There was some conflicting info among white northerners at that point: MN/ME/NH were fine for Biden, WI was close, and rural OH/IA were worse than 2016. IDK how I would have sorted it out if forced.
But there was no way to know that this red wasn't bleeding to PA

I guess by that point we did have the AZ early vote, which would have kept the non-Rust Belt path alive. But it's not like Biden was an overwhelming favorite in WI at that point either, and I do think this was well before the needle came around on Georgia

The sequence is key. Early on, Trump beat our expectations basically everywhere in NC/FL, and black turnout in rural GA was meh.
We then had to wait a while for big numbers from Atlanta, which eventually made clear something very different happened there https://twitter.com/newliberals/status/1325196203353468928?s=20

And indeed, something very different did happen there--at least with respect to either 2016, our expectations, and controlling for demographics.

If the sequence was different, you can imagine that we would have quickly seen what was happening in Georgia, but might have wrongly believed that Biden was going to do better than he did in CLT/Raleigh

Anyway, I think it's the best information we had at the time. Even when the needle did go to Biden in Georgia my first reaction was... skeptical, given the less-than-2018 results we had generally observed in other Sun Belt metros, including fast trending TX/AZ

And the needle did know way more than we did. It knew how many remaining votes were heavily Democratic absentee. It knew the Election Day vote left in DeKalb was in black areas, and so on.

I think the absentee vote created so much uncertainty about PA/MI that I didn't really think about it in those terms. After 1AM or so, I spent quite a bit of time on PA, and by 4AM or so I concluded that the race was over to my satisfaction https://twitter.com/Wertwhile/status/1325200211447177216?s=20

This article was finished around that time or maybe thereafter, and I didn't think there was much room for it to be wrong https://www.nytimes.com/2020/11/04/upshot/pennsylvania-election-results-ballots.html

This graphic--the one published in the article--is not the only graphical representation of the PA vote that we constructed that morning...

The other version, which we did not publish but has been updating internally here since early AM on Tuesday, used county x vote method data and featured a semi-circular dial with a long outstretched arm

Now that it's all over and would have been right, it would have been incredible to publish. But it was a tough number: Biden at >95% to win but only up 2 points. And if it was wrong, it would have been hanging out there for several stressful days

I think the article and the table was the more responsible approach