Bayesian approach for the zero-modified Poisson-Lindley regression model (Q2330491)

This paper proposes a zero-modified Poisson-Lindley regression model to fit overdispersed count data exhibiting inflation or deflation of zeros in the presence of covariates. By using the hurdle version of the zero-modified Poisson-Lindley distribution, separable likelihood functions for the parameter vectors are composed and the model is fitted without previous information about the zero-modification present in the dataset. Inference procedures are conducted under a Bayesian perspective employing an adaptation of the g-prior method. The random-walk Metropolis algorithm is applied for estimation via Markov Chain Monte Carlo sampling. A Monte Carlo simulation has been conducted to evaluate the performance of the developed methodology. For comparative purposes, the performance of maximum likelihood estimates was also assessed. The model was applied in the analysis of a real dataset of the number of bids received by a set of firms, and the impact of choosing different prior distributions for the regression coefficients was studied. A sensitivity analysis employing the Kullback-Leibler divergence was performed to detect influential points.