Which is the better 2024 predictor: polls or special elections?

Republicans hope the former, Democrats the latter. But it's complicated.

February 21, 2024, 3:29 PM

When it comes to predicting the 2024 election, it’s choose your own adventure. Republicans can point to national polls finding their likely nominee, former President Donald Trump, leading the likely Democratic nominee, President Joe Biden. Democrats can point to the success that their party has had in special elections since the beginning of 2023.

In the past, special elections have been more predictive of the next election than early general-election polls. But there are reasons to think 2024 could be different. Historically, general-election polls get more predictive during presidential primary season, when voters learn who their choices will be and get to know the candidates. But this year, virtually everyone in America already knows and has an opinion of the two likely nominees. According to 538’s polling averages, 95 percent of Americans have some opinion (either positive or negative) of Trump, and 95 percent have an opinion of how Biden is doing his job as well.*

As for the special elections, Nate Cohn of The New York Times has demonstrated that the types of people who show up to vote in low-turnout elections — whose demographics once favored Republicans — are now more likely to be Biden voters than Trump voters. The more inconsistent voters who might show up in November, though, are a more conservative bunch, Cohn’s analysis found.

As I wrote last fall, personally, I’m torn about which metric to put more stock in. But we’ve gotten some new data since then that makes the two seemingly contradictory indicators easier to reconcile.

From Jan. 1 through Sept. 19, 2023, Democrats won the average congressional or state-legislative special election** by a margin of 21 percentage points. However, the districts they were running in had an average base partisanship*** of just D+10. That means Democrats were punching above their weight in special elections by a whopping 11 points.

This looked like a continuation of the excellent special-election performances that Democrats posted in 2022 after the Supreme Court’s decision in Dobbs v. Jackson Women’s Health Organization that overturned abortion rights nationally.

Since September, though, the national political environment has moved toward Republicans. Biden’s approval rating has dipped slightly amid dissatisfaction with his handling of the Israel-Hamas war, immigration and more. It was also around this time that Trump took the lead in a rough average of general-election polls. And, as it turns out, Democrats’ advantage in special elections has evaporated too.

In 25 special elections since Nov. 7 (there were no specials between Sept. 19 and Nov. 7), Republicans have won by an average of 3 points. The districts in which the elections were held had an average base partisanship of R+2, meaning that, over the last few months, it’s actually Republicans who have been overperforming in special elections, albeit by an average of just 1 point. That’s exactly what the polls are saying right now.

Now, I don’t want to make too much of this. We can’t just throw out the special election results from the first three-fourths of 2023. After all, it is special-election overperformance from the entire election cycle, not just the last 12 months, that has historically proven predictive.**** And overall for the 2023-24 election cycle, Democrats are still overperforming in special elections by an average of 6 points.

But it is a reminder that there is still a lot of time left until voters head to the polls to choose the next president — time for more Republican-leaning special-election results that could bring that average even closer to the polls or, alternatively, time for the polls to shift to more closely match the special-election results.

In addition, the two metrics aren’t actually as contradictory as I’ve led you to believe. Historically, overperformance in special elections has been correlated with the national popular vote for the U.S. House — not the presidency — in the next regularly scheduled election. This nuance often gets lost in this debate: The U.S. House popular vote and the presidential popular vote are themselves correlated, but they are not the same thing. And interestingly, overperformance in special elections (as measured by Daily Kos Elections’s Special Elections Index) has virtually no correlation with the presidential popular vote:

PHOTO: Two scatterplots showing the average degree to which a party overperformed in special elections since the 1989-1990 election cycle compared with the final House and presidential popular vote margins in the general election.
Special elections predict House, not presidential, results
Amina Brown for 538

So one way to resolve this “conflict” would be if Trump wins the presidency, like the polls imply, but Democrats win the House, like the special elections imply. (Granted, this would be cold comfort to Democrats.)

This would require some split-ticket voting — which has become rarer and rarer in our increasingly polarized polity — but there is, in fact, some evidence that voters are more open to supporting congressional Democrats than Biden this year. Just look at polls so far this month that have asked about both the presidential race and the generic congressional ballot (i.e., which party people plan to support for Congress, without naming specific candidates). Trump leads the presidential polls by an average of 2 points, but the generic congressional ballot polls are deadlocked.

Of course, that’s still a 6-point difference between those six generic congressional ballot polls, which are tied, and the special-election results. And actually, the difference may be closer to 7 points, since 538’s more sophisticated rolling average of generic congressional ballot polls shows Republicans leading them by 1 point.

Still, that’s a much smaller gap than in September. And, crucially, the two indicators are now close enough that they could both be sort of right. Imagine Democrats win the House popular vote by 3 points. That would be consistent with the past track record of special elections: In the last three election cycles, special-election overperformance has been 1-3 points better for Democrats than the national House popular vote.

At the same time, Democrats winning the House popular vote by 3 points when polls show Republicans ahead by 1 point would be well within the range of a historically normal-sized polling error. Since 1998, polls of the generic congressional ballot have had a weighted-average***** error of 3.9 points.

The unfortunate reality is that neither polls nor special-election results are a precise enough indicator to give us the kind of certainty we crave in this era of historically close elections. All signs right now point to an election that could go either way, and if I had to guess, I don’t think that will change by November.


*All numbers as of 2:30 p.m. Eastern on Wednesday.

**Excluding races with no Democratic or Republican candidate and races in which a minor-party candidate received at least 10 percent of the vote.

***Defined as the average margin difference between how the district votes and how the country votes overall. It is calculated as 75 percent the district’s lean relative to the nation in the most recent presidential election and 25 percent its relative lean in the second-most-recent presidential election.

****Look no further than 2022, when the final midterm election results were not the “blue wave” suggested by the post-Dobbs special elections, but rather an evenly divided national popular vote that was closer to what the pre-Dobbs specials indicated.

*****To avoid giving individual pollsters too much influence if they happened to conduct a lot of polls, the polls are weighted by one over the square root of the number of generic congressional ballot polls that their pollster conducted.