SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Imprint / Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University

Contributions: Abstract

A taxonomy for predictor structures in multiple regression

Safir Yousfi
University of Heidelberg

The presented taxonomy of predictor structures is based on two dimensions. The first dimension (suppression vs. redundancy) refers to the explained variance. If the multiple coefficient of determination exceeds the sum of squared bivariate correlation suppression is present.

The classification with respect to the second dimension (consistency vs. divergence) depends on the consistency of the signs of the regression coefficient and the bivariate correlation with the criterion. The thresholds between the four basic constellations (consistent and divergent suppression, consistent and divergent redundancy) define further predictor structures (orthogonality, traditional suppression, exhaustive prediction, symmetry).

Each linear regression model with two predictors can be classified within this taxonomy. Generalisations for models with more predictors or criteria, categorical predictors and interactions are discussed as well as problems of the interpretation of divergence and suppression situations.