Who the Hell is Switching Sides in 2024 Anyway?

If you have followed the polls over the last six months, or if you just looked at the Solid Purple forecast of the election a few seconds ago, you’ve noticed that the numbers have stayed remarkably stable. Joe Biden, after winning the popular vote by about four and a half points, narrowly secured an Electoral College victory in 2020 by a handful of votes in a couple of swing states.

This time around, the popular vote is a jump ball. Depending on the survey, Biden might be up 2% or down 2%, but clearly a not-insignificant number of voters who supported him last time have decided they would rather have Donald Trump back in the White House, attempted insurrection and 34 felony convictions be damned, than another four years without any malarkey. Yet, Solid Purple shows the odds of either candidate winning the Electoral College at around 50/50, even though Biden would have lost the 2020 election if he had only won the popular vote by around 3 points.

So what’s the deal, Zach? How does that make any sense? And of all the elections to change your mind about which party to support, why would it be this one? Who are you people and where the fuck do you live?

I defer to the model on this one.

We can’t know until November. Hell, we might never know! We could wake up to a shift in vibes and then no one will ever find out what was causing the polls to suggest a once-in-a-century realignment along racial and generational lines despite this election being a rematch in an era of unprecedented polarization. But hey! The model was built to forecast the House as well as the Presidency, so we can see how it extrapolates the data I feed into it down to the district level. We might as well take a peek and see where the model thinks the shifts are occurring.

Upon first glance, it looks like Trump has gained ground just about everywhere, which is not terribly surprising given a national shift away from Biden in the ballpark of anywhere from 2-6% (for what it’s worth, the model thinks it’s a little over 3%). There are some minor shifts in Biden’s direction, but not in places that are especially helpful to him electorally. Getting 30% of the vote in the red part of a red state instead of 28% isn’t exactly a silver lining in a data set that is causing Democratic partisans to schedule their days around dry heaving.

But if we control for the shift in the national margin and pretend that there has been absolutely no change in the topline result, it suddenly becomes very obvious where support for Biden has theoretically eroded between 2020 and today.

Basically, other than the southern border districts of Texas, it's the cities. LA, San Francisco, Atlanta, Miami, New York, Greater DC. Places where Democrats have already seen some slippage in 2020 and 2022. But that only tells you so much. These areas have diverse populations with a lot of variables at play.

So I thought...hey, let’s sort the districts by a whole bunch of variables to see which ones are statistically significant. I decided to take a look at the racial demography of each district, the level of educational attainment, the median age, population density, geographic region, and whether or not Biden won that district in 2020. I hope you like data tables because I’ve made a lot of them!

Let’s start with looking at the 20 districts that have sprinted away from Biden according to the model. They are almost entirely concentrated within large population states like California, Texas, and New York. They’re mostly districts that Biden carried by Assad-like margins in 2020, and with a few exceptions, he still carries most of them. They’re much less white than the population as a whole, orders of magnitude more Latino, and voters in these districts are slightly less likely to have gone to college.

As for districts that may be swinging toward Biden, the opposite appears to be true on these dimensions. These districts are slightly whiter than the country as a whole, but significantly higher educated. Biden lost these districts by 4 in 2020, but would narrowly carry them now according to the forecast.

Race

Needless to say, homogenously white districts are very Republican. They also do not appear to be changing all that much between 2020 and now. The forecast projects the whitest districts in the country supporting Trump by 28 points on average, only a point or two from his 2020 margins. The only thing that sticks out here is how capital-D Democratic the state of Vermont is despite its racial demography.

In primarily Black districts, the forecast shows varying degrees of erosion across the board. Biden’s margins in these districts are, with the exception of Georgia’s 2nd Congressional district, healthy enough that he still is winning these areas by 35 points on average. If this holds, however, losing a seat like GA-02 would be shocking, and frankly, if I had to put money on it today, I would still confidently wager that Biden and Democrats down ballot carry that area.

The group the model is most pessimistic about for Biden is very clearly Latinos. Biden’s underperformance in the Rio Grande Valley relative to Hillary Clinton was one of the more head-turning outcomes in the 2020 race, and the forecast seems to suggest it will continue sprinting right. Before you shout “Latinos are not a monolith!” at me, I see you and I hear you, but many have suggested that, at least in these Texas districts, incumbents tend to perform particularly well, which is why the model has folks like scandal-plagued Henry Cuellar maintaining a slight edge despite Biden’s performance.

It would not surprise me if the methodological choices I made in building this thing out come back to bite me in the ass and Biden ends up doing pretty well in South Texas, if not the entire state.

Education

In areas with high levels of educational attainment, Biden did remarkably well in 2020 and the forecast believes little if anything has changed. These districts only reinforce observations made above - racially homogenous areas with high levels of college educated voters are set in stone, while more diverse districts are showing significant movement.

When you look at the districts where few if anyone has attended college, another wrinkle appears. Biden is losing votes across districts with low educational attainment regardless of the district’s racial composition. White voters without a college degree, Black voters without a college degree, Latino voters without a college degree...the model is under the impression that Democrats are hemorrhaging support with all of them.

Age

Many have speculated this cycle about Biden having a particularly hard time with young voters. You might notice in this table how they are shifting against him in most of these districts, except, of course, all four Utah seats where white, college-educated Mormon voters are the Spiders Georg of the American electorate.

The other districts might look familiar to you, because many of them have appeared elsewhere in this article! They’re heavily Black and Latino and many of them have low levels of educational attainment.

The starkest contrast between any two sets of related parameters in the forecast has to be in Florida, where the state’s rightward sprint in the last few cycles continues. The younger, bluer, more diverse areas of the state like Miami are changing rapidly, but the ruby red retirement communities? Well...they are not getting bluer, but they are not the reason why it suddenly transformed from being the swing state to supplanting Texas as Republican California.

Region, Population Density, and 2020 outcome

The platform I use to generate interactives does not seem to allow users to play around with including/excluding variables, or else you would be looking at a table sorting districts and their respective shifts based on geographic region, population density, and whether or not Biden won it in 2020. I will try to find a better way to share that data set with you all, but in the meantime, here is a summary of various groupings of districts and how the model thinks they’ve changed from 2020 to now.

This does not tell us a ton of new information that you could not have gathered from breaking the shifts down by race and education. Changes to Biden’s level of support in the forecast are negligible at best just about everywhere except in urban areas that Biden won by huge margins. Areas that he is still winning, to be clear, just not by the amounts needed to clinch re-election.

Key Takeaways

The polls, and with it, this forecast, reflect an environment where Donald Trump has secured the support of his base that turned out for him in 2016 and 2020. Joe Biden, however, despite having a much higher ceiling, has a lower floor. Democrats appear to have a problem concentrated almost entirely in sapphire blue areas, even as they do fine everywhere else. That would help explain why primaries and special elections, where the kinds of voters who live in more upscale, highly educated areas are overrepresented, project a vibrant and engaged Democratic Party that is not having any trouble with turnout or persuasion, even as the polls reflect a five-alarm fire.

I’m not a Democratic strategist - hell, I’m barely even an election forecaster - but on the one hand, you could see this as encouraging. All Biden has to do is shore up reliable Democratic constituencies that the party has an infrastructure and a proven track record of holding together. On the other hand, that same infrastructure should theoretically be preventing this from even being in the realm of possibility, and until this is proven to purely be a mirage, it’s got to keep them up at night.

I’ll leave it up to you to decide if those shifts are durable or even credible. On a gut level, the theory of racial depolarization, or “racedep,” to the extent that some surveys have demonstrated it, is baldly absurd, and does not seem to be supported by any non-polling data. Ultimately, model outputs are only as reliable as the inputs you feed it, and it’s not for me to decide if pollsters are picking up on something real or if it’s all a mirage. But if the polls are right...this is as good of a guess as any on what that means on a granular level.