Sensitivity Analysis

Sensitivity analysis quantifies how conclusions change when key assumptions are relaxed, especially unobserved confounding.

What it gives you

• A range of plausible policy values instead of a single point.
• A way to report robustness to hidden bias.

When to use it

• You cannot guarantee sequential ignorability.
• Overlap is weak or behavior logging is noisy.

Practical guidance

• Report both the point estimate and the sensitivity bounds.
• Be explicit about the sensitivity parameter and its meaning.
• Use the SensitivityPolicyValueEstimand to make the model explicit.
• Use synthetic benchmarks to build intuition before real data.