[R] multinomial conditional logit models

John Hendrickx john_hendrickx at yahoo.com
Wed Feb 12 14:00:03 CET 2003


A little while ago, I asked for some help in estimating "multinomial
conditional logit models". If you restructure your data, you can
estimate a multinomial logit as a conditional logit model. This gives
flexibility in imposing constraints on your dependent variable. One
application is to estimate a loglinear model for square tables such
as quasi-independence or quasi-symmetry with covariates.

Anyhow, I think I've sorted out most of the problems and I've posted
a sample program at
http://www.xs4all.nl/~jhckx/mcl/R/

There's also a sample program there on how to estimate models for
square tables, aka mobility models. Comments and suggestions on how
to do things more efficiently or elegantly in R are most welcome.

One problem that remains is that clogit in R doesn't produce the same
estimates as other programs like Stata. I've estimated this sample
model in several packages and they all have the same estimates to 3
decimal places accuracy. I've tried different convergence settings in
clogit but to no effect. I'd appreciate it if anyone could clarify
this.

Best regards,
John Hendrickx




More information about the R-help mailing list