A0396
Title: Causal inference with outcome-dependent missingness and self-censoring
Authors: Jacob Chen - Johns Hopkins University (United States)
Daniel Malinsky - Columbia University (United States)
Rohit Bhattacharya - Williams College (United States) [presenting]
Abstract: Unobserved confounding and missing data are two of the most common issues that arise in observational studies. In particular, if the outcome variable directly affects its own missingness status, i.e., it is "self-censoring", the resulting non-ignorable missingness along with unobserved confounding may lead to severely biased causal effect estimates. A test is proposed based on a randomized incentive variable offered to encourage reporting of the outcome (e.g., randomized gift cards) that can be used to verify identification assumptions that are sufficient to correct for both self-censoring and confounding bias. Concretely, the test confirms whether a given set of pre-treatment covariates is sufficient to block all sources of confounding between the treatment and outcome, as well as all associations between the treatment and missingness indicator after conditioning on the outcome. It is shown that under these conditions, the causal effect is identified by using the treatment as a "shadow variable" that allows us to correct for self-censoring. This leads to an intuitive inverse probability weighting estimator that uses a product of the treatment and response weights. The efficacy of the test is evaluated, and the estimator is downstreamed via simulations.
