ArticlePDF Available

An indepth actuarial assessment for wife assault recidivism: The Domestic Violence Risk Appraisal Guide

Authors:

Abstract and Figures

An actuarial tool, the Ontario Domestic Assault Risk Assessment (ODARA), predicts recidivism using only variables readily obtained by frontline police officers. Correctional settings permit more comprehensive assessments. In a subset of ODARA construction and cross-validation cases, 303 men with a police record for wife assault and a correctional system file, the VRAG, SARA, Danger Assessment, and DVSI also predicted recidivism, but the Hare Psychopathy Checklist (PCL-R) best improved prediction of recidivism, occurrence, frequency, severity, injury, and charges. In 346 new cases, ODARA and PCL-R independently predicted recidivism. An algorithm was derived for a combined instrument, the Domestic Violence Risk Appraisal Guide (DVRAG), and an experience table is presented (N=649). Results indicated the importance of antisociality in wife assault.
Content may be subject to copyright.
ORIGINAL ARTICLE
An Indepth Actuarial Assessment for Wife Assault Recidivism:
The Domestic Violence Risk Appraisal Guide
N. Zoe Hilton Æ Grant T. Harris Æ Marnie E. Rice Æ
Ruth E. Houghton Æ Angela W. Eke
Published online: 2 June 2007
American Psychology-Law Society/Division 41 of the American Psychological Association 2007
Abstract An actuarial tool, the Ontario Domestic Assault
Risk Assessment (ODARA), predicts recidivism using only
variables readily obtained by frontline police officers.
Correctional settings permit more comprehensive assess-
ments. In a subset of ODARA construction and cross-val-
idation cases, 303 men with a police record for wife assault
and a correctional system file, the VRAG, SARA, Danger
Assessment, and DVSI also predicted recidivism, but the
Hare Psychopathy Checklist (PCL-R) best improved pre-
diction of recidivism, occurrence, frequency, severity, in-
jury, and charges. In 346 new cases, ODARA and PCL-R
independently predicted recidivism. An algorithm was
derived for a combined instrument, the Domestic Violence
Risk Appraisal Guide (DVRAG), and an experience table is
presented (N = 649). Results indicated the importance of
antisociality in wife assault.
Keywords Domestic violence Risk assessment
Actuarial Psychopathy
As the 21st century began, the weight of empirical evi-
dence indicated that assessment tasks are performed more
accurately with the use of statistical rather than unstruc-
tured or informal methods. A meta-analysis by Grove and
colleagues (Grove et al. 2000) found that formal methods
(including statistical equations, actuarial tables, and algo-
rithmic programs) were more accurate than clinical judg-
ment (subjective, intuitive, and variable or unspecified
processes for combining information) to render predictions.
The advantage of statistical methods was consistent across
time, whether assessors were psychologists, whether clin-
ical information was available, and regardless of assessors’
experience and access to information. Grove et al. (2000)
reported that the size of this advantage varied with the
outcome of interest, and was highest for forensic outcomes
(e.g., inpatient assault, criminal recidivism). Ægisdo
´
ttir
et al. (2006) updated the meta-analysis focusing on
assessments pertinent to counseling psychology. They ob-
tained similar results across a variety of statistical formu-
lae, outcome criteria, and levels of clinician familiarity
with the data. Violence risk assessments yielded the
greatest superiority for statistical over informal methods.
Empirical support for the statistical prediction of vio-
lence raises optimism for similar performance in predicting
wife assault specifically. Huss and Langhinrichsen-Rohling
(2000) suggested that psychopaths represent a serious and
persistent subgroup of wife assaulters (see also Johnson
et al. 2006; Spidel et al. 2007). The Hare Psychopathy
Checklist (PCL-R; Hare 2003) is the accepted, standard
forensic tool for measuring psychopathy and, although not
designed as a risk assessment, its score has proven to be
one of the most robust predictors of violent and criminal
recidivism (Harris et al. 2001). It has the largest weight in
an actuarial tool for the assessment of the risk of violent
recidivism, the Violence Risk Appraisal Guide (VRAG,
The opinions expressed herein are ours and do not necessarily reflect
the opinions of SSHRC.
N. Z. Hilton (&) G. T. Harris M. E. Rice
R. E. Houghton
Research Department, Mental Health Centre, 500 Church Street,
Penetanguishene, ON, Canada L9M 1G3
e-mail: zhilton@mhcp.on.ca
Present Address:
R. E. Houghton
York Regional Police, Newmarket, ON, Canada
A. W. Eke
Ontario Provincial Police, Orillia, ON, Canada
123
Law Hum Behav (2008) 32:150–163
DOI 10.1007/s10979-007-9088-6
Harris et al. 1993), which has over 30 published replica-
tions (reviewed in Quinsey et al. 2006). The VRAG also
predicted violent recidivism by men with a history of se-
vere wife assault or murder, better than the PCL-R alone
(Hilton et al. 2001), and better than nonactuarial tools for
the prediction of spousal violence in a follow-up of wife
assaulters (Grann and Wedin 2002).
Domain-Specific Prediction
The Ontario Domestic Assault Risk Assessment (ODARA;
Hilton et al. 2004) is an actuarial risk assessment for wife
assault recidivism that can be scored by frontline police
officers on the basis of readily available information. It
contains thirteen empirically selected items, some appar-
ently specific to domestic relationships (prior domestic
violence, confinement of the victim, number of children,
perpetrator assaulted victim when she was pregnant, vic-
tim’s children from prior relationships, victim’s concern
about future assaults, and barriers to victim support), and
several common to the literature on the risk of antisocial
behavior in general, (prior correctional sentence, failure on
conditional release, substance abuse, threats of violence,
and two items pertaining to prior nondomestic violence).
ODARA score predicted wife assault recidivism in police
records, with an ROC area of .77 in construction, and .72 in
100 cross-validation cases. Because the ODARA was
originally intended for use by police officers, only infor-
mation ‘‘routinely available in the field’’ was considered
for inclusion (Hilton et al. 2004, p. 269), and information
more difficult to obtain, such as PCL-R score, was not.
Hilton et al. (2004) reported that total score on the
Spousal Assault Risk Assessment (SARA), a nonactuarial
assessment that can be scored from more indepth correc-
tional and clinical records (Kropp and Hart 2000), predicted
wife assault recidivism. The SARA can be considered a
candidate in the search for assessments to improve upon the
information routinely available to frontline police officers.
Similar methods were used to develop the Domestic Vio-
lence Screening Instrument (DVSI, Williams and Houghton
2004) which also contains items requiring some clinical
information (e.g., treatment participation) and the Danger
Assessment (DA; Campbell 1986, 1995) was associated
with any new wife assault (Goodman et al. 2000; Weisz
et al. 2000), although it is designed as a victim interview for
predicting lethal assault in particular.
The present research addressed how the risk of wife
assault recidivism is most accurately assessed when the
more complete information typically available to forensic
clinicians and criminal justice officials is available. Such
assessors typically have access to extensive information
about the perpetrator’s history of criminal and antisocial
behavior as well as victim reports and domain-specific
information (e.g., relationship history) not typically avail-
able in front line police work, and should use the most
accurate information available to assess risk. In this cir-
cumstance, use of a frontline risk assessment that does not
take account of indepth clinical information about antiso-
ciality could lead to suboptimal prediction.
In correctional contexts, including those involving sen-
tencing, parole, and community supervision, the concern is
often not only the likelihood of violence, but also its
severity. Detection of valid predictors might also be more
powerful using continuous measures of recidivism (Heckert
and Gondolf 2004). Instruments designed to predict
dichotomous outcomes (i.e., violent recidivism vs. not)
might not predict violence severity or such other continuous
outcomes as the number of recidivistic incidents, injury
caused, severity of violent behavior, or the seriousness of
charges incurred (though see Koziol-McLain et al. 2006).
The present study, therefore, examined whether the pre-
diction of dichotomous recidivism plus several indices of
the severity of recidivism could be enhanced by adding
indepth material to the easy-to-gather ODARA items.
In addition, correctional populations might be higher risk
than the entire population of offenders, and have less
opportunity to recidivate than offenders who receive only
community dispositions. We anticipated that incarceration
would result in attenuated predictive accuracy of the
instruments previously tested by Hilton et al. (2004
). In re-
cent research (Popham and Hilton 2006), convicted domestic
violence perpetrators were more likely to fall in the highest
ODARA risk category than the population of perpetrators
with a police record, of which they are a subgroup, raising the
need for greater discrimination in this subpopulation.
The Present Study
We examined whether the prediction of wife assault
recidivism and its severity could be improved by adding
more detailed clinical information to the ODARA. Seto
(2005) reported that predicting sex offenders’ recidivism
could not be improved by combining the most accurate
actuarial risk assessment with scores from other actuarial
tools; however, only tools specifically designed to predict
recidivism among sex offenders were considered. We tes-
ted tools designed specifically for domestic violence (the
SARA, DA, and DVSI), and indepth assessments known to
predict violence in general (PCL-R and VRAG). We began
with a sample of men with a police record of assault against
a female cohabiting partner or ex-partner, from the initial
ODARA construction and cross-validation, but used only
those cases that had a more detailed correctional system
case file (usually a probation or pre-sentence report).
Law Hum Behav (2008) 32:150–163 151
123
Results from this sample were tested in a new sample of
cases (not used in any previous development or validation)
drawn from three police sources, with additional clinical
material from correctional files.
Method
Inclusion Criteria, Data Bases, and Selection
From the three police records management systems described
below, we identified each male perpetrator in incidents
classified as domestic. For each perpetrator, we then isolated
the incident closest to, but no later than, December 31, 1996,
(Sample 1) or 1997 (Sample 2) in which, according to a
victim/witness report and other police evidence, he com-
mitted an act of physical assault or credible threat of death
with weapon in hand in the presence of a victim who was a
current or former wife or common-law wife (hereafter called
the index assault). Eligible cases were all those with evidence
of both an intimate relationship and an existing or prior
marital or cohabiting relationship. Incidents involving only
nonspousal victims or nonviolent acts were not eligible as the
index incident, and cases in which offenders and victims had
not lived together were also excluded, mainly because it was
not always discernible from police reports of assaults
involving noncohabiting relationships whether the relation-
ships were intimate. Eligibility did not require that offenders
be arrested or criminally charged for the index assault, but
research assistants only coded cases where they could agree,
based on the police reports, that an assault had clearly
occurred. In the case of multiple eligible incidents, the one
closest to the December 31 cut-off was coded as the index
assault.
Cases for Sample 1 were drawn from an electronic re-
cords management system used by the Ontario Provincial
Police as well as by approximately 50 municipal police
services. Most rural areas, Aboriginal communities, and
many municipalities in Ontario, Canada’s most populous
province, were represented in this archive. The records
included dispatch information and verbatim reports by
frontline officers; names of the perpetrator (suspect), vic-
tims, and witnesses; charges laid; and all details of the
investigation including statements by all parties. We were
provided a list of all entries classified as domestic, and
research assistants retrieved the records and determined
eligibility. We began with 689 cases (used for the con-
struction and cross-validation of the ODARA; Hilton et al.
2004) from which we selected all those with files compiled
by the Ministry of Corrections (e.g., pre-sentence reports,
psychological evaluations, psychosocial assessments, pro-
bation officer notes, etc.) as a result of charges pertaining to
the index or any other offense. The existence and location
of these files was ascertained from a Ministry offender
tracking data base. This selection process yielded a sample
of 303 men with a police report of wife assault all of whom
had a corrections file containing indepth clinical and psy-
chosocial information. The other cases from the 689 did
not have a correctional file and are not used in the present
study. Most index assaults occurred in 1996 (74%) or 1995
(19%); 75% were recorded as having resulted in criminal
charges.
For Sample 2 we obtained new cases, not used in
Sample 1 or any previous study, using the same identifi-
cation and selection process as in Sample 1. Approximately
half (n = 168, 49%) were retrieved from the same source as
used to compile Sample 1. The rest (n = 178, 51%) were
from similar records management systems maintained by
two urban police services in the Greater Toronto Area. All
cases were selected first from the police archives for eli-
gibility and then from the Ministry offender tracking
database for having a corrections file. In this sample of 346
cases, most index assaults occurred in 1996 (71%) or 1997
(17%); 90% resulted in criminal charges (possibly a result
of greater emphasis on charges in the metropolitan juris-
dictions).
For each perpetrator in Samples 1 and 2, we supple-
mented information on pre-index criminal history, index
disposition, and recidivism with criminal record data ob-
tained from a federal database maintained the Royal Cana-
dian Mounted Police, which records all criminal charges,
convictions, and criminal dispositions across Canada.
Procedure and Variables
All file information was retrieved within, or transported
securely to, a secure location in a police research unit. So-
ciodemographic variables, criminal history, items from
formal assessments, and outcomes were coded at the same
time using all documentation available regarding the index
and prior history. Coding was done by researchers with
many years’ experience coding such information, who also
trained and closely supervised two graduate assistants with
one to 3 years’ coding experience. We created a manual for
quantifying and coding all data available in the police re-
ports plus information in corrections files about childhood
history, adult functioning, relationships, and assessment.
Variables describing the two samples are shown in Table 1;
those requiring fuller explanations are described next.
Substance Abuse Score
As per scoring instructions for this ODARA item, 1 point is
given for each of eight variables similar to those predicting
violent recidivism in previous research (Harris et al. 1993):
perpetrator used alcohol immediately before or during the
152 Law Hum Behav (2008) 32:150–163
123
index assault, perpetrator used drugs immediately before or
during the index assault, perpetrator abused alcohol or
drugs in the days or weeks (up to a month) leading up to the
index date, perpetrator used alcohol or drugs more than
usual in the days or weeks (up to a month) leading up to the
index date, perpetrator is more angry or violent when using
alcohol or drugs, perpetrator was previously charged for an
offense while under the effects of alcohol, perpetrator had
an alcohol problem since age 18, and perpetrator had a
drug problem since age 18.
Prior Domestic Incidents
We recorded all police reports pertaining to incidents,
separate from and occurring before the index assault,
involving a forceful physical contact or threat of physical
harm by the perpetrator against the victim of the index
assault or a previous female partner with whom he had
lived or was living, or the partner’s child(ren). Similar
information was recorded for prior nondomestic incidents
(i.e., violence against any other persons).
Table 1 Sample
characteristics, shown as mean
(SD) or percent of sample
Note: * p \ .05, ** p \ .01,
*** p \ .001. All
characteristics except
recidivism were coded as of
date of index assault. Sample
differences indicated by (t-test
or v
2
)
Sample 1 Sample 2
Offender characteristics
Age (yr) 35.5 (10.1) 35.3 (10.0)
Unemployed (%) 14 14
Substance abuse score 1.94 (1.64) 1.60 (1.63)**
Number prior domestic incidents 0.16 (0.56) 0.27 (0.71)*
Prior criminal history score 9.95 (15.1) 16.3 (24.7)***
Violation of prior conditional release order (%) 42 27***
Violation of prior no-contact order (%) 8 4*
Total prior injury to female domestic partners 0.53 (1.50) 0.70(1.52)
Total prior injury to nondomestic victims 0.22 (1.23) 0.37(1.33)
Relationship characteristics
Victim age (yr) 32.4 (9.56) 32.4 (9.64)
Victim unemployed (%) 30.1 23.7
Duration of relationship (mo) 80.8 (83.9) 99.8 (106)*
Legally married at index (%) 37 45
Separated prior to index (%) 33 29
Perpetrator demonstrated jealousy (%) 14 11
Index assault details
Weapon used (%) 6 8
Perpetrator charged (%) 75 90***
Injury to victim 2.08 (0.89) 2.19 (0.97)
Formal assessments (potential range)
ODARA (0–13) 4.05 (2.15) 3.54 (2.00)*
SARA (0–40) 4.63 (4.94) 4.01 (4.00)
DA (0–15) 0.73 (1.20) 0.75 (1.23)
DVSI (0–30) 2.68 (2.38) 2.25 (2.09)*
PCL-R (0–40) 8.00 (6.81) 8.35 (6.67)
VRAG (–26 to +38) –2.17 (6.73) –3.66 (6.56)*
Wife assault recidivism
Any incident (%) 49 41
Number of incidents 0.83 (1.49) 0.65 (1.88)
Injury to partners (total score) 1.46 (2.20) 3.19 (3.40)***
Number of incidents with severe violence 0.30 (0.62) 0.70 (0.88)***
Cormier Lang score for recidivistic charges 1.46 (2.51) 1.69 (6.06)
Law Hum Behav (2008) 32:150–163 153
123
Prior Criminal History Score
We used the Cormier–Lang Criminal History score
(Quinsey et al. 2006), which captures the frequency and
severity of criminal conduct by totaling all charges ranging
from 1 (minor property offense) to 28 (homicide).
Violation of Prior Conditional Release Order
We considered any occasion on which the perpetrator,
while living in the community, disobeyed any order of a
criminal court, or civil court in the case of no-contact or-
ders. Examples of violation include: committing a new
criminal offense while on bail, probation, parole, or sus-
pended sentence; failure to appear for court or appointment
with a parole or probation officer; drinking or having
firearms when prohibited; contacting a person when pro-
hibited. We did not require criminal charges for the vio-
lation.
Injury
For all injury measures, we used a 5-point ordinal scale
derived from the Danger Assessment (Campbell, 1985,
p.105): 1 (no injuries or lasting pain), 2 (bruises, cuts, or
continuing pain), 3 (severe contusions, burns, broken
bones), 4 (head injury, internal injury, or permanent in-
jury), 5 (wounds from a weapon; e.g., stabbed, shot).
Severe Violence
We recorded whether the perpetrator committed acts de-
fined as severe on the Revised Conflict Tactics Scales
(CTS-2; Straus et al. 1996): punched, kicked, bit, hit with
something that could hurt, beat up, choked, slammed
against wall, burned or scalded, used a knife or gun; plus
similarly severe acts.
Violence Risk Appraisal Guide (VRAG)
The VRAG (Quinsey et al. 2006) was developed for male
offenders (Harris et al. 1993) and predicts violent recidi-
vism among forensic patients (Harris et al. 2002), nonfor-
ensic psychiatric patients (Harris et al. 2004), sex offenders
(Harris et al. 2003), and released prisoners (Glover et al.
2002). In construction, its 12 items, including PCL-R score,
all significantly and uniquely predicted violent recidivism
(Harris et al. 1993). The VRAG has repeatedly achieved
inter-rater reliability coefficients above .90 (Quinsey et al.
2006). In the present study, some items had no variance
(diagnosis of schizophrenia, female victim, and ever
married or equivalent), reducing the VRAG to a 9-item
modification. Previous modifications replicated the
VRAG’s accuracy, albeit with attenuated effect sizes (see
Harris and Rice 2003).
Hare Psychopathy Checklist-Revised (PCL-R) Score
The 20-item PCL-R (Hare 2003) is the standard tool for
forensic assessment of psychopathy. Although scoring is
often based on interview plus file information (Hare 2003),
we used file information alone, which yields reliable and
valid ratings, especially for violence prediction (Hare 2003;
Harris et al. 2001).
The Spousal Assault Risk Assessment (SARA)
The SARA (Kropp et al. 1999) is a structured professional
judgment scheme for domestic violence. Its 20 items,
gleaned from empirical and clinical literatures, are scored 0,
1, or 2. All 20 items were scored and summed for an item
total score. Although the SARA manual advises that inter-
views with the accused perpetrator and victim(s) be con-
ducted, the total score coded from file information has
achieved inter-rater reliability over .80 (Kropp and Hart
2000) and has predicted wife assault recidivism (Grann and
Wedin 2002; Heckert and Gondolf 2004; Hilton et al. 2004).
Danger Assessment (DA)
The DA (Campbell 1986, 1995) is an abuse history inter-
view plus a structured scale designed to assess the risk of
lethal wife assault. The items cover the offender’s domestic
and nondomestic violence history, access to weapons,
substance abuse, jealousy, sexual assault, threats, and the
victim’s fear of being killed. Test–retest reliability has
been at or above .89 in interviews (Campbell 1995). We
coded all 15 items and summed them for an item score
total. This unweighted score has been reported to predict
wife assault recidivism (e.g., Goodman et al. 2000; Heck-
ert and Gondolf 2004; Hilton et al. 2004; Weisz et al.
2000). In the present study, we deviated from the intended
implementation of the DA by scoring it from file material
rather than a victim interview.
Domestic Violence Screening Instrument (DVSI)
The DVSI (Williams and Houghton 2004) contains 12
items derived from a 34-item clinical guide collected on
over 9,000 cases analyzed for characteristics associated
with men’s history of repeated domestic violence, supple-
mented by literature reviews and focus groups to identify
items deemed to predict such violence. Each item has two
to four possible scores and the total score is the sum of the
item scores. In the development of the DVSI, this score had
an internal consistency of .71 (inter-rater reliability was not
154 Law Hum Behav (2008) 32:150–163
123
reported) and predicted severe wife assault recidivism,
ROC = .68 (Williams and Houghton 2004). The DVSI-R,
in which missing information was scored as 0, distin-
guished between first-time and repeat offenders being as-
sessed for violence towards partners or children,
ROC = .71 (Williams and Grant 2006). The DVSI-R also
added a structured clinical judgment of low, moderate or
high risk of violence, which was a significantly worse
predictor than raw item total score (Williams and Grant
2006) and was not used in the present study. The DVSI was
designed to be scored from documentary material. Because
it was published after our data were collected, we scored it
from other variables. Missing information was coded 0 as
in the DVSI-R. Two items (domestic violence treatment,
history of domestic violence restraining orders) could only
be scored dichotomously rather than trichotomously. One
item (drug or alcohol treatment) was missing for all cases,
making our implementation an 11-item modification.
Wife Assault Recidivism
Information about criminal and assaultive behavior occur-
ring after the index assault was obtained from police, cor-
rections, and criminal record reports, up to the end of 2001,
a mean of 5.10 years (SD = 1.44) post-index. Subsequent
assaults against a current or former wife or common-law
wife were deemed wife assault recidivism. We recorded
whether any incident occurred (dichotomous wife assault
recidivism), the number of such assaults in the follow-up
period, the number of incidents with severe violence, and
the Cormier-Lang score for all criminal offenses associated
with wife assault recidivism. Arrest, charges, or convictions
were not required to count as recidivism.
Reliability
Inter-rater reliability was measured by two research assis-
tants independently scoring a random subsample of 30
cases (previously reported by Hilton et al. 2004). Individ-
ual variables, including those comprising the ODARA, for
which agreement met or exceeded a Pearson correlation
coefficient of .80 (continuous measures) or kappa coeffi-
cient of .70 (categorical variables) in that test were re-
tained. Subsequently, ODARA score in 24 cases coded by
experienced coders, 10 of which were in addition scored by
completely novice assessors, all masked for outcome,
achieved inter-rater reliabilities .90. Dichotomous
recidivism coded independently by two research assistants
with all other information masked, also achieved an inter-
rater reliability > .90 (Hilton et al. 2004).
Results
Sample Characteristics and Recidivism
Table 1 shows the characteristics of Samples 1 and 2. The
samples differed on some variables (Table 1), and Sample
1 had significantly higher scores on three of the formal
assessments. Although a subset of the original ODARA
construction and cross-validation cases, Sample 1 used
only cases with a corrections file and thus represents a
higher risk group than previously published. As a result,
the predictive accuracy of the ODARA in this restricted-
range sample (recidivism base rate 49%) was marginally
lower than in the unselected, previous cross-validation
sample, r (303) = .30, p < .001, ROC area = .67, 95%
CI = .61–.73, suggesting there is benefit in developing an
improved prediction tool for a correctional sample.
Combining Assessments to Improve Predictive
Accuracy
As shown in Table 2, all assessments considered as poten-
tial additions to the ODARA in Sample 1 were significantly
and positively associated with dichotomous wife assault
recidivism (i.e., at least one post-index assault against a
current or former wife or common-law wife, documented in
a police report). Each assessment was also significantly
correlated with the ODARA in Sample 1: VRAG r = .52;
PCL-R, r = .57; SARA, r = .57; DA, r = .40; DVSI,
r = .49; all p < .001. Table 2 also shows the correlation
Table 2 Bivariate correlations between each candidate formal assessment and each outcome variable (nonsignificant correlations in italics) in
Sample 1 (n = 303)
VRAG PCL-R SARA DA DVSI
Dichotomous wife assault recidivism .19*** .22*** .18** .12* .17**
Number of recidivistic incidents .22*** .28*** .22*** .07 .24***
Total victim injury in recidivism .28*** .31*** .21*** .17** .18**
Number of CTS severe incidents .22*** .26*** .20** .07 .19**
Cormier-Lang score for recidivism .14* .23*** .18** .05 .26***
Note: * p < .05, ** p < .01, *** p < .001. Correlations are point-biserial for dichotomous recidivism. For comparison, the ODARA, constructed
partly on Sample 1, yielded correlations of .27 to .34, ps < .001
Law Hum Behav (2008) 32:150–163 155
123
between each candidate assessment and each of the con-
tinuous measures of recidivism. For each measure of
recidivism, we conducted a binary logistic (for dichotomous
recidivism) or linear (for continuous measures) regression
1
analysis for each formal assessment paired with the OD-
ARA. No formal assessment made an incremental and
independent improvement on the ODARA for dichotomous
recidivism (all p > .10). For each of the continuous mea-
sures of recidivism, only one formal assessment signifi-
cantly improved on the predictive accuracy of the ODARA.
For the number of recidivistic incidents, adding PCL-R
scores yielded a multiple R = .28, b = .13, F(1,
300) = 4.05, p < .05. For number of incidents with severe
violence, adding PCL-R scores yielded multiple R = .26,
b = .13, F(1, 282) = 3.75, p < .06; and for total recidivism
injury, adding PCL-R scores yielded multiple R = .31,
b = .22, F(1, 300) = 9.36, p < .01. Only adding DVSI
scores to the ODARA improved the prediction of Cormier-
Lang Criminal History score totaled across all recidivistic
incidents, multiple R = .26, b = .16, F(1, 300) = 6.88,
p < .01. Because the PCL-R was the most consistent con-
tributor to the prediction of continuous outcomes, was the
best bivariate predictor, and had yielded high levels of
predictive accuracy in separate previous studies of serious
domestic violence perpetrators (Hilton et al. 2001; Grann
and Wedin 2002), we selected it as most likely to improve
upon the ODARA in the prediction of wife assault recidi-
vism with indepth clinical and psychosocial information.
The DVRAG, an Indepth Assessment for Wife Assault
Recidivism
In order to optimize predictive accuracy and parsimonious
combination of assessments given adequate time and infor-
mation sources, the ODARA plus PCL-R score was identi-
fied as the basis of an indepth risk assessment. We used the
Nuffield weighting system to derive individual weights for
the 14 items (as used in the development of the VRAG;
Harris et al. 1993; Quinsey et al. 2006), whereby weights are
assigned to each value range of a variable according to its
deviation from the base rate. For each 5% (rounded) devia-
tion from the sample baserate in the recidivism rate among
offenders with a given value range, a weight of plus or minus
one is given to that value range. For six of the seven
dichotomous variables in Sample 1 (Prior conditional re-
lease failure, Threat, Confinement, Victim concern, Vio-
lence against others, Assault when pregnant) a score of zero
had a weight of 0. Prior correctional sentence was the
exception: its absence yielded a weight of –2 because per-
petrators scoring zero had a recidivism rate approximately
10% below the base rate. Offenders who made a Threat were
3% more likely than the base rate to recidivate, which was
rounded to 5% for a weight of +1. All item weights were
summed for each perpetrator, creating an assessment (see
Appendix) whose scores reflect deviation from the base rate
of wife assault recidivism (49%) in Sample 1.
The name Domestic Violence Risk Appraisal Guide
(DVRAG) reflected the similar development of this new
assessment to that of the VRAG. DVRAG scores were highly
related to each outcome variable in Sample 1: Pearson r
(n = 303) = .38, p < .001 for dichotomous wife assault
recidivism; .40, p < .001 for the number of recidivistic of-
fenses; .37, p < .001 for CTS severe domestic violence
recidivism; .39, p < .001 for total victim injury in recidi-
vistic offenses; and .33, p
< .001 for Cormier-Lang score.
For each correlation, DVRAG score was a statistically sig-
nificant improvement over ODARA score (z-score of dif-
ference > 1.65; Kanji 1993), and DVRAG score remained a
significant predictor of each outcome measure when con-
trolling for ODARA score, rs .19 to .31, ps < .01. The ROC
area for the predictive ability of DVRAG scores in Sample 1
was .71, SE = .03, corresponding to a Cohen’s d = .80 (Rice
and Harris 2005), and represented a statistically significant
improvement (p < .05, 1-tailed) over ODARA scores alone.
The 346 cases in Sample 2 were used to cross-validate
both the ODARA and DVRAG, because the cases were not
previously used to construct or validate either assessment.
The predictive validity (recidivism base rate = 41%) of the
ODARA was ROC area = .65 (SE = .03, 95% CI = .59 to
.71), d = .55; and of the DVRAG, ROC area = .70
(SE = .03, 95% CI = .64 to .75), d = .75; DVRAG score
represented an improvement in predictive validity over
ODARA score, p < .05, 1-tailed (Sample 2).
2
Table 3
shows the correlation between each of the ODARA and
DVRAG and each of the continuous outcome variables in
Sample 2. DVRAG correlations remained significantly
associated with each outcome when controlling for the
ODARA, and were significantly larger than those for the
ODARA (z-score of difference > 1.65) for dichotomous
wife assault recidivism, the number of recidivist incidents,
and their severity in terms of injury cases. The nonsignif-
icant improvement in predicting Cormier-Lang score for
recidivism (which is a count of criminal charges weighted
1
Because they are count variables and, therefore, skewed, some
statisticians would advise that the number of recidivistic incidents and
the number of severe incidents not be subjected to ordinary least-
squares regression. To check against any questionable conclusions,
we transformed these outcome variables using a Poisson transfor-
mation before analyses. As a second check, we subjected them to
Poisson loglinear analyses in conjunction with ODARA score and
each of the candidate assessments in a main effects analysis. In all
cases, the Poisson-based analyses yielded the same results as the
regression analyses reported here.
2
For the weighted ODARA items, without PCL-R, ROC = .68
(SE = .03, 95% CI = .62 to .74), d = .67, not a significant improve-
ment over the unweighted ODARA.
156 Law Hum Behav (2008) 32:150–163
123
for severity), and the lower predictive accuracy for this
outcome compared to measures not requiring criminal
charges, suggests that the DVRAG predicts actual reported
incidents rather than just police arrest decisions (which are
themselves influenced by many factors; Hilton et al. in
press). These results represent successful replications for
both the ODARA and DVRAG using cases not included in
the development or other testing of either instrument.
We combined the two samples to develop an actuarial
experience table for the DVRAG. In the combined sample
of 649 cases, mean DVRAG score was M = 2.88
(SD = 7.76), with a range of –10 to +37, and significant
predictive accuracy for wife assault recidivism (base
rate = 45%), ROC area = .70 (SE = .02, 95% CI ± .04),
p < .001, d = .75. The top scoring 3% of cases on the
DVRAG were all recidivists. The distribution of scores
also permitted identification of a category containing
approximately 3% of the lowest scoring cases, and subdi-
vision of remaining scores into seven equal sized catego-
ries at approximately the 20th, 40th, 60th, and 80th
percentiles, which yielded a statistically reliable increase in
recidivism rates at each score cut-off (Table 4). We eval-
uated the inter-rater reliability of DVRAG scores by having
two independent raters score the DVRAG (blind to recid-
ivism and each other’s scores) using 10 cases randomly
selected from the case files of a correctional treatment
program for domestic violence perpetrators. The Pearson
correlation coefficient was .92, and the intra-class corre-
lation (absolute agreement, random effects, single mea-
sures) was .90, both p < .001. This level of reliability for
the DVRAG and its overall standard deviation yielded a
standard error of measurement of 2.2 (Nunnally 1959). As
the DVRAG categories are no smaller than four points each
(Table 4), a perpetrator’s obtained score is likely to be
misclassified by no more than one category, 95% of the
time. Finally, we evaluated the scoring reliability of the
DVRAG by comparing the scoring by a research assistant
to that of an experienced forensic clinician with no explicit
DVRAG training, just the guidelines shown in the
Appendix, for 16 cases. The resulting Pearson correlation
was .83, indicating that the DVRAG is likely to be a reli-
able tool in regular forensic practices.
Direct comparisons of the predictive performance of the
DVRAG and ODARA with the other formal assessments in
the combined samples might be biased for several reasons:
almost half of the 649 cases were used to construct each of
the ODARA and DVRAG; some compromises from the
recommended scoring were required (e.g., interviews are
usual for the PCL-R, SARA, and DA); some DVSI items
were unavailable; and some VRAG items were invariant.
Yet, all showed statistically significant predictive validity
(Table 5). It is notable that the VRAG and PCL-R per-
formed at least as well as assessments designed specifically
for domestic violence risk assessment. This shared ability
to predict recidivism to at least some extent is not sur-
prising given the inter-correlations among the assessments
(Table 6), which in turn is partly attributable to some
overlapping items (e.g., nondomestic violence, conditional
release failure). On the other hand, different predictive
accuracy may be an inevitable consequence of some unique
items (e.g., treatment history, confinement of the victim).
3
Table 3 Cross-validation Predictive Accuracy for the ODARA and DVRAG, and z- scores of Difference Between Correlations of ODARA vs.
DVRAG with Outcomes, in Sample 2 (n = 346)
ODARA DVRAG z-score
Dichotomous wife assault recidivism ROC area .65 .70
Dichotomous wife assault recidivism .29 .36 1.68
Number of recidivistic incidents .36 .44 2.04
Total victim injury in recidivism .34 .41 1.74
Number of CTS severe incidents .26 .32 1.46
Cormier-Lang score for all recidivism .24 .29 1.15
Note: All rows under ODARA and DVRAG columns except first row are Pearson correlation coefficients (point-biserial for dichotomous
recidivism), all p < .001. z- scores p < .05 one-tailed except those in italics
3
We examined many other available variables reflecting perpetrators’
adult mental health, early adjustment, childhood abuse and neglect,
childhood exposure to domestic violence. The available data did not
permit the scoring of most psychiatric conditions (e.g., schizophrenia
symptoms, personality disorders). Other variables could be scored in
at least 200 cases, but were unrelated to wife assault recidivism (e.g.,
medical problems as an infant or young child; experienced childhood
corporal punishment, abuse, or neglect; witnessed domestic violence
as a child; prior head injury). Finally, a few variables scored for at
least 200 cases were correlated with dichotomous wife assault
recidivism: as an adult, the perpetrator exhibited procriminal attitudes
and values, r (346) = .15, p < .01, and attitudes unfavorable to con-
vention, r (335) = .21, p < .01; had been suspended or expelled from
elementary school, r (649) = .12, p < .01. All had been previously
identified as related to violent recidivism (Quinsey et al. 2006), but
none made an incremental improvement to DVRAG scores in pre-
dicting wife assault recidivism in either Sample 1, Sample 2, or both
samples combined. An exception, having been arrested under age 16,
r (649) = .10, p < .05, made a statistically significant improvement to
DVRAG score in 3 of 15 tests—to the prediction of total victim injury
in recidivism in the combined sample and in Sample 2, and prediction
of Cormier-Lang score for severity of recidivism in Sample 2.
Law Hum Behav (2008) 32:150–163 157
123
Because the five outcome measures yielded similar
levels of predictive accuracy, and were predicted by similar
variables, we computed the inter-correlations among all the
measures of wife assault recidivism. All were significantly
associated with each other, with correlations ranging from
.36 (total injury and Cormier Lang score) to .75 (dichoto-
mous wife assault and number of severe incidents), all
p<.01; however, they shared much less than half (on
average, 30%) of their total common variance.
Discussion
Previous research had resulted in a simple frontline actu-
arial domestic violence risk assessment for use by police
officers, victim service workers, and bail courts, the On-
tario Domestic Assault Risk Assessment (ODARA; Hilton
et al. 2004), whose 13 dichotomous items rely on easily
gathered information. The present study examined whether
this simple tool could be improved upon by adding tools
that require indepth psychosocial and clinical information
usually available to clinicians and forensic professionals,
including several formal assessments for domestic violence
(Spousal Assault Risk Assessment, SARA; Kropp and Hart
2000; Danger Assessment, DA; Campbell 1995; Domestic
Violence Screening Instrument, DVSI; Williams and
Houghton 2004), general violence (Violence Risk Apprai-
sal Guide, VRAG; Harris et al. 1993; Quinsey et al. 2006),
and lifecourse antisociality (Hare Psychopathy Checklist,
Table 4 Interpretation of DVRAG scores in the combined samples (N = 649)
Score Category Cumulative Proportion Overall Recidivism Rate 95%CI
–10 to –9 1 .02 .14 ±.21
–8 to –5 2 .22 .24 ±.07
–4 to –1 3 .43 .34 ±.08
0 to +3 4 .63 .44 ±.09
+4 to +10 5 .81 .51 ±.09
+11 to +23 6 .97 .71 ±.09
+24 to +41 7 100 1.0 0.0
Note: Observed DVRAG scores ranged from –10 to +37
Table 5 Accuracy of several formal risk assessments (plus the PCL-R) in predicting wife assault recidivism outcome variables in samples 1 and
2 combined (N = 649)
VRAG PCL-R SARA DA DVSI
Dichotomous wife assault recidivism ROC area .67 .66 .59 .56* .61
Dichotomous wife assault recidivism .29 .29 .21 .17 .20
Number of recidivistic incidents .32 .36 .27 .24 .22
Total victim injury in recidivism .31 .37 .22 .21 .17
Number of CTS severe incidents .25 .29 .23 .22 .14
Cormier-Lang score for all recidivism .23 .26 .24 .23 .20
Note: All rows except first row are Pearson correlation coefficients (point-biserial for dichotomous recidivism), p < .01 except where noted.
* p < .05. For comparison, the ODARA and DVRAG, constructed partly on Sample 1, yielded ROC areas for dichotomous recidivism in the
combined sample of .67 and .71, respectively; and correlations of .23 to .34, and .28 to .42, respectively, with the continuous outcome variables,
all ps < .001
Table 6 Inter-relationships (Pearson correlation coefficients) among formal assessments in the combined samples (N = 649)
ODARA VRAG PCL-R SARA DA DVSI DVRAG
ODARA .53 .55 .60 .43 .52 .87
VRAG .72 .43 .21 .31 .66
PCL-R .55 .36 .34 .72
SARA .61 .53 .63
DA .36 .46
DVSI .50
DVRAG
Note: All p < .001
158 Law Hum Behav (2008) 32:150–163
123
PCL-R; Hare 2003). In this sample of 649 cases with a
correctional file, all assessments showed a significant
ability to predict wife assault recidivism as operationalized
by at least some of several inter-related outcome mea-
sures. Among these tools, however, the PCL-R showed the
most promise in improving prediction over the ODARA
alone. Multivariate analyses in the development sample
(Sample 1) supported this finding.
The 14-item Domestic Violence Risk Appraisal Guide
(DVRAG) comprises the original ODARA items (but
scored continuously rather than dichotomously) combined
with the PCL-R. Item weights were based on each con-
stituent variable’s empirical association with dichotomous
wife assault recidivism in a construction sample of perpe-
trators with a police report of a domestic assault and a
corrections file. Several tests on cases masked for outcome
suggested that the DVRAG can be expected to have good
inter-rater reliability in routine forensic practice. Its pre-
dictive accuracy corresponded to a ‘‘large’’ effect (where
‘‘large’’ is a Cohen’s d .8, ROC area = .71, and ‘‘med-
ium’’ is d .5, ROC area = .64; Rice & Harris 2005)in
predicting wife assault recidivism in the development
sample of 303 cases, and its predictive accuracy was
maintained in a cross-validation sample of 346 cases and
the combined sample of 649 cases. The DVRAG per-
formed better than the ODARA and the other formal
assessments. No further improvement was made by adding
any other formal assessment or individual variable per-
taining to childhood abuse history, juvenile delinquency,
and adult adjustment. The combined sample of 649 sub-
jects was used to compile an actuarial experience table and
a practical scoring scheme (Appendix). Together, the
ODARA and DVRAG represent not a proliferation of
instruments but a coherent system of risk assessment. For
example, a police officer can score the ODARA in time for
a bail decision and a forensic clinician or probation officer
can subsequently score the DVRAG to provide an im-
proved assessment to aid sentencing, supervision, and
treatment decisions.
Assessing the Risk of Wife Assault Recidivism
The present results regarding optimal indepth assessment
of wife assault risk closely parallel construction of the
VRAG, which consists of several well established violence
risk factors empirically combined with the PCL-R. The
VRAG has yielded a large average effect in independent
replications (Quinsey et al. 2006; www.mhcp-research.
com/ragreps), especially when scored with high reliability
and without modification (Harris and Rice 2003). Together,
this research confirms the value of actuarial methods for
combining variables to construct decision support tools,
especially risk assessments. Three published domestic
violence risk assessments tested in the present study were
not developed actuarially, but relied on existing domestic
violence literature and clinical experience, neither of which
has sufficiently taken account of criminogenic variables in
previous violence prediction research (Hilton an Harris
2005; Hilton and Simmons 2001; Quinsey et al. 2006). In
addition, items with the heaviest weights in the DVRAG
tend to be those that reflect general antisociality. Thus,
rational selection and combination of items appeared to
have inadvertently concentrated too much on domain-
specific characteristics (e.g., jealousy) when measures
associated with and predictive of general criminality and
violence (e.g., nondomestic criminal history) would have
performed as well or better in predicting domestic violence
recidivism.
This is not to say that the nonactuarial assessments did
not predict wife assault recidivism. Indeed, we believe the
present results constitute a successful replication of SARA
total scores, provide further evidence that DA scores pre-
dict wife assault recidivism, and are the first independent
replication of the DVSI. On the other hand, we observed
less variance in the prevalence of wife assault recidivism
attributable to variables specific to domestic relationships
than to enduring general antisociality. This observation
might be qualified by the discovery of domain-specific
variables that account for wife assault but were not ade-
quately measured in the present study; however, the
inability of three domestic violence risk assessments,
developed using constructs identified in the 30-year-old
domestic violence literature, to outperform either the
VRAG or the PCL-R, leaves little optimism that such
variables will be found.
Most of the tested assessments predicted wife assault
recidivism across all outcome measures (dichotomous
recidivism, number of incidents, total injury score, inci-
dents with severe violence, and Cormier-Lang score),
contrary to some expectations that different outcomes re-
quire different predictors. The outcomes were significantly
intercorrelated but were not highly collinear. We infer that
this pattern of results reflects considerable amount of
measurement error in the evaluation of wife assault recid-
ivism. Such error could increase the difficulty of predicting
the specific behavior of wife assault, compared with pre-
dicting violence in general (a similar observation has been
made with respect to sexually violent recidivism; Rice
et al. 2006). Our use of several outcome measures allowed
us to replicate the predictive ability of previously published
assessments using new outcome measures. Also, we were
able to select, for the DVRAG, the assessment (PCL-R)
that most consistently and robustly added to the predictive
ability of the existing actuarial ODARA across these out-
comes, increasing confidence in the DVRAG’s applicabil-
ity to correctional and forensic clinical settings.
Law Hum Behav (2008) 32:150–163 159
123
The superior prediction yielded by assessments not de-
signed for domestic violence risk assessment confirms a
previous finding that the VRAG and PCL-R outperformed
the SARA in predicting wife assault recidivism (Grann and
Wedin 2002). These results are consistent with the
hypothesis that attitudes and actions specific to domestic
relationships play a minor etiologic role in wife assault. In
contrast, enduring antisociality—a fundamental, cross-
situational, and qualitatively distinct characteristic—is
responsible for a disproportionate amount of violent
behavior, including wife assault. Huss and Langhinrichsen-
Rohling (2006) have also reported that a measure of psy-
chopathy was associated with subtypes of men, within a
clinical treatment sample, responsible for the most extensive
violence against partners and other victims, even though the
psychopathy scores of their sample were limited in range.
We speculate that some apparently domain-specific vari-
ables might actually be triggers for the occurrence of a
violent act. For example, high levels of psychopathy best
predict who is likely to be a repetitive wife assaulter, but a
tendency towards sexual jealousy and proprietariness might
act as a proximal trigger, predicting when such violence is
most likely. Some apparently domain-specific variables on
the ODARA might also be proxies for general antisociality,
such as prior domestic assaults as well as nondomestic
(indicating criminal versatility) and assault during preg-
nancy (indicating callousness). This theoretical perspective
requires development and testing in further research.
Limitations
Although the DVRAG reliably rank ordered wife assaulters
with respect to their risk of wife assault recidivism, the
actuarial experience tables cannot automatically be applied
to new populations without knowledge of the base rate of
recidivism in those populations (Harris and Rice in press;
Mossman 2006). The present offenders all had a correc-
tional file and were more likely to have been charged than
the unselected sample of men with a police record of wife
assault used to construct the ODARA. The present proba-
bility estimates might over-estimate the rates of police
detected recidivism in other populations, especially in the
lower categories. For example, the lowest DVRAG cate-
gory has a probability of recidivism of 14% (and a wide
confidence interval) compared with only 5% in the lowest
ODARA category. The experience tables are less likely to
over-estimate risk for wife assaulters in the correctional
system, and we believe that the most appropriate applica-
tion of the DVRAG is to such a population—for whom
indepth clinical assessment information is available.
The present study was, nevertheless, limited by its
information sources. Criminal justice (police and
correctional) archives do not permit optimal assessment of
attitudes and emotional traits and could not, in the present
study, support the coding of psychiatric symptoms. Struc-
tured interviews and psychological assessment of the per-
petrator might improve the measurement of these
characteristics and, thereby, the predictive accuracy of all
the formal assessments in this study. This limitation would
apply as much to the PCL-R as to any of the other existing
assessments, yet it was a strong predictor. There is no basis
to expect additional clinical material to alter the conclusion
that the PCL-R was the best addition to the ODARA in
predicting wife assault recidivism. We did have several
sources of information to measure recidivism (criminal re-
cord, institutional and community corrections reports, and
police reports that included victim statements); however,
only officially documented behavior was defined as recid-
ivism. This definition maximizes the utility of the DVRAG
for criminal justice contexts. It might, though, under-
represent the total number of wife assault occurrences.
The predictive accuracy of all assessments might have
been limited by some restriction of the opportunity for
recidivism relative to previous research that included some
cases with no corrections record or even charges for the
index incident (Hilton et al. 2004). More cases were incar-
cerated for wife assault or other violence than in the OD-
ARA construction sample. Wife assault recidivism was not
eliminated by this intervention, but coincident with higher
scores on all the formal assessments, it likely caused some
attenuation of predictive performance in the available fol-
low-up period. Two assessments (the VRAG and DVSI)
might also have been hampered by scoring modifications.
The VRAG exhibited reduced variability compared with
populations of generally violent offenders (Quinsey et al.
2006). The DVSI, though not scored exactly as intended,
compared favorably with the other domestic violence tools,
especially in predicting criminal charges for recidivism. For
the PCL-R, SARA, and the DA, administration procedures
were altered by coding from files without conducting
interviews. The DA is intended to assess risk of lethal as-
saults which we did not test, although its best performance
was for injury, a necessary condition for lethal assault.
In conclusion, we present the Domestic Violence Risk
Appraisal Guide, DVRAG, a new actuarial indepth risk
assessment for wife assault recidivism, including scoring
details and experience table. DVRAG scores exhibited
good inter-rater reliability, and large, cross-validated ef-
fects in the prediction of several related outcomes reflect-
ing the occurrence, frequency, and severity of wife assault
recidivism. Prospective replication of the DVRAG scored
entirely masked for recidivism is desirable, although pre-
vious research has established that its components
(ODARA and PCL-R) exhibit predictive accuracy
when scored by assessors masked to outcome (Hare 2003;
Hilton et al. 2004). The success of the ODARA and PCL-R
160 Law Hum Behav (2008) 32:150–163
123
in predicting wife assault recidivism, and the predictive
value of their statistical combination, illustrates the power
of empirical methods in the construction of assessments for
forensic professionals, and the robustness of measures of
antisociality in predicting domain-specific violence.
Acknowledgments Funding for this research was provided by the
Social Sciences and Humanities Research Council (SSHRC) of
Canada. We are indebted to Detective Superintendent K. J. Lines for
her contribution to the development of the Ontario Domestic Assault
Risk Assessment and the collaborative research of the Ontario Pro-
vincial Police and Mental Health Centre Penetanguishene. We thank
the Ontario Ministry of Community Safety and Correctional Services
(MCSCS), Peel Regional Police, and York Regional Police for per-
mission to access information. In particular we extend our apprecia-
tion to the following for assistance accessing and managing file
information: Tina Gaspardy, Kathy Underhill, Greg Brown, and staff
of MCSCS Archives and Probation and Parole offices throughout
Ontario; Detective E. Gale and Detective A. Clewer of Peel Regional
Police; Detective Inspector K. Noakes of York Regional Police;
Constable C. Daunt of the Ontario Provincial Police (OPP), and the
OPP Behavioural Sciences Section data entry and information tech-
nology personnel. We also thank Catherine Cormier, Carol Lang,
Joseph Camilleri, Sonja Dey, Leslie Belchamber, Marnie Foster, Julie
McKay, and Kelly Rawson for research and administrative assistance,
and Matthew Huss and Michael Seto for helpful comments on earlier
versions of this article.
Appendix: Scoring the Domestic Violence Risk
Appraisal Guide (DVRAG)
Full scoring criteria are available from authors on request
except where noted.
1. Number of prior domestic incidents (assault on a
current or previous female cohabiting partner or her
children, recorded in a police occurrence report or
criminal record)
0= 1
1=0
2= +5
2. Number of prior nondomestic incidents (assault on
any person other than a current or previous female
cohabiting partner or her children, recorded in a po-
lice occurrence report or criminal record)
0= 1
1= +5
3. Prior correctional sentence of 30 days or more
No = 1
Yes = + 2
4. Failure on prior conditional release
No = 1
Yes = + 2
5. Threat to harm or kill at the index incident (threat of
physical harm made towards any person other than
himself)
No = 0
Yes = + 1
6. Confinement at the index incident (any attempt to
physically prevent the female victim from leaving the
scene of the incident)
No = 0
Yes = + 1
7. Victim concern (concern, fear, worry, or certainty
about possible future domestic assault, stated at the
time of the index incident)
No = 0
Yes = + 2
8. Number of children
1= 1
2= +1
9. Victim’s number of biological children from a pre-
vious partner
0= 1
1=0
2= +2
10. Violence against others (any assault on any person
other than a current or previous female cohabiting
partner or her children)
No = 0
Yes = + 8
11. Substance abuse score
One point is allotted for each of the following: alcohol
involved in the index incident, drugs involved in the index
incident, alcohol or drug abuse in days/weeks prior to
index incident, increased drug or alcohol use in days/
weeks prior to index incident, more angry or violent when
Law Hum Behav (2008) 32:150–163 161
123
using drugs or alcohol, alcohol involved in a prior criminal
offense, adult alcohol problem, adult drug problem.
1= 2
2= +2
12. Assault on victim when pregnant (index assault or
prior)
No = 0
Yes = + 5
13. Number of barriers to victim support
One point is allotted for each of the following: victim
has children aged £ 18 to care for; victim has no
telephone or transportation; victim is isolated geograph-
ically or from community; victim alcohol use in the
index incident or victim adult alcohol or drug problem.
0= 1
1=0
2= +4
14. Psychopathy Checklist-Revised Score (full scoring
criteria available in Hare, 2003)
9= 1
10 - 16 = + 1
17 = + 6
References
Ægisdo
´
ttir, S., White, M. J., Spengler, P. M., Maugherman, A. S., &
Anderson, L. A. et al. (2006). The meta-analysis of clinical
judgment project: Fifty-six years of accumulated research on
clinical versus statistical prediction. The Counseling Psycholo-
gist, 34, 383–390.
Campbell, J. C. (1986). Nursing assessment for risk of homicide with
battered women. Advances in Nursing Science, 8, 36–51.
Campbell, J. C. (1995). Prediction of homicide of and by battered
women. In J. C. Campbell (Ed.), Assessing dangerousness (pp.
96–113). Thousand Oaks, CA: Sage.
Glover, A. J. J., Nicholson, D. E., Hemmati, T., Bernfeld, G. A., &
Quinsey, V. L. (2002). A comparison of predictors of general
and violent recidivism among high risk federal offenders.
Criminal Justice and Behavior, 29, 235–249.
Goodman, L. A., Dutton, M. A., & Bennett, L. (2000). Predicting
repeat abuse among arrested batterers. Journal of Interpersonal
Violence, 15, 63–74.
Grann, M., & Wedin, I. (2002). Risk factors for recidivism among
spousal assault and spousal homicide offenders. Psychology,
Crime, and Law, 8, 5–23.
Grove, W. M., Zald, D. H., Lebow, B. S., Snitz, B. E., & Nelson, C.
(2000). Clinical versus mechanical prediction: A meta-analysis.
Psychological Assessment, 12, 19–30.
Hare, R. D. (2003). Hare psychopathy checklist-revised (2nd ed.).
Toronto, ON: Multi-Health Systems Inc.
Harris, G. T., & Rice, M. E. (2003). Actuarial assessment of risk
among sex offenders. Annals of the New York Academy of
Sciences, 989, 198–210.
Harris, G. T., & Rice, M. E. (in press). Characterizing the value of
actuarial violence risk assessment. Criminal Justice and Behav-
ior.
Harris, G. T., Rice, M. E., & Camilleri, J. A. (2004). Applying a
forensic actuarial assessment (the Violence Risk Appraisal
Guide) to nonforensic patients. Journal of Interpersonal Vio-
lence, 19, 1063–1074.
Harris, G. T., Rice, M. E., & Cormier, C. A. (2002). Prospective
replication of the Violence Risk Appraisal Guide in predicting
violent recidivism among forensic patients. Law and Human
Behavior, 26, 377–394.
Harris, G. T., Rice, M. E., & Quinsey, V. L. (1993). Violent
recidivism of mentally disordered offenders. Criminal Justice
and Behavior, 20, 315–335.
Harris, G. T., Skilling, T. A., & Rice, M. E. (2001). The construct of
psychopathy. In M. Tonry & N. Morris (Eds.), Crime and
justice: An annual review of research (pp. 197–264). Chicago:
University of Chicago Press.
Heckert, D.A., & Gondolf, E.W. (2004). Battered women’s
perceptions of risk versus risk factors and instruments in
predicting repeat reassault. Journal of Interpersonal Violence,
19, 778–800.
Hilton, N. Z., & Harris, G. T. (2005). Predicting wife assault: A
critical review and implications for policy and practice. Trauma,
Violence, and Abuse, 6, 3–23.
Hilton, N. Z., Harris, G. T., & Rice, M. E. (2001). Predicting violence
by serious wife assaulters. Journal of Interpersonal Violence, 16,
408–423.
Hilton, N. Z., Harris, G. T., & Rice, M. E. (in press). The effect of
arrest on wife assault recidivism, controlling for pre-arrest risk.
Criminal Justice and Behavior.
Hilton, N. Z., Harris, G. T., Rice, M. E., Lang, C., Cormier, C. A., &
Lines, K. J. (2004). A brief actuarial assessment for the
prediction of wife assault recidivism: The Ontario Domestic
Assault Risk Assessment. Psychological Assessment, 16, 267–
275.
Hilton, N. Z., & Simmons, J. L. (2001). The influence of actuarial risk
assessment in clinical judgments and tribunal decisions about
mentally disordered offenders in maximum security. Law and
Human Behavior, 25, 393–408.
Huss, M. T., & Langhinrichsen-Rohling, J. (2000). Identification of
the psychopathic batterer: The clinical, legal, and policy
implications. Aggression and Violent Behavior, 5, 403–422.
Huss, M. T., & Langhinrichsen-Rohling, J. (2006). Assessing
the generalization of psychopathy in a clinical sample of
domestic violence perpetrators. Law and Human Behavior, 30,
571–586.
Johnson, R., Gilchrist, E., Beech, A. R., Weston, S., Takriti, R., &
Freeman, R. (2006). A psychometric typology of U.K. domestic
violence offenders. Journal of Interpersonal Violence, 21, 1270–
1285.
Kanji, G. K. (1993). 100 statistical tests. London: Sage.
Koziol-McLain, J., Webster, D., McFarlane, J., Block, C. R., Ulrich,
Y., & Glass, N. et al. (2006). Risk factors for femicide-suicide in
abusive relationships: Results from a multisite case control
study. Violence and Victims, 21, 3–21.
Kropp, P. R., & Hart, S. D. (2000). The Spousal Assault Risk
Assessment (SARA) Guide: Reliability and validity in adult
male offenders. Law and Human Behavior, 24, 101–118.
Kropp, P. R., Hart, S. D., Webster, C. D., & Eaves, D. (1999). Spousal
assault risk assessment guide. NY: Multi-Health Systems Inc.
162 Law Hum Behav (2008) 32:150–163
123
Mossman, D. (2006). Another look at interpreting risk categories.
Sexual Abuse: A Journal of Research and Treatment, 18, 41–63.
Nunnally, J. C. (1959). Tests and measurements: Assessment and
prediction. New York: McGraw-Hill.
Popham, S., & Hilton, N. Z. (2006, October). The Ontario Domestic
Assault Risk Assessment (ODARA): Development and applica-
tion to corrections. Paper presented at the biennial conference of
the Ontario Ministry of Correctional Services and Community
Safety, North Bay, ON.
Quinsey, V. L., Harris, G. T., Rice, M. E., & Cormier, C. A. (2006).
Violent offenders: Appraising and managing risk. 2nd Ed.
Washington, DC: American Psychological Association.
Rice, M. E., & Harris, G. T. (2005). Comparing effect sizes in follow-
up studies: ROC, Cohen’s d and r. Law and Human Behavior,
29, 615–620.
Rice, M. E., Harris, G. T., Lang, C., & Cormier, C. A. (2006). Violent
sex offenses: How are they best measured from official records?
Law and Human Behavior, 30, 525–541.
Seto, M. (2005). Is more better? Combining actuarial risk scales to
predict recidivism among adult sex offenders. Psychological
Assessment, 17, 156–167.
Spidel, A., Vincent, G., Huss, M. T., Winters, J., Thomas, L., &
Dutton, D. (2007). The psychopathic batterer: Subtyping perpe-
trators of domestic violence. In H. Herve, & J. C. Yuille (Eds.),
The psychopath: Theory, research, and practice (pp. 327–342).
Mahwah, NJ: Lawrence Erlbaum.
Straus, M. A., Hamby, S. L., Boney-McCoy, S., & Sugarman, D. B.
(1996). The Revised Conflict Tactics Scales (CTS2). Journal of
Family Issues, 17, 283–316.
Weisz, A. N., Tolman, R. M., & Saunders, D. G. (2000). Assessing
the risk of severe domestic violence. Journal of Interpersonal
Violence, 15, 75–90.
Williams, K. R., & Grant, S. R. (2006). Empirically examining the
risk of intimate partner violence: The revised Domestic Violence
Screening Instrument (DVSI-R). Public Health Reports, 131,
400–408.
Williams, K. R., & Houghton, A. B. (2004). Assessing the risk of
domestic violence reoffending: A validation study. Law and
Human Behavior, 28, 437–455.
Law Hum Behav (2008) 32:150–163 163
123
... The total score can range from 0 to 13 and related percentiles reflect individuals' overall risk relative to other men with an IPV offense record. The ODARA predicts IPV recidivism with receiver-operating characteristic areas under the curve (AUC) ranging from .65 to .74 (Hilton & Harris, 2009;Hilton et al., 2008) and an average AUC of .67 to .69 (Messing & Thaller, 2013;van der Put et al., 2019). Recent studies validated the ODARA in racially diverse samples (Hegel et al., 2022;Radatz & Hilton, 2022). ...
... These results are generally consistent with the existing research highlighting the importance of criminal history and general antisociality as predictors of future offending, regardless of offense type (e.g., sexual offending, Barbaree et al., 2006;IPV, Hilton, 2021). This is also consistent with research demonstrating that measures of general criminality make a unique contribution to models of IPV recidivism; to wit, the addition of the Psychopathy Checklist-Revised (PCL-R; Hare, 2003) provides incremental predictive validity over the ODARA and has been considered especially useful for assessing relatively high-risk IPV offenders (Hilton et al., 2008). ...
Article
It is unknown whether existing intimate partner violence (IPV) risk assessment tools can predict recidivism within threat assessment samples. We examined the predictive validity for IPV, any violent, and general recidivism of four commonly used IPV risk appraisal tools (Ontario Domestic Assault Risk Assessment [ODARA], Spousal Assault Risk Assessment version 2 [SARA-V2], SARA version 3 [SARA-V3], and Brief Spousal Assault Form for the Evaluation of Risk [B-SAFER]) with 247 men charged with IPV and referred to a threat assessment service. Total scores of the ODARA and SARA-V2—but not SARA-V3 or B-SAFER—significantly predicted IPV recidivism and any violent recidivism. The SARA-V2 Criminal History subscale and the B-SAFER subscale of “Past” events—but no other subscales of the SARA-V2, B-SAFER, or SARA-V3—significantly predicted IPV recidivism. Although effect sizes were smaller than in past research, our results support the use of the ODARA and SARA-V2 with threat assessment IPV populations.
... Studies have shown that the ODARA can be reliably coded from archival data without a victim interview (Hilton, 2021). The measure has been found to have excellent interrater reliability, good internal consistency, and strong evidence of convergent validity (Hilton, 2021;Hilton et al., 2008). For the present study, the ODARA total score was used as a distal outcome to assess the relationship between individuals' identified class membership and actuarial risk score. ...
... A recent study by Hegel et al. [40] examined a predominantly Indigenous sample of individuals who had physically or sexually offended against their intimate partners and demonstrated that the ODARA was capable of predicting IPV recidivism at better than chance levels (AUC = 0.58-0.67). Similarly, Rettenberger and Eher [44] found that the ODARA and Domestic Violence Risk Appraisal Guide (DVRAG) [45] both accurately predicted domestic violence recidivism in sexually motivated IPV perpetrators (AUC = 0.71), as well as violent and general criminal recidivism. In terms of other validated risk assessment tools, the Spousal Assault Risk Assessment (SARA) [46] was able to predict IPV at levels greater than chance (AUC = 0.74) in those who had an index offense involving a severe or sexual assault against their intimate partners [47]. ...
Article
Full-text available
Purpose of Review The literature on sexual violence perpetration against an intimate partner is reviewed and synthesized. Intimate partner sexual violence (IPSV) is compared to other forms of interpersonal violence, and the heterogeneity of perpetrators is explored through an examination of proposed taxonomies. This review also addresses the applicability of existing risk assessment tools to IPSV perpetrators and identifies criminogenic needs of particular relevance to IPSV. Recent Findings Recent research suggests the perpetration of IPSV is heterogeneous in nature, and individuals who perpetrate IPSV present with criminogenic needs consistent with exclusively sexual and violent offenders. There is support for existing risk tools to predict recidivism, but they may not encompass all relevant risk factors specific to IPSV offending. Summary Commonality exists with other forms of interpersonal violence; however, the literature indicates that IPSV is complex and does not wholly resemble other sexual or violent offenses. Although further study is needed to fully understand IPSV perpetration, best practices in risk assessment and rehabilitation should employ tools that capture criminogenic needs.
... One of the main concerns of law enforcement agencies is to ensure the protection of victims who report their aggressors and to prevent a violent episode from occurring again (González-Álvarez et al., 2018). In this regard, several standardized measures that facilitate risk classification according to the statistical likelihood of recidivism (new reported IPVAW incident) have been developed in the international context for IPVAW offenders (e.g., the Brief Spousal Assault Form for the Evaluation of Risk version 2 [B-SAFER], Kropp et al., 1995; the Ontario Domestic Assault Risk Assessment [ODARA], Hilton et al., 2008; the Spousal Abuse Risk Assessment-Version 3 [SARA-V3], Kropp & Hart, 2015; and the Violence Risk Screening−Police Version [V-RISK-POL], Roaldset et al., 2017). Recent meta-analytical evidence supports the predictive value of such tools, especially actuarial instruments (overall AUC = 0.657; van der Put et al., 2019). ...
Article
Spanish intimate partner violence against women offender types (i.e., high instability/high antisociality, HiHa; low instability/high antisociality, LiHa; high instability/low antisociality, HiLa; low instability/low antisociality, LiLa) were matched with their police recidivism outcomes in a longitudinal study of 9,672 cases. Our goal was to examine whether these subtypes differed in (1) their recidivism rates, (2) the severity of the new violent episodes, and (3) the evolution of their risk levels. Results showed that individuals with high antisociality features where associated with the highest recidivism rates (26.5% HiHa; 22.6% LiHa), and higher likelihood of new severe violent episodes. HiHa offenders showed the highest risk over time, although the risk posed by all subtypes decreased during follow-up. Implications for police work are discussed.
... For example, Messing and Thaler (2013) (Hilton et al., 2008). Relatedly, a previous review indicated that assessments were administered incorrectly in the majority of observed cases, involving issues such as substitution or omission of items and misapplication of tests to specific settings (Messing & Thaler, 2013 A secondary aim of the study was to explore the role of variables that pertain to an offender's DVspecific history, as compared to more general characteristics and criminal history, in the prediction of DV recidivism. ...
Technical Report
Full-text available
Aim To develop an automated risk assessment tool that can be used to rapidly estimate custody-based DV offenders' likelihood of DV recidivism, using official administrative data that are routinely collected by Corrective Services NSW. Method The model development sample included all males in NSW who served a custodial sentence associated with one or more DV convictions between January 2013 and June 2017 (n = 6,100). A series of regression models were used to test predictors of DV recidivism and develop optimal estimates of recidivism probability. The final model was validated using bootstrapping techniques and various tests of predictive validity. Results Significant predictors of DV recidivism included age; alcohol and other drug problems; markers of a more extensive general criminal history; being released without the possibility of parole; and Indigenous status, as well as DV-specific variables such as prior sentences involving DV convictions and breaches of protection orders. The final estimation model, which we named the DV-TRAS, showed acceptable discrimination performance for DV recidivism (AUC = .660; 95% CI = .646-.675) that was significantly better than routine assessments of general recidivism risk. Bootstrapping techniques indicated satisfactory stability of the model across simulated samples. Conclusion The DV-TRAS appears to be a viable tool to support case management decision making for custody-based DV offenders in NSW. Key tests of predictive validity indicated accuracy in discriminating DV recidivists that was significantly better than general risk assessments, and similar to that of established manual assessments of offence-specific risk, while allowing for substantial time and resource savings in generating estimates.
... Several definitions of risk have existed in the literature as well as in practice. For instance, researchers have defined risk in terms of perpetrators' likelihood of committing DV, or victims' likelihood of experiencing violence in the future (Belfrage et al., 2012;Campbell et al., 2009;Eke et al., 2011;Hilton & Harris, 2009;Hilton et al., 2008;Messing et al., 2015;Nicholls, et al., 2013). However, there is no consistent view of what constitutes DV risk assessment. ...
Article
Assessing risk in domestic violence situations is foundational to ensuring safety. Although there is growing information about the reliability and validity of a variety of risk assessment tools across different practice contexts, there is a paucity of research on the feasibility and application of these tools in real world settings. The present qualitative study examined current practices in domestic violence risk assessment in Canada through a survey of professionals working across diverse sectors. Utilizing a thematic analysis of 255 open-text responses, this study presents several themes related to challenges identified at the systemic, organizational, and individual levels. Themes related to promising practices and the practical implication of risk assessment strategies are also explored.
Chapter
The rise in terrorist acts around the world has called for greater attention for health care professionals to predict and report those who may commit such violence. This focus raises significant ethical questions regarding the certainty with which such predictions can be made and concurrently, problems regarding breaches of confidentiality. In some jurisdictions, we see the increasing attention paid to mandatory disclosure when there is a reasonable suspicion of harm to self or others, even before an act of terror has occurred. The questions that need to be explored are (1) whether it is possible to make an accurate prediction of future use of violence; (2) if so, what are the potential outcomes of such disclosures; and (3) what are the ethical concerns for the health care profession in general as well as the individual professional? The European Union has passed laws under the collective heading of “Prevent” to identify those who have the potential to become part of a terrorist cell. Does such mandatory reporting actually stop and prevent terrorist acts from occurring, or, in a broader sense, does mandatory reporting actually reduce the potential for violence. To answer these questions, we will examine the ability to predict who will commit future violence using scientific research on risk assessment. Prediction of future violence using clinical interview methods alone has been found to be inaccurate more often than accurate prompting the development of structured interviews and ‘actuarials’ using data from researching violent offenders (Shapiro and Noe in Handbook of behavioral criminology. Springer, New York, 2018). Ethical considerations of whether the science of risk assessment yields sufficiently high enough accuracy to warrant violation of the patient’s privacy are explored. Does the violation of human rights especially in those where the predictions are inaccurate create an ethical crisis in health care?
Article
Actuarial risk assessment measures are often criticized because items are typically historical and do not capture potential change. Latent variable models are used to link historical risk factors to risk domains that may be the target of intervention. Using exploratory factor analysis, we explored the latent factors of the Ontario Domestic Assault Risk Assessment (ODARA) and the extent to which factors predict general, any violent, and IPV recidivism by conducting area under the receiver operating characteristic curve (AUC). We found that the ODARA contains three factors, which could be best attributed as antisocial patterns, victim vulnerabilities, and index offense-related. Antisocial Patterns significantly predicted all outcomes, whereas Victim Vulnerabilities only predicted general reoffending, and Index Offense did not reliably predict any of the recidivism outcomes. Moreover, Antisocial Patterns predicted all recidivism outcomes as well as the ODARA total. Additionally, Antisocial Patterns was able to predict any violent and general reoffending significantly better than Victim Vulnerabilities and Index Offense. Given that only Antisocial Patterns could predict IPV recidivism, our current understanding of factors unique to IPV needs further exploration to increase understanding and conceptualization of factors most strongly associated with IPV offenses, thereby improving the assessment of risk.
Article
Full-text available
This investigation studied the interaction between seven risk factors included in the police risk assessment of the VioGén System and found that these factors formed groups based on the dimensions of violence and psychopathology. The 171 femicides analysed were categorised into four groups: normalised (23.4%), violent (25.7%), pathological (18.7%), and pathological/violent (32.2%). These groups exhibited significant differences concerning their psychosocial profile and relationship dynamics. One of the main findings is the identification of the pathological type that had not been detected in previous typologies, thus highlighting the importance of the psychological factor when classifying the perpetrators of femicide. These results have important practical implications, as the classification of the aggressor could be a preliminary step taken before the risk assessment, which would make it possible to individualise predictions and improve the protection of the victims as well as the therapies and intervention programmes.
Article
Family abuse is a recurrent phenomenon within a select population of households. This form of abuse can include any physical or psychological harassment that occurs between family or household members, and often involves complex mental and emotional issues that are resistant to intervention. Traditional criminal justice strategies for combating this issue have evolved over time but have frequently demonstrated limited success. Within the past few decades, multiagency programs to address repeat family abuse have gained popularity. One such program, termed “second responders,” teams police officers with social service workers, victim advocates, or counselors to conduct follow‐up visits with victims of family abuse following a complaint. Second responders seek to educate victims about the cyclical nature of family abuse, engage in safety planning, and/or provide service referrals. These interventions are based on the premise that victims are more likely to be receptive to crime prevention opportunities immediately following victimization. Second responder interventions have received support from the US Department of Justice and their adoption has spread in both the United States and internationally, however, there remains little conclusive evidence on their effects. To update and extend the findings of the prior second responders systematic review and meta‐analysis by synthesizing the results of published and unpublished second responder evaluations through October of 2021. This review also examines the use of victim services as a secondary outcome and incorporates a number of additional moderator analyses. The Global Policing Database (GPD), a repository of all experimental and quasi‐experimental evaluations of policing interventions conducted since 1950, was searched using keywords related to second responder interventions and repeat family violence from 2004 to December 2019 (https://gpd.uq.edu.au/s/gpd/page/about). This search was also supplemented with additional strategies, such as reference harvesting of prior reviews, searching 2020 and 2021 volumes of leading academic journals, reviewing the reference lists of eligible studies, searching additional gray literature repositories focused on domestic violence, and consulting with eligible study authors. Eligible studies were required to include a treatment group that received the second responder intervention and a comparison group that did not. Assignment to these conditions could be either experimental or quasi‐experimental, but quasi‐experimental studies were required to use either matched comparison groups or multivariate analysis methods to control for confounding factors. Eligibility was limited to studies reporting on at least one measure of repeat family abuse, such as intimate partner violence, elder abuse, or general family abuse. Measures of repeat abuse could be based on either official (i.e., police data) or unofficial (i.e., victim survey data) data sources. Five new studies were identified between 2004 and 2019, all of which contained sufficient data for the calculation of at least one effect size. Along with the 10 studies included in the prior review, a total of 15 studies and 29 distinct effect sizes were analyzed across three outcome constructs. Effect sizes were calculated as logged odds ratios and results were synthesized using random effects models with restricted maximum likelihood estimation. Final results were exponentiated to represent the percentage point difference in the odds of a given outcome for treatment groups relative to control groups. Risk of bias was assessed using items adapted from the Cochrane Risk of Bias tools for experimental and quasi‐experimental studies. Eligible studies were generally considered to be of low risk of bias, however, issues with survey success/contact rates and the analytical approaches to these problems led to concern in several studies. These analyses suggest that second responder interventions produced no significant effects on either police or victim‐reported measures of repeat family abuse, in aggregate. However, findings from the more rigorous experimental studies indicated that second responder interventions were associated with a statistically significant 22% (95% confidence interval [CI] [1.04, 1.43]) increase in the odds of a police‐reported repeat family abuse incident, with no significant variability in individual study results. Additionally, studies that measured the use of victim services as a secondary outcome were associated with a statistically significant 9% (95% CI [1.02, 1.16]) increase in the odds of service use for treatment groups relative to control groups. Several study characteristics also proved to be important moderators of treatment effects. Increases in the speed of the second response were associated with significant decreases in the odds of a victim‐reported repeat incident, and studies that measured repeat family abuse using households were associated with significantly higher odds of a police‐reported repeat incident, compared to studies that used the same victim or victim/offender pairing more generally. Second responder interventions are undoubtedly appealing based on their logic and intentions. Yet, well‐intentioned programs with sound logic can still backfire, and the results of this updated review provide evidence that may be suggestive of a backfire effect. Even so, any firm conclusions from this review are limited by a lack of knowledge on the mechanisms operating in between the implementation of the second response intervention and the observed effects, as well as the small sample sizes involved in many analyses. While it seems clear that these programs are not producing any broad reductions in self‐reported victimization, the increase in police‐reported violence seen in experimental studies could indicate either a true increase in abuse or an increased willingness to call the police. The lack of observed impact on victim‐reported violence would suggest the latter, but without more specific measures, such conclusions should be avoided. If these results are indicative of increased reporting, however, many may consider this a desirable outcome, particularly given the often‐underreported nature of family abuse and the potential for increased reporting to lead to long‐term reductions in abuse. Furthermore, these results provide an indication that second responder programs can produce other intended effects, such as increasing the retention of victim services, and that the specific characteristics of these interventions may moderate their effects. It is unclear why elements such as the immediacy of the second response or the unit of analysis being evaluated would impact study results, but these observations are consistent with the theory that domestic violence interventions must capitalize on short windows of opportunity and create separation between victims and offenders to reduce exposure and subsequent victimization. This potential indicates a need for more research on second responder programs, but specifically research that examines these moderating characteristics and mechanisms. Even in light of this potential, second responder programs do not, on average, appear to reduce the prevalence of repeat family abuse. Given the presence of alternative (and possibly more effective) domestic violence interventions that now exist (e.g., Safe Dates, Shifting Boundaries, Green Dot, etc.), it seems that policymakers may wish to look elsewhere for efforts to reduce family abuse.
Article
Full-text available
Clinical predictions made by mental health practitioners are compared with those using statistical approaches. Sixty-seven studies were identified from a comprehensive search of 56 years of research; 92 effect sizes were derived from these studies. The overall effect of clinical versus statistical prediction showed a somewhat greater accuracy for statistical methods. The most stringent sample of studies, from which 48 effect sizes were extracted, indicated a 13% increase in accuracy using statistical versus clinical methods. Several variables influenced this overall effect. Clinical and statistical prediction accuracy varied by type of prediction, the setting in which predictor data were gathered, the type of statistical formula used, and the amount of information available to the clinicians and the formulas. Recommendations are provided about when and under what conditions counseling psychologists might use statistical formulas as well as when they can rely on clinical methods. Implications for clinical judgment research and training are discussed.
Article
Full-text available
A number of theoretical and empirical sources have proposed that a subgroup of domestically violent men exhibit more antisocial behavior, express more generalized violence, and are generally more resistant to mental health intervention than others. In a parallel literature, researchers have identified a subgroup of violent criminal offenders (i.e., psychopaths) that exhibit a number of similar characteristics to this more antisocial/generally violent group of batterers. Moreover, the offender literature on psychopathy describes the violence tendencies, physiological responses, cognitive impairments, interpersonal/affective characteristics, and treatment responsiveness of these individuals in much greater depth and breadth than the current domestic violence literature. The present article seeks to compare and contrast these two literatures, proposing that there is a subgroup of batterers that can be characterized as exhibiting significant psychopathic characteristics. The clinical, legal, and policy implications of identifying a subgroup of batterers in this manner also are explored.
Article
Full-text available
Assessing violence risk among wife assaulters is receiving increasing attention in the literature, but risk assessment tools specifically for this population are just beginning to be developed. The literature on wife assaulters suggests the importance of antisocial personality and behavior. The present study examines psychopathy; the Violence Risk Appraisal Guide (VRAG), a validated actuarial risk assessment tool for violent recidivism; and motives thought to be related to wife assault, in predicting violent recidivism among 88 men with a history of serious wife assault. Violent recidivism was lower among wife assaulters (24%) than among a larger sample of generally violent offenders (44%). Score on the Hare Psychopathy Checklist-Revised was a good predictor of subsequent violence, r = .35, and score on the VRAG was a significantly better predictor, r = .42, area under the curve (AUC) = .75. The prospects for predicting lethal wife assault and violence against specific victims are discussed.
Article
Full-text available
The accuracy of 10 risk measures in predicting general and violent recidivism among 106 federally sentenced male offenders was compared. During an average period of opportunity to reoffend of 713 days (SD = 601.38), 28 offenders recidivated nonviolently, and 34 recidivated violently. Common language effect sizes in discriminating violent recidivists from other offenders were .73 for the General Statistical Information on Recidivism–Revised and .72 for the Violence Risk Appraisal Guide. Effect sizes ranging from .58 to .68 were obtained for DSM-IV Conduct Disorder scored as a scale, the Violent Statistical Information on Recidivism–Revised, the Psychological Referral Screening Form, the Psychopathy Checklist–Revised total score and Factor 2, and the Childhood and Adolescent Taxon Scale. Effect sizes of .58 and .51 were obtained with the DSM-IV Antisocial Personality Disorder scored as a scale and the Psychopathy Checklist Factor 1, respectively.
Article
Full-text available
Multivariate techniques were used to derive and validate an actuarial instrument for the prediction of violent postrelease offenses by mentally disordered offenders. The 618 subjects were a heterogeneous group of men who had been charged with serious offenses. Approximately half had been treated in a maximum security psychiatric institution and the rest had been briefly assessed prior to imprisonment. The actuarial instrument consisted of 12 variables and significantly predicted violent outcome in each of five subgroups. The instrument's practical application and its use in clinical appraisals of dangerousness are discussed.