Title: On negative binomial regression and its variants
Authors: Thomas Yee - University of Auckland (New Zealand) [presenting]
Abstract: The negative binomial distribution (NBD) is a very popular alternative to the Poisson for handling over-dispersion. We look at some computational issues relating to estimating the parameters of the NBD and some of its variants. These variants include the NB-1 (whose variance function is proportional to the mean), the NB-H (whose ancillary parameter is modelled using the same explanatory variables as the mean), and the NB-P (a more flexible data-driven NBD variant). It is shown that the statistical framework described recently is flexible enough to include all these models with surprising ease. Some results from the author's VGAM R package will be included.