Simple regression-based procedures for taxometric investigations

Certain theories of psychopathology postulate the existence of distinct latent populations of individuals. By analogy with biology, we call such latent populations "taxa" and we call the statistical testing of such theories "taxometrics." Taxometric procedures are robust; they do not make restrictive distributional assumptions. However, they have two disadvantages for nonstatistician users: (1) they are developed via algebra hard for many nonstatistician users intuitively to accept; and (2) computational software is not widely available. We address these problems by presenting a simple taxometric procedure, MAXSLOPE, based on regression plots for pairs of variables. This procedure is easily implemented using commonly available software and is intuitively rather easy to understand. We apply it to two artificial datasets. One dataset, used to explain the graphs, shows clear-cut evidence of taxa. The other example shows less clear grouping structure and is used to show that the proposed graphical procedure works even in nonideal cases. Comparisons are made with currently used procedures of cluster analysis and multivariate normal mixture analysis.