NON NESTED MODEL SELECTION FOR SPATIAL COUNT REGRESSION

• Main Author: SURAHMAT (10112044), R.PRATHAMA
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/24113
• Institution: Institut Teknologi Bandung

Number of claims in in insurance data usually has many zeros, meaningly there is no claim from policy holder. Number of claims is discrete random variable. Usually used Poisson distribusion to model it. Random variable of Poisson distribussion has mean that equal to the variance. But, in insurance data variance is larger than mean. <br /> <br /> <br /> <br /> <br /> This case is called overdispersed. Poisson model is not match to model this case. Alternatively, used Binomial Negatif to model it. The other model that used is Zero In ated Poisson (ZIP) model. The other causes of this case is spatial heterogenity that is not observed. For that, spatial effect included to model, that is Conditonal Autoregressive (CAR). The context that used is Bayesian context. For parameter estimasion used Markov Chain Monte Carlo (MCMC) method. For model selection used DIC, Vuong test, and Clarke test.