Nate Cohn

Nate Cohn @Nate_Cohn

@gelliottmorris @BrendanNyhan @ben_golub we can't be sure; that's the point — PolitiTweet.org

Nate Cohn @Nate_Cohn

@gelliottmorris @BrendanNyhan @ben_golub all reasonable, though i'd add that it may not be as consistent with the polarization theory as you say — PolitiTweet.org

Nate Cohn @Nate_Cohn

@gelliottmorris @BrendanNyhan @ben_golub at the moment, your mean estimate looks awfully close to being outside your 95% CI from march? your modal estimate (and second likeliest) seem to be. — PolitiTweet.org

Nate Cohn @Nate_Cohn

@BrendanNyhan @ben_golub @gelliottmorris very few of these wind up in election years from 1948-2016. they nonetheless happen, and i can think of many creative ways to try and deal with it (including using 1948-2016 approval data) — PolitiTweet.org

Nate Cohn @Nate_Cohn

@BrendanNyhan @ben_golub @gelliottmorris we have some obvious cases, though some are debatable i suppose: covid; 9/01-3/03, 8/90-3/91, i'd argue the late 60s, 1940-1945, 1914-1920, arguably the whole era from 1856 through 1876 — PolitiTweet.org

Nate Cohn @Nate_Cohn

@BrendanNyhan @ben_golub @gelliottmorris go through every month since 1856 and identify how often the econ-based model would have been a dubious way of thinking about the 'fundamentals.' there are a lot, even in cases where it maybe kinda woulda 'worked' anyway (including right now). — PolitiTweet.org

Nate Cohn @Nate_Cohn

@BrendanNyhan @ben_golub @gelliottmorris i'd take issue with the idea that we're talking about unprecedented events; we're talking about events that didn't happen very often in the 18 election months from 1948-2016. — PolitiTweet.org

Nate Cohn @Nate_Cohn

As a result, I do think it's reasonable at this stage to stick with your historic prior (slight-to-modest GOP LV edge) over whatever self-report says today (though I do think they're consistent with it). — PolitiTweet.org

Nate Cohn @Nate_Cohn

5) Self-reported turnout well before an election isn't terribly accurate. I can't find it now, but we showed a steady increase in accuracy in our '18 polling, where voters gradually got better at assessing their propensity to vote. Shouldn't be a huge surprise. — PolitiTweet.org

Nate Cohn @Nate_Cohn

In our national poll, for example, the voters with >90% turnout scores (based on vote history), but who *aren't* almost certain to vote: Biden+13, 48-35 R+6 party reg 34-40 More generally, just 53% of reg GOP who back Biden are 'almost certain to vote' — PolitiTweet.org

Nate Cohn @Nate_Cohn

4) One interesting thing to mull: could a bigger Biden lead involve a bigger RV/LV gap, as Biden expands his support into Republican-tilting voters who don't usually vote for Democrats and wouldn't be terribly excited to vote for one? Just a thought. — PolitiTweet.org

Nate Cohn @Nate_Cohn

If I look at every poll reporting both, the gap gets even narrower--but it does seem to be there. That's consistent with a pretty long history of a GOP edge among LVs with respect to RVs, and I don't think it should be a surprise. — PolitiTweet.org

Nate Cohn @Nate_Cohn

3) Although 5 is at the higher end of the spectrum, I do think the balance of evidence continues to suggest that there is a GOP-tilt LV/RV gap, at least nationally. NYT/Siena national was Biden+11 (51-40) w 'almost certain' to vote (not our LV approach btw), so that's a 3 pt gap — PolitiTweet.org

Nate Cohn @Nate_Cohn

2) That said, this is the second consecutive ABC/Post poll to show a 5 point LV/RV gap. The May poll was 10 with RVs and 5 with LVs. So I'd be somewhat less inclined to dismiss it as noise; perhaps there's a pattern here, at least for this pollster — PolitiTweet.org

Nate Cohn @Nate_Cohn

So I don't think there's anything odd about a Monmouth poll (N=400 total!) showing a big LV/RV gap while our poll of PA (N=650 total!) doesn't really show one. That's just going to happen. Feel free to average it, etc. — PolitiTweet.org

Nate Cohn @Nate_Cohn

1) In general, self-reported LV screens will be noisy. Think about it. You've got two appx. N=400 samples of Biden and Trump voters. For each, the 'almost certain' to vote proportion has that +/- 5 MoE. So there's nothing weird about a lot of noise in the LV-RV gap, poll to poll — PolitiTweet.org

Nate Cohn @Nate_Cohn

One interesting thing about the ABC/Post poll is that Trump fares 5 points better among LVs (which appear to be those who say they're certain to vote) than RVs. A few thoughts — PolitiTweet.org

Nate Cohn @Nate_Cohn

If it holds up, it's actually a reversal from much of the last year+, when Trump's approval was better in the phone polls than online — PolitiTweet.org

Nate Cohn @Nate_Cohn

One emerging pattern at this moment is that Biden has a larger lead in telephone v online. Most recent post-6/1 live interview polls of RVs: Quinnipiac 15 NBC/WSJ 11 Monmouth 12 Suffolk 12 Marist 8 CNBC 9 NYT/Siena 13 Fox 12 CNN/SSRS 14 Avg 11.8 — PolitiTweet.org

Nate Cohn @Nate_Cohn

One emerging pattern at this moment is that Biden has a larger lead in telephone v online. Most recent post-6/1 live interview polls of RVs: Quinnipiac 15 NBC/WSJ 11 Monmouth 12 Suffolk 12 Marist 8 CNBC 9 NYT/Siena 13 Fox 12 CNN/SSRS 14 Avg 111.8 — PolitiTweet.org

Nate Cohn @Nate_Cohn

One fairly clear pattern at this point is that Biden's lead has a larger lead in the national telephone polls. Most recent post-6/1 live interview polls of RVs: Quinnipiac 15 NBC/WSJ 11 Monmouth 12 Suffolk 12 Marist 8 CNBC 9 NYT/Siena 13 Fox 12 CNN/SSRS 14 Avg 111.8 — PolitiTweet.org

Nate Cohn @Nate_Cohn

@tysonbrody i mean, at this point, repeating the 10 they had in May counts as good, relatively speaking — PolitiTweet.org

Nate Cohn @Nate_Cohn

I suppose we finally get ABC/Post tomorrow morning. Those COVID approval numbers for Trump were quite bleak on Friday--even worse than ours. It'll be interesting to see how that translates to the horserace. — PolitiTweet.org

Nate Cohn @Nate_Cohn

@gelliottmorris @NateSilver538 @jbview @DanRosenheck @collprof @Michael_Cobb68 Post civil war polarization arguably a lot more like today than 1948-1996, along with the events — PolitiTweet.org

Nate Cohn @Nate_Cohn

@DanRosenheck @gelliottmorris @NateSilver538 @collprof @Michael_Cobb68 in all sincerity, i think there's a fine case the fundamentals would be 'good' for a president who handled covid/protests with replacement level seriousness. i think his approval on the economy kinda speaks for itself. — PolitiTweet.org

Nate Cohn @Nate_Cohn

@DanRosenheck @gelliottmorris @NateSilver538 @collprof @Michael_Cobb68 i really think we should have some humility on the fundamentals, now and in general. is 'mass unemployment' really even remotely close to the center of trump's problem right now? — PolitiTweet.org

Nate Cohn @Nate_Cohn

@gelliottmorris @NateSilver538 @DanRosenheck @collprof @Michael_Cobb68 and that's ultimately where the road ends: you have to be able to justify why, based on limited (but often still meaningful in 00, 08, and yes 16) variance over 5 elections, you would expect a 9 point swing just once over the next 100. i think that's a bold case — PolitiTweet.org

Nate Cohn @Nate_Cohn

@gelliottmorris @NateSilver538 @DanRosenheck @collprof @Michael_Cobb68 yes, i certainly believe it worked well: there wasn't much variance in the last five elections; therefore, a model supposing little variance and no tail risks would do just fine and, yes, better than one that allows for such risks. but does that mean those risks don't exist? no — PolitiTweet.org

Nate Cohn @Nate_Cohn

@gelliottmorris @NateSilver538 @DanRosenheck @collprof @Michael_Cobb68 i think it's defensible to add a polarization term--it could have improved an '18 house model, for instance--tho it comes with risks. if your version yields *99%* on a 9 point lead in july--based on variance in, what, 5 polarized elections--then you may be going too far — PolitiTweet.org

Nate Cohn @Nate_Cohn

@gelliottmorris @NateSilver538 @DanRosenheck @collprof @Michael_Cobb68 yes, all elections have equal weight, but less polarized elections will have less weight in determining variance for 2020, just as, to take an example, 'black' respondents do not effect the mean for 'white' if you control for race in a linear model, despite equal weight — PolitiTweet.org