Functional Multiple Imputation for Longitudinal Data with Binomial Outcomes

• Activity Number: 512
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Stephanie Kliethermes*+ and Jacob Oleson
• Companies: University of Iowa and University of Iowa
• Keywords: Bayesian ; Functional Data ; Binomial Outcomes ; Missing Data ; Mixed Effects

Abstract

Functional mixed effects (FME) models and imputation methods provide flexibility in handling longitudinal data with non-parametric temporal trends and missing outcomes; but, these methods assume normality. We consider the situation with binomially distributed outcomes, specifically measured as percent correct. Although percent correct can be modeled assuming normality, estimates outside the parameter space are likely. We propose a Bayesian hierarchical binomial model on the number of correct responses received. The binomial model bounds the outcomes to realistic values and appropriately accounts for individual variability when imputing the missing values. We extend previous research on imputation methods in FME models where outcomes are normally distributed with varying degrees of missingness. Simulation studies advocate the usefulness of the binomial model particularly when outcomes occur on the boundary of the parameter space. The methods are demonstrated using a longitudinal study of cochlear implant users where we model the growth trajectory of individuals in their ability to recognize speech measured as percent correct