Title
Variable selection in the linear regression model with censored outcomes
Speaker
Brent Johnson, University of North Carolina
Abstract
Variable selection is an important topic in the statistical sciences with applications in many disciplines including epidemiology, engineering, and econometrics. Outcome censoring adds an additional layer of complication to the variable selection procedures and improper handling of such data leads to biased results and spurious conclusions. A biomedical application of variable selection with censored data occurs in clinical trials with staggered entry and subjects followed until the study endpoint or end of follow-up, whichever comes first. We discuss two ideas for variable selection: the first idea is a class of shrinkage estimators based on penalizing a vector of estimating equations while the second defines an objective stopping rule (through the addition of pseudo variables) which is then used in a final forward selection algorithm. The proposed procedures offer scientists and investigators semiparametric alternatives to methods based on the partial likelihood. We explore the operating characteristics of the proposed methods in small samples and illustrate their utility by applying the methods to real data.
Meet the speaker in Room 212 Cockins Hall at 4:30 p.m. Refreshments will be served.
