Nate Cohn

Nate Cohn @Nate_Cohn

RT @SeanTrende: My hot takes on contemporary Twitter discussions: (1) I think Trump is likely to lose. Maybe 1-in-5 chance he wins(?) (2)… — PolitiTweet.org

Nate Cohn @Nate_Cohn

@LeeMiringoff does the measures you've taken with respect to enhanced samples also mean excluding out of state telephone numbers? i wasn't clear on that from the ky experiment — PolitiTweet.org

Nate Cohn @Nate_Cohn

@kabir_here @CBSNewsPoll what's the relationship between the race ratings/colors and the MRP estimates? some, like TX/AK, don't seem to line up. and am i right in interpreting this to add up to a biden+10 national lead? https://t.co/0CJlnVqi9l — PolitiTweet.org

Nate Cohn @Nate_Cohn

RT @BenjySarlin: If you’d like to simulate the last time polls looked like this, Bob Dole’s 1996 campaign site is still up https://t.co/v6i… — PolitiTweet.org

Nate Cohn @Nate_Cohn

@LoganDobson Should be easy to tell — PolitiTweet.org

Nate Cohn @Nate_Cohn

Either way, we've got another week in the books where Biden is maintaining a commanding lead without really any serious evidence to the contrary. We haven't seen a lead so large last for so long since you were upgrading to Windows 95 — PolitiTweet.org

Nate Cohn @Nate_Cohn

With Echelon Insights coming in at Biden+15 and YouGov at Biden+10--both among LV--the apparent gap between live and online polling does't seem as stark as it did at a few points last week, especially if you toss in a few D+10 not D+15-type state poll results from CNN/NBC today — PolitiTweet.org

Nate Cohn @Nate_Cohn

@EdAsante77 regional weighting does very little to fix education — PolitiTweet.org

Nate Cohn @Nate_Cohn

Anyway, I think it's possible that something important may be going on here and the debate over education may have obscured it. I'd be curious to hear what @LeeMiringoff @maristpoll thinks here — PolitiTweet.org

Nate Cohn @Nate_Cohn

If so, this would certainly reduce the educational attainment of the sample (in the NYT/Siena poll of AZ, out of state numbers were 49% colplus) but come at a big loss of coverage and potential bias. Is that a factor? — PolitiTweet.org

Nate Cohn @Nate_Cohn

The solution is 'enhanced' records that include geographic information on the location of the telephone number, allowing you to call a WA area code who lives in AZ. Is Marist losing this group? Their study makes me concerned (see Treatment C: Cell) https://t.co/p5YW51B4KS — PolitiTweet.org

Nate Cohn @Nate_Cohn

Basically, RDD polls dial random telephone numbers. That was easy enough when everyone had the area code of their state. But now that many people have out of state area codes, how do you do an RDD poll without just calling everyone? — PolitiTweet.org

Nate Cohn @Nate_Cohn

Another possibility is that including enhanced records might be a good thing and improve coverage by allowing RDD pollsters to reach out of state numbers, and therefore Marist would be losing a large, mobile and well educated group — PolitiTweet.org

Nate Cohn @Nate_Cohn

Another possibility is that e enhanced records might be a good thing and improve coverage by allowing RDD pollsters to reach out of state numbers, and therefore Marist would be losing that relatively mobile and well educated group — PolitiTweet.org

Nate Cohn @Nate_Cohn

This appears to have the effect of significantly reducing the educational attainment of the sample, which Marist attributes to the higher SES of vendor-matched records, which certainly seems plausible. Are other RDD folks using only matched records? — PolitiTweet.org

Nate Cohn @Nate_Cohn

We don't do RDD so I can't really weigh in on this. But I'd be curious to hear from the RDD folks about some of the more arcane sampling choices described in the NBC/Marist memo, specifically abandoning 'enhanced' RDD sampling https://t.co/D5O8h9zhT6 — PolitiTweet.org

Nate Cohn @Nate_Cohn

Another possibility is that it's the result of a more dubious choice--basically, two wrongs canceling out to make a right. This seems unlikely, but if it's not weighting or luck then either Marist or the other RDD pollsters must be doing something different from the other — PolitiTweet.org

Nate Cohn @Nate_Cohn

There are basically a few possibilities. One is luck, but let's suppose that's not it. Another possibility is that this results from a good design choice. For instance, the NBC/Marist poll is weighted by income, which is correlated with education. Maybe that helps a ton? — PolitiTweet.org

Nate Cohn @Nate_Cohn

Earlier this year, I collected RDD data from CNN, ABC/Post, Monmouth, KFF. It came in at 44% with a degree among adults v. 32% weighted. Here, NBC/Marist poll lands at 35% among adults without weighting by education. So what's going on? — PolitiTweet.org

Nate Cohn @Nate_Cohn

Earlier this year, I compiled RDD polls from CNN, KFF, Monmouth, ABC/Post. They collectively came in at 47% colplus, unweighted, among registered voters and 44% among adults. The last Marquette poll was 46% unweighted — PolitiTweet.org

Nate Cohn @Nate_Cohn

That certainly checks an important box: this is not like the typical poll unweighted by education. But I have to confess that I find it very odd. I've gone through hundreds of polls, whether in xtbs or microdata. I can't remember seeing a sample that didn't need education weights — PolitiTweet.org

Nate Cohn @Nate_Cohn

One thing immediately stands out: despite not weighting by education, just 38 percent of registered voters in Arizona have a four-year degree. That's actually not too far off of the CPS-based targets for the college educated share of the electorate in Arizona — PolitiTweet.org

Nate Cohn @Nate_Cohn

NBC/Marist this morning has Biden+5 in AZ among RVs, which seems reasonable to me, but it's worth circling back to a methodological discussion from earlier this week, when Marist said they weren't weighting by education after a trial run in Kentucky https://t.co/TRapaFyi7p — PolitiTweet.org

Nate Cohn @Nate_Cohn

Or if you prefer, perhaps one could argue there's a noun-adjective gap between suburban/suburb. — PolitiTweet.org

Nate Cohn @Nate_Cohn

I like this, though one subtle question here is whether 'suburbs' are best understood in contrast to the 'cities' or 'urban areas.' A suburban part of a 'city' wouldn't count in the former but might in the latter. https://t.co/mkCxLi69EB — PolitiTweet.org

Nate Cohn @Nate_Cohn

@mayasweedler I wasn’t counting them, though there are individual cases where there’s a strong argument — PolitiTweet.org

Nate Cohn @Nate_Cohn

Density and education have independent effects on vote choice. White no college voters in the cities, for instance, are pretty competitive; conversely, so are Hispanic voters in rural areas https://t.co/uBIsaPmF0B — PolitiTweet.org

Nate Cohn @Nate_Cohn

@PaulJLockwood everything orange is suburban if it's not part of the central city (new york city/newark/jersey city) — PolitiTweet.org

Nate Cohn @Nate_Cohn

And we can drill into these distinctions when appropriate. Central city block groups with a density >10000 sq/mi are Biden 74, Trump 14 in our poll; those under that are Biden 59, Trump 28. — PolitiTweet.org

Nate Cohn @Nate_Cohn

And cann drill into these distinctions when appropriate. Central city block groups with a density >10000 sq/mi are Biden 74, Trump 14 in our poll; those under that are Biden 59, Trump 28. — PolitiTweet.org