Abstract
Identifying the motivation for violent behavior is important in risk assessment and risk management. This study examined previously archived data from the MacArthur Violence Risk Assessment Study to investigate the relationship between psychopathy and instrumental violence committed by civil psychiatric patients following hospital discharge. Analyses compared the traditional two-factor model of the Psychopathy Checklist: Screening Version (PCL:SV) to Cooke and Michie's revised three-factor model and Hare's revised 4-facet model. Results indicated that higher total scores on all models were associated with increased risk for instrumental membership. This remained the case after controlling for variables related to instrumental violence. When examining the individual contribution of factors/facets in all three models, those reflecting antisocial traits (Factor II, facet 4) were the only significant predictors of instrumental membership. The implications of these findings are discussed.
This paper is based on a completed master's thesis. Analyses were run from previously archived data from the MacArthur Violence Risk Assessment Study. We would like to thank Tracy Costigan, David DeMatteo, and Graham Thomas who assisted in critiquing the original manuscript.
Notes
a The four-facet model has the same total score as the two-factor model. The first three facets are the same as those of the three-factor model.
a A chi-square statistic was computed for dichotomous variables and a t-test analysis was run for continuous variables.
b Figures reported are Cramer's (for categorical variables) or Pearson's (for continuous variables).
∗p <.05.
∗ ∗ p <.01.
a The predicted change in odds for a unit increase on the PCL:SV.
b Nagelkerke R2 is reported for effect size.
c Values indicate the percentage of variance accounted for.
∗p <.05.
∗∗p <.01.
a The predicted change in odds for a unit increase on the PCL:SV.
∗p <.05.
∗∗p <.01.
∗∗∗p <.001.
1. Effect size was calculated by dividing the model chi-square (based on log-likelihood) by the original -2LL (log-likelihood before any predictors were entered). That is, R L 2 = − 2LL(Model)/− 2LL(Original) (CitationHosmer and Lemeshow, 1989).