Generalized Least Squares Methods for Bivariate Poisson Regression

We consider bivariate Poisson regression models to analyse bivariate counts obtained under a stratified sampling scheme. A hybrid maximum likelihood (ML)/generalized least squares (GLS) method is used to obtain estimates of the relevant parameters. The proposed two stage procedure is asymptotically equivalent to and computationally simpler than that based exclusively on maximum likelihood. We compare the results obtained under both methods via a numerical illustration with real data as well as via a simulation study. Keywords: Bivariate countsBivariate poisson distributionGeneralized least squaresTwo-stage procedures ACKNOWLEDGMENTS We appreciate the comments of the associate editor and two referees which effectively contributed to improve the final version of the paper. The authors are also grateful to the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) and Empresa Brasileira de Pesquisa Agropecuária (EMBRAPA) for financial support.