Potential Outcomes Framework

The potential outcomes framework is the approach to causal inference that treats each unit as having multiple possible outcomes — one for each treatment condition — of which only one is ever realized. The causal effect is the difference between the potential outcome under treatment and the potential outcome under control, a comparison that is fundamentally counterfactual because the two outcomes cannot both be observed for the same unit at the same time. This framework, formalized by Donald Rubin and building on the work of Jerzy Neyman, shifts the focus from causal structure to causal estimation: the question is not what causes what, but how much a specific intervention changes a specific outcome in a specific population. The framework is the engine of modern randomized controlled trials, propensity score matching, and difference-in-differences designs, but its power is purchased at the cost of a narrow scope: it excels at binary treatments and well-defined outcomes, and it struggles with systems where feedback, multiple interacting causes, and dynamic structure make the notion of a single treatment effect incoherent. For systems of that kind, the graphical methods of Judea Pearl and the dynamical methods of Convergent Cross Mapping are more natural tools.