Law and Human Behavior

, 33:497

The Psychological Inventory of Criminal Thinking Styles and Psychopathy Checklist: Screening Version as Incrementally Valid Predictors of Recidivism

Authors

    • Psychology Services, Federal Correctional Institution
Original Article

DOI: 10.1007/s10979-008-9167-3

Cite this article as:
Walters, G.D. Law Hum Behav (2009) 33: 497. doi:10.1007/s10979-008-9167-3

Abstract

A follow-up of 107 male federal prison inmates previously tested with the Psychological Inventory of Criminal Thinking Styles (PICTS) and Psychopathy Checklist: Screening Version (PCL:SV) was conducted to test the incremental validity of both measures. The PICTS General Criminal Thinking (GCT) score was found to predict general recidivism and serious recidivism when age, prior charges, and the PCL:SV were controlled. The PCL:SV, on the other hand, failed to predict general and serious recidivism when age, prior charges, and the PICTS were controlled. These findings support the hypothesis that content-relevant self-report measures like the PICTS are capable of predicting crime-relevant outcomes above and beyond the contributions of basic demographic variables like age, criminal history, and such popular non-self-report rating procedures as the PCL:SV.

Keywords

PICTSPCL:SVRecidivism predictionPrison inmates

A major tenet of the lifestyle theory of crime is that criminal behavior is a reflection of criminal thought process (Walters, 1990). In an effort to assess criminal thought process, Walters (1995) developed the Psychological Inventory of Criminal Thinking Styles (PICTS). The PICTS is designed to measure the eight thinking styles held to be instrumental in promoting and advancing a criminal lifestyle: Mollification, Cutoff, Entitlement, Power Orientation, Sentimentality, Superoptimism, Cognitive Indolence, and Discontinuity. Mollification is the process by which criminal actions are blamed on external events, whereas the Cutoff entails the rapid elimination of common deterrents to crime. Entitlement involves a sense of ownership, privilege, and conceit; Power Orientation, by contrast, is the unbridled pursuit of power and control over others. Sentimentality is an attempt to justify past criminal actions by engaging in seemingly benevolent behaviors, whereas Superoptimism is a belief in one’s ability to continue committing crime without experiencing the negative consequences. Cognitive Indolence involves taking short-cuts, acting impulsively, and lack of critical thinking; discontinuity, by comparison, is a lack of consistency in thought and action. The question, however, is how well does this measure of criminal thought process predict future criminal behavior?

Studies assessing the predictive validity of the PICTS indicate that most of the PICTS thinking style scales are capable of predicting future disciplinary infractions (Walters, 1996, 2005b; Walters & Elliott, 1999; Walters & Mandell, 2007; Walters & Schlauch, 2008) and recidivism (Palmer & Hollin, 2004; Walters, 1997, 2005a; Walters & Elliott, 1999). Predictive research on the PICTS is hampered, however, by several factors. First, the same thinking style scale rarely serves as the premier PICTS predictor in more than one study. Hence, the Cutoff (Walters, 1997, 2005a), Entitlement (Walters, 2005a; Walters & Elliott, 1999), Power Orientation (Walters, 1996), Sentimentality (Walters & Elliott, 1999), Superoptimism (Palmer & Hollin, 2004), and Cognitive Indolence (Healy & O’Donnell, 2006) scales have all been identified as the premier predictor or correlate of institutional adjustment or recidivism in one or more studies. As a consequence, the General Criminal Thinking (GCT) score was created by summing the raw scores of all eight PICTS thinking style scales. The GCT score enjoys higher levels of reliability and validity than any of the PICTS thinking style scales (Walters & Mandell, 2007; Walters & Schlauch, 2008).

A second limitation of PICTS predictive research is that inmates participating in these studies have often been program participants (Walters, 2005a, 2005b, Walters & Elliott, 1999) or inmates tested just prior to their release from prison (Palmer & Hollin, 2004). Program participants often record lower rates of disciplinary infractions and recidivism than non-participants (Wilson, Gallagher, & Mackenzie, 2000). Consequently, findings obtained with program participants may not generalize to non-program participants. The alternative strategy of testing general population inmates just prior to their release from prison is equally problematic if participants believe that staff will use the results to evaluate their readiness for release. Conversely, if the test results are kept strictly confidential, this raises questions about how well these findings generalize to real-life situations in which clinical staff have access to the PICTS. In an effort to resolve this problem several recent studies have used test results from inmates who completed the PICTS as part of a routine intake screening and uncovered evidence supporting the measure’s predictive validity (Walters & Mandell, 2007; Walters & Schlauch, 2008).

A third limitation of predictive research on the PICTS is that even though demographic variables like age and historical variables like prior disciplinary infractions are often controlled in these studies, participants’ scores on non-self-report rating procedures like the Psychopathy Checklist-Revised (PCL-R: Hare, 2003), Psychopathy Checklist: Screening Version (PCL:SV: Hart, Cox, & Hare, 1995), Level of Service Inventory-Revised (LSI-R: Andrews & Bonta, 1995), or Lifestyle Criminality Screening Form (LCSF: Walters, White, & Denney, 1991) are usually not controlled. Walters (2006) asserts that content-relevant self-report measures like the PICTS, Criminal Sentiments Scale (Andrews & Wormith, 1984), and Self-Appraisal Questionnaire (Loza, Dhaliwal, Kroner, & Loza-Fanous, 2000) possess adequate incremental validity when basic demographic measures like age and popular non-self-report rating procedures are controlled. This has been borne out in studies where the PICTS has been used to predict institutional adjustment. In these studies the PICTS has been found to possess incremental validity relative to the PCL:SV (Walters & Mandell, 2007), LSI-R:SV (Walters & Schlauch, 2008), and LCSF (Walters, 2005b) as a predictor of institutional adjustment. With the exception of two small-scale studies assessing the incremental validity of the PICTS relative to the LCSF, however, there are no incremental validity data on the PICTS as a predictor of recidivism (Walters, 1997; Walters & Elliott, 1999).

There is a large body of research showing that the PCL-R predicts violence and recidivism in psychiatric and criminal populations (Douglas, Vincent, & Edens, 2006). Although the literature on the PCL:SV is not as extensive as the literature on the PCL-R, research denotes that the PCL:SV may also be effective in predicting violence and recidivism. Studies indicate that the PCL:SV is capable of predicting institutional violence in male prison inmates (Belfrage, Fransson, & Strand, 2000), female jail inmates (Rogers et al., 2000), and male forensic patients (Hill, Rogers, & Bickford, 1996), as well as being capable of predicting community violence and recidivism in civil psychiatric patients (Skeem & Mulvey, 2001) and released inmates (Douglas, Yeomans, & Boer, 2005). There has been virtually no research, however, testing the incremental validity of the PCL:SV as a predictor of recidivism relative to content-relevant self-report measures like the PICTS.

The purpose of this study is to appraise the incremental validity of the PICTS and PCL:SV in a moderately sized sample (N > 100) of released federal prisoners. If, on the one hand, the PICTS is an incrementally valid predictor of criminal justice outcomes then it should be able to predict recidivism, just as it has previously predicted institutional adjustment (Walters & Mandell, 2007), above and beyond the contributions of age, prior criminality, and the PCL:SV. If, on the other hand, the PCL:SV is an incrementally valid predictor of criminal justice outcomes then it should be able to predict recidivism above and beyond the contributions of age, prior criminality, and the PICTS. It is hypothesized that the PICTS GCT score and PCL:SV total score or Cooke and Michie factor scores will continue to predict general and serious recidivism at Block 2 of a Cox regression analysis after age, number of prior charges, and the alternative measure (the PCL:SV in the case of the PICTS and the PICTS in the case of the PCL:SV) have been entered at Block 1.

Method

Participants

Participants were 107 male federal prisoners who completed the PICTS between August 2003 and February 2004, were scored on the PCL:SV as part of the Walters and Mandell (2007) investigation, and were released from custody at least 6 months before the end of the follow-up (August 21, 2008). Participants were, on average, 36.02 years (SD = 8.98) of age, with 11.08 years (SD = 2.10) of education. The ethnic breakdown for the sample was 30.8% White, 54.2% Black, and 15.0% Hispanic. Half of the participants (50.5%) listed their marital status as single, with 30.8%, 16.8%, and 1.9% identifying themselves as married, divorced/separated, and widowed, respectively. The modal instant offense for inmates participating in this study was a drug offense (35.5%), followed by robbery (17.8%), violence (15.9%), firearm violations (12.1%), miscellaneous offenses (11.2%), and property crimes (7.5%).

Measures

The Psychological Inventory of Criminal Thinking Styles (PICTS: Walters, 1995) is an 80-item self-report measure designed to assess the eight thinking styles that support a criminal lifestyle. Each PICTS item is scored on a four-point Likert-type scale (strongly agree, agree, uncertain, disagree) in which strongly agree responses contribute four points, agree responses three points, uncertain responses two points, and disagree responses one point to a scale. The one exception to this rule is the Defensiveness-revised (Df-r) scale, where items are reverse-scored (strongly agree = 1, agree = 2, uncertain = 3, strongly disagree = 4). The PICTS is composed of two 8-item validity scales—Confusion-revised (Cf-r) and Df-r—eight 8-item non-overlapping thinking style scales—Mollification (Mo), Cutoff (Co), Entitlement (En), Power Orientation (Po), Sentimentality (Sn), Superoptimism (So), Cognitive Indolence (Ci), and Discontinuity (Ds)—two content scales—Current Criminal Thinking (CUR) and Historical Criminal Thinking (HIS)—and four factor scales—Problem Avoidance (PRB), Infrequency (INF), Self-Assertion/Deception (AST), and Denial of Harm (DNH). The General Criminal Thinking (GCT) score, computed by summing the raw scores of all eight thinking style scales, is higher in reliability (Cronbach α = .93, test–retest r = .81–.93) and validity (correlations with institutional adjustment = .18–.38; correlations with release outcome = .23–.26) than any of the thinking style scales and will represent the PICTS in the present investigation. Sample items for each of the thinking style scales that make up the GCT are listed in Table 1.
Table 1

Sample items for the eight PICTS Thinking Style Scales

Scale

Sample item

Mollification (Mo)

I have told myself that I would never have had to engage in crime if I had had a good job.

Cutoff (Co)

I have used alcohol or drugs to eliminate fear or apprehension before committing a crime.

Entitlement (En)

The way I look at it, I’ve paid my dues and am therefore justified in taking what I want.

Power Orientation (Po)

When not in control of a situation I feel weak and helpless and experience a desire to exert power over others.

Sentimentality (Sn)

As I look back on it now, I was a pretty good guy even though I was involved in crime.

Superoptimism (So)

The more I got away with crime the more I thought there was no way the police or authorities would ever catch up with me.

Cognitive Indolence (Ci)

I tend to put off until tomorrow what should have been done today.

Discontinuity (Ds)

There have been times when I have made plans to do something with my family and then cancelled these plans so that I could hang out with friends, use drugs, or commit crimes.

The 12-item Psychopathy Checklist: Screening Version (PCL:SV: Hart et al., 1995) is a shortened version of the full 20-item PCL-R. Trained raters score each PCL:SV item using a three-point scale (0 = item does not apply; 1 = item applies to a certain extent; 2 = item applies). The PCL-R/SV is hierarchically organized with the individual PCL-R/SV items at the bottom, the total PCL-R/SV score at the top, and the PCL-R/SV factor scores in between. The PCL-R/SV items are divided into two parts which correspond with the two principal factors identified in research on the PCL-R: Part 1 (Affective/Interpersonal Traits) and Part 2 (Antisocial/Unstable Lifestyle). Although the two-factor model has received support in factor analytic studies on the PCL:SV, there is evidence that a three-factor model proposed by Cooke and Michie (2001) may also provide a good fit for the PCL:SV (Guy & Douglas, 2006). Each factor in the Cooke and Michie (2001) model is composed of three items: Factor 1: Arrogant and Deceitful Interpersonal Style (Superficial, Grandiose, Deceitful); Factor 2: Deficient Affective Experience (Lacks Remorse, Lacks Empathy, Doesn’t Accept Responsibility); Factor 3: Impulsive and Irresponsible Behavioral Style (Impulsive, Lacks Goals, Irresponsible). The reliability and validity of the PCL-R and PCL:SV are well documented (Hare & Neumann, 2006). Previously, the lone rater for this study and another trained rater independently scored the PCL:SV for 25 randomly selected cases and achieved moderately high inter-rater reliability (intraclass correlation coefficient = .75).

Procedure

Between August 2003 and February 2004, 213 inmates completed the PICTS as part of the routine intake procedure at a medium security federal prison. Because the PICTS data were collected for routine clinical purposes informed consent was not obtained. Permission was nonetheless solicited and granted by the Federal Bureau of Prisons’ Institutional Review Board to use the PICTS data for research and score the PCL:SV from an inmate’s central file. Of the 227 new arrivals at the institution where this study took place, 12 could not read well enough to complete the PICTS, 2 refused to be tested, 3 had insufficient chart data to score the PCL:SV, 6 produced invalid PICTS results, and 97 were not released at least 6 months prior to the end of the follow-up period. PICTS protocols with more than 10 unanswered items or a T-score on the Cf-r scale greater than 100 were considered invalid for the purpose of this study. There were five potential participants who failed to answer more than 10 PICTS items and one potential participant who scored above a T-score of 100 on the Cf-r. The remaining 107 inmates served as participants in this study.

The PCL:SV was scored from information contained in an inmate’s pre-sentence investigation (PSI) report. Before completing any ratings, the rater who scored all of the PCL:SVs for this study participated in a 3-day seminar where she was trained in all aspects of PCL-R/SV administration and scoring by a licensed doctoral-level clinical psychologist certified in the use of the PCL-R/SV. The rater for this study had no personal contact with the inmates on whom she scored the PCL:SV, was blind to the PICTS results, and had no access to the behavioral outcome data (disciplinary infractions or recidivism) used in either the present study or previous Walters and Mandell (2007) investigation. The PICTS was administered at intake and before any inmates were released from custody; the PCL:SV was scored 12–17 months after the PICTS and after almost 40% of the participants had been released from prison. This study can nonetheless be treated as a prospective investigation because the PCL:SV was based solely on a data source (PSI) that had been in existence prior to the onset of this study. Furthermore, all but one of the recidivism incidents occurred after the last PCL:SV had been scored.

The follow-up period began when the first inmate was released from custody (April 2, 2004) and ended on August 21, 2008. Time until the event (recidivism) or censoring (end of follow-up) was measured in months and ranged from 1 to 53 (M = 23.60, SD = 14.40, Md = 21.00). Recidivism data were obtained from a review of electronic files maintained at the Federal Bureau of Investigation (FBI) National Crime Information Center (NCIC) and Federal Bureau of Prisons (BOP) federal inmate database. Charges subsequent to an inmate’s release from custody were grouped into two categories: more serious charges and less serious charges. More serious charges were for offenses that someone without a prior criminal record would likely serve jail or prison time if convicted (murder, rape, robbery, assault, kidnaping, burglary, larceny, serious property damage, fraud, identity theft, possession of drugs with intent to distribute). Less serious charges were for offenses that someone without a prior criminal record would probably not serve time if convicted (simple drug possession, driving while intoxicated, shoplifting) or for violations that would not be considered criminal if committed by someone without a criminal record (parole/supervised release violation, absconding from supervision, escaping from a halfway house, firearm possession). This two-group demarcation corresponds roughly with the traditional felony-misdemeanor breakdown. Two sets of Cox regression survival analyses were performed, one for all charges (n = 54, 50.5%) and one for more serious charges (n = 34, 31.8%), in order to test the incremental validity of the PICTS and PCL:SV relative to age, prior criminality (number of prior charges), and each other.

Results

Table 2 provides descriptive statistics and intercorrelations for the seven predictor variables. The point biserial correlations between each predictor (age, prior charges, PICTS GCT, PCL:SV total score, PCL:SV factor 1 score, PCL:SV factor 2 score, PCL:SV factor 3 score) and the two dichotomized outcome measures of recidivism employed in this study (all charges, more serious charges) as well as the receiver operating characteristic (ROC) figures for each predictor are reproduced in Table 3. As the results listed in Table 3 indicate, prior charges and the GCT univariately predicted both general recidivism (total charges) and serious recidivism (more serious charges), whereas the PCL:SV factor 3 score univariately predicted serious recidivism.
Table 2

Descriptive statistics and correlations for the seven predictor variables

Predictor

Mean

SD

Range

Correlations

Priors

GCT

PCL-T

PCL-F1

PCL-F2

PCL-F3

Age

36.02

8.98

18–65

.35***

−.12

.21*

.24*

.34***

.04

Prior charges

11.46

9.20

1–55

 

.26**

.32***

.20*

.21*

.26**

PICTS GCT score

111.69

26.58

67–191

  

.21*

.10

.12

.22*

PCL:SV total score

13.29

4.22

3–23

   

.60***

.82***

.69***

PCL:SV F1 score

1.34

1.44

0–6

    

.43***

.26**

PCL:SV F2 score

3.47

1.60

0–6

     

.39***

PCL:SV F3 score

4.46

1.32

1–6

      

Note: PICTS GCT score Psychological Inventory of Criminal Thinking Styles General Criminal Thinking score, PCL:SV Psychopathy Checklist: Screening Version, F1 Factor 1, F2 Factor 2, F3 Factor 3, SD standard deviation, range = high and low scores for that particular measure in the current sample, N = 107

p < .05; ** p < .01; *** p < .001

Table 3

Point biserial correlations and ROC results for the seven predictors

Predictor

Outcome

All charges

More serious charges

r

AUC

r

AUC

Age

−.09 (.368)

.451 (.340–.561)

−.14 (.136)

.424 (.312–.537)

Prior charges

.24 (.014)

.648 (.544–.752)

.27 (.005)

.682 (.574–.791)

PICTS GCT score

.25 (.010)

.631 (.526–.736)

.26 (.006)

.655 (.546–.765)

PCL:SV total score

.01 (.915)

.506 (.396–.616)

.12 (.217)

.571 (.460–.682)

PCL:SV factor 1 score

−.12 (.220)

.445 (.336–.554)

−.05 (.622)

.488 (.373–.603)

PCL:SV factor 2 score

−.01 (.882)

.490 (.380–.600)

.04 (.688)

.515 (.402–.627)

PCL:SV factor 3 score

.12 (.227)

.554 (.444–.663)

.24 (.014)

.632 (.522–.742)

Note: Outcome = outcome measure, general recidivism (all charges) or serious recidivism (more serious charges); predictor = predictor variable; prior charges = number of prior charges; PICTS GCT score = Psychological Inventory of Criminal Thinking Styles General Criminal Thinking score; PCL:SV = Psychopathy Checklist: Screening Version; r = point-biserial correlation, the first number in the column is the point-biserial correlation and the second number (in parentheses) is the significance level of the correlation; AUC = area under the curve, the first number in the column is the AUC value and the range (in parentheses) is the 95% confidence interval for the AUC; N = 107

When age, prior charges, and the PCL:SV total score were entered at Block 1 and the PICTS GCT score was entered at Block 2 of a Cox regression analysis of general recidivism (all charges), significant effects were observed in both Blocks 1 (χ2 (3) = 7.95, p < .05) and 2 (χ2 (1) = 7.74, p < .01). Pairing the PICTS GCT with age and prior charges at Block 1 of a Cox regression analysis of general recidivism and entering the PCL:SV total score at Block 2 produced a significant Block 1 effect (χ2 (3) = 15.65, p < .01) but no Block 2 effect (χ2 (1) = 0.04, p = .85). Replacing the PCL:SV total score with the three Cooke and Michie (2001) factor scores did not substantially alter these results. The PICTS GCT score displayed incremental validity relative to age, prior charges, and the three PCL:SV factor scores (Block 2: χ2 (1) = 8.05, p < .01) but the three PCL:SV factor scores failed to achieve incremental validity relative to age, prior charges, and the PICTS GCT score (Block 2: χ2 (3) = 3.15, p = .37).

Entering age, prior charges, and the PCL:SV total score at Block 1 and the PICTS GCT score at Block 2 of a Cox regression analysis of serious recidivism (more serious charges) revealed significant effects at Blocks 1 (χ2 (3) = 13.88, p < .01) and 2 (χ2 (1) = 4.49, p < .05). When age, prior charges, and the PICTS GCT score were entered at Block 1 and the PCL:SV total score was entered at Block 2, Block 1 was significant (χ2 (3) = 17.32, p < .001) but Block 2 was not (χ2 (1) = 1.05, p = .31). Inserting the three PCL:SV factor scores into the regression equation did not significantly change the overall pattern of results. The PICTS GCT score achieved incremental validity when age, prior charges, and the three PCL:SV factor scores were controlled (Block 2: χ2 (1) = 3.89, p < .05) but the three PCL:SV factor scores failed to attain incremental validity when age, prior charges, and the PICTS GCT score were controlled (Block 2: χ2 (3) = 5.72, p = .13). The Wald and odds ratio results for age, prior charges, the PICTS, and the PCL:SV in all four analyses are presented in Table 4.
Table 4

Cox regression results for age, the PICTS GCT score, and the PCL-R/SV total and factor scores

 

Cox regression 1

Cox regression 2

Cox regression 3

Cox regression 4

Outcome

All charges

All charges

More serious charges

More serious charges

Predictor

Wald

exp(\( {{\upbeta}}_{\text{x}}^{\text{s}} \))/95% CI

Pw

Wald

exp(\( {{\upbeta}}_{\text{x}}^{\text{s}} \))/95% CI

Pw

Wald

exp(\( {{\upbeta}}_{\text{x}}^{\text{s}} \))/95% CI

Pw

Wald

exp(\( {{\upbeta}}_{\text{x}}^{\text{s}} \))/95% CI

Pw

Age

1.01

0.85(0.61–1.17)

.202

0.40

0.89(0.63–1.27)

.124

2.66

0.70(0.45–1.08)

.509

2.26

0.70(0.44–1.12)

.509

Prior charges

4.59*

1.30(1.02–1.66)

.444

4.34*

1.32(1.02–1.70)

.488

4.66*

1.38(1.03–1.86)

.432

5.42*

1.49(1.06–2.08)

.601

PICTS GCT

7.96**

1.49(1.13–1.96)

.791

8.34**

1.52(1.14–2.02)

.828

4.45*

1.44(1.03–2.03)

.526

3.87*

1.43(1.00–2.05)

.511

PCL:SV total

0.04

0.97(0.72–1.31)

.040

   

1.05

1.22(0.83–1.80)

.196

   

PCL:SV factor 1

   

2.40

0.76(0.54–1.07)

.477

   

2.39

0.68(0.42–1.11)

.572

PCL:SV factor 2

   

0.03

1.05(0.76–1.44)

.053

   

1.50

1.30(0.86–1.96)

.306

PCL:SV factor 3

   

0.70

1.15(0.83–1.58)

.161

   

2.44

1.48(0.91–2.40)

.586

Outcome = outcome measure, general recidivism (all charges) or serious recidivism (more serious charges); predictor = predictor variables; prior charges = number of prior charges; PICTS GCT = Psychological Inventory of Criminal Thinking Styles General Criminal Thinking score; PCL:SV = Psychopathy Checklist: Screening Version; Wald = Wald statistic with a χ2 distribution and one degree of freedom; exp(β) = exponent of the x-standardized (M = 0, SD = 1) coefficient in the form of an odds ratio (numbers below 1.00 indicate a negative relationship with the criterion and numbers above 1.00 indicate a positive relationship with the criterion); 95% CI = 95th percentile confidence interval for the exponent of the estimated coefficient; Pw = power as calculated with formulae from Hsieh & Lavori (2000)

p < .05; ** p < .01

Discussion

The purpose of this study was to assess the incremental validity of the PICTS and PCL:SV as predictors of recidivism. Walters (2006) notes that while content-diffuse self-report measures like the Minnesota Multiphasic Personality Inventory-2 (MMPI-2: Butcher et al., 2001) are often inferior to non-self-report rating procedures like the PCL-R/SV, LSI-R, and Violence Risk Appraisal Guide (VRAG: Quinsey, Harris, Rice, & Cormier, 1998) in predicting criminal justice outcomes, content-relevant self-report measures are often as effective as non-self-report rating procedures in making these types of determinations. One such procedure, the PICTS, has demonstrable incremental validity in predicting prison disciplinary adjustment relative to the PCL:SV (Walters & Mandell, 2007), LSI-R:SV (Walters & Schlauch, 2008), and LCSF (Walters, 2005b). The present study contributes to our understanding of the PICTS by showing that the PICTS GCT score successfully predicted both general and serious recidivism when entered into a Cox regression survival analysis behind age, prior criminality, and the PCL:SV. The statistically significant but modest effect recorded in this study suggests that a one standard deviation increase in the GCT led to a 49% increase in subsequent charges and a 44% increase in subsequent serious charges. These findings support the incremental validity of the PICTS and verify Andrews and Bonta’s (2003) contention that criminal cognition is a significant predictor of recidivism. The PCL:SV total score and Cooke and Michie (2001) factor scores, on the other hand, failed to demonstrate incremental validity after being entered into a Cox regression equation behind age, prior criminality, and the PICTS GCT score.

Two previous studies assessing the incremental validity of the PICTS GCT score as a predictor of recidivism relative to the LCSF produced mixed results. One small-scale recidivism study (N = 63) showed support for the incremental validity of the PICTS GCT score relative to the LCSF (Walters, 1997), whereas another small-scale recidivism study (N = 58) failed to generate a statistically significant incremental validity effect for the PICTS GCT score relative to the LCSF (Walters & Elliott, 1999). Using a moderately sized group of consecutively sampled prison inmates the current study uncovered evidence of incremental validity for recidivism prediction in the PICTS GCT score after another non-self-report rating procedure, the PCL:SV, was controlled. The current findings, in conjunction with previous incremental validity studies on institutional adjustment (Walters, 2005b; Walters & Elliott, 1999; Walters & Mandell, 2007; Walters & Schlauch, 2008) and non-incremental validity studies on recidivism (Palmer & Hollin, 2004; Walters, 2005a), indicate that the PICTS is capable of predicting clinically meaningful outcomes with modest efficacy.

In the current investigation the PCL:SV was less successful in predicting recidivism than it was in predicting disciplinary adjustment in a sample composed of many of the same participants (Walters & Mandell, 2007). The absence of a robust effect for the PCL:SV also conflicts with the largely positive results obtained when the PCL:SV has been used to predict recidivism in released prison inmates and forensic patients (Douglas et al., 2006). In a recent multi-sample investigation, Walters, Knight, Grann, and Dahle (2008) determined that factor 4 of the PCL-R/SV (antisocial behavior) was generally superior to the other three factor scores in predicting recidivism. Factor 4 was not included in the present investigation because it is not part of the Cooke and Michie (2001) model. Nevertheless, when factor 4 was added to the second and fourth Cox regression analyses the results remained unchanged, largely because the relationship between factor 4 and recidivism was weak.1 The reason why the PCL:SV in general and factor 4 of the PCL:SV in particular did not perform better in this study is unknown, although two possibilities suggest themselves. First, the fact that the PCL:SV was scored without benefit of an interview may have limited its ability to predict recidivism. Second, the relatively low power of the analyses, particularly the analyses where serious recidivism served as the outcome measure, may have interfered with the PCL:SV’s ability to predict recidivism.

The research design adopted in this study was limited in several respects. First, the manner in which recidivism was measured in this study (i.e., charges) is potentially problematic. Charges, although not free of bias, are much less biased than events occurring later in the criminal justice response sequence, such as convictions and incarceration (Mann, 1993). The downside to charge data is that they include a relatively large number of minor offenses. Consequently, a second set of analyses were conducted using a more stringent definition of recidivism, i.e., charges that if followed by conviction would likely lead to jail or prison time in an individual with no prior criminal record. Violent offenses were too infrequent (n = 12, 11.2%) and conviction status was too often absent from the present sample for either to serve as a meaningful definition of serious criminal recidivism. Fault, of course, could still be found with the decision to employ data from the NCIC to the extent that not all jurisdictions enter information into the system. These database limitations are mitigated somewhat by the supplemental use of the BOP data system and fact that incomplete data will create non-systematic (error) rather than systematic variance and therefore does not favor one predictor over another.

A second possible limitation of this study is the time span between administration of the PICTS and release from custody. The current investigation was the first PICTS recidivism study to use participants who were not program enrollees or who were not preparing for release. What the present study gains in clinical relevance and generalizability, however, it may lose in precision and predictive power. Outcome latency, as measured by the time span between administration of the PICTS and cessation of the follow-up, ranged from 52 to 60 months in the present study. Walters (in press) recently determined that the PICTS possesses maximum predictive efficacy for 48 months and loses much of its potency after 84 months. The time frame, beginning with administration of the PICTS and ending with termination of the follow-up, fell just outside the range of maximum predictive efficacy but well short of the range associated with minimum predictive efficacy. Even though the PICTS may have lost some of its potency near the end of the follow-up because of time-linked changes in criminal thinking, the loss was not enough to obfuscate the predictive efficacy of the PICTS.

Whereas the predictive effect achieved by the PICTS in this study may have been slightly attenuated by the 52–60 month gap between administration of the PICTS and completion of the follow-up, the effect observed in the current investigation is roughly equivalent to what has been observed in other predictive research on the PICTS. Univariate correlations between the PICTS GCT score and recidivism denote that the GCT accounted for 6–7% of the variance in recidivism recorded by participants in this study. An effect size of 6–7% is well within the range of effect sizes observed in previous prediction studies on the PICTS GCT score (3–14%: Walters, 1996, 1997, 2005a, 2005b; Walters & Elliott, 1999; Walters & Mandell, 2007; Walters & Schlauch, 2008). Every study that has examined the predictive efficacy of the PICTS has uncovered a significant yet modest univariate effect for the PICTS as a predictor of future recidivism and institutional adjustment. A small consistent effect, while it may account for only a small portion of the variance in an outcome, may nonetheless be important. Many doctors encourage their patients to take aspirin in the belief that taking aspirin reduces the risk of future heart attacks even though the coefficient of determination (r2) for aspirin in preventing future heart attacks is less than .01 (Rosenthal, 1994). Furthermore, a small consistent effect may be even more important when combined with other small consistent effects.

A third potential limitation of this study is that the study’s author was also the author of the PICTS, one of the predictor variables examined in this study. Allegiance effects, whereby the author of a theory or procedure finds more favorable results for his/her theory or procedure than unaffiliated or independent researchers, have been observed with different schools of psychotherapy (Staines & Cleland, 2007) as was as with different actuarial risk assessment procedures (Blair, Marcus, & Boccaccini, 2008). This implies that independent verification of the current results by researchers with no allegiance to either the PICTS or the PCL:SV is required.

The present results suggest several avenues for future research. First, investigators need to expand the number of non-self-report rating procedures against which the PICTS is compared. We currently have evidence that the PICTS GCT score possesses incremental validity relative to the PCL:SV (Walters & Mandell, 2007; present study), LSI-R:SV (Walters & Schlauch, 2008), and LCSF (Walters, 1997, 2005b; Walters & Elliott, 1999). What remains to be determined is whether the PICTS possesses incremental validity relative to formal risk assessment procedures like the VRAG and Historical/Clinical/Risk Management-20 (HCR-20: Webster, Douglas, Eaves, & Hart, 1997). Considering the modest effect sizes recorded by the PICTS when it has been used to predict future recidivism and institutional adjustment it is vital that we identify additional self-report and non-self-report rating procedures that may complement the PICTS. Finding the proper combination of self-report and non-self-report rating instruments is therefore a second avenue of research on PICTS incremental validity. The present study lends further credence to Walters’ (2006) assertion that content-relevant self-report measures like the PICTS are capable of providing useful predictive information above and beyond the contributions made by popular non-self-report rating procedures like the PCL:SV. If we are to take full advantage of the PICTS and other content-relevant self-report measures of criminality it is imperative that we expand the range of non-self-report rating procedures to which the PICTS is incrementally valid and find ways to combine the modest predictive effect of the PICTS with the modest predictive effects of popular non-self-report rating procedures like the PCL:SV, LSI-R:SV, and LCSF.

Footnotes
1

The univariate correlation between factor 4 of the PCL:SV and general recidivism (all charges) was −.01 (p = .925) and the univariate correlation between factor 4 and serious recidivism (more serious charges) was .09 (p = .361).

 

Copyright information

© American Psychology-Law Society/Division 41 of the American Psychological Association 2008