Yunfei Wang
Children’s National Health System, USA
Title: Implementing nonparametric residual bootstrap multilevel logit model with small number of level-2 units
Biography
Biography: Yunfei Wang
Abstract
It is a challenge to model hierarchically structured data with a small number of groups (e.g., level 2 units). When the number of groups is small, standard errors of parameter estimates in multilevel modeling tend to be biased downward, thus inflating the test statistics and the type I error. Although both parametric and nonparametric residual bootstrap approaches have been developed to deal with small number of groups in multilevel modeling with continuous response variables, there are limited approaches available with binary response variables. Here, we have developed a SAS macro by implementing nonparametric residual bootstrap multilevel logit model to analyze binary response variables. Using simulated data, our results show explicit advantage of the nonparametric residual bootstrap approach compared to the default estimator -- Residual Pseudo-likelihood (RSPL) – in SAS Proc GLIMIX with respect to modeling binary response variables in multilevel data with a small number of groups.