Parameter estimation and hypothesis tests in logistic model for complex correlated data

Observations are frequently generated in clinical trials from correlated multiple organs (or parts) of individuals. The statistical inference is little about conducting regression analysis based on such data. This paper first develops a logistic regression for correlated multiple responses using a stable correlation binomial (SCB) model. Then, we obtain maximum likelihood estimators (MLEs) of unknown parameters through a fast quadratic lower bound (QLB) algorithm. Further, likelihood ratio, score and Wald statistics are used to test the effect of covariates based on the MLEs. Finally, the QLB algorithm and asymptotic tests are evaluated through simulations and applied to real dental data.