Multiplicative correction of subject effect as preprocessing for analysis of variance.

Abstract

The procedure of repeated-measures ANOVA assumes the linear model in which effects of both subjects and experimental conditions are additive. However, in electroencephalography and magnetoencephalography, there may be situations where subject effects should be considered to be multiplicative in amplitude. We propose a simple method to normalize such data by multiplying each subject's response by a subject-specific constant. This paper derives ANOVA tables for such normalized data. Present simulations show that this method performs ANOVA effectively including multiple comparisons provided that the data follows the multiplicative model.