The reliability and validity of the violence risk scale - experimental version 1
Increasing concern for the safety of society has resulted in a heightened demand for instruments that can accurately predict the risk of violence. As a result, efforts to improve risk prediction techniques have intensified. However, the accurate prediction of violence has proved to be a difficult task, often plagued by inaccuracies and a general lack of validity (Douglas, MacFarlane, & Webster, 1996). The present study represents the first phase in a large scale evaluation of a newly developed risk assessment instrument, the Violence Risk Scale-Experimental Version 1 (VRS-E1, Wong & Gordon, 1996). The VRS-E1 was specifically designed to assess the risk of violence in offender populations. Using a sample of provincial and federal offenders (n=60), subjects were assessed with the VRS-E1 by two independent raters based on a file review and semi-structured interview. The interrater reliability of the VRS-E1 was determined based on the degree of concordance between the scores of two raters for each item and for the total scores using the Kappa (k) Coefficient of Agreement Statistic and Pearson's (r) Product Moment Coefficient. To assess the extent to which the VRS-E1 items are measuring the same or a similar construct, an analysis of the internal consistencies of the scale items was conducted using the Cronbach's alpha. Item analyses were also used to evaluate the independent contribution of each VRS-E1 item to the scale total score. The convergent validity of the VRS-E1 was evaluated by investigating the relationship of the VRS-E1 scores with the scores obtained on validated measures of criminal violence (i.e., the Criminal Career Profile analysis and the Psychopathy Checklist-Revised), criminal risk/need (i.e., General Statistical Information on Recidivism scale and Level of Service Inventory-Revised), and aggression/hostility (i.e., the Aggression Questionnaire and Interpersonal Behaviour Survey). Discriminant validity of the VRS-E1 was assessed by correlating the VRS-E1 scores with the aggressiveness and assertiveness domains of the IBS, and the correlation of VRS-E1 scores with violent and non-violent offending. Stepwise multiple regression analyses were used to evaluate the relative contribution of the VRS-E1 and comparison measures to the post-diction of violent and nonviolent criminal convictions based on official records. The VRS-E1 demonstrated good interrater reliability. The item by item interrater agreement was strong. The obtained Kappa value using all pairs of ratings (n= 1651) was in the 'good' concordance range (Fleiss, 1981). Over 58% of the ratings were in complete agreement and only 7% differed by more than one scale point. A very high alpha coefficient of .92 was obtained indicating that the VRS-E1 is internally consistent and that the items comprising the VRS-E1 are most likely measuring one unidimensional construct. The convergent validity of the VRS-E1 was supported by its significant relationship to validated measures of criminal/violent risk. The discriminant validity of the VRS-E1 was supported by significant correlations with violent but not with non-violent convictions, and with the IBS aggressiveness but not with the assertiveness domains. As well, the results indicate that the VRS-E1 outperforms the other risk assessment instruments in the postdiction of violent convictions. Among the comparison measures, the VRS-E1 contributed most significantly to the postdiction of violent convictions. Overall, the results suggest that the VRS-E1 can be rated reliably by individuals trained to administer it; the scale is internally consistent, and that the VRS-E1 is a valid instrument for assessing the risk of violence in offender populations.