Where the 2024 election stands today
There’s a lot of copium right now on Twitter about Biden’s chances. I wanted to write this post as an (hopefully) unbiased view of the current outlook for the 2024 race.
National Margin
Let’s begin with poll numbers. Thanks to umichvoter on Twitter for this chart:
(I’m going to disregard the two polls on 2/1 and 1/30, since if you include those you should probably also include a bunch of other polls from that time period)
Taking a raw average of these gets you a Trump+0.9 race. I’m generally of the opinion that if you have enough data you should just take the highest-quality pollsters and discard the rest, and if you do that (only taking A or A+ 538-rated pollsters) you get an average of Trump+4. However, there are only 4 of these A/A+ polls and they are older, so maybe this isn’t the right thing to do in this case. If we really care about recency, just taking the polls since 3/10, we get an average of Trump+0.75. The right thing to do of course would be to plug this all into a moving weighted average, but since I don’t care enough to do that, I’m going to just call it Trump+2 and move on.
What about special elections? Nathaniel Rakich compiled a nice list of results from special elections:
As you can see, if you take all the elections since the midterms, the average swing from their weighted partisan lean is D+5% while the median is D+8%. However, the starting point is very important; if you instead take everything from and including the 2023 elections on 11/7, you get an average swing of R+0 and a median swing of R+2. I think this is the correct way to do it; the winds have clearly shifted recently. Special election data is really noisy so take this with a heavy grain of salt - removing a few elections shifts both the mean and median by quite a bit - but if I had to put one number on it I’d go with the data I’d go with R+1, meaning that special elections imply that Trump should win by 1%. Which is in line with polls.
How much should you care about any of this, 8 months before the election?
G Elliott Morris of FiveThirtyEight (still feels funny to say) argues not much. He presents this graph, showing that at this point polls have a 8 point average error to the election.
He said on Twitter that the r^2 in this period for predicting november results from March polls is 0.25.
He also points out that there is more noise in this year’s polls than usual, implying a higher amount of non-sampling error than usual - about 1.5x as much.
I honestly don’t love results that go back to 1972. Let’s look at the most recent elections:
Which… yeah. This is still quite a lot of poll shifting. That said, there is quite a bit of mean reversion in this data, fluctuations that go back to a steady middle point. You have to be careful assessing mean reversion from the data - since you only know the mean after seeing all the data, a luxury you don’t have for future prediction - but it’s definitely real. The mean of this year’s polling for the past few months has been Trump+1.
Still, I don’t think any of these comparisons are quite apt, because we’ve never had a situation where both primaries are settled this early. There was some polling that like half of voters just didn’t believe it would be Trump v. Biden again; I haven’t seen anything recently, but I think this has probably alleviated somewhat since Nikki Haley dropped out and Biden and Trump won all the state primaries. We have never seen an election where voters know the candidates so well already, given that we went through this 4 years ago, and given that Trump has gotten constant media coverage these past 8 years. We also have never seen a presidential election where there are (likely) no debates, which IMO serve as a major reminder for voters to start paying attention.
My conclusion from all of this is that while national polls aren’t that predictive this far out, you really shouldn’t expect them to get much better until literally October. The data from 1972 to 2020 shows that average error goes down from 8 to 6 in June and then down to 4 in November, but IMO there are a lot of really good reasons to treat this year’s March as most years’ June.
The Fundamentals
I’ll be honest, I’m pretty skeptical of fundamentals. I think what you can draw from economic conditions is probably already reflected in topline polls, and the little that is not probably matters less than it did in the past due to increasing polarization.
But ok. David Shor writes that incumbent presidents get on average 51% of the vote. Even that probably applies less to 2024 because Trump has this quasi-incumbent status.
G Elliott Morris writes that an economics/presidential approval joint fundamentals model predicts a Trump+1.5 race. Though here too I’m not sure that we gain anything by using presidential approval to predict Trump v. Biden standing, when polling error across the two is likely correlated. A pure economics model would see Biden as the favorite, but it’s not clear how much we should take from that when economic sentiment is so meh.
How Much Biden Has To Win By
What the national margin has to be for Biden to win is a good question. In 2020, the tipping point state was Wisconsin at Biden+0.6, meaning that the Electoral College’s bias was R+3.9, so Biden had to win by 3.9 points in order to win. I expect this bias to shrink in 2024, given that the 2022 map had a smaller bias. I’m basing all this 2022 analysis on this great split ticket piece. In 2022, Republicans won the house popular vote by 1.5% after accounting for uncontested seats, while the tipping point state implied by house results (ignoring Alaska which will obviously vote R in 2024) was either Wisconsin at R+1.7 (if Biden wins NE-02 like he did in 2022) or Arizona at R+2.2 (if Biden does not win NE-02). I think it’s very likely he’ll win NE-02 by more than his national margin, since it is a well-educated secular district which will swing left, leading to a bias of R+0.1 - about even!
We have very little state polling to go off, so any prediction for the 2024 electoral college bias at this point consists of the 2020 state partisan leans, the 2022 state partisan leans, a bit from the 2016 partisan leans, crosstabs from 2024 polling, and just general hypothesized demographic shifts. The one other data source I would look at is party registration, which shows Republicans doing very well in Florida.
My guess is that the electoral map will look pretty similar to 2020, with a few changes. My guess is that 2024 will feature significant Democratic gains among secular voters like we saw in 2022, significant Republican gains among young and minority voters (especially Arabs & Muslims) like polls are suggesting, and general increasing education polarization like we’ve seen every year since 2016. I expect the battleground states to be WI, MI, PA, AZ and NV (though NV’s electoral votes matter less).
If I have to give a number, my overall median prediction for the partisan bias in 2024 is R+2.2. There’s more uncertainty around that number than there was in 2020; my 95% CI is probably like D+3 to R+7, which I could probably narrow by thinking more about this and talking to people.
(The tiny bit of state polling that we do have confirms this story)
Other Factors
Are there any other factors that we can point to for forecasting? I think there are three worth discussing.
The first is that, very plainly, Biden is better at politics than Trump is. He handily beat Trump during the first 2020 debate, and overall he has a capacity for thinking that Trump lacks.
The problem with this type of reasoning is that such better political skills should show up in the data, and I don’t think they really do. Biden’s approval rating is worse than Trump’s. 2020 saw a consistent lead for Biden, but that was probably a mirage all along, as he won by 4 points less than polls said and did much worse in swing states. Democrats’ wins in the 2022 midterms probably came in spite of him rather than because of him, shown by individual senate candidates outperforming the general environment. I think it is probably true that while Biden is generally better at politics than Trump, he faces unique media and other challenges that Trump lacks. I speculated on what some of these challenges are here.
The second is that voters have recently warmed up a bit to the economy, and maybe they’ll warm up more as time goes on:
(red line is where you’d expect economic sentiment to be based on economic factors)
This is really an unprecedented level of time lag between the economy improving and the ~vibes~ shifting, though. Could another 8 months be enough to snap voters back to reality? Maybe?
The third is Trump’s trials. The DC trial, the one that I thought had the highest chance of resulting in a conviction before November, will not finish before November because the Supreme Court has taken up Trump’s legal challenge - that as president he has total immunity - for the last day of their term.
I think it’s very unlikely that any of the other trials result in a conviction before November. The Florida case is being delayed to hell by the Trump-loving judge, though there is the off chance she gets kicked off the case. The Georgia case has been delayed because Nathan Wade, one of the prosecutors, got kicked off the case for his relationship with the main prosecutor Fanni Willis, and now Trump can appeal the decision that Fanni Willis gets to stay. The only one that seems plausible to finish before November is the NY case, which is on shaky legal ground and was just delayed 30 days because of some bullshit.
Let me expound on the NY case for a bit. It’s scheduled to begin in mid-April, which gives it enough time to finish if there are no more delays. The situation there is that he is being prosecuted for hush money payments to Stormy Daniels, but these aren’t a felony under NY State law unless they’re used to cover up another crime. The other crime being alleged here are campaign finance violations, which are federal crimes (albeit ones that are almost never enforced). NY State Courts have never ruled if hush money payments can be considered a felony if they’re used to cover up federal crimes, particularly ones he was never convicted of. But ok, the judge in this case is a liberal, and the NYS court has a liberal majority, so maybe the legal theory survives appeal before or after the case comes to conviction. Maybe all the other delay tactics tried by Trump fail in the state courts.
The other thing to note about this NY case is that it doesn’t carry mandatory prison sentence, and if Trump is treated like any other defendant, he probably wouldn’t face prison time as a first-time defender. If that happens, I expect voters to treats this like they do his civil law penalty; that is, not respond at all. But he may not be treated like just another defender by the liberal judge - he is, after all, Donald Trump.
But even if everything holds up in the state courts, SCOTUS could also very easily step in and throw out this NY case by referencing the “rule of lenity” that gives the defendant the leg up in ambiguous statutes (despite this rule of lenity almost never actually being used by the conservative majority), either before or after the case concludes. If it happens after the case concludes but before the election, it’s probably going to be an emergency order. I can very easily even imagine SCOTUS ruling in favor of Trump like 3 days before the election, creating a flurry of positive flurry that ends up helping him.
The Supreme Court seems to be on Trump’s side. The blatant manner in which they delayed the DC one along with the stupid logic they used to reject the 14th amendment challenge to Trump’s candidacy suggests that they really do not want legal issues to complicate the 2024 election. If the FL or GA cases unexpectedly look like they will finish before the election, the court could easily take up a bullshit challenge like they did in DC.
If he were convicted without an immediate overturning, how would it affect his chances? I found 4 polls, all from early February, that ask about the Trump v. Biden race if Trump is convicted:
Marist goes from a Biden+1 race to Biden+6. Biden+5 swing
NBC News goes from Trump+5 to Biden+2. Biden+7 swing
Harvard/Harris goes from Trump+6 to Trump+0/+4/+8 for the FL case / GA case / DC case respectively. average of Biden+2 swing
WSJ goes from to Trump+2 to Biden+4. Biden+6 swing
That’s a Biden+5 swing on average.
What is my prediction? I put the chance that one of the non-NY cases finishes with a conviction and without emergency SCOTUS order overturning it as like 3%, with a median swing of 2-3 points once Trump has the ability to swing the narrative. I give a similarly low chance that the NY case results in a conviction with jail time that is not overturned by the time of the election, and expect it to shift polls even less, maybe 1% on median. I don’t think a NY conviction followed by SCOTUS overturning or a NY conviction without prison sentence will do much.
Overall, the criminal trials don’t mean much.
Ok, that’s enough talk. Gun to my head, what’s my p(Trump victory)?
60%. There are reasons to be uncertain, but there are really no strong reasons to think things will shift in Biden’s direction.
(Or at least, that’s the number I get from thinking about it myself. In practice I’d also update off other people, which messes with public epistemics when other people then update off me.)
Trump really might do it again. Fuck.