Social Science Statistics Blog

The first is more of a public relations issue than an intellectual problem: Counterfactualists – I at any rate – apparently come on a bit strong at times. I’ve heard the term “counterfascism� (and left the room). I am told that this has to do with offering a simple operational definition for a notion – causality – that has defied a concise discourse for a few centuries too many. How can humble statistics propose a cure where respectable philosophy rails in confusion?

The second, more serious, issue relates to how far we want to go in dealing with the unobservable. The potential outcomes framework clearly and avowedly locates causal effects in the difference between potential outcomes, at least one of which remains unobservable (the “counterfactual' outcome). Direct observation of causal effects thus is impossible, although estimation is possible under certain well-defined circumstances. The exchange between A.P. Dawid (“Causal Inference without Counterfactuals�), Don Rubin, Jamie Robins, Judea Pearl, and others in JASA 1999 considers the problem at its most sophisticated. My conversations, shall we say, rarely reach such heights. But it’s eminently clear that many researchers are troubled to various degrees by admitting unobservable quantities into “science.� Positions here range from moderate empiricism to Vienna style positivism: “you either observe directly or you lie."

I’m in no place to offer solutions. But I do offer this complaint whenever the two issues are combined into a single charge--that counterfactualist potential outcomers are arrogant because they fancy themselves scientists when they deal in unobservable quantities. I’d say that the opposite is true: the potential outcomes framework of causality offers a cutting lesson in humility because it demonstrates the necessity of relying on unobservable (but not necessarily unestimable) quantities, not to mention strong prior theory, for a great many tasks dear to the scientific enterprise.