Arie Kapteyn, USC Dornsife Center for Economic and Social Research: First, let me start with a qualification about our results four years ago. It looked as if we did a good job because our national poll had Trump winning. But that, of course, was wrong: In the popular vote, Trump actually lost. So we didn’t do that well, and the reason why was our sample had an overrepresentation of people in rural states. Even though we weighted by education and past voting participation and all of that, we simply had too many Trump voters in the sample.

We have about 9,000 people in our panel [this year], and they answer questions every two weeks. And we ask them a number of things. We ask the probability that they will vote. We ask them the probability they’ll vote for Biden or Trump or someone else. But we now also ask them a question I think you’d always asked, Robert: “Who do you think your friends and neighbors will vote for?” We call it a “social-circle” question.

Now, we actually get a 10-point lead, nationally, for Biden over Trump. But if you look at the “social-circle” question, Biden only gets like a 5- or 6-point lead. One explanation for that may indeed be “social desirability.” In general — and certainly on the phone — people may still be a little hesitant to say to that they’re Trump voters.