Nate Silver, Trafalgar, And The Signal He's Lost The Plot
Trusting In Trafalgar? Hold On
So, part of my rebrand after November 3rd, 2020, is that I'm trying to be less of a dick in public. I'm trying to do this both for what can be described as personal advancement reasons - there's no need to be a dick, honestly - and also, like, what right do I have to be a dick to anyone after how shit my 2020 work was? On the whole, this hasn't been very hard for me, because I can still be snarky and funny without being gratuitous and mean, which I so often slipped into being in the past. And that mantra has led me to shelve a couple of columns that I would have previously written, because they were just dickish. And then I saw that 538 gave Trafalgar an A- pollster rating, so buckle in.
This is the least defensible thing 538 has ever done with one of their political quant analyses, and while I totally understand how it happened, it's infuriating. Trafalgar, for those who need the reminder, are the folks who had Brian Kemp winning Georgia by 12%, Ted Cruz winning Texas by 9%, who said "it's just not happening" in relation to Biden winning Georgia, who predicted Trump wins in Nevada and Arizona and Michigan and Pennsylvania, and who claimed that Trump was going to get something like 25% of the Black vote in Michigan the one time they ever released vote intention by race data.
This screenshot, stolen (with permission, to be clear) from this immaculate takedown of Trafalgar, was the final map put out by the head of Trafalgar on November 3rd. This is not a serious polling firm, it is a version of punditry wrapped up in pseudo-science. If you pick the Kansas City Chiefs every single week against the spread as a rule, and for 6 weeks they just destroy everyone, you're not smart, you're lucky. All Trafalgar does is bet on a pro-GOP polling miss and then if it comes, they're made men. If it doesn't, as it didn't in 2018 in Georgia and Texas, they just lie and claim that Cruz and Kemp had candidate specific problems that were obvious - except, not to you, because you had them winning by 9 and 12, respectively.
But my problems aren't with Trafalgar, really, because they're just unserious grifters trying to make a buck off the gullible and the unknowing. My problem is with Nate Silver, who just fucking validated these clowns with an A- because he can't realize that Trafalgar aren't actually good at this, but just manipulating a market opening.
By deciding that the only metric that mattered is distance from the pin, Silver incentives both herding and calculated deviation from the average. It incentivizes herding because if you have a nine point lead in a race where the average is 5, and you want to protect your reputation, your incentive is to move down to a 6 to avoid being too far away, but you've also encouraged Trafalagar-style consistent biases, where being 4 points right of the average at all times can get you an A-, just because you bet on Black and Black came up. There is nothing good about Trafalgar's process, nothing repeatable or valuable in an inherent, non-results oriented way.
In the span of 4 days, I lost soccer bets because of goals in the 96th, 94th, and 94th minutes against me. Was that a sign that I'm a horrible gambler who doesn't know soccer? No, it means I had a bad four days, and anyone who knows anything about anything knows the difference. They were good bets, just ones that didn't work out. Whatever, it is what it is, and I moved on, confident that I didn't have a process error on my mind. This kind of analysis from Nate et al would just skip over the fact that it was some of the worst luck you could have and just see three losses, because they're just doing the most surface level analysis.
An average error calculation misses all the nuances of polling work, from how Trafalgar was releasing campaign internals without noting that fact, the way that their head was promoting claims of voting fraud, or the fact that they are very clearly conducting garbage "polling," given what we know of their crosstabs from the one time they released them. And the thing that gets me the most mad? The old Nate Silver would have seen through them and immediately realized that this doesn't pass a basic smell test.
This isn't even about Trafalgar for me, this is about Nate Silver. Nate was - and still can - be a brilliant member of the class of public intellectuals, but man sometimes his shit doesn't pass a basic smell test. His model had Joe Biden winning Georgia while the GOP were favoured in the Georgia 7th, the kind of result that shows you don't know how to correlate your models properly. Their models did better than my predictions because they just shifted their modelling rightwards because of an economic and incumbency fundamental that said the race should be a tie, and so two errors - a polling miss and a broken fundamentals effect - cancelled each other out to some degree. Is that smart? Of course not, they just got lucky that their errors partially cancelled out.
The old Silver would have known immediately that these sorts of problems were signs of bad process - of noise, one might say, distracting from signal. The current Nate Silver is willing to accept the good faith of a pollster who advertises their lack of it. He has lost the plot, unfortunately. The argument his defenders would say is that his formula gave them X so they can't just arbitrarily make it Y, to which my response is a loud, long sigh. Of course they can, these are the galaxy brain geniuses that put the size of New York Times front page headlines into their uncertainty calculations. They make shit up all the time.
The name of the first piece I ever wrote about Trafalgar was Rebellion (Lies), a play on the Arcade Fire song of the same name. I'm not sure if there's any rebellion left there - they're just straight up liars. And Nate Silver just gave them all the credibly that a FiveThirtyEight A- grade conveys, as clear of a signal as we'll ever get that Nate cannot find his way anymore. It's sad to see it, but Silver has lost the plot - and this is just the latest of many, many signals of that fact.