Variance homogeneity in interactive regression: A clarifying note about data transformations.

Authors:
  • PBPhilip Bobko
    Philip Bobko
  • CRCraig Russell
    Craig Russell
Publisher website
Google Scholar
Abstract
A data transformation recently used by Stone and Hollenbeck (1989) is based on a faulty premise and should not be considered in future analyses. Specifically, these authors mistakenly equate the homoscedasticity assumption in regression analysis with the notion that subgroup variances need to be equal. We demonstrate that subgroup variances in regression can be legitimately different, owing to true main effects and interactions in the data. Therefore, any transformation addressing these differential subgroup variances may unwittingly remove true effects of the independent variables.