What the 2023 elections can — and can't — tell us about 2024

The 2023 elections were a victory for liberal governance. Democratic Gov. Andy Beshear won reelection in Kentucky. Ohio voted to enshrine abortion rights in its state constitution. Democrats won control of both chambers of the Virginia legislature and strengthened their hold on New Jersey's. The main bright spot for Republicans was that Republican Gov. Tate Reeves won reelection in Mississippi.

Yet it seemed like the only thing people cared about was what the results augured for the 2024 election. I get the impulse — I used to be one of those people who read way too much into these off-year election results. But after a Republican sweep in the 2021 elections failed to predict a mixed-but-generally-good-for-Democrats 2022 midterm, I took a closer look — and found that off-year election results aren't the best predictor of future elections. As a result, it's not clear what, if anything, we learned about 2024 from last week.

Kentucky and Mississippi are the worst oracles

Let's start with the easiest narrative to debunk: The idea that the results of the Kentucky and Mississippi gubernatorial elections tell us anything about the 2024 presidential election. After Beshear's victory, some observers noted the fact that, since 2003, the party that wins the Kentucky gubernatorial race has always gone on to win the White House the following year. But you should be wary of statistics like this one that claim something "always" happens or that rely on binary outcomes rather than margins. For example, the 2019 Kentucky gubernatorial race was extremely close and could have easily gone either way. If just 5,137 more people had voted for Republicans in that election, Kentucky's predictive streak would have been broken.

Instead, let's take a look at the margins in Kentucky gubernatorial races (and, for good measure, Mississippi gubernatorial races) since 1999 and how they compare to the margins of the national popular vote for president in the following year. It turns out that there's virtually no correlation.

This shouldn't shock anyone who followed this year's elections: Democrats won in Kentucky, and came close in Mississippi, because they had strong candidates with cross-party appeal. President Joe Biden, on the other hand, is unpopular, according to 538's tracker of his approval rating, and is likely not going to be competitive in deep-red states like these in next year's presidential race.

Take New Jersey and Virginia with a grain of salt

Next up: Are the results of New Jersey and Virginia legislative elections predictive of future elections? I used to believe they were good doppelgangers for federal House elections, at least: a large sample size of elections in a uniform campaign environment on a regularly scheduled date with a good mix of incumbents running for reelection and open seats. But when you look at their historical track record, New Jersey and Virginia seem like unreliable indicators at best.

After the 2017, 2019 and 2021 elections, I compared the Democratic or Republican winning margin in each New Jersey and Virginia legislative district that was actively contested by both parties to that district's base partisanship — essentially, how we would expect that district to vote if the national political environment were tied between Democrats and Republicans.* I then averaged these numbers to calculate the average Democratic or Republican overperformance in each chamber in each year. If that chamber's elections were good at predicting future U.S. House elections, that overperformance number should come close to matching the national House popular vote the following year. But since 2017 at least, it hasn't really done so.

Granted, the New Jersey and Virginia results do seem to gesture in the right general direction. The year with the biggest Republican overperformances was 2021, and 2022 was the only time in the last three cycles when Republicans won the House popular vote. Likewise, the year with the biggest Democratic overperformances (2017) preceded the cycle with the biggest Democratic popular-vote win (2018). But the exact overperformances don't do a good job of approximating the eventual House popular vote — just look at 2017-18, when the Virginia House missed by 7 percentage points. Plus, this is a pretty small sample size of elections, so we can't be confident that the directional pattern we think we're seeing will hold up over a longer time period.

According to this method, on average, the 2023 New Jersey state Senate election results were essentially neutral, while Republicans overperformed by about 3 points in both the Virginia state Senate and state House. In other words, despite the fact that Democrats won both chambers in Virginia, Republicans actually had a strong election night once you account for Virginia's blue hue (although, as Kyle Kondik of Sabato's Crystal Ball has pointed out, Virginia tends to vote more Republican down-ballot than it does for president). Democrats won each chamber by the slimmest possible majority — one seat (21-19 in the Senate, 51-49** in the House). If the directional pattern we identified above is true, then maybe this means 2024 will be somewhere between 2020's 3-point Democratic win in the House popular vote and 2022's 3-point Republican win. But again, these numbers have been such imprecise measures in the past, all this really tells us is what we already knew: Either party could win in 2024.

Trust the special elections?

In the end, the most predictive of 2023's elections may be some of the lowest-profile. Nov. 7 saw special elections take place in Rhode Island's 1st Congressional District as well as seven state-legislative districts (although one of those was uncontested). And as we've written many times, a party's overperformance in special elections (using the same methodology as I used for New Jersey and Virginia) has historically been pretty predictive of the House popular vote in the subsequent general election.

And those special election results were mixed, though overall they were better for Democrats than for Republicans. Democrats did better than base partisanship in four of the seven races, and the average overperformance on Nov. 7 was D+3.

That was actually a departure from what special elections had been telling us prior to November. Between January and September, Democrats had been performing spectacularly in special elections, putting up margins that were, on average, 11 points bluer than base partisanship. We don't know whether Democrats' more muted performance on Nov. 7 was due to a change in the national environment or whether it was just noise in the data, but it is at least interesting that it was more in line with polls of the generic congressional ballot (which show absolutely no sign of a D+11 national environment) and with the results in New Jersey and Virginia.

That said, it is special elections all cycle long that have historically proved predictive, not just special elections on odd-year Election Days, and with Nov. 7's special elections factored in, Democrats are still overperforming base partisanship by an average of 9 points on the year.

So after all that analysis, I'm afraid that we're not much closer to answering the question of what last week's elections mean for future ones. Some of 2023's election results clearly tell us nothing about 2024; others may tell us something, but we can't really be sure; still others have historically been quite useful but are pointing in confusing directions this year. In other words, I don't recommend changing your 2024 outlooks based on what happened last week.

Instead, take the 2023 elections for what they were: a series of Democratic and liberal victories that were important in their own right. More than 25 million people live in Kentucky, Mississippi, New Jersey and Virginia and will be directly impacted by the results there. Ohio guaranteed abortion rights to 2.6 million reproductive-age women and legalized marijuana for 8.7 million adults. In Pennsylvania, Democrats notched key wins in judicial and county-level races that will matter for the 2024 presidential race not because of what they portend, but because they will affect how that election is administered logistically and how ballots are counted. These things are the true legacies of the 2023 election.

Footnotes

*A district's "base partisanship" is the average difference between how the district voted and how the country voted overall in the last two presidential elections, with the most recent presidential election weighted 75 percent and the second-most recent presidential election weighted 25 percent. This is admittedly an imperfect exercise in Virginia, where absentee ballots were not allocated by precinct until very recently, which meant that presidential results by legislative district were only an estimate. However, Daniel Donner of Daily Kos Elections did a similar analysis that compared the statewide popular vote for the Virginia House to the commonwealth's base partisanship, and he arrived at the same conclusion.

One more note: I didn't bother doing this exercise for elections for New Jersey Assembly, the state's lower chamber, which are also held in odd-numbered years. Assembly districts elect two legislators each, and all candidates in the district run in the same race (the top two vote-getters are elected). This irregular arrangement makes Assembly races too different from typical elections to reliably analyze.

**One district in the Virginia House, the 82nd, still does not have a projected winner according to the Associated Press. However, Republican Kim Taylor's 78-vote lead is probably large enough to withstand a recount.