Scalp distributions of event-related potentials: an ambiguity associated with analysis of variance models

Analysis of variance (ANOVA) interactions involving electrode location are often used to assess the statistical significance of differences between event-related potential (ERP) scalp distributions for different experimental conditions, subject groups, or ERP components. However, there is a fundamental incompatibility between the additive model upon which ANOVAs are based and the multiplicative effect on ERP voltages produced by differences in source strength. Using potential distributions generated by dipole sources in spherical volume conductor models, we demonstrate that highly significant interactions involving electrode location can be obtained between scalp distributions with identical shapes generated by the same source. Therefore, such interactions cannot be used as unambiguous indications of shape differences between distributions and hence of differences in source configuration. This ambiguity can be circumvented by scaling the data to eliminate overall amplitude differences between experimental conditions before an ANOVA is performed. Such analyses retain sensitivity to genuine differences in distributional shape, but do not confuse amplitude and shape differences.