Checking the assumptions in mixed-model analysis of variance: a residual analysis approach