Applying Stein Shrinkage: Theory and Experiment

Rudolf Beran
University of California, Davis

Stein's fundamental insights on shrinkage estimation, initially viewed 
as a theoretical curiosity, introduced a powerful class of 
regularization strategies that reduce risk. This talk considers indirect 
applications of Stein shrinkage to arrays of means in linear models 
through adaptive penalized least squares and adaptive hypercube 
estimators. Some limitations of current theory are noted. Numerical 
experiments bring out real-world considerations that underlie effective 
practical uses of Stein shrinkage.