Causal Inference Without Counterfactuals: Comment

Judea Pearl

1. BACKGROUND

The field of statistics has seen many well-meaning crusades against threats from metaphysics and other heresy. In its founding prospectus of 1834, the Royal Statistical Society resolved “to exclude carefully all Opinions from its transactions and publications-to confine its attention rigorously to facts.” This clause was officially struck out in 1858, when it became obvious that facts void of theory could not take statistics very far (Annals of the Royal Statistical Society 1934, p. 16). Karl Pearson launched his own metaphysics “red scare” about causality in 1911: “Beyond such discarded fundamentals as ‘matter’ and ‘force’ lies still another fetish amidst he inscrutable arcana of modern science, namely, the category of cause and effect” (Pearson 1911, p. iv). Pearson’s objection to theoretical concepts such as “matter” and “force” was so fierce and his rejection of determinism so absolute that he consigned statistics to almost a century of neglect within the study of causal inference. Philip Dawid was one of a handful of statisticians who boldly protested the stalemate over causality: “Causal inference isone of the most important, most subtle, and most neglected of all the problems of statistics” (Dawid 1979). In the past two decades, owing largely to progress in counterfactual, graphical, and structuralnalyses, causality has been transformed into a mathematical theory with well-defined semantics and well-founded logic, and many practical problems that were long regarded as either metaphysical or unmanageable can now be solved using elementary mathematics. (See Pearl 2000 for a gentle introduction to the counterfactual, graphical, and structural equation approaches to causality.) In the article, Professor Dawid welcomes the new progress in causal analysis but expresses mistrust of the quasi-deterministic methods by which this progress has been achieved. Attitudes of suspicion toward counterfactuals and structural equation models are currently pervasive among statisticians, and Dawid should be commended for bringing such concerns into the open. By helping to dispel misconceptions about counterfactuals, Dawid’s article may well have rescued statistics from another century of stagnation over causality.

View full paper in PDF here