Comment on J. Neyman and causal inference in experiments and observational studies: ”On the application of probability theory to agricultural experiments. Essay on principles. Section 9” [Ann. Agric. Sci. 10 (1923), 1–51]. (English) Zbl 0955.01559

Summary: From the text: Without a doubt, Neyman’s paper cited in the heading is an important, but previously unposted, milestone in statistics. My belief is that the proposal to evaluate procedures over their repeated-sampling randomization-based distributions is uniquely Neyman’s. Had it not been for his attributions to the contrary, I would have thought that the proposal to use the physical act of randomization in experimental design was previewed here as well. Finally, with respect to his definition of causal effects, although the underlying implicit definition was relatively common prior to 1923, Neyman certainly appears to be the first to formalize it. However, neither he nor other writers in the next half-century seem to have applied this notation for potential outcomes to observational studies for causal effects, instead using the generally inferior observed-outcome notation, and providing no formal statement of a treatment assignment mechanism exhibiting possible dependence on the potential outcomes. In contrast, in the last dozen years, since the publication of the papers referenced in the introduction by Dabrowska and Speed, this framework, with explicit statements of its associated assumptions and explicit modeling of nonrandomized assignment mechanisms, has been applied in a variety of disciplines, often with an attendant increase in clarity. As with Neyman’s 1923 formalization, I have no doubt that these refinements were ‘in the air’, and I am glad to have been a contributor to their exposition and development.


01A75 Collected or selected works; reprintings or translations of classics
01A60 History of mathematics in the 20th century
Full Text: DOI