2015/10/06 by Stijn Vansteelandt, Vanessa Didelez, Vansteelandt, Stijn +1 · 2 citations
Mathematics · Decision Sciences · #Advanced Statistical Methods and Models #Statistical Methods and Inference #Probabilistic and Robust Engineering Design
paper · pdf · doi:10.48550/arxiv.1510.01770
Two-stage least squares (TSLS) estimators and variants thereof are widely\nused to infer the effect of an exposure on an outcome using instrumental\nvariables (IVs). They belong to a wider class of two-stage IV estimators, which\nare based on fitting a conditional mean model for the exposure, and then using\nthe fitted exposure values along with the covariates as predictors in a linear\nmodel for the outcome. We show that standard TSLS estimators enjoy greater\nrobustness to model misspecification than more general two-stage estimators.\nHowever, by potentially using a wrong exposure model, e.g. when the exposure is\nbinary, they tend to be inefficient. In view of this, we study double-robust\nG-estimators instead. These use working models for the exposure, IV and outcome\nbut only require correct specification of either the IV model or the outcome\nmodel to guarantee consistent estimation of the exposure effect. As the finite\nsample performance of the locally efficient G-estimator can be poor, we further\ndevelop G-estimation procedures with improved efficiency and robustness\nproperties under misspecification of some or all working models. Simulation\nstudies and a data analysis demonstrate drastic improvements, with remarkably\ngood performance even when one or more working models are misspecified.\n