Assessing Impact of Unobserved Confounders with Sensitivity Index Probabilities through Pseudo-Experiments

TL;DR

This paper introduces nonparametric indices to measure the influence of unobserved confounders in causal inference, validated through pseudo-experiments and real-world data, aiming to improve understanding of confounding effects.

Contribution

The study proposes novel nonparametric indices for assessing unobserved confounders' impact, validated via pseudo-experiments and real data, advancing causal inference methodology.

Findings

Indices reflect true confounder impact

Validated through pseudo-experiments

Applicable to real-world data

Abstract

Unobserved confounders are a long-standing issue in causal inference using propensity score methods. This study proposed nonparametric indices to quantify the impact of unobserved confounders through pseudo-experiments with an application to real-world data. The study finding suggests that the proposed indices can reflect the true impact of confounders. It is hoped that this study will lead to further discussion on this important issue and help move the science of causal inference forward.