Is another, perhaps less important, difference that the z-scaling approach does "handle" (or remove) heteroscedasticity between subjects - at least in some fashion - whereas an lmer() model with random intercept would assume that each subject's observations have the same variance? DEJ