A Righty On Why The Criticism Of Nate Silver Is Generally Stupid

Pay attention because this is one of the rare occasions in which I’m on the same side as Paul Krugman.

Kruggie (as I like to call him), or Little Pauly Krugster (as I also like to call him), incurred some righty guffaws for writing this column in which he wrings his hands over the criticism of Nate Silver’s ’538′ electoral projections. These complaints, he frets, represent a ‘war on objectivity’:

This is really scary. It means that if these people triumph, science — or any kind of scholarship — will become impossible. Everything must pass a political test; if it isn’t what the right wants to hear, the messenger is subjected to a smear campaign.

Oh, my stars and garters! The horror! The masses might not automatically believe numbers spit out by Smart People like Pauly K.! That’s so ‘really scary’ that I just well I never!

Now, as you can see, I don’t quite share Krugman’s obsession with erecting barriers around the Smart People class and sheltering their output from all criticism. Indeed, as you know, I find the campaign of Enforced Belief In Official Numbers more than a little creepy, and I believe that All large calculations are wrong anyway.

So why am I on the same side as Krugman? Because the criticism of Nate Silver and his oeuvre has just gotten stupid.

Maybe this is still just sour grapes from my having been embarrassingly hoodwinked by some of the initial righty poll-debunking, but the more articles and ‘exposes’ of Nate Silver and his ‘bias’ I read, the more I scratch my head at their content (or lack thereof).

To get one thing out in the open, no, I have not studied Nate Silver’s model. I am sure I don’t know many of the details. However, I can imagine how I would do it if I were building such a model – and all indications are, Silver’s basically just gone and done the exact same thing I would do.

Maybe part of the problem is a misconception of the purpose of such a model. Implicitly, the purpose of the model is not to tell people what to think about who will win, or even necessarily to predict the winner. The purpose of such a model is to translate public polling information into a best-estimate of who will win, consistent with that data. It is a type of equivalence: because the polls say X, that is equivalent to a future in which Obama has a Y% chance of winning.

It is a translation operation such a model is performing: it is just translating the data from polls (Obama leads by X% in state S) into the language of Who Will Actually Win The Presidency, constrained by that data (in whatever way).

In that sense, the lefties are absolutely correct when they look at the gripes about Silver and see them as righties idiotically arguing with data. Does Nate Silver control all these national and state polls from all these different organizations? No? Because (presumably) all he’s doing is putting them all inside a giant Monte Carlo simulation and letting it spit out the results.

Now. Does this mean there is no room whatsoever to criticize Silver’s model? Of course not. Any such model will inherently depend on lots of subjective choices about parameters, factors, distributions, fudge factors(?), how much weight to give to this data point vs. that data point, etc. etc. Sure. And Silver’s model is surely no exception. This inevitability is part and parcel of why I think All large calculations are wrong. Make no mistake, Silver’s model is ‘wrong’.

But that goes without saying. (All large calculations are wrong.) The question is not whether his model is wrong (it is), the question is whether it is less-wrong than viable alternatives. What are the viable alternatives? Jonah Goldberg’s model?

As we know, Nate Silver’s model predicted the 2008 outcome quite well. This is undeniably a point in his favor. If righties want to criticize Silver’s model (I don’t see why – what’s the point?), they have to confront its earlier success.

But more importantly, to make a convincing case for rejecting the model’s output, they’ll need to drill into the actual details and subjective choices inside his model (whatever fudge factors, etc.), and not merely show that they are subjective, but show that other choices are both more rational and would lead to a different output.

This is, with very few exceptions, what I am not seeing from the right. Which is why the criticisms are stupid.

And even if you do make a valid criticism that is based on critiquing some tangible detail of his model construction, all you’ve really done is established the principle that: Someone else could construct a model where they choose fudge factor F2 instead of Silver’s fudge factor F1, and it would show Romney as the predicted winner.

Which is swell, but ultimately doesn’t say all that much. These are all just models, attempts to translate (imperfect) data into an (inherently imperfect) estimate of an event related to that data. The only real test of such models, is reality.

In other words, Election Day is an experiment. It might indeed allow us to reject the null-hypothesis that Nate Silver’s model is a reasonable one. But until then, when all we have are all these righty columns, blogs, and tweets griping about it, the hypothesis seems alive and kicking to me.

About these ads