Rationale, Procedures, and Applications for Decomposition of Explained Variance in Multiple Regression Analyses

Abstract

Publication of studies utilizing multiple regression (MR) statistical techniques is on the rise in communication journals, but researchers have been remiss in reporting an important aspect of MR results: identification of the unique and non-unique (joint, incremental, or common) influences of independent variables on the dependent variable. This article explicates three methods of decomposing R2 and demonstrates, both by analogy and contrast to ANOVA, the importance of partitioning MR explained variance. Particular attention is paid to the relevance of variance decomposition techniques for explanation, prediction, and control in communication theory as well as the interpretive and the evaluative roles variance decomposition analyses should have in communication research. Finally, the specific steps entailed in performing each partitioning method are reviewed, and worked examples are supplied from the message/attitude/behavior research literature.