Key Takeaways
- The variance in horse-race polling is essentially completely one-sided depending how unified the Democratic sample is.
- Polling needs to happen consistently right up until the very end (Election Day), which is not normal industry practice, because in recent elections Democrats have not been coming home until the moment they actually vote.
- There is a group out there saying they are crossover Democratic voters, but they are either coming home to the Dems in the end or staying home in these off-year elections. Figuring out if their turnout actually means they do vote for Trump – or stay home – is the most important factor this cycle.
Recent 2024 presidential polls have shown a lot of different scenarios within the same margin of error with either a tied race or a slight Trump lead in the popular vote. At the same time, it seems like there has been a constant string of high-profile elections where Democrats do well or even outperform the polling like in the NY-03 special.
This article attempts to answer the question of why the polls show good numbers for Republicans that don’t always bear out in final results and why current polling might be off.
We’ll start with a list of recent head-to-head elections, and what stood out during our close study of the results was the recalled vote unity. Most of the current polling is Registered Voters, but even the Likely Voter makeups aren’t different enough to be the cause of the variance between polls – but this factor is. For polls that had crosstabs by the recalled vote, which asked a respondent who they voted for in 2020, we used it to make the table below.
The Trump voter margin is how many people who voted for Trump 2020 are sticking with him again. Let’s say 90% in 2020 Trump voters were voting for Trump, which means currently 10% aren’t. That would give a value of 90 – 10 = 80%. The Difference column is how much more (or less in Emerson’s case) united Trump voters are behind him than Biden’s 2020 voters (shown with the same calculation but in an inverse number) are lining up behind him. The Overall Margin column is the head-to-head Trump vs. Biden margin.
If you think you see a trend in shading, you would be right. The next set of plots has the table above in the top left. The top right plots out the Difference column in the x-axis, and the y-axis is the Overall Ballot. This shows a pretty clean line that almost every poll falls on which basically says that the Difference is what influences the overall margin. The bottom charts are the split by party and show the amount of unification Trump voters have has no effect on the topline, but the Biden voter unification difference is the key trend.
With this, we can make the assumption that how unified Biden’s 2020 voters are is the most important aspect of the 2024 election, and how united it is in each pollster’s sample matters for the result.
In this case, we want to examine in recent elections how many Democrats eventually voted for the Democrat. The New York Times’ Nate Cohn is actually the one who really assigns Democratic strength in the post-Dobbs special-election era to there being no persuasion in the end. This would imply that Democrats are “coming home” in special and off-year elections so far, which is much different than current polling for Biden would imply.
So, the next question becomes, when does the polling become more accurate to measure this phenomenon?
We in the industry make some assumptions on when people make up their mind. Some punditry mentions Labor Day before the election, pollsters switch to their likely voter (LV) screens usually around the conventions, and the CNN exit poll has about 5% of people in the last week of the election in 2020 and only a quarter starting in September.
There are other ways to measure when people get serious about their votes. In the chart below from 2016, the last time we had a presidential election with high-polling third parties, we see the reported share of the vote going to minor parties in state polls begin to drop rapidly by the end of September after being stagnant the whole time before. However, it didn’t really land on the accurate result until the day before the election.
A similar election to this 2016 trend was the 2022 Oregon governor’s race where a typically blue state saw a moderate Independent candidate taking a lot of the vote with a Republican lead that saw a Democratic win in the end. Final polling before the election was close, so the Democrat winning wasn’t surprising, but why did it take so long?
The next plot shows the dots of reach 2022 Oregon poll’s margin where the higher and bluer it is, the better for Democrats and vice-versa. The black line represents the cumulative votes counted on that day for the governor’s race with early voting, so there is a slight delay on when there were actual votes but paints a good picture on when voting happened. Republican leads didn’t get knocked down until the start of early voting, and the final polls themselves weren’t accurate until half of the people had voted, including the vast majority of Democrats.
We would like to posit a new theory in more recent elections: Democrats and Biden are not seen favorably by a noticeable number of people who voted for Biden in 2020, and so they are not “coming home” until the moment they cast their ballot. This is what has been causing polls to be off in the Republican direction. Because of the consequences of their last-minute decision to vote for the Democrat, it would also have a disparate impact in the battleground races, which is where polls were off the most. It’s also another reason to pay serious attention to the Republican candidate’s ballot share, not just the margin between the Republican and Democratic candidates in a race.
To back this up, we also have data for six recent internal polls conducted by Cygnal in recent battleground districts that have been anonymized. For each of these, we took a poll a month out and then another poll one day before the election. Trump 2020 voters stayed with the Republican candidate at equal levels, and the named Democrat on the ballot did see modest gains among Biden 2020 voters.
However, when we split the sample from the poll that was taken one day out into Biden 2020 voters who had voted or had “yet to vote”, there was a massive gap, almost all in the double digits. In these elections, about one-third to one-half of the Biden 2020 voters had already voted. The “yet-to-vote” recalled-Biden voters had not moved much in the average or individual races, but the ones who did vote by the time of the poll backed the Democrat candidate by wide margins. We also saw this effect was consistently larger among Biden 2020 voters without a college degree.
This could mean two possibilities, or likely a combination of two factors. The first is that in a special election, more partisan voters are the ones showing up, so it would make sense that they are breaking towards the Democrat on the ballot. This would be good news for Biden and the Democrats if the voters feel pressure to vote for him. The second is that the “yet-to-vote Biden 2020 voters don’t come home but instead in these specials are staying home. That will take further voter file analysis and is certainly something that could be happening though you would expect smaller Democrat turnout overall.
Either way, we can use this for a few lessons about polling and targeting strategy this cycle. If Biden 2020 voters aren’t going to come home until their ballot is cast, it’s going to be a long election with stagnant polls until the very end. Therefore, we should be polling right up until Election Day to measure the level of Democratic Party unity, and we should be doing more to highlight the difference in “yet to vote” and “already voted” numbers, something Cygnal has several custom visuals depicting for clients.
For Republicans, it means that you can’t rely on the numbers being inflated by Democratic crossover votes; instead focus on hitting Republican and Independent voter goals or else you will find yourself underperforming your final poll margin (again, why we have to pay attention to ballot share now more than in past elections).
And lastly, there should be a special emphasis placed not as much on persuasion but more on turnout. Early polling should be about setting up an image so that when things finally do solidify at the end, you are not losing as many GOP voters, or you are in a good position to pick up possible Republican voters that have left.
These “yet-to-vote”, not-as-partisan Biden 2020 voters are either sitting out the special elections or voting Democratic at the last moment, and they could potentially be up for grabs to each side. The level of differential turnout is sure to play a bigger role, even if it’s not what public pollsters are showing as the difference right now.