Nate Cohn

Nate Cohn @Nate_Cohn

Yeah, I think there are actually two flips there. MIL city v burbs; North v. East. Thanks for pointing out; will try nd correct. https://t.co/IAnL24fNLv — PolitiTweet.org

Nate Cohn @Nate_Cohn

As an aside, we had Smith+9 in MN -- the same as the presidential race, but with big underlying, regional differences. Biden better in the MSP metro; Smith running better in rural areas. A bit of downballot reversion to something more like the traditional MN map, as is typical — PolitiTweet.org

Nate Cohn @Nate_Cohn

NH primary was on Tuesday, the day we went into the field. Would have done it otherwise. https://t.co/vUHaNnUsVg — PolitiTweet.org

Nate Cohn @Nate_Cohn

Only 6% are undecided across the four states, on average, the rest who aren't supporting a major party candidate say they're voting for a listed minor party candidate or said someone else/won't vote. https://t.co/KJn7a56DKR — PolitiTweet.org

Nate Cohn @Nate_Cohn

As an aside, NH is Trump's worst state by recalled 2016 vote choice--it's a Clinton/Trump tie there, like IRL. Another reminder that this metric may not be as clear of a way to judge whether a poll is 'good' or 'bad' for a candidate as you might assume. — PolitiTweet.org

Nate Cohn @Nate_Cohn

I'm reminded of our Michigan poll last October, where we had some similar challenges on our first swing through, without the benefit of knowing just how the file would play out. It's possible that a more tailored design would yield better results than the default approach — PolitiTweet.org

Nate Cohn @Nate_Cohn

The third possibility is that it's not noise, exactly, but a reflection of some real difficulties. As I said last night, we really struggled there; the voter file was surprisingly bad. And it's our first time polling there--the last top battleground where we've never done a poll — PolitiTweet.org

Nate Cohn @Nate_Cohn

Another possibility is that it's noise. This is a small sample at N=450, and there's no way around losing precision at that point, even though we have decent party variables. Maybe if we polled it again, the exact same way today, we'd show a number that came closer to our prior. — PolitiTweet.org

Nate Cohn @Nate_Cohn

One is that it's just right, and NH is tougher than assumed. It's hard to say, given the absence of serious polling there. I'd note the last poll was Biden+8... but wasn't weighted by education. This poll would have shown Biden+7 without education weights — PolitiTweet.org

Nate Cohn @Nate_Cohn

State by state, MN/WI/NV were all within 1 or 2 points of what I would have guessed at the outset. NH is quite a bit better for Trump than I would have guessed, and there are three main possibilities worth considering — PolitiTweet.org

Nate Cohn @Nate_Cohn

Taken together, Biden's up 6 across the four states--which were about 1.2 points right of the country in '16. So I'd say it's basically in line with the national polling, which currently shows Biden ~ +7.5 — PolitiTweet.org

Nate Cohn @Nate_Cohn

The big catch: despite all of that, Trump still doesn't really even lead on average on law and order or violent crime. That's probably in part because they think Biden's better at unifying the country, handling protests, and think Trump encourages violence — PolitiTweet.org

Nate Cohn @Nate_Cohn

The polls suggest that a lot of the Trump law and order pivot succeeded... but with a big catch. --Voters split on whether law and order > COVID to their vote --They say Biden *supports* defunding the police --They say Biden hasn't done enough to condemn violent riots, by a lot — PolitiTweet.org

Nate Cohn @Nate_Cohn

Biden maintains a lead in Wisconsin and Trump's three best pick-up opportunities, among likely voters MN: Biden 50, Trump 41 NH: Biden 45, Trump 42 NV: Biden 46, Trump 42 WI: Biden 48, Trump 43 https://t.co/s4K1VIPozS — PolitiTweet.org

Nate Cohn @Nate_Cohn

If we had a contemporaneous account, like some panel polls out there, that would be different. But those facts alone are enough to preclude weighting on it. And our June polls--wildly good for Biden--also showed Trump doing *better* than '16 in recalled vote — PolitiTweet.org

Nate Cohn @Nate_Cohn

On one thing we can validate, for ex: people are absolutely horrible at reporting whether they voted. Half who didn't say they did, maybe because they don't take the question literally or not. I also know, based on hard data, that people removed from the file since 16 tilt Dem — PolitiTweet.org

Nate Cohn @Nate_Cohn

We don't weigh on it for two main of reasons: --We don't know what proportion of the current registered voters are Clinton/Trump. People die. Move. etc. --We don't trust people to recall this accurately and, more importantly, in an unbiased way https://t.co/ysRJczOxYG — PolitiTweet.org

Nate Cohn @Nate_Cohn

In general, the churn is focused on the Dem side. Young, nonwhite voters drop out at disproportionate rates, pbly just due to avg turnout and mobility, but also *join* the electorate at disproportionate rates. That's true even when young/nonwhite turnout go up. — PolitiTweet.org

Nate Cohn @Nate_Cohn

And already in mail voting requests, usually a high turnout group of voters, there are some states where ~15% of the requests are from 2016 nonvoters. If your assumption is that the 2020 electorate = everyone from 2016 plus a few, that's just not the way elections work — PolitiTweet.org

Nate Cohn @Nate_Cohn

And already in mail voting requests, usually a high turnout group of voters, there are some states where ~15% of the requests are from 2016 nonvoters. If your assumption is that the 2020 electorate = everyone from 2020 plus a few, that's just not the way elections work — PolitiTweet.org

Nate Cohn @Nate_Cohn

It implies a normal nonvoters turnout: even no change in the aggregate turnout rate usually involves quite a bit of churn. 10 percent of the likely electorate voted in '18 or the 20 primary not '16, for instance. https://t.co/bAdeWJK9zV — PolitiTweet.org

Nate Cohn @Nate_Cohn

Average RV/LV gap: Trump+1.3. Seems reasonable, though who knows — PolitiTweet.org

Nate Cohn @Nate_Cohn

College educated share of the likely electorate: MN: 44 (a bit higher than my prior tbh) NH: 42 NV: 36 WI: 36 — PolitiTweet.org

Nate Cohn @Nate_Cohn

This is interesting: self-reported 2016 vote among likely voters (balance is other or didn't vote '16). We don't weight on self-reported vote, unlike some other pollsters MN: Trump 41, Clinton 38 NH: Trump 41, Clinton 41 NV: Trump 37, Clinton 36 WI: Trump 43, Clinton 37 — PolitiTweet.org

Nate Cohn @Nate_Cohn

Sample size: MN 814 WI 760 NV 462 NH 445 The two-tier sample size approach here was the plan, given newsworthiness and the absence of party registration in WI/MN, which means you really count on larger sample to get yourself into shape — PolitiTweet.org

Nate Cohn @Nate_Cohn

Some methodological outtakes to tide you over: party ID/party reg of electorate the likely electorate: NV D36-R32/D39-R34 NH D22-D23/D28-R31 MN D33-R30/No party reg WI D28-R30/No party reg — PolitiTweet.org

Nate Cohn @Nate_Cohn

Data's in and final, probably looking at appx t-minus 1 hour. Go walk your dog or something https://t.co/vEW5D8PJrr — PolitiTweet.org

Nate Cohn @Nate_Cohn

RT @ByMikeBaker: The big cities with the worst air quality on Earth right now, according to the site IQAir. https://t.co/iUL1VyyqMi — PolitiTweet.org

Nate Cohn @Nate_Cohn

TLDR: we can't predict turnout with precision. It can cause certain kinds of biases. There is limited reason to think biases in turnout in '18 will be the same in 20, and therefore there's limited reason to assume the distribution of bias in our polls will follow same pattern — PolitiTweet.org

Nate Cohn @Nate_Cohn

One persistent issue, though, is that Latino voters are straight up likelier to be undecided at this stage. That will bias a poll toward Republicans, all else equal. Not 10 points, of course. But yeah I'd expect it. — PolitiTweet.org