Discrepancy between estimated marginal means and predicted values in ANCOVA
I am using SPSS/PASW Statistics and when I compare the estimated marginal means that are output as part of an analysis of covariance (ANCOVA) with the means for saved, predicted values, I find that the sets of means are not equal. Can you tell me why these values are different in SPSS?
Resolving the problem
The estimated marginal means saved from the ANCOVA model are also known as covariate-adjusted means in the ANCOVA literature. Covariate-adjusted means for all groups are calculated using the grand mean value(s) of the covariate(s) by default (or any other specific value(s) you specify instead). However, predicted values for individual cases use each case's actual covariate value(s). That is the reason for the discrepancies you observed. The means of the predicted means will only equal the estimated marginal means if the means of the covariates for each group are the same as the grand means of the covariates (or the specific value(s) at which you're requested the covariate(s) to be fixed in computing the EMMEANS).