I look forward to tracking the work that @fiftyplusone.news will be doing, but people who are interested in public opinion should read the good work that is being done *within* individual polls in addition to seeing how they fit in the big picture.

Let's not lose the trees for the forest.

5/5

The American National Election Study (the longest running study of voters with waves going back nearly 80 years) very rarely does "well" in predicting the result of the election, but it is an incredibly valuable resource for understanding how public opinion has shifted in the post-WWII era

4/

that the primary purpose of opinion polling is to predict the result of a particular election, and when there are high-profile "misses" on this front (e.g. the overconfident projections from polling aggregators in 2016), it undermines the public's confidence in the whole enterprise

3/

This is especially apparent in the various modeling exercises that take place during election season. While it is an interesting exercise to try to forecast the results of the presidential election based on the available polling. I worry that it contributes to the idea

2/

Here's what I see when looking only at the F2F interviews vs the inclusion of the web sample:

The trend moves in the same direction, but you get a bump of about 5 pp when including the web sample.

Do we know if there was a methodology shift in the 2024 web survey? There have been big mode differences in the past between self administered modes and interviewer mediated surveys

I've heard some objections to this approach based on accessibility (e.g. speech-to-text uses the same paste functions). Have you given any thought to how to allow valid uses of pasting?