Overview
Paul Holland (1940–2022) was a statistician at Educational Testing Service (ETS) known for his 1986 paper "Statistics and Causal Inference" (Journal of the American Statistical Association, 81(396): 945–960), which gave the Rubin potential outcomes framework its canonical econometric exposition and introduced the phrase "no causation without manipulation." The paper formalizes the conditions under which causal effects are identified from statistical associations, distinguishes causation from association, and establishes why the causal effect of attributes (race, sex) cannot be estimated in the Rubin sense — only the effect of treatments that could hypothetically be manipulated.
Key Contributions / Features
- "No causation without manipulation": The causal effect Y(1)−Y(0) is only well-defined for treatments that could, in principle, be applied to any unit; attributes that cannot be randomized (race, sex, ability) are not "causes" in the potential outcomes sense — they are background variables that define subpopulations, not manipulable treatments
- Statistics and Causal Inference (1986): Canonical exposition of the Rubin Causal Model in the statistics literature; introduced the "science" (Neyman model for randomized experiments) and "super-population" framework and showed the Neyman, Fisher, and Rubin frameworks are equivalent under randomization
- Rubin-Holland model: Holland's formalization made the potential outcomes framework standard in statistics education; the model is now routinely called the Rubin Causal Model (RCM) or Rubin-Holland model
Related
Sources