Statistical Inference For Longitudinal Data With Missing Data Based On The Conditional QIF

In the past decades, analysis of longitudinal data has been a hot research field in statistics, with strong emphasis on application in biological and health sciences. For complete longitudinal data, there are richful statistical models in application, such as GEE(generalised estimating equations),PQL(penalized quasi-likelihood), QIF(quadratic inference function) and so on. But the problem of dealing with missing values in longitudinal study is also very common. In this paper, we introduce generalized linear mixed-effects models with three missing-data mechanisms(Little&Rubin[1]).Based on the conditional quadratic inference functions, we compare the parameter estimators with three different methods of filling missing values(zero,mean,tail value).We also give a generalized score-type method to test whether the missing values are ignorable or not based on the conditional quadratic inference functions.