Using IRT to separate measurement bias from true group differences on homogeneous and heterogeneous scales: an illustration with the MMPI.