Nate Cohn vs. Nate Silver

[I spent the winter writing songs about getting better]

Cohn is far better.

[Alcon]

I like Cohn, but I'm curious to know your deep, surely methodology-based reasoning on this one.

[Kingpoleon]

I like Cohn, but I'm curious to know your deep, surely methodology-based reasoning on this one.

I'm guessing: "DUDEFEST>>>>>>>girls>guys"

[I spent the winter writing songs about getting better]

I like Cohn, but I'm curious to know your deep, surely methodology-based reasoning on this one.

That he hasn't been basically completely full of sh!t this election season? I made this because he got in a recent Twitter feud with Silver.

[Xing]

Cohn is better this year, since he actually seems to factor in state fundamentals.

[Alcon]

I like Cohn, but I'm curious to know your deep, surely methodology-based reasoning on this one.

That he hasn't been basically completely full of sh!t this election season? I made this because he got in a recent Twitter feud with Silver.

I don't think Silver has been totally full of crap this election season.  He blew analyzing the primary, but he's been pretty forthright about why and how he's changed his methodology.  I mean, look at the difference in how FiveThirtyEight and the Upshot's General election models treat uncertainty and volatility.  Silver actually seems to have done a solid job of that, while I'll say Cohn and a few others were a bit too 2008/2012 in terms of including uncertainty in their evaluation.  I didn't really watch Cohn during the primary season so I can't evaluate how he handled it vs. Silver.  But I think they're both worth reading and considering, and as much as I disagreed with Silver throughout the primary and think he may have been too committed to his theories, I'm not willing to blackball him, especially based on how reasonably he seems to have evaluated his mistakes.

[mencken]

The nomenklatura seem to have forgotten that Nate Silver's inaccurate assessment of the primaries was a result of him departing from unbiased number crunching and parroting conventional wisdom. He has since learned his lesson, unlike most of this board apparently.


Let us settle this question empirically. Here are the probabilities each analyst assign to each state as of this morning:




On Election Night we can compare the accuracy of their predictions by adding the squared error of each. For example, when Clinton wins Illinois, Cohn gets a demerit of (1-0.98)^2, while Silver gets a demerit of (1-0.94)^2. Add these together for all 50 states and see who has the most total squared error.

[I spent the winter writing songs about getting better]

Rhode Island is only 83% likely to vote Dem?

What the hell is wrong with this guy?

[TheDeadFlagBlues]

Write-in: Sam Wang.

[Sprouts Farmers Market ✘]

Rhode Island is only 83% likely to vote Dem?

What the hell is wrong with this guy?

It's like you don't understand how this works???

[Antonio the Sixth]

Silver provides a lot more detail about his methodology, and I appreciate that. Also, if I have to choose between a model that seems underconfident (Silver's) and one that seems overconfident (Cohn's), I'd rather be cautious and choose the former.

[mencken]

Bump:

Erc pointed out some flaws in my methodology, namely that many of the errors are correlated and it fails to distinguish a narrow victory from a landslide. Nevertheless, I think this is still a good heuristic measure of who had the better model six weeks out. Counting Michigan and New Hampshire as ties in light of their narrow margins, Cohn gets a demerit of ~2.74 compared to Silver's ~2.48. Giving Michigan and New Hampshire to the respective likely winners yields ~3.16 and ~3.03, respectively.

Even without the correlation effects Erc mentioned (although really that should be to Silver's advantage since he built interstate correlations into his model), it looks too close to definitively say which forecaster was better. However, the preponderance of evidence suggests Silver did a slightly better job of forecasting the uncertainty of his calls.