Generalized Empirical Likelihood in Partially Linear Modes for Longitudinal With Non-monotone Missing Data

To study the estimation in partially linear models for longitudinal with non-monotone missing data, based on quadratic inference functions, the generalized empirical likelihood method is used to estimate the regression coefficients and the baseline function, and the corresponding maximum empirical likelihood estimators are derived. The empirical log-likelihood ratios are proven to be asymptotically chi-squared, and the corresponding confidence regions and intervals are then constructed. The numerical study is conducted to compare the finite sample behavior of the generalized empirical likelihood and the normal approximation-based method.