Nate Cohn

Nate Cohn @Nate_Cohn

A Montana poll! Daines+1 in the Senate Trump+8, 51-43 Tie in the US House Gianforte+7 for Gov -- We'll have a second opinion tmro from NYT/Siena https://t.co/4sUZAX4NST — PolitiTweet.org

Nate Cohn @Nate_Cohn

Anyway, take it together and I think there are a lot of little things that add up to us having somewhat more DK than other pollsters, but probably fewer DK than you think — PolitiTweet.org

Nate Cohn @Nate_Cohn

We do the latter and hold out DKs in weighting. I know that at least some other pollsters collapse DK into other, and since those respondents are likelier to DK on many other questions, it would have a big effect on the presidential undecideds as well — PolitiTweet.org

Nate Cohn @Nate_Cohn

A possible factor--and I genuinely don't know how prevalent this is--how do you handle dk in your weighting? For ex: let's say I weight by race to 70w/10b/10h/10o, but 5% DK. Do I collapse the 5 into 'other', and keep the same targets? Or do I weight to 66.5,9.5,9.5,9.5, 5 dk? — PolitiTweet.org

Nate Cohn @Nate_Cohn

I'm fairly confident that 2018 voters represent a much smaller share of our samples than other polls, based on the extent of the effort it takes to reach them. — PolitiTweet.org

Nate Cohn @Nate_Cohn

Relatedy: we go out of our way to sample low turnout voters, who haven't voted in recent elections and weight on past turnout. They're also more likely to be undecided. In our polling so far this year, 2018 voters are 5% undecided; those who didn't vote in 2018 are 8.5% undecided — PolitiTweet.org

Nate Cohn @Nate_Cohn

So another survey would be cutting out those 21/14/13 groups out entirely, while we do still assign them some probability to vote, increasing the undecided share of the sample — PolitiTweet.org

Nate Cohn @Nate_Cohn

Pct. undecided by intention to vote, taking all NYT/Siena 2020 polling together since September: Almost certain to vote 5.6 Very likely 7 Somewhat likely 21.8 Not too likely 14 Not at all likely 13 — PolitiTweet.org

Nate Cohn @Nate_Cohn

One factor is that we have a probablistic LV screen. So if you tell us that you're somewhat likely to vote, you still have a chance to vote--while another poll cuts you out entirely. In our polling, these folks are way likelier to be undecided. — PolitiTweet.org

Nate Cohn @Nate_Cohn

Ok, but even beyond that you might think that 6% is too high. IDK if that's true, but let's suppose it is. Here are a few other factors that at least could be involved. — PolitiTweet.org

Nate Cohn @Nate_Cohn

And yes, those folks who say they won't vote very well might, based on our history of looking at validated turnout. They might just leave it blank. Maybe they'll choose a candidate. IDK. But many tell us in the immediately prior question that they intend to vote. — PolitiTweet.org

Nate Cohn @Nate_Cohn

To take an example in Kansas president, we've got Trump 48.2 Biden 40.6 Jorgensen 3.6 respondent volunteers some other candidate: 1 respondent volunteers they won't vote for pres: .6 Actual undecided: 5.9 — PolitiTweet.org

Nate Cohn @Nate_Cohn

Second, there are usually a lot fewer undecided voters than you think! In this Kansas poll, it's 6 percent undecided in both the president and the Senate race--not the 12 or whatever that it looks like. — PolitiTweet.org

Nate Cohn @Nate_Cohn

Just for starters: we do push undecided voters, twice. If you don't know your presidential vote we ask 'if you had to decide today?' And then we ask a second 'do you lean'-type question. — PolitiTweet.org

Nate Cohn @Nate_Cohn

Just for starters: we do push leaners, twice. If you don't know your presidential vote we ask 'if you had to decide today?' And then we ask a second 'do you lean'-type question. — PolitiTweet.org

Nate Cohn @Nate_Cohn

I get this question a lot, and I have some answers but maybe not one that explains all of it https://t.co/bZuckeqkkh — PolitiTweet.org

Nate Cohn @Nate_Cohn

Republicans still hold a modest lead in deeply conservative Kansas, according to a new Times/Siena poll. President: Trump 48, Biden 41 Senate: Marshall 46 (R), Bollier (D) 42 https://t.co/iJJyVlxlWz — PolitiTweet.org

Nate Cohn @Nate_Cohn

@arnonmishkin @SienaResearch IDK what other people do, but we do push leaners 2x--once in the initial question and then a separate leaner question. I think there are a lot of elements of our design that tend to increase non-major cand share, like the number of low turnout voters and listing minors — PolitiTweet.org

Nate Cohn @Nate_Cohn

@PatrickRuffini no i mean in your poll respondents — PolitiTweet.org

Nate Cohn @Nate_Cohn

Regardless of the ultimate outcome, I do think Democrats / their wealthiest megadonors will probably be kicking the… https://t.co/FanqXpXcfR — PolitiTweet.org

Nate Cohn @Nate_Cohn

Trump's approval rating in the 2018 exit poll was.... worse in TX than AZ/FL/OH. It's been highly competitive the w… https://t.co/tuvbYDNTe4 — PolitiTweet.org

Nate Cohn @Nate_Cohn

@b_schaffner @blfraga and what does this yield for over all self-reported race (w 2 part race/ethnicity question) in these states v. voter file race — PolitiTweet.org

Nate Cohn @Nate_Cohn

@PatrickRuffini how close does your self-reported race in the VRA states come to the Census-based estimate you would have used there? — PolitiTweet.org

Nate Cohn @Nate_Cohn

@b_schaffner @blfraga I do think this is about more than turnout, fwiw, like the Hispanic overreport in our poll — PolitiTweet.org

Nate Cohn @Nate_Cohn

Maybe I'll try and make some kind of self-report based estimate that models our post-VF weighted self-reported race in GA/FL/NC/SC as a function of the census-based estimates, and fall back on that as an alternative to just weighting on the census-based est — PolitiTweet.org

Nate Cohn @Nate_Cohn

@mattyglesias FL has self-reported voter file race. It would be an issue in TX as well, though I think here they might cancel out to an extent, as maybe we'd have more Hispanic voters but fewer black voters than we'd guess. I think the high black, no hispanic states (say MI) would be scariest — PolitiTweet.org

Nate Cohn @Nate_Cohn

And.... I'm only looking at partial data this AM, but I am looking at a state where the gap is a bit larger right now lol, and I'm hoping it shrinks when all of the interviews are completed. — PolitiTweet.org

Nate Cohn @Nate_Cohn

And indeed, that's what our poll shows: basically the same pattern and of a similar magnitude to our SC/NC/FL/GA data, when comparing our self-report to the census-based target. What will be tough: if we run into a state where the gap is quite a bit larger. — PolitiTweet.org

Nate Cohn @Nate_Cohn

I am not sure this is the right choice. But I think the hardest, coldest fact I have is that we *would* have been biased toward too many voter-file black voters in SC/GA/NC/FL and too few Hispanics if we weighted on self-report. I'd expect that nationwide if we had the same data. — PolitiTweet.org

Nate Cohn @Nate_Cohn

If you look at our last US national survey, you'll note it's 11% black and 13% Hispanic. That's a result of this. If we had weighted to the census, it would have been 12.5 Black, 10.5 Hispanic or whatever, and Biden would have eeked up to a 10 point lead (though gaining <1 pt) — PolitiTweet.org