Modeling Poverty Rates with Generalized Poisson Regression

Abstract

Poisson distribution which is the 'benchmark' distribution for count response has equidispersion characteristic. It means that the variance is equal to the mean. Fitting Poisson distribution on any count response which does not satisfy equidispersion can produce an underestimate or overestimate standard error. A Generalized Poisson (GP) distribution which is an extension of Poisson distribution can accommodate the possibility of equidispersion, overdispersion as well as underdispersion for count response. This research aims to employ the Generalized Poisson Regression (GPR) for analyzing the relationship between poverty rates in the East Java Province (Indonesia) and some potentially explanatory variables. The adequacy of the GPR model over the other count regression model, namely binomial negative regression model, also assessed in this research.