Chapter 3 of 10All chapters
Chapter 3 of 10
When you cannot experiment
Observational evidence.
The limitation
Astronomy, epidemiology and economics often cannot assign treatments. They compare groups that differ for their own reasons, which is where confounders live.
- Statistical adjustment helps and never fully substitutes for randomisation.
- Natural experiments exploit an accident that assigned people almost randomly.
Strength of evidence
Consistency across studies, a dose response relationship, a plausible mechanism and correct time order all raise confidence when an experiment is impossible.