When researchers test simple effects (the effect of factor at one level of the other factor), it’s typical to use MSwithin as the error term (i.e., the error term from the omnibus ANOVA). Why is this...

When researchers test simple effects (the effect of factor at one level of the other factor), it’s typical to use MSwithin as the error term (i.e., the error term from the omnibus ANOVA). Why is this practice popular? Why not use the pooled variance associated with only the means that figure into the simple-effect analyses? (Hint: there were two reasons. One deals with power; the other deals with precision)