Mean-Centering and the Interpretation of Anova and Moderated Regression

Joachim Vosgerau, Carnegie Mellon, US
Hubert Gatignon, INSEAD, France
ANOVA and moderated regression can yield seemingly different effects when applied to the same data. In ANOVA, main effects are estimated at their means, and interaction effects are restricted to be symmetric relative to the means. In moderated regression, constant effects are estimated at zero, and no specific pattern of interaction is imposed. As a consequence, main effects in ANOVA cannot be interpreted if the independent variables are truly categorical. We discuss how to interpret effects in moderated regression, and when to mean-center variables in moderated regression (mean-centering can overcome the arbitrary origin of interval scales but does not reduce multicollinearity).
[ to cite ]:
Joachim Vosgerau and Hubert Gatignon (2007) ,"Mean-Centering and the Interpretation of Anova and Moderated Regression", in NA - Advances in Consumer Research Volume 34, eds. Gavan Fitzsimons and Vicki Morwitz, Duluth, MN : Association for Consumer Research, Pages: 537-540.