Title: Estimating extra zeros proportion in different variability conditions of a regression count model
Authors: Antonio Jose Saez-Castillo - Universidad de Jaen (Spain) [presenting]
Antonio Conde-Sanchez - Universidad de Jaen (Spain)
Ana Maria Martinez-Rodriguez - University of Jaen (Spain)
Abstract: Several identification problems in the fit of a count data model appear when the dataset includes a certain proportion of extra-zeros and is additionally affected by a lack of equi-dispersion: commonly, the zero inflation may be confused with over-dispersion, and under-dispersion may be also hidden if the model is not adequate. The zero inflated hyper-Poisson regression model permits to independently manage the existence of extra zeros and over- and/or under-dispersion in presence of covariates. A simulation study has shown its capability to adequately estimate the proportion of extra-zeros and to distinguish the existence of over- and under-dispersion. This model is tested in different real datasets which present different variability structure, and its fits are compared with those provided by other common models, such as the zero-inflated Poisson or the zero-inflated negative binomial.