Sensitivity Analysis

Every statistical model relies on assumptions.

• "We assumed the missing data was random."
• "We assumed the outliers didn't matter."
• "We adjusted for age and sex."

But what if you were wrong?

Breaking the Model

Sensitivity Analysis is like a stress test. You re-run the math with different assumptions.

1. Missing Data: What if everyone who dropped out actually died? Does the drug still work?
2. Outliers: What if we remove the one patient who lived to 120? Does the average survival change?
3. Adjustments: What if we don't adjust for age?

Robust vs. Fragile

• Robust: You change the assumptions, and the result stays the same. The drug works no matter how you slice it.
• Fragile: You change one tiny assumption, and the p-value jumps to 0.06. The result disappears.

The Diagnosis

Good papers include a sensitivity analysis. They say: "We tested this 5 different ways, and the answer was always Yes."

Bad papers pick the one specific set of assumptions that gave the answer they wanted.