A weighted simulation-based estimator for incomplete longitudinal data models

Recently, Li and Wang (2012a,b) and Wang (2007) have proposed a simulation-based estimator for generalized linear and nonlinear mixed models with complete longitudinal data. This estimator is constructed using the simulation-by-parts technique which leads to the unique feature that it is consistent even using finite number of simulated random points. This paper extends the methodology to deal with incomplete longitudinal data by applying the inverse probability weighting method for the monotone missing-at-random response data. The finite sample performance of this estimator is investigated through simulation studies and compared with the multiple imputation approach.