COMPARISON WILLIAMS METHOD AND BETA-BINOMIAL IN OVERDISPERSION OF LOGISTIC REGRESSION: A CASE OF INDONESIA GENERAL ELECTION DATA 2014

摘要

Democratization in Indonesia so far has resulted in increasingly rational voters. The rational voters in each district or city of Indonesia are varied due to many factors. The system of election in Indonesia today is direct election system in which every citizen has freedom to vote the preferred candidates or even not to vote at all. There were 12 political parties participated in the legislative election in 2014, whereas in the presidential election there were two pairs of president and vice-president candidates competed.udThis research was aimed to obtain models, at the district level, that properly relate the votes were gained by the two candidates and other variables such as human development index, the results of legislative election as specially coalition of political parties voting results. Since the vote data was binary and showed over-dispersion then a logistics model accounting for over-dispersion was utilized. An over-dispersion problem is present whenever observations which might be expected to correspond to the binomial distribution may have greater variance than ni πi (1-πi).In this research the William’s method and beta-binomial regression were used to overcome the problem. The result showed that the Williams method provided better estimates when was compared to beta-binomial regression. ududkeyword: Logistic regression, Overdispersion, Williams’Method, Beta-Binomial Regression, General Election