Welcome Daily Dish  and Marginal Revolution readers!

For your first blog posts, you want to pick subjects that are snappy, topical, concise.  Something sexy.  That's why I've chosen to write today about....causal inference.  (I'll return to the question I raised at the end of my first post later on.  When I have an actual readership, I'll try to stick to my word better.)

More often than not, this blog will be about fact-checking arguments and reviewing the right way to measure things.  But other posts will assess various claims made in think tank and academic studies about how X affects Y.  How does family structure affect child outcomes? How does ideology affect voting?  How does discrimination affect the wage gap between men and women?  You ought to know up front how much of a grinch I am when it comes to establishing causality, and why.

There are academic treatises on the proper way to infer causal relationships, but I want to lay it out in a way that will be accessible to all y'all, whether you currently own a graphing calculator or not.  So with the left-brained in mind, I'm going to initially frame the discussion using a dramatic depiction of a debate between an influential Harvard statistician named Don Rubin and a dead physicist named Sir Isaac Newton:

(Scene: Rubin and Newton shooting pool.  ZZ Top's "Tush" plays on jukebox.)

Newton: Did you catch the Real Housewives of New Jersey reunion the other night?

Rubin: (looks up, eyeing Newton suspiciously)  Two ball in the side pocket.

Newton: Nice shot.  (profane word muttered under breath)

Rubin: So to pick up where we left off the other day, I still say that we can only talk about cause and effect if things turn out differently than they would have absent the cause.  For instance, saying that my incredible talent in pool is the "effect" of my mother making me watch all those billiards tournaments on ESPN is saying that if she had not made me watch all those tournaments, I would be stinking up the pool hall.  If, however, the counterfactual—what would have happened absent my mother making me watch billiards—would have been that my father would have made me watch the tournaments anyway, then we can't really say that my mother caused me to be the pool player I am today.  Even if she had not made me watch pool, I would still be just as good as I am.  My mother's efforts would have had no "effect" on my pool talent—they would not have "caused" it.

Newton: Bullstein!  Look, you just sank the two ball, right?  Your brain sent a signal that led your muscles to apply force to the cue stick, that force was transferred to the ball, and the ball went into the pocket—your actions caused the effect.  If the counterfactual would have been that I would have knocked the ball in myself, that doesn't change the fact that what actually happened was that you caused the two ball to go into the corner pocket!

Rubin: Six ball in the corner pocket. (misses)  OK, let me try from another angle (the argument, not the shot).  Let's think in terms of a pressing social policy question—whether divorce is harmful to children.  I say that it all depends on the counterfactual.  A child experiencing a divorce may end up worse off than where she was before the divorce, but if the counterfactual would have been that her parents would have been miserable, combative, and disruptive then she might not have turned out any better.  In that case, divorce would have had no effect on her.  A policy that had prevented her parents from divorcing would not have helped her!

Newton: That may be, but that doesn't mean the divorce didn't actually cause the child do turn out worse.  It just means that unhappy parents would have caused the child to do worse in the absence of divorce causing it. (scratches, yells unprintable word)

Rubin (sighs in exasperation): OK, let me try one more time with a medical example.  You run a clinical trial.  You give half the participants a little blue pill and everyone else gets a placebo.  The people who get the little blue pill all lose their hair.  We know that the blue pill caused the hair loss because we know that if they had taken the placebo instead, they'd have their hair just like the people who really did take the placebo.  That's the counterfactual.

Newton (cracks pool stick over knee): Dammit Rubin!  Run another trial!  Give half the people the little blue pill and give half of them an otherwise identical red one!  Everyone loses their hair!  Do you then say that the blue pill did not cause hair loss?  Of course not!  Both the red pill and the blue pill cause hair loss!

Rubin: but--

Newton: I am NOT DONE.  Sure, the logic of experiments—getting assigned to a treatment or control group randomly—helps to establish causality, but the causality either operates or it does not—independently of how we identify it.  The girl who is harmed by divorce does not give a flying (censored) that she would have been harmed by fighting parents if the divorce hadn't happened.  (pacing the pool hall now) Counterfactual-shmounterfactual!

(Chorus line enters stage left.  ZZ Top enters stage right.  All sing):  Counterfactual!  Shmounterfactual!  What the girl cares about is the Actual!....


After intermission (which may last a couple of days—again, day job...), I'll come back around to the point.  The gist is that social science has over-corrected in addressing its previous inattention to counterfactual logic (which is crucial for policy questions and powerful in certain applications) and is now uncritically adopting the views I put in Rubin's mouth above (which I'm caricaturing somewhat).  On the other hand, at least Rubin's acolytes are able to convincingly establish causality in a limited number of contexts.  Most of those suspicious of him have no idea where to begin in establishing causality outside those contexts.

And for the record, I am not high on poppers.