In many forensic psychiatric hospitals, patients’ mental health is monitored at regular intervals. Typically, clinicians score patients using a Likert scale on multiple criteria including hostility. Having an overview of patients’ scores benefits staff members in at least three ways. First, the scores may help adjust treatment to the individual patient; second, the change in scores over time allows an assessment of treatment effectiveness; third, the scores may warn staff that particular patients are at high risk of turning violent, either before or after release. Practical importance notwithstanding, current practices for the analysis of mental health scores are suboptimal: evaluations from different clinicians are averaged (as if the Likert scale were linear and the clinicians identical), and patients are analyzed in isolation (as if they were independent). Uncertainty estimates of the resulting score are often ignored. Here we outline a quantitative program for the analysis of mental health scores using cultural consensus theory (CCT; Anders & Batchelder, 2015). CCT models take into account the ordinal nature of the Likert scale, the individual differences among clinicians, and the possible commonalities between patients. In a simulation, we compare the predictive performance of the CCT model to the current practice of aggregating raw observations and, as an alternative, against often-used machine learning toolboxes. In addition, we outline the substantive conclusions afforded by the application of the CCT model. We end with recommendations for clinical practitioners who wish to apply CCT in their own work.
Hierarchical paired comparison modeling, a cultural consensus theory approach
