Statistical modeling for mortality data using local generalized poisson regression model

Mortality data by age are known have a non linier pattern likes a bath-tub shape. There are some parametric regression models developed to reveals the relationship between age and mortality rates. However using such models needs more efforts including a lots of parameter needed in the models and also numerical instabilities. To overcome this difficulties, recently, researchers give much attention to nonparametric regression models. Instead of assuming some restricted regression function, this approach allows for more flexible and robust of smooth function. There are several nonparametric model have been studied intensively, including Kernel and Smoothing Spline Regression. In this paper we developed a new approach by using local polynomial modeling. This model is a likelihood based model assuming Generalized Poisson distribution for the number of deaths at specific age. Using Generalized Poisson distribution instead of Poisson distribution makes this model robust for over or under dispersion problems. We apply this model to Indonesians mortality data based on the result from Population Census 2010, and found that this model performs well.