Whenever the density for an observation does not have a closed form, numerical integration of the latent continuous variables is required. For binary outcomes, see Bock and Aitkin in Psychometrika. I don't have the references handy. The response here is logit not censored but the idea is the same. For information about censored outcomes, see:
Maddala, G.S. (1983). Limited-dependent and qualitative variables in econometrics. Cambridge: Cambridge University Press.
Marco posted on Thursday, January 26, 2006 - 6:17 am
Hello Linda, hello Bengt,
when estimating sig-between with MLR, does that imply that the non-normality of the raw data S-T is "removed" in sig-between? It seems to be important, since sig-between could only be analyzed with non-robust ML.