StellarWhen Can We Conclude Causation?
The key question: When can we say one thing CAUSES another?
Rule: We can infer causation ONLY from a well-designed randomized experiment.
Why? Random assignment distributes all variables — both measured and unmeasured — approximately equally across treatment groups. If the groups differ significantly in the response variable after treatment, the treatment caused the difference.
Summary of conclusions:
| Study Type | Can Conclude Association? | Can Conclude Causation? |
|---|---|---|
| Observational study | Yes | No (confounding possible) |
| Randomized experiment | Yes | Yes (if properly designed) |
Important caveat: Even with a well-designed experiment, we can only generalize to the population represented by the experimental units. If subjects are volunteers, results may not generalize broadly.