Title: Shrinkage estimation in the presence of missing data
Authors: Christian Heumann - Ludwig-Maximilians-University Munich (Germany) [presenting]
Abstract: Shrinkage estimators such as the LASSO or RIDGE are now popular tools for automatic variable selection in regression analysis. The inference problem has also been studied and various proposals have been made, e.g. Multi-Split-Sampling. An interesting problem is how to get confidence intervals in the presence of missing covariates, when missing is MAR (missing at random) and the data have been multiply imputed. We discuss several approaches to that problem and evaluate the proposed methods in an extensive simulation study.