Modeling Overdispersion for Clustered Binomial Data

• 발행기관: 한국자료분석학회
• 저자명: Kwang Mo Jeong Hyun Yung Lee
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.15 No.5, 2343~2356쪽, 전체 14쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2013.10.30

영문 초록

Overdispersion is a very common phenomenon that occurs when binomial responses exhibit larger variance than that permitted by the binomial distribution. When binary responses are clustered within a subject the independence of Bernoulli trials cannot be assumed because the observations within a cluster are positively correlated with each other in general. This fact may cause the extra variation of binomial distribution as discussed by Kupper, Haseman (1978). In this paper we suggest various overdispersion models for clustered binomial responses in the presence of covariate variables. The quasi-likelihood method, the zero-inflated binomial model, the beta-binomial model, and the random effects model are the available approaches. We also discuss the goodness-of-fit statistics assessing the overdispersion models. The goodness-of-fit of overdispersion models will be compared through several examples.

목차

1. Introduction

2. Review on Goodness-of-fit Test for Overdispersed Binomial Data

3. Goodness-of-fit Test for Clustered Binomial GLMs

4. A Monte Carlo Study

5. Summary and Comments

References

키워드

clustered binomial data overdispersion quasi-likelihood zero-inflated binomial model beta-binomial model generalized linear mixed model