From the man who runs 538 since Nate Silver left:
Another note on the polls: There is an abnormally high level of non-sampling error in polls right now. In Summer/Fall POTUS campaigns we normally see non-sampling error of around 1.5%—implying a total MOE of around 6% for the average poll. But extra error is about 2.5% today. 1/2
— G Elliott Morris (@gelliottmorris) February 20, 2024
Notably, accounting for this extra error does not meaningfully change the state of play — Trump has a slight ~2pt lead — but it *does* significantly decrease the confidence we have in his lead. This is an image of a model I'm currently testing, showing 90% uncertainty intervals pic.twitter.com/g6IQ57iuTa
— G Elliott Morris (@gelliottmorris) February 20, 2024
I'm a little busy working on big projects rn, but if I weren't I'd (1) look at what economic and political fundamentals suggest about Nov and then (2) compare forecasts from models factoring in polled matchups & name rec, candidate favs and ideology, and demographics of defectors
— G Elliott Morris (@gelliottmorris) February 19, 2024
Anyway, it's an overdetermined problem, and my humble opinion is that there are too many people slinging takes that are far too certain on the matter
— G Elliott Morris (@gelliottmorris) February 19, 2024