POISSON, NEGATIVE BINOMIAL, AND ZERO-INFLATED POISSON MODEL: CASE STUDY ON NUMBER OF DOCTOR VISITS IN AUSTRALIA

• Main Author: Natalie Elvaretta, Wilona
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/42416
• Institution: Institut Teknologi Bandung

An individual’s frequency of visiting a doctor to counsel about one’s health, in a certain period of time, could be an indicator of the quality of the healthcare service given. How often a person visits a doctor is influenced by several factors, such as: one’s socio-economic status; whether or not one has a health insurance; and one’s financial condition. The number of visits to a doctor is an example of a count data. Often, a Poisson model is used to model the count data. However, a count data may contain a large number of zeros. If there is a high frequency of zeros in a data set, it may cause an overdispersion; a case where the variance is larger than the mean. Due to its equidispersion property, applying a Poisson model to an over-dispersed data, is not appropriate. A Negative Binomial probability model could be used to model an over-dispersed data; whereas a zero-inflated Poisson model could be used to model a data set with excess zeros. In this final project, a Poisson, a Negative Binomial, and a zero-inflated Poisson model are applied to the number of doctor visits data. The comparison of the three models are carried out by analysing the values of the AIC and the BIC, and using the Vuong test. Based on those values, a Negative Binomial regression model is selected with the AIC value equals to 6,422.982 and the BIC value equals to 6,495.081. By conducting the Wald test, with a 5% significance level, it was found that the factors which are significant are: gender; insurance ownership status; and the variables which related to one’s health condition.