Every election puts the media at odds with Nate Silver, the celebrated out election forecaster who derides Washington punditry for being devoid of data.
But as the election comes to a close, the Huffington Post has written a lengthy criticism of Silver for trending his site’s forecast toward Donald Trump. And Politico reports that “Nate Silver Rages at Huffington Post Editor in 14-part Tweetstorm.”
That might be a polite way to put it, though:
This article is so fucking idiotic and irresponsible. https://t.co/VNp02CvxlI
— Nate Silver (@NateSilver538) November 5, 2016
The Huffington Post’s Ryan Grimm had accused Silver of “unskewering the polls,” a term that the right-wing had used before the Mitt Romney-Barack Obama election. Conservatives had claimed polls were skewed and then launched an effort to “unskewer” them. Of course, it turned out there was nothing wrong with the polls and Obama really would win.
Grimm said Silver’s forecasting model, showing the election tightening and Trump on the rise, is “causing waves of panic among Democrats.”
“He may end up being right, but he’s just guessing,” wrote Grimm. “A ‘trend line adjustment’ is merely political punditry dressed up as sophisticated mathematical modeling.”
Boy, did that set off Silver. Here’s the rest of his response on Twitter.
The reason we adjust polls for the national trend is because **that's what works best emperically**. It's not a subjective assumption.
— Nate Silver (@NateSilver538) November 5, 2016
It's wrong to show Clinton with a 6-point lead (as per HuffPo) when **almost no national poll shows that**. Doesn't reflect the data.
— Nate Silver (@NateSilver538) November 5, 2016
Every model makes assumptions but we actually test ours based on the evidence. Some of the other models are barley even empirical.
— Nate Silver (@NateSilver538) November 5, 2016
There are also a gajillion ways to make a model overconfident, whereas it's pretty hard to make one overconfident.
— Nate Silver (@NateSilver538) November 5, 2016
If you haven't carefully tested how errors are correlated between states, for example, your model will be way overconfident.
— Nate Silver (@NateSilver538) November 5, 2016
Not just an issue in elections models. Failure to understand how risks are correlated is part of what led to the 2007/8 financial crisis.
— Nate Silver (@NateSilver538) November 5, 2016
There's a reasonable range of disagreement. But a model showing Clinton at 98% or 99% is not defensible based on the empirical evidence.
— Nate Silver (@NateSilver538) November 5, 2016
We constantly write about our assumptions and **provide evidence** for why we think they're the right ones. https://t.co/IhLKXdxGGK
— Nate Silver (@NateSilver538) November 5, 2016
That's what makes a model a useful scientific & journalistic tool. It's a way to understand how elections work. Not just about the results.
— Nate Silver (@NateSilver538) November 5, 2016
The problem is that we're doing this in a world where people—like @ryangrim—don't actually give a shit about evidence and proof.
— Nate Silver (@NateSilver538) November 5, 2016
The philosophy behind 538 is: Prove it. Doesn't mean we can't be wrong (we're wrong all the time). But prove it. Don't be lazy.
— Nate Silver (@NateSilver538) November 5, 2016
And especially don't be lazy when your untested assumptions happen to validate your partisan beliefs.
— Nate Silver (@NateSilver538) November 5, 2016
He followed up with this defense of his rant:
When you go low, I go high 80% of the time, and knee you in the balls the other 20% of the time.
— Nate Silver (@NateSilver538) November 5, 2016
Trump has in the past proven a problem for Silver’s modeling. Silver issued an apology of sorts after the Republican primary, when FiveThirtyEight regularly dismissed Trump’s chances of winning. He has at times during the general election put Trump’s chances much higher than they are now, warning President Trump was a real possibility. The latest FiveThirtyEight election forecast as of 12:23 p.m. PT had Trump with a 35 percent chance of winning the election.
READER COMMENTS ( )