Crime Analysis Report

BRIEF REPORT

An Item Response Theory Analysis of the Psychological Inventory of Criminal Thinking Styles: Comparing Male and Female

Probationers and Prisoners

Glenn D. Walters Kutztown University

An item response theory (IRT) analysis of the Psychological Inventory of Criminal Thinking Styles (PICTS) was performed on 26,831 (19,067 male and 7,764 female) federal probationers and compared with results obtained on 3,266 (3,039 male and 227 female) prisoners from previous research. Despite the fact male and female federal probationers scored significantly lower on the PICTS thinking style scales than male and female prisoners, discrimination and location parameter estimates for the individual PICTS items were comparable across sex and setting. Consistent with the results of a previous IRT analysis conducted on the PICTS, the current results did not support sentimentality as a component of general criminal thinking. Findings from this study indicate that the discriminative power of the individual PICTS items is relatively stable across sex (male, female) and correctional setting (probation, prison) and that the PICTS may be measuring the same criminal thinking construct in male and female probationers and prisoners.

Keywords: item response theory, Psychological Inventory of Criminal Thinking Styles, federal probationers

Supplemental materials: http://dx.doi.org/10.1037/pas0000014.supp

As a consequence of its status as one of the Big Four risk factors for general and violent recidivism (Andrews & Bonta, 2010), criminal thinking has attracted the attention of researchers and clinicians alike (Walters, 2009). Among the factors responsible for the growing popularity of criminal thinking in corrections as well as in the criminal justice field overall is its relevance to classifi- cation, prediction, and intervention. There is evidence, for in- stance, that criminal thinking, in conjunction with certain status (demographic, incarceration, and mental health) variables, can be useful in classifying offenders for both treatment and management purposes (Mandracchia & Morgan, 2012). Moreover, a recent meta-analysis found that general criminal thinking (GCT) success- fully predicted recidivism above and beyond the contributions of age and criminal history (Walters, 2012). Criminal thinking has also been shown to be helpful in assessing treatment readiness (Taxman, Rhodes, & Dumenci, 2011) and change as a function of treatment involvement (Walters, 2002). All of these factors sug- gest that criminal thinking is an important criminal justice con-

struct that should be assessed whether an offender is housed in prison or being supervised in the community.

To understand the latent structure of constructs measured with a psychological test, researchers often turn to item response theory (IRT). Unlike classical test theory, in which the focus is on the general characteristics of a test, IRT is concerned with how the individual items on a test assess a latent trait or construct (�). Adopting an IRT framework consequently requires a shift in emphasis from the overall test score to the individual items and from group- or aggregate-level data to individual-level data (Lord & Novick, 1968). Given that there is no single IRT model, the manner in which the properties of the person and the item are examined will vary according to the model the researcher adopts. Samejima’s (1969) graded response model constructs parametric logistic functions to assess these properties (Hambleton, 1989). The parameter logistic functions included in Samejima’s two- parameter graded response model are labeled discrimination and threshold. Whereas the discrimination parameter appraises the strength of relationship between a test item and the latent trait the test is designed to measure, the threshold parameter locates the position of the item along the latent trait continuum (Baker & Kim, 2004). These two parameters provide valuable information on a person’s responses to the test items and how well these items assess the latent trait allegedly measured by the test.

Walters, Hagman, and Cohn (2011) used IRT and confirmatory factor analysis (CFA) with the Psychological Inventory of Crim-

This article was published Online First June 30, 2014. Glenn D. Walters is the author of the Psychological Inventory of

Criminal Thinking Styles (PICTS) and receives remuneration from the sale of the PICTS manual.

Correspondence concerning this article should be addressed to Glenn D. Walters, Department of Criminal Justice, Kutztown University, Kutztown, PA 19530-0730. E-mail: walters@kutztown.edu

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

Psychological Assessment © 2014 American Psychological Association 2014, Vol. 26, No. 3, 1050 –1055 1040-3590/14/$12.00 http://dx.doi.org/10.1037/pas0000014

1050

inal Thinking Styles (PICTS: Walters, 1995), a test designed to measure criminal thinking. The CFA results from this study con- firmed the hypothesis that the criminal thought process, as mea- sured by the PICTS, was hierarchically organized and composed of two first-order factors (proactive and reactive criminal thinking) and one second-order or superordinate factor (GCT). The IRT analysis revealed that the eight items from the PICTS Sentimen- tality (Sn) scale did not load onto the superordinate GCT factor, whereas nearly all of the items from the seven remaining PICTS thinking style scales achieved moderate to good discrimination between individuals high and low on the criminal thinking latent trait continuum. Further analysis revealed that these items achieved greater precision at higher rather than at lower levels of the continuum.

Although the Walters et al. (2011) study has potentially impor- tant implications for theory, research, and practice, the generaliz- ability of these findings to other populations and settings needs to be verified. As a case in point, all participants were from a single medium-security federal prison and were of moderate to moder- ately high risk. The purpose of the current study was to test the generalizability of the previous study’s results to male and female probationers and female prisoners. Three hypotheses were tested in this study. The first hypothesis predicted that both sex (male, female) and setting (prison, probation) would produce a differen- tial item functioning (DIF) effect resulting in the need to analyze male and female and prison and probation data separately. The second hypothesis predicted that items from the PICTS Sn scale would not load onto a GCT factor. The third hypothesis proposed that the discrimination and threshold parameter estimates obtained from probationers would correlate significantly with the discrim- ination and threshold estimates obtained from incarcerated offend- ers, even though the threshold function would be shifted right due to a lower endorsement of criminal thinking items in the probation samples.

Method

Participants

Starting in November 2009, the Federal Probation and Pretrial Services began routinely administering the PICTS to all federal offenders under community supervision (i.e., probation and super- vised release). Participants for the current study were 26,831 federal probationers from across the United States who had been administered the PICTS between November 2009 and June 2013: 19,067 men with a mean age of 39.76 years and 7,764 women with a mean age of 37.05 years. In the vast majority of cases, the probation officer administered the PICTS in his or her office to individual probationers. For those participants who completed the PICTS more than once, only the first administration was included in this study.

Participants for the male prison group came from a single medium-security federal facility and consisted of 3,039 men (2,872 prisoners with no more than one missing PICTS item who had been included in the original Walters et al., 2011, investigation and 167 prisoners with more than one missing PICTS item who had not been included in the original Walters et al., 2011, inves- tigation). These individuals had completed the PICTS as part of a routine intake procedure (over 95% completion rate) sometime

between November 2003 and December 2010. Male prisoners had a mean age of 35.00 years. Participants in the female prison group were 227 female prisoners with a mean age of 33.63 years—127 from a medium-security state facility and 100 from a multicustody federal facility—who had completed the PICTS as part of a re- search study (Walters, Elliott, & Miscoll, 1998).

Measure

The PICTS (Walters, 1995) is an 80-item self-report measure designed to assess the criminal thought process. Each item is rated on a 4-point Likert-type scale (1 � disagree, 2 � uncertain, 3 � agree, 4 � strongly agree) and items are combined to form 10 different scales: two 8-item validity scales (Confusion–revised [Cf-r] and Defensiveness-revised [Df-r]) and eight 8-item thinking style scales (Mollification [Mo], Cutoff [Co], Entitlement [En], Power Orientation [Po], Sn, Superoptimism [So], Cognitive Indo- lence [Ci], and Discontinuity [Ds]). A GCT score can be obtained by summing the eight individual thinking style scales, but because Walters et al. (2011) determined that only seven of these scales loaded onto the general factor (Sn being the exception), there are two possible GCT models: GCT64 (comprising all 64 items from the eight thinking style scales) and GCT56 (comprising all items from the 7 thinking style scales other than Sn). Mo, En, Po, and So are combined to form the first-order proactive factor and Co, Ci, and Ds are combined to form the first-order reactive factor. Reli- ability and validity data for and descriptions of the eight PICTS thinking style scales can be found in Table 1.

Procedure

The current study followed a four-step procedure. The first step was to determine whether the construct assessed by the PICTS is sufficiently unidimensional to justify IRT analysis. Unidimension- ality was evaluated with two bifactor CFA in which all of the PICTS items (64 or 56) were loaded onto a general factor and several subfactors (32 on the proactive factor and 24 on the reactive factor [GCT64 and GCT56] and eight on sentimentality [GCT64 only]) simultaneously. Model fit was assessed using gen- eral rules of thumb suggested by Hu and Bentler (1999): compar- ative fit index (CFI) and Tucker–Lewis Index (TLI) values greater than or equal to .95 and root-mean-square error of approximation (RMSEA) value less than or equal to .06.

The second step of the procedure was to test for DIF. This was accomplished by including sex (male � 0, female � 1) and setting (probation � 1, prison � 2) as covariates in an IRT analysis of the full 30,097-participant sample using the multiple-indicator multiple-cause (MIMIC) structural equation modeling procedure from MPlus 5.0 (Muthén & Muthén, 1998 –2007). MIMIC uses CFA and structural equation modeling to determine whether one or more covariates (sex and setting in the current study) have differ- ential IRT loadings, as a whole, on a latent construct (in this case, criminal thinking as measured by the GCT). The accuracy of MIMIC for DIF is well established (Woods, 2009).

In the third step of the procedure, male and female probationers and prisoners were compared on age, race (White, non-White), and the PICTS Df-r and GCT64 scores. The IRT analyses comprised the fourth step of the procedure. Two-parameter logistic IRT analysis using Samejima’s (1969) graded response model was

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

1051IRT OF PICTS IN PROBATIONERS

computed with MPlus 5.0 (Muthén & Muthén, 1998 –2007). Two models were fitted in the current study: a model composed of all 64 thinking style items (GCT64) and a model in which the eight Sn items were excluded (GCT56). Items were calibrated using a max- imum likelihood with robust standard errors estimation procedure and expectation maximization convergence was set at .001. These convergence criteria were satisfied for the graded response model in all four analyses. Eight randomly selected items (one from each scale, including Sn) were removed from the GCT64 to allow model fit comparisons between GCT64 and GCT56 given that even in comparisons involving nonnested models, the number of items in each model should be the same (Burnham & Anderson, 2002).

The proportion of missing values for individual PICTS items in the two probation samples ranged from 0.01% to 0.48%, and over 96% of the probationers did not have a single missing value on any of the 64 PICTS items. The proportion of missing values for individual PICTS items in the male prison sample ranged from 0.6% to 2.4%, with 81% of the sample not having a single missing value on any of the 64 PICTS items. The proportion of missing values for individual PICTS items in the female prison sample ranged from 0.0% to 1.8%, and 85% of the sample did not have a single missing value on any of the 64 PICTS items. Missing data were handled with full information maximum likelihood.

Results

Bifactor CFA using weighted least squares means and variance adjusted (WLSMV) estimation, was performed on both versions of the PICTS. The GCT64, �

2(1087) � 69,899, p � .001, achieved poor fit on one indicator (CFI � .63) and good fit on the other two indicators (TLI � .96, RMSEA � .046). The GCT56, �

2(884) � 51,553, p � .001, also displayed poor fit on the CFI (.70) and good fit on the TLI (.97) and RMSEA (.044). Whereas the results were mixed, the majority of indicators supported unidimensionality.

DIF was testing by including sex (0 � male, 1 � female) and setting (1 � probation, 2 � prison) as covariates in an IRT analysis of the GCT64 using all 30,097 participants. The results of the MIMIC analysis showed clear evidence of DIF for both sex, t � 20.63, p � .001, and setting, t � 47.00, p � .001. The IRT analyses were accordingly performed on four different groups of participants: male probationers (n � 19,067), male prisoners (n �

3,039), female probationers (n � 7,764), and female prisoners (n � 227).

Prior to conducting the IRT analyses, male probationers and prisoners and female probationers and prisoners were compared on age, race (White, non-White), the PICTS Df-r scale, and the PICTS GCT64 score. Male probationers were older (M � 39.76 years, SD � 13.43, vs. M � 35.00 years, SD � 9.87), more often White (49.5% vs. 17.2%), more defensive in completing the PICTS (Df-r: M � 22.97, SD � 2.90, vs. M � 21.12, SD � 4.31), and less inclined toward criminal thinking (GCT64: M � 75.75, SD � 16.00, vs. M � 112.64, SD � 28.10) than were male prisoners (p � .001). Likewise, female probationers were older (M � 37.05 years, SD � 11.90, vs. M � 33.63 years, SD � 8.44) and more often White (41.9% vs. 33.0%), with higher scores on the Df-r (M � 21.17, SD � 2.59, vs. M � 17.48, SD � 4.62) and lower scores on the GCT64 (M � 79.92, SD � 17.75 vs. M � 112.37, SD � 26.94) compared to female prisoners (p � .001).

Table 2 lists the average discrimination parameter estimates for the PICTS GCT56 and 8 Sn scale items. The results indicate that the Sn items did a significantly poorer job of discriminating between individuals at different levels of the criminal thinking latent trait dimension than did the other seven PICTS thinking styles. The GCT56 also achieved significantly better fit than the GCT64 when eight items were removed from GCT64 to make it comparable in size to the GCT56 and differences on the Akaike information criterion and Bayesian information criterion were cal- culated (see Table 2). Bivariate fit, in the form of standardized residuals for pairs of items, was modest. Composite reliability estimates (Raykov, 1997) for the GCT64 and GCT56, on the other hand, were good in all four samples.

When the parameter estimates for male and female proba- tioners were compared with the parameter estimates for male and female prisoners, the threshold (b1–3) parameters were shifted noticeably to the right, suggesting that probationers were less likely than prisoners to endorse the criminal thinking items on the PICTS (see Table 3). Even so, the threshold values correlated .87 to .97 across sex and .83 to .91 across setting. Discrimination parameter estimates also correlated strongly across sex and setting. Information function analysis revealed that the PICTS items achieved greater precision at higher levels

Table 1 The PICTS Thinking Style Scales: Reliability, Validity, and Description

Scale n REL VAL Description

Mo 8 .81 .12 Externalizing blame for past criminal actions through justification, rationalization, and excuse making. En 8 .80 .19 Giving oneself permission to commit a crime out of necessity, uniqueness, or privilege. Po 8 .80 .16 Desire to gain a sense of power and control over others. So 8 .83 .14 Belief in the ability to avoid the negative consequences of a criminal lifestyle indefinitely. Co 8 .82 .21 Rapid elimination of deterrents to crime through a sense of frustration or simple phrase (e.g., “the hell with it”). Ci 8 .79 .16 Impulsive decision making characterized by failure to critically evaluate personal plans and ideas. Ds 8 .85 .20 Easy distractibility leading to failure to maintain commitments and follow through on initial intentions. Sn 8 .73 .18 Denying harm done to others by performing good deeds and engaging in artistic pursuits.

Note. PICTS � Psychological Inventory of Criminal Thinking Styles; Scale � PICTS thinking style scale; Mo � Mollification; En � Entitlement; Po � Power Orientation; So � Superoptimism; Co � Cutoff; Ci � Cognitive Indolence; Ds � Discontinuity; Sn � Sentimentality; REL � test–retest reliability after 2 weeks in 50 medium-security prison inmates; VAL � mean (validity) effect size (Walters, 2002) in the form of point-biserial correlations between each PICTS scale and the presence of an institutional infraction or recidivism (k � 12); Description � brief description of the thinking style the scale is designed to measure.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

1052 WALTERS

of the criminal thinking trait dimension than at lower levels of the criminal thinking trait dimension, with the information function for GCT56 peaking at 1.75 in the male probation sample, 1.70 in female probation sample, 1.80 in the male prison sample, and 1.75 in the female prison sample.

Discussion

The purpose of this study was to cross-validate previous IRT findings obtained in a study on male medium-security federal prisoners (Walters et al., 2011). As predicted, the earlier results generalized to a national sample of male and female federal probationers. Under current federal guidelines, probation is re- served for individuals with little to no criminal history who have committed a nonviolent, non-drug-related offense (United States Sentencing Commission, 2012, §3E1.1). Accordingly, the vast majority of individuals on federal probation can be classified as low risk. Despite being low risk and endorsing fewer criminal thinking items than prisoners, male and female probationers achieved results that confirmed the earlier Walters et al. (2011) findings. First, there were DIF effects for both sex and setting, necessitating separate IRT analyses for male and female proba- tioners and prisoners. Second, items from the Sn scale failed to consistently load onto the GCT factor, supporting previous results from the Walters et al. (2011) study and justifying removal of these items from the GCT score. Third, discrimination and threshold parameter estimates recorded by low-risk male and female proba- tioners were comparable to those recorded by higher risk male and female prisoners despite threshold estimates that were prominently shifted to the right because of probationers’ weaker endorsement of criminal thinking items.

The results of the current IRT analyses indicated that one of the criminal thinking styles proposed by Walters (1995) as a core feature of criminal thinking, sentimentality, consistently failed to load onto the GCT factor. This would seem to suggest that al- though sentimentality may be part of the criminal lifestyle (Wal- ters, 1990), it is probably not a cardinal feature of criminal think- ing. A practical implication of these results, then, is that GCT should be scored by summing the Mo, Co, En, Po, So, Ci, and Ds scales, and leaving the Sn scale out of the equation. No more than two out of eight items from any one of the other seven thinking style scales (except for the Mo, En, and So scales in the female prison sample) failed to achieve adequate discrimination in the current study, compared with five to seven of the Sn items. Another practical issue is whether certain items can and should be removed from the PICTS. Except for Items 1 (En) and 5 (So), however, most of the items from the 56-item version of the PICTS

Table 2 Comparisons Between the PICTS GCT64 and GCT56 Across the Four Samples

M-probation F-probation M-prison F-prison

Discrimination parameter GCT56

M 1.49 1.39 1.48 1.12 SD 0.42 0.48 0.46 0.42 % � 1.00 85.7% 83.9% 82.1% 62.5%

8 Sn items M 0.71 0.60 0.93 0.50 SD 0.46 0.46 0.55 0.45 % � 1.00 25.0% 12.5% 37.5% 12.5%

AIC fit comparison GCT64a 1,252,802 559,279 294,582 26,120 GCT56 1,203,617 543,582 292,925 26,013 �AIC 49,185 15,697 1,657 107

BIC fit comparison GCT64a 1,254,554 560,830 295,924 26,884 GCT56 1,205,376 545,141 294,274 26,780 �BIC 49,178 15,689 1,650 104

Bivariate model fit GCT64 29.6% 13.1% 24.0% 0.6% GCT56 29.7% 14.2% 23.0% 0.4%

Composite reliability GCT64 .971 .966 .972 .950 GCT56 .971 .967 .971 .952

Note. PICTS � Psychological Inventory of Criminal Thinking Styles; M-probation � male probationers (n � 19,067); F-probation � female probationers (n � 7,764); M-prison � male prisoners (n � 3,039); F-prison � female prisoners (n � 227); GCT64 � 64-item general criminal thinking score; GCT56 � 56-item general criminal thinking score (minus the eight Sn items); Sn � Sentimentality scale; AIC � Akaike information criterion; BIC � Bayesian information criterion; Bivariate model fit � percentage of standardized residuals with p � .01; % � 1.00 � proportion of items with a discrimination parameter of 1.00 or higher. a Includes 56 items (64 original items � 8 randomly removed items, one from each thinking style scale).

Table 3 Comparison of IRT Parameters for the 64 PICTS Items in Male and Female Probationers and Prisoners

Men Women Sex Setting

Probationers Prisoners Probationers Prisoners Probation Prison Men Women

Par M SD M SD M SD M SD t(63) r t(63) r t(63) r t(63) r

a 1.39 0.49 1.41 0.50 1.28 0.51 1.04 0.47 8.04� .98� 8.79� .77� �0.70 .92� 5.50� .76�

b1 1.94 1.45 0.77 1.10 1.68 1.51 �0.05 1.11 6.08 � .97� 11.90� .88� 14.59� .91� 16.29� .83�

b2 3.06 1.56 1.68 1.28 2.61 1.52 0.65 1.14 7.43 � .95� 13.91� .89� 16.74� .91� 18.40� .83�

b3 4.68 1.58 3.37 1.31 4.12 1.52 2.30 1.03 12.46 � .97� 12.82� .87� 16.07� .91� 16.77� .84�

Note. Sample sizes were as follows: male probationers � 19,067, male prisoners � 3,039, female probationers � 7,764, female prisoners � 227. PICTS � Psychological Inventory of Criminal Thinking Styles; IRT � item response theory; Sex � comparisons between sex (men vs. women) for probation and prison separately; Setting � comparisons between settings (probation vs. prison) for men and women separately; Par � parameter; a � discrimination parameter; b � threshold parameters; t(63) � paired-samples t test with 63 degrees of freedom. � p � .001.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

1053IRT OF PICTS IN PROBATIONERS

achieved reasonable discrimination and respectable coverage of the criminal thinking construct. Additional research is required to determine how to maximize the utility of the PICTS.

A long-standing limitation of the PICTS is the lack of data on female offenders. Prior to the current study, published research on the PICTS included two relatively small female samples: a group of 127 female state prisoners and a group of 100 female federal inmates. In the current study, these two samples were combined to create the female prisoner sample against which the female probationers were compared. A MIMIC analysis, in which sex served as a covariate, revealed the presence of a significant sex DIF effect. This effect, although statistically significant and consistent with a gendered interpretation of crime, was of questionable clinical utility in that the discrimi- nation and threshold parameter estimates for male and female probations correlated .95 to .98. Correlations between the pa- rameter estimates for male and female prisoners were also significant but lower in magnitude (.77–.89), perhaps because of the smaller number of participants in the two prison samples relative to the two probation samples. Overall, these results suggest that the PICTS is measuring criminal thinking in both men and women and the manner in which it measures this construct is similar across sex and setting, as indicated by roughly comparable levels of discrimination and position (threshold) on the criminal thinking trait continuum in male and female probationers and prisoners.

Despite the presence of a large sample of male and female probationers and the use of a sophisticated statistical model derived from IRT, the current study has several limitations. One such limitation is the absence of risk information on partici- pants in the current study. Criminal history and scores on the Lifestyle Criminality Screening Form (Walters, White, & Den- ney, 1991) were available for the two prison samples but not for the federal probationers. It was simply assumed that the federal probationers were of uniformly low risk given that only defen- dants with minimal criminal histories and nonviolent offenses were eligible for federal probation at the time this study was being conducted. A more thorough evaluation of the relation- ship between risk level and PICTS performance, however, requires risk estimates on each participant. A second potential limitation of this study is that because probationers earned significantly higher Df-r validity scale scores than prisoners, probationer scores on the thinking style portion of the test may have been artificially suppressed by a lack of openness in acknowledging the presence of criminal thinking. Even if true, the IRT results showed that the relative position of items on the criminal thinking dimension varied minimally between those who completed the PICTS in prison and those who completed it while on probation. Hence, whereas lower risk and higher defensiveness probably account for the lower endorsement of PICTS thinking style items by probationers relative to individ- uals who completed the PICTS in prison, the nature of criminal thinking, as measured by the PICTS, was similar across the two populations and settings.

Results from the current study and the previous Walters et al. (2011) investigation indicate that the PICTS achieves greater precision at higher rather than lower levels of the criminal thinking trait dimension. What this means is that the PICTS is more sensitive to stronger as opposed to weaker expressions of

criminal thinking and, as such, does a better job of discrimi- nating between moderate, high, and very high levels of criminal thinking than it does of discriminating between moderate, low, and very low levels of criminal thinking. One way to improve on the PICTS’s ability to assess the full range of criminal thinking would be to replace poorly discriminating items with items specifically designed to differentiate between lower lev- els of the trait dimension. In all likelihood, however, this would entail increasing the size of the instrument because only a few PICTS items (i.e., Items 1 and 5) have no real value in identi- fying criminal thinking. Another possibility would be to expand the rating scale used to evaluate individual PICTS items from its current four-option format (disagree, uncertain, agree, strongly agree) to a five-option format (strongly disagree, disagree, uncertain, agree, strongly agree). Providing respon- dents with the opportunity to differentiate between disagree and strongly disagree responses may also provide for greater dis- crimination at the lower end of the trait dimension.

References

Andrews, D. A., & Bonta, J. (2010). The psychology of criminal conduct (5th ed.). New Providence, NJ: Bender.

Baker, F. B., & Kim, S.-H. (2004). Item response theory: Parameter estimation techniques (2nd ed.). New York, NY: Marcel Dekker.

Burnham, K. P., & Anderson, D. R. (2002). Model selection and multi- modal inference: A practical information-theoretic approach. New York, NY: Springer.

Hambleton, R. K. (1989). Principles and selected applications of item response theory. In R. L. Linn (Ed.), Educational measurement (3rd ed., pp. 147–200). London, England: Collier Macmillan.

Hu, L.-T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alterna- tives. Structural Equation Modeling: A Multidisciplinary Journal, 6, 1–55. doi:10.1080/10705519909540118

Lord, F. M., & Novick, M. R. (1968). Statistical theories of mental test scores. Reading, MA: Addison-Wesley.

Mandracchia, J. T., & Morgan, R. D. (2012). Predicting offenders’ crim- inogenic cognitions with status variables. Criminal Justice and Behav- ior, 39, 5–25. doi:10.1177/0093854811425453

Muthén, B., & Muthén, L. (1998 –2007). Mplus user’s guide (5th ed.). Los Angeles, CA: Muthén & Muthén.

Raykov, T. (1997). Estimation of composite reliability for congeneric measures. Applied Psychological Measurement, 21, 173–184. doi: 10.1177/01466216970212006

Samejima, F. (1969). Estimation of latent ability using a response pattern of graded scores. Psychometika Monographs, 34(4, Pt. 2, Whole No. 17).

Taxman, F. S., Rhodes, A. G., & Dumenci, L. (2011). Construct and predictive validity of criminal thinking scales. Criminal Justice and Behavior, 38, 174 –187. doi:10.1177/0093854810389550

United States Sentencing Commission. (2012, November). Guidelines manual. Washington, DC: Author.

Walters, G. D. (1990). The criminal lifestyle: Patterns of serious criminal conduct. Newbury Park, CA: Sage. doi:10.4135/9781483325569

Walters, G. D. (1995). The Psychological Inventory of Criminal Thinking Styles: Part I. Reliability and preliminary validity. Criminal Justice and Behavior, 22, 307–325. doi:10.1177/0093854895022003008

Walters, G. D. (2002). The Psychological Inventory of Criminal Thinking Styles (PICTS): A review and meta-analysis. Assessment, 9, 278 –291. doi:10.1177/1073191102009003007

Walters, G. D. (2009). Criminal thinking. In M. McMurran & R. C. Howard (Eds.), Personality, personality disorder, and violence (pp. 281–295). Chichester, England: Wiley.

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

1054 WALTERS

Walters, G. D. (2012). Criminal thinking and recidivism: Meta-analytic evidence on the predictive and incremental validity of the Psychological Inventory of Criminal Thinking Styles (PICTS). Aggression and Violent Behavior, 17, 272–278. doi:10.1016/j.avb.2012.02.010

Walters, G. D., Elliott, W. N., & Miscoll, D. (1998). Use of the Psycho- logical Inventory of Criminal Thinking Styles in a group of female offenders. Criminal Justice and Behavior, 25, 125–134. doi:10.1177/ 0093854898025001008

Walters, G. D., Hagman, B. T., & Cohn, A. M. (2011). Toward a hierar- chical model of criminal thinking: Evidence from item response theory and confirmatory factor analysis. Psychological Assessment, 23, 925– 936. doi:10.1037/a0024017

Walters, G. D., White, T. W., & Denney, D. (1991). The Lifestyle Crim- inality Screening Form: Preliminary data. Criminal Justice and Behav- ior, 18, 406 – 418. doi:10.1177/009385489101800400

Woods, C. M. (2009). Evaluation of MIMIC-model methods for DIF testing with comparison to two-group analysis. Multivariate Behavioral Research, 44, 1–27. doi:10.1080/00273170802620121

Received January 5, 2014 Revision received May 23, 2014

Accepted May 28, 2014 �

T hi

s do

cu m

en t

is co

py ri

gh te

d by

th e

A m

er ic

an P

sy ch

ol og

ic al

A ss

oc ia

ti on

or on

e of

it s

al li

ed pu

bl is

he rs

. T

hi s

ar ti

cl e

is in

te nd

ed so

le ly

fo r

th e

pe rs

on al

us e

of th

e in

di vi

du al

us er

an d

is no

t to

be di

ss em

in at

ed br

oa dl

y.

1055IRT OF PICTS IN PROBATIONERS