Liam Plunkett & the tragedy of social science

England's narrow defeat in the second test raises two important issues in the social sciences.

I'm thinking of the use of Liam Plunkett as a nightwatchman in the second innings - a role he played to the letter by virtue of getting out to the last ball of the day.

This poses the old question of whether nightwatchmen work.Charles Davis has argued that they don't on the grounds that tailenders used as nightwatchmen average as much as they do when they go in as tailenders.

But is this the best counterfactual? The nightwatchman's scores as a tailender might not be the right one because there's a selection bias here: nightwatchmen are usually used when batting conditions are tough - when bowlers are on top and the light is fading. To maintain one's average in such circumstances is creditable. And in England's case this week, had Moeen Ali had to go in on Monday night he might have gotten out and not played what was very nearly a match-saving innings. Using other counterfactuals. Anantha Narayanan says the use of a nightwatchman has been a "great success".   

This problem of counterfactuals is common in the social sciences, especially macroeconomics. It's sometimes said that we can't know for sure whether particular policies would have worked because we don't know what the counterfactual was. Others, though, say that one virtue of economic models is that they can generate plausible counterfactuals against which to judge policy.

But there's a second issue raised by the use of nightwatchmen. To see it, ask: what mechanism might cause them to work?

One plausible one here is the Pygmalion effect. If you make tail-end charlies believe they can bat by pushing them up the order, they will bat better, just as if you label kids as smart they'll do better at school. John Lambie, manager of Partick Thistle had this in mind when, on being told a player of his was so badly concussed that he didn't know who he was replied: "Tell him he's Pele and get him back on."

But there's an alternative mechanism - overconfidence. If you tell people they're good when they're not they'll be emboldened to over-exted themselves and so fail - as Plunkett did when he thought he could off-drive Herath.

This, though, proves a point Jon Elster made - that in the social sciences, "the opposite of a profound truth is another profound truth" precisely because there are countervailing mechanisms.
This point, of course, applies to very many political issues. Take just three:

 - Taxes. If you raise taxes on the rich, they might emigrate or evade the taxes or just stop working. Or they might have to work harder or longer to make up for the loss of post-tax income. Higher taxes might or might not deter effort.

 - Hospitals. Should we have more smaller hospitals, as Simon Stevens suggests? Maybe, because there are diseconomies of scale in bigger hospitals. Or maybe not, becuase there are economies of scale so smaller hospitals are inefficient.

 - Syria. If young British Muslims go to fight in Syria, will they return as radicalized terrorists? Or will they instead be killed there, or simply get their aggression out of their system and return as decent citizens?

In all these cases, anyone could make a case either way, and newspapers pay blowhards good money to do so. To be confident of such opinions, though, is to be guilty of what Jordan Ellenberg calls "false linearity" - a failure to see that offsetting mechanisms create hump-shaped curves.

What matters, though, is empirical evidence. But because of that problem of counterfactuals, this will sometimes be elusive. Many of those blowhards should remember Wittgenstein: "Whereof one cannot speak, thereof one must be silent."

Comments

[Luke]

"Charles Davis has argued that they don't on the grounds that tailenders used as nightwatchmen average as much as they do when they go in as tailenders."

Wrong measure. First, the nightwatchman role takes away a tailender's main scoring stroke, the wild swipe. Second, a nightwatchman's job is not to score runs but to stay there.

So balls faced without dismissal might be a better measure. But that's not perfect. A 50 ball overnight nought from a nightwatchmen is perfect. At other times you want your no11 to demoralise tired bowlers with a five ball cameo involving two sixes and and a couple of edges through the slips, before running himself out to prevent the bowlers getting any credit.

It is, as you say, hard to prove one way or another.

[Andy]

I haven't done a deep statistical analysis to prove this, but if I was going to send in a nightwatchman, I'd always send in a bowler who's had a poor outing with the ball. Chances are, he's hungrier to contribute to the team's cause than the bloke who's just taken 5 and already feels like he's done enough. Broad or Jordan would've been a better bet in the context than Plunkett.

[Jeremy Norton]

Great article you have here. I like your point of view. I agree with your closing remarks. More people should think this way.