Skip to content

Lesson 6 of 6 · 12 min

Sensitivity analysis

Measure how strong unmeasured confounding would need to be to overturn a result.

Sensitivity analysis asks how wrong the assumptions could be before the decision changes. Report the threshold, not just the point estimate.

For event studies, inspect pre-trends and try alternative comparison groups. For uplift models, test placebo treatments.

Key points

  • Report the confounding strength that would flip the decision

Knowledge check

Effect estimation

Two-period difference-in-differences

Treated regions go from 60% to 66% employment; comparison regions go from 58% to 61%. What is the DiD estimate in points?

Answers are checked anonymously unless you sign in.

Sign in to mark this lesson complete and resume where you left off.