Hi all, I want to run a multivariable logistic multinomial model with about 10 predictors (most of them continuous). The dependent variable has 4 categories (I think we could collapse it into 3 if necessary).

At any rate, when I run the model I get the following warning:

"There are 580 (66.8%) cells (i.e., dependent variable levels by subpopulations) with zero frequencies. The log-likelihood value cannot be further increased after maximum number of step-halving. The NOMREG procedure continues despite the above warning(s). Subsequent results shown are based on the last iteration. Validity of the model fit is uncertain. The NOMREG procedure continues despite the above warning(s). Subsequent results shown are based on the last iteration. Validity of the model fit is uncertain."

So the questions are: What can I do to get around this? Should I categorize all my predictors? Is there another solution? I imagine that if I use loglinear models i would run into the same problems. Can anyone suggest a book on categorical data analysis? P

-------------------------- Pablo Mora, Ph.D.