Efficient Targeted Estimation Using Instrumental Variables

• Activity Number: 524
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Boriska Toth*+ and Mark van der Laan
• Companies: University of California at Berkeley and University of California at Berkeley
• Keywords: semiparametric estimation ; causal inference ; clinical trials ; instrumental variables ; targeted maximum likelihood ; treatment effect

Abstract

The method of instrumental variables can be used to obtain an unbiased estimate of a causal effect in the presence of unmeasured confounding between a treatment and outcome. We shed light on the limitations and potential of this method by using a novel estimator that is, in some sense, optimal. We use a targeted maximum likelihood estimator (TMLE), which is semiparametric, asymptotically efficient, and a substitution estimator. We derive a TMLE estimator for the causal effect of treatment, as well as the assumptions needed for unbiased estimation and identifiability. A crucial question concerning the use of instrumental variables is whether the gain in bias reduction compensates the blowup in variance as compared to using a biased estimator that doesn't account for unmeasured confounding. We answer this question both analytically and empirically. We find that in regions of high unmeasured confounding and a strong instrument, the instrumental variable-based estimator is superior. We further compare the TMLE estimator to other major approaches to estimating the causal effect, concluding that TMLE performs favorably.