Having discussed the results and prediction markets, let me address a different question I saw brought up last night. If the polls have been this wildly off in Wisconsin and Michigan, how can we be confident that Democrats are set for a banner showing in November?
The best answer to this is that you shouldn’t be confident, or, at the very least, certain of anything. But the comparison is mostly apples to oranges. Midterm elections are fairly high turnout affairs and we have lots of historical data about turnout patterns. So the composition of the electorate is fairly predictable. Fairly. Primaries are quite different. It’s a bit like saying you have a hundred kids who are going to be in the school lunch room — a table with the jocks, one with the nerds, one each for the goths (dating myself), the richies and the stoners — and they’re going to vote for student body president. But now let’s say only 20 students are going to show up. It’s a very different result if it’s the goths and nerds who show up vs. the jocks and the richies.
That’s a bit how polling a primary election works. And it’s even more the case in open primary states like Michigan and Wisconsin. So to extend our analogy, a few of the 20 students may be from another school. These primary polls are mostly models, or how you’re modeling the electorate. If your model is far off it almost doesn’t matter how well you polled.
That’s the case with every poll to a degree — but far, far more so with these primaries. EPIC-MRA, the premier pollster in Michigan, didn’t even release a primary poll in its final poll before the election, opting instead to focus on the general. So yes, don’t take anything for granted. Polls can be wrong. But mostly these particular misses shouldn’t make you any less trustful of polls than you already should have been. They’re qualitatively different than the challenges of polling the actual election.
Two other points to consider.
One is differential response. This seems to have played a real roll in Michigan and quite possibly in Wisconsin. This is a related but different issue from the modeling one. Basically maybe Francesca Hong’s supporters were just way more over-the-top and revved up in their support and thus more eager to answer pollsters. But no matter how over-the-top you are, you can still only vote once. This may have overlapped with some communities being harder to reach — rural voters for instance. From my limited observations last night, Crowley was doing better in rural areas. This issue, too, is one that is more manageable in a higher-turnout election. But it’s one that, in general, has led polls in a number of recent elections to underestimate Republicans by very small but sometimes significant margins.
Finally, there are specific issues with the Wisconsin election that we can’t ignore. The apparent winner, Crowley, only got back into the race three weeks ago. Here’s the chain of events. Lt. Gov. Sara Rodriguez was sort of the establishment choice who relied heavily on her experience and managerial prowess. But it was a crowded field and her campaign hadn’t really caught fire. Then last month it came out that she actually didn’t have any money for what was supposed to be a crucial ad buy. The best explanation Rodriguez could come up with was that she trusted her campaign manager and that the campaign manager had probably double counted all the contributions. (If this seems so bizarre and you think you must be missing something … you’re not. It was that bizarre.) But the upshot was that this meant A) that she had no money, when she was supposed to be the flush candidate and B) that it just wrecked confidence that she had the managerial ability or trustworthiness or something to be governor. Upshot, she drops out.
At this point, outgoing Gov. Tony Evers (D) gets Crowley to get back into the race and starts trying to get everyone to consolidate around Crowley against Hong. (Crowley was in originally, but hadn’t caught fire and dropped out.) He gets back in on July 18th. Mandela Barnes —who had been Evers’ first lieutenant governor and lost a narrow race against Sen. Ron Johnson (R-WI) — dropped out on July 31st. Then on top of all of that, the state’s voters have been fed a steady diet of news over the last 10 days or so — the upshot of which, rightly or wrongly, was that nominating Hong was all but gifting the governor’s mansion to the GOP. And not just any Republican, but an arch-election denier, Rep. Tom Tiffany (R-WI). So it’s difficult to know how much of this is a mammoth polling error vs a decisive shift away from Hong in the week or so since the last polls were in the field.
The last point on that front is we could be looking at something like what happened to Sen. Bernie Sanders (I-VT) in the 2020 election. The polls were saying that Hong was up by more than 20 points. But the final polls put her support in the mid-to-low 40s. The final results show her taking just over 39% of the vote. That’s not that different at all. It’s just that what had been a split field of non-progressive candidates almost entirely consolidated around Crowley. That kind of shift, especially after Barnes left the race, is exactly the kind of thing you’d imagine would happen in the final days when a majority could not be persuaded to support Hong AND voters were getting hit with an avalanche of commentary that Democrats were throwing away the race if they didn’t consolidate behind a single candidate.
None of this is meant as a rearguard defense of polling. This was clearly a huge miss. But these very sui generis final weeks of the campaign created mammoth volatility in the electorate itself, quite apart from errors in polling.
Probably the best way to look at all of this is that for all the drama and surprise, largely an artifact of the polling, the outcomes of these two races were very, very similar. It’s just that Abdul El-Sayed ended up a tiny fraction ahead and Francesca Hong ended up an even tinier fraction behind. Big picture, both of these races were more or less a tie between a conventional, mainstream Democrat and a candidate of the progressive left.
