on a 1 (disagree strongly) to 7 (agree strongly) scale. Nine items from this measure form a scale measuring the personality characteristics associated with the disorder (Factor 1; internal consistency = .47, M = 34.4, SD = 6.72), and 13 items comprise a scale that indexes the behavioral tendencies in psychopathy (Factor 2; internal consistency = .77, M = 40.1, SD = 12.4).

APSD (Frick & Hare, 2001)

The APSD is a 20-item measure that was created to assess the construct of psychopathy indexed in the PCL-R in an instrument suitable for children and adolescents. Like the PCL-R, each item is scored on a 0 (not at all true) to 2 (definitely true) scale. Six items comprise the Callous-Unemotional traits scale (internal consistency = .54; M = 3.11, SD = 1.87), and 10 items comprise the Impulsive-Conduct Problems scale (internal consistency = .64; M = 5.58, SD = 2.87).

PAI-ANT (Morey, 1991)

The PAI-ANT scale is a 24-item measure of psychopathy that is composed of three facet scales. ANT–Antisocial Behaviors (ANT-A; M = 6.54, SD = 2.54) assesses the overt antisocial behavior history present in psychopathy, ANT–Egocentricity (ANT-E; M = 4.85, SD = 3.60) was designed to measure the narcissistic superiority characteristic of psychopathy, and ANT–Stimulus-Seeking (ANT-S; M = 7.06, SD = 4.47) indexes the predilection toward risk taking and novelty seeking in psychopathy. Each item is scored on a 0 (false) to 3 (very true) scale. Salekin et al. (2001) showed that PAI-ANT scores were preferentially related to symptoms of Cluster B personality disorders, particularly those of antisocial personality disorder (r = .70). PAI-ANT scores have been noted to relate strongly to PPI total scores (Edens, Poythress, & Watkins, 2001) and to PCL-R total scores (Edens, Hart, Johnson, Johnson, & Olver, 2000). However, neither PAI-ANT nor any of its subscales seem to be related to the unique variance in PCL-R Factor 1, and the ANT-E scale does not correlate with total or factor scores of the PCL-R (Edens et al., 2000).

Personality Diagnostic Questionnaire–4+ (PDQ-4+; Hyler, 1994)

The PDQ-4+ was designed to provide a self-report measure of symptoms of all personality disorders described in the DSM-IV (American Psychiatric Association, 1994) as well as symptoms of several experimental personality disorders. Each item is scored true or false. Diagnoses based on previous versions of the PDQ have been shown to be moderately related to diagnoses obtained from two structured interviews (average κs were .36 and .42; Hyler, Skodol, Kellman, Oldham, & Rosnick, 1990; Hyler, Skodal, Oldham, Kellman, & Doidge, 1992). In another study, PDQ symptoms of each personality disorder were correlated with symptoms obtained from a diagnostic interview (rs between .24 and .55; Zimmerman & Coryell, 1990).

Revised Interpersonal Adjective Scales: Big Five Version (IASR-B5; Trapnell & Wiggins, 1990)

The IASR-B5 is a 124-item inventory of adjectives that assesses both the interpersonal circumplex and the Big Five models of personality. Each adjective is rated on a Likert-type scale ranging from 1 (extremely inaccurate) to 8 (extremely accurate). A total of 64 items make up the interpersonal circumplex (8 items per octant), which is anchored by dimensions of Dominance (M = 0.05, SD = 1.11) and Love (M = 0.29, SD = 1.22); there are 20 items on each index of the remaining Neuroticism (M = 4.44, SD = 0.96), Openness (M = 5.44, SD = 0.84), and Conscientiousness (M = 3.47, SD = 1.07) dimensions of the Big Five. Trapnell and Wiggins (1990) showed that each domain score from the IASR-B5 correlated about .7 with its corresponding domain score on the NEO-PI. In that study, the median absolute value of the correlations of each IASR-B5 domain with other NEO-PI domain scores was .14, suggesting that IASR-B5 scores exhibited good convergent and discriminant validity.

Procedure

Participants completed all questionnaires in a single session after providing informed consent. The order of administration of each questionnaire was randomly assigned. Participants were encouraged to ask for clarification if needed during the testing session, and a research assistant was present at all times during testing. Questionnaires were scored by computer after the session was complete.

Data Analysis

The replicability of the factor structures of the PPI, SRP-II, and APSD were explored with common factor analysis, followed by varimax rotation.¹ To best approximate simple structure (Thurstone, 1940), factor loadings above .4 were considered notable. Congruence coefficients (Gorsuch, 1983) were computed to determine the similarities between the obtained factor solutions and published factor solutions for each instrument. Procrustean rotations using the desired published loading matrix as the target matrix, which can be used profitably in lieu of confirmatory factor analysis (CFA) to confirm the factor structures of personality instruments (McCrae, Zonderman, Costa, Bond, & Paunonen, 1996), were also used to examine the congruence coefficients resulting from theory-guided factor rotations. We used SPSS version 11.0 for all analyses except the Procrustean factor analyses, which were conducted in Version 6.5 of Matlab.

We used exploratory factor analysis (EFA) instead of CFA for a number of reasons. First, although CFAs can suggest modifications of an a priori structure, they are less useful for suggesting entirely new factor structures for an inventory, as the results of EFA can. Hence, to maintain a consistent data analytical strategy across the results for the initial and revised factor structures for each instrument, we used exploratory analyses throughout this study. In addition, CFAs have failed to replicate the structures of well-validated normal-range personality inventories, such as the NEO-PI-R and MPQ, the structures of which are replicated through exploratory analyses (Church & Burke, 1994; McCrae et al., 1996). Furthermore, in studies of both normal-range (Church & Burke, 1994) and psychopathic (Lynam, Whiteside, & Jones, 1999) personality self-reports, CFAs often need to employ cross-loadings that are inconsistent with or contrary to the theoretical structures to provide an adequate fit to the data. Finally, confirmatory Procrustean rotations of EFA results can recover personality instrument structures that confirmatory analyses cannot (McCrae et al., 1996). For these reasons, we chose to use EFA (with varimax and Procrustean rotations) to achieve the goals of CFAs with a method that is more suitable for the peculiarities of personality self-reports.

The psychopathy factor scores used in the initial correlational and regression analyses were calculated from previously published factor structures.² For each correlational analysis, the difference between the factors of each instrument in their correlations with each variable was assessed statistically using Steiger’s (1980) formula for the test of differences between dependent correlations. To control for experiment-wise error rate, an alpha level of .005 was used to evaluate the statistical significance of all results. When the differences among genders and races in the correlations of psychopathy factors with criterion variables were analyzed with r-to-z tests, only 4 out of the 672 comparisons emerged as significant at the .005 level. Because this number of significant comparisons is near the number that would be expected by chance (3.32) at the alpha level set for this study, data were collapsed across gender and racial groups in all analyses reported below.

RESULTS

Psychopathy Measures: Initial Published Factor Scores

EFAs and congruence with prior work

Table 1 displays the results of common factor analyses, followed by varimax rotation, that were performed on the subscales of the PPI. As in Benning et al. (2003), when all factors with eigenvalues greater than 1 were extracted, three factors emerged (eigenvalues of 2.70, 1.88, and 1.10), explaining 71.0% of the covariance among PPI subscales. The first factor was recognizable as Impulsive Antisociality, the second as Fearless Dominance, and the third was marked by a single subscale, Coldheartedness (cf., Benning et al., 2003). When two factors were extracted, Impulsive Antisociality and Fearless Dominance emerged, respectively, with Coldheartedness failing to load appreciably on either factor. When Coldheartedness was omitted and all factors with eigenvalues greater than 1 were extracted, two factors were obtained (eigenvalues = 2.65 and 1.81), again resembling Impulsive Antisociality and Fearless Dominance, respectively, and explaining 63.8% of the total covariance among PPI subscales.³, ⁴ Factor congruence coefficients comparing these final factor loadings with those described in Benning et al. (2003) were .98 for Fearless Dominance and .99 for Impulsive Antisociality, demonstrating that the two factor structures are essentially equivalent. A Procrustean rotation yielded the same factor loading pattern with the same congruence coefficients (.98 for Fearless Dominance and .99 for Impulsive Antisociality).

TABLE 1.

Results of Common Factor Analyses of Psychopathic Personality Inventory (PPI) Subscales

	Eigenvalues > 1			Two Factors		Without Coldheartedness
PPI Subscale	A	B	C	A	B	A	B
Impulsive Nonconformity	.66	.43	−.04	.68	.38	.70	.37
Blame Externalization	.64	−.18	−.17	.64	−.23	.63	−.26
Machiavellian Egocentricity	.68	.14	.16	.68	.17	.67	.11
Carefree Nonplanfulness	.56	−.07	.18	.53	.00	.53	−.07
Fearlessness	.53	.60	−.03	.54	.54	.57	.55
Stress Immunity	−.31	.74	.28	−.31	.85	−.25	.77
Social Potency	.07	.62	.03	.09	.58	.11	.62
Coldheartedness	.11	.12	.89	.07	.29	—	—

NOTE: Factor loadings greater than .4 are in bold.

Because only one third of the total items in the SRP-II are used in computing factor scores, Table 2 gives the results of common factor analyses that were performed on the items of the SRP-II that Hare (1991) rationally assigned to each factor. Inspection of the scree plot suggested that a two-factor solution would be optimal; these two factors (eigenvalues 3.96 and 2.10) accounted for 27.6% of the covariance among SRP items. However, SRP-II Factor 1 did not emerge clearly from the varimax-rotated solution (congruence coefficient .61). Only three Factor 1 items had notable loadings on the appropriate factor from Hare’s (1991) rational factors, and four items from SRP-II Factor 1 had their strongest loading with SRP-II Factor 2 items. SRP-II Factor 2 appeared to emerge more clearly (congruence coefficient .76), with all a priori items from this factor having their strongest loading on this factor. Nevertheless, these congruence coefficients are below even the liberal minimum of .85 required to consider factors equivalent (Cooke & Michie, 2001). To examine whether rotation to a target factor matrix would improve the loading patterns of the SRP-II items, the extracted factors were subjected to a Procrustean rotation, using a unit-weighted item-loading matrix as the target matrix. Although the factor loadings did not change noticeably, the congruence coefficient for SRP-II Factor 2 (.93) exhibited substantial improvement; however, SRP-II Factor 1 still had a notably smaller congruence coefficient (.75) than would be considered indicative of factor invariance. This pattern of results is consistent with other factor analytic work on the SRP-II, which has found the SRP-II Factor 1 scale to be less reliable and replicable than the Factor 2 scale (Williams & Paulhus, 2004).

TABLE 2.

Results of Common Factor Analyses of Self-Report Psychopathy–II Scale Items

	Varimax		Procrustes
Item Number	A	B	A	B
6	−.17	.67	−.22	.67
10	−.09	.06	−.08	.09
19	.05	.28	.04	.26
25	.06	.59	.00	.62
30	.12	.37	.07	.32
31	.27	.03	.26	.09
47	−.16	.50	−.22	.50
53	.21	.11	.19	.16
60	.32	−.01	.31	−.01
2	.33	.06	.32	.06
7	.37	−.06	.38	−.03
11	.67	.06	.68	.16
17	.54	−.06	.53	−.01
18	.30	−.14	.32	−.10
23	.58	.16	.56	.20
28	.44	.05	.42	.07
29	.50	−.06	.49	−.06
32	.47	.17	.45	.20
38	.36	−.10	.36	−.05
39	.33	−.15	.34	−.14
50	.55	.04	.54	.09
58	.54	−.10	.55	−.03

NOTE: Factor loadings greater than .4 are in bold. Loadings less than .4 that represent the strongest loading of each item on either factor are in italics.

Table 3 displays the results of common factor analyses that were performed on the items of the APSD. Inspection of the scree plot suggested that either a three- (eigenvalues of 3.94, 1.89, and 1.49) or two-factor solution would be tenable. The three-factor solution explained 36.6% of the total covariance among APSD items and roughly corresponded with that found in Frick, Bodin, and Barry (2000) after varimax rotation. However, the congruence coefficients for the Narcissism (.83), Impulsivity (.57), and Callous-Unemotional (.78) scales were all below the most liberal guidelines for considering factors equivalent (Cooke & Michie, 2001). A Procrustean rotation using the 18 items in the community sample loading matrix in Frick et al. (2000) as the target matrix worsened the congruence coefficient for Impulsivity (.51) and did not notably improve the congruence coefficients for Narcissism (.83) or Callous-Unemotional traits (.79).⁵ Because of the low congruence for the Impulsivity factor, and because these low congruence coefficients mirrored those in Frick et al. (2000), the two-factor solution, which explained 29.1% of the covariance among APSD items, was examined next. After varimax rotation, two factors emerged that were roughly comparable to those in Frick et al. (1994). However, the congruence coefficients for Impulsive-Conduct problems (.83) and Callous-Unemotional traits (.67) were again below minimally acceptable values for equivalence to the original factors. A Procrustean rotation using the 16 items included in the loading matrix presented in Frick et al. (1994) as the target matrix did not notably improve the congruence coefficients for Impulsive-Conduct problems (.84) or Callous-Unemotional traits (.70).

TABLE 3.

Results of Common Factor Analyses of Antisocial Process Screening Device Items

	Three-Factor						Two-Factor
	Varimax			Procrustes			Varimax		Procrustes
Item Number	A	B	C	A	B	C	A	B	A	B
1	.51	.16	.07	.40	.08	.34	.48	.06	.49	.15
2	.09	.51	.05	—	—	—	.39	.09	.34	.10
3	.09	−.06	.47	−.01	.49	.07	.02	.46	−.02	.46
4	.38	.20	−.04	.21	.00	.38	.42	−.04	.42	.03
5	.31	.18	.21	.30	.22	.18	.35	.22	.29	.27
6	.18	.66	.17	—	—	—	.55	.21	—	—
7	−.19	−.10	−.15	−.12	−.14	−.15	−.21	−.15	—	—
8	.51	.06	.05	.35	.07	.35	.42	.04	.45	.12
9	.27	.24	−.06	.22	−.04	.27	.37	−.05	.34	−.01
10	.24	.52	.17	.96	.01	−.25	.52	.21	—	—
11	.32	.29	.05	.34	.07	.25	.44	.06	.43	.13
12	.02	.21	.54	.21	.50	−.06	.14	.55	.04	.53
13	.09	.58	−.07	.34	−.08	.13	.43	−.01	.38	−.01
14	.34	.38	.16	.38	.18	.28	.51	.18	.46	.25
15	.50	.15	−.11	.23	−.06	.51	.47	−.11	.53	−.03
16	.61	.11	.18	.39	.19	.43	.51	.16	.54	.24
17	.31	.01	−.01	.11	.00	.31	.24	−.02	—	—
18	.06	.03	.80	.14	.83	−.08	.04	.80	−.06	.85
19	.29	.26	.00	.28	.03	.25	.40	.01	.36	.08
20	−.05	.02	.28	.06	.22	−.10	−.03	.29	−.07	.23

NOTE: Factor loadings greater than .4 are in bold. Loadings less than .4 that represent the strongest loading of each item on any factor are in italics.

Notably, men had higher scores than women on each psychopathy factor from each instrument, all ts > 3.65, ps < .001. With the exception of Caucasian participants having higher mean scores on SRP-II Factor 2 than did African American participants, t(139) = 3.40, p < .001, there were no significant racial differences in mean psychopathy factor scores among Hispanic, African American, and Caucasian participants, ts < 2.45, ps > .015.

Psychopathy factor intercorrelations

Table 4 gives the intercorrelations among the previously published two-factor solutions of the three main measures of psychopathy in this sample. PPI Fearless Dominance related approximately equally to Factors 1 and 2 of the SRP-II. PPI Impulsive Antisociality related strongly to only Factor 2 of the SRP-II, as well as to both factors of the APSD. Factor 1 of the SRP-II was not related to either factor of the APSD, whereas Factor 2 of the SRP-II was related preferentially to the Impulsive-Conduct factor of the APSD. The Callous-Unemotional factor of the APSD was not related to Fearless Dominance; it related much more strongly to Impulsive Antisociality, SRP-II Factor 2, and APSD Impulsive-Conduct problems. The factors of the PPI had significantly different relations with the factors of the other two measures. The factors of the SRP-II differed significantly in their relations with Impulsive Antisociality and both APSD factors, and the APSD factors differed significantly in their relations with Impulsive Antisociality and SRP-II Factor 2. In sum, PPI Fearless Dominance and SRP-II Factor 1 highly related to each other, and PPI Impulsive Antisociality, SRP-II Factor 2, and the APSD factors all related similarly to each other.

TABLE 4.

Intercorrelations of Psychopathy Factor Scores

	PPI		SRP-II		APSD
Psychopathy Factor	Fearless Dominance	Impulsive Antisociality	Factor 1	Factor 2	Callous-Unemotional	Impulsive Conduct
PPI
Fearless Dominance	—	.21**	.50**	.46**	.08	.16*
Impulsive Antisociality	.21**	—	−.11c	.68**d	.49**e	.64**f
SRP-II
Factor 1	.50**a	−.11b	—	.06	.05	−.10
Factor 2	.46**a	.68**b	.06	—	.38**e	.61**f
APSD
Callous-Unemotional	.08a	.49**b	.05c	.38**d	—	.42**
Impulsive Conduct	.16*a	.64**b	−.10c	.61**d	.42**	—

NOTE: ns range from 315 to 326. PPI = Psychopathic Personality Inventory; SRP-II = Self-Report Psychopathy–II Scale; APSD = Antisocial Process Screening Device. Within each instrument’s factors, coefficients with the same subscript do not differ significantly from each other.

^*p < .005.

^**p < .001.

This pattern of intercorrelations suggests that there may be common higher-order factors underlying the factors of these different psychopathy scales. In a joint factor analysis of all six psychopathy factors, followed by varimax rotation, two broad factors with eigenvalues greater than 1 emerged (eigenvalues 2.76 and 1.48) that explained 70.7% of the covariance among psychopathy factor scales. The following scales loaded strongly on the first factor (with primary factor loadings in parentheses after each scale): PPI Impulsive Antisociality (.88), SRP-II Factor 2 (.79), APSD Callous-Unemotional traits (.53), and APSD Impulsive-Conduct problems (.76). The second factor comprised PPI Fearless Dominance (.89) and SRP-II Factor 1 (.57). No scale had a secondary loading greater than .3 on the other factor in this analysis.

Table 5 displays the correlations of the factors of each of the above-noted self-report psychopathy instruments with the subscales of the PAI-ANT scale. PPI Fearless Dominance and APSD Callous-Unemotional traits related approximately equally to the Antisocial behavior and Egocentricity facets of PAI-ANT, although PPI Fearless Dominance was related more strongly to the Sensation-Seeking facet of PAI-ANT than APSD Callous-Unemotional traits. SRP-II Factor 1 was not significantly related to any facet of the PAI-ANT. In contrast, SRP-II Factor 2, PPI Impulsive Antisociality, and APSD Impulsive-Conduct problems correlated significantly with all facets of PAI-ANT. The factors of the PPI differed significantly in their relations with ANT-A and ANT-E, whereas the APSD factors differed in their relations only with ANT-S. The factors of the SRP-II significantly differed in their relations with all PAI-ANT subscales. In summary, PPI Impulsive Antisociality, SRP-II Factor 2, and APSD Impulsive-Conduct problems were strongly related to all PAI-ANT scales; PPI Fearless Dominance and APSD Callous-Unemotional traits were less strongly related to PAI-ANT scale scores; and SPR-II Factor 1 was uncorrelated with any PAI-ANT scores.

TABLE 5.

Correlations of Psychopathy Factors With Personality Assessment Inventory (PAI) Antisocial (ANT) Subscales

	PPI		SRP-II		APSD
PAI Scale	Fearless Dominance	Impulsive Antisociality	Factor 1	Factor 2	Callous-Unemotional	Impulsive Conduct
ANT-A	.32**a	.55**b	.11c	.60**d	.33**	.39**
ANT-E	.25**a	.57**b	.11c	.53**d	.40**	.52**
ANT-S	.55**	.50**	.13c	.71**d	.25**e	.52**f

NOTE: ns range from 315 to 325. PPI = Psychopathic Personality Inventory; SRP-II = Self-Report Psychopathy–II Scale; APSD = Antisocial Process Screening Device; ANT-A = PAI-ANT–Antisocial Behavior; ANT-E = PAI-ANT-Egocentricity; ANT-S = PAI-ANT–Sensation Seeking. Within each instrument’s factors, coefficients with the same subscript do not differ significantly from each other.

^*p < .005.

^**p < .001.

Psychopathy factors and personality symptoms and traits

Table 6 shows the relationships among the various psychopathy factors and self-reported symptoms of DSM-IV (American Psychiatric Association, 1994) personality disorders. Both PPI Fearless Dominance and SRP-II Factor 1 were inversely related to symptoms of the anxious-fearful Cluster C personality disorders, along with paranoid and self-defeating personality features. In addition, SRP-II Factor 1 had significant inverse relations with schizotypal personality disorder and overall impairment, whereas PPI Fearless Dominance did not. Consistent with their relations with the PAI-ANT subscales, PPI Fearless Dominance was significantly related to APD symptoms, whereas SRP-II Factor 1 was not. In contrast, PPI Impulsive Antisociality, SRP-II Factor 2, and APSD Impulsive-Conduct problems all exhibited significant positive relations with symptoms of most personality disorders, and in addition, all were positively related to overall impairment. The Callous-Unemotional component of the APSD also showed positive associations with symptoms of nearly all personality disorders. With the exception of schizoid personality symptoms, the factors of the PPI differed significantly in their relations with PDQ-4+ variables, as did the two SRP-II factors. The APSD factors differed significantly in their relations with symptoms of schizoid, dependent, and passive/aggressive personality disorders as well as symptoms of all Cluster B personality disorders except APD.

TABLE 6.

Correlations of Psychopathy Factors With Personality Disorder Questionnaire–4+ (PDQ-4+) Symptoms

	PPI		SRP-II		APSD
PDQ-4+Scale	Fearless Dominance	Impulsive Antisociality	Factor 1	Factor 2	Callous-Unemotional	Impulsive Conduct
Schizoid	.03	.03	.13	.03	.20**e	−.06f
Schizotypal	−.06a	.32**b	−.28**c	.23**d	.24**	.24**
Paranoid	−.17*a	.26**b	−.27**c	.21**d	.28**	.24**
Avoidant	−.40**a	.21**b	−.39**c	.02d	.17*	.20**
Dependent	−.38**a	.28**b	−.47**c	.04d	.09e	.27**f
Obsessive-Compulsive	−.25**a	.17*b	−.31**c	.10d	.13	.18*
Histrionic	−.17*a	.33**b	−.37**c	.21**d	.11e	.42**f
Narcissistic	−.11a	.42**b	−.24**c	.33**d	.27**e	.49**f
Borderline	−.02a	.50**b	−.30**c	.46**d	.27**e	.49**f
Antisocial	.26**a	.54**b	−.02c	.61**d	.35**	.42**
Passive/Aggressive	−.17*a	.44**b	−.29**c	.25**d	.24**e	.45**f
Self-Defeating	−.25**a	.35**b	−.42**c	.14d	.18*	.29**
Sadistic	.14a	.42**b	.03c	.44**d	.34**	.47**
Impairment	−.12a	.53**b	−.27**c	.32**d	.35**	.44**
Fake Good	.05a	−.15b	.13c	−.15d	−.08e	.22**f
Fake Bad	.00a	.34**b	−.09c	.24**d	.16*	.17*

NOTE: ns range from 302 to 321. PPI = Psychopathic Personality Inventory; SRP-II = Self-Report Psychopathy–II Scale; APSD = Antisocial Process Screening Device. Within each instrument’s factors, coefficients with the same subscript do not differ significantly from each other.

^*p < .005.

^**p < .001.

Following Salekin et al. (2001), Figure 1 depicts the locations of each psychopathy factor in the interpersonal space defined by IASR-B5 dimensions of dominance and love. PPI Fearless Dominance and SRP-II Factor 1 occupied the assured-dominant octant of the interpersonal circumplex, whereas APSD Callous-Unemotional traits and PPI Impulsive Antisociality resided in the coldhearted octant. SRP-II Factor 2 and APSD Impulsive-Conduct problems were located in the arrogant-calculating octant. The factors of the PPI had the highest communalities (or relationships with the Dominance and Love axes combined) with the interpersonal circumplex, and the APSD factors had the lowest overall interpersonal communalities. SRP-II Factor 2 had a notably larger interpersonal communality than did SRP-II Factor 1. Table 7 displays the correlations of the various psychopathy with the Big Five personality dimensions in the IASR-B5, along with the regression coefficients for the Big Five dimensions as predictors of each psychopathy factor. PPI Fearless Dominance and SRP-II Factor 1 were related to high Dominance, low Neuroticism, and high Conscientiousness; PPI Fearless Dominance was additionally related to high Openness. APSD Callous-Unemotional traits were correlated only with low Love, although PPI Impulsive Antisociality, SRP-II Factor 2, and APSD Impulsive-Conduct problems were also strongly correlated with low Love. PPI Impulsive Antisociality was additionally related to high Neuroticism; SRP-II Factor 2 and APSD Impulsive-Conduct problems were instead correlated with high Dominance. The PPI factors differed in their relations with all Big Five dimensions, whereas the SRP-II factors differed in their relations with Neuroticism and Love. In the zero-order correlations, the APSD factors did not differentially relate to any Big Five dimensions.

FIGURE 1.

Projections of Initial Psychopathy Factors Onto the Interpersonal Circumplex

TABLE 7.

Correlations and Regression Weights of Revised Interpersonal Adjective Scales: Big Five Version (IASR-B5) Domains in Predicting Psychopathy Factor Scores (N = 276)

	PPI		SRP-II		APSD
PAI Scale	Fearless Dominance	Impulsive Antisociality	Factor 1	Factor 2	Callous-Unemotional	Impulsive Conduct
IASR-B5 Domain
Neuroticism	−.42**a	.16*b	−.52**c	.01d	.02	.15
Dominance	.61**a	.10b	.39**	.38**	.04	.20**
Openness	.36**a	−.05b	.06	.14	−.10	−.02
Love	−.13a	−.54**b	−.11c	−.40**d	−.40**	−.40**
Conscientiousness	.18*a	−.07b	.23**	.02	.05	.07
Regression weights
Constant	.55	−1.1	48**	35**	3.3**	3.8
Neuroticism	−.89 (−.36)**	.41 (.14)	−3.1 (−.45)**	.52 (.04)	.02 (.01)	.56 (.19)*
Dominance	.91 (.44)**	.20 (.08)	1.7 (.29)**	3.6 (.33)**	.00 (.00)	.53 (.21)*
Openness	.63 (.22)**	.01 (.00)	−.30 (−.04)	.97 (.06)	−.11 (−.05)	−.20 (−.06)
Love	−.27 (−.14)*	−1.2 (−.53)**	−.54 (−.10)	−3.8 (−.37)**	−.60 (−.39)**	−.86 (−.36)**
Conscientiousness	.01 (.01)	−.14 (−.05)	.58 (.09)	−.37 (−.03)	.11 (.06)	.21 (.08)

NOTE: PPI = Psychopathic Personality Inventory; SRP-II = Self-Report Psychopathy–II Scale; APSD = Antisocial Process Screening Device. Standardized regression weights are presented in parentheses. Within each instrument’s factors, coefficients with the same subscript do not differ significantly from each other.

^*p < .005.

^**p < .001.

In the regression analyses, PPI Fearless Dominance (R = .72) and SRP-II Factor 1 (R = .61) were predicted better by the IASR-B5 domains than were APSD Callous-Unemotional traits (R = .41), PPI Impulsive Antisociality (R = .57), SRP-II Factor 2 (R = .53), or APSD Impulsive-Conduct scores (R = .47). PPI Fearless Dominance was significantly predicted by low Neuroticism, high Dominance, high Openness, and low Love; SRP-II Factor 1 was significantly predicted by low Neuroticism and high Dominance; and APSD Callous-Unemotional traits were significantly predicted by low Love, as was PPI Impulsive Antisociality. SRP-II Factor 2 was significantly predicted by low Love and high Dominance, which with high Neuroticism were also the significant predictors of APSD Impulsive-Conduct problems.

Psychopathy Measures: Revised Factor Scores

Because the factor analytic results suggested ways in which the structures of these instruments might be improved, we conducted additional analyses to create more stable and interpretable factor structures for each instrument. We used the results from the initial EFAs as the starting points for purifying the structures of the PPI and the APSD. However, the factor analytic results for the SRP-II did not use all available items from the instrument to establish its factor structure. Hence, we used the correlations of each SRP-II item with the interpersonal circumplex as an additional starting point for selecting items that would form more coherent, theoretically meaningful psychopathy factors. We did not calculate Procrustes rotations or congruence coefficients for the resulting factor solutions, because there are no prior published factor loading matrices against which to compare those in the present study. Although these analyses may capitalize on chance in the current sample and likely require replication in other samples before the modifications proposed may be used with confidence in other samples, they represent at least a first step in refining the factor structures of these instruments.

PPI Fearlessness and Fearless Dominance

Because the Fearlessness subscale of the PPI had a secondary loading on Impulsive Antisociality in this study, a result that has been evident to a lesser degree in other studies (Benning et al., 2003; Patrick, Edens, Poythress, Lilienfeld, & Benning, 2004; Ross et al., 2004), an exploratory common factor analysis followed by varimax rotation was performed on the 19 items that compose PPI Fearlessness to clarify the factor structure of this scale. Of the three factors that emerged (eigenvalues of 5.64, 1.49, and 1.23), two loaded only on Fearless Dominance in subsequent factor analyses. The items composing these factors (Numbers 19, 26, 59, 111, 118, 142, and 181 on the first factor; Numbers 5, 42, and 98 on the second factor) were combined into a revised Fearlessness scale (Fearlessness-REV, α = .81). When Fearlessness-REV was entered into a common factor analysis with the other PPI subscales, followed by varimax rotation, Fearlessness-REV loaded more strongly on Fearless Dominance (.57) than Impulsive Antisociality (.32), suggesting that Fearlessness-REV is a purer marker of Fearless Dominance than the original Fearlessness scale. Using the Fearlessness-REV scale in computing Fearless Dominance scores resulted in a smaller, non-significant correlation of Fearless Dominance with Impulsive Antisociality; this revised Fearless Dominance score also correlated .99 with initial Fearless Dominance scores. Not surprisingly, the pattern of criterion variable associations described above for Fearless Dominance was nearly identical for Fearless Dominance scores calculated with the Fearlessness-REV scale. Therefore, we do not report additional analyses for the revised Fearless Dominance factor alone.

Revised SRP-II factors

Because the factor structure of the SRP-II has proven particularly difficult to replicate using exploratory means alone (Williams & Paulhus, 2004), we used analytical and theoretical considerations to generate new SRP-II factors. First, we dropped all SRP-II Factor 1 items that did not load at least .30 on Factor 1 in the initial exploratory analysis (namely, Items 10, 19, 31, 53, and 60). To create a new set of Factor 1 items, we included items that correlated at least .30 with original SRP-II Factor 1 scores and less than .15 with original SRP-II Factor₂ scores; these items were 12, 15, and 19. Because SRP-II Factor 1 was located in the assured-dominant octant of the interpersonal circumplex, additional Factor 1 items were included if they correlated at least .30 with Dominance and less than .15 with both Love and scores for the arrogant-calculating octant. These items included 35 and 52. The resulting revised SRP-II Factor 1 contains nine items, with an internal consistency of .74 (M = 39.9, SD = 9.24). Scores on this factor correlated .78 with original SRP-II Factor 1 scores.

The construct embodied in SRP-II Factor 2 seems to have drifted into the arrogant-calculating octant (see Figure 1), compared with other constructs designed to measure the more overtly antisocial features in psychopathy, such as PCL-R Factor 2 and Impulsive Antisociality in the PPI, which reside along the low Love axis of the interpersonal circumplex. To remedy this drift, items in SRP-II Factor 2 that correlated more than .25 with interpersonal arrogant-calculating octant scores and that correlated at least .10 more with arrogant-calculating than low Love scores were dropped; these included Items 2, 11, 17, 23, 32, 50, and 58. Next, Items 20, 37, 57, and 60, which correlated more than .20 with low Love scores and .20 or less with arrogant-calculating scores, were added to SRP-II Factor 2. To flesh out this 10-item SRP-II Factor 2 scale, items correlating more than .30 with the 10-item total score—namely, Items 16, 21, and 40—were added. Finally, because Item 45 had its strongest relationship with the coldhearted octant of all the interpersonal octants, it was added to the revised SRP-II Factor 2. This 14-item factor had an internal consistency of .73 (M = 31.9, SD = 10.9), and scores on this factor correlated .76 with scores on the original SRP-II Factor 2.

When the 23 items from the revised SRP-II factors were entered into a common factor analysis, followed by varimax rotation, inspection of the scree plot revealed two strong first factors (first five eigenvalues = 3.41, 3.14, 1.77, 1.30, and 1.22). All of the items from the revised SRP-II Factor 1 emerged with primary loadings above .3 on the second factor of this analysis (median primary loading = .49), and no item had a secondary loading greater than .25 on the first factor (median secondary loading = .06). Likewise, all revised SRP-II Factor 2 items had their strongest loading on the first factor of this analysis, with all items having a loading of at least .25 (median primary loading = .40) and no item having a secondary loading above .20 on the second factor (median secondary loading = .07). Notably, these factors were uncorrelated, even after a promax rotation (r = .01). Because the original and revised SRP-II factors shared only about 60% of their variance, the analyses performed for the original SRP-II factors are repeated below using the revised SRP-II factors described above.

Modifying the APSD factors

Unfortunately, the factors of the APSD do not come from a larger item pool, as do the factors of the SRP-II, nor are a sufficient number of items represented in each of the factors to accommodate substantial item deletions, as was done for PPI Fearlessness. Furthermore, the two-factor solution in the APSD would be difficult to refine because only four items have primary loadings on the Callous-Unemotional factor in the current study, compared with the 16 items with primary loadings on the Impulsive-Conduct factor. However, the three-factor structure of the APSD may be improved in this study because a number of items loaded on each of the three factors in the initial factor analyses. Therefore, we modified the factors of the three-factor structure of the APSD in an attempt to refine their psychometric structure and external correlates.

From the exploratory analyses, Items 1, 8, 15, and 16 loaded strongly together; in addition, Items 4 and 5 loaded above .3 on this factor and had loadings of .2 or less on the other two factors. These items were combined to yield an Impulsive Narcissism factor (internal consistency = .66; M = 2.79, SD = 2.05), because Items 1 and 4 loaded on the Impulsivity factor in Frick et al.’s (2000) three-factor structure of the APSD, and Items 5, 8, 15, and 16 loaded on the Narcissism factor. Items 2, 6, 10, and 13 loaded strongly on the second factor, which we named Deceitful Risk Taking (internal consistency = .68; M = 1.71, SD = 1.62), because these items index a predilection toward illegal activities, lying, conning, and taking risks. The third factor featured strong loadings from Items 3, 12, and 18 (internal consistency = .63; M = 1.02, SD = 1.28), making it a reduced set of items from the Callous-Unemotional factor in Frick et al. (2000).

Revised psychopathy factor intercorrelations

In all analyses involving these revised psychopathy factors, rs greater than .15 in absolute magnitude are significant. As shown in Table 8, the revised SRP-II factors show better convergent and discriminant validity with respect to the factors of the PPI than the original SRP-II factors. Revised SRP-II Factor 1 was much more strongly related to PPI Fearless Dominance than PPI Impulsive Antisociality, and revised SRP-II Factor 2 displayed strong, positive relations only with PPI Impulsive Antisociality. However, all three revised factors of the APSD correlated more strongly with PPI Impulsive Antisociality and revised SRP-II Factor 2 than they did with PPI Fearless Dominance or revised SRP-II Factor 1. Interestingly, the revised Callous-Unemotional traits scale correlated weakly with the other two APSD factors, suggesting that it measured unique variance in the psychopathy construct not represented in PPI Fearless Dominance or revised SRP-II Factor 1. Of the three APSD factors, only Deceitful Risk Taking had a significant relationship with PPI Fearless Dominance, although it did not correlate significantly with revised SRP-II Factor 1.

TABLE 8.

Intercorrelations of Revised Psychopathy Factor Scores

	PPI		SRP-II		APSD
Psychopathy Factor	Revised Fearless Dominance	Impulsive Antisociality	Revised Factor 1	Revised Factor 2	Revised Callous-Unemotional	Callous-Deceitful Risk Taking	Impulsive Narcissism
PPI
Revised Fearless Dominance	—	.14	.57**c	.24**d	.03e	.37**f	.00e
Impulsive Antisociality	.14**	—	−.24c	.69**d	.26**e	.55**f	.53**f
SRP-II Revised
Factor 1	.57**a	−.24b**	—	−.04	−.02	.03	−.13
Factor 2	.24**a	.69**b	−.04	—	.26**e	.53**f	.43**e, f
APSD
Revised Callous-Unemotional	.03a	.26**b	−.02c	.26**d	—	.18**	.13
Deceitful Risk Taking	.37**a	.55**b	.03c	.53**d	.18**	—	.36**
Impulsive Narcissism	.00a	.53**b	−.13c	.43**d	.13e	.36**f	—

NOTE: ns range from 322 to 325. PPI = Psychopathic Personality Inventory; SRP-II = Self-Report Psychopathy–II Scale; APSD = Antisocial Process Screening Device. Within each instrument’s factors, coefficients with the same subscript do not differ significantly from each other.

^*p < .005.

^**p < .001.

The revised SRP-II factor scales had largely the same relationships with the PAI-ANT scales as the original factors, with revised SRP-II Factor 2 correlating strongly with each PAI-ANT scale (rs from .49 to .65) and revised SRP-II Factor 1 correlating negligibly with each PAI-ANT scale (rs from −.01 to .10). The APSD Deceitful Risk Taking scale had a profile of PAI-ANT scale correlations similar to that of the original APSD Impulsive-Conduct Problems scale, because all of the correlations with the PAI-ANT scales were large and strong (rs from .54 to .69). The pattern of correlations for the APSD Impulsive Narcissism scale was similar to that for the original APSD Callous-Unemotional Traits scale, with the strongest correlation obtained for ANT-E (r = .48) and approximately equal correlations for ANT-A (r = .23) and ANT-S (r = .28). Curiously, the revised APSD Callous-Unemotional Traits scale had relatively weak correlations with all of the PAI-ANT scales; its only significant correlation was with the overtly antisocial ANT-A scale (r = .23).

Revised psychopathy factors and personality symptoms and traits

Like the original SRP-II Factor 1 scale, the revised SRP-II Factor 1 acted as a reversed marker of most PDQ-4+ personality disorder symptoms (significant rs from −.52 to −.29). It also had nonsignificant associations with sadistic, antisocial, and schizoid personality disorder symptoms. Revised SRP-II Factor 2 demonstrated relationships with PDQ-4+ personality disorder symptoms similar to those for the original SRP-II Factor 2—that is, significant correlations with all Cluster B disorders (rs from .23 to .66) and with paranoid (r = .19) and schizotypal (r = .23) Cluster A disorders. Similar to APSD Impulsive-Conduct problems, APSD Impulsive Narcissism showed positive correlations with nearly all of the PDQ-4+ symptom clusters (significant rs from .18 to .48; r = −.07 with schizoid symptoms). However, the relations observed for the original APSD Callous-Unemotional Traits scale separated across the remaining revised APSD scales. APSD Deceitful Risk Taking exhibited more selective relationships with symptoms of the Cluster B disorders (rs from .24 to .56; except for histrionic personality disorder, r = .08) and with sadistic personality disorder (r = .48). The revised APSD Callous-Unemotional Traits scale showed significant relationships with symptoms of Cluster A disorders (rs from .17 to .20) as well as antisocial (r = .25) and sadistic (r = .18) personality disorder symptoms.

Figure 2 diagrams the positions of the revised psychopathy factors within the interpersonal circumplex. PPI Impulsive Antisociality was joined by revised SRP-II Factor 2, revised APSD Callous-Unemotional Traits, and APSD Impulsive Narcissism in or at the border of the coldhearted octant. Revised PPI Fearless Dominance and revised SRP-II Factor 1 both resided in the assured-dominant octant, and APSD Deceitful Risk Taking was situated in the arrogant-calculating. With respect to the IASR-B5 scales, the revised SRP-II factors had normal-range personality correlates similar to those for the original SRP-II factors, except that revised SRP-II Factor 1 was also significantly related to Openness (r = .20), and revised SRP-II Factor 2 was much less strongly related to Dominance (r = .19). Again, APSD Impulsive Narcissism had personality correlates similar to those of original APSD Impulsive-Conduct problems. The revised APSD Callous-Unemotional Traits scale was less strongly related to low Love (r = −.23) than the original APSD Callous-Unemotional Traits scale, although low Love was still its only significant normal-range personality correlate. As might be expected from its conning, manipulative, and sensation-seeking item content, APSD Deceitful Risk Taking correlated with Dominance (r = .35) and Love (r = −.37) approximately equally.

FIGURE 2.

Projections of the Revised Psychopathy Factors Onto the Interpersonal Circumplex

The significant predictors of revised SRP-II Factor 1 (R = .73) and Factor 2 (R = .52) were unchanged from the original factors, except that SRP-II Factor 2 was no longer predicted by high Dominance (β = .14). Likewise, the predictors of revised APSD Callous-Unemotional traits (R = .24) and APSD Impulsive Narcissism (R = .45) were similar to those of the original APSD Callous-Unemotional traits and Impulsive-Conduct problems scales, respectively. Consistent with its zero-order correlations with the IASR-B5 scales, the only significant predictors of APSD Deceitful Risk Taking (R = .49) were high Dominance (β = .30) and low Love (β = −.35).

DISCUSSION

In this study, the two-factor structure of the PPI was recovered from an EFA followed by varimax rotation with a high degree of congruence with the loadings presented in Benning et al. (2003). Factor 2 of the SRP-II achieved congruence with the structure reported in Hare (1991) after a Procrustean rotation, although SRP-II Factor 1 did not. The factors of the APSD in this sample were not congruent with those presented in Frick et al. (1994, 2000), even after Procrustean rotations. Fearless Dominance shared approximately 25% of its variance with both SRP-II factors; and Impulsive Antisociality, SRP-II Factor 2, and both APSD factors shared 17% to 46% of their variance. Contrary to prediction, SRP-II Factor 1 scores were not related to any facets of psychopathy as measured by PAI-ANT or antisocial personality disorder symptoms; however, following predictions, the remaining factors correlated to varying degrees with all PAI-ANT subscales and antisocial personality disorder symptoms.

Consistent with our hypotheses, PPI Fearless Dominance and SRP-II Factor 1 were negatively related to symptoms of Cluster C personality disorders on the PDQ-4+. However, these psychopathy factors were also negatively related to symptoms of other Cluster B personality disorders, which conflicted with our hypotheses. In line with predictions, Impulsive Antisociality, SRP-II Factor 2, and APSD Impulsive-Conduct problems were positively related to Cluster B personality disorder symptoms and impairment. However, we did not initially predict that these factors would be positively related to nearly all other personality disorder symptoms. Furthermore, we found that APSD Callous-Unemotional traits were significantly related to symptoms of all Cluster A and B personality disorders (including schizoid personality disorder, consistent with our prediction). Within the interpersonal version of the Big Five, consistent with our hypotheses, Fearless Dominance and SRP-II Factor 1 were related to high Dominance and low Neuroticism. Counter to our hypotheses, SRP-II Factor 1 was not significantly predicted by low Love, although PPI Fearless Dominance was. Also following our predictions, Impulsive Antisociality, SRP-II Factor 2, APSD Impulsive-Conduct problems were related to low Love. In addition, APSD Callous-Unemotional traits were related only to low Love.

Revisions of each instrument’s structures yielded factors that were largely similar to the original factors, although revised SRP-II Factor 2 was related only to low Love in the interpersonal circumplex. The APSD broke into three factors, and the revised APSD Callous-Unemotional traits scale had weaker relationships with external criterion variables than did the original scale. In contrast, the APSD Impulsive Narcissism factor had patterns of correlations that were highly similar to those for the original APSD Impulsive-Conduct Problems scale, whereas the APSD Deceitful Risk Taking scale represented a new factor that had specific relationships with Cluster B personality disorders and equal relations with Dominance and Love in the interpersonal circumplex.

Intrapersonal and Interpersonal Differences Among Psychopathy Factor Constructs

Because of their negative correlations with symptoms of the anxious-fearful personality disorders and their negative correlations with Neuroticism, PPI Fearless Dominance and SRP-II Factor 1 both appear to index the “absence of ‘nervousness’ or other psychoneurotic manifestations” that Cleckley (1941/1988) viewed as one of the hallmarks of psychopathy. In contrast, APSD Callous-Unemotional traits seem to index a factor of general psychological distress, as evidenced by its positive manifold of correlations with personality disorder symptoms. Scores on PPI Fearless Dominance were specifically related to antisocial personality disorder symptoms and were predicted by low Love, suggesting that despite their relatively weak intercorrelation, both factors of the PPI measure some of the antisocial deviance that makes psychopathy a clinical and forensic problem. Consistent with its lower communality with the interpersonal circumplex dimensions and its not being predicted by low Love, SRP-II Factor 1 did not exhibit these relations.

Contrary to prediction, neither PPI Fearless Dominance nor SRP-II Factor 1 was positively related to narcissistic personality symptoms, and SRP-II Factor 1 was not related to PAI ANT-E scores. However, regression analyses describing the Big Five structures of PDQ-4+ narcissism and PAI ANT-E scores clarified these findings. The only significant Big Five predictors of PDQ-4+ narcissism were high Neuroticism (β = .35) and low Love (β = −.25), indicating that PDQ-4+ narcissism lacks the primary component of pathological narcissism: high Dominance, which is what binds narcissism to the core personality features of psychopathy (Benning et al., 2005; Hart & Hare, 1989). Similarly, although high Dominance emerged as a significant predictor of PAI ANT-E, low Love was more than doubly important in predicting PAI ANT-E scores (βs = .19 and .41, respectively), which would reduce substantially the expected relations of PAI ANT-E with PPI Fearless Dominance and SRP-II Factor 1.

Curiously, although the SRP-II and APSD were designed as alternative measures of the PCL-R, their factors appear to measure rather different constructs, perhaps because their target assessment populations differ. Within the interpersonal circumplex, the SRP-II appears to measure a construct associated more with the Dominance axis of the interpersonal circumplex (Salekin et al., 2001) than with the PCL-R (Harpur et al., 1989). SRP-II Factor 1 resides in the assured-dominant octant, whereas PCL-R Factor 1 occupies the arrogant-calculating octant (Harpur et al., 1989). Likewise, SRP-II Factor 2 projected onto the arrogant-calculating octant in this study, whereas PCL-R Factor 2 projects onto the middle of the coldhearted octant (Harpur et al., 1989). The implication is that total SRP-II scores have migrated from the low Love side of the arrogant-calculating octant to its high Dominance side (Salekin et al., 2001). Conversely, the constructs of the APSD appear to be more heavily directed toward the Love axis than the PCL-R. In particular, the Callous-Unemotional traits in the APSD appear to have moved from the arrogant-calculating octant into the middle of the coldhearted octant. However, the Impulsive-Conduct Problems scale appears to have moved from the cold-hearted octant that PCL-R Factor 2 occupies into the arrogant-calculating octant, where one might expect the Callous-Unemotional Traits scale to reside.

Finally, although different versions of the same psychopathy factors largely demonstrated better agreement among each other than did the psychopathy measures considered by Hare (1985), consistent with the factors sharing similar places within the nomological network, it is clear that the constructs that are measured by each instrument are not isomorphic. It is unclear whether the variance that is not shared among similar factors essentially represents measurement error or whether it instead reflects substantive divergence in the constructs measured by each instrument. In addition, two of the three psychopathy measures had null or minimal factor intercorrelations. Although the independence of these factors replicates prior work on these instruments (Benning et al., 2003; Williams & Paulhus, 2004), these results raise questions as to the nature of the higher order psychopathy construct that these instruments were designed to measure. Alternatively, these results may indicate that these self-report measures are poor measures of the higher order psychopathy constructs embodied in widely accepted clinician ratings of psychopathy, such as the PCL-R and its derivatives. Indeed, there are studies suggesting that self-report measures of psychopathy do not capture certain factors (Edens et al., 2000; Murrie & Cornell, 2002) or much of the total scores (Lee, Vincent, Hart, & Corrado, 2003) of these well-validated measures of psychopathy.

Adaptive and Maladaptive Characteristics in Psychopathy

Because the constructs embodied in the PPI factors are uncorrelated in both community and incarcerated samples (Benning et al., 2005), it seems more likely that the lack of correlation between the factors of the PPI and the SRP-II is driven by divergences in construct content rather than by peculiarities of this sample. One reason that the factors of the PPI and SRP-II may be uncorrelated is that they may measure adaptive and maladaptive features of psychopathy. In this study, the factors of the PPI and the SRP-II showed opposing relations with the Impairment scale of the PDQ-4+, along with its fake good and bad scales. In each case, PPI Fearless Dominance and SRP-II Factor 1 related negatively to impairment and faking bad, and they related positively to faking good, implying that they may be markers of adaptive psychological functioning. In contrast, PPI Impulsive Antisociality and SRP-II Factor 2 related positively to impairment and faking bad and negatively with faking good, implying that they are likely markers of maladaptive psychological functioning. As Krueger, Hicks, and McGue (2001) found that adaptive prosocial and maladaptive antisocial behaviors were orthogonal to each other and had unique genetic etiologies and personality correlates, these results suggest that the adaptive and maladaptive personality traits in psychopathy specifically are orthogonal to each other, perhaps indicating that they have separate etiologies. Also, this interpretation implies that when both factors have maladaptive antisocial content, the factors will correlate with each other.

This prediction is borne out when the two factors of the APSD are examined with respect to their maladaptive content. Personality characteristics similar to the low Love that characterized the two main factors of the APSD are related to persistent, maladaptive criminal behavior (Krueger et al., 1994) and externalizing disorders (Krueger, Caspi, Moffitt, Silva, & McGee, 1996). Hence, it seems likely that the factors of the APSD assessed two facets of the maladaptive personality features of psychopathy in this undergraduate sample without assessing many of the adaptive features of psychopathy. Even the Deceitful Risk Taking construct from revised three-factor structure of the APSD appears to have substantial mal-adaptive implications, based on its significant relationship to the Negative Self-Presentation and Impairment scales on the PDQ-4+. However, it should also be remembered that the APSD was originally designed for use in children as an observer rating in samples with substantial base rates of antisocial behavior rather than as a self-report in undergraduate samples. Therefore, although the factorial concerns that have been raised in these results parallel those in Frick et al. (1994, 2000), they may be of limited applicability to the target population for this instrument. The results of the current study also call into question the validity of the APSD as a measure of psychopathy in undergraduates because it may assess only the maladaptive features of psychopathy in this population.

Nevertheless, the revised factor structure of the APSD provided clues as to what maladaptive features PPI Fearless Dominance and SRP-II Factor 1 lack that have been considered important in psychopathy. In PPI Fearless Dominance and SRP-II Factor 1, there are a number of items that index a person’s ability to influence others’ opinions or behavior through persuasion or other socially acceptable means that lie solely along the agentic Dominance axis of the interpersonal circumplex. In contrast, the APSD Deceitful Risk Taking factor indexes less socially acceptable ways of influencing others through manipulative conning and lying—such as the pathological lying/conning-manipulative testlet in Cooke and Michie’s (2001) work on the PCL-R and Levenson, Kiehl, and Fitzpatrick’s (1995) Primary Psychopathy scale—and lies between the high Dominance and low Love axes of the interpersonal circumplex. The most aggressive and least socially acceptable ways of influencing others—which include PPI Impulsive Antisociality, SRP-II Factor 2, and APSD Impulsive-Conduct problems and which employ outright confrontation instead of sly manipulation—are located solely along the low Love axis of the interpersonal circumplex. Thus, as the constructs involved in psychopathy become less socially acceptable, they seem to shift from high Dominance to low Love in the interpersonal circumplex. This interpretation implies that the manipulative and deceitful content of PCL-R Factor 1, when combined with its mixture of features that may be associated exclusively with either high Dominance (e.g., glibness and superficial charm, grandiosity) or low Love (e.g., callousness, lack of remorse), would be responsible for anchoring it in the arrogant-calculating octant of the interpersonal circumplex.

Limitations and Future Directions

Although the current study investigated the construct validity of a number of self-report psychopathy measures, Hare (1985) noted that a substantial method factor separated clinical ratings and self-report measures of psychopathy. Future studies should include examination of the validity of these factors of psychopathy compared with clinical ratings of similar two-factor measures. Including clinician-rated psychopathy measures would allow comparison of these factors of psychopathy to those in the PCL:SV (for nonincarcerated samples) or PCL-R (for incarcerated samples) and would also permit an examination of whether self-reports and clinician ratings of psychopathy factors would share enough variance to segregate into conceptual factors within the nomological network of psychopathy or whether they would instead separate into methodological factors. Furthermore, the construct validity of these self-report psychopathy factors should be examined with interview-based diagnostic symptoms of personality disorders, because there are indications in the literature that the PDQ-4+ may not be a suitable replacement for a structured clinical interview (Fossati et al., 1998, Wilberg, Dammen, & Friis, 2000; but see Davison, Leese, & Taylor, 2001).

Finally, this sample consisted entirely of college undergraduates. Although this work is useful for understanding the facets of psychopathy within a noninstitutionalized sample, future work should incorporate incarcerated participants to examine how these self-reported factors of psychopathy perform in the population in which most psychopathy research has been performed. Furthermore, despite the similarities of the psychometric correlates between groups in this study, it should be noted that there may be interactions in laboratory task performance between members of different genders (cf. Vitale, Smith, Brinkley, & Newman, 2002) or racial groups (cf. Brinkley, Schmitt, Smith, & Newman, 2001) and psychopathy factor scores. Our conservative criterion used to detect significant effects in this study necessarily reduced our power in detecting them. Furthermore, the same scores on psychopathic traits may have different implications for social functioning (which was not assessed in this study) across genders and cultures. For example, in societies mirroring conditions found in post–World War II United States, the dominance and assertiveness that are hallmarks of psychopathy may be perceived as more socially adaptive for men and less so for women (Twenge, 2001). Likewise, the selfish and manipulative behavior that individuals with psychopathy often display may lead to their being more ostracized and socially disadvantaged within collectivist cultures, compared to cultures that are more individualistic (Oyserman, Coon, & Kemmelmeier, 2002).

Nevertheless, the divergent relationships of these factors of psychopathy to criterion variables in this study imply that people high on each of the factors of these instruments may show unique behavioral and emotional deficits in the laboratory. For example, consistent with their fearlessness and dominant interpersonal presentation, those high in PPI Fearless Dominance and SRP-II Factor 1 may show deficits in fear-potentiated startle (cf. Benning, 2004) and amygdala activity (cf. Gordon, Baird, & End, 2004), and they may be less reactive to social stressor tasks, such as giving a self-critical speech in the laboratory. In contrast, consistent with research on similar psychopathy factors strongly associated with low Love or Agreeableness (Brinkley et al., 2001; Lynam et al., 1999), those high in PPI Impulsive Antisociality, SRP-II Factor 2, or APSD Impulsive-Conduct problems may show deficits in passive avoidance learning and other behavioral tasks. Hence, the results of this study imply that examining the constructs embodied in these psychopathy instruments separately will aid researchers in understanding the mechanisms that underlie each factor of this perplexing disorder.

Biographies

Stephen D. Benning, M.A., is a doctoral candidate in the Clinical Science and Psychopathology Research program at the University of Minnesota at Twin Cities. His research interests include the structure of emotion and psychopathy in addition to psychophysiological measures of appetitive processing and positive emotionality.

Christopher J. Patrick, Ph.D., is the Starke R. Hathaway Distinguished Professor of Psychology at the University of Minnesota at Twin Cities, where he serves as director of the Clinical Science and Psychopathology Research Training Program and as director of the National Institute of Mental Health Neurobehavioral Training Program. His research interests include the affective and cognitive neuroscience of psychopathy and externalizing psychopathology.

Randall T. Salekin, Ph.D., is an associate professor in the Department of Psychology at the University of Alabama. He received his Ph.D. at the University of North Texas and completed his internship at Yale University. His research interests include understanding the development of psychopathic traits and the potential negative outcomes the syndrome has for society.

Anne-Marie R. Leistico, M.A., is a clinical child psychology graduate student at the Univeristy of Alabama. Her research interests include the assessment of psychopathic traits and risk for dangerousness.