Construction and validation of risk assessments in a six-year follow-up of forensic patients: a tridimensional analysis

Abstract

Evaluations of risk were conducted for 162 Canadian mentally disordered criminal defendants through the assembly of actuarial data, scores from special-to-purpose psychometric instruments, and scaled global predictions of dangerousness to others by clinicians and nonclinical raters. Violent conduct by participants was tracked across legal and medical institutions and the community for a subsequent 6 years, with aggregate violence base rates reaching 62%. Decisions about risk were strongly associated with participant attributes and presentations during forensic interviews, but neither linear regression equations involving background and scale items nor direct discretionary judgments could account for more than 25% of variance in the frequency of outcome violence. Predictive accuracy maximized after 3 years, and was strongest for hospital-based violence. Professional clinicians were no more accurate than nonclinical raters. Implications of these findings for the sociolegal control of violence, and for the resurgent "second generation" of risk research, are explored.