From the Democratic firms' assessment of 2020 polling error: I find this piece particularly vindicating. I've been saying from the start that likely voter modeling is part of the issue.

https://www.democracydocket.com/2021/04/revisiting-polling-for-2021-and-beyond/

People ignored me and preferred to listen to louder [male] voices saying it's all sample. It IS *also* sample. But it is also that we cannot predict a population that doesn't exist yet!!!

I'm glad this is starting to get out there. Never rely on people who don't show their work.

If campaign polls using extensive histories have these problems, it's likely magnified in public polls not using registration data.

This point is also extremely important: We don't want to overcorrect. There's something about Trump on the ballot here, and we'll need to know how that impacts us all moving forward.

To clarify, they do say they think the sample/measurement error is larger than the turnout error. My point has been that it's both, and I've been written off for saying that. They are also saying it's both.