Configural frequency analysis of dependent samples for intra-patient treatment comparisons

The dependent sample Configural Frequency Analysis (CFA; cf. von Eye, 1990) a non-parametrical test, is suggested as an alternative to the parametrical t-test for dependent samples. The dependent sample CFA may be used, for example, for detecting intra-patient treatment effects in pre-post or cross-over designs. Contrary to the t-test the dependent sample CFA can be applied to nominal (i.e., binary), ordinal, or continuous response variables. Furthermore, the proposed test may be generalized to two and more response variables. Significant test results or identified cells may be interpreted as discrimination types. Discrimination types are patterns of discrete or polychotomous response variables which contain frequencies occurring more often than expected under a null hypothesis (H ₀) of no treatment effects. The test for discrimination types follows the rationale of the Lehmacher (1980) simultaneous sign test which compares row with column marginals in a square contingency table. The dependent sample CFA is illustrated using examples from pharmacological research.