Sensitivity analysis for incomplete data via unmeasured confounding

TL;DR

This paper introduces a sensitivity analysis method for incomplete data, modeling missingness as unmeasured confounding, and provides a measure called MinNI to quantify the impact of nonignorability on estimates.

Contribution

It proposes a novel approach to sensitivity analysis by assuming missingness arises from unmeasured confounding rather than the usual selection model, with a new sensitivity index MinNI.

Findings

MinNI quantifies the minimum nonignorability needed to alter estimates.

The method applies to proportion estimates and can extend to complex scenarios.

Provides a framework for assessing robustness of inferences to missing data mechanisms.

Abstract

We present a method to analyze sensitivity of frequentist inferences to potential nonignorability of the missingness mechanism. Rather than starting from the selection model, as is typical in such analyses, we assume that the missingness arises through unmeasured confounding. Our model permits the development of measures of sensitivity that are analogous to those for unmeasured confounding in observational studies. We define an index of sensitivity, denoted MinNI, to be the minimum degree of nonignorability needed to change the mean value of the estimate of interest by a designated amount. We apply our model to sensitivity analysis for a proportion, but the idea readily generalizes to more complex situations.