Discussion: ‘the Scientific Model of Causality’

Heckman advocates an approach to causal inference that draws upon structural modeling of the outcome(s) of interest (which he calls scientific), and he contrasts this approach sharply with that arising out of the statistical literature on experimentation. Drawing extensively on several previous papers—for example, Heckman (1997, 2000, 2001) and Heckman and Navarro-Lozano (2004)—Heckman goes even further here, arguing that the statistical literature on causal inference is incomplete because it does not attempt to model the process by which subjects are selected into treatments (or what statisticians have called the ‘‘treatment assignment mechanism’’) and that this literature confounds the task of defining parameters with the tasks of identifying and estimating these parameters. I shall return to these points later. But whereas Heckman distinguishes sharply between these approaches (and hence between certain literatures in economics and statistics), on balance I find the similarities in the approaches he discusses much more profound than the dissimilarities. To elaborate, while there has been and continues to be much philosophical