Josh Marshall writes:

Elections are hard to predict. But even with that, some of the notional “surprises” we’re seeing [on Election Day, 2025] are less surprises than a measure of GOP dominance over current press narratives. People were looking for an upset in New Jersey. Nate Silver’s Silver Bulletin speculated that New Jersey might be moving toward becoming the next swing state. In fact, Rep. Mikie Sherrill (D) currently appears on track to crush Republican Jack Ciattarelli. A similar failure of conventional wisdom appears to be unfolding in the Virginia Attorney General’s race. A lot of D.C. insiders had convinced themselves that a controversy over some intemperate texts (not nothing but fairly close to it) had doomed his campaign. As recently as a couple days ago, betting markets (which are proxies for conventional wisdom) gave his opponent Jason Miyares 3-to-1 odds of victory. Jones now appears on his way to a clear though not resounding victory with a 3-to-4 percentage point margin.

Marshall continues:

These results aren’t terribly surprising. You’d expect Democratic gubernatorial candidates to do well in blue states in a climate where the Republican president is deeply unpopular. . . . The issue, again, is the power of Republican political narratives currently have over the elite political press. . . .

Maybe. But maybe it’s simpler than that; it’s just that journalists were reacting to the Republicans outperforming the polls in 2024.

In our election forecasts, we include terms to allow for the possibility of systematic state and national polling errors. There’s no perfect way to do this adjustment, and I’ve heard that some of the private polls in 2024 did better than the public polls that we and others were aggregating. The point is that there’s uncertainty.

Lots of people, even quantitatively-minded people, can’t seem to handle this uncertainty. For example, our 2024 forecast was criticized for not “communicating that a Trump landslide was a significant probability.” But the election was very close–nothing like a landslide at all!

The relevant point here is that people seem to be able to handle a point forecast (“Here’s who we think will win”) or complete uncertainty (“We know nothing at all”) but have difficulty with anything in between (“We’re pretty sure the election will be close, but it could go either way”). And I think what happens sometimes is that people take forecasts with uncertainty and place them in one of these two bins. For example, when we (and Nate, and others) gave our forecasts in 2024, the public–even some quantitatively-trained members of the public–were inclined to either treat our forecasts as deterministic (our forecast odds are 50/50 so we’re essentially predicting an exact tie) or as empty (our forecast odds are 50/50 so we’re making no predictions at all).

Now let’s get back to those 2025 elections. Democrats had comfortable polling leads in all these races, but pundits had been burned a year earlier, so a natural first step was to take the polling errors from 2024 and apply them to the 2025 polls. In the event, the polling errors were mostly in the other direction. On election eve, a reasonable forecast would be to say that the Democratic candidates were likely to win handily, they might win by even more than expected, or the races might have been close, with some Republican upsets not being out of the question. It’s hard for me to compare this to the conventional wisdom because that’s not written down in any one place.

Uncertainty is hard, especially given that people flip between thinking of forecasts as deterministic and thinking of forecasts as being completely uncertain. It’s related to the well-known challenge of producing wide-enough uncertainty intervals.