Harris getting fairly close to even split on favorability, per 538 tracking. Trump's improved a little following assassination attempt/RNC, has settled at about -9 net. Trump won in 2016 despite having worse favorability than Clinton (both were poor & lower than what we see now)

Another thing that bothers me: assuming non-college whites in Georgia and Wisconsin are the same. The cleavage people proxy for when dividing voters by race + education is largely about religion

high impact example of what can happen when someone conflates a probability statement reported as a percentage with an estimate of vote share. (this is why we use frequency statements, which do come with their own drawbacks but tend to produce this type of error less often)

Nate makes a crucial point: Just because you can go to a website and fill out a survey does not mean your submission is actually used when the company runs new numbers and publish a poll. Good firms verify respondents and have ways to check for BS. (Not saying they’re perfect.)

Imagine if we had election forecasts built using TikTok likes or X user bios… People don’t appreciate how good polls really are at what they do (ie compared to the alternatives)

Can You Guess How Americans Feel About Harris’s Platform? projects.fivethirtyeight.com/kamala-harris-…

Pope Gregory enumerated 7 deadly sins. He was wrong. There are actually 13

a wonky contribution to forecast uncertainty is specification uncertainty. forecasters make independently reasonable modeling decisions that result in different outputs, so you can think of any individual forecast as a draw from a _distribution_ of possible forecast outcomes

If you’re on polling Twitter you’ve probably seen some criticism recently of a pollster called ActiVote, which has been releasing a lot of polls conducted on their own smartphone app. So Nathaniel Rakich actually called them up to ask about methods: abcnews.go.com/538/smartphone…