Generalizability of Psychopathy Network Analysis Findings to Scores Assigned to Individuals Convicted of a Sex Offense

Abstract

Despite being well-researched, the conceptualization of psychopathy incites much debate within the field. Results from network analysis can inform these debates by graphically and quantitatively depicting the core characteristics of the construct of psychopathy. Existing network studies with Psychopathy Checklist—Revised (PCL-R) scores suggest the affective traits are most central to the construct of psychopathy, but more studies are needed. The current study examined network models developed using data from risk assessments of individuals convicted of a sex offense (N = 615). Findings corroborate some aspects of previous network studies in that affective features were most central to the construct and antisocial traits were least central, but there were instances of traits with notably higher centrality (e.g., Pathological Lying, Need for Stimulation, and Impulsivity) or lower centrality (e.g., Shallow Affect) than in prior research, suggesting that trait centrality may vary depending on the sample and evaluation context.

Keywords

psychopathy PCL-R network analysis field study individuals convicted of a sex offense

Although there are various conceptualizations of psychopathy and many measures for assessing psychopathic traits (see e.g., Drislane et al., 2018; Lilienfeld et al., 2012; Patrick, 2019; Skeem & Cooke, 2010), forensic practitioners rely extensively on one of these measures—the Psychopathy Checklist–Revised (PCL-R; Hare, 1991, 2003)—for assessing psychopathic traits in the field (Boccaccini et al., 2017; Neal & Grisso, 2014; Viljoen et al., 2010). The 20-item, clinician-scored PCL-R organizes psychopathic traits into a two-factor model underpinned by four facets: Affective, Interpersonal, Lifestyle, and Antisocial (Hare, 2003; see Table 1 for a list of PCL-R items). Although the affective component is widely considered to be at the core of psychopathy (Lynam & Miller, 2015; Verschuere & te Kaat, 2019), the more antisocial and behavioral traits assessed by the PCL-R (e.g., Poor Behavioral Controls, Early Behavioral Problems, Juvenile Delinquency, and Revocation of Conditional Release) stir much debate. Some scholars argue that antisocial behaviors are central to the construct of psychopathy based on both psychometric and developmental psychopathology research (Hare & Neumann, 2010). Others argue that antisocial behavior is merely a “downstream correlate” of other more important psychopathy components (Skeem & Cooke, 2010, p. 443). The importance of other components of the PCL-R, such as impulsivity, has also been questioned due to their lack of universal application to all individuals with psychopathic traits (Poythress & Hall, 2011).

Table 1:

Descriptive Statistics for PCL-R Facet and Item Scores

PCL-R Score	M	SD	Cronbach’s α
Facet 1: Interpersonal	5.52	1.97	.77
1. Superficial charm	0.78	0.77
2. Grandiose	0.75	0.78
4. Pathological lying	0.93	0.74
5. Manipulative	1.04	0.79
Facet 2: Affective	5.30	1.30	.75
6. Lack of remorse	1.38	0.69
7. Shallow affect	1.00	0.73
8. Lack of empathy	1.17	0.75
16. Failure to accept responsibility	1.43	0.71
Facet 3: Lifestyle	5.57	1.70	.67
3. Need for stimulation	0.90	0.76
9. Parasitic lifestyle	0.61	0.70
13. Lack of long-term goals	0.62	0.70
14. Impulsivity	1.20	0.72
15. Irresponsibility	1.23	0.73
Facet 4: Antisocial	4.54	3.50	.61
10. Poor behavioral controls	1.06	0.82
12. Early behavioral problems	0.73	0.81
18. Juvenile delinquency	0.81	0.87
19. Revocation of release	1.41	0.87
20. Criminal versatility	0.74	0.78
Other
11. Promiscuous sexual behavior	1.56	0.69	—
17. Many short-term marital relations^a	0.62	0.75	—
Factor 1	8.84	3.97	.82
Factor 2	9.32	4.27	.74
PCL-R total	19.98	7.34	.83

Note. N = 615 for items on the four facets. N = 614 for Promiscuous Sexual Behavior. PCL-R = Psychopathy Checklist—Revised.

Item not included in network models.

Recent research suggests that results from psychopathy network analyses studies may be useful for informing these debates. These studies allow researchers to identify the traits that are at the core of psychopathy by investigating how the traits assessed by the measure relate to one another in terms of connectivity and importance. Findings from the small number of existing PCL network studies suggest that traits from the Affective facet tend to be the most central to the construct of psychopathy and traits from the Antisocial facet tend to be the least central; however, these studies differ with respect to their conclusions about which specific traits are the most and least central (Preszler et al., 2018; Verschuere et al., 2018). This study adds to this small, but growing literature using network analyses to evaluate the centrality of the traits assessed by the PCL-R when scores have been assigned for real-world decision-making in the field, where the psychometric properties of scores on the PCL-R often differ from those of scores assigned for research purposes (Edens & Boccaccini, 2017).

Network Analysis

Network analysis is a statistical technique developed to examine pathways between symptoms (Borsboom & Cramer, 2013; McNally, 2016). Network analysis results provide a visual depiction of the symptoms of a disorder and how strongly they relate to one another. More specifically, symptoms (e.g., PCL-R items) are depicted as nodes, and the relation between two nodes is represented by a straight line, or edge, that connects those two symptoms. Stronger associations between nodes are represented by thicker edges. Network analysis results also provide numerical indices that quantify the centrality of each symptom. For example, strength is the sum of the absolute values of correlations between a node and other nodes within the network. Symptoms with higher centrality values are more likely to be at the core of the construct. Traits not at the core (i.e., in the periphery) show low or even negative scores across the centrality measures.

Centrality findings from network models provide evidence of the connectedness of the traits of psychopathy, and understanding connectedness among the traits of a disorder such as psychopathy is useful for at least two reasons. First, traits that are highly central can be thought of as being more necessary to the construct. Although findings from other types of analyses, such as item response theory and factor analysis, provide seemingly similar types of information, those types of analyses usually focus on the associations among traits assessing the same underlying latent factor (e.g., a specific PCL-R factor or facet) or the association among those latent factors. Network analysis allows for an examination of the association among individual traits, irrespective of latent variable structure or properties.

Second, centrality findings may have significant implications for criminal justice policy and practice. From a Risk-Need-Responsivity (RNR; Andrews et al., 1990) treatment framework perspective, centrality findings may help practitioners identify responsivity indicators that clinicians can target to facilitate engagement in the treatment process, a predictor of treatment progress (J. S. Levenson & Macgowan, 2004) and satisfaction (J. S. Levenson et al., 2009, 2010). Indeed, researchers have found that affective facet traits influence treatment progress among highly psychopathic individuals convicted of a sex offense (Olver et al., 2011; Sewall & Olver, 2019), suggesting that affective elements of psychopathy (responsivity indicators in the RNR framework) must be managed during treatment, while the more behavioral elements are changed (S. Wong & Hare, 2005; S. C. P. Wong, 2015). Affective traits with high centrality in network models may be those that are the most important to target for management during treatment.

Existing Psychopathy Network Analysis Findings

Two published studies have reported network analysis findings for the PCL-R. The first study used data from 1,559 individuals ordered to either prison or substance abuse treatment in several states across the United States, 3,954 individuals residing in state prisons in Wisconsin, and 1,937 violent mentally disordered individuals convicted of a crime under mandatory inpatient treatment in the Netherlands (Verschuere et al., 2018). The second study used data from 277 California forensic hospital patients who had been found incompetent to stand trial, not guilty by reason of insanity, or had been classified as a “Mentally Disordered Offender” (i.e., too dangerous for release secondary to mental illness; Preszler et al., 2018). In both of these studies, researchers conceptualized each individual PCL-R item as a “symptom” (i.e., node) of the larger construct of psychopathy. They then constructed networks to evaluate the associations between these symptoms.

Across both studies, items from the Affective facet were among the most central in the networks; however, the specific items that were the most central differed between samples. Among the U.S. correctional samples, Lack of Empathy and Shallow Affect were the most central symptoms (Verschuere et al., 2018). In the U.S. forensic sample, Lack of Remorse was the most central trait; however, centrality values were also strong for Lack of Empathy (Preszler et al., 2018). Similarly, both studies found that items from the Antisocial facet were among the least central items. Each study also found items from other facets to lack centrality. For example, Revocation of Release (Antisocial facet) was among the least central across both U.S. correctional samples and Lack of Long-Term Goals (Lifestyle facet) was among the least central in one U.S. correctional sample (Verschuere et al., 2018). In contrast, the least central items in the U.S. forensic patient sample included Revocation of Release from the Antisocial facet, as well as Pathological Lying, Manipulative, and Impulsivity from the Interpersonal facet (Preszler et al., 2018).

Together, these two existing PCL-R network analysis studies point to some generalizability in findings at the facet level, with high centrality for affective traits and lower centrality for traits from the Antisocial, Interpersonal, and Lifestyle facets, but a lack generalizability at the item/symptom level. In other words, facet centrality tends to show stability across samples despite some item-level variability.

As the PCL-R is scored by clinicians, PCL-R network analysis findings necessarily provide information about the centrality of psychopathy traits as scored by clinicians using one specific psychopathy measure. Findings from a recent network analysis study of self-report psychopathy measure scores from undergraduate students help to support the generalizability of the main PCL-R network findings (Tsang & Salekin, 2019). Specifically, researchers used network analyses with scores from the Levenson Self-Report Psychopathy Scale (LSRP; M. R. Levenson et al., 1995), the Psychopathic Personality Inventory (PPI; Lilienfeld & Andrews, 1996), and the Self-Report Psychopathy Scale-II (SRP-II; Hare et al., 1989). Similar to PCL-R network findings, affective traits were among the most central when examining all traits assessed by the four psychopathy measures; however, in contrast to PCL-R network findings, items assessing manipulativeness (Interpersonal facet), irresponsibility (Lifestyle facet), and impulsivity (Lifestyle facet) were the most central in their item-level psychopathy networks.

Together, findings from these three network analysis studies show some similarities (e.g., affective traits being most central), but some potentially important differences (e.g., inconsistent strength for Shallow Affect, Conning/Manipulative). More network analysis research is needed, especially research using different types of patient and justice-involved samples.

Current Study

Our goal was to examine the generalizability of PCL-R network analysis findings to PCL-R scores assigned to individuals convicted of a sex offense who were evaluated for civil commitment as Sexually Violent Predators (SVPs). SVP laws allow for the postrelease civil commitment of individuals convicted of a sex offense who are believed to be at an especially high risk for reoffending due to a mental or behavioral abnormality that makes it difficult for them to control their offending behavior (Miller et al., 2005). The PCL-R is one of the most commonly used measures in SVP evaluations (Boccaccini et al., 2017).

There are several reasons why network findings in this sample of individuals being evaluated for SVP commitment may differ from previous research. SVP evaluators administer the PCL-R as just one part of a broader evaluation to help answer questions about impairment and risk. Individuals being evaluated for SVP commitment know they are participating in an evaluation that will influence decisions about their future and have clear reasons to engage in impression management (i.e., to avoid SVP commitment). Existing network analysis findings are generally based on psychopathy measure scores assigned in the structured context of a research study, where the evaluator’s ultimate goal is to arrive at the correct PCL-R score and evaluees know that their results will only be used for research purposes (Edens & Boccaccini, 2017). These factors could influence both evaluator (e.g., interview content, focus of record review) and evaluee (e.g., openness, honesty) behavior, and lead to network analysis findings that have implications for assessment and treatment of individuals convicted of a sex offense.

We examined two network models for the study. The first model followed recent PCL-R network research and included only the 18-items falling onto one of the four PCL-R facets (Preszler et al., 2018). The second model included a 19th item, Promiscuous Sexual Behavior, which researchers have included in some prior network studies and found to be generally low in centrality (Verschuere et al., 2018). The Promiscuous Sexual Behavior item was designed to identify individuals whose sexual relationships may be impersonal, characterized be infidelity, and potentially coercive, with high scores indicating illegal sexual behavior, not necessarily illegal sexual behavior, or both (Hare, 2003). We included this item in a second network model to examine if it emerges as more central in an SVP evaluation sample than it has in prior research.

Method

Participants and Procedure

The PCL-R scores used in this study come from 615 males convicted of a sex offense who were evaluated for commitment as SVPs by state-contracted evaluators in Texas between 1999 and 2018. In Texas, criteria for SVP commitment include that the individual is a “repeat sexually violent offender” who “suffers from a behavioral abnormality that makes the person likely to engage in a predatory act of sexual violence” (Texas Health and Safety Code § 841.003). Behavioral abnormality evaluators are required to perform a clinical interview and assess for psychopathy (Texas Health and Safety Code § 841.023). Although the behavioral abnormality statute does not require evaluators to use the PCL-R (Hare, 1991, 2003), the referral letter state-contracted evaluators receive from the state’s Department of Criminal Justice instructs them to administer the “Hare Psychopathy Checklist—Revised (PCL-R).”

The mean age among the SVP evaluees at the time of evaluation was 45.5 (SD = 11.7). Evaluees (100% male) were identified by evaluators as white (n = 317, 51.5%), black (n = 166, 27.0%), Latino (n = 114, 18.5%), or from another racial/ethnic background (n = 7, 1.1%; missing n = 11, 1.8%). Evaluators reported whether evaluees had at least one child victim (n = 402, 65.4%) or no child victims (n = 184, 29.9%; missing n = 29, 4.7%), but did not provide any detailed information about offending history (e.g., number of offenses, number of victims, different types/ages of victims).

The research team collected deidentified data directly from the SVP evaluators for a series of ongoing projects examining evaluator differences in PCL-R scoring (Boccaccini et al., 2008, 2014). Evaluators entered PCL-R item scores and basic demographic and victim information (any child victim versus no child victims) about the individuals they evaluated into an Excel worksheet and sent the completed worksheets to the research team. We contacted 21 state-contracted evaluators at the onset of the study and 12 provided data for a total of 619 cases. The mean number of evaluations per evaluator was 51.3 (SD = 36.2, range = 9–139). Eleven evaluators reported having completed at least one formal PCL-R training workshop, with most having completed a training workshop given by Robert Hare, Adelle Forth, or Stephen Hart (Rufino et al., 2012). We have no information about the training experience of the 12th evaluator, who scored eight of the 615 individuals in the study.

Each evaluee must have had a history of two or more qualifying sex offenses to be considered for civil commitment. Potentially qualifying offenses include: Indecency with a child (sexual contact), sexual assault, aggravated sexual assault, aggravated kidnapping (with intent to sexually abuse or violate), burglary (with intent to commit a sexual offense mentioned above), a murder that is determined beyond a reasonable doubt to have been based on sexually motivated conduct, and attempt, conspiracy, or solicitation to commit any offense mentioned above (Texas Health & Safety Code § Chapter 841).

Materials

The PCL-R is a 20-item clinician-scored measure of psychopathy. Evaluators rate each item on a scale from 0 to 2, with higher scores reflecting higher levels of that trait. The items are separated into two factors, Interpersonal-Affective (Factor 1) and Lifestyle-Antisocial (Factor 2), and four facets, Interpersonal (Facet 1), Affective (Facet 2), Lifestyle (Facet 3), and Antisocial (Facet 4). Two items (Promiscuous Sexual Behavior and Many Short-Term Marital Relationships) do not load on either factor or any facet. We followed previous research by excluding these items from our first network analysis model (Preszler et al., 2018), but included Promiscuous Sexual Behavior in a second model to allow for an examination of the centrality of impersonal and potentially coercive sexual behavior within the psychopathy network. Table 1 provides descriptive statistics for the PCL-R facets and items used in this study.

Eighty of the 619 SVP evaluees had at least one missing PCL-R item score for the 18 items included in the network analysis. We used prorating procedures from the PCL-R: Second Edition Manual (Hare, 2003) to estimate scores for these items (i.e., prorated facet score minus the sum of the items on the facet with scores). Four cases were missing too many item scores to prorate. We prorated a total of 111 item scores (Facet 1 = 20, Facet 2 = 3, 40 from Facet 3 = 40, Facet 4 = 48) across the remaining 76 individuals with missing items. Thus, the final sample included 615 individuals, with 18 item scores, yielding a total of 11,070 item scores. Only 1.0% (111/11,070) of these scores were prorated. As one of the 615 individuals was missing a score for the Promiscuous Sexual Behavior item, our second network model included scores from 614 evaluees.

Research suggests that rater agreement is generally weaker for PCL-R scores in Texas SVP evaluations than reported in the PCL-R manual (Boccaccini et al., 2014). Although the PCL-R manual reports rater-agreement coefficients in the ICC₁ = .70 to .90 range for PCL-R scores (Hare, 2003), Texas SVP studies have reported lover values for total (ICC_A,1 = .40–.55), Factor 1 (ICC_A,1 < .20), and Factor 2 (ICC_A,1 < .60) scores (Boccaccini et al., 2014; Edens et al., 2010).

Analytic Strategy

We followed the framework of the network analytic strategy used by Preszler et al. (2018) to allow for direct comparisons between our findings and those from the most recent published study in this area. First, we estimated a partial correlation network using an EBIC Least Absolute Shrinkage and Selection Operator (LASSO) network. Though Preszler et al. (2018) used the adaptive lasso procedure for inducing sparsity, more recent network research has used the EBIC method, which has the benefit of using the extended Bayesian information criterion fit index to guide the regularization. In this graphical representation of the connection between all nodes, edges represent the direct relations between nodes. The thicker the edge, the stronger the magnitude of the association between these two symptoms (Epskamp & Fried, 2017). Second, to further elaborate on the graphical representation of the relation between symptoms, we calculated strength and expected influence (EI) centrality measures. These centrality measures help to identify the role of specific symptoms across the network. High strength centrality, which is the sum of the absolute value of the edge weights of a particular node, indicate strong direct relationships with other nodes in the structure. EI is similar to strength in that it measures the sum of the edge weights of a particular node; however, EI maintains negative associations instead of summing absolute values. The values of strength and EI are presented as z scores, meaning the higher the value of the measure, the more central the symptom is to the network. All analyses were performed using the qgraph package (Epskamp et al., 2012) within R.

Results

Network Analysis: 18 Items

Network Structure

Figure 1 displays the adaptive LASSO network of the 18 PCL-R items included in prior network studies. The placement of the nodes represent how correlated each node is with the other nodes in the network and the strength of that correlation is reflected in the thickness of edge. The total number of possible edges is 153 (18 × 17/2) and of these possible edges, the network produced 80 edges (i.e., 52% of the possible edges).

Figure 1:

EBICglasso Network Graph of 18 PCL-R Items

In Figure 1, the interpersonal and affective traits are most closely clustered together. Specifically among the Interpersonal facet, Superficial Charm and Grandiose Sense of Self-Worth were strongly connected, among the Affective facet, Lack of Remorse was strongly connected with Lack of Empathy and Failure to Accept Responsibility, and among the Antisocial facet, Juvenile Delinquency and Early Behavioral Problems were strongly connected. The lifestyle symptoms were not as closely clustered. Among the Lifestyle facet, Need for Stimulation and Impulsivity were strongly connected as well as Parasitic Lifestyle and Irresponsibility. The connections between the traits in each respective facet were positive; however, the network produced three weak negative edge weights: between Grandiose Sense of Self Worth and Irresponsibility, between Failure to Accept Responsibility and Juvenile Delinquency, and between Juvenile Delinquency and Manipulative.

Centrality

Table 2 displays the z scores of strength and EI. Nonparametric bootstrapped stability analyses indicated correlation-stability (CS) coefficients (nBoots = 5,000; correlation = .7) to be acceptable for strength (.67) and EI (.75). See Epskamp et al. (2018) for details on nonparametric boostrapping and acceptable CS-coefficient levels.

Table 2:

Centrality and Cluster Measure z Scores for 18 Item Network

Symptom	EI	Strength
Interpersonal facet
1. Superficial charm	−0.17	−0.25
2. Grandiose	−0.29	0.11
4. Pathological lying	1.12	1.04
5. Manipulative	0.54	0.55
M	0.30	0.31
Affective facet
6. Lack of remorse	2.12	2.05
7. Shallow affect	−0.60	−0.69
8. Lack of empathy	1.04	0.97
16. Failure to accept responsibility	0.24	0.55
M	0.70	0.72
Lifestyle facet
3. Need for stimulation	1.06	0.99
9. Parasitic lifestyle	−0.51	−0.60
13. Lack of long-term goals	−1.25	−1.35
14. Impulsivity	0.69	0.62
15. Irresponsibility	−0.12	0.06
M	−0.03	−0.06
Antisocial facet
10. Poor behavioral controls	−0.19	−0.28
12. Early behavioral problems	0.19	0.11
18. Juvenile delinquency	−0.79	−0.40
19. Revocation of release	−2.03	−2.13
20. Criminal versatility	−1.04	−1.13
M	−0.77	−0.77

Note. All scores are z scores. EI = Expected Influence. N = 615.

Four items, Pathological Lying, Lack of Remorse, Lack of Empathy, and Need for Stimulation, had the strongest EI and strength values (z scores > 1.00 and .97, respectively). When looking at the centrality of the respective facets, the items on the Affective facet were highly central in both EI (M = .70) and strength (M = .72). These averages suggests this facet may be the most central facet to the psychopathy network as measured by the PCL-R. Interestingly, Shallow Affect which falls among the Affective facet was low across EI (z score = −.60) and strength (z score = −.69).

Lack of Long-Term Goals, Revocation of Release, and Criminal Versatility, were consistently low across EI and strength (z scores < −1.00). Juvenile Delinquency was among the least central items across the EI index (z score = −.79) and Shallow Affect was among the least central items across the strength index (z score = −.69). The Antisocial facet proved to be the least central facet across EI (M = −.77) and strength (M = −.77), suggesting the Antisocial facet is not as central to the psychopathy network as measured by the PCL-R. Early Behavioral Problems was the only item among the Antisocial facet with positive centrality values on EI and strength (z score = .19 and .11, respectively), but these values were relatively small.

Network Analysis Including Promiscuous Sexual Behavior (19 Items)

Figure 2 displays the adaptive LASSO network of 19 psychopathy symptoms as assessed by the PCL-R. The model includes the 18 items from the previous model and includes Promiscuous Sexual Behavior as a 19th item. The total number of possible edges is 171 (19 × 18/2) and of these possible edges, the network produced 93 edges (i.e., 54% of the possible edges). Table 3 displays the z scores of strength and EI for the 19 item network analysis. Nonparametic bootstrapped stability analyses indicated CS coefficients (nBoots = 5,000; correlation = .7) to be acceptable for strength (.67) and EI (.67). Overall, adding a Promiscuous Sexual Behavior node did little to change the overall structure of the network (Figure 2) or the centrality of the other nodes (Table 3).

Figure 2:

EBICglasso Network Graph of 19 PCL-R Items

Table 3:

Centrality and Cluster Measure z Scores for 19 Item Network

Symptom	EI	Strength
Interpersonal facet
1. Superficial charm	–0.16	−0.30
2. Grandiose	–0.19	0.08
4. Pathological lying	1.22	1.10
5. Manipulative	0.49	0.54
M	0.34	0.36
Affective facet
6. Lack of remorse	2.17	2.06
7. Shallow affect	–0.61	−0.76
8. Lack of empathy	0.99	0.86
16. Failure to accept responsibility	0.16	0.52
M	0.68	0.67
Lifestyle facet
3. Need for stimulation	1.24	1.11
9. Parasitic lifestyle	–0.55	−0.70
13. Lack of long-term goals	–1.25	−1.41
14. Impulsivity	0.69	0.55
15. Irresponsibility	0.15	0.43
M	0.06	−0.00
Antisocial facet
10. Poor behavioral controls	–0.28	−0.19
12. Early behavioral problems	0.32	0.24
18. Juvenile delinquency	–0.79	−0.31
19. Revocation of release	–2.01	−2.18
20. Criminal versatility	–0.51	−0.66
M	–0.65	−0.62
11. Promiscuous sexual behavior	–1.07	−0.98

Note. All scores are z scores. EI = Expected Influence. N = 614.

Promiscuous Sexual Behavior appeared to be among the least central items across both EI and strength (z score = −1.07 and −.98, respectively) in this 19-symptom model (Table 3). These findings suggest Promiscuous Sexual Behavior is not highly central to the construct of psychopathy as measured by the PCL-R, even among a sample of relatively high-risk individuals convicted of a sex offense. Promiscuous Sexual Behavior was most closely connected to Criminal Versatility, both from the Antisocial facet, followed by Need for Stimulation and Irresponsibility from the Lifestyle facet (Figure 2).

Discussion

Our goal was to examine the generalizability of PCL-R network analysis findings to field scores assigned for real-world risk-assessment decision-making in SVP cases. Table 4 provides comparisons across these PCL-R network studies, listing the 19 PCL-R items from our final network model from highest strength to lowest strength, and providing the rank order of those same items from prior PCL-R network studies.

Table 4:

Rank Order of Strength Ratings for PCL-R Items From Current and Prior Network Analysis Studies

PCL-R item	Individuals convicted of a sex offense/field study	Forensic patients/research study	Discorded convicted persons/field study	Convicted persons/research study^a	Convicted persons/research study^b
6. Lack of remorse	1	1	3	3	3
3. Need for stimulation	2	17	6	12	10
4. Pathological lying	3	8	12	8	5
8. Lack of empathy	4	3	4	1	1
14. Impulsivity	5	13	9	18	7
5. Conning/manipulative	6	12	4	4	2
16. Failure to accept responsibility	7	4	10	9	15
15. Irresponsibility	8	5	1	16	17
12. Early behavioral problems	9	2	15	11	5
2. Grandiose	10	10	10	7	12
10. Poor behavioral controls	11	11	12	4	4
1. Superficial charm	12	9	6	4	11
18. Juvenile delinquency	13	6	15	16	13
20. Criminal versatility	14	6	6	13	7
9. Parasitic lifestyle	15	14	2	14	9
7. Shallow affect	16	18	19	2	13
11. Promiscuous sexual behavior	17	—	20	10	16
13. Lack of long-term goals	18	15	12	14	18
19. Revocation of release	19	16	15	19	19
17. Short-term relationships	—	—	18	20	19

Note. Items in the same column with the same rank had identical strength values in the study. PCL-R = Psychopathy Checklist—Revised.

NIMH sample. ^bWisconsin sample.

Generalizability of Affective and Antisocial Facet Items Across Samples

The findings summarized in Table 4 show several areas of generalizability across studies (i.e., field vs. research) but also some areas of inconsistency. For example, our findings provide continued support for the generalizability of affective features as being highly central across studies using both the PCL-R (Preszler et al., 2018; Verschuere et al., 2018) and self-report measures (Bronchain et al., 2019; Tsang & Salekin, 2019). Items assessing lack of remorse, lack of empathy, and unconcern have consistently high strength ratings across studies, regardless of study design (i.e., field vs. research), sample characteristics (i.e., individuals convicted of a sex offense vs. individuals convicted of a general offense vs. forensic patients), and type of measure (i.e., self-report or clinician-scored). These findings support assertions in the larger literature regarding the importance of affective traits to the construct of psychopathy (Verschuere & te Kaat, 2019).

The centrality of these affective traits has potentially significant implications for treatment. The literature is cautiously optimistic regarding the treatability of individuals convicted of a sex offense who score high on psychopathy; however, treatment success appears to depend on whether treatment is tailored to the responsivity of the individual (Olver & Wong, 2009; Sewall & Olver, 2019). Affective psychopathy traits influence treatment progress (Olver et al., 2011; Sewall & Olver, 2019) and, therefore, need to be appropriately managed for successful behavioral change among those with high levels of psychopathy (S. Wong & Hare, 2005; S. C. P. Wong, 2015). Our findings for the centrality of Lack of Remorse and Lack of Empathy and those from prior network studies highlight the need to tailor treatment to manage these specific affective traits in treatment.

Our findings are also consistent with existing studies in showing that the items from the Antisocial facet tend to be less central than items from other facets (Preszler et al., 2018; Verschuere et al., 2018). These consistent findings are noteworthy given debates about the extent to which antisocial behaviors are central to the construct (Hare & Neumann, 2010; Skeem & Cooke, 2010). Thus far, all network analysis studies investigating the PCL suggest that, compared to other elements of the disorder, the Antisocial facet is peripheral to psychopathy.

Unique Findings in the Sample Individuals Convicted of a Sex Offense

The three PCL-R items that emerged as having higher strength values in our field sample of individuals convicted of a sex offense than in other samples were Pathological Lying, Need for Stimulation, and Impulsivity. As the PCL-R is a clinician-scored instrument, scores reflect clinicians’ perceptions of an individual’s traits, and the network models derived from these scores are therefore influenced by those perceptions. If evaluators in one context focus on different traits than evaluators in another context, it could help to explain why centrality findings for some items vary across samples, and why some items emerged as highly central in our sample of individuals evaluated for SVP commitment. For example, the prominence of Pathological Lying in our sample is consistent with the prominent role of denial in the broader literature examining risk assessment among individuals convicted of a sex offense. Indeed, clinicians’ perceptions of an evaluee’s level of denial are correlated with their opinions about risk (e.g., r = .44; McCallum et al., 2017), despite the fact that meta-analyses have repeatedly shown that denial is not a significant predictor of future offending (d = .02, Hanson & Morton-Borgoun, 2005; Mann et al., 2010). Perhaps this misplaced reliance on denial in risk assessment of individuals convicted of a sex offense leads Pathological Lying to become more central to evaluators’ views of the psychopathy construct within the context of SVP evaluations. In other words, the centrality of Pathological Lying in the current psychopathy network model may be a byproduct of SVP evaluators focusing on denial and minimization within the broader context of risk-assessment evaluations of individuals convicted of a sex offense.

The centrality of Pathological Lying could also be attributable to other aspects of the SVP evaluation context. Individuals evaluated for SVP commitment know that they are facing potentially lifelong postrelease confinement and evaluators may be especially likely to cross-check information obtained via interview. In these real-world cases, evaluees may lie more often, and be caught lying more often, than individuals participating in research studies. Clear evidence of dishonesty in these cases—likely indicated by discrepancies between interview responses and collateral records—may ultimately influence scoring across the PCL-R (e.g., Failure to Accept Responsibility, Conning/Manipulative, Lack of Remorse, Shallow Affect; see Figure 1).

The high centrality of Pathological Lying in our network indicates that ratings on this item were strongly associated with the scoring of other items, more so than in prior studies. One possible explanation for this finding is that this is akin to the type halo/horn effect that researchers have reported in other diagnostic contexts, with evidence for one symptom influencing ratings for other symptoms that should not have been affected (DeVries et al., 2017; Mumma, 2002). In the context of this study, an over-emphasis on denial may have inappropriately influenced scoring across the entire PCL-R. For example, scores on items with higher centrality within this sample (e.g., Lack of Remorse, Lack of Empathy, Failure to Accept Responsibility) could have also been influenced by behaviors and attitudes consistent with denial (rationalization of offending behavior, minimization of severity). If true, this potential confounding of denial and scoring across multiple PCL-R items might explain the relatively small association between PCL-R scores and recidivism among individuals convicted of a sex offense, especially in field studies (Harris et al., 2017; Hawes et al., 2013), and would suggest a need for training PCL-R evaluators to avoid this type of systematic measurement error when evaluating individuals convicted of a sex offense.

The relatively high centrality of Need for Stimulation and Impulsivity in our sample might also be attributable to the unique context of SVP evaluations, where evaluators are asked to identify individuals who lack the ability to control sexually dangerous behavior (Miller et al., 2005). The variability in strength values for these items across PCL-R network studies (Table 4) may support those questioning the need for impulsivity items in psychopathy measures (Poythress & Hall, 2011). Moreover, although some research with self-report measures and nonjustice-involved samples suggests that impulsivity may be highly central to the construct (Tsang & Salekin, 2019), other research using similar methods does not (Bronchain et al., 2019). These inconsistent findings for impulsivity may be less problematic for risk assessment of individuals convicted of a sex offense than those for pathological lying. Indeed, meta-analytic findings suggest that impulsivity and self-regulation problems are meaningful predictors of sexual recidivism (Hanson & Morton-Borgoun, 2005; Mann et al., 2010).

Promiscuous Sexual Behavior

Because our sample was comprised entirely of individuals convicted of a sex offense, we included the Promiscuous Sexual Behavior item from the PCL-R into a second set of analyses due to its potential relevance to the sample. The network analysis with this item showed it to be among the least central items to the construct of psychopathy, which is generally consistent with findings from other studies that have included the item (Table 4). The item’s strongest connection was Criminal Versatility. Low centrality for the Promiscuous Sexual Behavior item is consistent with its exclusion from PCL-R factor models, suggesting that it is not strongly associated with other psychopathic traits as measured by the PCL-R.

It is, however, also possible that a sample of individuals convicted of a sex offense may not be the best setting for evaluating associations with this item. Deviant and illegal sexual behavior are taken into consideration when scoring item 11, with the PCL-R manual noting that those with high scores may have charges or convictions for sexual assault (Hare, 2003). Thus, it is not surprising that the mean score for the Promiscuous Sexual Behavior item was 1.56, which was higher than any other item (Table 1).

Limitations and Future Directions

This study differed from previous PCL-R network analyses studies in two fundamental ways: This was the first study to focus on a sample of individuals convicted of a sex offense, and this was the first study in which the PCL-R scores had significant consequences for evaluees’ legal cases. One factor that may affect the generalizability of our findings is that the data come from a select subset of presumably high risk individuals convicted of a sex offense. The mean PCL-R total score of 19.98 indicates that our sample was similar to those used in other network studies with respect to overall levels of psychopathy (e.g., M = 20.09, 22.54, and 23.30 by Verschuere et al., 2018). However, individuals in SVP samples tend to have multiple prior offenses and relatively high scores on measures specifically designed to predict reoffending, raising questions about the generalizability of the findings to other samples of individuals convicted of a sex offense.

PCL-R scores assigned in the context of SVP evaluations could have been affected by aspects of the evaluation process itself, including the referral questions in SVP evaluations and evaluees being motivated to avoid long-term SVP commitment. These and other aspects of field settings make it difficult to interpret whether differences between our network findings and previous studies are due to something about individuals convicted of a sex offense, SVP evaluations, or other aspects of real-world forensic assessment.

Another limitation is that we did not have detailed information about the evaluees’ victims or offense-related behaviors. PCL-R research with individuals convicted of a sex offense generally shows that those who offend only against children tend to have lower scores than those who offend against adults or both adults and children (Brown et al., 2015; Skovran et al., 2010). Furthermore, research suggests this difference is consistent across total, factor (Brown et al., 2015; Skovran et al., 2010), and facet scores (Brown et al., 2015). These findings raise questions about whether network models might vary depending on victim type.

An additional limitation is that the study focused on field scores from only one psychopathy assessment tool (PCL-R). Although the PCL-R is the most widely used psychopathy measure in clinical-forensic practice (Boccaccini et al., 2017; Neal & Grisso, 2014; Viljoen et al., 2010), it would be useful to investigate how psychopathic symptoms relate to one another in field settings when assessed using other psychopathy assessment tools. Network analysis findings based on self-report scores from nonfield settings reveal that interpersonal and affective traits are the most central domains of those networks (Bronchain et al., 2019; Tsang & Salekin, 2019), suggesting that high centrality findings for these traits are not unique to the raters of the PCL-R. However, more research is needed, especially research that conceptualizes psychopathy across different administration formats and among diverse populations (Preszler et al., 2018; Verschuere et al., 2018).

Conclusion

Although our study contained limitations, it is among the first to explore the relation between items on the PCL-R using network analysis with data collected from risk-assessment evaluations of individuals convicted of a sex offense. At the facet level, our findings support those from prior research in that the Affective facet was found to be the most central, whereas the Antisocial facet was found to be the least central (Preszler et al., 2018; Verschuere et al., 2018). At the item level, we found Lack of Remorse to be the most central and Revocation of Release to be the least central item in the network across both centrality indices. This suggests that the Affective facet, specifically Lack of Remorse, may lay at the core of the construct, and the more antisocial behaviors, particularly Revocation of Release, may not be as fundamental. Overall, network analysis provides a unique way of evaluating how the traits assessed by the PCL-R relate to one another in terms of connectivity and importance. Future research should continue to explore these relations across diverse settings and using various psychopathy measures.

Footnotes

ORCID iDs

Jonathan Preszler

Marcus T. Boccaccini

Gabriele Trupp earned her MA in Forensic Psychology from John Jay College of Criminal Justice and is now in her fourth year of the Clinical Psychology PhD program at Sam Houston State University. She focuses on research that can be used to better inform clinical practice.

Jonathan Preszler received his PhD in clinical psychology from Washington State University. He is currently a neuropsychology fellow at the University of Pittsburgh Medical Center.

Marcus T. Boccaccini is a professor of psychology at Sam Houston State University. His research program focuses on the reliability and validity of forensic assessment procedures and opinions.

David K. Marcus is a professor and Chair of the Department of Psychology at Washington State University. His research interests include psychopathy, spitefulness, and other dark personality traits.

Jorge G. Varela is a professor of psychology at Sam Houston State University, where he serves as the Director of Clinical Training for their doctoral program in Clinical Psychology.

Darrel B. Turner works in private practice in Forensic and Clinical Psychology. He is licensed in Texas and Louisiana, and his primary area of expertise is in sexual offending perpetrated by adult males and females.

References

Andrews

D. A.

Bonta

Hoge

R. D.

(1990). Classification for effective rehabilitation: Rediscovering psychology. Criminal Justice and Behavior, 17(1), 19–52. https://doi.org/10.1177/0093854890017001004

Boccaccini

M. T.

Chevalier

Murrie

D. C.

Varela

J. G.

(2017). Psychopathy Checklist-Revised use and reporting practices in sexually violent predator evaluations. Sexual Abuse: A Journal of Research and Treatment, 29(6), 592–614. https://doi.org/10.1177/1079063215612443

Boccaccini

M. T.

Murrie

D. C.

Rufino

K. A.

Gardner

B. O.

(2014). Evaluator differences in Psychopathy Checklist-Revised factor and facet scores. Law and Human Behavior, 38(4), 337–345. https://doi.org/10.1037/lhb0000069

Boccaccini

M. T.

Turner

D. B.

Murrie

D. C.

(2008). Do some evaluators report consistently higher PCL-R scores than others? Findings from a statewide sample of sexually violent predator evaluations. Psychology, Public Policy, and Law, 14(4), 262–283. https://doi.org/10.1037/a0014523

Borsboom

Cramer

A. O. J.

(2013). Network analysis: An integrative approach to the structure of psychopathology. Annual Review of Clinical Psychology, 9, 91–121. https://doi.org/10.1146/annurev-clinpsy-050212-185608

Bronchain

Raynal

Chabrol

(2019). The network structure of psychopathic personality traits in a non-institutionalized sample. Personality and Individual Differences, 146, 41–45. https://doi.org/10.1016/j.paid.2019.03.044

Brown

A. R.

Dargis

M. A.

Mattern

A. C.

Tsonis

M. A.

Newman

J. P.

(2015). Elevated psychopathy scores among mixed sexual offenders: Replication and extension. Criminal Justice and Behavior, 42(10), 1032–1044. https://doi.org/10.1177/0093854815575389

DeVries

L. N.

Hartung

C. M.

Golden

T. L.

(2017). Negative halo effects in parent ratings of ADHD and ODD. Journal of Psychopathology and Behavioral Assessment, 39(2), 179–188. https://doi.org/10.1007/s10862-016-9560-z

Drislane

L. E.

Brislin

S. J.

Jones

Patrick

C. J.

(2018). Interfacing five-factor model and triarchic conceptualizations of psychopathy. Psychological Assessment, 30(6), 834–840. https://doi.org/10.1037/pas0000544.supp

10.

Edens

J. F.

Boccaccini

M. T.

(2017). Taking forensic mental health assessment “out of the lab” and into “the real world”: Introduction to the special issue on the field utility of forensic assessment instruments and procedures. Psychological Assessment, 29(6), 599–610. https://doi.org/10.1037/pas0000475

11.

Edens

J. F.

Boccaccini

M. T.

Johnson

(2010). Inter-rater reliability of the PCL-R total and factor scores among psychopathic sex offenders: Are personality features more prone to disagreement than behavioral features? Behavioral Sciences and the Law, 28(1), 106–119. https://doi.org/10.1002/bsl.918

12.

Epskamp

Borsboom

Fried

E. I.

(2018). Estimating psychological networks and their accuracy: A tutorial paper. Behavior Research Methods, 50(1), 195–212. https://doi.org/10.3758/s13428-017-0862-1

13.

Epskamp

Cramer

A. O.

Waldorp

L. J.

Schmittmann

V. D.

Borsboom

(2012). qgraph: Network visualizations of relationships in psychometric data. Journal of Statistical Software, 48(4), 1–18. https://doi.org/10.18637/jss.v048.i04

14.

Epskamp

Fried

E. I.

(2017). A tutorial on regularized partial correlation networks 2018. arxiv. http://arxiv.org/abs/1607.01367

15.

Hanson

R. K.

Morton-Borgoun

K. E.

(2005). The characteristics of persistent sexual offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical Psychology, 73(6), 1154–1163. https://doi.org/10.1037/0022-006X.73.6.1154

16.

Hare

R. D.

(1991). The Hare Psychopathy Checklist–Revised (PCL-R). Multi-Health Systems.

17.

Hare

R. D.

(2003). Hare PCL-R (2nd ed.). Multi-Health Systems.

18.

Hare

R. D.

Harpur

T. J.

Hemphill

J. F.

(1989). Scoring pamphlet for the Self-Report Psychopathy Scale: SRP-II. Simon Fraser University.

19.

Hare

R. D.

Neumann

C. S.

(2010). The role of antisociality in the psychopathy construct: Comment on Skeem and Cooke (2010). Psychological Assessment, 22(2), 446–454. https://doi.org/10.1037/a0013635

20.

Harris

P. B.

Boccaccini

M. T.

Rice

A. K.

(2017). Field measures of psychopathy and sexual deviance as predictors of recidivism among sexual offenders. Psychological Assessment, 29(6), 639–651. https://doi.org/10.1037/pas0000394

21.

Hawes

S. M.

Boccaccini

M. T.

Murrie

D. C.

(2013). Psychopathy and the combination of psychopathy and sexual deviance as predictors of sexual recidivism: Meta-analytic findings using the psychopathy checklist-revised. Psychological Assessment, 25(1), 233–243. https://doi.org/10.1037/a0030391

22.

Levenson

J. S.

Macgowan

M. J.

(2004). Engagement, denial, and treatment progress among sex offenders in group therapy. Sexual Abuse: A Journal of Research and Treatment, 16(1), 49–63. https://doi.org/10.1177/107906320401600104

23.

Levenson

J. S.

Macgowan

M. J.

Morin

J. W.

Cotter

L. P.

(2009). Perceptions of sex offenders about treatment: Satisfaction and engagement in group therapy. Sexual Abuse: A Journal of Research and Treatment, 21(1), 35–56. https://doi.org/10.1177/1079063208326072

24.

Levenson

J. S.

Prescott

D. S.

D’Amora

D. A.

(2010). Sex offender treatment: Consumer satisfaction and engagement in therapy. International Journal of Offender Therapy and Comparative Criminology, 54(3), 307–326. https://doi.org/10.1177/0306624X08328752

25.

Levenson

M. R.

Kiehl

K. A.

Fitzpatrick

C. M.

(1995). Assessing psychopathic attributes in a noninstitutionalized population. Journal of Personality and Social Psychology, 68(1), 151–158. https://doi.org/10.1037/0022-3514.68.1.151

26.

Lilienfeld

S. O.

Andrews

B. P.

(1996). Development and preliminary validation of a self-report measure of psychopathic personality traits in noncriminal populations. Journal of Personality Assessment, 66(3), 488–524. https://doi.org/10.1207/s15327752jpa6603_3

27.

Lilienfeld

S. O.

Patrick

C. J.

Benning

S. D.

Berg

Sellbom

Edens

J. F.

(2012). The role of fearless dominance in psychopathy: Confusions, controversies, and clarifications. Personality Disorders: Theory, Research, and Treatment, 3(3), 327–340. https://doi.org/10.1037/a0026987

28.

Lynam

D. R.

Miller

J. D.

(2015). Psychopathy from a basic trait perspective: The utility of a five-factor model approach. Journal of Personality, 83(6), 611–626. https://doi.org/10.1111/jopy.12132

29.

Mann

R. E.

Hanson

R. K.

Thornton

(2010). Assessing risk for sexual violence: Some proposals on the nature of psychologically meaningful risk factors. Sexual Abuse: A Journal of Research and Treatment, 22(2), 191–217. https://doi.org/10.1177/1079063210366039

30.

McCallum

K. E.

Boccaccini

M. T.

Bryson

C. N.

(2017). The influence of risk assessment instrument scores on evaluators’ risk opinions and containment recommendations. Criminal Justice and Behavior, 44(9), 1213–1235. https://doi.org/10.1177/0093854817707232

31.

McNally

R. J.

(2016). Can network analysis transform psychopathology? Behaviour Research and Therapy, 86, 95–104. https://doi.org/10.1016/j.brat.2016.06.006

32.

Miller

H. A.

Amenta

A. E.

Conroy

M. A.

(2005). Sexually violent predator evaluations: Empirical evidence, strategies for professionals, and research directions. Law and Human Behavior, 29(1), 29–54. https://doi.org/10.1007/s10979-005-1398-y

33.

Mumma

G. H.

(2002). Effects of three types of potentially biasing information on symptom severity judgments for major depressive episode. Journal of Clinical Psychology, 58(10), 1327–1345. https://doi.org/10.1002/jclp.10046

34.

Neal

T. M. S.

Grisso

(2014). Assessment practices and expert judgment methods in forensic psychology and psychiatry: An international snapshot. Criminal Justice and Behavior, 41(12), 1406–1421. https://doi.org/10.1177/0093854814548449

35.

Olver

M. E.

Stockdale

K. C.

Wormith

J. S.

(2011). A meta-analysis of predictors of offender treatment attrition and its relationship to recidivism. Journal of Consulting and Clinical Psychology, 79(1), 6–21. https://doi.org/10.1037/a0022200

36.

Olver

M. E.

Wong

S. C. P.

(2009). Therapeutic responses of psychopathic sexual offenders: Treatment attrition, therapeutic change, and long-term recidivism. Journal of Consulting and Clinical Psychology, 77(2), 328–336. https://doi.org/10.1037/a0015001

37.

Patrick

C. J.

(2019). Handbook of psychopathy (2nd ed.). Guilford Press.

38.

Poythress

N. G.

Hall

J. R.

(2011). Psychopathy and impulsivity reconsidered. Aggression and Violent Behavior, 16(2), 120–134. https://doi.org/10.1016/j.avb.2011.02.003

39.

Preszler

Marcus

D. K.

Edens

J. F.

McDermott

B. E.

(2018). Network analysis of psychopathy in forensic patients. Journal of Abnormal Psychology, 127(2), 171–182. https://doi.org/10.1037/abn0000325

40.

Rufino

K. A.

Boccaccini

M. T.

Hawes

Murrie

D. C.

(2012). When experts disagreed, who was correct? A comparison of PCL-R scores from independent raters and opposing forensic experts. Law and Human Behavior, 36(6), 527–537. https://doi.org/10.1037/h0093988

41.

Sewall

L. A.

Olver

M. E.

(2019). Psychopathy and treatment outcome: Results from a sexual violence reduction program. Personality Disorders: Theory, Research, and Treatment, 10(1), 59–69. https://doi.org/10.1037/per0000297

42.

Skeem

J. L.

Cooke

D. J.

(2010). Is criminal behavior a central component of psychopathy? Conceptual directions for resolving the debate. Psychological Assessment, 22(2), 433–445 . https://doi.org/10.1037/a0008512

43.

Skovran

L. C.

Huss

M. T.

Scalora

M. J.

(2010). Sexual fantasies and sensation seeking among psychopathic sexual offenders. Psychology, Crime & Law, 16(7), 617–629. https://doi.org/10.1080/10683160902998025

44.

Texas Health & Safety Code § 841.000-841.150. (2000).

45.

Tsang

Salekin

R. T.

(2019). The network of psychopathic personality traits: A network analysis of four self-report measures of psychopathy. Personality Disorders: Theory, Research, and Treatment, 10(3), 246–256. https://doi.org/10.1037/per0000319.supp

46.

Verschuere

Te Kaat

(2019). What are the core features of Psychopathy? A protypicality analysis using the Psychopathy Checklist-Revised (PCL-R). Journal of Personality Disorders, 34(3), 410–419. https://doi.org/10.1521/pedi_2019_33_396

47.

Verschuere

van Ghesel Grothe

Waldorp

Watts

A. L.

Lilienfeld

S. O.

Edens

J. F.

Skeem

J. L.

Noordhof

(2018). What features of psychopathy might be central? A network analysis of the Psychopathy Checklist-Revised (PCL-R) in three large samples. Journal of Abnormal Psychology, 127(1), 51–65. https://doi.org/10.1037/abn0000315

48.

Viljoen

J. L.

McLachlan

Vincent

G. M.

(2010). Assessing violence risk and psychopathy in juvenile and adult offenders: A survey of clinical practices. Assessment, 17(3), 377–395. https://doi.org/10.1177/1073191109359587

49.

Wong

Hare

R. D.

(2005). Guidelines for a psychopathy treatment program. Multi-Health Systems.

50.

Wong

S. C. P.

(2015). Treatment of violence prone individuals with psychopathic personality traits. In Livesley

Dimaggio

Clarkin

(Eds.), Integrated treatment of personality disorder: A modular approach (pp. 345–376). Guilford Press.