The present analysis is devoted to making an empirically based choice among alternate causal explanations. This entails making causal inferences from statistical correlations. While this might, at one time, have constituted a heresy, I believe that the procedure to be followed here will soon be a part of statistical orthodoxy.

This is not the place for an extended philosophical discussion of the problem of causality. Yet I would like to make my position on the problem as clear as concise presentation will permit. My basic sympathies are with that school which argues that scientifically relevant causal explanation inheres only in our theories, i.e., that the explained event takes the shape which it does because our postulates and logic preclude any other shape on pain of being themselves incorrect. However, the development of such theory, containing such postulates, is usually the product of an inspired insight on the part of one thoroughly immersed in the manifestations of the empirical phenomenon under consideration. The production and verification of such insight in a systematic and reproducible way is the goal of inductive research. Where controlled experimentation is possible, Mill's canons may apply. Where such experiments are either impossible or impracticable, statistical inference becomes necessary. It is in this situation that the present approach, based upon a model developed by Herbert Simon and others, seems justified.

Simon's model is designed to capture the asymmetry in our notions of causality. When one speaks of A as a cause of B, one usually has in mind a unidirectional forcing, and not merely a covariation, or phased covariation.