Jonathan Chavez

3/ To put it mildly, these numbers do not line up with what we're seeing almost anywhere else. If CCES is right, and everyone else is wrong, we're in for a long few weeks. On balance, I would tend to trust other polls over CCES on this one if forced. Why?

4/ Polling old people online are hard. Really, really hard. Every large sample, national high-quality online poll that I do, the first concern I have is "how am I going to get/weight/stratify" the sample of those over the age of 50.

5/ No matter how hard I try on my sample stratification, inevitably I will find a new thing I did not consider and have to weight for it. In this case, I'm really hoping the 65+ population was not properly weighted for online propensity in the CCES, and also for Facebook usage.

6/ (As a side note, moving forward, I think all online pollsters should be weighting for Facebook usage by age cohort. If you get this wrong, you've ruined your poll). So that's my biggest concern - the olds, and like @redistrict, I'm going to be looking at Sumter County in FL.

7/ The Hispanic vote is also a pretty big wild card. It's not hard at all to explain why Hispanic men have moved towards Trump. We've seen in our years of polling on the military and patriotism, appeals to patriotism and fast paths to Americanization resonate with Hispanic men.

8/ It's clear Hispanics have moved towards Trump. Venezuelans, too. Maybe other Hispanics have as well, but the current polling showing that Hispanic men have moved hard towards Trump strikes me as near the outer limit of plausible movement.

9/ The 2016 evidence that exit polls and polling understated Clinton's advantage is pretty convincing. Additionally, I don't know how you accurately poll the Rio Grande Valley right now. They went through a COVID outbreak that was NYC-level bad. That's a big deal.

10/ So if we're in for a surprise tomorrow, the first place I'm looking is the RGV, and I think that potential error is two-tailed. I don't actually think it makes much of a difference anywhere but TX though (AZ is going to be won/lost among whites in the Maricopa suburbs).

11/ Finally on the shy Trump voter. I've spent countless hours in the data looking for it over the last few weeks, and at this point, it feels like a snipe hunt. Every theory of where they would show up when I look further into the data comes back null.

12/ Turnout is so boffo with GOP and DEMs you really can't crank the GOP turnout up enough to have it come from there. It's already cranked to 11. What you then have to do is say, well there's this cadre of IND/Unaffiliated where they would come from.

13/ There's a big problem with that, and that's younger voters who are strongly DEM leaning choosing to register as IND and not as DEM. Simply put, the IND universe is far more favorable to DEMs than in the past.

14/ NV is a prime example of this. Automatic voter reg has increased this effect, but a lot of those young INDs who have voted are voting like Ds (by mail) and polling young people is the one thing I'm most confident in. Those folks lean DEM.

15/ Add it all up: Misses are absolutely possible, but it's easier to miss a moving target than a stationary one and this race has not moved. Additionally, because old people have higher phone response rates, the risk of error from them is slightly lower.

16/ In '16, my biggest worry was that white working-class voters were being missed, and Clinton bore all of the downside risks. I don't feel that way this year. There are still some things I'm worried about.

17/ The CCES seniors numbers spooked me, and Hispanics (and to a lesser extent Asians) outside of Florida are a bit of a wildcard. But we can't forget: 18-21-year-olds being added to the electorate alone makes Clinton's vote totals work for a Biden win.