Date of Award
Doctor of Philosophy (PhD)
James H. Geer
Legal reforms initiated in 1975 led to an increasing reliance upon mental health professionals to make predictions concerning violent behavior among the mentally ill, giving way to a series of violence prediction studies. However, these studies were noted to have major limitations, including unacceptably low rates of correct classification, high rates of misclassification, use of static, unavailable, or impractical predictors, and failure to cross-validate findings. Further, the bulk of these studies focused on predicting post-hospitalization violence, as opposed to inpatient violence, which can have important ramifications in treatment settings. The current study therefore examined the ability of forensic patients' current self report on theoretically relevant and readily available variables (i.e., anger, psychotic symptoms, moral associations, self-esteem, attributional style, IQ) to postdictively discriminate among patients who committed a violent act (N = 40) and did not commit a violent act (N = 40) during the first 2 weeks of admission. By using all of the identified predictors, a true positive rate of 72.5% and true negative rate of 70% was obtained. Applying these results to a separate sample of newly admitted patients (prediction phase), it was found that 87.5% were correctly predicted to engage in violent behavior, with 75% correctly predicted to not engage in violent behavior during the first 2 weeks of admission. Applying recommended criteria for computing validity and efficacy using local base rates, it was found that the resulting prediction scheme was both valid and effective at predicting violent behavior during the first 2 weeks of hospitalization. Moreover, it was found that the presence of hallucinations, patients' self-report of anger (decreased control and proneness to expression), and nature of commitment charge were of relatively greater importance in discriminating violent from nonviolent patients. The implications of these findings are discussed in relation to noted shortcomings of the study, therapeutic interventions for this population, and future prediction studies of violent behavior.
Martin, Mac Marcantel, "Predicting Short -Term Violence Among Forensic Inpatients." (2001). LSU Historical Dissertations and Theses. 382.