Generalized quasi-likelihood method for longitudinal binary data with measurement error

In this thesis, we propose an approach to correct the estimation of the bias of the model parameters when using a generalized quasi-likelihood method to analyze longitudinal binary data with measurement errors. The measurement errors are assumed to follow a normal distribution with an unknown variance, which can be estimated by repeated observations or taken from previous similar studies. An approximation method proposed by Monahan and Stefanski (1992)is used to obtain the expectation of an unknown function involved in the calculation of the means and covariance, which will be used later to construct the estimating functions of the GQL. A simulation study is carried out in the aim of investigating the small sample performance of the proposed approach. The results of an intensive simulation study show that the proposed approach works very well in all configurations. The efficiency gain of the proposed method, as compared to the naive use of GQL is remarkable. The proposed method has great potential to be widely used to analyze data from social, economical and biomedical studies