Calibrated Sensitivity Analysis for the Instrumental Variables Method for Observational Studies

• Activity Number: 392
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yenchih Hsu*+ and Scott A Lorch and Dylan Small
• Companies: University of Pennsylvania and University of Pennsylvania and The Wharton School
• Keywords: Causal inference ; Causal inference ; Sensitivity intervals ; Two stage least squares

Abstract

The instrumental variables (IV) method is an approach to estimating a causal relationship between a treatment and an outcome based on an observational study in the presence of unmeasured confounders. A valid IV is a variable that affects the treatment, has no direct effect on the outcome except through its effect on the treatment and is independent of the unmeasured confounders given measured confounders. There is often concern that an IV is not valid in the sense that it is correlated with unmeasured confounders. A sensitivity analysis is used to examine the impact of the violation. In many available sensitivity analysis methods, the sensitivity parameter that describes the invalidity of the proposed IV is on an absolute scale. It may be difficult for a subject matter expert to specify a plausible range of values for the sensitivity parameter on this absolute scale. We develop an approach that calibrates the value of the sensitivity parameter to the observed covariates that are thought to be related to the unmeasured confounders that are making the proposed IV invalid and to be more interpretable to subject matter experts. We will illustrate our method using a neonatology study.