Article Critique NOTES

Assessing the Proactive and Reactive Dimensions of Criminal Thought Process: Divergent Patterns of Correlation With Variable- and Person-Level Measures of Criminal Risk and Future Outcome

Glenn D. Walters

Department of Criminal Justice, Kutztown University

ABSTRACT The goal of this study was to determine whether measures of proactive and reactive criminal thinking display divergent patterns of correlation with outside criteria. A sample of 3,039 male medium-security federal prisoners who completed the Psychological Inventory of Criminal Thinking Styles (PICTS) served as participants in this study. Despite being highly correlated (r¼ .75), the PICTS proactive and reactive scales displayed divergent patterns of correlation with the eight risk/outcome measures. As predicted, the proactive scale corresponded with lower crim- inal risk, older age of first conviction, and decreased odds of prior substance misuse and mental illness, whereas the reactive scale corresponded with higher criminal risk, earlier age of first con- viction, greater odds of prior substance misuse and mental illness, and more evidence of subse- quent arrest. Contrary to predictions, the proactive scale was associated with increased rather than decreased commission of disciplinary infractions in prison. When participants with elevated proactive scores were compared to participants with elevated reactive scores on the eight risk/out- come variables, the results revealed that the two profiles were moderately negatively correlated. Thus, although proactive criminal thinking is associated with below-average criminal risk and below-average future negative outcomes, reactive criminal thinking does just the opposite.

ARTICLE HISTORY Received 11 March 2018 Revised 11 July 2018

The proactive and reactive dimensions of criminal thought process (i.e., how rather than what an offender thinks) in Walters’s (2012) two-dimensional model of adult criminal thinking has its foundation in prior research on proactive and reactive childhood aggression. Like proactive and react- ive childhood aggression (Dodge & Coie, 1987; Poulin & Boivin, 2000), proactive and reactive criminal thinking over- lap extensively with one another yet appear to represent dis- tinct concepts or processes (Walters, Hagman, & Cohn, 2011; Walters & Yurvati, 2017). In other words, although correlating .50 or higher with each other, proactive and reactive aggression (Martinelli, Ackermann, Bernhard, Freitag, & Schwenck, 2018) and proactive and reactive crim- inal thinking (Walters, 2007) consistently display divergent patterns of association with various outside criteria, such as hostile attribution biases. A developmental progression is therefore proposed in which the instrumentality of proactive aggression gives rise to the planned and calculated features of antisocial cognition, referred to as proactive criminal thinking, and the impulsivity of reactive aggression gives rise to the reckless and emotional features of antisocial cog- nition, referred to as reactive criminal thinking (Walters, 2005). Taken as a whole, the two dimensions of criminal thought process explain the complex nature of crime and

the paradox of highly correlated scales that form divergent associations with the same external criteria.

Just as proactive and reactive childhood aggression have different external correlates (Koolen, Poorthuis, & van Aken, 2012; Swogger, Walsh, Maisto, & Conner, 2014; Urben et al., 2018), so, too, do proactive and reactive criminal thinking correlate differentially with the same external crite- ria. Research has fairly consistently demonstrated that react- ive criminal thinking correlates better with indexes of criminal risk, as represented by scores on the Lifestyle Criminal Screening Form (Walters, 1995; Walters & Elliott, 1999) and the second factor of the Psychopathy Checklist (Walters & Di Fazio, 2016), than does proactive criminal thinking. There is also evidence that whereas reactive crim- inal thinking mediates the past crime—future drug use rela- tionship, proactive criminal thinking does not (Walters, 2016). When it comes to predicting recidivism, proactive and reactive criminal thinking appear to correlate similarly with subsequent offending (see Walters, 2012), but the effect size of the reactive scale typically exceeds the effect size of the proactive scale when both scales are included as predic- tors in the same regression equation (Walters & Lowenkamp, 2016). Finally, although reactive criminal thinking tends to outperform proactive criminal thinking in

CONTACT Glenn D. Walters [email protected] Department of Criminal Justice, 361 Old Main, Kutztown University, Kutztown, PA 19530-0730. � 2018 Taylor & Francis Group, LLC

JOURNAL OF PERSONALITY ASSESSMENT 2020, VOL. 102, NO. 2, 223–230 https://doi.org/10.1080/00223891.2018.1508469

predicting institutional adjustment (Folk et al., 2016; Walters & Geyer, 2005), the opposite effect has also been found (Walters & Mandell, 2007).

Does the fact that measures of proactive and reactive aggression and criminal thinking overlap extensively mean that these scales are assessing the same construct, are redun- dant to one another, or do not warrant separate treatment and interpretation? Some might argue that it depends on the level of association between the two variables, yet two varia- bles can correlate extensively and still not be measuring the same construct (Cronbach & Meehl, 1955). Hence, a high correlation between two scores on a psychometric instrument should be considered a necessary but not sufficient condition for concluding that the two scores are measuring the same construct. Before it can be concluded that two scales are measuring the same construct, similar patterns of convergent and discriminant correlation should be observed between scores on these two scales (Smith, 2005; Westen & Rosenthal, 2005). Hence, if two scales correlate similarly with the same set of external criteria then it is more likely they are measur- ing the same construct, but if the scales achieve dissimilar patterns of correlation with the same set of external criteria then it is more likely that they are measuring different con- structs. The purpose of this investigation was to determine whether a criminal thinking measure designed to assess pro- active and reactive criminal thought process exhibits divergent patterns of correlation with external criteria despite a high degree of intercorrelation.

The Psychological Inventory of Criminal Thinking Styles (PICTS; Walters, 1995) is designed to assess criminal thought process by providing scores on scales of proactive and react- ive criminal thinking. As previously stated, proactive criminal thinking represents the planned, calculated, and emotionless features of the criminal thought process, whereas reactive criminal thinking encompasses the impulsive, irrational, and emotional aspects of the criminal thought process. Walters and Yurvati (2017) examined the construct validity of the proactive and reactive scales of the PICTS by correlating them with three putative measures of proactive criminal thought or cognitive insensitivity (Moral Disengagement: Bandura et al., 1996; Offending, Crime, and Justice Neutralization scale: Hamlyn et al., 2003; Denver Youth Survey [DYS] Neutralization scale: Huizinga & Jakob-Chien, 1998) and three putative measures of reactive criminal thought or cognitive impulsivity (Weinberger Adjustment Inventory–Impulse Control: Weinberger & Schwartz, 1990; National Longitudinal Survey of Youth–Child Risk-Taking scale: Center for Human Resource Research, 2009; DYS Impulsivity scale: Huizinga & Jakob-Chien, 1998). Zero-order correlations and regression coefficients revealed that the PICTS proactive scale corresponded significantly better with three putative proactive measures than with three putative reactive measures, whereas the PICTS reactive scale corre- sponded significantly better with three putative reactive meas- ures than with three putative proactive measures.

Because proactive criminal thinking encompasses the planned and calculated aspects of antisocial cognition and

reactive criminal thinking subsumes the impulsive and irre- sponsible aspects, a reasonable assumption is that reactive criminal thinking will be more closely tied to criminal risk factors and the negative consequences of a criminal lifestyle than proactive criminal thinking. In other words, the impul- sive and reckless nature of reactive criminal thinking makes it far more likely that the individual will engage in less suc- cessful patterns of criminality and be at greater risk for detection by law enforcement than the duplicity that evolves from proactive criminal thinking. This is discussed in the childhood aggression literature, where the aggressive actions of children who score higher on measures of reactive aggres- sion have a greater likelihood of coming to the attention of parents and school officials than the aggressive actions of children who score higher on measures of proactive aggres- sive (Card & Little, 2006; Rieffe et al., 2016). Although dif- ferences between proactive and reactive aggression have been consistently found at the variable level, the research is mixed when it comes to comparisons made at the person level (Carroll, McCarthy, Houghton, O’Connor, & Zadow, 2018; Smeets et al., 2017). Accordingly, this study examined differences in proactive and reactive criminal thinking at both the variable and person levels.

This study

The purpose of this investigation was to determine whether reactive criminal thinking, because of its impetuous and irresponsible nature, is more closely tied to criminal history risk than proactive criminal thinking, despite a moderate to high degree of intercorrelation between the two forms of criminal thought process. In the previously mentioned Walters and Yurvati (2017) study, proactive and reactive latent factors achieved divergent patterns of correlation with alternate measures of proactive and reactive criminal think- ing despite correlating .65 with each other. In the present study, a large group of incarcerated felons who had been administered the PICTS within several weeks of entering a medium-security federal prison were evaluated for criminal risk and future behavioral problems using both variable-level and person-level data. It was predicted that proactive and reactive criminal thinking would display divergent patterns of correlation at both the variable and person levels.

The research questions that drove this study were both conceptual and practical. Conceptually, this study was designed to determine whether proactive criminal thinking is less apt to be associated with criminal risk and poor out- comes than reactive criminal thinking, presumably because it is less subject to detection by law enforcement, just as proactive aggression is less subject to detection by parents and school officials than reactive aggression (Card & Little, 2006; Rieffe et al., 2016). Practically, this study was designed to determine whether administering measures of both pro- active and reactive criminal thinking is worthwhile, given an extensive degree of overlap between the two scales. It was hypothesized that historical measures of criminal risk (e.g., prior convictions, substance misuse) and prospective

224 WALTERS

measures of negative outcome (i.e., institutional misconduct and recidivism) would correlate positively with (variable- level analysis) and be above average (person-level analysis) on the reactive scale and correlate negatively with and be below average on the proactive scale.

Method

Participants

The sample for this study consisted of 3,039 male inmates who completed the PICTS as part of a routine intake evaluation for inmates entering a medium-security federal prison sometime between March 2003 and August 2010. This number represents over 95% of all inmates admitted to this medium-security insti- tution during the time period in which data were collected. The average age of participants at the time of evaluation was 35.0 years (SD¼ 9.87) and the racial and ethnic breakdown was 63.0% African American, 18.4% Hispanic, 17.2% White, 0.8% Asian, and 0.6% Native American.

Measures

The PICTS is an 80-item self-report measure designed to assess eight criminal thinking patterns or styles: mollifica- tion, cutoff, entitlement, power orientation, sentimentality, superoptimism, cognitive indolence, and discontinuity (Walters, 1995). Seven of the eight PICTS thinking style scales have been found to load onto one of two higher order factors referred to as proactive (mollification, entitlement, power orientation, and superoptimism) and reactive (cutoff, cognitive indolence, and discontinuity) criminal thinking. Whereas proactive criminal thinking reflects the planned, calculated, and callous or unemotional features of antisocial cognition, reactive criminal thinking reflects the impulsive, irresponsible, and emotional features. The internal consist- ency, stability, and predictive and construct validity of the PICTS dimensional scales (proactive and reactive) have received support in several studies conducted over the last several years (Walters, 2012).

Eight variables served as dependent variables in this study. Four of the variables were criminal history or criminal risk indicators: number of prior convictions, age at first conviction (in years), total score based on retrievable items from the Lifestyle Criminality Screening Form (LCSF; Walters, White, & Denney, 1991), and Facet 4 (Antisocial) of the Psychopathy Checklist–Revised (PCL–R; Hare, 2003). The PCL–R items were scored exclusively from file data (presentence investiga- tion report [PSI]) and were restricted to Facet 4 because these were the only items addressed with regularity in the PSI. Fifty randomly selected cases were independently rated on Facet 4 of the PCL–R by a second rater. These ratings were then com- pared to the original ratings using a two-way mixed effects model (absolute agreement, average measures). The results revealed that the raters achieved an above-average level of interrater agreement on the Facet 4 measure (intraclass correl- ation coefficient [ICC]¼ .84).

The last four dependent variables were prior substance misuse (yes–no), prior mental illness (yes–no), number of disciplinary reports received for institutional infractions dur- ing a 1- to 76-month (M¼ 30.03) period of incarceration, and number of subsequent arrests experienced during a 1- to 76-month (M¼ 25.33) follow-up. The regression analyses performed on the disciplinary reports and subsequent arrests outcome measures included time at risk in prison and time at risk in the community, respectively, as covariates, along with age and race. For the profile comparison portion of the study, number of disciplinary reports received was divided by number of months (time at risk) in federal prison to cre- ate a rate of disciplinary infractions indicator, and subse- quent arrests were divided by number of months (time at risk) in the community to create a rate of subsequent arrests indicator.

Data collection

Descriptive statistics were computed for the two independ- ent variables (proactive and reactive criminal thinking) and eight dependent variables (prior convictions, age at first con- viction, LCSF total score, Facet 4 of PCL–R, prior substance misuse, prior mental illness [schizophrenia, bipolar disorder, major depression], disciplinary infractions, and subsequent arrests) included in this study. Data for the independent variable came from the PICTS and data for the dependent variables came from a review of electronic files maintained by the Federal Bureau of Prisons (presentence investigation report, disciplinary files) or other federal law enforcement agencies (FBI National Crime Information Center). Data were complete for all measures except for subsequent arrests. This was because only 1,435 members of the study cohort had been released from custody at the time the arrest outcome data were being collected. The use of these data for research purposes was approved by the Federal Bureau of Prisons and Kutztown University institutional review boards.

Data analysis

Data were analyzed at both the variable and person levels. Eight regressions were performed at the variable level, one for each dependent variable in this study. The three continu- ous dependent variables (age at first conviction, LCSF total score, and PCL–R Facet 4 score) were assessed with stand- ard regression and a maximum likelihood (ML) estimator. The two dichotomous dependent variables (substance misuse and mental illness) were assessed with binomial logistic regression analysis and the three count-dependent variables (prior convictions, disciplinary reports, and subsequent arrests) were assessed with negative binomial regression. In the latter two regressions, a maximum likelihood with robust standard errors (MLR) estimator was employed. Age (in years) and race (White¼ 1, non-White¼ 2) were included as covariates in all eight regressions, whereas time spent in prison served as a third covariate in the regression

DIVERGENT PATTERNS OF CORRELATION 225

equation predicting disciplinary reports and time at risk in the community was added to the regression equation pre- dicting subsequent arrests. All analyses were performed with Mplus 8.1 (Muth�en & Muth�en, 1998–2017).

The second step of the data analysis entailed assigning individual cases to four patterns using clinical guidelines provided in the PICTS manual (Walters, 2013) and then performing several person-level analyses. The four patterns used in this study were an elevated proactive pattern (P�T score of 60, R<T score of 60), an elevated reactive pattern (P<T score of 60, R�T score of 60), an elevated proactive and reactive pattern (P�T score of 60 and R�T score of 60), and an unelevated pattern (P<T score of 60, R<T score of 60). The outcome measures were first standardized (z scores) and then the mean scores for each outcome were calculated. These mean score profiles were then compared across the four groups of patterns using the double-entry ICC (McCrae, 2008). This was done to determine the degree to which the four groups differed from one another on the eight outcome measures.

Results

Descriptive statistics for the two independent variables and eight dependent variables used in this study are summarized in Table 1. An intercorrelational matrix of the eight depend- ent variables revealed a modest degree of association between variables (M¼ .18, SD¼ .18, range¼ .02�.57), with the highest correlations (.42�.57) occurring between the four criminal history indicators (prior convictions, age at first conviction, LCSF total score, PCL–R Facet 4 score). The two independent variables (PICTS proactive and react- ive scales) correlated at r¼ .75.

Variable-level analyses

Table 2 summarizes the variable-level results attained by P and R in each of the eight regression analyses. P was associ- ated with reduced odds of achieving four outcomes (LCSF total score, PCL–R Factor 4 score, substance misuse, and mental illness), increased odds of achieving two outcomes (age at first conviction and disciplinary reports), and non- significant results on two outcomes (prior convictions and subsequent arrests). R was associated with increased odds of achieving six outcomes (LCSF total score, PCL–R Facet 4 score, substance misuse, mental illness, prior convictions, and subsequent arrests), reduced odds of achieving one out- come (age at first conviction), and nonsignificant results for one outcome (disciplinary reports).

With the exception of the association between higher P and increased odds of disciplinary infractions, these results are fully congruent with the research hypothesis tested in this study. Whereas the standardized regression coefficients were small to modest, the odds ratios obtained from the binomial logistic regression and negative binomial regression analyses were even smaller. These latter results consequently provide meaningful support for the conceptual goal of this study but are not particularly informative when it comes to

the practical goal of using the P and R scales to predict dichotomous and count risk and outcome measures.

Person-level analyses

The outcome profiles of individuals achieving elevated scores (T� 60) on the proactive scale, the reactive scale, the proactive and reactive scales, and neither scale are presented in Table 3 as person-level analyses. Assessing strength of relationship with the double-entry ICC, it was noted that the proactive and reactive patterns achieved a moderately strong inverse correlation with one another. Although the reactive pattern achieved a strong positive correlation with the dual elevation pattern, in which both P and R were ele- vated, the proactive pattern correlated positively, although only weakly, with the unelevated pattern.

The person-level results provide support for both the conceptual and practical objectives of this study. A correl- ation of –.56 between the risk/outcome patterns for inmates who elevated the proactive scale alone and risk/outcome pat- terns for inmates who elevated both the proactive and react- ive scales compared to a correlation of .69 between the risk/ outcome patterns for inmates who elevated the reactive scale alone and risk/outcome patterns for inmates who elevated both scales is striking. Coupled with the fact that the risk/ outcome patterns for the proactive group correlated minim- ally yet positively with the risk/outcome patterns for the unelevated group and the risk/outcome patterns for the reactive group correlated negatively with the risk/outcome patterns for the unelevated group, this suggests that inmates who elevated only the proactive scale were more similar to inmates who did not elevate either scale, whereas inmates who elevated only the reactive scale were more similar to inmates who elevated both scales.

Table 1. Descriptive statistics for the 10 variables included in this investigation.

Variable n M SD Range

Prior convictions 3,039 4.28 2.63 0–30 Age at first conviction 3,039 20.74 5.93 7–62 LCSF total score 3,039 4.57 1.88 0–10 PCL–R Facet 4 score 3,039 3.15 1.92 0–10 Disciplinary reports 3,039 1.44 2.63 0–41 Subsequent arrests 1,435 1.21 1.62 0–14 Proactive dimension 3,039 52.48 13.66 32–128 Reactive dimension 3,039 43.06 13.35 24–96

n No. (%) No. (%)

Substance misuse (yes–no) 3,039 2,055 (67.6%) 984 (32.4%) Mental illness (yes–no) 3,039 476 (15.7%) 2,563 (84.3%)

Note: Variable¼ postdicted or predicted dependent variable or one of the independent variables; prior convictions¼ prior criminal convictions; age at first conviction¼ age at time of first conviction; LCSF total score¼ total score from the Lifestyle Criminality Screening Form; PCL–R Facet 4 score¼ Facet 4 (antisocial) score from the Psychopathy Checklist–Revised; disciplinary reports¼ number of disciplinary reports received in prison con- trolling for time at risk; subsequent arrests¼ number of subsequent arrests following release from prison after controlling for time at risk; proactive dimension¼ Psychological Inventory of Criminal Thinking Styles (PICTS) Proactive (P) scale score; reactive dimension¼ PICTS Reactive (R) scale score; substance misuse¼ history of prior substance misuse versus no history of prior substance misuse; mental illness¼ history of mental illness versus no history of mental illness; n¼ participants with nonmissing data.

226 WALTERS

Discussion

As anticipated, the PICTS proactive and reactive scales were highly correlated, with the strength of correlation suggesting that the two scales shared more than half their variance in common. Despite extensive overlap, the scales displayed divergent patterns of association with six measures of crim- inal risk and two measures of future criminal outcome using both variable- and person-level analyses. In nearly every case the reactive scale was associated with higher levels of criminal risk and negative outcome, whereas the proactive scale was associated with lower levels of criminal risk and negative outcome. Hence, the reactive scale was associated with an above-average number of prior convictions, an ear- lier age of first conviction, higher LCSF and PCL–R risk scores, more evidence of prior substance misuse and mental illness, and a greater likelihood of subsequent arrest, whereas the proactive scale was associated with a below-average number of prior convictions, a later age of first conviction, lower LCSF and PCL–R risk scores, less evidence of prior substance misuse and mental illness, and significantly higher

levels of institutional infractions. These results are largely consistent with the notion that reactive criminal thinking, by virtue of its impulsive and irresponsible nature, is more likely to be associated with higher criminal risk and a greater proportion of future crime-related problems. These results corroborate prior findings from the Walters and Yurvati (2017) study, which also used the PICTS to assess criminal thought process, and suggest that proactive and reactive criminal thinking are distinct constructs, despite their overlap. The one inconsistent finding (i.e., above-aver- age institutional misconduct in relationship to proactive criminal thinking) warrants further discussion.

The relationship between institutional misconduct and proactive criminal thinking reminds us that proactive crim- inal thinking is not simply a less discriminating version of reactive criminal thinking. It was hypothesized that pro- active criminal thinking would correlate negatively with criminal risk and show better outcomes than reactive crim- inal thinking because it is not saddled with the impulsivity and low self-control that afflict reactive criminal thinking. It is for this reason that individuals with profiles in which only proactive criminal thinking is elevated might be less subject to detection by law enforcement than individuals with pro- files in which only reactive criminal thinking is elevated. Why, then, was proactive criminal thinking associated with a higher rate of institutional misconduct than reactive crim- inal thinking? Although prior research indicates that pro- active criminal thinking is associated with lower levels of institutional misconduct relative to reactive criminal think- ing (Folk et al., 2016; Walters & Geyer, 2005), there is at least one other study that agrees with the results reported here (Walters & Mandell, 2007). It is possible that the struc- ture provided by prison diminishes the role of reactive crim- inal thinking in the behaviors that lead to prison misconduct, such that proactive criminal thinking is just as likely to be associated with the violation of prison rules as reactive criminal thinking, if not more so, because in such a highly structured environment stealth and subterfuge are less likely to provide protection. That institutional miscon- duct correlated with proactive criminal thinking might mean that proactive criminal thinking is just as problematic as

Table 3. Mean scores and double-entry intraclass correlations for participants with elevated proactive profiles, elevated reactive profiles, elevated proactive and reactive profiles, and nonelevated profiles.

High Pa High Rb High P & Rc Nonelevatedd

Group means Prior convictions �0.153 0.200 0.198 �0.041 Age at first conviction 0.039 �0.201 �0.104 0.036 Substance misuse �0.013 0.183 0.117 �0.038 Mental illness �0.028 0.375 0.169 �0.067 LCSF total score �0.096 0.324 0.209 �0.062 PCL–R Facet 4 score �0.190 0.202 0.108 �0.024

Disciplinary reports rate 0.159 �0.013 0.132 �0.033 Subsequent arrests rate �0.050 �0.017 0.102 �0.011 Double-entry intraclass correlations High P �.55 �.56 .19 High R .69 �.44 High P & R �.77

Note. Group means¼ z scores; High P¼ participants with proactive (P) T scores �60 and reactive (R) T scores <60; High R¼ participants with P T scores <60 and R T scores �60; High P & R¼ participants with P T scores �60 and R T scores �60; nonelevated¼ participants with P T scores <60 and R T scores <60. LCSF¼ Lifestyle Criminality Screening Form; PCL–R¼ Psychopathy Checklist–Revised.

an¼ 191. bn¼ 256. cn¼ 353. dn¼ 2,239.

Table 2. Regression results for the proactive and reactive dimension scores.

Variable Proactive dimension Reactive dimension

Continuous outcomes b [95% CI] b z p b [95% CI] b z p Age at first conviction 0.037 [0.016, 0.058] 0.084 3.41 <.001 �0.063 [�0.085, �0.042] �0.143 �5.84 <.001 LCSF total score �0.017 [�0.025, �0.010] �0.125 �4.55 <.001 0.034 [0.026, 0.041] 0.240 8.86 <.001 PCL–R Facet 4 score �0.018 [�0.026, �0.011] �0.131 �4.76 <.001 0.028 [0.020, 0.035] 0.191 7.00 <.001

Dichotomous outcomes b [95% CI] OR z p b [95% CI] OR z p Substance misuse �0.011 [�0.020, �0.002] 0.971 �2.47 .014 0.023 [0.013, 0.032] 1.023 4.80 <.001 Mental illness �0.017 [�0.028, �0.005] 0.984 �2.92 .003 0.035 [0.024, 0.046] 1.036 6.28 <.001

Frequency count outcomes b [95% CI] exp(b) z p b [95% CI] exp(b) z p Prior convictions �0.002 [�0.005, 0.000] 0.998 �1.92 .054 0.007 [0.005, 0.010] 1.007 6.45 <.001 Disciplinary reports 0.008 [0.002, 0.015] 1.008 2.49 .013 0.001 [�0.006, 0.008] 1.001 0.31 .756 Subsequent arrests 0.001 [�0.006, 0.007] 1.001 0.16 .871 0.008 [0.001, 0.015] 1.008 2.22 .027

Note: Age (in years) and race (1¼White, 2¼ non-White) were included in each of the eight regressions as covariates; in addition, time at risk in federal prison served as a covariate in the disciplinary reports regression and time at risk in the community served as a covariate in the subsequent arrests regression. Variable¼ postdicted or predicted dependent variable; continuous outcomes were subjected to least squares multiple regression, dichotomous outcomes were subjected to binomial logistic regression analysis, and frequency count outcomes were subjected to negative binomial regression; b [95% CI]¼ unstandardized coefficient with the 95% confidence interval, b¼ standardized coefficient in least squares regression; OR¼ logistic regression odds ratio; exp(b)¼ incidence rate ratio; z¼Wald Z-test, p¼ significance level of the Wald Z-test.

DIVERGENT PATTERNS OF CORRELATION 227

reactive criminal thinking, although in less structured situa- tions someone with a proactive PICTS profile might have a better chance of avoiding detection by law enforcement than if they were in a more structured situation. This possibility requires further study.

Proactive and reactive aggression and criminal thinking

It should be noted that the results reported here place pro- active and reactive criminal thinking squarely within the broader context of research on proactive and reactive aggression. Although a fairly extensive body of research exists in support of the argument that proactive and reactive aggression represent distinct processes despite being highly correlated (Polman, Orobio de Castro, Koops, van Boxtel, & Merk, 2007), some researchers have questioned the mean- ingfulness of the proactive—reactive distinction in aggressive behavior (Bushman & Anderson, 2001). One reason for the skepticism is the degree of overlap and lack of orthogonality between the two constructs. Because much of the research on proactive and reactive aggression has been conducted at the variable level, researchers have started studying the pro- active—reactive question with both variable-level and per- son-level data. Adopting this approach, Smeets et al. (2017) observed variable-level differences between proactive and reactive aggression but failed to find consistent support for person-level differences. Carroll et al. (2018), by comparison, found meaningful distinctions between proactive and react- ive aggression at both the variable and person levels. This study is more in line with the Carroll et al. (2018) results in identifying meaningful differences between proactive and reactive criminal thinking at both the variable (regression analyses) and person (elevation patterns) levels despite a high degree of intercorrelation. In fact, the person-level find- ings were even stronger than the variable-level results in this study. This suggests that proactive and reactive criminal thinking, although not identical to proactive and reactive aggression, can be understood and studied within the larger context of the proactive—reactive aggression literature.

It would be a mistake to conclude on the basis of these results that proactive criminal thinking is less dangerous or problematic than reactive criminal thinking. In many ways, proactive criminal thinking might be more dangerous and more problematic than reactive criminal thinking. The fact that proactive criminal thinking is less likely to lead to immediate negative consequences than reactive criminal thinking—in other words, that criminal behavior inspired by proactive criminal thinking has a greater likelihood of going undetected, at least initially—does not make it innocuous. We need only consider the instrumental/proactive—expres- sive/reactive breakdown of homicide motives to find a paral- lel in another area of criminology to illustrate this point. In an early study on instrumental—expressive motives for homicide, Miethe and Drass (1999) discovered that 36% of the situational factors they examined were unique to instru- mental homicides, 30% were unique to expressive homicides, and 34% were common to both forms of homicide. Similar to proactive and reactive criminal thinking, instrumental

and expressive homicide are more different than they are similar, despite the fact many homicides are driven by a combination of instrumental and expressive motives (Adjorlolo & Chan, 2017). Just because instrumental homi- cides are more difficult to solve and are more likely to go unsolved than expressive homicides makes them no less worthy of law enforcement attention (Salfati & Bateman, 2005). The same could be said for proactive and reactive criminal thinking, where the risk and outcome effects might be stronger for reactive criminal thinking but where the degree of support for a criminal lifestyle is equal across these two dimensions of criminal thought process.

Theoretical and practical implications

There are both theoretical and practical implications to these results. One theoretical implication is that despite their high intercorrelation (.75 in this study), the proactive and reactive scales of the PICTS appear to be measuring different con- structs. Results from the Walters and Yurvati (2017) study revealed that the PICTS proactive and reactive scales were assessing latent constructs with features that reflected the pro- active (planned, calculated, and callous) and reactive (impul- sive, irresponsible, and emotional) dimensions of criminal thought process, respectively. According to the results reported here, scores on the PICTS proactive and reactive scales corre- lated differentially with criminal risk and outcome. With one notable exception, the proactive scale correlated negatively with several criminal risk measures, whereas the reactive scale corre- lated positively with these same measures and subsequent arrests. When the mean profiles of risk and outcome measures were compared for PICTS with elevated proactive criminal thinking and elevated reactive criminal thinking, the outcome was a moderately strong inverse double-entry ICC. A practical implication that can be drawn from these results is that the PICTS proactive and reactive scales potentially provide infor- mation useful in evaluating and managing prison inmates. Individuals with elevations on either scale are at risk for future problems, although the problems will differ depending on the relative elevation of each scale. Interventions differ depending on whether reactive (e.g., problem solving and cognitive skills training) or proactive (e.g., moral education and cognitive restructuring) criminal thinking is elevated, so a comprehen- sive evaluation will be of assistance in establishing the appro- priate treatment for whichever pattern is present.

Limitations

In closing, it is important to consider several study limita- tions. First, the sample consisted of male inmates housed in a single medium-security federal correctional institution. As such, the generalizability of these results to female prisoners, nonincarcerated offenders, state and jail inmates, and felons housed in low- or high-security facilities cannot be assumed. The generalizability of the recidivism findings is also an issue because inmates serving longer sentences were less likely to have been released from confinement and included in the recidivism analyses than inmates serving shorter

228 WALTERS

sentences. A second potential limitation of this study is that all eight dependent variables came from official records, a procedure that could have limited the scope and depth of analysis. A deeper analysis could have produced richer infor- mation through inmate self-report and the inclusion of dependent variables that assess offender attitudes (criminal thought content), expectancies, and attributions. Third, the PICTS was administered at a single point in time (i.e., intake). PICTS administered at a later date, after the inmate had become more accustomed to incarceration, or at mul- tiple times to assess changes in antisocial cognition might have painted a more accurate or representative picture of the inmate’s criminal thought process. Fourth, the procedure used to assess similarity between outcome profiles—the dou- ble-entry ICC—is one of the more popular approaches to determining the extent to which the scatter, elevation, and shape of the different outcome profiles corresponded with one another. It has been argued that the double-entry ICC’s superiority to alternative procedures has not been demon- strated, but neither is there evidence that it is inferior to these other procedures (Furr, 2010). Sixth, the effect sizes for the dichotomous and count outcomes were very small, although it should be noted that in each case these were regression coefficients that controlled for both age and race.

Conclusion

In this study, findings from variable- and person-level analy- ses confirmed that the constructs of proactive and reactive criminal thinking, despite extensive overlap, are distinct, separate, and meaningful entities and that scales based on these constructs could have practical utility in assessing offender risk and predicting future outcome.

References

Adjorlolo, S., & Chan, H. C. (2017). The nature of instrumentality and expressiveness of homicide crime scene behaviors: A review. Trauma, Violence, & Abuse, 18, 119–133.

Bandura, A., Barbaranelli, C., Caprara, G. V., & Pastorelli, C. (1996). Mechanisms of moral disengagement in the exercise of moral agency. Journal of Personality and Social Psychology, 71, 364–374.

Bushman, B. J., & Anderson, C. A. (2001). Is it time to pull the plug on the hostile versus instrumental aggression dichotomy? Psychological Review, 108(1), 273–279.

Card, N. A., & Little, T. D. (2006). Proactive and reactive aggression in childhood and adolescence: A meta-analysis of differential relations with psychosocial adjustment. International Journal of Behavioral Development, 30, 466–480.

Carroll, A., McCarthy, M., Houghton, S., O’Connor, E. S., & Zadow, C. (2018). Reactive and proactive aggression as meaningful distinctions at the variable and person level in primary school-aged children. Aggressive Behavior, 44, 431–441.

Center for Human Resource Research. (2009). NLSY79 user’s guide. Columbus, OH: CHRR NLS User Services, The Ohio State University.

Cronbach, L. J., & Meehl, P. E. (1955). Construct validity in psycho- logical tests. Psychological Bulletin, 52, 281–302.

Dodge, K. A., & Coie, J. D. (1987). Social-information processing fac- tors in reactive and proactive aggression in children’s peer groups. Journal of Personality and Social Psychology, 53, 1146–1158.

Folk, J. B., Disabato, D. J., Daylor, J. M., Tangney, J. P., Barboza, S., Wilson, J. S., … Holwager, J. (2016). Effectiveness of a self-adminis- tered intervention for criminal thinking: Taking a chance on change. Psychological Services, 13, 272–282.

Furr, R. M. (2010). The double-entry intraclass correlation as an index of profile similarity: Meaning, limitations, and alternatives. Journal of Personality Assessment, 92(1), 1–15.

Hamlyn, B., Maxwell, C., Hales, J., & Tait, C. (2003). The 2003 Crime and Justice Survey (England and Wales) technical report. London, UK: National Centre for Social Research/BMRB Social Research.

Hare, R. D. (2003). The Hare Psychopathy Checklist–Revised Manual (2nd ed.). Toronto, Canada: Multi-Health Systems.

Huizinga, D., & Jakob-Chien, C. (1998). The contemporaneous co- occurrence of serious and violent juvenile offending and other prob- lem behaviors. In R. Loeber & D. P. Farrington (Eds.), Serious and violent juvenile offenders: Risk factors and successful interventions (pp. 47–67). Thousand Oaks, CA: Sage.

Koolen, S., Poorthuis, A., & van Aken, M. A. G. (2012). Cognitive dis- tortions and self-regulatory personality traits associated with pro- active and reactive aggression in early adolescence. Cognitive Therapy and Research, 36, 776–787.

Martinelli, A., Ackermann, K., Bernhard, A., Freitag, C. M., & Schwenck, C. (2018). Hostile attribution bias and aggression in chil- dren and adolescents: A systematic literature review on the influence of aggression subtype and gender. Aggression and Violent Behavior, 39, 25–32.

McCrae, R. R. (2008). A note on some measures of profile agreement. Journal of Personality Assessment, 90, 105–109.

Miethe, T. D., & Drass, K. A. (1999). Exploring the social context of instrumental and expressive homicides: An application of qualitative comparative analysis. Journal of Quantitative Criminology, 15(1), 1–21.

Muth�en, B., & Muth�en, L. (1998–2017). Mplus user’s guide (8th ed.). Los Angeles, CA: Muth�en and Muth�en.

Polman, H., Orobio de Castro, B., Koops, W., van Boxtel, H. W., & Merk, W. W. (2007). A meta-analysis of the distinction between reactive and proactive aggression in children and adolescents. Journal of Abnormal Child Psychology, 35, 522–535.

Poulin, F., & Boivin, M. (2000). Reactive and proactive aggression: Evidence of a two-factor model. Psychological Assessment, 12, 115–122.

Rieffe, C., Broekhof, E., Kouwenberg, M., Faber, J., Tsutsui, M. M., & G€uro�glu, B. (2016). Disentangling proactive and reactive aggression in children using self-report. European Journal of Developmental Psychology, 13, 439–451.

Salfati, C. G., & Bateman, A. L. (2005). Serial homicide: An investiga- tion of behavioural consistency. Journal of Investigative Psychology and Offender Profiling, 2, 121–144.

Smeets, K. C., Oostermeijer, S., Lappenschaar, M., Cohn, M., van der Meer, J. M. J., Popma, A., … Buitelaar, J. K. (2017). Are proactive and reactive aggression meaningful distnctions in adolescents? A variable- and person-based approach. Journal of Abnormal Child Psychology, 45(1), 1–14.

Smith, G. T. (2005). On construct validity: Issues of method and meas- urement. Psychological Assessment, 17, 396–408.

Swogger, M. T., Walsh, Z., Maisto, S. A., & Conner, K. R. (2014). Reactive and proactive aggression and suicide attempts among crim- inal offenders. Criminal Justice and Behavior, 41, 337–344.

Urben, S., Habersaat, S., Pihet, S., Suter, M., Ridder, J., & St�ephan, P. (2018). Specific contributions of age of onset, callous-unemotional traits and impulsivity to reactive and proactive aggression in youths with conduct disorders. Psychiatric Quarterly, 89(1), 1–10.

Walters, G. D. (1995). The Psychological Inventory of Criminal Thinking Styles: Part I. Reliability and preliminary validity. Criminal Justice and Behavior, 22, 307–325.

Walters, G. D. (2005). Proactive and reactive aggression: A lifestyle view. In J. P. Morgan (Ed.), Psychology of aggression (pp. 29–43). Hauppauge, NY: Nova Science Publishers, 2005.

Walters, G. D. (2007). Measuring proactive and reactive criminal think- ing with the PICTS: Correlations with outcome expectancies and hos- tile attribution biases. Journal of Interpersonal Violence, 22, 371–385.

DIVERGENT PATTERNS OF CORRELATION 229

Walters, G. D. (2012). Crime in a psychological context: From career criminals to criminal careers. Thousand Oaks, CA: Sage.

Walters, G. D. (2013). The Psychological Inventory of Criminal Thinking Styles (PICTS) professional manual. Allentown, PA: Center for Lifestyle Studies.

Walters, G. D. (2016). Mediating the distal crime-drug relationship with proximal reactive criminal thinking. Psychology of Addictive Behaviors, 30(1), 128–137.

Walters, G. D., & Di Fazio, R. (2016). Psychopathy and the criminal lifestyle: Similarities and differences. In C. B. Gacono (Ed.), The clin- ical and forensic assessment of psychopathy: A practitioner’s guide (2nd ed., pp. 433–444). New York, NY: Routledge.

Walters, G. D., & Elliott, W. N. (1999). Predicting release and discip- linary outcome with the Psychological Inventory of Criminal Thinking Styles: Female data. Legal and Criminological Psychology, 4(1), 15–21.

Walters, G. D., & Geyer, M. D. (2005). Construct validity of the Psychological Inventory of Criminal Thinking Styles in relationship to the PAI, disciplinary adjustment, and program completion. Journal of Personality Assessment, 84, 252–260.

Walters, G. D., Hagman, B. T., & Cohn, A. M. (2011). Toward a hier- archical model of criminal thinking: Evidence from item response

theory and confirmatory factor analysis. Psychological Assessment, 23, 925–936.

Walters, G. D., & Lowenkamp, C. T. (2016). Predicting recidivism with the Psychological Inventory of Criminal Thinking Styles (PICTS) in community-supervised male and female offenders. Psychological Assessment, 28, 652–659.

Walters, G. D., & Mandell, W. (2007). Incremental validity of the Psychological Inventory of Criminal Thinking Styles and Psychopathy Checklist: Screening Version in predicting disciplinary outcome. Law and Human Behavior, 31, 141–157.

Walters, G. D., White, T. W., & Denney, D. (1991). The Lifestyle Criminality Screening Form: Preliminary data. Criminal Justice and Behavior, 18, 406–418.

Walters, G. D., & Yurvati, E. (2017). Testing the construct validity of the PICTS proactive and reactive scores against six putative meas- ures of proactive and reactive criminal thinking. Psychology, Crime and Law, 23(1), 1–14.

Weinberger, D. A., & Schwartz, G. E. (1990). Distress and restraint as superordinate dimensions of self-reported adjustment: A typological perspective. Journal of Personality, 58, 381–417.

Westen, D., & Rosenthal, R. (2005). Improving construct validity: Cronbach, Meehl, and Neurath’s ship. Psychological Assessment, 17, 409–412.

230 WALTERS

Copyright of Journal of Personality Assessment is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.