Interesting. Polling errors tend not to be autocorrelated across presidential elections, but I also think Trump does seem to be special. Trumpier voters are simply less likely to pick up the phone. In 2016, people noticed that polls that weighted by college education were more accurate. People weighted by college education in 2020, and the polls were ... even less accurate. I haven't heard why exactly, although the "Democrats were more likely to be quarantining" thing might be true. I would like to see what happens if you weight by recalled 2020 vote, particularly in Wisconsin. That can sometimes lead to inaccuracy, but might be justified.
I still think the sample size here is n=2 and based on that a lot of people think it's a 100% lock that Trump outperforms. That's probably too confident. The circumstances now are very different however, and pollsters trying to prevent last time could potentially over-correct. That said, I bet it's 70% likely Trump overperforms his polls on the whole and is maybe 50% likely to over-perform by at least a full point. Nate Silver said he wouldn't take -110 on T over-performing, but I think that's not confident enough. The 2022 and 2018 polls were not wrong b/c Trump wasn't on the ballot and Trumpier voters likely stayed home.
Interesting analysis. Looking forward to seeing what goes down in Wisconsin!
why do you think this apparent and persistent bias is not priced in into betting markets?
Interesting. Polling errors tend not to be autocorrelated across presidential elections, but I also think Trump does seem to be special. Trumpier voters are simply less likely to pick up the phone. In 2016, people noticed that polls that weighted by college education were more accurate. People weighted by college education in 2020, and the polls were ... even less accurate. I haven't heard why exactly, although the "Democrats were more likely to be quarantining" thing might be true. I would like to see what happens if you weight by recalled 2020 vote, particularly in Wisconsin. That can sometimes lead to inaccuracy, but might be justified.
I still think the sample size here is n=2 and based on that a lot of people think it's a 100% lock that Trump outperforms. That's probably too confident. The circumstances now are very different however, and pollsters trying to prevent last time could potentially over-correct. That said, I bet it's 70% likely Trump overperforms his polls on the whole and is maybe 50% likely to over-perform by at least a full point. Nate Silver said he wouldn't take -110 on T over-performing, but I think that's not confident enough. The 2022 and 2018 polls were not wrong b/c Trump wasn't on the ballot and Trumpier voters likely stayed home.