Variable Shrinkage and Selection via Measurement Error Selection Likelihoods
Kyle R. White
North Carolina State University
Coauthors: Len Stefanski, Yichao Wu
This paper presents a novel approach to shrinkage and variable selection by exploiting measurement error attenuation as opposed to standard approaches including penalized likelihoods. When applied to linear regression assuming joint normality for the predictors and response, this very different modeling perspective is proven to recover the LASSO solution path exactly in some cases and provides similar sparse solution paths in other cases depending on the choice of measurement error likelihood. This method can also be applied to generalized linear models through regression calibration. Simulations compare these measurement error likelihood methods with the current standards in variable selection.
Date received: August 9, 2012
Copyright © 2012 by the author(s). The author(s) of this work and the organizers of the conference have granted their consent to include this abstract in Topology Atlas. Document # cbfm-67.