A Monte Carlo evaluation of Bobko's ordinal interaction analysis technique

Abstract

In a two-factor design, interactions are typically analyzed by analysis of variance (ANOVA). Bobko (1986) has suggested an alternative ordinal-interaction technique that might avoid spurious main effects and show more power than the classical ANOVA. In this study I (a) compared the classical and ordinal techniques in terms of Type I error rate and power under normally distributed homogeneous and heterogeneous populations, and (b) determined the effect of a population non-null main effect on the ordinal technique's Type I error rate. The ordinal technique showed a substantial power superiority over the classical technique under variance homogeneity, although it did have a power cap of less than 100%. Its Type I error rate under variance heterogeneity, however, was not stable and was susceptible to non-null main effects. (PsycInfo Database Record (c) 2025 APA, all rights reserved)