Goodness-of-Fit for the Clustered Binomial Models

• 발행기관: 한국자료분석학회
• 저자명: 정광모(Kwang Mo Jeong)
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.17 No.4, 1725~1737쪽, 전체 13쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2015.08.30

영문 초록

We sometimes encounter the violation of independence between binary responses within subjects in clustered data that is commonly observed in cluster randomized trials. Under this situation the standard statistical methods that assume binomial distribution are inappropriate to use. In cluster randomized trials we have a target for comparing homogeneity of proportions between several treatment groups. The standard Pearson chi-squared statistic applied to the clustered data has a tendency to inflate type I error. Many researchers studied various kinds of adjusted chi-squared statistics in analyzing clustered binary data by incorporating the variance inflation and the intra-class correlation. In this paper we suggest the likelihood ratio test and the Wald test based on generalized linear mixed model. We explain the testing procedure via practical examples and also compare the performances through a Monte Carlo study.

목차

1. Introduction

2. Framework of Clustered Binary Data

3. Test Statistics

4. A Monte Carlo Study

5. Concluding Remarks

References

키워드

adjusted chi-squared statistic clustered binary data correlation between binary outcomes likelihood ratio test Wald test