Primary Submission Category: Sensitivity Analysis
Design sensitivity and its implications for weighted observational studies
Authors: Melody Huang, Daniel Soriano, Samuel Pimentel,
Presenting Author: Melody Huang*
Increasingly, observational studies are being used to answer causal questions in the social and biomedical sciences. Estimating causal effects in observational settings often requires an assumption that unmeasured confounding is absent. This assumption cannot usually be checked empirically, and violations are often plausible. Recent work has introduced different sensitivity analyses to assess the potential impact of an unobserved confounder on a study’s results post hoc. However, sensitivity to unmeasured confounding is not typically a primary consideration in designing the treated-control comparison. We introduce a framework allowing researchers to explicitly optimize robustness to omitted variable bias at the design stage using a measure called design sensitivity. Design sensitivity, which describes the asymptotic power of a sensitivity analysis, allows researchers to transparently compare the impact of different estimation strategies on sensitivity. We show how this general framework applies to two commonly-used sensitivity models, the marginal sensitivity model and the variance-based sensitivity model. By comparing design sensitivities, we interrogate how key features of weighted designs, including estimands and model augmentation, impact robustness to unmeasured confounding, and how impacts differ for the two different sensitivity models. We illustrate the proposed framework on a study examining drivers of support in the Colombian FARC peace agreement.