Definition
Plain language
Rerunning an analysis with different reasonable choices to see whether the conclusion survives.
As stated in the literature
Robustness analysis varying modeling assumptions, inclusion criteria, or parameters to test whether a reported effect persists.
Also called: sensitivity checks
Why it matters: It separates findings that are real from findings that only exist because of one convenient analysis choice.
For example, if dropping the three oddest data points or using a slightly different cutoff date makes a reported improvement vanish, the original conclusion was fragile.
Heard on the show
“The authors' sensitivity checks support the bend, but extending that curve into the future is a separate assumption.”Episode 287 — Can You Measure Research Taste If The AI Isn't Allowed To Code?