A Comparison of Bootstrap Confidence Intervals for Multi-level Longitudinal Data Using Monte-Carlo Simulation

Longitudinal investigations, where subjects are followed over time, have played an increasingly prominent role in medicine, health, and psychology in the last decades. This chapter will address inference for a two-level mixed model for a longitudinal study where observational units are clustered at both levels. Bootstrap confidence intervals for model parameters are investigated under the issues of non-normality and limited sample size of the original data. A one stage case-resampling bootstrap will be established for constructing confidence intervals by sampling clusters with replacement at the higher level. A two-stage case-resampling bootstrap will be developed by sampling clusters with replacement at the higher level and then sampling with replacement at the lower level also. Monte-Carlo simulations will be utilized to evaluate the effectiveness of these bootstrap methods with various size clusters for the mixed-effects model in terms of bias, standard deviation and confidence interval coverage for the fixed effects as well as for variance components of the random effects . The results show that the parametric bootstrap and cluster bootstrap at the higher level perform better than the two-stage bootstrap . The bootstrap methods will be applied to a longitudinal study of preschool children nested within classrooms.