Vous êtes sur la page 1sur 26

Criminal Justice and Behavior

http://cjb.sagepub.com

Self-Report Measures of Psychopathy, Antisocial Personality, and Criminal


Lifestyle: Testing and Validating a Two-Dimensional Model
Glenn D. Walters
Criminal Justice and Behavior 2008; 35; 1459 originally published online Sep 10, 2008;
DOI: 10.1177/0093854808320922
The online version of this article can be found at:
http://cjb.sagepub.com/cgi/content/abstract/35/12/1459

Published by:
http://www.sagepublications.com

On behalf of:
International Association for Correctional and Forensic Psychology

Additional services and information for Criminal Justice and Behavior can be found at:
Email Alerts: http://cjb.sagepub.com/cgi/alerts
Subscriptions: http://cjb.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations http://cjb.sagepub.com/cgi/content/refs/35/12/1459

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

SELF-REPORT MEASURES OF PSYCHOPATHY,


ANTISOCIAL PERSONALITY, AND
CRIMINAL LIFESTYLE
Testing and Validating a Two-Dimensional Model
GLENN D. WALTERS
Federal Correctional Institution, Schuylkill, Pennsylvania

This article reports results from five studies. Exploratory factor analysis was used to select indicators from the Psychological
Inventory of Criminal Thinking Styles, Levenson Self-Report Psychopathy scales, and Personality Assessment
InventoryAntisocial Features Scale. The 10 indicators were subjected to confirmatory factor analysis, the results of which
show that the two-dimensional model (proactive, reactive) achieves significantly better fit than a general one-factor model
and a two-factor social learning model (criminal thinking, antisocial behavior) with 521 medium-security and 116 maximumsecurity inmates. The construct validity of the two-dimensional model is confirmed in a path analysis pairing (a) proactive
scales with positive outcome expectancies for crime and (b) reactive scales with hostile attribution biases. Implications for a
unified theory of aggression and criminality are discussed.
Keywords: Personality Assessment Inventory; Levenson Self-Report Psychopathy; Psychological Inventory of Criminal
Thinking Styles; proactive; reactive

hereas research and practice in forensic psychology have grown at an unprecedented


rate, theory has failed to keep pace with new developments in the field. Forensic
researchers and practitioners who are looking for theoretical inspiration and guidance must
consequently find both in theories from related disciplines or in general psychological principles that overlook the intricacies of forensic psychology research and practice. Theory is
barely mentioned in three recently published textbooks on forensic psychology (Bartol &
Bartol, 2004; Goldstein, 2007; Weiner & Hess, 2006), and the two families of theory that
receive the most attention in these bookspersonality models (Cleckley, 1976; Hare, 1996)
and behavioral models (Andrews & Bonta, 1998; Robins, 1966)are seriously flawed as
general explanations of crime and forensic psychology (see Walters, 2004). From physics
to psychology, it is well known that to remain viable, a field must be grounded in substantive theory. Psychopathy, antisocial personality, and criminal lifestyle are three constructs that
have been offered as possible psychological explanations for criminal behavior. The similarities between these three constructs are striking and so suggest that they share structural
AUTHORS NOTE: I would like to thank Matthew Geyer, Charles Schlauch, and Patti Walters for their assistance in collecting and entering data for this project. The assertions and opinions contained herein are my private views and should not be construed as being official or as reflecting the views of the Federal Bureau of
Prisons or the U.S. Department of Justice. Address all correspondence to Glenn D. Walters, Psychology
Services, FCI-Schuylkill, PO Box 700, Minersville, PA 17954-0700; e-mail: gwalters@bop.gov.
CRIMINAL JUSTICE AND BEHAVIOR, Vol. 35 No. 12, December 2008 1459-1483
DOI: 10.1177/0093854808320922
2008 International Association for Correctional and Forensic Psychology

1459
Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1460

CRIMINAL JUSTICE AND BEHAVIOR

and content features. If research could demonstrate that these three crime-related constructs
lie along the same dimension or dimensions, then perhaps we would have the beginnings of
a substantive theory of forensic psychology to guide research and practice in the field.

A TWO-DIMENSIONAL MODEL OF CRIMINALITY

A critical first step in developing a theoretical model is to determine the underlying, or


latent, structure of the construct on which the model is based, and one way to do so is with
taxometric analysis. The taxometric method (Meehl, 1995; Ruscio, Haslam, & Ruscio,
2006) allows researchers to gauge whether the latent structure of a construct is categorical
(taxonic) or continuous (dimensional). An early taxometric study on the Psychopathy
ChecklistRevised (PCL-R; Hare, 2003; for a reference to the abbreviations used in this
article, see appendix) showed signs of taxonic structure (Harris, Rice, & Quinsey, 1994),
but more recently conducted and more methodologically sound studies have produced results
more congruent with a dimensional interpretation of the latent structure of psychopathy as
measured by the PCL-R/PCL:SV (Edens, Marcus, Lilienfeld, & Poythress, 2006; Guay,
Ruscio, Knight, & Hare, 2007; Walters, Duncan, & Mitchell-Perez, 2007; Walters, Gray,
et al., 2007), the Psychopathic Personality Inventory (Lilienfeld & Andrews, 1996; Marcus,
John, & Edens, 2004), and the Levenson Self-Report Psychopathy scales (LSRP;
Levenson, Kiehl, & Fitzpatrick, 1995; Walters, Brinkley, Magaletta, & Diamond, in press).
Dimensional results have also been obtained when the taxometric method has been applied
to measures of antisocial personality (Marcus, Lilienfeld, Edens, & Poythress, 2006; Walters,
Diamond, Magaletta, Geyer, & Duncan, 2007) and criminal lifestyle (Walters, 2007a;
Walters & McCoy, 2007).
Once research has established that crime-related constructs such as psychopathy, antisocial personality, and criminal lifestyle have a dimensional, rather than taxonic, latent structure, the next step is to determine the content of these underlying dimensions. Research in
developmental psychology may be of benefit in identifying both the number and the nature
of dimensions shared by psychopathy, antisocial personality, and criminal lifestyle. In a
principal components analysis regarding teacher ratings of student aggression, Dodge and
Coie (1987) uncovered two factors, which they labeled proactive aggression and reactive
aggression. The two-dimensional model of childhood aggression was replicated in a series
of confirmatory factor analyses (Poulin & Boivin, 2000), and scales that were designed to
measure proactive and reactive aggression in children have a moderately high and relatively
narrow range of intercorrelationspecfically, .77 to .83 (Day, Bream, & Pal, 1992; Dodge
& Coie, 1987; Hubbard et al., 2002; Poulin & Boivin, 2000; Price & Dodge, 1989). Despite
a high degree of intercorrelation between factors and a belief on the part of some investigators that the proactivereactive breakdown has outlived its usefulness (Bushman &
Anderson, 2001), the factors each exhibit a differential pattern of correlation with outside
criteria: proactive aggression with positive outcome expectancies for aggression and reactive
aggression with hostile attribution biases (Crick & Dodge, 1996). This same countervailing
relationship has been observed in incarcerated juvenile delinquents (Smithmyer, Hubbard,
& Simons, 2000) and adult prison inmates (Walters, 2007b).
The two-dimensional model advanced in this article holds that proactive and reactive
criminality are psychological functions (motives) with developmental roots in proactive and
Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES

1461

reactive childhood aggression. Although these two functions derive from different theoretical perspectivesproactive criminality/aggression from socialcognitive learning theory
(Bandura, 1986) and reactive criminality/aggression from frustrationaggression response
theory (Berkowitz, 1993)they share a great deal in common and thus overlap extensively.
Amygdala and orbital frontal cortex dysfunction have been implicated in both proactive and
reactive aggression, although the nature of the deficit and the actual brain pathways involved
in each pattern appear to differ (Blair, 2004). In addition, proactive and reactive aggression
are correlated with peer rejection and delinquency; however, reactive aggression is associated with emotional dysregulation and poor social adjustment, and proactive aggression is
associated with better psychological and social adjustment (Card & Little, 2006). The twodimensional model proposes that the motives that drive childhood aggression extend into
adult criminality and exist in the form of two overlapping dimensions: proactive aggression/
criminality and reactive aggression/criminality. These two dimensions, despite a moderate to
high degree of intercorrelation, demonstrate semidistinct patterns of association. Proactive
aggression/criminality correlates with positive outcome and efficacy expectancies for
aggression/crime, and reactive aggression/criminality correlates with poor socialemotional
adjustment and hostile attribution biases for aggression/crime.
Indicators from self-report measures of criminal lifestyle, antisocial personality, and psychopathy can be organized in ways other than the proactivereactive breakdown proposed
by the two-dimensional model of criminality. One alternate conceptualization is to assign all
the criminal thinking indicators to one factor and all the antisocial behavior indicators to a
second factor. Social learning theory has been used to explain aggression (Bandura, 1973)
and criminality (Akers & Jensen, 2006). Five core assumptions underpin social learning
theory: Learning is a social process; learning is an internal process; behavior is directed
toward particular goals; behavior eventually becomes self-regulated; and reinforcement and
punishment have direct (behavioral) and indirect (cognitive) effects (Bandura, 1986). As such,
social learning theory provides a bridge, or transition, between behavioral learning theories
and cognitive learning theories (Ormrod, 1999). Because social learning theory focuses on
the cognitive and behavioral aspects of learning, it is more apt to divide indicators from selfreport measures of criminal lifestyle, antisocial personality, and psychopathy along cognitive and behavioral lines than along proactive and reactive lines, despite serving as the
conceptual foundation for proactive aggression/criminality. Consequently, one alternate
model against which the two-dimensional model (proactive, reactive) is compared is a social
learning alternative composed of cognitive and behavioral factors.

SELECTING INDICATORS

The Psychological Inventory of Criminal Thinking Styles (PICTS; Walters, 1995) is a


well-researched self-report measure of criminal thinking. Previous studies have shown that
the PICTS can be partitioned into two general factors (see Walters, 2005a) that Egan,
McMurran, Richardson, and Blair (2000) label wilful criminality and lack of thoughtfulness.
The five scales that Egan et al. assigned to the wilful criminality factor (Mollification [Mo],
Entitlement [En], Power Orientation [Po], Sentimentality [Sn], and Superoptimism [So])
appear to reflect proactive criminal thinking, whereas the three scales that they assigned to
the lack of thoughtfulness factor (Cutoff [Co], Cognitive Indolence [Ci], Discontinuity [Ds])
Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1462

CRIMINAL JUSTICE AND BEHAVIOR

apparently reflect reactive criminal thinking. Results from several exploratory and confirmatory factor analyses, however, suggest that Sn may belong to a third factor, commonly
referred to as denial of harm (Walters, 1995, 2002, 2005a). One question posed by this article is whether or not to include the Sn in the present analyses given its uncertain status with
respect to the proactivereactive dimensions and whether it loads as well onto a general
criminal thinking factor as do the other seven PICTS scales.
The LSRP scales (Levenson et al., 1995) were created to assess psychopathy in nonincarcerated populations, but they have also been used in incarcerated populations (Brinkley,
Schmitt, Smith, & Newman, 2001). In an effort to create self-report indices comparable to
Factor 1 (callous and remorseless use of others) and Factor 2 (chronic antisocial lifestyle)
of the PCL-R, Levenson et al. (1995) divided the 26 LSRP items into primary and secondary
psychopathy scales, with the primary scale serving as a proxy for Factor 1 and the secondary
scale as a proxy for Factor 2. Whereas the LSRP Secondary Psychopathy Scale (LSRP-SP)
is a reasonably good index of the antisocial behavior tapped by Factor 2 of the PCL-R,
questions have been raised about the construct validity of the LSRP Primary Psychopathy
Scale (LSRP-PP; Lilienfeld & Fowler, 2006). Chief among these concerns is the absence
of a meaningful relationship between (a) the LSRP-PP and low trait anxiety (Levenson et al.,
1995; McHoskey, Worzel, & Szyarto, 1998) and (b) the scales tendency to correlate higher
with Factor 2 of the Psychopathic Personality Inventory than with Factor 1 (Lilienfeld,
Skeem, & Poythress, 2004; Wilson, Frick, & Clements, 1999). In addition, nearly one third
of the LSRP-PP items are reverse scored, which may introduce error into the responses of
individuals with lower reading skills and less motivation than that of the normative college
sample. Consequently, another goal of this article is to determine whether the LSRP-PP
should be included in subsequent analyses.
The Antisocial Features Scale (ANT) of the Personality Assessment Inventory (PAI;
Morey, 2007) is composed of 24 items designed to assess the behavioral and personality
characteristics of antisocial personality and psychopathy. There are three ANT subscales:
Antisocial Behaviors (ANT-A), Egocentricity (ANT-E), and Stimulus Seeking (ANT-S).
According to Morey (2007), elevated ANT-E scores reflect inflated self-importance, callousness, low social anxiety, and the desire to satisfy personal goals and impulses at the
expense of others. This description suggests that ANT-E measures proactive, as opposed to
reactive, criminality. By contrast, ANT-S seems to align more closely with reactive criminality, as indicated by high levels of recklessness, impulsivity, and novelty seeking in those
who score high on this subscale (Morey, 2007). To the extent that ANT-A is a catalogue of
a persons level of prior antisocial activity, it could reflect either proactive or reactive criminality. The question posed by Study 2 is whether ANT-A, like the LSRP-PP, loads sufficiently well onto an antisocial behavior factor to be retained in this study.

HYPOTHESES

Five hypotheses were tested in this article, one for each study:
Hypothesis 1: Sn will be the lowest loading indicator on a general criminal thinking factor in an
exploratory factor analysis of the eight PICTS scales, thus justifying its removal from subsequent analyses.

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES

1463

Hypothesis 2: The LSRP-PP and the ANT-A will be the two lowest loading indicators on a general antisocial behavior factor in an exploratory factor analysis of the five LSRP/ANT indicators, thus justifying their removal from subsequent analyses.
Hypothesis 3: Grouping the PICTS, LSRP, and ANT indicators into proactive and reactive factors
(two-dimensional model) will provide a significantly better fit for data collected on a large
sample of medium-security federal prisoners than will loading all the indicators onto a single
factor (general model) or grouping the indicators into criminal thinking and antisocial behavior
factors (social learning model).
Hypothesis 4: Results from Study 3 (Hypothesis 3) will be replicated in a smaller sample of maximumsecurity federal prisoners.
Hypothesis 5: The construct validity of the two-dimensional model will be supported in a path
analysis of correlations between the proactive and reactive dimensions and the measures of
positive outcome expectancies for crime and hostile attribution biases.

STUDY 1
METHOD

Participants. Participants for this first study included 625 male federal prisoners who
completed the PICTS during a routine intake procedure at a medium-security federal prison.
These 625 PICTS protocols had never been included in any previous factor analyses of the
PICTS (Walters, 1995, 2002, 2005a). The mean age of the participants was 35.11 years
(SD = 9.04), and the average educational level was 11.33 years (SD = 1.94). Ethnically,
over half the sample was Black (58.9%, n = 368), with Whites making up 24.5% (n = 153);
Hispanics, 15.5% (n = 97); and Asian/Native Americans, 1.1% (n = 7). The majority of participants listed their marital status as single (65.1%, n = 407), followed by married (22.4%,
n = 140), divorced/separated (11.2%, n = 70), and widowed (1.3%, n = 8). The modal
instant offense in this sample was drugs (43.8%, n = 274), followed by illegal weapons
(17.1%, n = 107), robbery (14.4%, n = 90), miscellaneous offenses (11.2%, n = 70), violence
(9.1%, n = 57), and property crimes (4.3%, n = 27).
Measure. The PICTS is a self-report inventory with 80 items rated on a 4-point Likert-type
scale: Strongly agree responses earn a respondent 4 points; agree, 3 points; uncertain,
2 points; and disagree, 1 pointexcept for the DefensivenessRevised Scale, which is
reverse scored (strongly agree = 1, disagree = 4). Aside from the two 8-item validity scales
(ConfusionRevised [Cf-r] and DefensivenessRevised), the PICTS generates scores for
eight nonoverlapping 8-item thinking-style scales (Mo, Co, En, Po, Sn, So, Ci, and Ds), four
10-item factor scales (Problem Avoidance, Infrequency, Self-Assertion/Deception, and
Denial of Harm), two content scales (Current and Historical), two composite scales (Proactive
Criminal Thinking and Reactive Criminal Thinking), and one general score (General Criminal
Thinking). The present investigation focuses on the eight thinking-style scales, all of which
have been found to possess adequate reliability (r = .73.93 after 2 weeks; r = .47.86 after
1012 weeks), internal consistency ( = .54.79) and validity (unweighted mean correlations
of .12.20 with institutional adjustment/recidivism; Walters, 2002, 2006).
Procedure. The PICTS is routinely administered to inmates within 2 weeks of their arrival at
the institution where this study took place. Of the 687 inmates who arrived at the institution

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1464

CRIMINAL JUSTICE AND BEHAVIOR

during an 18-month period, 2 refused to be tested; 21 could not read well enough to complete
the PICTS; 24 left more than 10 PICTS items blank; 6 produced extreme scores on the
ConfusionRevised Scale (T-score > 100); and 9 selected the uncertain option for all 80
PICTS items. Eliminating these 62 individuals from the study resulted in a final sample of
625 participants. Informed consent was not required because administration of the PICTS was
a routine clinical procedure. Nonetheless, institutional review board approval was obtained
for the use of these data in research. A principal-axis factor analysis of a single general factor
(criminal thinking) was conducted using the eight PICTS scales as indicators.
RESULTS

A single factor was extracted from a principal-axis factor analysis of the eight PICTS
scales using 625 male inmates of a medium-security federal prison. The first factor accounted
for 63.12% of the total variance in the eight thinking styles (eigenvalue = 5.05), and the second factor accounted for less than 10% of the variance in the thinking-style scales (eigenvalue =
0.79). Factor loadings on the first factor (general criminal thinking) were as follows: Mo = .753,
Co = .789, En = .795, Po = .799, Sn= .628, So = .765, Ci = .823, and Ds = .724.
DISCUSSION

As predicted, the Sn scale was the weakest loading indicator in an exploratory factor
analysis of the eight PICTS scales. The Sn scale has not loaded particularly well onto the
proactive or reactive factor in previous factor analytic research (Walters, 1995, 2005a), and
in the present study it was the weakest correlating indicator when the eight PICTS scales were
loaded onto a general criminal thinking factor. As such, it was dropped for the confirmatory
factor analyses in Studies 3 and 4 and the construct validity analyses in Study 5.

STUDY 2
METHOD

Participants. Participants included 1,702 federal prisoners (n = 1,221 males, n = 481


females) who completed the LSRP and PAI as part of a national mental health prevalence
study conducted in 14 federal correctional institutions (Diamond & Magaletta, 2006). The
average age of each participant was 34.29 years (SD = 9.59) with 11.17 years of education
(SD = 2.54). The ethnic breakdown was as follows: White, 32.0% (n = 544); Black, 40.1%
(n = 683); Hispanic, 26.8% (n = 456); Asian/Native American, 1.1% (n = 19). Over a third
of the sample (37.7%, n = 600) had never been married, with 40.2% (n = 639) describing
their current marital status as married or common law, 20.7% (n = 329) as divorced or separated, and 1.5% (n = 24) as widowed. The majority of participants came from low-security
institutions (55.6%, n = 947), with 25.4% (n = 432) emanating from medium-security institutions and 19.0% (n = 323) from high-security institutions.
Measures. The LSRP is a 26-item self-report inventory designed to assess psychopathy in
nonincarcerated populations. The first 16 items measure primary psychopathy (i.e., LSRP-PP;
affective and interpersonal features), and the last 10 items measure secondary psychopathy
Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES

1465

(i.e., LSRP-SP; chronic antisocial lifestyle). Each LSRP item is rated on a 4-point Likert-type
scale (disagree strongly, disagree somewhat, agree somewhat, agree strongly), with seven of
the items being reversed scored to control for various response style or test-taking sets, such
as social desirability. Reliability (LSRP-PP: = .82; LSRP-SP: = .63) and validity (correlations with the PCL-R and passive avoidance errors) have been found to be satisfactory
(Brinkley et al., 2001; Epstein, Poythress, & Brandon, 2006), although concerns have been
raised about the construct validity of the LSRP-PP (Lilienfeld & Fowler, 2006).
The PAI is a 344-item self-report measure in which each item is rated on a 4-point scale
(1 = very true, 2 = mainly true, 3 = slightly true, 4 = false). In this study, only the three 8-item
nonoverlapping subscales of the ANT were employed: ANT-A, ANT-E, ANT-S. Internal
consistency (as measured by the alpha coefficient []) and testretest reliability (as measured after 2428 days [r]) are satisfactory for the ANT-A ( = .73.80, r = .80.86), ANT-E
( = .63, r = .70.79), and ANT-S ( = .69.77, r = .78.84), and there is evidence for both
the reliability and the validity of the ANT in correctional and forensic samples (Edens &
Ruiz, 2005).
Procedure. The LSRP and PAI were normally administered during a single testing session, although one test was occasionally administered several days after the rest of the test
battery. Spanish versions of each test were available for Spanish-speaking inmates who
could not read English. There were 225 inmates from the national mental health prevalence
study who completed both the LSRP and PAI but were not included in the final sample of
1,702 participants because they achieved T-scores of 80 or higher on the PAI Inconsistency
Scale, 80 or higher on the PICTS Infrequency Scale, or 92 or higher on the PAI Negative
Impression Scale. Informed consent was obtained from the inmates who participated in the
original national mental health prevalence study, and institutional review board approval
was sought and obtained for the use of these data in research. A principal-axis factor analysis of a single general factor (antisocial behavior) was conducted using the two LSRP scales
and three ANT subscales.
RESULTS

A single factor was extracted from the five LSRP/ANT indicators in a sample of 1,702
male and female inmates from 14 federal facilities. The first factor accounted for 50.66%
of the total variance in the eight thinking styles (eigenvalue = 2.53), and the second factor
accounted for 17.64% of the variance in the PICTS scales (eigenvalue = 0.88). Factor loadings on the first factor (general antisocial behavior) were as follows: LSRP-PP = .554,
LSRP-SP = .574, ANT-A = .562, ANT-E = .681, and ANT-S = .721.
DISCUSSION

The hypothesis for Study 2 was that the LSRP-PP and the ANT-A would be the two
weakest loading indicators on a general antisocial behavior factor when the five indicators
from the LSRP and ANT were subjected to an exploratory factor analysis in a large group
of federal prisoners. The results support this hypothesis and suggest that the LSRP-PP and
the ANT-A could be removed from subsequent analyses because their status as proactive or
reactive measures is uncertain and they do not load particularly well onto the antisocial
behavior factor of the social learning model.
Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1466

CRIMINAL JUSTICE AND BEHAVIOR

STUDY 3
METHOD

Participants. A group of 521 male inmates from a medium-security federal prison


located in the northeastern United States served as participants in this study. Each participant produced a complete and valid PICTS (no more than 10 unanswered items and a
ConfusionRevised Scale T-score of 100 or less), LSRP (no more than 2 unanswered
items), and ANT (no more than 2 unanswered items), all of which were administered routinely within 2 weeks of an inmates arrival at the institution. Validity indices for the PAI
were unavailable because the 24 ANT items were administered separately as a single instrument rather than imbedded in the larger 344-item PAI. The average age of inmates in this
sample was 34.64 years (SD = 9.97), and the mean educational level was 11.42 years (SD =
1.46). The ethnic breakdown was as follows: Black, 68.1% (n = 355); White, 16.5% (n =
86); Hispanic, 13.4% (n = 70); and Asian/Native American, 2.0% (n = 10). Marital status
was as follows: single, 74.9% (n = 390); married, 16.1% (n = 84); divorced, 8.3% (n = 43);
and widowed, 0.8% (n = 4). The modal confining offense was drugs (45.3%, n = 236), followed by miscellaneous offenses, such as firearms and fraud (32.8%, n = 171), robbery
(13.1%, n = 68), violence (6.0%, n = 31), and property crimes (2.9%, n = 15). The present
sample was independent of previous samples used to test the taxometric structure of the
PICTS (Walters, 2007a; Walters & McCoy, 2007), LSRP (Walters et al., in press), and ANT
(Walters, Diamond, et al., 2007).
Measures. Seven of the eight 8-item PICTS scales (Mo, Co, En, Po, So, Ci, Ds) served as
indicators in this study. Research indicates that PICTS scales possess adequate to good reliability (r = .73.93 after 2 weeks; r = 47.86 after 1012 weeks), internal consistency
( = .54.79), unidimensionality (precision of = .01.03),1 and validity (unweighted mean
correlations of .12.20 with institutional adjustment/recidivism; Walters, 2006). The 10-item
LSRP-SP was the eighth indicator employed in this study. Internal consistency ( = .67, precision of = .02) is adequate, and validity is reasonable (Brinkley et al., 2001; McHoskey
et al., 1998) for the LSRP-SP. Two of three ANT subscales, ANT-E and ANT-S, also served
as indicators in this study. In this study, the 24 ANT items were administered as a 24-item
inventory rather than as part of the full PAI. Unidimensionality, as measured by the precision
of the alpha coefficient, was satisfactory in the present sample of participants: ANT-E (.02)
and ANT-S (.03).
Procedure. The PICTS, LSRP, and ANT were administered in random order to all participants during a single testing session, although in 10% to 15% of cases, one of the measures
was completed several days after the other measures. Only inmates who could read English
were included in the investigation. Overall, 4 inmates refused to be tested, and 74 inmates participated in the testing but were excluded from the final sample because of reading, language,
and education difficulties (less than 6 years of formal schooling; n = 38), random responding
(n = 7), missing data (n = 26), or achievement of a T-score of greater than 100 on the PICTS
ConfusionRevised Scale (n = 3). PICTS protocols with no more than 10 missing items and
LSRP and ANT protocols with no more than 2 missing items were included in this investigation. Valid protocols with missing items were prorated by (a) calculating an average item

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES

1467

score for the items that were completed and (b) adding this number (or 2 times this number,
in the case of 2 missing items) to the raw total for the scale.
Informed consent was not required, because testing was a routine clinical procedure at
the institution where this study took place. Nonetheless, institutional review board approval
was obtained from the Bureau of Prisons for the use of these data in research. Following
approval, data were fit to three models: The one-factor model (M1) loaded all 10 indicators
onto a single latent factor; the two-dimensional (proactivereactive) model (M2) loaded
Mo, En, Po, So, and ANT-E onto a proactive latent factor and Co, Ci, Ds, LSRP-SP, and
ANT-S onto a reactive latent factor; the social learning model (MSL) loaded Mo, Co, En,
Po, So, Ci, and Ds onto a criminal thinking latent factor and LSRP-SP, ANT-E, and ANT-S
onto an antisocial behavior latent factor. Each model was estimated with maximum likelihood, and metrics were set at 1.00 for the first pathway between a latent factor and an
observed variable (indicator) and between each error term and observed variable. All analyses were conducted with a structural equation modeling (SEM) program (Amos 4.0;
Arbuckle & Wothke, 1999).
Maximum likelihood estimation (MLE) is the most common method for estimating the
coefficients in SEM analysis, although several assumptions must be met before this estimation approach can be used. First, the sample size must be adequate. As such, sample size was
determined to be adequate, using a power analysis (MacCallum, Browne, & Sugawara,
1996). Second, MLE assumes multivariate normality because highly nonnormal data can
lead to inflated-model chi-square values and downwardly biased parameter standard errors
(Bentler & Chou, 1987). Third, the sample covariance matrix that MLE attempts to reproduce assumes linearity (Klein, 2005). Univariate and multivariate normality, as well as
homoscedasticity (homogeneity of covariance matrices), were tested to determine whether
data should be transformed, and 90% confidence intervals (CIs) for the standardized regression coefficients were calculated with 2,000 bootstrapped trials. Finally, MLE assumes that
the models have been validly specified. Critical ratios (CRs) for the unstandardized regression coefficients and standardized residual covariances were consequently computed. CR
values were calculated by dividing the regression or factor estimate by the standard error of
the estimate (with scores above 1.96 denoting a significant effect at the .05 level). Fit statistics employed in the present investigation included the model chi-square, the comparative fit
index (CFI; Bentler, 1990), the root mean square error of approximation (RMSEA; Browne
& Cudeck, 1993), and the Akaike information criterion (AIC; Akaike, 1987).
RESULTS

Normality and homoscedasticity. Table 1 lists the means, standard deviations, ranges,
skew, and kurtosis of the 10 indicator variables. All such indicators showed signs of significant univariate skew (CR > 1.96), and Mardias (1974) coefficient of multivariate kurtosis
revealed the presence of significant multivariate kurtosis (CR = 14.68). Heteroscedasticity
was assessed by constructing a multiple linear regression in which Mo was arbitrarily
selected as the outcome variable and the other 9 indicators served as predictor variables.
With a procedure described by Pryce (2005), the unstandardized residuals of the multiple
regression were saved, and Levenes (1960) Test for the Equality of Variances was run on
each indicator, divided at the median into a high-scoring group and a low-scoring group. A

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1468

CRIMINAL JUSTICE AND BEHAVIOR

TABLE 1:

Descriptive Statistics for the 10 Indicators in Study 3

Raw Scores

Percentile Ranks

Indicator

Range

SD

Skew

PICTS
Mo
Co
En
Po
So
Ci
Ds
LSRP-SP
ANT-E
ANT-S

8-26
8-31
8-29
8-29
8-28
8-32
8-32
10-37
0-22
0-22

13.01
13.55
13.25
12.49
15.00
16.02
15.44
21.14
4.01
6.28

4.12
5.12
3.84
4.08
4.11
4.76
5.02
5.12
3.88
3.94

0.82
0.87
0.86
1.15
0.71
0.34
0.58
0.26
1.33
0.86

Kurtosis

0.18
0.11
0.57
1.13
0.14
0.35
0.04
0.34
1.93
0.60

Skewa

Kurtosisb

0.02
0.04
0.01
0.03
0.02
0.00
0.00
0.00
0.03
0.01

1.23
1.27
1.20
1.24
1.21
1.20
1.20
1.20
1.24
1.20

Note. N = 521. See appendix for all abbreviations used in the article.
a. Standard error of skew was .11.
b. Standard error of kurtosis was .21.

significant Levene F test was interpreted as a sign of heteroscedasticity, based on the presence of significantly different group variances. Heteroscedasticity was observed in all 10
indicators. Hence, nonnormality and heteroscedasticity were characteristic of the indicators
used in this study. According to research, under conditions of nonnormality and heteroscedasticity, ranking methods and transformations are useful (Zwick, 1986), and a percentileranking transformation may be a particularly powerful and reliable method for transforming
nonnormal and heteroscedastic data (Zimmerman & Zumbo, 2005).
Three data transformations were examined in an effort to determine which did the best
job of normalizing the distribution and homogenizing the sample covariances. A base-e logarithmic transformation of the data produced significant univariate skew on 10 indicators, 1
heteroscedastic indicator, and a significant coefficient of multivariate kurtosis (CR = 3.96).
A square root transformation, however, produced 8 skewed indicators, 8 heteroscedastic
indicators, and a significant coefficient of multivariate kurtosis (CR = 5.24). The percentilerank transformation was the only data transformation procedure to yield no skewed indicators, no heteroscedastic findings, and a nonsignificant coefficient of multivariate kurtosis
(CR = 1.60). The platykurtotic distribution produced by the percentile-rank transformations
(negative kurtosis; see Table 1) is a consequence of the rectangular nature of the percentile
distribution, although this feature of the percentile-rank transformation does not impede its
ability to serve as an effective proxy for statistical analysis (Zimmerman & Zumbo, 2005).
Accordingly, percentile-rank transformations were employed in this study.
Regression weights and residual covariances. The first column of Table 2 reproduces the
unstandardized regression weights for the coefficient pathways in a confirmatory factor
analysis of percentile-rank transformation indicators organized into the two-dimensional
model, followed by the standard error of estimate, the CR, the standardized coefficient, and
the 90% bootstrapped CI of the standardized coefficient. All CRs in the two-dimensional
model were significant at the .001 level; likewise, all CRs in the one-factor and social learning models were significant at the .001 level.

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES


TABLE 2:

1469

Regression Weights Obtained With the Two-Dimensional Model: Study 3

Regression Path

Estimate

SE

CR

Mo Proactive
En Proactive
Po Proactive
So Proactive
ANT-E Proactive
Co Reactivea
Ci Reactive
Ds Reactive
LSRP-SP Reactive
ANT-S Reactive
Proactive Reactive

1.000
1.012
1.040
0.987
0.811
1.000
0.959
0.935
0.795
0.665
448.57

0.059
0.059
0.059
0.060

0.042
0.043
0.046
0.048
38.57

17.12***
17.66***
16.67***
13.56***

22.61***
21.76***
17.24***
13.74***
11.63***

(90% CI)
.75
.76
.78
.74
.61
.86
.82
.80
.68
.57
.84

(.71.79)
(.72.80)
(.75.82)
(.70.78)
(.56.66)
(.83.88)
(.79.85)
(.76.83)
(.63.72)
(.51.62)
(.80.88)

Note. Estimate = unstandardized regression coefficient; (90% CI) = standardized coefficient and 90th-percentile
biased corrected confidence interval of the standardized coefficient (B = 2,000).
a. Set to 1; no CR possible.
***
p < .001.

TABLE 3:

Goodness-of-Fit Statistics: Study 3

Model

2 (N = 521)

df

CFI

RMSEA (90% CI)

AIC

AIC

wAIC

312.72***
170.20***
259.85***

35
34
34

.90
.95
.92

.124 (.111.136)
.088 (.075.101)
.113 (.100.126)

352.72
212.20
301.85

140.52
0.00
89.65

.0000000
.9999999
.0000000

M1
M2
MSL

Note. Model = type of model based on number and configuration of factors; M1 = one-factor model (general
dimension); M2 = two-dimensional model (proactive and reactive); MSL = social learning model (criminal thinking,
antisocial behavior); AIC = difference between AIC values obtained by the different models, with the lowest value
set at 0; wAIC = Akaike weight.
***
p < .001.

Standardized residual covariances measure the difference between the sample covariance matrix and the implied covariance matrix, divided by the sample covariance standard
deviation. Discrepancies were defined by a CR significant at the .05 level (CR > 1.96).
There were two significant discrepancies in the two-dimensional model (between ANT-E
and ANT-S and between So and LSRP-SP), five significant discrepancies in the one-factor
model, and two significant discrepancies in the social learning model.
Goodness-of-fit indices. Results produced by the goodness-of-fit indices are listed in Table 3.
Whereas the two-dimensional model displayed adequate fit on the CFI and RMSEA, the
one-factor and social learning models displayed borderline to poor fit. Direct comparisons
between models using the AIC statistic (final two columns of Table 3) reveal a highly significant difference in relative fit between the individual models. An AIC difference (AIC) of
less than 2 is considered nonsignificant, whereas differences of 2-4, 4-7, 7-10, and more than
10 provide weak, definite, strong, and very strong evidence, respectivelythat is, the model
with the lower AIC value is superior to the model with the higher AIC value (Burnham &
Anderson, 2002). Differences of 140.52 and 89.65 are therefore highly significant and so
indicate very strong evidence that the two-dimensional model provides a significantly better
fit for the data than either the one-factor model or the social learning model.

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1470

CRIMINAL JUSTICE AND BEHAVIOR

DISCUSSION

The results of this third study furnish preliminary support for the hypothesis that two
correlated dimensionsproactive and reactive criminalityunderpin popular crime-related
constructs, such as psychopathy, antisocial personality, and criminal lifestyle. Results are
consistent in showing (a) fair to modest absolute fit for the two-dimensional model and
highly significant (p < .001) paths between each indicator and its assigned factor (proactive
or reactive) and (b) modest to poor absolute fit for the two alternate models (one-factor
model, social learning model) with which the two-dimensional model was compared. When
direct nonnested comparisons were made (AIC), the two-dimensional model proved superior to both alternate models. Overall, the results of this study support the hypothesis that
self-report measures of psychopathy, antisocial personality, and lifestyle criminality share
two general dimensions. Whether these two dimensions reflect proactive and reactive criminality, however, requires further investigation, in terms of cross-validating these preliminary
findings and testing the construct validity of the 10 indicators as measures of proactive and
reactive criminality.

STUDY 4
METHOD

Participants. Participants were 116 maximum-security male inmates who were administered the PICTS, LSRP, and ANT as part of the standard intake procedure for a unit-based
psychology program held in a U.S. federal penitentiary in the mid-Atlantic region of the
United States. As in Study 3, informed consent was not sought, given the clinical nature of
the data collection procedures, although institutional review board approval was obtained
for the use of these data in research. The average participant in this study was 35.03 years
of age (SD = 8.49) and had accumulated 11.22 years of education (SD = 1.44). Ethnically,
62.9% (n = 73) of the participants were Black; 30.2% (n = 35), White; 5.2% (n = 6),
Hispanic; and 1.7% (n = 2), Asian/Native American. Over three quarters of the sample characterized their marital status as single (n = 94, 81.0%), with the remainder of the sample
being composed of married (n = 16, 13.8%) and divorced (n = 6, 5.2%) participants. The
modal confining offense was robbery (27.6%, n = 32), followed by violent crimes (26.7%,
n = 31), drugs (22.4%, n = 26), firearms (18.1%, n = 21), property crimes (3.4%, n = 4),
and miscellaneous offenses (1.7%, n = 2).
Measures. The 10 indicators from the PICTS, LSRP, and ANT of the previous study
were employed as indicators in the present investigation, although there was one noteworthy administrative difference between the two studies. Whereas the ANT was administered
as a stand-alone procedure in Study 3, it was completed as part of the full PAI in the present
study. Consequently, scores on the PAI Inconsistency Scale, the PICTS Infrequency Scale,
and the PAI Negative Impression Scale were available to assess protocol validity. The
following were screened out of the sample: PAI protocols with more than 20 omitted items,
a PAI Inconsistency Scale T-score of 80 or more, a PICTS Infrequency Scale T-score of 80
or more, a PAI Negative Impression Scale T-score of 92 or more, PICTS protocols with
more than 10 omitted items, or a ConfusionRevised Scale T-score of more than 100. These
criteria resulted in the elimination of two protocols from the present study.
Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES


TABLE 4:

1471

Descriptive Statistics for the 10 Indicators in Study 4

Raw Scores

Percentile Ranks
a

Indicator

Range

SD

Skew

PICTS
Mo
Co
En
Po
So
Ci
Ds
LSRP-SP
ANT-E
ANT-S

824
829
825
824
826
832
830
1037
019
017

13.34
15.34
14.95
13.09
16.53
17.91
16.82
20.84
3.94
5.93

3.87
5.33
4.12
3.89
4.30
4.83
5.23
5.20
3.50
3.92

0.54
0.26
0.16
0.56
0.40
0.01
0.44
0.35
1.50
0.89

Kurtosis

0.44
0.86
0.82
0.56
0.33
0.36
0.73
0.26
3.17
0.01

Skewa

Kurtosisb

0.02
0.02
0.00
0.02
0.00
0.00
0.00
0.00
0.02
0.02

1.22
1.24
1.20
1.22
1.19
1.21
1.20
1.20
1.22
1.20

Note. N = 116.
a. Standard error of skew was .22.
b. Standard error of kurtosis was .45.

Procedure. The procedure for the present investigation was identical to that utilized in
the first study, except that the sample was composed of male inmates from a maximumsecurity penitentiary, instead of male inmates from a medium-security federal correctional
institution.
RESULTS

Normality check. Table 4 lists the means, standard deviations, ranges, skew, and kurtosis
of the 10 indicator variables. Eight indicators (Mo, Co, En, Po, So, Ds, ANT-E, ANT-S)
show signs of significant univariate skew (CR > 1.96); Mardias (1974) coefficient of multivariate kurtosis reveals significant multivariate kurtosis (CR = 4.41); and two indicators
(Ds, LSRP-SP) dichotomized at the median display significant heteroscedasticity (p < .05).
Percentile-rank transformations, however, display no signs of univariate skew, multivariate
kurtosis (CR = 1.56), or heteroscedasticity. Percentile-rank transformations were employed
accordingly in this study.
Regression weights and residual covariances. Table 5 lists the unstandardized regression
estimates, standard errors, and CRs for the regression and covariance paths of the 10 indicators. All CRs in the one-factor, two-dimensional, and social learning models were significant
at the .001 level. The standardized regression coefficients for each path in the two-dimensional
model and the 90% bootstrapped CI are also reported. There was one significant standardized residual covariance in the two-dimensional model (between ANT-E and ANT-S), one
significant standardized residual covariance in the one-factor model, and no significant
standardized residual covariances in the social learning model.
Goodness-of-fit indices. When goodness-of-fit indices were applied to the three models,
there was evidence of modest fit for the two-dimensional model and generally poor fit for
the one-factor and social learning models (see Table 6). Direct comparisons (AIC) between
the two-dimensional model and the two competing models indicate strong and very strong
Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1472

CRIMINAL JUSTICE AND BEHAVIOR

TABLE 5:

Regression Weights Obtained With the Two-Dimensional Model: Study 4

Regression Path

Estimate

SE

CR

Mo Proactive
En Proactive
Po Proactive
So Proactive
ANT-E Proactive
Co Reactivea
Ci Reactive
Ds Reactive
LSRP-SP Reactive
ANT-S Reactive
Proactive Reactive

1.000
0.945
0.929
0.856
0.505
1.000
0.945
1.017
0.541
0.527
448.99

0.129
0.129
0.130
0.131

0.100
0.098
0.113
0.113
82.95

7.30***
7.20***
6.60***
3.85***

9.45***
10.38***
4.80***
4.68***
5.41***

(90% CI)
.76
.71
.70
.64
.38
.83
.79
.85
.45
.44
.86

(.62.85)
(.60.80)
(.61.78)
(.50.75)
(.21.55)
(.73.90)
(.69.86)
(.76.91)
(.29.58)
(.27.57)
(.76.94)

Note. Estimate = unstandardized regression coefficient; (90% CI) = standardized coefficient and 90th-percentile
biased corrected confidence interval of the standardized coefficient (B = 2,000).
a. Set to 1; no CR possible.
***
p < .001.

TABLE 6:

Goodness-of-Fit Statistics: Study 4

Model

2 (N = 236)

df

CFI

RMSEA (90% CI)

AIC

AIC

wAIC

87.76***
72.34***
81.04***

35
34
34

.89
.92
.90

.114 (.085.145)
.099 (.067.131)
.110 (.079.141)

127.76
114.34
123.04

13.42
0.00
8.70

.0018332
.9860960
.0127206

M1
M2
MSL

Note. Model = type of model based on number and configuration of factors; M1 = one-factor model (general
dimension); M2 = two-dimensional model (proactive and reactive); MSL = social learning model (criminal thinking,
antisocial behavior); AIC = difference between AIC values obtained by the different models, with the lowest value
set at 0; wAIC = Akaike weight.
***
p < .001.

evidence that the two-dimensional model provides a significantly better fit for the data than
that of the social learning and one-factor models, respectively.
DISCUSSION

The modest shrinkage in absolute goodness-of-fit from Study 3 to Study 4 may have
more to do with reduced power than weak theory. The reason is that the sample in Study 4
had much less power to accept models with close fit (0.36) and reject models with not-close
fit (0.22) than that of the sample in Study 3, where the power to accept models with close
fit and reject models with not-close fit both exceeded 0.90. Taken as a whole, these findings confirm the hypothesis that a two-dimensional model with proactive and reactive latent
dimensions may have value in explaining criminality as assessed by offender self-report.
Like that of Study 3, the absolute fit of the two-dimensional model to the data in this study
was less impressive than the relative fit of the two-dimensional model, in comparison to the
one-factor and social learning models. Even though the present study successfully crossvalidated the relationships observed in Study 3 using a small group of penitentiary inmates,
it still did not answer one very important question: namely, how can we be sure that the two
latent factors in the two-dimensional model actually represent proactive and reactive criminality? To answer this question, a fifth study was conducted to test the construct validity
of the proactive and reactive dimensions of the two-dimensional model.
Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES

1473

STUDY 5
METHOD

Participants. Participants for the fifth study were male inmates who completed testing during the second half of the year, when data were being collected for the third study and when
a change in policy added two new measures to the standard intake test battery: the Outcome
Expectancies for Crime Inventory (OEC; Walters, 2003) and the Hostile Attribution Bias
(HAB) measure. There were 356 inmates who were processed into the institution during this
6-month period. Of this number, 3 refused to be tested; 20 could not read English; 8 experienced significant reading problems or had fewer than 6 years of education; 3 produced invalid
PICTS protocols (2 left more than 10 items unanswered and 1 had a T-score of more than 100
on the ConfusionRevised Scale); 2 left more than 2 items blank on the LSRP; 3 left more
than 2 items blank on the ANT scale; 23 left more than 2 items blank on the OEC; and 3 left
1 or more items blank on the HAB. This resulted in a final sample of 291 male inmates, with
a mean age of 33.56 years (SD = 9.26) and mean educational level of 11.48 years (SD = 1.38).
The ethnic breakdown for the sample was as follows: Black, 68.0%; White, 15.5%; Hispanic,
13.7%; and Asian/Native American, 2.7%. Over three quarters of the sample (77.0%) listed
their marital status as single, with another 15.1%, 6.9%, and 1.0% stating that they were married, divorced, and widowed, respectively. Nearly half the sample was serving time for a drug
offense (45.7%), with 13.4% serving time for robbery, 3.4% for violence, 3.1% for a property
crime, and 34.4% for a miscellaneous offense.
Measures. In addition to the PICTS, LSRP, and ANT, the OEC and HAB were administered. The OEC lists 16 potential outcomes for crime, divided into 12 anticipated positive
outcomes (acceptance, approval, control, excitement, freedom, love, power, prestige, purpose,
respect, security, status) and four anticipated negative outcomes (death, jail/prison, loss of
family, loss of job). Respondents are each instructed to rate a crime they have committed
(i.e., the confining offense or some other criminal act) on the basis of the outcomes they
would currently anticipate receivingthat is, if they were living in the community and
committed the crime right now.
A 7-point rating scale is used with the OEC, reflecting degree of belief in which the outcome will occur across similar situations (1 = never, 7 = always). The sum of the ratings
from the 12 positive outcome expectancy items constitutes the OEC-POS score (range =
12-84). Internal consistency for the OEC-POS was strong in the present sample ( = .90),
and previous research indicates that scores on the OEC-POS fall precipitously as a consequence of an inmates involvement in a therapeutic intervention designed to reduce positive
outcome expectancies for crime (Walters, 2003).
The HAB consists of three vignettes similar to situations used in previous studies on hostile attribution biases in children (Dodge, Price, Bachorowski, & Newman, 1990) but which
take place in an adult correctional facility. The first vignette asks respondents to interpret
the intentions of an inmate who bumps into them as he passes the respondent in the commissary line. Using a 5-point Likert-type scale, the respondent signifies the degree to which
he believes the bump was intentional (1 = definitely unintentional, 5 = definitely intentional). The second and third vignettes involve being struck in the back with a basketball
and being reprimanded by a lieutenant (supervisory staff member) for having ones shirt

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1474

CRIMINAL JUSTICE AND BEHAVIOR

untucked (when other inmates with untucked shirts are seemingly ignored). These two
vignettes were also rated using the 5-point Likert-type scale previously described. The total
score for the three vignettes (HAB-TOT) served as an outcome measure in this study
(range = 315), with higher scores indicating greater hostile attributional bias. Internal
consistency in the present sample was modest when using Cronbachs alpha coefficient
( = 54) but moderate when interitem correlations were calculated (r = .27.31). Initial
validation of the HAB shows that although it correlates with a putative measure of reactive
criminal thinking (Co), it fails to correlate with a putative measure of proactive criminal
thinking (En; Walters, 2007b).
Procedure. Percentile-rank converted scores on the PICTS, LSRP, ANT, OEC-POS, and
HAB-TOT were fit to two models: a theory-congruent model and a theory-incongruent
model. The theory-congruent model made Mo, En, Po, So, and ANT-E correlates of OECPOS and Co, Ci, Ds, LSRP-SP, and ANT-S correlates of HAB-TOT. The theory-incongruent
model made Mo, En, Po, So, and ANT-E correlates of HAB-TOT and Co, Ci, Ds, LSRP-SP,
and ANT-S correlates of OEC-POS. The prediction was that the theory-congruent model
would demonstrate significantly better fit than that of the theory-incongruent model. MLE
was used to estimate the coefficients in an SEM recursive path analysis of proactive and reactive predictors and outcome expectancy and hostile attribution bias outcomes, as computed
by Amos 4.0. Covariance curves were drawn between each of the predictor variables given
the intercorrelated nature of proactive and reactive criminal thinking and behavior.
Unstandardized regression coefficients and the standardized residual covariances of the
comparison between the sample and the implied covariance matrices were computed as CRs
in which the regression estimate was divided by the standard error of the estimate (with
scores above 1.96 denoting a significant effect at the .05 level). Four principal fit indices
were also calculated: the model chi-square, the CFI, the RMSEA, and the AIC. Whereas the
CFI and RMSEA were used to assess the absolute fit of the theory-congruent and theoryincongruent models, the AIC was used to assess the relative fit of the two models.
RESULTS

Regression weights. Table 7 lists the unstandardized and standardized regression weights
for the pathways between the predictor variables (five proactive and five reactive) and the two
outcome measures in the theory-congruent model. Table 8 lists the unstandardized and standardized regression weights for pathways between the predictor variables and the two outcome measures in the theory-incongruent model. The results indicate five significant CRs in
the theory-congruent model, four of which were in the predicted direction and one of which
was in the opposite direction: So OEC-POS, ANT-E OEC-POS, Co HAB-TOT, Ds
HAB-TOT (a negative relationship, contrary to predictions), and ANT-S HAB-TOT.
There were no significant pathways in the theory-incongruent model.
Goodness-of-fit indices. Table 9 lists the goodness-of-fit results for the two models. The
chi-square was significant for the theory-incongruent model but not for the theory-congruent
model, denoting better fit for the latter. Both models achieved CFI values in the good-fit
range, but whereas the RMSEA value for the theory-incongruent model indicated modest
fit, the RMSEA value for the theory-congruent model displayed good fit. Moreover, the

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES


TABLE 7:

1475

Regression Weights for the Theory-Congruent Model: Study 5

Regression Path
Mo OEC-POS
En OEC-POS
Po OEC-POS
So OEC-POS
ANT-E OEC-POS
Co HAB-TOT
Ci HAB-TOT
Ds HAB-TOT
LSRP-SP HAB-TOT
ANT-S HAB-TOT

(90% CI)

Estimate

SE

CR

0.015
0.058
0.072
0.162
0.190
0.248
0.012
0.163
0.078
0.171

0.073
0.080
0.074
0.076
0.064
0.085
0.079
0.082
0.074
0.064

0.20
0.73
0.97
2.13*
2.97**
2.92**
0.16
1.98*
0.97
2.67**

.02
.06
.07
.16
.19
.25
.01
.16
.08
.17

(.10, .14)
(.08, .19)
(.05, .21)
(.03, .30)
(.08, .30)
(.13, .39)
(.11, .13)
(.29, .02)
(.04, .19)
(.06, .27)

Note. Estimate = unstandardized regression coefficient; (90% CI) = standardized coefficient and 90th-percentile
biased corrected confidence interval of the standardized coefficient (B = 2,000).
*
p < .05. **p < .01.

TABLE 8:

Regression Weights for the Theory-Incongruent Model: Study 5

Regression Path
Mo HAB-TOT
En HAB-TOT
Po HAB-TOT
So HAB-TOT
ANT-E HAB-TOT
Co OEC-POS
Ci OEC-POS
Ds OEC-POS
LSRP-SP OEC-POS
ANT-S OEC-POS

Estimate

SE

CR

0.126
0.076
0.056
0.015
0.057
0.095
0.106
0.046
0.125
0.114

0.075
0.083
0.077
0.079
0.066
0.086
0.080
0.083
0.072
0.065

1.68
0.92
0.73
0.19
0.86
1.10
1.33
0.55
1.73
1.75

(90% CI)
.13
.08
.06
.02
.06
.09
.11
.05
.12
.11

(.00.25)
(.06.21)
(.07.19)
(.12.15)
(.06.17)
(.05.24)
(.03.23)
(.19.10)
(.00.24)
(.02.22)

Note. Estimate = unstandardized regression coefficient; (90% CI) = standardized coefficient and 90th-percentile
biased corrected confidence interval of the standardized coefficient (B = 2,000).

TABLE 9:

Goodness-of-Fit Statistics: Study 5

Model

2 (N = 291)

df

CFI

RMSEA (90% CI)

SRC

AIC

AIC

wAIC

M2-TC
M2-TI

8.06
38.26***

11
11

1.00
0.98

.000 (.000.047)
.092 (.062.125)

0
3

142.06
172.26

0.00
30.20

.9999997
.0000003

Note. M2 = two-dimensional model in which Mo, En, Po, So, and ANT-E make up the proactive dimension and Co,
Ci, Ds, ANT-S, and LSRP-SP make up the reactive dimension; TC = theory congruent (i.e., proactive scales predict
OEC-POS and reactive scales predict HAB-TOT); TI = theory incongruent (i.e., proactive scales predict HAB-TOT
and reactive scales predict OEC-POS); SRC = number of significant discrepancies in the standardized residual
covariances (p < .05, CR > 1.96); AIC = Akaike Information Criterion; AIC = difference between AIC values
obtained by the TC and TI models; wAIC = Akaike weight.
***
p < .001.

upper limit of the 90% RMSEA CI fell in the poor-fit range for the theory-incongruent
model and in the good-fit range for the theory-congruent model. There were three discrepancies (CR > 1.96, p < .05) in the theory-incongruent model (So OEC-POS, ANT-E
OEC-POS, ANT-S HAB-TOT) but no discrepancies in the theory-congruent model.

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1476

CRIMINAL JUSTICE AND BEHAVIOR

Although the standardized pathway coefficients, chi-square, RMSEA, and standardized


residual covariance results imply that the theory-congruent model fit the data better than
the theory-incongruent model, none of these procedures permit direct model comparison.
To compare the models directly, the difference between the model AIC values must be calculated. The difference between AIC values obtained in the present investigation (AIC = 30.20)
furnishes very strong evidence (AIC > 10) that the theory-congruent model fit the data
better than the theory-incongruent model.
DISCUSSION

The results of Study 5 furnish support for the construct validity of a two-dimensional
model of crime-related cognition and behavior. A path analysis of 10 predictor variables from
the PICTS, LSRP, and ANT, classified as proactive or reactive and regressed onto selfreported positive outcome expectancies for crime and hostile attribution biases, reveal that
theory-congruent pairings (proactive with outcome expectancies and reactive with hostile
attribution biases) achieve significantly better fit than that of theory-incongruent pairings
(proactive with hostile attribution biases and reactive with outcome expectancies). Consistent
with prior research on childhood aggression (Crick & Dodge, 1996), juvenile delinquency
(Smithmyer et al., 2000), and adult criminality (Walters, 2007b), proactive criminal thinking
and behavior are differentially associated with positive outcome expectancies for crime,
whereas reactive criminal thinking and behavior are differentially associated with hostile
attribution biases. These findings furnish preliminary support for the construct validity of the
proactivereactive breakdown proposed by the two-dimensional model.

GENERAL DISCUSSION

The studies described in this article were inspired by research indicating that crimerelated constructs such as psychopathy, antisocial personality, and criminal lifestyle have a
dimensional, rather than taxonic, latent structure (Edens et al., 2006; Marcus et al., 2006;
Walters, 2007a). It was reasoned that if all three constructs are dimensional, then perhaps
they also share one or more dimensions. To test this possibility, a two-dimensional model
with proactive and reactive latent factors was constructed and compared to a one-factor
model in which all three constructs were loaded onto a single latent factor and a social learning model composed of two latent factors: criminal thinking and antisocial behavior. The
results of an SEM analysis of self-report data evaluating all three constructs (psychopathy,
antisocial personality, criminal lifestyle) in a reasonably sized sample of medium-security
male prison inmates and a small cross-validation sample of maximum-security male prison
inmates show modest to adequate fit for the two-dimensional model from an analysis of
regression weights, residual covariances, and goodness-of-fit indices. The analysis also
revealed that the two-dimensional model is superior to the one-factor and social learning
models in direct comparisons using the AIC relative fit measure. The present results consequently indicate that a two-dimensional model comprising proactive and reactive latent factors may underpin crime-related constructs like psychopathy, antisocial personality, and
criminal lifestyle while casting doubt on alternate one-factor and social learning models.

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES

1477

A principal implication of this study is that it provides support for a unified theory of
antisocial behavior, from childhood aggression to adult criminality. Besides the fact that
childhood aggression and adult criminality are dimensional rather than taxonic (Dodge,
Lochman, Harnish, Bates, & Pettit, 1997; Walters, 2007a), there are at least three other
points on which the two constructs converge. First, confirmatory factor analyses indicate
that two dimensionswhat Dodge and colleagues (Crick & Dodge, 1996; Dodge &
Coie, 1987; Dodge et al., 1997) and Walters (2007b) refer to as proactive and reactive
do a better job of accounting for childhood aggression (Poulin & Boivin, 2000) and adult
criminality (Walters, 2007b) than that of a one-factor model and a social learning model
with cognitive and behavioral factors. Second, these two dimensions are highly correlated. The rating scales that have been used to classify children as proactive or reactive
have been found to correlate with each other (.41 to .90), although the majority of correlations cluster between .77 and .83 (Day et al., 1992; Dodge & Coie, 1987; Hubbard et al.,
2002; Price & Dodge, 1989; Poulin & Boivin, 2000). Intercorrelations between the
PICTS proactive and reactive composite scales, however, range from .60 to .72 (Walters,
2006; Walters & Mandell, 2007). Some might argue that this level of intercorrelation
makes the dimensions redundant. However, in the present study, as well as in previous
studies on aggressive children (Crick & Dodge, 1996), juvenile delinquents (Smithmyer
et al., 2000), and adult prisoners (Walters, 2007b), the two dimensions have correlated
differentially with positive outcome expectancies for aggression/crime and hostile attribution biases in ways that are consistent with theory and congruent with the construct
validity of the individual measures.
Card and Little (2006) assert that correlations between proactive and reactive aggression
are largely a consequence of the restricted range of the methods used (i.e., rating scales in
childhood aggression research and self-report measures in the present study). Whereas there
is a strong likelihood that shared method variance is partially responsible for the height of
the correlation between proactive and reactive aggression/criminality, I would contend that
it does not fully account for this correlation. Grounded in similar neurobiological, psychological, and sociological processes, proactive and reactive aggression/criminality are structurally, functionally, and developmentally related. The two-dimensional model being
advocated in this article can be considered one level of a larger theory in which aggression
and criminality are hierarchically organized, with a general tendency to aggress against others and violate the rules of society at the top of the hierarchy. Below this would be the
proactive and reactive functions of aggression/criminality and, below that, the specific attributions, expectancies, goals, values, and thinking styles that facilitate aggression and criminality. Results from the current investigation indicate that the general tendency toward
aggression/criminality does not adequately explain the overlap among psychopathy, antisocial personality, and criminal lifestyle and that we need to look to lower levels in the hierarchy (proactive, reactive) to gain a better understanding of the relationships between these
three crime-related constructs.
The current results also have implications for clinical practice. Correctional assessment
and classification should consider both proactive and reactive criminality when evaluating
inmates. Because reactive criminality leads to more overt and obvious forms of acting-out
behavior, it is more likely to be disruptive to the orderly running of a correctional institution.
Yet whereas reactive criminality did a better job of predicting the total number of disciplinary

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1478

CRIMINAL JUSTICE AND BEHAVIOR

reports received by prisoners in one study (Walters, 2005b), proactive criminality did a
better job of predicting prison-based aggressive behavior in another study (Walters &
Mandell, 2007). Consequently, when evaluating an inmate for programming or classification purposes, both forms of criminality should be taken into account. In classifying
inmates using the proactivereactive model, it will be important to avoid using a categorical scheme in which inmates are classified as proactive or reactive, because research indicates that proactive and reactive criminality, like proactive and reactive childhood
aggression, are highly correlated dimensions. Accordingly, most individuals will be high or
low on both dimensions. Another practical implication of the present findings is that proactive and reactive criminality may require different forms of intervention. Whereas reactive
criminality appears to respond to behaviorally oriented skill development techniques and
programs (Walters, 2008), proactive criminality will probably require more cognitively oriented interventions in which outcome expectancies for crime and criminal goals are targeted. Many correctional programs address skill deficits by providing inmates with anger
and stress management training (Young, Dembo, & Henderson, 2007), but few programs
address the criminogenic features of proactive criminal thinking.
Although the proactivereactive breakdown that defines the two-dimensional model may
seem to parallel the Factor 1Factor 2 structural breakdown of the PCL-R, the similarities
are more apparent than real. Both components of the two-dimensional model share more in
common with Factor 2, the behavioral or chronic antisocial lifestyle component of psychopathy, than with Factor 1, the personality or callous and remorseless use-of-others component of psychopathy. Moreover, the one original indicator from the present investigation
that seems to share the most in common with Factor 1 of the PCL-R (the LSRP-PP) failed
to load sufficiently onto the general antisocial behavior factor of the social learning model
and demonstrate unambiguous allegiance to either the proactive or reactive factors of the
two-dimensional model to justify including it in this series of five studies. In fact, proactive
and reactive scales from both the PICTS (Walters, 2002; Walters & Mandell, 2007) and the
ANT (Salekin, Rogers, Ustad, & Sewell, 1998; Walters, Duncan, & Geyer, 2003) consistently correlate better with Factor 2 of the PCL-R/PCL:SV than Factor 1 of the PCL-R/
PCL:SV. Therefore, despite superficial similarities between the proactive and reactive factors of the two-dimensional model and Factors 1 and 2 of the PCL-R, the two models
should not be confused.
A potential limitation of the current set of studies is that they were conducted on male
inmates from medium- and maximum-security federal prisons who were administered a
series of self-report questionnaires. A reasonable question at this juncture is whether these
results generalize to female offenders and to both male and female nonoffenders. The issue
of whether indicators from the PICTS, LSRP, and ANT apply to nonoffenders is neither
moot nor trivial. All three measures have been used in nonoffender samples, including the
PICTS (see Walters & McCoy, 2007). The LSRP, in fact, was developed for the express
purpose of creating a self-report measure that could be used with nonincarcerated participants (Levenson et al., 1995). If it can be shown that the present findings generalize to
nonoffender populations, then perhaps it would indicate that the proactivereactive differentiation has value in describing behaviors and motivations beyond childhood aggression
and adult criminality. It will also be important to know whether the current findings generalize to non-self-report measures of psychopathy, antisocial personality, and criminal

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES

1479

lifestyle. Conducting a confirmatory factor analysis of individuals administered the PCL-R


as a measure of psychopathy, the Structured Clinical Interview for DSM-IV Axis II
Personality Disorders (First, Gibbon, Spitzer, Williams, & Benjamin, 1997) as a measure
of antisocial personality, and the Lifestyle Criminality Screening Form (Walters, White, &
Denney, 1991) as a measure of criminal lifestyle may shed additional light on the validity
of the proactivereactive model, although these measures may be incapable of generating
a sufficient number of proactive indicators to conduct the analyses. In short, further
research is required to determine the generalizability of the present findings.
Diminished power is a limitation shared by Studies 4 and 5. Study 4 had a sample size
of 118, and Study 5 had to be computed with 11 degrees of freedom, because covariance
curves were drawn between each predictor to account for the high degree of intercorrelation between predictors in the two-dimensional model. A study with 291 participants and
11 degrees of freedom has modest power to accept close-fitting models ( 0.41) and reject
not-close-fitting models ( 0.30; MacCallum et al., 1996). However, the theory-congruent
and theory-incongruent models had equivalent power, and the theory-congruent model
achieved significantly better fit than that of the theory-incongruent model in direct comparisons between the two models. Accounting for all the covariances between the predictor
variables may have artificially elevated certain fit measures such as the CFI, but once again,
the theory-congruent and theory-incongruent models shared this same advantage, and the
theory-congruent model was clearly the better-fitting model.
Study 5 was also limited by the fact that the outcome measures were exclusively selfreport. Retrospective behavioral measures, such as prior arrests for proactive crimes (robbery, burglary) and reactive crimes (assault, domestic violence), were unavailable for a
large portion of the participants in this study, and prospective behavioral measures, such as
proactive and reactive disciplinary reports, have proved unreliable and thus require more
information about the offense than what is normally available.
The development, appraisal, and refinement of theoretical models form the essence of
science. A two-dimensional model of criminality believed to underlie such popular crimerelated concepts as psychopathy, antisocial personality, and criminal lifestyle received
moderate support in the present investigation. Alternate models and explanationssuch as
a general factor model and a social learning model composed of cognitive and behavioral
factorswere evaluated and found to be lacking. Furthermore, results obtained with male
inmates of a medium-security federal prison were cross-validated on a group of male
inmates of a maximum-security federal prison, with modest shrinkage in model fit. The
theoretical value of the two-dimensional model in clarifying the connection between
childhood aggression and adult criminality, however, requires longitudinal research. Such
studies, if they are not already being done, will take decades to complete. In the meantime,
researchers can contribute to the development and evaluation of theoretical paradigms
such as the two-dimensional model by expanding the focus to other offender and nonoffender populations, by using validated behavioral outcome measures, and by creating
alternative explanatory models against which the proactivereactive model can be evaluated. There is no reason why psychopathy, antisocial personality, and criminal lifestyle
cannot individually contribute to our understanding of criminal behavior now that it has
been shown that proactive and reactive criminality may account for a certain portion of the
variance shared by each.

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1480

CRIMINAL JUSTICE AND BEHAVIOR

Appendix: Abbreviations
AIC: Akaike information criterion
ANT: Antisocial Features Scale
ANT-A: Antisocial Features ScaleAntisocial Behaviors subscale
ANT-E: Antisocial Features ScaleEgocentricity subscale
ANT-S: Antisocial Features ScaleStimulus Seeking subscale
CFI: comparative fit index
CI: confidence intervals
CR: critical ratio
HAB: Hostile Attribution Bias
HAB-TOT: Hostile Attribution Biastotal score from three vignettes
LSRP: Levenson Self-Report Psychopathy Scale
LSRP-PP: Levenson Self-Report Psychopathy ScalePrimary Psychopathy
LSRP-SP: Levenson Self-Report Psychopathy ScaleSecondary Psychopathy
MLE: maximum likelihood estimation
OEC: Outcome Expectancies for Crime Inventory
OEC-POS: Outcome Expectancies for Crime Inventorypositive outcome expectancy items
PAI: Personality Assessment Inventory
PCL-R: Psychopathy ChecklistRevised
PCL:SV: Psychopathy Checklist: Screening Version
PICTS: Psychological Inventory of Criminal Thinking Styles (eight scales)
Mo: Mollification Scale
Co: Cutoff Scale
En: Entitlement Scale
Po: Power Orientation Scale
Sn: Sentimentality Scale
So: Superoptimism Scale
Ci: Cognitive Indolence Scale
Ds: Discontinuity Scale
RMSEA: root mean square error of approximation
NOTE
1. Precision of alpha is calculated as the standard error of item intercorrelations (Cortina, 1993), with higher standard
errors suggesting a greater likelihood of multidimensionality. The precision of alpha values for the rationally derived
Psychological Inventory of Criminal Thinking Styles scales are comparable to the precision of alpha values obtained for the
factorially derived Psychopathy ChecklistRevised factor (.02) and facet scores (.02.05, N = 409; Walters, Duncan, &
Mitchell-Perez, 2007).

REFERENCES
Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52, 317-332.
Akers, R. L., & Jensen, G. F. (2006). Empirical status of social learning theory of crime and deviance: The past, present, and
future. In F. T. Cullen, J. P. Wright, & K. R. Blevins (Eds.), Taking stock: The status of criminological theory (Vol. 15,
pp. 37-76). New Brunswick, NJ: Transaction Books.
Andrews, D. A., & Bonta, J. (1998). The psychology of criminal conduct (2nd ed.). Cincinnati, OH: Anderson.
Arbuckle, J. L., & Wothke, W. (1999). AMOS 4.0 users guide. Chicago: SPSS/SmallWaters.
Bandura, A. (1973). Aggression: A social learning analysis. Englewood Cliffs, NJ: Prentice Hall.
Bandura, A. (1986). Social foundations of thought and action. Englewood Cliffs, NJ: Prentice Hall.

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES

1481

Bartol, C. R., & Bartol, A. M. (2004). Introduction to forensic psychology. Thousand Oaks, CA: Sage.
Bentler, P. M. (1990). Comparative fit indexes in structural models. Psychological Bulletin, 107, 238-246.
Bentler, P. M., & Chou, C. P. (1987). Practical issues in structural modeling. Sociological Methods and Research, 16, 78-117.
Berkowitz, L. (1993). Aggression: Its causes, consequences, and control. Philadelphia: Temple University Press.
Blair, R. J. R. (2004). The roles of orbital frontal cortex in the modulation of antisocial behavior. Brain and Cognition,
55, 198-208.
Brinkley, C. A., Schmitt, W. A., Smith, S. S., & Newman, J. P. (2001). Construct validation of a self-report psychopathy scale:
Does Levensons Self-Report Psychopathy Scale measure the same construct as Hares Psychopathy ChecklistRevised?
Personality and Individual Differences, 31, 1021-1038.
Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. In K. A. Bollen & J. S. Long (Eds.), Testing
structural equation models (pp. 136-162). Newbury Park, CA: Sage.
Burnham, K. P., & Anderson, D. R. (2002). Model selection and multi-model inference: A practicalinformationtheoretic
approach (2nd ed.). New York: Springer-Verlag.
Bushman, B. J., & Anderson, C. A. (2001). Is it time to pull the plug on the hostile versus instrumental aggression dichotomy?
Psychological Review, 108, 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.
Cleckley, H. (1976). The mask of sanity (5th ed.). St. Louis: Mosby.
Cortina, J. M. (1993). What is coefficient alpha? An examination of theory and applications. Journal of Applied Psychology,
78, 98-104.
Crick, N. R., & Dodge, K. A. (1996). Social information-processing mechanisms in reactive and proactive aggression. Child
Development, 67, 993-1002.
Day, D. M., Bream, L. A., & Pal, A. (1992). Proactive and reactive aggression: An analysis of subtypes based on teacher
perceptions. Journal of Clinical Child Psychology, 21, 210-217.
Diamond, P. M., & Magaletta, P. R. (2006). The Short-Form Buss-Perry Aggression Questionnaire (BPAQ-SF): A validation
study with federal offenders. Assessment, 13, 227-240.
Dodge, K. A., & Coie, J. D. (1987). Social-information processing factors in reactive and proactive aggression in childrens
peer groups. Journal of Personality and Social Psychology, 53, 1146-1158.
Dodge, K. A., Lochman, J. E., Harnish, J. D., Bates, J. E., & Pettit, G. S. (1997). Reactive and proactive aggression in school
children and psychiatrically impaired chronically assaultive youth. Journal of Abnormal Psychology, 106, 37-51.
Dodge, K. A., Price, J. M., Bachorowski, J. A., & Newman, J. P. (1990). Hostile attributional biases in severely aggressive
adolescents. Journal of Abnormal Psychology, 99, 385-392.
Edens, J. F., Marcus, D. K., Lilienfeld, S. O., & Poythress, N. G. (2006). Psychopathic, not psychopath: Taxometric evidence
for the dimensional structure of psychopathy. Journal of Abnormal Psychology, 115, 131-144.
Edens, J. F., & Ruiz, M. A. (2005). PAI Interpretive Report for Correctional Settings (PAI-CS) professional manual. Lutz,
FL: Psychological Assessment Resources.
Egan, V., McMurran, M., Richardson, C., & Blair, M. (2000). Criminal cognitions and personality: What does the PICTS
really measure? Criminal Behaviour and Mental Health, 10, 170-184.
Epstein, M. K., Poythress, N. G., & Brandon, K. O. (2006). The Self-Report Psychopathy Scale and passive avoidance learning:
A validation study of race and gender effects. Assessment, 13, 197-207.
First, M. B., Gibbon, M., Spitzer, R. L., Williams, J. B., & Benjamin, L. (1997). Structured Clinical Interview for DSM-IV
Axis II personality disorders (SCID-II). Washington, DC: American Psychiatric Press.
Goldstein, A. M. (Ed.). (2007). Forensic psychology: Emerging topics and expanding roles. New York: John Wiley.
Guay, J.-P., Ruscio, J., Knight, R. A., & Hare, R. D. (2007). A taxometric analysis of the latent structure of psychopathy:
Evidence for dimensionality. Journal of Abnormal Psychology, 116, 701-716.
Hare, R. D. (1996). Psychopathy: A construct whose time has come. Criminal Justice and Behavior, 23, 25-54.
Hare, R. D. (2003). The Hare Psychopathy ChecklistRevised manual (2nd ed.). Toronto, Ontario, Canada: Multi-Health
Systems.
Harris, G. T., Rice, M. E., & Quinsey, V. L. (1994). Psychopathy as a taxon: Evidence that psychopaths are a discrete class.
Journal of Consulting and Clinical Psychology, 62, 387-397.
Hubbard, J. A., Smithmyer, C. M., Ramsden, S. R., Parker, E. H., Flanagan, K. D., Dearing, K. F., et al. (2002). Observational,
physiological, and self-report measures of childrens anger: Relations to reactive versus proactive aggression. Child
Development, 73, 1101-1118.
Klein, R. B. (2005). Principles and practice of structural equation modeling (2nd ed.). New York: Guilford.
Levene, H. (1960). Robust tests for the equality of variance. In I. Olkin (Ed.), Contributions to probability and statistics:
Essays in honor of Harold Hotelling (pp. 278-292). Palo Alto, CA: Stanford University Press.
Levenson, M. R., Kiehl, K. A., & Fitzpatrick, C. M. (1995). Assessing psychopathic attributes in a noninstitutionalized population. Journal of Personality and Social Psychology, 68, 151-158.

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

1482

CRIMINAL JUSTICE AND BEHAVIOR

Lilienfeld, S. O., & Andrews, B. P. (1996). Development and preliminary validation of a self-report measure of psychopathic
personality traits in noncriminal populations. Journal of Personality Assessment, 66, 488-524.
Lilienfeld, S. O., & Fowler, K. A. (2006). The self-report assessment of psychopathy: Problems, pitfalls, and promises. In
C. J. Patrick (Ed.), Handbook of psychopathy (pp. 107-132), New York: Guilford.
Lilienfeld, S. O., Skeem, J. L., & Poythress, N. G. (2004, March). Psychometric properties of self-report psychopathy measures. Paper presented at the annual meeting of the American Psychology and Law Society, Scottsdale, AZ.
MacCallum, R. C., Browne, M. W., & Sugawara, H. M. (1996). Power analysis and determination of sample size for covariance structure modeling. Psychological Methods, 1, 130-149.
Marcus, D. K., John, S. L., & Edens, J. F. (2004). A taxometric analysis of psychopathic personality. Journal of Abnormal
Psychology, 113, 626-635.
Marcus, D. K., Lilienfeld, S. O., Edens, J. F., & Poythress, N. G. (2006). Is antisocial personality disorder continuous or
categorical? A taxometric analysis. Psychological Medicine, 36, 1571-1582.
Mardia, K. V. (1974). Applications of some measures of multivariate skewness and kurtosis in testing normality and robustness
studies. Sankhya, 36B, 115-128.
McHoskey, J. W., Worzel, W., & Szyarto, C. (1998). Machiavellianism and psychopathy. Journal of Personality and Social
Psychology, 74, 192-210.
Meehl, P. E. (1995). Bootstraps taxometrics: Solving the classification problem in psychopathology. American Psychologist,
50, 266-275.
Morey, L. C. (2007). The Personality Assessment Inventory (PAI): Professional manual (2nd ed.). Lutz, FL: Psychological
Assessment Resources.
Ormrod, J. E. (1999). Human learning (3rd ed.). Upper Saddle River, NJ: Prentice Hall.
Poulin, F., & Boivin, M. (2000). Reactive and proactive aggression: Evidence of a two-factor model. Psychological
Assessment, 12, 115-122.
Price, J. M., & Dodge, K. A. (1989). Reactive and proactive aggression in childhood: Relations to peer status and social context dimensions. Journal of Abnormal Psychology, 17, 455-471.
Pryce, G. (2005). Inference and statistics in SPSS. Glasgow, UK: GeeBeyJey.
Robins, L. N. (1966). Deviant children grown up. Baltimore: Williams & Wilkins.
Ruscio, J., Haslam, N., & Ruscio, A. M. (2006). Introduction to the taxometric method: A practical guide. Mahwah,
NJ: Lawrence Erlbaum.
Salekin, R. T., Rogers, R., Ustad, K. L., & Sewell, K. W. (1998). Psychopathy and recidivism among female inmates. Law
and Human Behavior, 22, 109-128.
Smithmyer, C. M., Hubbard, J. A., & Simons, R. F. (2000). Proactive and reactive aggression in delinquent adolescents:
Relations to aggression outcome expectancies. Journal of Clinical Child Psychology, 29, 86-93.
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. (2002). The Psychological Inventory of Criminal Thinking Styles (PICTS): A review and meta-analysis.
Assessment, 9, 278-291.
Walters, G. D. (2003). Changes in outcome expectancies and criminal thinking following a brief course of psychoeducation.
Personality and Individual Differences, 35, 691-701.
Walters, G. D. (2004). The trouble with psychopathy as a general theory of crime. International Journal of Offender Therapy
and Comparative Criminology, 48, 133-148.
Walters, G. D. (2005a). How many factors are there on the PICTS? Criminal Behaviour and Mental Health, 15, 273-283.
Walters, G. D. (2005b). Predicting institutional adjustment with the Lifestyle Criminality Screening Form and Psychological
Inventory of Criminal Thinking Styles. International Journal of Forensic Mental Health, 4, 63-70.
Walters, G. D. (2006). The Psychological Inventory of Criminal Thinking Styles (PICTS) professional manual. Allentown,
PA: Center for Lifestyle Studies.
Walters, G. D. (2007a). The latent structure of the criminal lifestyle: A taxometric analysis of the Lifestyle Criminality
Screening Form and Psychological Inventory of Criminal Thinking Styles. Criminal Justice and Behavior, 34, 1623-1637.
Walters, G. D. (2007b). Measuring proactive and reactive criminal thinking with the PICTS: Correlations with outcome
expectancies and hostile attribution biases. Journal of Interpersonal Violence, 22, 371-385.
Walters, G. D. (2008). Anger management training in incarcerated male offenders: Differential impact on proactive and reactive criminal thinking. Manuscript submitted for publication.
Walters, G. D., Brinkley, C. A., Magaletta, P. R., & Diamond, P. M. (in press). Taxometric analysis of the Levenson SelfReport Psychopathy scale. Journal of Personality Assessment.
Walters, G. D., Diamond, P. M., Magaletta, P. R., Geyer, M. D., & Duncan, S. A. (2007). Taxometric analysis of the Antisocial
Features Scale of the Personality Assessment Inventory in federal prison inmates. Assessment, 14, 351-360.
Walters, G. D., Duncan, S. A., & Geyer, M. D. (2003). Predicting disciplinary adjustment in inmates undergoing forensic
evaluation: A direct comparison of the PCL-R and PAI. Journal of Forensic Psychiatry and Psychology, 14, 382-393.

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Walters / FACTOR AND PATH ANALYSIS OF SELF-REPORT MEASURES

1483

Walters, G. D., Duncan, S. A., & Mitchell-Perez, K. (2007). The latent structure of psychopathy: A taxometric investigation
of the Psychopathy ChecklistRevised in a heterogeneous sample of male prison inmates. Assessment, 14, 270-278.
Walters, G. D., Gray, N. S., Jackson, R. L., Sewell, K. W., Rogers, R., Taylor, J., et al. (2007). A taxometric analysis of the
Psychopathy Checklist: Screening Version (PCL:SV): Further evidence of dimensionality. Psychological Assessment, 19,
330-339.
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., & McCoy, K. (2007). Taxometric analysis of the Psychological Inventory of Criminal Thinking Styles in
incarcerated offenders and college students. Criminal Justice and Behavior, 34, 781-793.
Walters, G. D., White, T. W., & Denney, D. (1991). The Lifestyle Criminality Screening Form: Preliminary data. Criminal
Justice and Behavior, 18, 406-418.
Weiner, I. B., & Hess, A. K. (Eds.). (2006). The handbook of forensic psychology (3rd ed.). New York: John Wiley.
Wilson, D. L., Frick, P. J., & Clements, C. B. (1999). Gender, socialization, and psychopathic traits in a college sample.
Journal of Psychopathology and Behavioral Assessment, 21, 221-235.
Young, D. W., Dembo, R., & Henderson, C. E. (2007). A national survey of substance abuse treatment for juvenile offenders.
Journal of Substance Abuse Treatment, 32, 255-266.
Zimmerman, D. W., & Zumbo, B. D. (2005). Can percentiles replace raw scores in the statistical analysis of test data?
Educational and Psychological Measurement, 65, 616-638.
Zwick, R. (1986). Rank and normal scores alternatives to Hotellings T2. Multivariate Behavioral Research, 21, 169-186.

Glenn D. Walters, PhD, currently serves as drug program coordinator at the Federal Correctional Institution, Schuylkill,
Pennsylvania. His research interests fall into three areas: the genetic correlates of crime, substance abuse, and problem gambling; psychological assessment of offenders, with an emphasis on criminal thinking and psychopathy; and development of
an overarching theory of criminal behavior. He has published more than 180 articles and book chapters and is the author of
14 books, including The Criminal Lifestyle (1990), Criminal Belief Systems (2002), and Lifestyle Theory: Past, Present, and
Future (2006).

Downloaded from http://cjb.sagepub.com at Biblioteca di Ateneo on February 2, 2009

Vous aimerez peut-être aussi