Two-Stage Estimation of Structural Instrumental Variable Models with Coarsened Data

• Activity Number: 431
• Type: Contributed
• Date/Time: Tuesday, August 2, 2016 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Byeong Yeob Choi* and Jason Fine and Maurice Alan Brookhart
• Companies: The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
• Keywords: Accelerated failure time model ; Coarsened data ; Endogeneity ; Instrumental variables ; Rank-based estimator ; Unmeasured confounder

Abstract

Two-stage least squares is a popular estimation method for structural equations models that accommodate unmeasured confounders. In such models, both the outcome and exposure are assumed to follow linear models conditional on the measured confounders and instrumental variables, which impacts the outcome only via its relation with the exposure. We consider coarsened data, where both the outcome and exposure may be incompletely observed, which includes the important special case where both the outcome and exposure are censored event times. A general class of two-stage minimum distance estimators is proposed that separately fits the linear models for the outcome and the exposure and then estimates the true exposure effect on the outcome using a reduced form model. An optimal minimum distance estimator is identified and shown to be theoretically superior to the usual two-stage least squares estimator with fully observed data. Simulation studies demonstrate that the methods perform well with realistic sample sizes. We used our approach in a study of the comparative effectiveness of colon cancer treatments.