NON-NESTED SPATIAL COUNT REGRESSION MODEL SELECTION IN HEALTH INSURANCE

• Main Author: Jordy
• Format: Final Project
• Language: Indonesia
• Subjects: Matematika
• Online Access: https://digilib.itb.ac.id/gdl/view/33875
• Institution: Institut Teknologi Bandung

In the most cases, Poisson distribution is used for a count regression model. In this paper, not only Poisson Distribution that examined, but also another distribution such as Generalized Poisson, that capable of modeling overdisperion, and Zero- Inflated Generalized Poisson, that capable of modeling excess zeros in response distribution. Then I also add spatial effect to the regression model. With the addition of these spatial effects, Bayesian approached is considered which allows the modeling for a spatial dependency pattern. The addition of spatial effects of each location caused the model to be having a lot of parameters. Thus MCMC algorithm is used to estimate the parameters. Because the models to be compared come from different distribution models so the models are categorized as a nonnested models. To compare the models that are non-nested, we use Vuong test and Clarke test. Provided that the Generalized Poisson distribution is a better distribution to other models for the data that used in this paper.