Intimate partner violence (IPV) has profound and widespread health and economic implications at an individual, familial, and societal level. Violence risk assessment measures offer an evidence-informed approach to ascertain the degree of threat an abuser poses. Risk assessments are an essential means of informing professionals and victims alike regarding the nature and intensity of services required to help keep a victim safe as well as a transparent and defensible indicator of the rationale for intervening with an abuser (e.g., sentencing, probation conditions, required treatment). Violence risk assessment measures have been in existence for several decades and there has been a proliferation of IPV specific measures developed. However, there is little agreement in the literature with regard to the most appropriate approach (actuarial, structured clinical judgment) nor which specific measure has the strongest empirical validation behind it, leaving clinicians and policy makers with little clear guidance.
The state of knowledge regarding risk assessment for IPV was summarized through a systematic examination of all English publications from westernized nations from 1990 to 2011. Three search engines (PsychINFO, Science Direct, and Social Science Index) identified 3361 potentially relevant articles. After dropping duplicates, examining titles and abstracts and removing articles that did not explicitly examine risk assessments for IPV we were left with a total of 39 articles. Our specific objectives were to: (a) identify all IPV risk assessment measures and relevant approaches (e.g., pilot tools; women’s appraisals of their partner’s risk, the application of non-IPV specific measures to IPV populations/risk); (b) describe briefly the purpose, development and use of the various risk assessment approaches; and (c) report the state of the validation of the various measures and evaluate the psychometric properties of these diverse approaches (inter-rater reliability, convergent validity, predictive validity).
The review yielded studies reporting on the validity and reliability of eight IPV specific actuarial instruments and three general actuarial risk assessment measures. The range of area under the curve (AUC) values reported for the validity of the Ontario Domestic Assault Risk Assessment (ODARA; Hilton, et al., 2004) predicting recidivism was good to excellent (0.64 – 0.77). The single study that reported on the Domestic Violence Risk Appraisal Guide (DVRAG; Hilton, Harris, Rice, Houghton, & Eke, 2008) reported an AUC = 0.70 (p < .001). The inter-rater reliability for both instruments was excellent. The Domestic Violence Screening Inventory (DVSI, Williams & Houghton, 2004) and Domestic Violence Screening Inventory – Revised (DVSI-R; Williams & Grant, 2006) were found to be good predictors of new family violence incidents and IPV recurrence (AUC range 0.61 – 0.71). Three studies examined the Psychopathy Checklist – Revised (PCL-R; Hare, 1991, 2003) and Violence Risk Appraisal Guide (VRAG; Harris, Rice, & Quinsey, 1993; Quinsey, Harris, Rice, & Cormier, 2006), neither of which are IPV specific, reporting AUCs ranging from 0.66 – 0.71 and 0.67 - 0.75, respectively. The Level of Service Inventory – Revised (LSI-R; Andrews & Bonta, 1995, 2000, 2001) and Level of Service Inventory – Ontario Revision (LSI-OR; Andrews, Bonta, & Wormith, 1995) were discussed in four articles, reporting two AUC values of 0.50 and 0.73, both of which were predicting IPV recidivism.
Two structured professional judgment instruments were included in the review, the Spousal Assault Risk Assessment guide (SARA; Kropp, Hart, Webster, & Eaves, 1994, 1995, 1999, 2008) and the Brief Spousal Assault Form for the Evaluation of Risk (B-SAFER; Kropp, Hart, & Belfrage, 2005). The SARA research reports nine AUCs ranging from 0.52-0.65. The interrater reliability (IRR) for the SARA was excellent for total scores, good for the summary risk ratings, and poor for the critical items. Although neither of the articles examining the B-SAFER reported the predictive validity of the instrument one did report the IRR based on 12 cases with a mean interclass coefficient (ICC) of 0.57.
The current literature for the Danger Assessment (DA; Campbell, 1986; Campbell, Webster, & Glass, 2009) and unstructured Victim Appraisals do not provide a clear picture of the validity and reliability of these two approaches to ascertaining an abuser’s future risk of IPV. The DA has the largest body of literature behind it, but there are limitations in the research that inhibit a clear determination of the psychometric properties of the measure, thus far. Victim appraisals of the risk of future IPV show some evidence of predictive accuracy, even exhibiting greater predictive validity than some risk assessment instruments (Heckert & Gondolf, 2004) and adding significantly to regression models (Weisz et al., 2000); however, further research is needed to determine the best means with which to collect the victim’s reports and determining the conditions (e.g., stalking) and characteristics of victims that should be considered (e.g., PTSD, substance use).
Overall, the literature reveals moderate postdictive/predictive accuracy across measures with little evidence to support one as being highly superior to others, particularly given the heterogeneity of perpetrators and victims, study limitations, and the small body of empirical literature to date. Although lethal assault (which might reflect femicide, filicide, and/or familicide) is of greatest concern, the necessary evidentiary basis for recommending a measure to assess for risk of lethal IPV violence is highly limited (also see Bowen, 2011; Guo & Harstall, 2008; Hart & Watt, 2008).
Several themes emerged when we examined the synthesized literature: (1) There is a relatively small body of empirical evidence evaluating IPV violence risk assessment measures. (2) The need for continued advancements in the methodological rigor of the research including prospective studies, research that compares multiple measures within single studies, and research that uses large samples and appropriate outcome indicators. Particularly challenging is studying the predictive validity of measures intended to predict femicide. There also is a need for increased consistency in adhering to measure guidelines in evaluation studies (i.e., coding the measures in the manner recommended), obtaining criterion data from multiple and reliable sources and using outcome data that matches the intended use of the risk assessment measures. (3) A need to extend the investigation of the validity and reliability beyond North American borders and expand the cross-validation research to diverse samples (e.g., Gay- Lesbian –Bisexual-Transgendered; male victims/female perpetrators). (4) A particularly exciting development in IPV risk assessment research is evidence that risk assessments can serve to reduce risk levels (Belfrage et al., 2011).
In terms of clinical implications, the review demonstrates the considerable promise of several IPV risk assessment measures but generally reveals modest postdictive/predictive accuracy for most measures. Limited evidence for the superiority of actuarial vs. SPJ measures was evident. Similarly, IPVspecific risk assessment measures were not found to consistently outperform general violence risk assessment measures; however, we would recommend considerable caution in interpreting this finding given the small number of studies examining non-IPV measures and the fact that only one of those studies actually used IPV recidivism as the outcome criterion. We would assert this may largely be a reflection of poor study designs and procedures (e.g., not using the SARA in the preferred manner), particularly given meta-analyses in the risk assessment field have found context/outcome specific measure to have an advantage over non-population/offence specific measures (Singh, Grann, & Fazel, 2011). Given the challenges in comparing across studies and the heterogeneity of partner abusers it seems premature to recommend one preferred assessment measure/approach to clinicians. Victim appraisals, while the research has a considerable ways to go, were found to have clinical relevance. However, preliminary evidence suggests that clinicians may want to be particularly cautious when working with some sub-groups when taking into account victims’ perceptions (e.g., PTSD symptoms, substance use, stalking and severe abuse experienced) and supplement the woman’s input with an additional structured assessment.
When clinicians and administrators are faced with the challenge of determining which measure(s) to use to assess risk of IPV they should carefully consider the purpose of the assessment (Heilbrun, 2009). Assessors also should take into account the context, setting, and resources when evaluating which measure best suits their needs. For instance, some structured professional judgment measures (e.g., SARA and PCL-R) may be more resource intensive than most actuarial measures making them inappropriate for certain circumstances (e.g., police responders; also see Coid et al., 2009) . In addition, many of the measures considered here require extensive professional training and expertise of the evaluator (e.g., PCL-R). Finally, consideration must be given to the characteristics of the population to be assessed (e.g., age, gender, ethnicity, socio-economic status) and the extent to which a measure has been cross-validated in similar samples is required (Heilbrun,2009).When validation research and empirical evidence is limited evaluators should be particularly cautious in the interpretation of results and should make the limitations clearly evident to stakeholders in their risk assessment documentation and communications. It is also important that measures are used as intended (e.g. administering measures and conducting interviews, reviewing criminal records and clinical files); therefore, if the setting and context does not lend itself to accessing the required information and sufficient time to complete each recommended component of an assessment the measure may not yield accurate information. In particular, assessors want to be clear about the outcome of concern (verbal abuse, physical abuse, severe violence, stalking, femicide?) and knowledgeable about relevant base rates (Heilbrun, 2009).
This review is intended to provide both researchers and clinicians with a comprehensive review of the state of the IPV risk assessment field; as such we were inclusive of studies by not excluding published findings based on study quality but rather describing the limitations of available research. Reflecting that objective and the heterogeneous nature of the research available we have provided a narrative review and did not provide common effect sizes via a meta-analysis. We also included only studies published in English from Westernized nations, published in peer-reviewed journals thereby limiting the generalizability of our conclusions. Based on the available literature, we are also unable to provide guidance on the clinical relevance and utility of these instruments with female perpetrators, male victims, and in same-sex relationships due to the lack of studies using relevant populations. The field at present is limited by the small number of studies that have addressed each instrument and due to diverse methodological limitations. The extant literature lacks prospective, longitudinal studies, studies comparing multiple instruments, studies that reflect the intended outcome and/or that utilize multiple sources of data to code outcome criteria, and studies that code the measures in the manner intended/include all items, thereby making conclusions tentative. For instance many studies of the SARA relied on file reviews in the absence of an interview. Also, the criterion variables either did not match the intended use of the instrument (e.g. the Danger Assessment was used to measure re-abuse) or relied on criminal records or self report, but rarely both. In recognizing these limitations we hope to guide future research. For the same reason we did not perform a meta analysis of the heterogeneous literature.
In conclusion, there is considerable room for further IPV risk assessment research. In particular, studies examining the incremental validity of using IPV specific variables or measures once taking into account general risk predictors (incremental validity) , prospective studies, and rigorous designs comparing multiple measures in single studies and using relevant criterion variables are required.