On Poisson-exponential-Tweedie models for ultra-overdispersed data

Abstract: We introduce a new class of Poisson-exponential-Tweedie (PET) mixture in the framework of generalized linear models for ultra-overdispersed count data. The mean-variance relationship is of the form $m+m^{2}+\phi m^{p}$, where $\phi$ and $p$ are the dispersion and Tweedie power parameters, respectively. The proposed model is equivalent to the exponential-Poisson-Tweedie models arising from geometric sums of Poisson-Tweedie random variables. In this respect, the PET models encompass the geometric versions of Hermite, Neyman Type A, P\'{o}lya-Aeppli, negative binomial and Poisson inverse Gaussian models. The algorithms we shall propose allow us to estimate the real power parameter, which works as an automatic distribution selection. Instead of the classical Poisson, zero-shifted geometric is presented as the reference count distribution. Practical properties are incorporated into the PET of new relative indexes of dispersion and zero-inflation phenomena. Simulation studies demonstrate that the proposed model highlights unbiased and consistent estimators for large samples. Illustrative practical applications are analyzed on count datasets; in particular, PET models for data without covariates and PET regression models. The PET models are compared to Poisson-Tweedie models showing that parameters of both models are adopted to data.