Issues in Evaluating Model Fit With Missing Data