296 – Methods in High-Dimensional Regression
Multiple Imputation for High-Dimensional Mixed Incomplete Data Using a Factor Model
Thomas R. Belin
University of California at Los Angeles
Ren He
University of California at Los Angeles Fielding School of Public Health
One strategy for producing imputations for high-dimensional incomplete data is to model associations among variables using a factor-analysis framework, thereby avoiding concerns with a more general association structure where some parameters are poorly estimated. Song and Belin (2004) pursued such a strategy for continuous outcomes; here we propose a similar strategy allowing for mixed data types (continuous, binary, ordinal and nominal). We describe an MCMC approach for fitting the model, and our method is compared in several simulation settings to available-case analysis and a rounding method.