Title: Variable selection and estimation in generalized linear models with measurement error
Authors: Liqun Wang - University of Manitoba (Canada) [presenting]
Abstract: The variable selection and estimation problems in linear and generalized linear models are studied when some of the predictors are measured with error. We demonstrate how measurement error (ME) affects the selection results and propose regularized instrumental variable (RIV) methods to correct for the ME effects. We show that the proposed methods have the oracle property in a linear model and we derive their asymptotic distribution under general conditions. We also investigate the performances of the methods in generalized linear models. Our simulation studies show that the RIV methods outperform the naive method in both linear and some generalized linear models. Finally, the proposed method is applied to a real dataset.