This answer does a great job of showing that in main-effect-only models (i.e., models with no interaction; y ~ SEX (=boys,girls) + Schoolgender (=same_sex,coed)), one predictor can change holding the other predictor constant at ANY of its levels (e.g., holding SEX constant at either boys or girls).
But then, what is the benefit of controlling for a predictor (SEX) in a main-effect-only model, when we can hold it constant at ANY of its levels? (Indeed, is it correct to refer to a categorical moderator like SEX as a control moderator or the terminology is reserved for continuous predictors only?)
