Thread by @Nate_Cohn

I see a lot of talk about this article, and I don't think it's really worth any attention my part. I will take this as an opportunity to highlight something fairly tragic about this election: we don't really know what happened in the detail we'd like

https://twitter.com/PostOpinions/status/1332692621052039172?s=20

Obviously, the election outcome is quite a bit different than in 2016--at least for the purposes of determining control of government.
But from the standpoint of the numbers, this was just not a very different election from 2016 in many of the states that matter most--like PA/WI

We're talking about, what, a ~2 pt swing nationwide and in many of the critical battleground states? Without fantastic data, it's just too small for us to decompose that modest movement in terms of turnout, changes in the composition of eligible voters, changes in attitudes, etc

We have *some* precise data. We have exact results by county and precinct. We have voter file turnout data of varying quality.
This can help us say a few broad things, like changes turnout and shifts that correlate with geography

With this data, we can see that Biden gained in well-educated suburbs, and especially traditionally GOP ones like ATL/KS/DAL/IND. We can say that Trump did well in Latino areas. We can say black turnout growth lagged non-black turnout surge, etc.

But there are actually a lot of things we can't say with this data. Age and gender are two great examples. They're just not very strongly correlated with geography.

We really have no hard, exact data that tells us about shifts among these groups. To figure this stuff out--and tbh, to increase our confidence in some of our geographic/ecological inferences earlier--we have to rely on survey data.

Unfortunately, this is a terrible, terrible year for survey data. The exit polls have always been bad and inappropriate for exacting analysis. This year they're both worse and incomparable to earlier exit poll data, rendering them.. even more useless

Usually, most of the exacting analysis is done with pre-election survey data! Unfortunately, this year's pre-election survey data is as bad as it's been in a competitive election in a very, very long time--and in ways that are quite material for unpacking the story

We know the national surveys were off by 4 points or so. Worse, we have reason to believe they're off in very specific ways--like a severe underestimate of Trump's strength among white working class voters

But we that's only what we know about, based on the limited, exact data I mentioned earlier. We have no idea whether they were also off by a lot on say, young voters or women.

I think we have every reason to assume they were off by a lot on older voters. But we don't have proof of that, and we don't know by how much.

Unfortunately, I don't think we're going to resolve these questions to the extent we could in prior elections, even when all of the data arrives, and that's even before considering the extent that the modest changes from 2016 raised the level of precision we needed to do well

With all of that cause for humility in mind, the assumptions underpinning the opinion piece cited at the outset are... really going beyond what we can do with the data. And it's pretty unlikely given what we do know