Kingston

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

Psychology, Crime & Law

ISSN: 1068-316X (Print) 1477-2744 (Online) Journal homepage: http://www.tandfonline.com/loi/gpcl20

The relationship between mental illness and


violence in a mentally disordered offender sample:
evaluating criminogenic and psychopathological
predictors

Drew A. Kingston, Mark E. Olver, Melissa Harris, Brad D. Booth, Sanjiv Gulati
& Colin Cameron

To cite this article: Drew A. Kingston, Mark E. Olver, Melissa Harris, Brad D. Booth, Sanjiv Gulati
& Colin Cameron (2016): The relationship between mental illness and violence in a mentally
disordered offender sample: evaluating criminogenic and psychopathological predictors,
Psychology, Crime & Law, DOI: 10.1080/1068316X.2016.1174862

To link to this article: http://dx.doi.org/10.1080/1068316X.2016.1174862

Accepted author version posted online: 06


Apr 2016.
Published online: 19 Apr 2016.

Submit your article to this journal

Article views: 47

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=gpcl20

Download by: [Temple University Libraries] Date: 13 June 2016, At: 14:28
PSYCHOLOGY, CRIME & LAW, 2016
http://dx.doi.org/10.1080/1068316X.2016.1174862

The relationship between mental illness and violence in a


mentally disordered offender sample: evaluating criminogenic
and psychopathological predictors
Drew A. Kingstona,b, Mark E. Olverc, Melissa Harrisa, Brad D. Bootha, Sanjiv Gulatia and
Colin Camerona
a
Integrated Forensic Program, Royal Ottawa Health Care Group, Brockville Mental Health Center, Brockville,
Ontario, Canada; bInstitute of Mental Health Research, Ottawa, Ontario, Canada; cDepartment of Psychology,
University of Saskatchewan, Saskatoon, Saskatchewan, Canada
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

ABSTRACT ARTICLE HISTORY


The importance of mental illness as a risk factor for violence has Received 19 October 2015
been debated with significant implications for mental health Accepted 28 March 2016
policy and clinical practice. In offender samples, psychopathology
KEYWORDS
tends to be unrelated to recidivism, although some researchers Offender; mental health;
have noted that this relationship may be dependent upon certain diagnoses; recidivism
moderating factors. In the present, prospective investigation,
psychopathology is examined as predictors of recidivism in 121
provincially sentenced (i.e. less than 2 years) mentally disordered
offenders. Results indicated that psychopathological predictors
were generally poor predictors of recidivism in univariate and
multivariate analyses. Consistent with our hypotheses, age of
onset of criminal activity was a significant moderating factor on
the relationship between mental illness and recidivism, although
results were not in the expected direction for certain classes of
mental illness. Results are discussed in the context of a social
learning model of crime and in terms of the treatment of mentally
disordered offenders.

There has been a long-standing belief by the public that individuals with a mental illness
are prone to aggressive behavior and are generally more dangerous than individuals
without a mental illness. This belief is underscored by the fact that mental illness is signifi-
cantly overrepresented in the criminal justice system (Corrado, Cohen, Hart, & Roesch,
2000; Diamond, Wang, Holzer, Thomas, & Cruser, 2001; Fazel & Danesh, 2002). Epidemio-
logical studies have estimated the prevalence of mental illness in correctional settings to
be between 10% and 40%; however, when the definition includes substance use disorders
(SUDs) and some personality disorders, estimates increase to between 80% and 90%.
Moreover, the rates of incarceration of individuals with a mental illness are increasing,
which has undoubtedly contributed to a growing interest in identifying the determinants
of violence in this population. In this paper, we examined several putative risk factors
derived from two competing theoretical explanations used with mentally disordered
offenders.

CONTACT Drew A. Kingston drew.kingston@theroyal.ca Integrated Forensic Program, Royal Ottawa Health Care
Group, Brockville Mental Health Center, 1804 Highway 2, Brockville, Ontario, Canada K6V 5W7
© 2016 Informa UK Limited, trading as Taylor & Francis Group
2 D. A. KINGSTON ET AL.

Relationship between mental illness and violence


There has been an ongoing debate regarding the extent to which mental illness is a risk
factor for violence (Markowitz, 2011; Monahan, 1981). On the one hand, many research-
ers subscribe to the psychopathological theory of criminal behavior, whereby untreated
mental illness, particularly schizophrenia and other psychotic disorders, is considered a
direct cause of criminal behavior. A number of studies and quantitative reviews have
supported this perspective. For example, Brennan, Mednick, and Hodgins (2000) exam-
ined psychiatric hospitalizations and official arrests for violence in a large Danish cohort
of 358,180 people. Psychiatric diagnoses included schizophrenia, organic brain syn-
drome (e.g. dementia, psychosis associated with cerebral infection), affective psychosis,
which included bipolar disorder, and other psychoses (e.g. unspecified). Results indi-
cated that individuals with a psychiatric diagnosis were more likely to be arrested for
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

a violent offense as compared to those who had never been hospitalized. Among
men, the highest risk of violence was observed among those with an organic brain syn-
drome, followed by those diagnosed with schizophrenia. Among women, the highest
risk of arrest was evident among those diagnosed with schizophrenia. Schizophrenia
was the only diagnosis associated with violence in both men and women, after control-
ling for concurrent disorders, including substance use and personality disorders. Two
more recent meta-analyses similarly reported that psychosis was associated with an
increased risk of violence and was particularly salient when the psychosis was
comorbid with a SUD (Douglas, Guy, & Hart, 2009; Fazel, Gulati, Linsell, Geddes, &
Grann, 2009).
In contrast to the studies cited earlier, there are a number of studies which have failed
to identify a significant link between psychosis or other forms of mental illness and vio-
lence (Appelbaum, Robbins, & Monahan, 2000; Elbogen & Johnson, 2009; Witt, van
Dorn, & Fazel, 2013). Some researchers have argued that methodological diversity
across studies accounts for much of the discrepant findings reported in the literature.
One possible confounding factor is sample composition, such that mental illness is a
risk factor for violence among the general population but this effect does not generalize
to offenders. In their meta-analysis on the relationship between psychosis and violence,
Douglas et al. (2009) showed that the association between psychosis and violence was
pronounced among community samples (OR = 3.46) but was no longer meaningful
when restricted to forensic psychiatric (OR = 0.91) or correctional samples (OR = 1.27).
The vast majority of studies using offender samples have shown psychiatric diagnoses
to be unrelated to recidivism and that the predictors of recidivism are largely shared
between mentally disordered offenders and non-disordered offenders. Consequently,
others have adopted social psychological explanations of crime developed with non-
mentally disordered offenders. Andrews and Bonta (1994, 2010), for example, presented
a social learning theory of criminal behavior, called the General Personality and Cognitive
Social Learning (GPCSL) model. According to the GPCSL, there are eight robust predictors
of criminal behavior that reside within the individual or their immediate social learning
environment: criminal history, pro-criminal companions, pro-criminal attitudes, antisocial
personality pattern, education/employment, family/marital, substance abuse, and
leisure/recreation. Mental health variables were not considered significant predictors of
criminal behavior.
PSYCHOLOGY, CRIME & LAW 3

In a meta-analysis of 58 studies of mentally disordered and non-disordered offenders,


psychiatric diagnosis was not associated with recidivism and psychosis was inversely
related to recidivism (Bonta, Law, & Hanson, 1998). In an updated meta-analysis, Bonta,
Blais, and Wilson (2013) reported that the central eight risk factors were important predic-
tors of general and violent recidivism among mentally disordered offenders, whereas clini-
cal variables, such as schizophrenia were not significant predictors. Since the publication
of this latest quantitative review, several studies have been conducted that continue to
support the notion that mental illness fails to reliably predict recidivism in offender
samples and that the best predictors are consistent with the GPCSL model (Kingston,
Olver, Harris, Wong, & Bradford, 2015; Rezansoff, Moniruzzaman, Gress, & Somers, 2013;
Skeem, Winter, Kennealy, Louden, & Tatar, 2014).
Despite the fact that mental illness tends to be a weak and non-significant predictor
of recidivism in offender samples, there are clearly those incidents in which individuals
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

commit crimes that are attributable to their mental health symptoms. A classic example
is an assault that is primarily motivated by delusions or hallucinations. Skeem,
Manchak, and Peterson (2011) presented a conceptual framework that differentiated
between offenders whose crimes are a direct result of mental illness and those
whose offending behavior is mediated by general risk factors, such as those presented
in social learning models of criminal behavior. Several studies have shown that mental
illness is a direct predictor of criminal behavior in a relatively small proportion of
offences and that the proportion is even smaller among internalizing disorders
(e.g. depression) as compared to disorders with externalizing features, such as
bipolar disorder (Peterson, Skeem, Hart, Vidal, & Keith, 2010; Peterson, Skeem,
Kennealy, Bray, & Zvonkovic, 2014). Skeem et al. (2011) suggested that the extent to
which the relationship of mental illness to violence is mediated or direct depends on
any number of moderating factors. Recently, Walters and Crawford (2014) examined
a large (n = 2627) offender sample and found that history of violence was a significant
moderator of the relationship between mental illness and violence, manifested in the
institution and upon release to the community.
Age of onset of criminal behavior was identified as another potentially important mod-
erating factor. Age of onset and persistence of offending have been identified as typolo-
gical features of offenders with mental illness. Hodgins (2008) presented two primary
types of offenders with mental illness.1 Early start offenders display an early pattern of anti-
social and criminal behavior that commences prior to the onset of mental illness, whereas
late start offenders present with no history of criminal behavior prior to the onset of the
mental illness; however, after the onset of the illness, they have repeated episodes of crim-
inal behavior, including aggression.
Silver (2006) noted that early start offenders have more exposure to criminogenic risk
factors and are more deeply embedded in criminogenic environments than late start
offenders. Indeed, as compared to late start offenders, early start offenders have more
diverse criminal activity, have been convicted for more nonviolent and violent crimes,
and have more problems with substance abuse, particularly during adolescence
(Mathieu & Côté, 2009; Tengström, Hodgins, & Kullgren, 2001). Early start offenders
show an increased prevalence of antisocial personality disorder diagnoses (Goethals,
Willigenburg, Buitelaar, & van Marle, 2008; Mathieu & Côté, 2009; Tengström et al., 2001),
psychopathic traits (Tengström et al., 2001), and history of prison infractions (Vitelli, 1997).
4 D. A. KINGSTON ET AL.

Early start offenders are more likely to have criminally involved parents, have parents
who are substance abusers, and are more likely to have been placed outside of their
parent’s home and into foster or group home care (Alltucker, Bullis, Close, & Yovanoff,
2006; Jones, Van den Bree, Ferriter, & Taylor, 2010; Laajasalo & Häkkänen, 2005; Mathieu &
Côté, 2009; Tengström et al., 2001). Although both early start and late start offenders rep-
resent mentally disordered offenders, they are likely to follow different trajectories to end
up in the criminal justice system, with different risk factors and different intervention
needs.

Present study
In this prospective study, we compared the predictive accuracies of GPCSL and psycho-
pathological variables in the prediction of institutional misconduct and criminal recidivism
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

in a sample of provincially sentenced mentally disordered offenders. We hypothesized


that factors identified by the GPCSL would predict institutional misconduct and criminal
recidivism and that psychopathological predictors would not show substantive predictive
utility, particularly after controlling for GPCSL variables. We also examined the proposal by
Skeem et al. (2011) that age of onset would moderate the relationship between mental
illness and violence and, as such, would differentially impact the importance of mental
illness as a risk factor for criminal recidivism. We predicted that mental illness would be
a significant risk factor for late start offenders and that the relationship would be negligible
for early start offenders.

Method
Participants
Participants were a sample of 121 offenders admitted to the Secure Treatment Unit (STU),
Royal Ottawa Health Care Group, an inter-ministerial treatment facility for mentally ill
offenders serving provincial (i.e. less than 2 years) jail sentences. Full ethical approval
was obtained from the Royal Ottawa Health Care Group and the Ministry of Community
Safety and Correctional Services. Offenders are referred to the STU from their home insti-
tution based on suspected mental illness. At the STU, offenders primarily interact with
mental health professionals who target mental health and criminogenic needs relevant
to the individual. Treatment may consist of pharmacotherapy, psychoeducational
groups, and group psychotherapy.
Offenders served an aggregate average sentence of 348.6 days (SD = 166.4 days)
and were admitted to the STU between March 2011 and June 2013. The average
length of stay was 113.3 days (SD = 72.3 days). With regard to their index offense,
22% of the sample (n = 27) committed a sexual offense, whereas 42% of the sample
(n = 51) committed a non-sexual, violent offense. The remainder of the sample com-
mitted nonviolent offenses, such as break and enter and driving under the influence.
Participants ranged in age from 20 to 66 years, with an average age of 34.4 years
(SD = 11.7 years). The majority of the sample was Caucasian (n = 87, 71.9%), followed
by Aboriginal (n = 16, 13.2%), Black (n = 12, 9.9%), other (n = 4, 3.3%), and unknown
(n = 2, 1.7%).
PSYCHOLOGY, CRIME & LAW 5

Materials
Level of Service/Case Management Inventory
The Level of Service/Case Management Inventory (LS/CMI; Andrews, Bonta, & Wormith,
2004) is an actuarial risk-need assessment tool that was designed to appraise recidivism
risk, identify criminogenic needs, and inform recommendations for treatment and case
management. The LS/CMI is part of a broader family of tools that fall under the rubric
of the Level of Service (LS) scales, with a large number of variations in existence.
Wormith (2011) noted that in 2010 alone, there were 1,085,647 ‘officially declared admin-
istrations’ of the LS scales, making it the most widely used group of risk assessment tools
on the planet (p. 80). The LS/CMI includes 43 items organized around the ‘central eight’ risk
factors identified as the strongest correlates of criminal conduct and which are embedded
within the GPCSL model described earlier.
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

A number of studies have been conducted demonstrating the instrument’s sound psy-
chometric properties, including its reliability and predictive accuracy for a range of out-
comes (e.g. Brews, 2009; Rettinger & Andrews, 2010; Wormith, Hogg, & Guzzo, 2012)
and with mentally disordered offenders specifically (Girard & Wormith, 2004; Skeem
et al., 2014). In a recent meta-analysis of 128 studies comprising 137,931 offenders,
Olver, Stockdale, and Wormith (2014) reported the LS scales to significantly predict
general and violent recidivism (r = .29 and .23, respectively); these effects were upheld
across gender and ethnicity. In the present sample, the LS/CMIs were completed for
each offender by trained correctional staff and were included with the referral package
sent to the STU. The internal consistency of the LS/CMI in the present sample was satis-
factory (α = .78).

Brief Psychiatric Rating Scale


The Brief Psychiatric Rating Scale (BPRS; Overall & Gorham, 1962) is a widely used clinician-
rated tool intended to assess common psychopathological symptoms, particularly those
seen in psychotic patients. The tool is comprised of 18 items (i.e. symptom constructs)
that are rated on a scale from 1 (symptom is not present) to 7 (the symptom is extremely
severe), which can be combined into a total score. The BPRS has been shown to be clini-
cally valid when used by mental health workers across varying levels of professional skill
(McGorry, Goodwin, & Stuart, 1988; Ventura, Green, Shaner, & Liberman, 1993) with a con-
sistent underlying structure (Shafer, 2005). BPRS scores were completed by one of several
different psychiatrists conducting the intake assessment. Scores were based on a compre-
hensive interview with the offender and collateral file information, which included pre-
vious mental health and correctional reports. BPRS scores were determined during
routine clinical practice and therefore, it was not possible to evaluate inter-rater reliability.
In the present sample, the internal consistency of the BPRS was satisfactory (α = .87).

Diagnosis
All possible diagnoses were included in the analyses and were coded as present or not
present based on current presentation. Diagnoses were based on Diagnostic and statistical
manual of mental disorders, fourth edition, text revision (DSM-IV-TR) criteria (American Psy-
chiatric Association [APA], 2000) and were assigned by the intake psychiatrist at this facil-
ity. As with the BPRS, it was not possible to evaluate inter-rater reliability. Diagnoses were
6 D. A. KINGSTON ET AL.

collapsed into relevant categories including the presence of a SUD, schizophrenia or other
psychotic disorder, a mood disorder, an anxiety disorder, a personality disorder, and a non-
substance related mental disorders (NSMDs). NSMDs included any Axis I mental health
diagnosis that did not also have a co-occurring SUD. In this study, we partitioned sub-
stance use from non-SUDs because the former is more heavily weighted toward the
GPCSL model and general antisocial characteristics (also see Rezansoff et al., 2013).
Finally, we assessed the relevance of any comorbidity as well as concurrent diagnoses;
that is, a mental disorder that is comorbid with an SUD.

Age of onset and outcome measures


Age of onset of criminal activity was determined based on offense history data collected
from the Canadian Police Information Center (CPIC) records. CPIC is a national record of
criminal arrests and convictions. Both charges and/or convictions were used as this
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

more accurately reflects offence history as it is common for some offenses to be


dropped or pled down to lesser charges.
In this study, we used institutional misconduct and post-release recidivism as our primary
outcome measures. Institutional misconducts were defined as any incident of physical
aggression while incarcerated at this facility. Any incident of physical aggression that is
observed by staff is required to be documented and entered into the electronic Incident
Management System (eIMS). The number of incidents of physical aggression was extracted
from eIMS for each participant. These incidents were recoded into a yes/no binary variable.
Recidivism was defined as a return to provincial correctional supervision on a new con-
viction within two years of the completion of a provincial sentence to incarceration. In this
study, criminal recidivism was defined as any recidivism (violent, nonviolent, or sexual).
Violent recidivism included criminal offenses categorized as homicide and related,
serious violent, and assault and related. Nonviolent recidivism covered any type of
offence that was neither violent, nor sexual and included break and enter and related,
traffic/import drugs, arson/property damage, among others. Sexual recidivism included
offenses classified as violent sexual and nonviolent sexual.

Data analytic plan


The analyses proceeded in several phases. First, in light of the heterogeneity of BPRS scale
content and to aid interpretation of study findings, we began by conducting a principal
components analysis of its scale items to reduce them to a smaller number of homo-
geneous symptom groupings for subsequent analysis. The principal components analysis
(PCA) would also serve as a psychometrically meaningful exercise to compare to previous
PCAs of the tool. Second, we computed point biserial correlations between binary (yes–no)
diagnosis and LS/CMI and BPRS total and need/component scores to examine the risk rel-
evance of the diagnostic categories; that is, diagnosis associated with increased risk/need
should demonstrate positive correlations with the LS/CMI and its domains, while those
with little or no risk relevance should demonstrate weak or negative associations. Similarly,
diagnoses associated with high levels of symptomatology and impairment should demon-
strate strong positive associations with the BPRS and its component domains.
Third, we examined the predictive accuracy of LS/CMI and BPRS total and need/com-
ponent scores for the three criterion variables operationalized above through receiver
PSYCHOLOGY, CRIME & LAW 7

operating characteristic (ROC) curve analysis. ROC generates an Area Under the Curve
(AUC) statistic ranging from 0 to 1.0 representing the extent to which a randomly selected
recidivist would have a higher score on the measures than a randomly selected non-reci-
divist. Fourth, we examined the prediction of outcome criteria by binary mental disorder
diagnosis through Cox regression survival analysis to control for individual differences in
follow-up time. Cox regression, in turn, provides a hazard ratio (e B) which represents the
predicted change in the hazard of the unwanted outcome (i.e. recidivism) for every one
unit change in the predictor variable. Values for e B above 1.0 indicate that increases in
the predictor are associated with increases in the hazard of the outcome, while values
below 1.0 denote an inverse relationship, that is, increases in the predictor are associated
with decreases in the hazard of the outcome. Fifth, we examined age as a potential mod-
erating factor on the relationship between mental illness and recidivism via Cox regression
and Kaplan–Meier survival analyses.
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

Results
As indicated earlier, the suspected presence of a mental illness was the primary inclusion
criteria to the STU. Approximately 82% were diagnosed with an SUD (100/121). The vast
majority also exhibited some form of comorbidity; that is, the presence of an SUD with
another mental disorder (77.7%; n = 94) or any comorbidity, regardless of type of
mental disorder (89.3%, n = 108). Approximately, 40% of participants were diagnosed
with schizophrenia or another psychotic disorder (n = 48). A similar proportion of individ-
uals were diagnosed with a mood disorder (n = 43), an anxiety disorder (n = 52), and a per-
sonality disorder (n = 46). Approximately, 15% of participants were diagnosed with an
NSMD (n = 18); that is, an Axis I disorder without any co-occurring SUD.
The follow-up time ranged up to 486 days with an average of 344.9 days (SD = 53.45
days). Approximately 25% of the sample (n = 30) re-offended violently, whereas half of
the sample (52.9%; n = 64) recidivated with a criminal offense. Of note, no one in the
present sample was detected of committing a sexual re-offense.
Prior to the main analyses, BPRS items were subjected to an exploratory factor analysis
with principal components extraction followed by varimax rotation. Factors yielding an
eigenvalue >1 were selected and scree plots were examined to confirm factor selections.
Individual factor coefficients for each subject were then derived based on the global factor
structure. Items with factor coefficients ≥0.5 were considered to load on a respective
factor. A high Kaiser–Meyer–Olkin measure of sampling adequacy (KMO = .809) confirmed
the validity of using factor analyses for structure detection. The four principal components
(Eigenvalues: 3.7, 3.5, 2.7, and 2.5) cumulatively accounted for 68% of the total variance.
We labeled the four components as follows: Hostility–agitation (20.3% of the total
variance – Somatic Concern, Grandiosity, Hostility, Suspiciousness, Uncooperativeness,
Excitement, and Disorientation), Thought Disorder (19.4% of the overall variance –
Conceptual Disorganization, Mannerisms and Posturing, Hallucinatory Behavior, and
Unusual Thought Content), anxiety–depression (14.9% of the total variance – Anxiety,
Guilt Feelings, Tension, and Depressive Mood), and Anergia (13.4% of the total variance
– Emotional Withdrawal, Motor Retardation, and Blunted Affect). These components are
similar to previous research on the BPRS (see Burger, Yonker, Calsyn, Morse, & Klinkenberg,
2003; Crippa, Sanches, Hallak, Loureiro, & Zuardi, 2002). Subsequent analyses are
8 D. A. KINGSTON ET AL.

conducted using BPRS PCA components. The component loading matrix is presented
in Table 1.

Profiles of psychiatric symptoms and criminogenic need as a function of


diagnosis
Table 2 presents descriptive statistics for the diagnostic subgroups and total sample on
the LS/CMI and BPRS scores, including criminogenic need domains and the BPRS
components.
In turn, point biserial correlations between binary diagnosis (yes–no diagnosis present)
with LS/CMI and BPRS domains are presented in Table 3; several themes were noteworthy.
First, SUD and dual diagnosis with SUD generated the highest number of significant posi-
tive correlations with LS/CMI need, correlating significantly with four of the central eight
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

domains, as well as the total score; by contrast, neither diagnosis correlated significantly
with the BPRS or its component scores. Second, personality disorder diagnosis and any
dual diagnosis were associated with more serious criminal histories and higher LS/CMI
total scores; personality disorder diagnosis was also associated with antisocial pattern
and negative companions while any dual diagnosis was associated with pro-criminal atti-
tudes; however, the former evinced weak or negative associations with BPRS symptoma-
tology, while the latter was associated with hostility–agitation and total symptomatology.
The remaining diagnoses appeared to have limited risk relevance. For instance, NSMD was
inversely associated with most LS/CMI domains and total score, while mood and anxiety
diagnoses evinced weak or negative associations with criminogenic domains. Although
anxiety was predictably associated with higher BPRS scores on anxiety–depression and
inversely associated with thought disorder, these three diagnostic categories otherwise
did not have strong associations with the BPRS. Finally, schizophrenia diagnoses were
associated significantly with poor use of leisure/recreation and negative companions,
but no other LS/CMI domain or its total score; however, schizophrenia diagnosis was

Table 1. Rotated component matrix of BPRS item scores.


BPRS item Hostility–agitation Thought disorder Anxiety–depression Anergia
Hostility .836 −.165 .221 −.014
Excitement .751 .288 .185 −.316
Grandiosity .637 .401 −.080 .012
Somatic concern .625 .205 .279 .034
Uncooperativeness .575 .392 −.088 .365
Suspiciousness .522 .402 .309 .364
Disorientation .461 .418 −.178 .120
Unusual thought .406 .801 −.028 .018
Hallucinatory behavior −.066 .746 .223 .018
Mannerisms and posture .322 .685 −.176 .305
Conceptual disorganization .464 .652 −.245 .298
Guilt feelings −.022 −.102 .809 −.068
Depressed mood .021 −.173 .784 .160
Anxiety .277 .222 .746 .081
Tension .383 .446 .560 .030
Motor retardation .078 .014 −.086 .907
Blunted affect −.196 .196 .229 .776
Emotional withdrawal .168 .516 .185 .613
Note: Items loading on a given component in italics. Items loading highest on a given component are highlighted in bold
italics and included in the computation of component scores.
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

Table 2. Descriptive statistics (Means and SD) of diagnostic groups on criminogenic and psychopathological predictors.
Measure NSMD SUD Schizophrenia Mood disorder Anxiety disorder Personality disorder Any DD DD with SUD Total sample
Criminogenic predictors (n = 16) (n = 99) (n = 46) (n = 43) (n = 51) (n = 44) (n = 105) (n = 93) (n = 118)
(LS/CMI needs and total score)
Criminal history 4.2 (2.5) 6.2 (1.7) 6.0 (1.8) 5.7 (2.0) 5.7 (2.3) 6.6 (1.4) 6.0 (1.9) 6.2 (1.7) 5.8 (2.0)
Education/Employment 4.5 (2.9) 5.9 (2.3) 5.9 (2.2) 5.0 (2.8) 5.5 (2.5) 6.1 (2.2) 5.8 (2.5) 5.9 (2.3) 5.7 (2.5)
Family/Marital 2.3 (1.0) 2.3 (1.2) 2.4 (1.1) 2.3 (1.2) 2.3 (1.2) 2.3 (1.2) 2.3 (1.2) 2.3 (1.2) 2.3 (1.2)
Leisure/Recreation 1.7 (0.5) 1.8 (0.5) 1.9 (0.4) 1.7 (0.6) 1.7 (0.5) 1.8 (0.5) 1.8 (0.5) 1.8 (0.5) 1.8 (0.5)
Companion 1.2 (1.0) 2.2 (1.1) 2.4 (1.1) 1.7 (1.5) 1.8 (1.2) 2.5 (1.1) 2.1 (1.1) 2.3 (1.1) 2.1 (1.1)
Pro-criminal attitudes 1.9 (1.4) 1.9 (1.4) 2.2 (1.3) 1.7 (1.4) 1.7 (1.4) 2.1 (1.4) 1.9 (1.4) 1.9 (1.4) 1.9 (1.4)
Substance abuse 2.3 (2.0) 5.6 (1.9) 5.4 (2.2) 4.8 (2.1) 4.9 (2.5) 5.3 (2.2) 5.3 (2.1) 5.7 (1.9) 5.1 (2.3)
Antisocial pattern 1.9 (1.2) 2.1 (1.2) 2.1 (1.1) 1.9 (1.3) 2.0 (1.3) 2.3 (1.1) 2.1 (1.2) 2.1 (1.2) 2.1 (1.2)
LS/CMI total score 19.9 (8.7) 28.0 (7.3) 28.1 (7.1) 24.8 (8.5) 25.6 (9.1) 29.0 (6.8) 27.3 (7.9) 28.1 (7.4) 26.6 (8.1)
Psychopathological predictors (BPRS PCA) (n = 17) (n = 88) (n = 44) (n = 38) (n = 47) (n = 39) (n = 95) (n = 82) (n = 108)
Hostility–agitation 14.8 (8.8) 14.4 (6.4) 16.8 (8.2) 15.8 (7.9) 13.9 (5.7) 14.9 (7.0) 15.1 (6.9) 14.8 (6.5) 14.4 (6.8)
Thought disorder 6.9 (4.3) 7.6 (4.8) 10.8 (5.6) 6.8 (4.6) 5.8 (2.9) 7.4 (4.7) 7.8 (4.9) 7.8 (5.0) 7.5 (4.7)
Anxiety–depression 12.4 (4.1) 11.4 (4.3) 11.0 (3.8) 12.6 (4.4) 12.9 (3.8) 10.2 (4.1) 11.5 (4.1) 11.6 (4.2) 11.5 (4.3)

PSYCHOLOGY, CRIME & LAW


Anergia 4.6 (2.8) 5.5 (3.1) 6.8 (3.8) 4.9 (2.4) 4.7 (2.2) 5.4 (3.5) 5.4 (3.1) 5.6 (3.2) 5.3 (3.0)
BPRS total 38.6 (13.2) 38.9 (14.2) 45.3 (16.5) 39.8 (14.9) 37.3 (9.8) 37.8 (13.9) 39.7 (14.2) 39.6 (14.3) 38.6 (14.0)
Note: LS/CMI, Level of Service/Case Management Inventory; BPRS, Brief Psychiatric Rating Scale; NSMD, non-substance related mental disorder; SUD, substance use disorder; DD, co-morbid diag-
noses. Schizophrenia includes a diagnosis of Schizophrenia and other psychotic disorders (e.g. Schizoaffective disorder, Delusional Disorder).

9
10
D. A. KINGSTON ET AL.
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

Table 3. Bivariate associations (Point Biserial Correlations) between binary diagnosis and score on criminogenic and psychopathological predictors.
Measure NSMD SUD Schizophrenia Mood disorder Anxietydisorder Personality disorder Any DD DD with SUD
Criminogenic predictors (LS/CMI needs and total score)
Criminal history −.33*** .36*** .07 −.06 −.08 .29** .22* .34***
Education/Employment −.19* .23* .07 −.21* −.06 .13 .12 .20*
Family/Marital .01 .04 .05 .02 .00 .05 .11 .05
Leisure/Recreation −.05 .16 .19* −.09 −.09 .07 .11 .08
Companion −.31*** .37*** .23* −.24** −.18 .28** .10 .36***
Pro-criminal attitudes −.01 .05 .18 −.12 −.14 .14 .23* −.01
Substance abuse −.49*** .54*** .10 −.07 −.06 .09 .01 .53***
Antisocial pattern −.04 .06 .02 −.10 −.05 .19* .07 .04
LS/CMI total score −.33*** .40*** .15 −.17 −.11 .23* .25** .36***
Psychopathological predictors (BPRS PCA)
Hostility–agitation .02 .01 .29** .15 −.07 .06 .25** .09
Thought disorder −.06 .06 .57*** −.11 −.31*** −.01 .15 .10
Anxiety–depression .09 −.07 −.11 .18 .28** −.24* .01 .02
Anergia −.10 .15 .41*** −.09 −.18 .01 .11 .16
BPRS total .00 .03 .39*** .06 −.09 −.05 .20* .12
Note: *p < .05. **p < .01. ***p < .001. n = 118 for r with LS/CMI, n = 108 for r with BPRS. LS/CMI, Level of Service/Case Management Inventory; BPRS, Brief Psychiatric Rating Scale; NSMD, non-
substance related mental disorder. SUD, substance use disorder. DD, co-morbid diagnoses. Schizophrenia includes a diagnosis of Schizophrenia and other psychotic disorders (e.g. Schizoaffective
disorder and Delusional disorder).
PSYCHOLOGY, CRIME & LAW 11

significantly associated with high levels of psychiatric symptomatology, with moderate to


large positive correlations with most BPRS domains, including the total score.

Diagnostic, psychiatric, and criminogenic predictors of recidivism


In Table 4, we present AUC values for criminogenic variables (i.e. LS/CMI items and total
score) and psychopathological variables based on BPRS scores for the prediction of insti-
tutional misconduct, violent recidivism, and criminal recidivism.
None of the criminogenic or psychopathological variables significantly predicted insti-
tutional misconduct. However, several criminogenic variables were associated with recidi-
vism. Criminal history was the only LS/CMI item associated with violent recidivism. Criminal
history was also associated with criminal recidivism, as was Leisure/Recreation, Compa-
nions, Substance Abuse, Antisocial Pattern, and total LS/CMI score. With regard to psycho-
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

pathological predictors, hostility–agitation was associated with both violent and criminal
recidivism. Anergia was also significantly associated with violent recidivism.
Cox regression analyses were also conducted on the LS/CMI items and BPRS factors in
predicting community recidivism. Cox Regression estimates relative risk ratios (hazard
rates) but, importantly, controls for variations in time-at-risk. Results were largely similar
with the AUC analyses reported earlier with the following exceptions. Anergia was no
longer a significant predictor of violent recidivism and hostility–agitation failed to
predict criminal recidivism.
Cox regression survival analysis was conducted to examine the predictive validity of
psychiatric diagnoses for violent and criminal recidivism, while adjusting for individual
differences in follow-up time (see Table 5). For categorical predictors such as DSM diagno-
sis, Exp(β), is the ratio of the estimated hazard for a case with the characteristic to that of a
case without the characteristic (i.e. relative risk).

Table 4. Predictive accuracy of criminogenic and psychopathological variables.


Institutional
violence Violent recidivism Criminal recidivism
Measure N AUC 95% CI AUC 95% CI AUC 95% CI
Criminogenic predictors (LS/CMI needs and total score)
Criminal history 118 .59 [.48, .70] .69** [.59, .79] .77*** [.69, .86]
Education/Employment 118 .49 [.37, .61] .47 [.35, .59] .59 [.48, .69]
Family/Marital 118 .54 [.43, .66] .52 [.41, .64] .58 [.48, .69]
Leisure/Recreation 118 .50 [.39, .62] .53 [.41, .65] .62* [.51, .72]
Companion 118 .54 [.43, .65] .51 [.40, .63] .63* [.53, .73]
Pro-criminal attitudes 118 .56 [.45, .67] .47 [.35, .59] .51 [.41, .62]
Substance abuse 118 .56 [.45, .67] .59 [.48, .70] .67** [.57, .77]
Antisocial pattern 118 .48 [.37, .60] .55 [.44, .66] .64** [.53, .73]
LS/CMI total score 118 .55 [.44, .67] .56 [.45, .67] .70*** [.60, .79]
Psychopathological predictors (BPRS PCA)
Hostility–agitation 108 .60 [.48, .73] .69** [.57, .82] .63* [.52, .73]
Thought disorder 109 .58 [.45, .71] .58 [.44, .72] .56 [.46, .67]
Anxiety–depression 109 .53 [.40, .67] .57 [.44, .70] .55 [.44, .66]
Anergia 109 .52 [.40, .64] .66** [.53, .78] .58 [.47, .68]
BPRS total 109 .58 [.44, .71] .68** [.55, .81] .60 [.49, .71]
LS/CMI, Level of Service/Case Management Inventory; BPRS, Brief Psychiatric Rating Scale; AUC, Area under the Curve.
*p < .05.
**p < .01.
***p < .001.
12
D. A. KINGSTON ET AL.
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

Table 5. Cox regression survival analysis: predictive validity of psychiatric diagnoses.


Violent recidivism Criminal recidivism
95% CI 95% CI
Diagnostic predictor B SE Wald p eB LL UL B SE Wald p eB LL UL
NSMD −0.57 .61 0.89 .347 0.56 0.171 1.859 −1.04 .47 4.95 .026 0.36 0.142 0.884
SUD 0.40 .54 0.56 .454 1.50 0.522 4.284 1.00 .43 5.43 .020 2.72 1.173 6.307
Schizophrenia 0.47 .37 1.63 .202 1.59 0.779 3.259 0.50 .26 0.04 .843 1.05 0.639 1.732
Mood disorder 0.18 .37 0.24 .625 1.20 0.578 2.491 −0.14 .26 0.27 .603 0.87 0.521 1.461
Anxiety disorder −0.15 .37 0.16 .686 0.86 0.414 1.786 −0.17 .26 0.46 .498 0.84 0.509 1.389
Personality disorder 0.25 .37 0.46 .497 1.29 0.624 2.646 0.44 .25 3.11 .078 1.56 0.952 2.552
Any DD 1.35 1.02 1.75 .186 3.84 0.524 28.225 0.78 .52 2.28 .131 2.18 0.792 6.008
DD with SUD 0.74 0.54 1.90 .168 2.10 0.731 6.006 0.83 .36 5.27 .022 2.29 1.128 4.628
Note: N = 121. NSMD, non-substance related mental disorder. SUD, substance use disorder. DD, co-morbid diagnoses. Schizophrenia includes a diagnosis of Schizophrenia and other psychotic
disorders (e.g. Schizoaffective disorder and Delusional Disorder). Significant p-values are in bold font.
PSYCHOLOGY, CRIME & LAW 13

None of the psychiatric diagnoses predicted violent recidivism and few predicted any
criminal recidivism. The 95% confidence intervals are quite large for the Exp(β) values in
light of the binary nature of the diagnostic predictors. With regard to criminal recidivism,
NSMD was significantly related to outcome, χ 2(1, N = 121) = 5.40, p = .020. Specifically, the
presence of an NSMD decreased the hazard rate by 64% (e−1.04). In contrast, the presence
of an SUD was significantly related to outcome, χ 2(1, N = 121) = 5.90, p = .015 and
increased the hazard rate by 172% (e1.00). Similarly, the presence of a dual diagnosis (i.e.
mental illness with an SUD) was related to criminal recidivism, χ 2(1, N = 121) = 5.57,
p = .018 and increased the hazard rate by 129% (e0.83).
Cox regression analyses were conducted to examine the extent to which binary diag-
nosis predicted violent and any criminal recidivism after controlling for criminogenic
need (see Kingston, et al., in press). In this case, we used LS/CMI total score minus the crim-
inal history subcomponent, so that the quantity would reflect the sum total of crimino-
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

genic needs. For each regression, binary diagnosis and LS/CMI score (minus criminal
history) were entered simultaneously and their unique relationships to outcome exam-
ined. In short, while the LS/CMI criminogenic need score significantly uniquely predicted
any criminal recidivism for all analyses (e B = 1.04–1.06, ps = .040–.004), none of the diag-
nostic categories uniquely significantly predicted this outcome. However, the LS/CMI crim-
inogenic need score did not significantly uniquely predict violent recidivism nor did any of
the diagnostic categories.

Interrelationships of age, psychiatric symptomatology, risk, and recidivism


We conducted our next set of analyses to examine the importance of age as a potential
moderating variable of associations between mental illness (i.e. BPRS total score and
binary mental health diagnosis) and recidivism.

BPRS analyses
Cox regression survival analyses were conducted to examine the relationship of psychiatric
symptomatology (BPRS total score), age at first sentence, and their interaction to violent
and general criminal recidivism. BPRS total score was chosen given that this is an
overall measure of psychiatric symptomatology and it significantly predicted both com-
munity recidivism outcomes, while age at first sentence is a good proxy for onset of crim-
inal behavior. For instance, in bivariate analyses, age at first sentencing date was a
significant predictor of any criminal (r = −.28, p = .002) as well as violent (r = −.22,
p = .016) recidivism. LS/CMI total score was entered in the final block of the analyses as
a means of controlling for baseline risk; the results are reported in Table 6. First, higher
overall levels of psychiatric symptomatology (BPRS score) significantly predicted increased
violent and criminal recidivism (block 1) and continued to predict both outcomes after
controlling for age at first sentence (block 2); the latter also significantly predicted any
criminal recidivism and trended toward significance in the prediction of community vio-
lence. Their interaction (block 3) did not attain significance in the prediction of either
outcome, and the nature of this relationship remained unchanged after adding the LS/
CMI total score into the final step (block 4) of the analyses. The BPRS total score,
however, continued to predict any criminal recidivism at these subsequent steps
(ps = .042 and .055, respectively) of the analysis.
14
D. A. KINGSTON ET AL.
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

Table 6. Cox regression survival analyses: examination of incremental relationship of psychiatric symptomatology and age at first sentence to recidivism and their
interaction with and without controls for baseline risk.
Violent recidivism Criminal recidivism
95% CI for e B 95% CI for e B
Regression model B SE Wald p e B
Lower Upper B SE Wald p e B
Lower Upper
Block 1
BPRS total .043 .013 11.34 .001 1.04 1.018 1.070 .024 .009 6.30 .012 1.02 1.005 1.043
Block 2
BPRS total .036 .013 8.07 .004 1.04 1.011 1.063 .020 .009 4.75 .029 1.02 1.002 1.040
Age first sentence −.097 .052 3.51 .061 .91 .820 1.005 −.058 .025 5.44 .020 .94 .898 .991
Block 3
BPRS total −.021 .061 0.12 .729 .98 .868 1.104 .062 .030 4.14 .042 1.06 1.002 1.129
Age first sentence −.265 .189 1.96 .162 .77 .529 1.112 .036 .066 0.30 .582 1.04 .911 1.180
BPRS × age interaction .004 .004 0.93 .335 −.003 .002 1.94 .164
Block 4
BPRS total −.015 .064 0.05 .816 .99 .869 1.117 .056 .029 3.69 .055 1.06 .999 1.119
Age first sentence −.308 .198 2.42 .120 .74 .498 1.083 .052 .064 0.67 .412 1.05 .930 1.194
BPRS × age interaction .003 .004 0.79 .375 −.002 .002 1.78 .183
LS/CMI total −.053 .034 2.50 .114 .95 .888 1.013 .033 .023 2.04 .153 1.03 .988 1.081
Note: BPRS, Brief Psychiatric Rating Scale; LS/CMI, Level of Service/Case Management Inventory. Significant p-values are in bold font.
PSYCHOLOGY, CRIME & LAW 15

To unpack these recidivism analyses further, we dichotomized the age at first sentence
and BPRS variables. For age at first sentence (M = 19.97 years, SD = 8.91) we used an age
cutoff of 18 years so that the distinction would be those offenders who had any youth sen-
tence (i.e. under 18 years of age, n = 63) in contrast to those who did not (i.e. age 18 or
above, n = 58).2 The mean BPRS score was 38.65 (SD = 14.0) and so a cut score of 40
was employed (nobody had a score of 39). Of note, BPRS total score approached signifi-
cance in the prediction of criminal recidivism among individuals with a youth sentence
(AUC = .63, p = .117, 95%CI = .48–.78) but not individuals without a youth sentence
(AUC = .49, p = .936, 95%CI = .32–.67). The two dichotomous variables were used to
create four BPRS × Age groups and failure rates of violent and general criminal recidivism
were examined via Kaplan–Meier survival analysis.
As seen in Figure 1, individuals under age 18 at first sentence who scored relatively high
on the BPRS (Group 3) had the highest and fastest rate of violent as well as any criminal
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

recidivism. More specifically, pairwise comparisons demonstrated this group to have sig-
nificantly higher failure rates for any criminal recidivism than individuals without a prior
youth conviction, whether they scored higher on the BPRS, log rank χ 2(1, N = 46) = 6.92,
p = .009, or did not, log rank χ 2(1, N = 58) = 14.13, p < .001. Individuals with a youth criminal
history and high BPRS scores also had significantly higher rates of violent recidivism than
lower scoring individuals without a youth criminal history, log rank χ 2(1, N = 58) = 8.15,
p = .004. Individuals with a youth conviction but scoring lower on the BPRS had signifi-
cantly higher rates of general recidivism than individuals scoring lower on the BPRS
without a youth conviction, log rank χ 2(1, N = 75) = 5.13, p = .024. No other group differ-
ences attained significance. In short, psychiatric symptomatology was associated with
increased recidivism among individuals with an earlier onset of criminal history; not so
for individuals with a later onset.

Figure 1. Survival analysis: Rates of criminal and violent recidivism over time (days) as a function of
psychiatric symptomatology (BPRS cut score) and age at first sentence. Note: Figure 1a Group 3 has
significantly higher rates of criminal recidivism than Groups 2 and 4 and Group 1 has significantly
higher rates of criminal recidivism than Group 2. No other group differences are significant.
Figure 1b Group 3 has significantly higher rates of violent recidivism than Group 2. No other group
differences are significant.
16 D. A. KINGSTON ET AL.

Mental health diagnosis


The second set of analyses examined age as a possible moderator of the association
between binary diagnosis and recidivism. Cox regressions survival analyses were thus con-
ducted entering age at first sentence and binary diagnosis in the first step, followed by an
age × diagnosis interaction term in the second step, to examine if young age at onset of
criminal behavior potentially moderated the relationship of mental health diagnosis to
recidivism. Age at first sentence significantly predicted criminal recidivism in the first
step for all analyses (e B = .932–.945, ps = .005–.028), while none of the diagnostic cat-
egories uniquely significantly predicted criminal recidivism. Upon entering the interaction
term in the second step, however, a significant interaction was found between age at first
sentence and any personality disorder (Wald = 7.86, p = .005) and any anxiety disorder
(Wald = 3.97, p = .046). No other interaction terms were significant. These analyses were
repeated for violent recidivism as an outcome variable; again age at first sentence
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

uniquely predicted violent recidivism (e B = .893–.904, ps = .019–.040), but none of the


binary diagnoses nor the interaction terms significantly predicted this outcome.
Kaplan–Meier survival analyses were then conducted to examine the association
between binary mental disorder diagnosis and early vs. late onset of criminality (i.e.
youth vs. no youth disposition) for those diagnoses that demonstrated significant inter-
actions above (personality disorder and anxiety disorder). First, there were no differences
in rates of recidivism among individuals with prior youth sentences with (69.2%) or
without (67.6%) personality disorders; however, among individuals first sentenced after
age 18, individuals without a personality disorder diagnosis had significantly lower rates
of recidivism (26.3%) than those with such a diagnosis (55.0%), log rank χ 2(N = 58) =
4.60 p = .032. In addition, individuals without a prior youth sentence or personality dis-
order (PD) diagnosis had significantly lower rates of criminal recidivism than both
groups who had prior youth sentences (with PD) log rank χ 2(N = 64) = 13.96 p < .001,
(no PD) log rank χ 2(N = 75) = 13.58 p < .001. Second, there were no differences in rates
of recidivism among individuals with prior youth sentences with (69.0%) or without
(67.6%) anxiety disorders; however, among individuals first sentenced after age 18, indi-
viduals with an anxiety disorder diagnosis had lower rates of recidivism (21.7%) than
those without an anxiety disorder diagnosis (45.7%) at χ 2(N = 58) = 3.39, p = .065. In
addition, individuals with an anxiety disorder diagnosis had significantly lower rates of
criminal recidivism than both groups who had prior youth sentences (with anxiety dis-
order) log rank χ 2(N = 52) = 12.35 p < .001, (no anxiety disorder) log rank χ 2(N = 57) =
11.80 p < .001. In all, it seems having any personality disorder served as a risk factor
among individuals with older onset of criminal behavior, while having an anxiety disorder
was associated with lower rates of recidivism among such individuals (Figure 2).

Discussion
In the present, prospective study, we examined the relative predictive accuracy of psycho-
pathological indicators and criminogenic risk factors on violent and criminal recidivism in a
sample of provincially sentenced mentally disordered offenders. Consistent with the
results of the most recent meta-analyses (Bonta, Blais, & Wilson, 2014) and subsequent
studies (Kingston et al., 2015; Rezansoff et al., 2013), the best predictors of criminal recidi-
vism were consistent with the central eight risk factors identified within the GPCSL.
PSYCHOLOGY, CRIME & LAW 17
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

Figure 2. Survival analysis: Rates of criminal recidivism over time (days) as a function of psychiatric
symptomatology (binary DSM diagnosis) and age at first sentence. Note: Figure 2a Group 1 has signifi-
cantly lower rates of criminal recidivism than Groups 2–4. No other group differences are significant.
Figure 2b Group 2 has significantly lower rates of criminal recidivism than Groups 3 and 4 and is lower
than Group 1 at p = .065. Group 1 has significantly lower rates of criminal recidivism than Group 4 and is
lower than Group 3 at p = .052. No other group differences approached or attained significance.

Psychopathological predictors were generally poor predictors of criminal behavior in both


univariate and multivariate analyses, with a few exceptions. The only diagnostic categories
that seemed to be meaningfully associated with criminal recidivism were NSMDs (which
were inversely related to criminal recidivism) and SUDs, which are part of the central
eight risk factors embedded within the GPCSL. In turn, SUD-related diagnoses and
NSMD demonstrated meaningful associations with LS/CMI total and criminogenic need
scores, in the same direction as their univariate associations with recidivism, further attest-
ing to their risk relevance (i.e. SUD associated with greater risk and need, while NSMD
associated with lower risk and need).
Next, we examined the importance of age of onset of criminal behavior as a potential
moderator variable. As noted earlier, Skeem et al. (2011) proposed age of onset of criminal
activity as a variable that may differentiate the extent to which the relationship between
mental illness and criminal behavior is direct or fully mediated. An important distinction
has been made with regard to offenders who commence with criminal activity at a
relatively young age versus those who start relatively late (Hodgins, 2008; Laajasalo &
Häkkänen, 2005; Mathieu & Côté, 2009; Tengström et al., 2001; Vitelli, 1997). Consistent
with past research (Silver, 2006; Tengström et al., 2001), we hypothesized that mental
illness would be a significant predictor of criminal behavior for late start offenders, as com-
pared to early start offenders. This hypothesis was not supported, and in fact we found the
opposite. Although age of onset of criminal behavior in the present sample was a signifi-
cant moderator, results indicated that psychopathology predicted criminal recidivism
among individuals who had received a youth sentence (i.e. first sentenced under 18
years of age) but not individuals who were first sentenced at age 18 or older, and thus
who had no youth criminal history.
18 D. A. KINGSTON ET AL.

This pattern seemed to be observed with some specific classes of mental disorder.
Regardless of diagnosis for any of the mental disorder categories, individuals with an
earlier onset of criminal behavior had the highest rates of recidivism; different patterns
emerged, though, for individuals with a later onset of criminal behavior as per Skeem
et al. (2011), at least for some diagnoses. For instance, finer grained analyses of any per-
sonality disorder demonstrated that individuals with such a diagnosis had significantly
higher rates of criminal recidivism than those without such a diagnosis specifically
among individuals with a later onset of criminal activity. Moreover, the opposite pattern
was seen for anxiety diagnoses; individuals with an anxiety diagnosis with late onset of
criminal behavior had lower rates of recidivism than individuals without such a diagnosis.
Such findings are consistent with the possibility of anxiety disorders perhaps even mitigat-
ing risk among late onset offenders while personality disorder increases risk in this group.
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

Limitations
Several limitations of the present study need to be considered. The sample was relatively
small (N = 121) and the follow-up time was relatively short, with an average of 344 days.
This is offset by the fact that the base rate of criminal recidivism in the present sample
was relatively high (52.5%), to yield sufficient power for many analyses. Moreover, as
noted earlier, this sample consisted of provincial offenders and by definition received a
shorter sentence length (i.e. 2 years less a day) than federal offenders. As such, our
results may not be generalizable to offenders who receive longer prison sentences.
Another limitation is that inter-rater reliability of the BPRS and psychiatric diagnoses
were not available. There is accumulating evidence of specific problems with the reliability
of some diagnoses in particular and this may have influenced the predictive accuracy of
these variables. Despite the possible diagnostic errors, we feel this methodology is ecolo-
gically valid with the results having applicability in routine clinical practice.
With regard to the prediction of future behavior, it is of note that we were unable to
determine whether an individual was presenting with active symptoms at the time of
re-offense. This is particularly important as recent evidence shows that the relationship
between mental illness and recidivism can vary within a particular offender across time
(Peterson et al., 2014). Indeed, instability of the illness in the offender, treatment resistant
symptoms, medication non-adherence, and perhaps, positive treatment response were
not measured in this investigation. As we were unable to control for the presence or
phase of an individual’s illness at the time of recidivism, it is a potentially confounding vari-
able. Perhaps mental illness is best conceptualized as an acute dynamic risk factor, such
that certain active symptoms of mental illness may play a role in the timing of a criminal
offense. For example, Hanson and Harris (1998) discussed the important connection
between acute, rapidly changing risk factors (e.g. mood and intoxication) and recidivism.
In a sample of 409 sexual offenders matched on psychiatric symptoms, the authors found
that although recidivists and non-recidivists did not differ on measures of general psycho-
logical symptoms, the recidivists tended to show an increase in psychological symptoms
just prior to re-offending.
A final limitation of this study is that youth records are sealed if an individual remains
offense-free for a varying period of time (from 2 months to 5 years, depending on the
nature of the youth charges) upon turning 18. The youth charges would only appear on
PSYCHOLOGY, CRIME & LAW 19

the national criminal record if an adult was charged or convicted of a crime between the
ages of 18 and 21. Approximately one quarter of the offenders in the present sample had
no charges or convictions between 18 and 21; they could have had criminal charges as
youth, which would affect our coding of age of onset of criminal behavior, but these
charges did not appear in the record.

Conclusions
Identifying the predictors of criminal behavior directly informs psychological interven-
tions. The Risk, Need, and Responsivity (RNR) model has been the most widely used
and validated model of offender rehabilitation (Andrews & Bonta, 2010 Andrews et al.,
1990) although its applicability to mentally disordered offenders needs further attention
(Morgan et al., 2012). The RNR model outlines the importance of treating criminogenic
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

needs – that is, dynamic offender characteristics, that when changed, are associated
with reduced recidivism rates. The belief that mental illness is a criminogenic need has
led to a number diversion programs (e.g. mental health courts). Although such programs
have been successful in reducing the rates of incarceration and increasing general access
to mental health services, the ability of these programs to reduce recidivism has been
mixed at best (Sarteschi, Vaughn, & Kim, 2011; Skeem et al., 2011) and is particularly
weak for those programs weighted more heavily toward mental health models, as
opposed to criminal justice based models (Morgan et al., 2012). In contrast, there are a
number of programs that are consistent with the RNR model and such programs have
shown reductions in recidivism (Andrews, Zinger, et al., 1990; Dowden, Blanchette, &
Serin, 1999; Hanson, Bourgon, Helmus, & Hodgins, 2009).
The results of the present study support previous meta-analyses and lend further
support to the prominence of the central eight risk factors as compared to mental
health variables. However, mental illness was a relevant predictor for a small subset of
offenders (although not whom we originally hypothesized). Nevertheless, these findings
support the notion that treating mental illness may prevent a minority of criminal behavior
(also see Peterson et al., 2014) but again, targeting criminogenic needs, such as substance
abuse, is current best practice in offender rehabilitation.
Despite the notion that mental illness may not be directly associated with criminal
activity in the vast majority of cases, mental illness is likely an important treatment
target for most, if not all, mentally disordered offenders. Indeed, there is evidence that
mentally disordered offenders present with more general risk factors than non-mentally
disordered offenders (Skeem et al., 2014). Moreover, consistent with the RNR model,
mental illness is also likely best conceptualized as a responsivity factor, such that some
individuals with active symptoms of mental illness may find it difficult to engage in treat-
ment or attend to the treatment content. Therefore, targeting such symptoms may be an
important first step in the treatment process. Some programs integrating both mental
health and criminogenic factors have been developed with some positive preliminary
results in terms of mental health and criminal justice outcomes (Morgan, Kroner, Mills,
Bauer, & Serna, 2013). Addressing mental health symptoms may also help to promote
improvement on an individual’s identified criminogenic needs. For example, managing
mental health symptoms may allow one to make more adequate use of leisure time, resist-
ing urges to turn to substance use to manage symptoms, and to obtain employment, all of
20 D. A. KINGSTON ET AL.

which have been shown to reduce the likelihood of recidivism. More work needs to be
done on how best to integrate approaches in treating mental illness and general risk
factors in promoting both symptom improvement and recidivism reduction. Finally, meth-
odologically rigorous treatment outcome studies with mentally disordered offenders are
needed to identify best practices with this population.

Notes
1. Hodgins (2008) also described a third type of mentally disordered offender who shows a
chronic course of illness with no aggressive behavior following the onset of the disorder.
After one to two decades, these individuals engage in serious violent behavior (e.g.
murder) that is generally directed toward those who care for them. Research suggests that
this group is relatively small in number relative to the other two types.
2. We found an age 15 cutoff (as commonly reported in the literature) to establish early vs. late
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

starters was quite limiting in this sample and did not discriminate recidivists from non-recidi-
vists as accurately as any youth disposition vs. no youth disposition. In the present sample, any
youth disposition (under age 18) was thus more informative than younger youth dispositions
compared to all other dispositions.

Acknowledgment
We would like to thank Michael Seto for his comments on an earlier version of this paper.

References
Alltucker, K. W., Bullis, M., Close, D., & Yovanoff, P. (2006). Different pathways to juvenile delinquency:
Characteristics of early and late starters in a sample of previously incarcerated youth. Journal of
Child and Family Studies, 15(4), 479–492. doi:10.1007/s10826-006-9032-2
American Psychiatric Association. (2000). Diagnostic and statistical manual of mental disorders
(4th ed., text rev.). Washington, DC: Author.
Andrews, D. A., & Bonta, J. (1994). The psychology of criminal conduct. Cincinnati, OH: Anderson.
Andrews, D. A., & Bonta, J. (2010). The psychology of criminal conduct (5th ed.). New Providence, NJ:
Anderson.
Andrews, D. A., Bonta, J., & Wormith, J. S. (2004). Level of service/case management inventory (LS/CMI).
Toronto, ON: Multihealth Systems.
Andrews, D. A., Zinger, I., Hoge, R. D., Bonta, J., Gendreau, P., & Cullen, F. T. (1990). Does correctional
treatment work? A clinically relevant and psychologically informed meta-analysis. Criminology, 28,
369–404.
Appelbaum, P. S., Robbins, P. C., & Monahan, J. (2000). Violence and delusions: Data from the
MacArthur violence risk assessment study. American Journal of Psychiatry, 157(4), 566–572.
doi:10.1176/appi.ajp.157.4.566
Bonta, J., Blais, J., & Wilson, H. A. (2013). The prediction of risk for mentally disordered offenders: A
quantitative synthesis. Public Safety Canada. Corrections Research: User report.
Bonta, J., Blais, J., & Wilson, H. A. (2014). A theoretically informed meta-analysis of the risk for general
and violent recidivism for mentally disordered offenders. Aggression and Violent Behaviour, 19,
278–287. doi:10.1016/j.avb.2014.04.014
Bonta, J., Law, M., & Hanson, K. (1998). The prediction of criminal and violent recidivism among men-
tally disordered offenders: A meta-analysis. Psychological Bulletin, 123(2), 123–142. doi:10.1037/
0033-2909.123.2.123
Brennan, P. A., Mednick, S. A., & Hodgins, S. (2000). Major mental disorders and criminal violence in a
Danish birth cohort. Archives of General Psychiatry, 57(5), 494–500. doi:10.1001/archpsyc.57.5.494
PSYCHOLOGY, CRIME & LAW 21

Brews, A. L. (2009). The level of service inventory and female offenders: Addressing issues of reliability and
predictive validity (Unpublished master’s thesis). University of Saskatchewan, Saskatoon,
Saskatchewan, Canada.
Burger, G. K., Yonker, R. D., Calsyn, R. J., Morse, G. A., & Klinkenberg, W. D. (2003). A confirmatory factor
analysis of the Brief Psychiatric Rating Scale in a homeless sample. International Journal of Methods
in Psychiatric Research, 12, 192–196.
Corrado, R. R., Cohen, I., Hart, S., & Roesch, R. (2000). Comparative examination of the prevalence of
mental disorders among jailed inmates in Canada and the United States. International Journal of
Law and Psychiatry, 23(5), 633–647. doi:10.1016/S0160-2527(00)00054-6
Crippa, J. A. S., Sanches, R. F., Hallak, J. E. C., Loureiro, S. R., & Zuardi, A. W. (2002). Factor structure of
Bech’s version of the brief psychiatric rating scale in Brazilian patients. Brazilian Journal of Medical
and Biological Research, 35, 1209–1213.
Diamond, P. M., Wang, E. W., Holzer III, C. E., Thomas, C., & Cruser, des A. (2001). The prevalence of
mental illness in prison. Administration and Policy in Mental Health Services Research, 29(1), 21–
40. doi:10.1023/A:1013164814732
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

Douglas, K. S., Guy, L. S., & Hart, S. D. (2009). Psychosis as a risk factor for violence to others: A meta-
analysis. Psychological Bulletin, 135(5), 679–706. doi:10.1037/a0016311
Dowden, C., Blanchette, K., & Serin, R. C. (1999). Anger management programming for federal male
inmates: An effective intervention (Research Report R-82). Ottawa, ON: Correctional Service of
Canada.
Elbogen, E. B., & Johnson, S. C. (2009). The intricate link between violence and mental disorder:
Results from the national epidemiological survey on alcohol and related conditions. Archives of
General Psychiatry, 66(2), 152–161. doi:10.1001/archgenpsychiatry.2008.537
Fazel, S., & Danesh, J. (2002). Serious mental disorder in 23000 prisoners: A systematic review of 62
surveys. The Lancet, 359, 545–550. doi:10.1016/s0140-6736(02)07740-1
Fazel, S., Gulati, G., Linsell, L., Geddes, J. R., & Grann, M. (2009). Schizophrenia and violence: Systematic
review and meta-analysis. Plos Med, 6(8). doi:10.1371/journal.pmed.1000120
Girard, L., & Wormith, J. S. (2004). The predictive validity of the level of service inventory-Ontario revi-
sion on general and violent recidivism among various offender groups. Criminal Justice and
Behavior, 31, 150–181.
Goethals, K., Willigenburg, L., Buitelaar, J., & van Marle, H. (2008). Behaviour problems in childhood
and adolescence in psychotic offenders: An exploratory study. Criminal Behaviour and Mental
Health, 18, 153–165. doi:10.1002/cbm.688
Hanson, R. K., Bourgon, G., Helmus, L., & Hodgins, S. (2009). The principles of effective correctional
treatment also apply to sexual offenders: A meta-analysis. Criminal Justice and Behaviour, 36(9),
865–891. doi:10.1177/0093854809338545
Hanson, R. K., & Harris, A. (1998). Dynamic predictors of sexual recidivism (User Report JS42–82/1998-
01E). Ottawa: Department of the Solicitor General of Canada.
Hodgins, S. (2008). Violent behavior among people with schizophrenia: A framework for investi-
gations of causes, and effective treatment and prevention. Philosophical Transactions of the Royal
Society B, 363. doi:10.1098/rstb.2008.0034
Jones, R. M., Van den Bree, M., Ferriter, M., & Taylor, P. J. (2010). Childhood risk factors for offending
before first psychiatric admission for people with schizophrenia: A case-control study of high
security hospital admissions. Behavioural Sciences and the Law, 28, 351–365. doi:10.1002/bsl.885
Kingston, D. A., Olver, M. E., Harris, M., Wong, S. C. P., & Bradford, J. M. (2015). The relationship
between mental disorder and recidivism in sexual offenders. International Journal of Forensic
Mental Health, 14, 10–22. doi:10.1080/14999013.2014.974088
Laajasalo, T., & Häkkänen, H. (2005). Offence and offender characteristics among two groups of
Finnish homicide offenders with schizophrenia: Comparison of early- and late-start offenders.
Journal of Forensic Psychiatry and Psychology, 16(1), 41–59. doi:10.1080/14789940412331327679
Markowitz, F. E. (2011). Mental illness, crime, and violence: Risk, context, and social control.
Aggression and Violent Behaviour, 16(1), 36–44. doi:10.1016/j.avb.2010.10.003
22 D. A. KINGSTON ET AL.

Mathieu, C., & Côté, G. (2009). A modelization of differences between early- and late-starter French
Canadian offenders. International Journal of Forensic Mental Health, 8, 25–32. doi:10.1080/
14999010903014705
McGorry, P. D., Goodwin, R. J., & Stuart, G. W. (1988). The development, use, and reliability of the brief
psychiatric rating scale (nursing modification) – An assessment procedure for the nursing team in
clinical and research settings. Comprehensive Psychiatry, 29, 575–587.
Monahan, J. (1981). Predicting violent behaviour: An assessment of clinical techniques. Washington, DC:
Sage Publications.
Morgan, R. D., Flora, D. B., Kroner, D. G., Mills, J. F., Varghese, F., & Steffan, J. S. (2012). Treating offen-
ders with mental illness: A research synthesis. Law and Human Behavior, 36, 37–50.
Morgan, R. D., Kroner, D. G., Mills, J. F., Bauer, R. L., & Serna, C. (2013). Treating justice involved persons
with mental illness. Criminal Justice and Behavior, 41, 902–916.
Olver, M. E., Stockdale, K. C., & Wormith, J. S. (2014). Thirty years of research on the level of service
scales: A meta-analytic examination of predictive accuracy and sources of variability.
Psychological Assessment, 26, 156–176. doi:10.1037/a0035080
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

Overall, M. E., & Gorham, D. R. (1962). The brief psychiatric rating scale. Psychological Reports, 10, 799–
812. doi:10.2466/pr0.1962.10.3.799
Peterson, J., Skeem, J. L., Hart, E., Vidal, S., & Keith, F. (2010). Analyzing offense patterns as a function
of mental illness to test the criminalization hypothesis. Psychiatric Services, 61(12), 1217–1222.
doi:10.1176/ps.2010.61.12.1217
Peterson, J. K., Skeem, J., Kennealy, P., Bray, B., & Zvonkovic, A. (2014). How often and how consist-
ently do symptoms directly precede criminal behaviour among offenders with mental illness?. Law
and Human Bahviour, 439–449. doi:10.1037/hb0000075
Rettinger, L. J., & Andrews, D. A. (2010). General risk and need, gender specificity, and the recidivism
of female offenders. Criminal Justice and Behavior, 37, 29–46. doi:10.1177/0093854809349438
Rezansoff, S. N., Moniruzzaman, A., Gress, C., & Somers, J. (2013). Psychiatric diagnoses and multiyear
criminal recidivism in a Canadian provincial offender population. Psychology, Public Policy, and
Law, 19(4), 443–453. doi:10.1037/a0033907
Sarteschi, C. M., Vaughn, M. G., & Kim, K. (2011). Assessing the effectiveness of mental health courts: A
quantitative review. Journal of Criminal Justice, 39, 12–20.
Shafer, A. (2005). Meta-analysis of the brief psychiatric rating scale factor structure. Psychological
Assessment, 17(3), 324–335. doi:10.1037/1040-3590.17.3.324
Silver, E. (2006). Understanding the relationship between mental disorder and violence: The need for
a criminological perspective. Law and Human Behavior, 30(6), 685–706. doi:10.1007/s10979-006-
9018-z
Skeem, J. L., Manchak, S., & Peterson, J. K. (2011). Correctional policy for offenders with mental illness:
Creating a new paradigm for recidivism reduction. Law & Human Behaviour, 35, 110–126. doi:10.
1007/s10979-010-9223-7
Skeem, J. L., Winter, E., Kennealy, P. J., Louden, J., & Tatar II, J. R. (2014). Offenders with mental illness
have criminogenic needs, too: Toward recidivism reduction. Law and Human Behaviour, 38(3),
212–224. doi:10.1177/0032885511415226
Tengström, A., Hodgins, S., & Kullgren, G. (2001). Men with schizophrenia who behave violently: The
usefulness of an early- versus late-start offender typology. Schizophrenia Bulletin, 27(2), 205–218.
Retrieved from http://resolver.scholarsportal.info.proxy.bib.uottawa.ca/resolve/05867614/v27i0002/
205_mwswbvoaelot.xml.
Ventura, J., Green, M. F., Shaner, A., & Liberman, R. P. (1993). Training and quality assurance with the
brief psychiatric rating scale: The drift busters. International Journal of Methods in Psychiatric
Research, 3, 221–244.
Vitelli, R. (1997). Comparison of early and late start models of delinquency in adult offenders.
International Journal of Offender Therapy and Comparative Criminology, 41(4), 351–357. doi:10.
1177/0306624X97414005
Walters, G. D., & Crawford, G. (2014). Major mental illness and violence history as predictors of insti-
tutional misconduct and recidivism: Main and interaction effects. Law and Human Behaviour, 38(3),
238–247. doi:10.1037/lhb0000058
PSYCHOLOGY, CRIME & LAW 23

Witt, K., van Dorn, R., & Fazel, S. (2013). Risk factors for violence in psychosis: Systematic review and
meta-regression analysis of 110 studies. PLoS ONE, 8(2). doi:10.1371/journal.pone.0055942
Wormith, J. S. (2011). The legacy of D. A. Andrews in the field of criminal justice: How theory and
research can change policy and practice. International Journal of Forensic Mental Health, 10, 78–
82. doi:10.1080/14999013.2011.577138
Wormith, J. S., Hogg, S., & Guzzo, L. (2012). The predictive validity of a general risk/needs assessment
inventory on sexual offender recidivism and an exploration of the professional override. Criminal
Justice and Behavior, 39, 1511–1538. doi:10.1177/0093854812455741
Downloaded by [Temple University Libraries] at 14:28 13 June 2016

You might also like