Application of Auxiliary Variable in Response Mean Estimation for Incomplete Longitudinal Data

Abstract

This paper proposes an application of estimator for response mean Y given the value of an auxiliary variable X under simple linear regression model for incomplete longitudinal data and monotone missing data patterns. The proposed estimator, called Conditional Maximum Likelihood Estimator (CondMLE), is adapted from an Anderson’s factored likelihood function. Monte Carlo simulations was repeated 2,000 times for each situations in comparison of the coefficient of variations (CV) derived from CondMLE and Anderson’s estimator which generally regards no auxiliary variable in estimations. Essentially, regarding the results of the simulation study, CondMLE presented smaller CV than Anderson’s estimator, for sample size of 20, 30 and 50, regardless differences in the percentages of missing data and correlation coefficients of the response variables in the two occasions.