A methodology for improving efficiency estimation based on conditional mix-GEE models in longitudinal studies

Abstract

Estimating random effects accurately is crucial since it reflects the subject-specific effect in longitudinal studies. In this paper, we develop a new methodology for improving the efficiency of fixed-effects and random-effects estimation based on conditional mix-GEE models. The advantage of our proposed approach is that the serial correlation over time was accommodated in estimating random effects. Meanwhile, the normality assumption for random effects is not required. In addition, according to the estimates of some mixture proportions, the true working correlation matrix can be identified. The feature of our proposed approach is that the estimators of the regression parameters are more efficient than CCQIF, cmix-GEE and CQIF approaches even if the working correlation structure is not correctly specified. In theory, we show that the proposed method yields a consistent and more efficient estimator than the random-effect estimator that ignores correlation information from longitudinal data. We establish asymptotic results for both fixed-effects and random-effects estimators. Simulation studies confirm the performance of our proposed method.

KEYWORDS:

• Conditional mix-GEE
• Random effects
• Quadratic inference functions
• Longitudinal data

Additional information

Funding

This work was supported by the National Natural Science Foundation of China under Grant 11690012; National Natural Science Foundation of China under Grant 11631003; Education Department Science and Technology Project of Jilin Province under Grant JJKH20190728KJ; and the Key Program of Jilin University of Finance and Economics under Grant 2018Z01.