Should Causality Be Defined in Terms of Experiments ?

The paper reviews what has come to be called the Rubin-Holland model of causation. That model takes as its starting point experimentation and defines causation as a hypothetical value: a difference score between the observation of a unit in control and experimental condition. We discuss several different implementations of the model that Rubin and Holland have suggested. We present several other possible implementations with special emphasis of quasi-experiments. Finally, we note that causality can alternatively be better defined in terms of the absence of spuriousness.