The only case in which it would be appropriate to compare item parameters (or resulting trait level scores) estimated separately in two or more groups is when the researcher can assume the groups have the same distribution on the latent trait variable.