A0672
Title: Instrumental variable method in regularized regression with mismeasured predictors
Authors: Liqun Wang - University of Manitoba (Canada) [presenting]
Abstract: Regularization methods are widely used in high-dimensional regression models, and most methods are developed for situations where all variables are correctly and precisely measured. However, in real data analysis, measurement error is common. The variable selection and estimation problems are studied in linear and generalized linear models where some of the predictors are not directly observable. How measurement error impacts the selection results is demonstrated, and regularized instrumental variable methods are proposed to correct the effects of measurement error. The proposed methods are consistent in selection and estimation, and their asymptotic distributions are derived under general conditions. The performance of the methods is also investigated through Monte Carlo simulations, and they are compared with the naive method, which ignores measurement errors. Finally, the proposed method is applied to a real dataset.
