🤖 AI Summary
In observational studies, residual overt bias (due to inadequate measurement-based confounder matching) and hidden bias (from unmeasured confounding) often jointly distort causal inference; conventional Rosenbaum-type sensitivity analysis, which neglects their coupling, yields overly conservative conclusions. This paper proposes the first sensitivity analysis framework that jointly accounts for both overt and hidden bias in a coherent manner: leveraging iterative convex programming, it constrains hidden bias within a feasible region compatible with the observed matching structure—without requiring parametric assumptions on treatment or outcome models. Simulation studies demonstrate substantial gains in statistical power over existing methods. Empirical analyses on real-world datasets confirm the framework’s robustness and practical utility. An open-source R package implementing the method is publicly available to facilitate reproducibility and adoption.
📝 Abstract
Matching is one of the most widely used causal inference designs in observational studies, but post-matching confounding bias remains a challenge. This bias includes overt bias from inexact matching on measured confounders and hidden bias from the existence of unmeasured confounders. Researchers commonly apply the Rosenbaum-type sensitivity analysis framework after matching to assess the impact of these biases on causal conclusions. In this work, we show that this approach is often conservative because the solution to the Rosenbaum-type sensitivity model may allocate hypothetical hidden bias in ways that contradict the overt bias observed in the matched dataset. To address this problem, we propose an iterative convex programming approach that enhances sensitivity analysis by ensuring consistency between hidden and overt biases. The validity of our approach does not rely on modeling assumptions for treatment or outcome variables. Extensive simulations demonstrate substantial gains in statistical power of sensitivity analysis, and a real-world data application illustrates the practical benefits of our approach. We have also developed an open-source R package to facilitate the implementation of our approach.
