Reading focus
Syllabus: What Would Justify a Causal Claim?
Translate the potential-outcome notation into a specific person, intervention, comparison and time.
Why you’re reading it. Define the effect before deciding how to estimate it.
A question to consider. For a study-group intervention, what is the missing outcome for a student who attends, and what could another student legitimately tell us about it?
Why this reading
Rubin defines a treatment effect through outcomes that would occur under alternative conditions. The paper makes the missing comparison explicit and uses it to explain both the value of randomization and the demands of nonrandomized inference.
Before you read
Potential outcomes describe what the same unit would experience under each treatment at a specified time. Only one can be observed in that comparison. An average effect can differ from every individual’s effect. Random assignment to conditions and random sampling from a population are separate operations.
Read for
- Follow how the definition of an effect determines the comparison the analysis needs.
- Distinguish an estimator’s average behavior over possible assignments from the accuracy of one realized study.
- Ask what knowledge of assignment or selection would be required to interpret a nonrandomized comparison.
Consider afterwards
- Why would measuring one person before and after an intervention fail, by itself, to reveal both potential outcomes for the same time?
Editorial context
The paper develops a framework and examines examples; it does not make every comparison causally interpretable. Its notation clarifies the question but leaves substantive assumptions about treatments, selection and relevant units to be justified. Estimating an average effect also leaves the mechanism and transfer to other populations unresolved.