Type of Document	Dissertation
Author	Li, Yinmei
Author's Email Address	yli1@lsu.edu
URN	etd-03312006-100431
Title	Power Analysis for a Mixed Effects Logistic Regression Model
Degree	Doctor of Philosophy (Ph.D.)
Department	Pathobiological Sciences (Veterinary Medical Sciences)
Advisory Committee	Advisor Name Title Daniel Scholl Committee Co-Chair James Miller Committee Co-Chair Giselle Hosgood Committee Member Luis Escobar Committee Member Martin Hugh-Jones Committee Member Guoli Ding Dean's Representative
Keywords	bovine immunodeficiency virus logistic-normal model clustered binary data power analysis
Date of Defense	2006-03-21
Availability	unrestricted
Abstract In herd health studies, the mixed effects logistic regression model with random herd effects are commonly used for modeling clustered binary data. These models are well developed and widely used in the literature, among which is the logistic-normal regression model. In contrast to the rich literature in modeling methods, the sample size/power analysis methods for such mixed effects models are sparse. The sample size/power analysis method for the logistic-normal regression model is not readily available. This study is to develop a power analysis/sample size estimation method for the logistic-normal regression model. Extended from the sample size method for the likelihood ratio test in the generalized linear models (Self et al., 1992), a power analysis method for the logistic-normal model is developed based on a noncentral chi-square approximation to the distribution of the likelihood ratio statistic. The method described in this dissertation can be applied to both exchangeable and non-exchangeable responses. The power curves are presented with respect to the change of each of the planning values while holding other planning values fixed for two examples of the logistic-normal model containing one random cluster effect. The results from this proposed sample size/power analysis method for the logistic-normal model were compared to the results from the method for the fixed effects logistic regression model. For a given total sample size and the same applicable planning values, the power for the logistic-normal regression model is smaller than that for the fixed effects logistic regression model, suggesting that the minimum required sample size calculated from using the method for the fixed effects model is too small to achieve the desired power when the logistic-normal model is to be used in data analysis.
Files	Filename       Size       Approximate Download Time (Hours:Minutes:Seconds)     28.8 Modem   56K Modem   ISDN (64 Kb)   ISDN (128 Kb)   Higher-speed Access    Li_dis.pdf 2.18 Mb 00:10:06 00:05:11 00:04:32 00:02:16 00:00:11