As far as I'm concerned, election polls have become a worthless byproduct of a different age (since 2016!). Just like buggy whips.
Well, at least you can repurpose buggy whips...
A long thread about polls and modeling going on at
@zeynep on Twitter. A lot of useless arguing going on, but not all is useless.
Are probabilities forecasts? Damn right they are. Anything purporting to provide information about the future is a forecast.
Is counting the number of states polls got right really meaningful? (Nate Silver) Not so much. Tufecki points out that one could predict most states ahead of time without using polls. I'd add that not all states are of equal importance. The polling in large electoral count states matters far more, like Florida.
Are the models/polling verified at all? Nah. Tufecki's comparison to weather forecasts is worthwhile. For instance, a few years ago the NHC started assigning probabilities to tropical disturbances on their 3 and 5 day chances of becoming tropical storms. After a couple of years, the data showed that they were consistently underestimating the chances of their medium probablity group (needed more high) so they changed their procedures. We don't really get that kind of feedback since national elections are so variable in nature, the mechanics are not well understood, and they just aren't well understood even after the fact.
Polling and models feedback into the system they are trying to predict.
Polling and models will continue anyway.