Estimating Random Coefficients Models Given Covariates Having Ignorable Missing Values and Random Effects

• Activity Number: 645
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Yongyun Shin*+
• Companies: Virginia Commonwealth University
• Keywords: Random Coefficients Model ; Maximum Likelihood Estimation ; Conditional Expectation ; the EM algorithm ; Ignorable Missign Data

Abstract

In a random coefficients model where children are nested within schools, the outcome and covariates have ignorable missing values, and a covariate subject to missingness is normally distributed. A conventional approach to efficient handling of missing data is via efficient estimation of a joint distribution of all variables subject to missingness conditional on all of the covariates completely observed. A covariate subject to missingness and having a random effect (covariateMR), however, yields the joint model that is not normally distributed and difficult to estimate well. This article introduces efficient estimation of the random coefficients model by maximum likelihood and multiple imputation. The idea is to efficiently estimate the desired model given the covariateMR and impute missing outcomes at stage 1; and efficiently estimate the covariateMR model given the stage-one imputation and impute missing covariateMRs at stage 2. The two stages alternate until a convergence criterion is satisfied. The approach will be illustrated with a simulation study, and a national sample of children analyzing the random coefficient of log household income on reading achievement.