Nate Silver's The Signal and The Noise

Nate Silver's been in the news a lot these last few days: looking at some stories, you'd think he'd won the election, not Mr. Obama. A statistician, his rigorous polling analysis riled, then humiliated political pundits, whose imaginary political horse-race was rejected by Silver's cold, hard numbers.

And what numbers they were. His "prediction"--though really just the most likely probability among many scenarios offered by his model--nailed the electoral college total on the night.

I've just read his book, The Signal and The Noise, and while it isn't as approachable as you might expect, that's what makes it interesting. Silver's insistence is that the quintillions of bytes of data at our disposal actually make prediction harder, not easier, and that we should not place too much stock in forecasts. Nor is it enough for us to to indulge brilliant but unanchored insights of the type offered by Malcolm Gladwell and Freakanomics. Planning for the relative probabilities of many possible outcomes is more useful than going all-in on specific predictions.

He takes us through our failures and successes. We're still lost at sea when it comes to disaster prediction, but (as he demonstrated last week) quite good at forecasting election results. Silva homes in on the work of mathematician Thomas Bayes as the right approach to understanding how statistics are most rationally interpreted to human ends.

By the end, though, it's surprising how unfulfilling it all seems. The Insight Industry's easy answers sure do go down easier than a dissertation on Big Data's ineluctable disinterest in them. Here, we are instead shown how prediction-makers screw up by treating their work as esoteric, by hiding uncertainties and inexactitudes to appear precise and decisive.

Which brings us back to the pundits, red-faced but curiously immobile in the stew of their failures. I don't quite buy that accepting these complexities will fill the chasm between their inanity and numerate reason. Though a place that seems rich with human promise, what emerges from the dark is a kind of interpretive accounting: much less interesting than whatever probabilities may finally be offered, and harder to sell than pompous old men just telling their peers how it is. It's such a difficult story to tell: even if you accept it, it's always less fun than the one where Joe Scarborough was made a total fool of by a brilliant young statistician.

I'm not helping, I know. But the pundits aren't going anywhere.

The Signal and the Noise: Why So Many Predictions Fail-but Some Don't [Amazon]