Causation and Statistical Inference

Abstract

Abstract In the applied statistical literature, causal relations are often described equivocally or euphemistically as ‘risk factors’, or as part of ‘dimension reduction’. The statistical literature also tends to speak of ‘statistical models’ rather than of causal explanations, and to say that parameters of a model are ‘interpretable’, often means that the parameters make sense as measures of causal influence. These ellipses are due in part to the use of statistical formalisms for which a causal interpretation is wanted but unavailable or unfamiliar, and in part to a philosophical distrust of attributions of causation outside experimental contexts, misgivings traceable to the disciplinary institutionalization of claims of influential statisticians, notably Karl Pearson and Ronald Fisher. More candid treatments of causal relations have recently emerged in the theoretical statistical literature.