MEPX
Chapter 5 of 10All chapters
  1. 1What correlation measures
  2. 2The five explanations
  3. 3Confounders
  4. 4Experiments
  5. 5Natural experiments
  6. 6Weighing causal evidence
  7. 7Selection effects
  8. 8Regression to the mean
  9. 9Language and reporting
  10. 10Acting under uncertainty

Chapter 5 of 10

Natural experiments

Randomness found rather than made.

The idea

Sometimes an outside event assigns people almost at random: a lottery, a policy boundary, an arbitrary cut-off date. That variation can be exploited to estimate an effect.

  • Difference in differences compares change over time between affected and unaffected groups.
  • Regression discontinuity compares people either side of an arbitrary threshold.

The assumptions

Each method rests on assumptions that can fail, and they should be stated and tested rather than assumed. The credibility of the estimate is the credibility of those assumptions.