Split Ticket

Entering 2020, it was nearly impossible to find a national poll that had Joe Biden trailing Donald Trump. Polling averages all showed every Democratic candidate leading the then-incumbent Republican president by several points, with the question being the size of the margin. Commentary discussed whether the Republicans even had a shot at holding the White House.

Today, the picture could not look more different. Polling averages consistently show Biden trailing Trump, with the president tanking among core Democratic constituencies such as minorities and young voters. Troublingly for Biden, his favorables are now worse than Trump’s, and many polls show that a majority of Americans believe he is simply too old for another term.

At the same time, however, off-season election results consistently show a different picture, with Democrats frequently overperforming their 2020 margins in several key elections in 2022 and 2023 alike. Moreover, the cataclysmic crash among young voters and minorities is nowhere to be seen when we examine the actual results — post-election studies actually found Democrats improved on their 2020 margins with Hispanics and young voters while mostly holding steady with Black voters in 2022.

These two findings seem irreconcilable and beg the question: Why do polls paint a different picture from the election results we’ve been seeing?

For decades, the maxim was simple: the higher the turnout, the better Democrats would do. This is the origin for some electoral assumptions of many analysts — for example, the notion that Republicans get a bump in polling around Labor Day, attributed to polls switching from “registered voter” to “likely voter” screens. This was borne out of a pattern where Democrats would consistently do better in polls that did not screen for propensity or engagement.

Those screens do meaningfully change the composition of a survey’s electorate. A sizable share of eligible Americans do not vote, especially in non-presidential elections. According to the Pew Research Center, while two-thirds of the voting-eligible population cast ballots in 2020, less than half did so in 2018 and 2022. The “dropoff” group, which votes in presidential years but not midterm cycles, usually disproportionately consists of traditionally-Democratic constituencies, and it is precisely this group that Biden seems to be bleeding support with.

Thanks to data posted publicly by Nate Cohn of the New York Times, we can glean some valuable insights into how voting propensity is related to support shares in polling. In particular, it is abundantly clear that almost all of Biden’s slippage in polling comes from his slide with disengaged voters.

Per data shared by Cohn, we can see the turnout levels across cycles among respondents to New York Times/Siena College polls that were registered to vote at the time of the given election. “Any primary” means that the voter voted in any primary, up to and including 2022, and “any midterm” means that the voter voted in any midterm election, up to and including 2022. From this, we can see that participation among voters generally tends to decrease as we traverse the intersections of the categories described above.

The question then becomes: How have Biden’s and Trump’s support levels held up among each of these groups? If we limit our analysis to the pool of voters in each group that actually voted in 2020, we can see from Cohn’s data that Biden loses significantly more support dependent on the levels of political engagement.

Among the most engaged partisans, Biden experiences practically no slippage. But among the other two groups, we see more and more defections. Although Biden retains more of his highly-engaged, partisan voters than Trump does, he experiences far more defection among the voters who rarely show up.

This is best shown by a single, striking fact: Across all Times/Siena polls conducted last year, per Cohn, Biden leads by 4 among voters who have cast a ballot in a primary or a midterm. Among all other registered voters, Trump leads by 10.

These findings help explain the gap between special election results and polling. Special elections have characteristically low voter turnout. This means high-propensity Democrats, with little to no defection from 2020, are overrepresented, holding Democratic margins steady when they turn out. High-propensity Republicans, on the other hand, are markedly less “loyal”, meaning that among a high voting propensity electorate, Democratic candidates are effectively more popular than with the electorate at large, especially in solidly-Democratic constituencies. It’s just that those who would defect are much less likely to turn out for off-season elections.

This would help reconcile the discrepancy between election results and presidential polling, as the electorates observed in presidential years are quite different compositionally from the ones seen in interim elections. It would also go a long way towards explaining why Democrats still failed to claim a popular vote victory in November 2022, despite overperforming in special elections by about 5 points just two months before the midterms. It’s simply easier to believe that millions more Republicans who behaviorally don’t turn out for special elections chose to turn out for the general than it is to believe that the political environment jumped some ten points to the right in a little over two months.

It’s also reinforced by data recently published by the New York Times, which shows how special elections voter pools are compositionally different from the pool of registered voters at large. According to the aforementioned Times data, special elections tend to have a more Democratic electorate than general elections do, even after controlling for compositional differences. For example, white, college-educated voters who turn out for special elections are more likely to support Democratic candidates than the ones that do not show up.

None of this is to say that November’s political environment is guaranteed to be Republican-leaning. For example, in 2022, millions more Democrats stayed home than Republicans, and it is very likely that Democrats would have won the national popular vote if there was greater turnout. In presidential elections, anti-incumbent partisan turnout bias decreases significantly, as voters of all stripes and creeds turn out at higher rates, smoothing out the electorate to better reflect the country as a whole.

It should also not be concluded that there has been no loss for Democrats among those who voted for Biden in 2020. What the New York Times data suggests is precisely that low-propensity Democrats, who haven’t spoken up at the ballot box in soon-to-be four years, are precisely the ones Biden has lost support from. Given this, it is actually very possible that the turnout deficit is reversed in 2024 — if voters are not enthusiastic about either Biden or Trump, then they may choose to skip the election entirely, and this may meaningfully hurt Republicans given the margins we observe among those disengaged voters in polling.

It is also possible that Biden’s position is thus much stronger than the poll toplines seem. Many of the voters that he has lost support from backed Barack Obama, Hillary Clinton, and Biden himself and, importantly, already voted against Trump twice. Making gains with a disengaged and currently disenchanted group of voters may not be easy, but their recent voting record suggests that reversion to the Democratic column is certainly possible, if not likely.

It is impossible to validate this hypothesis until the election, and there are reasons that the polls could be entirely unrepresentative; infrequent and disengaged voters are very difficult to reach in polls and even more difficult to construct a representative sample with, and it is very possible that the ones being reached right now are simply not representative of the overall electorate. Perhaps those upset with the Biden administration are more likely to respond and make their grievances known.

Alternatively, perhaps low-engagement voters are simply behaving differently, in part due to their belief of what the matchup will actually be — a third of likely voters aren’t sure that the matchup will be between Biden and Trump, according to polling conducted by Data For Progress. Given what we know about voter engagement, this number is almost certainly higher among registered voters.

These findings may look odd, but it is important to remember that early polls are famously not predictive, partly because campaigns exist to remind voters of the qualities and stances of each candidate. As campaign season accelerates and the matchup solidifies in the public eye, it may be the case that disengaged voters will begin to behave more like engaged voters, as they are reminded of why they voted the way they did four years ago (incidentally, this would also be a reason that special elections continue to hold some directional value).

Unlike the keenest observers, most Americans are significantly less engaged with the daily political intrigue and developments which can reinforce partisan fervor. Polling data suggests that if Biden is to win re-election, however, then he will have to make gains with this silent supermajority from where he currently stands, or hope that they simply stay home.

Editor note: A special thanks to Nate Cohn of the New York Times for sharing polling data with us for use in this piece, both privately and publicly.