This paper questions sensitivity analysis reliability.
Data & Methods
➡️ Using Monte Carlo simulations, we assess the robustness of variables across different data-generating processes.
Key Findings
➡️ Determinants and confounders' correlation significantly impact inference validity.
➡️ Variables with strong outcome influence are more likely to be correctly identified as robust.
➡️ Leamer's extreme bounds analysis and Bayesian model averaging reduce false positives compared to other methods.
➡️ Sensitivity test results depend heavily on the specific definition of robustness chosen.
Why It Matters
➡️ Researchers should focus on variables meeting inferential criteria rather than relying solely on sensitivity tests.
