Dear Olivia, thank you very much for the insight. Actually my problem is that for certain combinations of the levels of the variables I have no observations, i.e. in the mexico example that's equivalent to not having an observation for "vote88=1", "pristr=1", "othcok=2","othsocok=2" Is there a way to deal with that? Thanks Alejandro Olivia Lau <olau(a)fas.harvard.edu> escribió: Dear Alejandro, If your data isn't full-rank, it means that you have one of the following problems: 1) You have one or more perfectly collinear variables. (e.g., x2 = x3 + 4) 2) You have a variable that consists either entirely of NA values or 0s. Some basic data checking (try summary(), is.na(), etc) will clear this up. See http://gking.harvard.edu/zelig/docs/Data_Sets.html for some guildelines. Best, Olivia ----- Original Message ----- From: Alejandro Buren To: zelig(a)latte.harvard.edu Sent: Monday, August 15, 2005 5:36 PM Subject: [zelig] non full-rank models in mlogit Thanks to all, I have overcome my previous problem. I'm writing now because I have data which doesn't fulfill the vglm requirement of being full-rank. Is there a way to fit a multinomial logit non full-rank model? Thanks, Alejandro __________________________________________________ Correo Yahoo! Espacio para todos tus mensajes, antivirus y antispam ¡gratis! ¡Abrí tu cuenta ya! - http://correo.yahoo.com.ar