Statistical Approaches to Suicidal Risk Factor Analysis

Suicide research is a particularly difficult area primarily because of the base rate problem and inadequate case finding. Traditional item-analytic and multiple regression or discriminant function data-analytic methods in suicide research are criticized on several technical grounds, including capitalization on chance, failure to cross-validate, and confusion of the degree of relationship with its statistical significance. These errors are further confounded when the research data base misrepresents the very low true base rate. However, the most serious defect in item-analytic and both conventional and stepwise multiple regression procedures is their failure to take into account the causal structure of suicide risk factors. Setwise hierarchical multiple regression/correlation analysis is offered as an effective tool for suicide research. It capitalizes on the powerful general data-analytic features of regression analysis, but does so in a way that represents the causal structure of the putative risk factors. The more complex methods of causal models analysis are also recommended. I do not believe, however, that progress in the understanding of suicide lies mainly in the improvement of the statistical procedures employed. Even with optimal procedures, the amount by which we can expect to increase the predictability of suicidality using psychosocial risk factors is not likely to be large. Recent research in the biochemistry of suicide offers some hope. If to the psychosocial factors now employed we can add relevant biological factors and their interactions with psychosocial factors, we may be able to develop the causal models necessary for the understanding, prediction, and prevention of suicide.