Nate Silver

Nate Silver @NateSilver538

Yeah, people treating it as some Law of Nature that Republicans overperform their polls in MI/PA/WI need to recall that exactly the opposite happened in 2012. — PolitiTweet.org

Nate Silver @NateSilver538

RT @cjane87: My treatise: Donald Trump’s campaign is too dang Online. https://t.co/BOAstGu8EW — PolitiTweet.org

Nate Silver @NateSilver538

RT @jbarro: I think people overstate the extent to which this fight is about electoral politics. Biden doesn’t want to pack the court but w… — PolitiTweet.org

Nate Silver @NateSilver538

RT @galendruke: We discussed several of the ideas that constitutional scholars have debated on this recent podcast. One of the challenges… — PolitiTweet.org

Nate Silver @NateSilver538

RT @ClareMalone: Well, I guess my latest story is timed well to Biden's court comments out this morning... What's on the Democratic agenda… — PolitiTweet.org

Nate Silver @NateSilver538

We actually had HRC at ~70% on Election Day but also 30% chance are supposed to happen 30% of the time, that's what the probabilities are for. https://t.co/8oqu1OmvAz — PolitiTweet.org

Nate Silver @NateSilver538

We actually had HRC at 70% on Election Day but also 30% chance are supposed to happen 30% of the time, that's what the probabilities are for. https://t.co/8oqu1OmvAz — PolitiTweet.org

Nate Silver @NateSilver538

I understand the impulse to constantly make comparisons to 2016 but the whole point of a model like 538's is to not only make comparisons to 2016 but also to *all other presidential elections* for which we have useful data†. † 1880, 1936 or 1972 for various parts of the model — PolitiTweet.org

Nate Silver @NateSilver538

A map of where the candidates are >50% in our polling average. (Gray = neither candidate above 50). https://t.co/jrUKgB26b2 — PolitiTweet.org

Nate Silver @NateSilver538

The data in the Midwest has been a bit more confusing. You can find a few comparatively bright spots for Trump in WI and OH polls recently, but not many in MI or IA. It's probably mostly noise. — PolitiTweet.org

Nate Silver @NateSilver538

We also learned that high-quality polls in PA are continuing to cluster around a 6 or perhaps 6.5 point Biden lead, and high quality polls in Florida are +3 or +4 Biden, on average. That's what our model thought all along rather than being thrown off the scent by spammy polls. — PolitiTweet.org

Nate Silver @NateSilver538

Overall, though, this was a day that brought some clarity. One of Trump's worst days of state polling in a while, but one of his better days for national polls, closing what had been a gap between them. Both now seem to point to perhaps a 9 or 9.5 point Biden lead nationally. — PolitiTweet.org

Nate Silver @NateSilver538

Obviously a fairly awful result for Trump in Michigan, but these other ones he can live with given that Fox News state polls have actually had pretty good numbers for Biden this year. — PolitiTweet.org

Nate Silver @NateSilver538

So while one can see why this would reduce polling error (if voters lock in their choice earlier, that's nice for pollsters) and that's something to consider when evaluating the chance of a Trump comeback, I think there are still some real challengers for pollsters here. — PolitiTweet.org

Nate Silver @NateSilver538

The caveat (mine, not theirs) is that we haven't seen the huge partisan split in the mail vote in the past like we have this year. If somehow you get the mail vs. in-person calibration off, that could contribute to some pretty large polling errors. — PolitiTweet.org

Nate Silver @NateSilver538

A JP Morgan note finds states with more mail/absentee voting tend to have *less* polling error. I looked at this too and found that it didn't have much of an effect, although maybe polls were slightly better in these states. There's a big caveat though. https://t.co/6WGAnr57GU https://t.co/3sPcVg4vck — PolitiTweet.org

Nate Silver @NateSilver538

@NickRiccardi I mean... it was 9-10 points to the right of the country in the midterms. A different electorate in presidential years but maybe it's just getting a bit redder, other things held equal. — PolitiTweet.org

Nate Silver @NateSilver538

@NickRiccardi 1) The RCP average includes a lot of spammy GOP-leaning polls in Florida, whereas the higher quality polls in FL have shown Biden with a slightly larger lead. He's probably up 9-10 nationally and 3-4 in FL, which is not *that* big a gap. — PolitiTweet.org

Nate Silver @NateSilver538

How polls handle the early vote (Do they even ask about early voting? Do they treat early voters the same as other "likely voters"?) could contribute to some pretty big house effects down the stretch run here, I'd think. An especially important time to look at poll averages. — PolitiTweet.org

Nate Silver @NateSilver538

Fairly big registered voter / likely voter splits in this poll, which favor Biden. He's +5 among RVs in PA (rather than +10 among LVs) and +2 among RVs in FL in these polls (rather than +4 among LVs). — PolitiTweet.org

Nate Silver @NateSilver538

RT @laurabronner: A final debate poll, with @elena___mejia @bycoffe and @juruwolfe, to track how the debate affects the race. Since the s… — PolitiTweet.org

Nate Silver @NateSilver538

Iowa, Ohio and Georgia all now have almost exactly 50:50 odds in our forecast. Trump still a 2:1 favorite in Texas. https://t.co/MfPiotv2br — PolitiTweet.org

Nate Silver @NateSilver538

@joshtpm I mean, Texas is a state where I'd still expect Republicans to win the close calls, but also a state where it's pretty hard to project turnout. So I can see why he's nervous although he's *probably* OK. — PolitiTweet.org

Nate Silver @NateSilver538

With Suffolk (Biden +6) and Quinnipiac (Biden +8, but treat it more like +6 given their house effect) out in PA today, though, it does go to show that the higher-quality polls show a more solid Biden lead there than the spammy stuff that dominates certain other polling averages. — PolitiTweet.org

Nate Silver @NateSilver538

A bit of mean-reversion here. Quinnipiac had Biden up 13 in Pennsylvania in their previous poll, but Trump ahead by 5 in Texas. These are closer to where Q polls have usually been this year, i.e. maybe 2 points better for Biden than the averages in those states. — PolitiTweet.org

Nate Silver @NateSilver538

Getting in the weeds here, but why is winning NC an equally good sign for Biden as winning FL, even though FL has more electoral votes? Because NC's demographics more closely resemble the Midwest than those other states, so NC might tell us more about PA, etc. — PolitiTweet.org

Nate Silver @NateSilver538

That starts to depend more on which state we're talking about. Biden is 88% to win if he wins FL but loses AZ and NC. Biden is 89% to win if he wins NC but loses AZ and FL. But only 72% to win if he wins AZ but loses FL and NC. — PolitiTweet.org

Nate Silver @NateSilver538

One other question is what is Biden wins some of these states but not others. Biden is ~99% to win if he wins *any two* of FL, NC and AZ but Trump wins the other one. But what if he wins just one of the three? — PolitiTweet.org

Nate Silver @NateSilver538

I also wonder whether this accounts for some of the gap between the higher-quality polls and the lower-quality ones (higher-quality polls have been better for Biden lately). The higher-quality polls may have more sophisticated ways to account for early voting. — PolitiTweet.org

Nate Silver @NateSilver538

This is pretty interesting ... starting to see more likely voter models where Democrats *gain* ground relative to the registered voter version. — PolitiTweet.org