The use of the $l_z$ and $l_z^*$ person-fit statistics and problems derived from model misspecification