Nate Silver

Nate Silver @NateSilver538

RT @ABCPolitics: .@NateSilver538 says “things look a little better on the margin for Trump, but there is one number that should really worr… — PolitiTweet.org

Nate Silver @NateSilver538

I get colleges are in a *real* tough spot. If you went through the rest of the sequence without the student-blaming, I'd probably ¯\_(ツ)_/¯. But blaming 18-year-olds who just had their senior year ruined for not completely self-isolating (while paying full tuition!) is gross. — PolitiTweet.org

Nate Silver @NateSilver538

What *some* colleges are doing isn't so far from a bait-and-switch. Lay out implausible conditions under which college could commence w/o COVID spread. Get tuition on premise of an in-person experience. Blame students when the conditions are inevitably violated. Then go online. — PolitiTweet.org

Nate Silver @NateSilver538

If I were Biden I'd think it was positive ROI to drop like $100K on ads in Albany and Buffalo in the hopes that Trump would spend $5M in return. — PolitiTweet.org

Nate Silver @NateSilver538

Bad use of polling. — PolitiTweet.org

Nate Silver @NateSilver538

@BenjySarlin @jbarro @IChotiner I think this is right, and I also wonder how certain experts would react. — PolitiTweet.org

Nate Silver @NateSilver538

@dandrezner Yeah, nothing really. And there haven't been very many state polls in general lately. — PolitiTweet.org

Nate Silver @NateSilver538

So there's not yet much suggestion of a major bounce (maybe a minor one) although I'd also note that telephone polls typically show bigger bounces after news events than online polls like Morning Consult. I don't know that we'll get any hi-quality phone polls until after the RNC. — PolitiTweet.org

Nate Silver @NateSilver538

The only post-convention poll we have is from Morning Consult, which had Biden +9 vs. +8 beforehand. There's also a Redfield & Wilton poll conducted Wed. and Thurs. *during* the convention that has Biden +10, vs. +7 beforehand. — PolitiTweet.org

Nate Silver @NateSilver538

Biden has inched up to +8.8 over Trump in our national polling average, his biggest lead since July 20, although not yet much change since before the conventions. https://t.co/cy51vc5isJ https://t.co/y5wWyysWUP — PolitiTweet.org

Nate Silver @NateSilver538

@BenjySarlin But I worry a lot more about the long-term costs of prolonged lockdowns than a lot of other people do, I think. I also worry about the morbidity costs of COVID (i.e. "long COVID") more than most others, though. — PolitiTweet.org

Nate Silver @NateSilver538

@BenjySarlin No that's fair and to be sure there was a lot of magical thinking about how soon these things could be put into place given our federal government. — PolitiTweet.org

Nate Silver @NateSilver538

@BrendanNyhan A model that uses approval ratings is not a fundamentals model; approval is the dependant variable! — PolitiTweet.org

Nate Silver @NateSilver538

@BenjySarlin I have complicated feelings on this. I kind of wonder if we presumed that people wouldn't put up with it, we might have made more progress on strategies like testing and tracing. Instead locking things down/keeping them closed response became the default response in blue states. — PolitiTweet.org

Nate Silver @NateSilver538

But these "fundamentals" models have been terrible when used to actually predict elections when they didn't know the results in advance. Rather than using them, you literally would have been better off just guessing the election would be 50:50. https://t.co/uzhRzj9s2F — PolitiTweet.org

Nate Silver @NateSilver538

Of course, rather than accepting some intrinsic degree of ambiguity and uncertainty, you can test out 300 different model specifications on a sample size of n=10 and find something that "fits" the data. And even manage to convince yourself that you've found a persuasive "story". — PolitiTweet.org

Nate Silver @NateSilver538

1992? The economy went into recession in 1990. But it ended by March 1991, and the economy was recovering, though slowly (especially in data published at the time vs. later revised). Bush was coming off the extremely popular Gulf War. Not at all obvious why he collapsed so badly. — PolitiTweet.org

Nate Silver @NateSilver538

1988? Bush going for a 3rd straight term for Republicans amid an average economy. VERY similar to 2000 and 2016, which were photo finishes. (Incumbent party lost the Electoral College but won the popular vote). No particular reason to expect a Bush landslide. — PolitiTweet.org

Nate Silver @NateSilver538

See, this is why the "fundamentals" are kind of overrated. It's not at all obvious that they would have favored Bush in 1988 or Clinton in 1992 **unless you know the outcomes in advance**. — PolitiTweet.org

Nate Silver @NateSilver538

You might say "it *was* a real struggle" and I guess it was in some states but others locked down for months. And there's some tension between how lockdowns poll (popular!l) and people's revealed preferences. But overall Americans have been more lockdown-tolerant than I expected. — PolitiTweet.org

Nate Silver @NateSilver538

I was sort of the opposite, LOL. I kind of assumed during that time Americans wouldn't put up with the sort of lockdowns they had in China. Then by March I thought lockdowns would last a few weeks then it would be a real struggle to get people to abide by them after that. — PolitiTweet.org

Nate Silver @NateSilver538

As a result of the adjustment, Biden has gone from a 72.6% chance of winning the Electoral College to a 72.5% chance in our forecast. So, yeah, not a big deal. But we wanted to disclose it nonetheless. — PolitiTweet.org

Nate Silver @NateSilver538

tl;dr: a good rule-of-thumb is that partisan polls are biased by 3-4 points in favor of the party the group supports (closer to 3 points for POTUS and 4 for Congress). Our presidential averages had been adjusting them by 5 points, though, so we've scaled that back slightly. — PolitiTweet.org

Nate Silver @NateSilver538

That's because the previous adjustment was based on our study of campaign *internal* polls released to the public (i.e. if the Biden or Trump campaigns released a poll). Those tend to be more biased than the broader universe of partisan-affiliated polls as we now define them. — PolitiTweet.org

Nate Silver @NateSilver538

However, the adjustment our model was making to these polls was overly harsh: about 5 points (so e.g. a Biden +7 poll from a D partisan group would be treated as Biden +2), when the adjustment should have been closer to 3 points. — PolitiTweet.org

Nate Silver @NateSilver538

As an aside, partisan polls of the presidential race have been more common this year. They make up about 20% of our state poll dataset, for instance. Previously, you saw plenty of partisan polls in Congressional races, but relatively few for POTUS. — PolitiTweet.org

Nate Silver @NateSilver538

But some ostensibly partisan polls play it straight down the fairway, while some ostensibly nonpartisan polls have pronounced biases. So the model will shift away from the prior as it collects more data. It mostly affects cases when we don't have much data from that pollster. — PolitiTweet.org

Nate Silver @NateSilver538

Previously, we had not used partisan polls in our averages at all. But we began incorporating them in 2018 and have now carried that over to 2020. Our model starts out with a prior that these polls are biased, unlike partisan polls, where the assumption is that they're unbiased. — PolitiTweet.org

Nate Silver @NateSilver538

Ready for a boring early Friday night thread? We're making a verrrrrry small tweak to our polling averages, slightly reducing the adjustment we make partisan-affiliated polls. Here's how we define a partisan poll: https://t.co/37N8bDqsnB https://t.co/emn6OHrPFR — PolitiTweet.org

Nate Silver @NateSilver538

@ClareMalone @BrianDSilver1 Y — PolitiTweet.org