Another look at the (im-)precision of individual risk estimates made using actuarial risk assessment instruments

We investigated the precision of individual risk estimates made using actuarial risk assessment instruments (ARAIs) by discussing some major conceptual issues and then illustrating them by analyzing new data. We used a standard multivariate statistical procedure, logistic regression, to create a new ARAI based on data from a follow-up study of 90 adult male sex offenders. We indexed predictive precision at the group level using confidence intervals for group mean probability estimates, and at the individual level using prediction intervals for individual probability estimates. Consistent with past research, ARAI scores were moderately and significantly predictive of failure in the aggregate, but group probability estimates had substantial margins of error and individual probability estimates had very large margins of error. We conclude that, without major advances in our understanding of the causes of violence, ARAIs cannot be used to estimate the specific probability or absolute likelihood of future violence with any reasonable degree of precision or certainty. The implications for conducting violence risk assessments in forensic mental health are discussed.