Predictive models in psychoanalysis

Psychoanalytic phenomena involve complex patterns of behavior which are difficult to model effectively and without oversimplification. This paper presents an approach to modeling which is appropriate for variables treated in psychoanalysis, and it describes a method for reducing the number of variables without losing essential information. The models used are from elementary catastrophe theory and are geometric in form. The approach is scientific in the strict sense that the models generate predictions which can be tested objectively. We illustrate this with an example involving the clinical phenomenon of affect-response. We present a ten-variable model based on the compact double cusp; the model generates an unforeseen, non-trivial, testable prediction.