Washington Primary Points Toward Another Nailbiter

As any consumer of polling knows, polls have error margins. A sophisticated consumer of polling knows that polls have sources of error beyond simple sampling error. Polls can be off because not enough of a certain group of people respond to them, or because people misunderstand the question, or because people lie. 

Some combination of these factors has obviously been at work in election polling over the past few cycles. Most famously, former President Donald Trump has overperformed his polls significantly both times he ran for office. Whether this time will be a third time is a matter for another piece, but it does add some uncertainty to the mix.

There are, however, some data points that aren’t dependent upon public polling. One of these is special elections, although this has become a more difficult data point as the Democrats’ coalition has become more upscale. 

The second is the Washington primary. This is something that I’ve written about since 2010, and it was one of the first indicators I had in 2020 that the election was likely to be much closer than the public polling suggested.

The idea is this: In Washington, all candidates of all parties run on the same ballot. For example, the Senate primary this year featured five Democrats and six Republicans.

As it turns out, the results of this primary pretty closely mirror the eventual election results. This makes some degree of sense. The primary is held in August when people’s impressions of the race begin to come into focus. All candidates, including incumbents, are encouraged to run serious campaigns. An incumbent caught sleeping can lose, or may show weakness that encourages additional challenges in the future.

All told, this has been a remarkable predictor for the fall. The year in which it really performed poorly was 1998, but more about that later.

So what happened this year? We can start with the results of the primary from last year, along with the eventual outcomes:

What do we see this year? Obviously, we don’t know the final results yet, but we can look at the primary outcomes:

To my eye, this looks a lot like 2022. Obviously, the Democratic vote share is up in District 3. Republicans did re-nominate their 2022 nominee, Joe Kent, but the Republican primary there was less about Never Trump vs. MAGA. There’s probably less crossover voting for the Democrat than in 2022. 

We can, however, build a predictive model off of this. I’ve done this since 2010, although the model has evolved somewhat over time. The idea is this: We go back to 1992, and we compare the vote in the primary against the vote in the fall. We can then get a sense of, on average, how far Democrats run ahead of their primary vote share (or Republicans, it makes no difference). Third parties are discarded for the purposes of this analysis. Note that we also exclude races where two Democrats or two Republicans emerge from the summer primary.

We can then take the results for the 2024 primary and make predictions as to where things will land in the fall. Here are the predictions, with 95% credible intervals [NOTE: CREDIBLE is the correct term of art here, not confidence]:

These are interesting results, but they aren’t necessarily the be-all-end-all (unless you’re a fan of 8th District Rep. Kim Schrier, who should be feeling pretty good right now about her district). You can see the track record of the approach in the following table:

It’s a pretty good fit. 

Now, what’s more interesting is that if you take these estimates and compare them to Democratic vote share from the previous presidential election, you can derive a pretty good rule of thumb for how things will turn out nationally for the Democrats.

For example, in 2022, the model suggested that Democrats would run about 3.8 points behind President Biden’s 2020 vote share. On a national basis, Biden won 52.5% of the vote in 2020, so if we extrapolate nationally, this worked out to a prediction that Democrats would win about 48.7% of the national vote share. They won 48.6% of the national vote share.

You can see as well that it is directionally pretty sound. Democrats are predicted to run way behind presidential vote share in 1994 and 2010. Some years, like 2012, appear to be big misses, but in fact, House Democrats did run quite a bit behind Obama’s 2008 performance. There are some big misses, notably in 2016, but as a rule of thumb, it performs pretty well.

So what does this tell us for 2024? The model is predicting Democrats running about two points behind Biden’s 2020 vote share. Since Biden won by four-and-a-half points, it points toward a pretty close vote outcome in the House that could go either way. In a year where the popular vote is tied up, I think we’d expect a pretty close presidential election as well.

There are three objections. The first is that Washington is pretty far to the left of the nation and certainly is further to the left of the nation than it was in 1992. But since we’re using the in-district presidential results as our baseline, the model accounts for that. So a prediction that Republicans will get 48% of the vote in the Seattle suburbs (i.e., WA-08) has a very different meaning than it would have in 1992. Today that points toward a mediocre result for Republicans; in 1992, that would have been disastrous.

Second, and more seriously, we might note that things can change between the primary and Election Day. This is true. There’s a reason the model performs poorly in 1998, for example, as there was a pretty big swing in there with the Clinton impeachment. We might think that because Democratic enthusiasm is so high right now, perhaps this is the high point for Democrats.

The problem is that these things are difficult to predict, as are the directions of the change. We might have anticipated in 2022, for example, that the recency of the Dobbs decision would fade and that Republicans would perform better. From time to time, these types of analyses will be right, but over time, they will not be.

Finally, “pretty close” is probably not what you’re looking for here. But, as they say, there’s a reason we hold the elections. We’re trying to predict a pretty complex ecosystem; heuristics are often the best we can do. Having a pretty good guess that anyone predicting a blowout is likely off base, based upon actual elections, is a pretty good start for the fall.