Please indulge me as I take a minute today to complain about poll questions that ask people whether they'd be more or less likely to vote for a candidate if the candidate did X thing.

Things people are generally bad at:
-explaining how they decide their votes
-entertaining counterfactuals
-understanding probability

This asks them to do all three!

But the even bigger problem is that people aren't going to be answering the question as asked.

Like, say you find a person who's a die-hard Democrat. They still have Clinton bumper stickers. The chance they're voting for Trump again is zero.

A pollster calls them up, and asks how...I don't know, Trump's hypothetical support for a bill to legalize kicking puppies would affect their vote.

Now....they weren't going to vote for him before, and they won't if he signs the law, so really it won't affect their vote at all. But that's not what they'll say.

They'll say it makes you less likely to vote for him, either to register their dislike of Trump, or their dislike of puppy-kicking, or both.

Still my favorite dumb case-in-point example of this...

It's not that these questions can't be useful, especially if there are others for comparison, but I think it's generally a mistake to take the results at absolute face value.

Anyways, back to your regularly scheduled programming!

A great demonstration of this principle in action -- most people who said hearings made them more likely to think Trump committed impeachable offenses *already believed that*, and vice versa. https://twitter.com/laurabronner/status/1308930285443125251