North Carolina (NYT/Siena): Biden +4

[kwabbit]

They have a +6 Trump 2016 sample size, even though he won the state by 3 points.

Biden may be leading by more than 4.

Does no one know realize that recalled vote is always going to seem to oversample the previous winner? For whatever reason, people are not good at recalling their vote. People who did not vote will say they did vote, mostly for the previous winner, Trump, and some who voted for Clinton will say they voted for Trump. It sounds absurd that someone wouldn't remember who they voted for, but this is a phenomenon that is present in basically all polls that ask for recalled vote.

Given that Trump won by about 4 in NC, you would expect about a +6 or +7 on recalled vote. NYT/Siena, perhaps the best pollster in the business, has not been oversampling Trump voters on literally every poll because the recalled vote is more favorable than the 2016 result. They have the actual vote history of their respondents!

[kwabbit]

They have a +6 Trump 2016 sample size, even though he won the state by 3 points.

Biden may be leading by more than 4.

I don't understand why, but nearly every NYT/Siena poll has had Trump 2016 sample higher than what it really was and the GOP party ID edge, even in states where Ds have a clear party ID edge

Because recalled vote, if sampled correctly, will always overestimate the previous winner by about 3 points of margin. People are straight up bad at recalling their vote, but over a large sample they do so in a predictable way.

From what Cohn says, this is because Trump-leaning independents have begun identifying as Republicans. If the previous composition of the electorate was 35D/35I/30R, now it's closer to 35D/30I/35R. The independents have just shifted over. This is why Trump is getting crushed even more among independents than would be expected.