Abstract
A concern about the accuracy of the SPSS analysis of covariance program (ANCOVA) has been noted in the literature. This concern stems from different results obtained when using the classical experimental, regression, and hierarchial solutions in a "balanced" factorial design. This paper demonstrates that orthogonal effects are not to be expected following the "removal of the covariate." Three different methods of least squares were applied to an analysis of covariance problem in a 3 × 3 factorial experiment with equal cell frequencies. The three solutions yielded different sums of squares for the main effects even when the correlation between the covariate and dependent variable is negligible and the cell frequencies are equal. The importance of this fact is stressed for researchers using ANCOVA with factorial designs.
