FINITE BIPREDICTION TESTS IN TREATMENT-RESPONSE CONTINGENCY TABLES

Matthias J. Müller (Germany)

Abstract

Finite partial association or biprediction tests are proposed for treatment response contingency tables with small 2 × 2 subsamples embedded in c × 2 or r × c treatment response tables. The tests are based on the Freeman-Halton principle and compared with the asymptotic Kimball-Castellan tests in case of low frequency 2 × 2 subtables. Biomedical applications are illustrated by two-sample immunological response patterns to vaccination, and by a three-sample example of treatment-induced improvement ratings in psychiatry. Further applications related to repeated observations and bivariate response patterns are discussed.

Keywords and phrases: finite tests, configural frequency analysis, partial association, contingency tables, categorical data, non-parametric tests.