You are on page 1of 10

Australian Psychologist

ISSN: 0005-0067 (Print) 1742-9544 (Online) Journal homepage: https://www.tandfonline.com/loi/rapy20

Prediction of the risk of male sexual reoffending in


Australia

Alfred Allan, Deborah Dawson & Maria M. Allan

To cite this article: Alfred Allan, Deborah Dawson & Maria M. Allan (2006) Prediction of
the risk of male sexual reoffending in Australia, Australian Psychologist, 41:1, 60-68, DOI:
10.1080/00050060500391886

To link to this article: https://doi.org/10.1080/00050060500391886

Published online: 03 Feb 2007.

Submit your article to this journal

Article views: 359

View related articles

Citing articles: 6 View citing articles

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


https://www.tandfonline.com/action/journalInformation?journalCode=rapy20
Australian Psychologist, March 2006; 41(1): 60 – 68

Prediction of the risk of male sexual reoffending in Australia

ALFRED ALLAN1, DEBORAH DAWSON2, & MARIA M. ALLAN1


1
Clinical Forensic Psychology Program, Edith Cowan University, Joondalup and 2Offender Program Branch, Western
Australian Department of Justice, Perth, Western Australia, Australia

Abstract
This paper reports the findings of a retrospective study designed, primarily, to investigate the predictive accuracy of the
Rapid Risk Assessment for Sexual Offence Recidivism (RRASOR); the Static 99 and two models developed in Western
Australia, namely the Violent Offender Treatment Program Risk Assessment Scale (VOTPRAS) and the 3-Predictor model
on a Western Australian sample of violent and nonviolent sexual offenders. A secondary aim was to establish whether the
instruments are equally valid for Indigenous and non-Indigenous, and violent and nonviolent sexual offenders. The data of
538 convicted sexual offenders, who were assessed by the Sex Offender Treatment Program of the Western Australian
Department of Justice from 1987 to 2002, were used. The predictor variables were the total scores obtained for each
instrument and the outcome variable a conviction in a court for a further sexual offence and, in the case of the VOTPRAS,
also a further violent offence. In general the predictive accuracy of the 3-Predictor model was the best, followed by the Static
99, the RRASOR and last, the VOTPRAS. The study provides tentative support for the argument that risk assessment tools
that were developed overseas should not be used with Indigenous people without further research and that different
assessment tools should be developed for violent and nonviolent sexual offenders respectively.

The prediction of the risk of sexual reoffending are strict actuarial instruments that use algorithms to
behaviour is an important task for psychologists combine historical, and therefore static (unchange-
working in corrections settings. Probably the most able) items to predict sexual recidivism. The SVR-
widely used method of doing this is the unstructured 20, on the other hand, is an example of an
clinical judgement approach. In response to criticism instrument that uses a structured professional judg-
(e.g., Hanson, 1998) of the reliance on clinical ment approach that takes into account both static
expertise a number of instruments have been and dynamic (changeable) factors that have been
developed to assist psychologists undertaking risk empirically identified as predictors of sexual recidi-
assessments. Canadian research underpins most of vism in a specific population. The advantage of
the instruments available, such as the Sexual structured professional judgment instruments is that
Offender Risk Assessment Guide (SORAG; they allow psychologists to take into account current
Quinsey, Lalumiere, Rice, & Harris, 1995); Sex circumstances and clinical observations. Instead of a
Violence Risk-20 (SVR-20; Boer, Wilson, Gauthier, simple summation of risk factors, all the factors are
& Hart, 1997); Rapid Risk Assessment for Sexual considered and integrated (Boer, Hart, Kropp, &
Offence Recidivism (RRASOR; Hanson, 1997; Webster, 1997). Both types of instruments, but
Hanson & Bussiere, 1996); and the Sexual Offender especially the actuarial instruments, rely heavily on
Need Assessment Rating (SONAR; Hanson & specific normative samples, and consequently care
Harris, 2000a). A notable exception is an English should be taken when administering them in other
instrument, the Structured Anchored Clinical Judge- cultures and countries (Allan & Dawson, 2002;
ment (SACJ; Grubin, 1998) which was combined Amenta, Guy, & Edens, 2003; Lennings, 2003).
with the RRASOR to form the Static 99 (Hanson & The Offender Program Branch (OPB) has been
Thornton, 1999, 2000). Most of these instruments assessing and treating Western Australian sexual

Correspondence: A. Allan, Clinical Forensic Psychology Program, Edith Cowan University, 100 Joondalup Drive, Joondalup, WA 6027, Australia.
E-mail: a.allan@ecu.edu.au
ISSN 0005-0067 print/ISSN 1742-9544 online Ó The Australian Psychological Society Ltd
Published by Taylor & Francis
DOI: 10.1080/00050060500391886
Prediction of risk of male sexual reoffending 61

offenders since 1987. In the absence of Australian- than that of the SVR-20 (De Vogel, de Ruiter, van
developed instruments aimed at assessing the risk of Beek, & Mead, 2004). The Swedish studies (Lång-
reoffending by male sexual offenders, psychologists ström, 2004; Sjöstedt, 2002; Sjöstedt & Grann,
working for OPB use the Violent Offender Treat- 2002; Sjöstedt & Långström, 2001, 2002) indicated
ment Program Risk Assessment Scale (VOTPRAS; that both instruments show modest accuracy regard-
Ward & Dockerill, 1999). This instrument was ing sexual recidivism, with the Static 99 marginally
developed in Western Australia for the assessment outperforming the RRASOR (Table 1), a finding
of violent offenders’ eligibility to a violent offenders’ consistent with Canadian results (Barbaree, Seto,
treatment program. Other names used for this Langton, & Peacock, 2001; Hanson & Thornton,
instrument include Level of Service Need Inventory 2000; Table 1). The Långström (2004) finding of
(LOSNI); Violent Offender Risk Assessment Scale possible race differences in Sweden in the predictive
(VORAS) and Violent Offender Risk Treatment validity of the RRASOR and the Static 99, particu-
Assessment Scale (VOTRAS). To prevent confusion larly in respect of African Asian offenders (Table 1),
we decided, in consultation with one of the devel- is of particular importance to this study.
opers of the instrument (K. Howells, personal After considering the meagre literature regarding
communication, 28 March 2005), to follow Ward the validity of risk prediction instruments with
and Dockerill (1999). Little has been published Indigenous people in Canada, Långström concluded
about the VOTPRAS, but five of its seven items are that ‘‘further studies of the accuracy of risk assess-
related to previous and current criminal offences and ment instruments across immigration status and
two to alcohol and drug use. These items were ethnicity are needed’’ (p. 117).
unevenly weighted. The VOTPRAS has not been These European cross-validation studies suggest
cross-validated and the reported accuracy rates are that there may be some shrinkage in validity if the
based on a limited sample of 50 Indigenous and RRASOR and Static 99 are used outside North
152 non-Indigenous violent offenders (Ward & America. In particular the Långström (2004) find-
Dockerill, 1999; Table 1). ings imply that it would be inappropriate to use the
Although no published cross-validation studies of RRASOR and Static 99 in Australia without a
the international instruments mentioned above have thorough investigation of their validity for local
to date been undertaken in Australia, a number have populations, especially with regard to Indigenous
been undertaken in European countries in respect of Australians (Allan & Dawson, 2002, 2004), who are
the RRASOR and the Static 99, both of interest to overrepresented in the criminal justice system
this study. In a Dutch study the predictive validity of (Baker, 2001; Walker & McDonald, 1995). The
the Static 99, while still modest, was notably poorer culture of Indigenous people is notably different

Table 1. Predictive accuracy of risk assessment instruments for sexual reoffending among adult male sexual offenders in different countries

AUCa of different instruments

Study Sample RRASOR Static 99 VOTPRAS 3-P

Allan & Dawson (2002) WA Indigenous .97


Barbaree et al. (2001) Canadian .73 .68
Dempster (1998) Canadian .77
De Vogel et al. (2004) Dutch .71
Hanson (1997) Canadian .71
Hanson & Thornton (2000) Canadian and United Kingdom .68 .71
Långström (2004) Full Swedish .73 .75
Nordic .76 .76
Non-nordic European .77 .79
African Asian .48 .50
Sjöstedt & Långström (2001) Swedish .72 .76
Sjöstedt (2002) Swedish .73
Sjöstedt & Grann (2002) Swedish .73 .75
Ward & Dockerill (1999) WA Indigenous and non-Indigenousb .72 – .76c

Note. AUC ¼ area under the receiver operating characteristics curve; RRASOR ¼ Rapid Risk Assessment for Sexual Offence Recidivism;
3-P ¼ 3-Predictor model; VOTPRAS ¼ Violent Offender Treatment Program Risk Assessment Scale.
a
AUC is typically used as index in the assessment of accuracy in prediction of recidivism. This method is discussed in the Statistical Analysis
section, here it is adequate to say that Sjöstedt and Grann (2002) considered an AUC of 5.60 but 5.70 as marginal, 5.70 but 5.80 as
modest, 5.80 but 5.90 as moderate and 5.90 as high.
b
Violent offenders.
c
Violent reoffending based on varying time at risk.
62 A. Allan et al.

from that of non-Indigenous people. It cannot be good (Table 1), a finding that should be interpreted
assumed that an instrument that was developed for with great caution given that only 39 of the sample
people from a different culture would be valid in an of 109 reoffended during the study period. When
Indigenous Australian population. 92 non-Indigenous offenders (of whom 38 had
The retrospective study reported in this paper reoffended in the study period) were added to the
was undertaken to investigate the predictive accu- Indigenous group, the 3-Predictor model was still
racy of a number of instruments taking into account found to be highly accurate, leaving very little
the race of offenders. This was done as part of an variance unaccounted for. Two factors prompted
ongoing study and to assist the OPB of the Western the use of the 3-Predictor model in the present
Australian Department of Justice in identifying an study: (a) it served as a locally developed model
appropriate instrument to use to predict the risk of with which to compare the two internationally
sexual recidivism. Predictive accuracy was investi- developed instruments, and (b) it provided an
gated as a function of violent sexual and nonviolent opportunity to determine whether the same levels
sexual offending because we believe that violent and of accuracy reported for this instrument in Table 1
nonviolent sexual offenders should be treated as two could be obtained for a larger sample of Indigenous
distinct groups when their risk of sexual offending is and non-Indigenous sexual offenders.
assessed and that a clear distinction should also be
made between the risk of violent and nonviolent
sexual offending. This belief is based on previous Method
research with juvenile sexual offenders that sug-
Databases
gested that violent sexual offenders form a distinct
group that is likely to commit both sexual and Data used in this study were collected as part of a
nonsexual violent offences (e.g., homicide) later comprehensive and ongoing research project aimed
(Allan, Allan, Marshall, & Kraszlan, 2003; also see at validating and developing risk prediction instru-
Rubinstein, Yeager, Goodstein, & Lewis, 1993). ments for Australian offender populations. The
This is confirmed by the Allan and Dawson (2002) OPB’s offender files contain extensive and detailed
study and reports by psychologists working for OPB information ranging from presentence reports, victim
that they see qualitative differences between violent impact statements, judges’ comments, psychiatric
and nonviolent sexual offenders. and psychological reports to therapist notes. As part
When the study commenced OPB was considering of the project, a team of trained coders systematically
adopting the RRASOR as its instrument of choice to coded the data in the offender files and captured it in
assess the risk of sexual reoffending. The Static 99, electronic format. At the time this study was under-
which was found to have modest predictive accuracy taken two databases had been completed. The first
with sexual offenders in Canada, the United King- database contains information about 67 psychosocial
dom (Hanson & Thornton, 1999, 2000), the variables for 1838 Indigenous sexual and violent
Netherlands (De Vogel et al., 2004) and Sweden offenders. The second database contains information
(Långström, 2004; Sjöstedt, 2002; Sjöstedt & about 242 psychosocial variables for 1000 Indigen-
Grann, 2002; Sjöstedt & Långström, 2001), was ous and non-Indigenous sexual offenders. Coders
included after the commencement of data collection used definition manuals that defined each variable.
when the OPB identified it as a viable alternative to These definitions were developed with reference to
the RRASOR. As a consequence of the late inclusion the literature and each coder was trained until 85%
the data collected and analysed to evaluate this concordance was reached for each factor with all
instrument came from a smaller sample. coders. The research team worked in consultation
We also examined the predictive validity of an with an Indigenous Advisory Committee (Allan
experimental instrument, the 3-Predictor model & Dawson, 2004; Fong, Merson, Tittums,
(Allan & Dawson, 2002). This model was devel- O’Loughlin, & Yavu-Kama-Harathunian, 2002)
oped in the course of an earlier Western Australian and where the definition of a factor was deemed
retrospective study that examined the factors that culturally inappropriate by this committee, an
predict whether Indigenous male sexual offenders appropriate definition was negotiated. Some defini-
would reoffend violently and sexually, respectively. tions important for an understanding of the study are
We found that for sexual offenders (violent and briefly provided next.
nonviolent) the three factors that best predicted
sexual reoffending were all dynamic, namely un-
Definitions
realistic long-term goals, unfeasible release plans,
and poor coping skills prior to release (these Indigenous and non-Indigenous. Offenders who iden-
constructs are defined in the Method section). tified themselves as Aboriginal or Torres Strait
The predictive accuracy of this instrument was very Islanders at assessment were defined as Indigenous;
Prediction of risk of male sexual reoffending 63

offenders identifying themselves as of any other led to the assessment (index offence) ranged from
origin were defined as non-Indigenous. wilful exposure to rape. The mean age at first offence
was 21 and at time of the index offence, 31. The
Sexual offence. Any offence identified by OPB as an average time at risk for this sample was 9 years 3
offence requiring a sexual offender intervention months and during this period 179 reoffended in a
program was deemed to be a sexual offence nonviolent sexual manner; 69 in a violent sexual
irrespective of whether the person had undergone manner and 239 in a nonsexual violent manner.
such a program. Offences considered requiring such Due to the fact that a smaller number of variables
intervention included unwanted and illegal contact were coded in the first database than the second, the
of a sexual nature and those with a covert sexual sample sizes vary for the different analyses. The data
motive, such as administration of a stupefying drug of these 538 offenders remained after the exclusion
and abduction with intent. of the data of a number of offenders from the two
databases. Data of Indigenous offenders known to be
Violent and nonviolent sexual offences. Sexual of- deceased was excluded for cultural reasons. The data
fences involving assault causing physical harm to the of offenders who had been used in the development
victim were regarded as violent sexual offences; sexual of the 3-Predictor model (Allan & Dawson, 2002)
offences in the absence of assault causing physical and in a pilot study for the present study (Bell, 2002;
harm were regarded as nonviolent sexual offences. To Bell & Allan, 2002) were also not used to avoid
meet the criteria for sexual violence the act of violence inflating accuracy rates.
had to be supported by medical reports, material facts
or sentencing remarks in the relevant file.
Procedure
Time at risk. This was calculated from the senten- The four instruments (RRASOR, VOTPRAS, Static
cing date of the original offence to the sentencing or 99 and 3-Predictor model) were scored using data
incarceration date of the subsequent offence (ex- available in the databases. In order to ensure
cluding prison time). consistency one research assistant scored each case
while at least one other research assistant checked
Poor coping skills. This was coded if there was the scores. Where there was a difference of opinion
evidence on file that the offender had used alcohol or the scorers, in consultation with the senior research
other maladaptive behaviours as a coping strategy. assistant, discussed the case until consensus was
For Indigenous offenders the offender may state he reached. The senior research assistant also did a
uses one or more maladaptive behaviours to deal random audit of the scoring of cases where there was
with hurting inside (Memmott, Stacy, Chambers, & no difference of opinion. Data on recidivism was
Keys, 2001). retrieved from the relevant Western Australian and
national databases and offenders were followed up
Unfeasible release plans. This was coded when there until their last recorded conviction. When we inves-
was evidence on file that the offender did not have tigated the accuracy of prediction for all sexual
feasible or realistic release plans when he was offenders, we included all future sexual offences, but
released from prison or court, for example, where for the VOTPRAS an additional investigation was
the offender planned to return to a high-risk undertaken to determine the accuracy of prediction for
environment such as where they would be in close subsequent nonsexual violent offences. These include
contact with young children. violence with or without physical harm, but not
behaviour involving acts of a sexual nature. For the
Unrealistic long-term goals. This was coded if there separate investigations of the nonviolent and violent
was evidence on file that the offender was unable to groups in the sample we included future nonviolent
plan for the future in a realistic way. For example, sexual and violent sexual offences, respectively.
plans in respect of relationships and work (pattern of A subsequent offence that was committed while on
meaningful activity for Indigenous offenders) that bail, but adjudicated at the same time as the index
were clearly nonachievable given the offender’s offence, was classified as a reoffence.
history and circumstances.
Statistical analysis
Participants
In the current study we were concerned with how
The coded data of 538 convicted sexual offenders (347 well the total score on the various risk instruments
Indigenous and 191 non-Indigenous; 161 violent and predicted group membership (reoffenders and non-
377 nonviolent) taken from the two databases men- reoffenders). This was assessed using a classification
tioned above were used in this study. The offences that matrix generated by discriminant function analysis
64 A. Allan et al.

(DFA) that depicts the number of cases that were that all reoffenders had been incorrectly classified
classified correctly and those that were misclassified. (nonreoffenders 100% accurately classified). The
Each participant’s discriminant score (generated via AUC was .52 for nonviolent sexual offenders and .70
total scores on each instrument) was utilised to for violent sexual offenders, with all reoffenders
determine the posterior probabilities of being in each classified incorrectly and all nonreoffenders classified
of the two groups (Tabachnick & Fidell, 2001). correctly in both groups. The prediction accuracy for
Although not reported here due to a focus on the sexual offences in general was not investigated as the
comparison between the various tools and differing results were very different for nonviolent and violent
offender groups, sensitivity and specificity for various sexual offences.
cut-off points were assessed using receiver operating As would be expected of a violence prediction tool,
characteristics (ROC). when only subsequent nonsexual violent offending
Despite limitations (Amenta et al., 2003; Dolan & was considered for the total sample, an AUC of .79
Doyle, 2000) ROC analysis appears to be the was achieved. The classification accuracy was 72.2%
preferred method to investigate the predictive value for nonreoffenders and 73.8% for reoffenders.
of prediction tools (Dolan & Doyle, 2000; Gardner, The VOTPRAS was more accurate in the pre-
Lidz, Mulvey, & Shaw, 1996; Mossman, 1994a, diction of sexual reoffending for non-Indigenous
1994b, 2000; Mossman & Somoza, 1991; Rice & (AUC ¼ .53) than Indigenous offenders (AUC ¼ .48).
Harris, 1995a, 1995b, 1997; Steadman, 2000). The Classification accuracy for both groups was poor, with
ROC curve is a plot of the true positive rate all reoffenders being incorrectly classified (100%
(sensitivity) as a function of the false positive rate classification accuracy for nonreoffenders). Given
(1 minus specificity). The area under the curve the notable difference in the prediction accuracy for
(AUC) is a statistical index that represents the average violent and nonviolent sexual offenders it would have
difference in true positive and false positive rates been preferable to investigate violent and nonviolent
across all possible cut-offs. The AUC can range from Indigenous and non-Indigenous sexual offenders
0 to 1 (implying perfect performance), with an AUC separately, but the sample size was too small for
of .5 indicative of no better than random prediction. separate investigations.
In the case of the prediction of sexual reoffending, When only the prediction of subsequent nonsexual
the AUC represents the probability that a randomly violent offending was considered, the VOTPRAS
selected person who commits a new sexual offence had an AUC of .65 for Indigenous offenders and .75
has a higher score on the specific instrument than a for non-Indigenous offenders. Classification accu-
randomly selected person who does not. racy for non-Indigenous offenders was 87.8% for
reoffenders and 47.5% for nonreoffenders. For
Indigenous reoffenders the accuracy was 96.3% for
Results reoffenders and 25.9% for nonreoffenders.
VOTPRAS
RRASOR
A sample of 538 was used to investigate the
VOTPRAS. As Table 2 shows, the AUC for all In comparison to the VOTPRAS, greater rates of
sexual offenders was .53, and DFA analysis revealed accuracy were found for the RRASOR in respect to

Table 2. Predictive accuracy of risk assessment instruments for sexual reoffending among adult male sexual offenders in Western Australia

AUC of different instruments

n RRASOR VOTPRAS Static 99 3-P

All sexual offenders 538 .71 .53 .84c


Race
Non-Indigenous 191 .74 .53
Indigenous 347 .65 .48
Offence type
Nonviolent sexuala 377 .72 .52
Violent sexualb 161 .46 .70
Non-Indigenous nonviolent sexuala 144 .78

Notes. AUC ¼ area under the receiver operating characteristics curve; RRASOR ¼ Rapid Risk Assessment for Sexual Offence Recidivism;
3-P ¼ 3-Predictor model; VOTPRAS ¼ Violent Offender Treatment Program Risk Assessment Scale.
a
Prediction of nonviolent sexual reoffending.
b
Prediction of violent sexual reoffending.
c
n ¼ 226.
Prediction of risk of male sexual reoffending 65

sexual reoffending. The AUC was .71 for all sexual & Långström, 2001) and the Netherlands (De Vogel
offenders (n ¼ 538). Classification accuracy was good et al., 2004). Despite its modest AUC (.78) the Static
for nonreoffenders (92.6%) but poor for reoffenders 99 was the next most accurate instrument after the 3-
(27.1%). The results were similar for nonviolent Predictor model, but once again the findings must be
sexual offenders with an AUC of .72, and 91.6% of viewed with caution due to the small sample size and
nonreoffenders classified correctly and 31% of the fact that it consisted of only non-Indigenous
reoffenders. In contrast, the AUC was .46 for violent nonviolent sexual offenders.
sexual offenders, with only 2.9% of reoffenders As in most other studies, the Static 99 out-
classified correctly even though 98.7% of nonreof- performed the RRASOR in terms of predictive
fenders were classified correctly. accuracy (Hanson & Thornton, 2000; Långström,
For Indigenous offenders the AUC was .65 and 2004; Sjöstedt & Grann, 2002; Sjöstedt & Lång-
again classification accuracy was poor for reoffenders ström, 2001), but for a contra result see Barbaree
(nonreoffenders ¼ 95%; reoffenders ¼ 17.1%). For et al. (2001). Even with Indigenous people in the
non-Indigenous offenders the AUC was .74 and sample the predictive accuracy of the RRASOR
classification accuracy was 91.3% for nonreoffenders (AUC ¼ .71) is in accordance with that found in
and 33.6% for reoffenders. Canada (Barbaree et al., 2001; Dempster, 1998;
Hanson, 1997; Hanson & Thornton, 2000) and
Sweden (Långström, 2004; Sjöstedt, 2002; Sjöstedt
Static 99
& Grann, 2002; Sjöstedt & Långström, 2001). It was
Due to the late inclusion of the Static 99 in the study nevertheless considerably poorer in respect of the
we were unable to examine differences as a function Indigenous offenders when a discrete analysis was
of ethnicity and offence type. Therefore the following done (AUC ¼ .65).
results are for a small sample (n ¼ 144) of non- Given that the VOTPRAS was developed for
Indigenous nonviolent sexual offenders. The AUC violent offenders it is not surprising that compared
was .78 and classification accuracy was moderate for to the other instruments, its predictive accuracy was
reoffenders (52.5%), but good for nonreoffenders poor for sexual offenders. However, what is notable
(91.3%). is that its predictive accuracy was much better for
violent (AUC ¼ .70) than for nonviolent
(AUC ¼ .52) sexual offenders. The predictive accu-
3-Predictor
racy was poor for both Indigenous and non-
There were 226 offenders of the 538 in the sample Indigenous sexual offenders, but slightly better for
who had data available for the testing of the 3- the non-Indigenous group (AUC of .53 vs. .48). It
Predictor model. These offenders represented a should be noted that the predictive accuracy in
combined group of Indigenous and non-Indigenous respect of violent reoffending was better for the
offenders and nonviolent sexual and violent sexual sexual offenders in the present study than that found
offenders. Due to the small sample size discrete by Ward and Dockerill (1999) for violent offenders
analyses were not possible for the various groups. (AUC of .79 vs. .73).
The 3-Predictor model achieved high levels of While we acknowledge that more research is
accuracy (AUC ¼ .84). The classification accuracy required, the results with respect to the RRASOR
was 90% for nonreoffenders and 75% for reoffen- provide support for our belief that instruments
ders. The accuracy rate for reoffenders increased to developed in other countries should preferably not
82.5% when violent sexual offenders were removed be used with Indigenous people without a proper
from the sample. investigation. The AUC for non-Indigenous Western
Australians in this study was .74, very similar to that
found in Canada where the RRASOR was developed
Discussion
(Table 1), while it was only .65 for Indigenous
If the relative predictive accuracy of the four people in our study.
instruments is compared it is clear that the 3- The findings of this study also support the opinion
Predictor model was the best predictor of sexual of the OPB clinicians that there are qualitative
reoffending in Western Australia despite shrinkage differences between violent and nonviolent sexual
(AUC ¼ .84). However, the findings in respect of the offenders, and our own conclusion that these two
3-Predictor model should be interpreted cautiously groups require different assessment instruments
given the relatively small sample. (Allan & Dawson, 2002). The predictive accuracy
The predictive accuracy of the Static 99 is in of the RRASOR, a sexual offending risk assessment
accordance with that found in Canada (Barbaree instrument, was poor (AUC ¼ .46) for violent sexual
et al., 2001; Hanson & Thornton, 2000); Sweden offenders compared to nonviolent sexual offenders
(Långström, 2004; Sjöstedt & Grann, 2002; Sjöstedt (AUC ¼ .72). In contrast the VOTPRAS, which was
66 A. Allan et al.

developed to assess the eligibility for treatment of portant. Future research should in particular
high-risk violent offenders, was much better at investigate whether the combination of the dynamic
predicting reoffending by violent sexual offenders factors of the 3-Predictor model and the static
(AUC ¼ .70) compared to the AUC of .52 for factors measured by an instrument such as the
nonviolent sexual offenders. Static 99 would predict better than instruments
These disparate findings for violent and nonviolent based exclusively on static or dynamic factors.
sexual offenders add support to the notion that group Some of the limitations of the present study have
differences should be considered in the development, already been referred to. In particular, the relatively
and examination of the predictive accuracy, of risk small number of participants made it impossible to
assessment instruments (Miller, Amenta, & Conroy, investigate the predictive accuracy of all instruments
2005). Despite earlier findings that there are no separately for Indigenous and non-Indigenous offen-
offender type differences in actuarial prediction of ders, and violent and nonviolent sexual offenders.
sexual recidivism (Hanson & Thornton, 2000; Other limitations common to studies in which data
Sjöstedt & Långström, 2001), Bartosh, Garby, are collected retrospectively from official records
Lewis, and Gray (2003) found that the effectiveness were that there were missing data and that the
of the four instruments they examined were related accuracy of some records was uncertain. This was
to offender type. Although our study does not make dealt with by comparing our data with that from
the more traditional distinction between different other official sources where possible. It is also
types of sexual offenders, it does indicate that further impossible to be certain how representative the
research is warranted examining possible group sample is of the total population of sexual offenders
differences. in Western Australia because the use of convictions
The relative superiority of the VOTPRAS for as recorded in official records is notoriously unreli-
violent sexual offenders in this study also suggests able (Friendship & Thornton, 2001; Mayhew, Elliot,
that other instruments used to assess the risk of future & Dowds, 1989; Taylor, 1999). To compound the
violent behaviour, such as the Psychopathy Check- problem in the present study, not all sexual offenders
list – Revised (PCL-R; Hare, 1991), may be more convicted by the courts are assessed by OPB and
accurate in predicting the risk of violent sexual therefore the data of some offenders may not be
reoffending than instruments such as the RRASOR, included in the databases we used. It should also be
developed specifically to assess the risk of sexual noted that because Australian Indigenous people
offending in general. We are aware of findings that the come from a large number of distinct language, skin
PCL-R on its own is not a good predictor of sexual and geographical subgroups the findings of the
recidivism in general (e.g., Barbaree et al., 2001), but present study cannot be generalised to all Australian
it is well-established that violent sexual offenders are Indigenous people.
more likely to be convicted of a new nonsexual violent Nevertheless, the present study provides support
offence than a new sexual offence (Allan et al., 2003; for the contention that risk assessment instruments
Hildebrand, De Ruiter, & De Vogel, 2004; Prentky, developed in other countries and cultures should
Lee, Knight, & Cerce, 1997; Rubinstein et al., 1993; not be applied in Australia, and particularly with
Sjöstedt & Långström, 2002). Indigenous people, without proper investigation.
It is notable that the 3-Predictor model is entirely The study further demonstrated that it is possible to
based on dynamic risk factors. This is in sharp develop an instrument specifically for an Australian
contrast with the Hanson and Bussiere (1998) population of Indigenous and non-Indigenous sex-
finding, when they reviewed 61 sexual offending ual offenders that is more accurate than other
studies, that long-term recidivism was best pre- instruments currently in use. It further confirms
dicted by static factors. However, Hanson and that instruments consisting of dynamic factors are
Harris (2000b) found that even after controlling useful predictors of risk, and suggests that dynamic
for pre-existing differences in static risk factors, factors may play an important role in the sexual
dynamic factors continued to be strongly associated reoffending of sexual offenders. Finally, it supports
with recidivism. Furthermore, studies on nonsexual our belief that it is necessary to develop different
offenders have found that dynamic risk factors risk prediction instruments for violent and
predict recidivism at least as well as static variables nonviolent sexual offenders.
(Gendreau, Little, & Goggom, 1996; Zamble &
Quinsey, 1997). Given that the 3-Predictor model is
Acknowledgments
still under development, our finding that the three
best predictors of sexual reoffending were all This research was supported by an Edith Cowan
dynamic should be treated with caution. Never- University and Western Australian Department of
theless it indicates that research in respect of the Justice collaborative grant. However, the views
dynamic risk factors for sexual offenders is im- expressed in this paper and inferences made are
Prediction of risk of male sexual reoffending 67

those of the authors and do not necessarily reflect the Fong, N., Merson, N., Tittums, J., O’Loughlin, H., &
official opinion of either Edith Cowan University or Yavu-Kama-Harathunian, C. (2002). Indigenous advisors: A
‘‘blueprint’’ for the future. Paper presented at the Probation and
the Department of Justice. The authors acknowledge Community Corrections: Making the Community Safer,
the contribution of two anonymous reviewers, Perth, Western Australia, 23 – 24 September, 2002.
Fran Bell, Mary Edwards, Peter Marshall, Ranjini Friendship, C., & Thornton, D. (2001). Sexual reconviction for
Sathyanath and the student research assistants work- sexual offenders discharged from prison in England and Wales.
British Journal of Criminology, 41, 285 – 292.
ing on the Offender Program Edith Cowan research
Gardner, W., Lidz, C. W., Mulvey, E. P., & Shaw, E. C. (1996).
project. Clinical versus actuarial predictions of violence in patients with
mental illness. Journal of Consulting and Clinical Psychology, 64,
602 – 609.
References Gendreau, P., Little, T., & Goggom, C. E. (1996). A meta-
analysis of the predictors of adult offender recidivism: What
Allan, A., Allan, M. M., Marshall, P., & Kraszlan, K. (2003). works! Criminology, 34, 575 – 607.
Recidivism among male juvenile sexual offenders in Western Grubin, D. (1998). Sex offending against children: Understanding the
Australia. Psychiatry, Psychology and Law, 10, 359 – 378. risk (Police research series paper No. 99). London: Home
Allan, A., & Dawson, D. (2002). Developing a unique risk of violence Office.
tool for Australian Indigenous offenders (No. CRC 6/00-01). Hanson, R. K. (1997). The development of a brief actuarial scale for
Canberra: Criminology Research Council. sexual offence recidivism (User report No. 97 – 04). Ottawa:
Allan, A., & Dawson, D. (2004). Assessment of the risk of reoffending Ministry of the Solicitor General of Canada.
by Indigenous male violent and sexual offenders (Trends and issues Hanson, R. K. (1998). What do we know about sexual
in crime and criminal justice No. 280). Canberra: Australian offender risk assessment? Psychology, Public Policy, and Law,
Institute of Criminology. 4, 50 – 72.
Amenta, A. E., Guy, L. S., & Edens, J. F. (2003). Sex offender risk Hanson, R. K., & Bussiere, M. T. (1996). Predictors of sexual
assessment: A cautionary note regarding measures attempting offender recidivism: A meta-analysis. Retrieved 24 March, 1997,
to quantify violence risk. Journal of Forensic Psychology Practice, from http://www.sgc.gc.ca
3, 39 – 50. Hanson, R. K., & Bussiere, M. T. (1998). Predicting relapse: A
Baker, J. (2001). The scope for reducing indigenous imprisonment rates meta-analysis of sexual offender recidivism studies. Journal of
(Contemporary Issues in Crime and Justice No. 55). Sydney: Consulting and Clinical Psychology, 66, 348 – 362.
New South Wales Bureau of Crime Statistics and Research. Hanson, R. K., & Harris, A. (2000a). The sex offender need
Barbaree, H. E., Seto, M. C., Langton, C. M., & Peacock, E. J. assessment rating (Sonar): A method for measuring change in risk
(2001). Evaluating the predictive accuracy of six risk instru- levels. Retrieved 10 April, 2000, from http://www.sgc.gc.ca
ments for adult sex offenders. Criminal Justice and Behavior, 28, Hanson, R. K., & Harris, A. J. R. (2000b). Where should we
490 – 521. intervene? Dynamic predictors of sexual offense recidivism.
Bartosh, D. L., Garby, T., Lewis, D., & Gray, S. (2003). Criminal Justice and Behavior, 27, 6 – 15.
Differences in the predictive validity of actuarial risk assess- Hanson, R. K., & Thornton, D. (1999). Static 99: Improving risk
ment in relation to sex offender type. International Journal of assessments for sex offenders. Retrieved 16 August, 2002, from
Offender Therapy and Comparative Criminology, 47, 422 – 438. http://www.sgc.gc.ca
Bell, F. A. (2002). Sexual and violent recidivism in Western Hanson, R. K., & Thornton, D. (2000). Improving risk assess-
Australian sex offenders: Assessing the predictive accuracy of risk ments for sexual offenders: A comparison of three actuarial
assessment. Unpublished Bachelor of Arts (Psychology) Hon- scales. Law and Human Behavior, 24, 119 – 136.
ours thesis, Edith Cowan University, Perth, Western Australia. Hare, R. D. (1991). The Hare Psychopathy Checklist – Revised
Bell, F. A., & Allan, A. (2002, 23 – 24 September). Predicting sexual Manual. Toronto: Multi-Health Systems.
and violent recidivism in Western Australian sexual offenders. Paper Hildebrand, M., De Ruiter, C., & De Vogel, V. (2004).
presented at the Probation and Community Corrections: Psychopathy and sexual deviance in treated rapists: Association
Making the Community Safer Conference, Perth, Western with sexual and nonsexual recidivism. Sexual Abuse: A Journal
Australia. of Research and Treatment, 16, 1 – 24.
Boer, D. P., Hart, S. D., Kropp, P. R., & Webster, C. D. (1997). Långström, N. (2004). Accuracy of actuarial procedures for
Manual for the Sexual Violence Risk – 20: Professional guidelines for assessment of sexual offender recidivism risk may vary across
assessing risk of sexual violence. Vancouver: Mental Health, Law ethnicity. Sexual Abuse: A Journal of Research and Treatment, 16,
and Policy Institute Simon Fraser University. 107 – 120.
Boer, D. P., Wilson, R. J., Gauthier, C. M., & Hart, S. D. (1997). Lennings, C. J. (2003). The use of the RSV-20 in a forensic
Assessing risk of sexual violence: Guidelines for clinical sample: A research note. International Journal of Forensic
practice. In C. D. Webster, & M. A. Jackson (Eds.), Psychology, 1, 147 – 153.
Impulsivity: Theory, assessment, and treatment (pp. 327 – 342). Mayhew, P., Elliot, D., & Dowds, L. (1989). The British crime
New York: The Guilford Press. survey (Home Office Research Study No. 111). London: Home
De Vogel, V., de Ruiter, C., van Beek, D., & Mead, G. (2004). Office.
Predictive validity of the SVR-20 and Static 99 in a Dutch Memmott, P., Stacy, R., Chambers, C., & Keys, C. (2001).
sample of treated sex offenders. Law and Human Behavior, 28, Violence in Indigenous communities. Canberra: Commonwealth
235 – 251. Attorney-General’s Department.
Dempster, R. J. (1998). Prediction of sexually violent recidivism: Miller, H. A., Amenta, A. E., & Conroy, M. A. (2005). Sexually
A comparison of risk assessment instruments. Unpublished violent predator evaluations: Empirical evidence, strategies for
Master of Arts (Psychology), Simon Fraser University, professionals, and research directions. Law and Human
Vancouver. Behavior, 29, 29 – 54.
Dolan, M., & Doyle, M. (2000). Violence risk prediction. Clinical Mossman, D. (1994a). Assessing predictions of violence: Being
and actuarial measures and the role of the Psychopathy accurate about accuracy. Journal of Consulting and Clinical
Checklist. British Journal of Psychiatry, 177, 303 – 311. Psychology, 62, 783 – 792.
68 A. Allan et al.

Mossman, D. (1994b). Further comments on portraying the Sjöstedt, G., & Grann, M. (2002). Risk assessment: What
accuracy of violence prediction. Law and Human Behaviour, is being predicted by actuarial prediction instruments?
18, 587 – 593. International Journal of Forensic Mental Health, 1, 179 –
Mossman, D. (2000). Commentary: Assessing the risk of violence 183.
– Are ‘‘accurate’’ predictions useful? The Journal of the Sjöstedt, G., & Långström, N. (2001). Actuarial assessment of sex
American Academy of Psychiatry and the Law, 28, 272 – 281. offender recidivism risk: A cross-validation of the RRASOR
Mossman, D., & Somoza, E. (1991). ROC curves, test accuracy, and the Static-99 in Sweden. Law and Human Behavior, 25,
and the description of diagnostic tests. Journal of Neuropsy- 629 – 645.
chiatry and Clinical Neurosciences, 3, 330 – 333. Sjöstedt, G., & Långström, N. (2002). Assessment of risk
Prentky, R. A., Lee, A. F. S., Knight, R. A., & Cerce, D. (1997). for criminal recidivism amongst rapists: A comparison
Recidivism rates among child molesters and rapists: A metho- of four different measures. Psychology, Crime and Law, 8,
dological analysis. Law and Human Behavior, 21, 635 – 659. 25 – 40.
Quinsey, V. L., Lalumiere, M. L., Rice, M. E., & Harris, G. T. Steadman, H. J. (2000). From dangerousness to risk assessment of
(1995). Predicting sexual offences. In J. C. Campbell (Ed.), community violence: Taking stock at the turn of the century.
Assessing dangerousness: Violence by sex offenders, batterers, and Journal of the American Academy of Psychiatry and the Law, 28,
child abusers (pp. 114 – 137). Thousand Oaks, CA: Sage. 265 – 271.
Rice, M. E., & Harris, G. T. (1995a). Psychopathy, schizophrenia, Tabachnick, B. G., & Fidell, L. S. (2001). Using multivariate
alcohol abuse, and violent recidivism. International Journal of statistics (4th ed.) Boston: Allyn & Bacon.
Law & Psychiatry, 18, 333 – 342. Taylor, R. (1999). Predicting reconvictions for sexual and violent
Rice, M. E., & Harris, G. T. (1995b). Violent recidivism: offences using the revised offender group reconviction scale.
Assessing predictive validity. Journal of Consulting and Clinical (Research Findings No. 104). London: Home Office Research
Psychology, 63, 737 – 748. and Statistics Directorate.
Rice, M. E., & Harris, G. T. (1997). Cross-validation and Walker, J., & McDonald, D. (1995). The over-representation of
extension of the Violence Risk Appraisal Guide for child Indigenous people in custody in Australia (Trends and issues in
molesters and rapists. Law and Human Behavior, 21, 231 – 241. crime and criminal justice No. 47). Canberra: Australian
Rubinstein, M., Yeager, C. A., Goodstein, C., & Lewis, D. O. Institute of Criminology.
(1993). Sexually assaultive male juveniles: A follow-up. Ward, A., & Dockerill, J. (1999). The predictive accuracy of the
American Journal of Psychiatry, 150, 262 – 265. Violent Offender Treatment Program Risk Assessment Scale.
Sjöstedt, G. (2002). Violent recidivism among sexual offenders: Risk Criminal Justice and Behavior, 26, 125 – 140.
factors and assessment procedures. Unpublished Doctor of Zamble, E., & Quinsey, V. L. (1997). The criminal recidivism
Philosophy dissertation, Karolinska Institutet, Stockholm. process. New York: Cambridge University Press.

You might also like