NBINREG: Stata module to estimate negative binomial regression models

Here is the first version of a maximum liklihood negative binomial with cluster, robust, and score options. Initial values are calculated my a call to poisson. Two scores are produced: 1) the normal B-based scores, and 2) the score based on the alpha parameter. The latter score necessitates a digamma function, which is approximated here until a formal one is developed by Statacorp. I have not added the comparison with Poisson here, but it takes only a little extra code. Otherwise, the output is the same as other ML algorithms. A caveat: if the true value of alpha is close to 0, then the model is really Poisson. There may be convergence problems. This is true with any NB type program. Actually, when modeling it is probably best to start with Poisson - and if found to be overdispersed then use the NB. The value of the robust option here is quite important. It may be quite likely that an instance of overdispersed Poisson data is not truly negative binomial either. A clustering effect, for instance, which may cause overdispersion, may not take a gamma shape. If so, then the robust options becomes important.