[Limdep Nlogit List] Logged Variables in Negative Binomial Panel Regression
sb4 at u.washington.edu
Thu May 25 12:29:53 EST 2006
I am running negative bionomial regression models on panel data. I have transformed some of my independent variables (on the RHS) to their logarithmic values to obtain better distributions. However, I find that while the models with the raw variables run quite well, models with the logarithmically transformed variables usually fail to converge. The error message I typically get is #143 (estimated variance matrix of estimates is singular). However, this problem is unique to the negative binomial models and not to any other regression models (OLS, Tobit, Logit, Poisson) that I'm also running. Is this occurrrence due to some property of the negative binomial model? Would I be better off not transforming any of the independent variables? Would really appreciate any suggestions/ insights in this regard.
More information about the Limdep