The Unit of Analysis: Group Means Versus Individual Observations

Abstract

This paper shows that the common recommendation to use group means when there may be nonindependence among observational units is unnecessary, unduly restrictive, impoverishes the analysis, and limits the questions that can be addressed in a study. When random factors are properly identified and included in the analysis, the results (Fs and critical Fs) are identical in balanced ANOVA designs, irrespective of whether group means or individual observations are employed. The use of individual observations also allows the exploration of other interesting questions pertaining to interaction and generalizability. In addition, the pooling strategy can be considered. Thus, the question of the proper experimental unit or unit of analysis for treatment effects is answered directly, correctly, and implicitly when the proper statistical model is employed.