Dem
50
Ties
1
GOP
49
As long as we are talking about models and polls, let's look again at the question of what happened in Michigan and Wisconsin. The Senate primary polls in Michigan and the gubernatorial primary polls in Wisconsin showed progressive candidates doing vastly better than they actually did. Why? The New York Times very own Nate (Cohn) took a stab at figuring out what went wrong and whether any polls can be trusted anymore.
Let us begin with some general remarks. First, primaries are always hard to poll because people aren't really tuned in yet and it is very hard to guess who will actually vote and who will not. People who answer the poll may say they will vote, but in the end don't vote because November is far off and they have better things to do now. This could lead to an actual electorate that is very different from the sample.
Second, in the general election, many people know exactly who they will vote for because one has a (D) and one has an (R). If they are operating in good faith, they tell the pollster the truth. For a primary, these markers are absent and people may genuinely not know much about the candidates and could change their mind easily later based on new information. Primary choices are not as deeply ingrained as (D) vs. (R) general elections.
Third, telephone response rates are dreadful nowadays. The old model of just dialing phone numbers at random simply doesn't work anymore. Many people have their phone set up to reject all calls not from people in their contact list. Pollsters are trying to deal with this by contacting potential respondents by text message and online. This definitely increases the response rate, which is good. But it strongly overweights people who respond to unknown text messages and people who are terminally online. The people who do respond tend to be younger, better-educated, more highly engaged, and more progressive than the average Democrat. This introduces a strong progressive bias into the sample. This is related to the "non-response" bias we saw in 2016, when many Trump supporters hung up on the pollster instantly but Hillary Clinton's supporters politely answered all the questions. Cohn has some recent data on this. In the NYT poll of the mayoralty race in NYC, 70% of the respondents who were polled by text message supported Zohran Mamdani but only 54% of those polled by a telephone call supported him.
Cohn also notes that Michigan and Wisconsin were very different. One size does not fit all. In Michigan, there were no neutral primary polls in the final days (because Abdul El-Sayed was ahead by 15+ points, so why spend the money?). But there was a Glengariff Group general election poll released a week before the primary that had Mike Rogers (R) crushing El-Sayed 49-39 but had Haley Stevens eking out a narrow win. That could have shocked some El-Sayed supporters into thinking: "I want to send a message, but even more, I want a Democrat to be elected." Many voters could have changed their minds days before the election and there were no late primary polls to capture that shift.
Wisconsin was a different kettle of fish cheese. First Lt. Gov. Sara Rodriguez (D-WI) was in, then she was
out. Then Mandela Barnes was in, then he was out. David Crowley was in, then out, then in again. It is not hard to
imagine that with the moderate candidates coming and going like flies, they wouldn't poll well. The results bear this
out. Polls had the progressive Francesca Hong at roughly 44%. She got 39%, so her number is pretty much on target. What was
surprising is that David Crowley did far, far better than predicted. But his polling could have reflected many voters'
uncertainty about whether he was in or out. Still, by Primary Day, he was the only moderate left, so all the moderates
voted for him in the end. Also, in the final week, the resurfacing of Hong's old
tweets
about canceling Thanksgiving (too colonial) probably cost her some votes. And remember, Wisconsin is the mother of all
swing states, so there are a lot of moderates. The kids in Madison are noisy, but they don't represent the entire state
well. The moral of the story here is that when there is one progressive in for the duration and the moderates keep
coming and going, the polls are not likely to capture the support for the moderates well. It might be true in the other
direction, though there aren't any clear-cut recent cases of progressives shuttling in and out of a major election.
Usually, the left coalesces around a candidate pretty early, and that candidate stays the course. (V)