Hostage to Fortune

In for a penny, in for a dollar. My predictions for Tuesday

Nov 04, 2024

I’m going to write a post that contradicts what I said on TV a few days earlier. Which I guess officially makes me a pundit.

On Thursday last week, I was on BBC Newsnight talking about the US election. I’ve had a quite a ride over the past few months on that show. Four different hosts (gotta catch em all) and four very different periods in the election.

My first visit was the night of the Biden/Trump debate, a portentous moment, after which it seemed that the race was heading inexorably to Trump’s second term. My second visit was the first night of the Democratic National Convention, about six weeks later when everything had changed - a new Democratic nominee, riding high on a small polling lead, having ruthlessly swatted aside her potential challengers. Then I showed up again on October 23rd, probably at the peak of Trump’s vibes-driven recovery in the polling, to talk about whether Trump really was an authoritarian fascist, as John Kelly had claimed. And then finally, I rocked up last Thursday as Harris started to pull back into a tie, or maybe post-Madison Square Garden, a lead.

That mad roller coaster should make anyone feel cautious before making clear predictions. Indeed, I literally said on Thursday to a live national audience, they the race was a coin-flip and anyone telling you that someone is definitely going to win is ‘trying to sell you something’.

I am not trying to sell you something. But…

I still think the outcome is very uncertain. However, I also think it’s useful to make predictions and see what you get right and what you get wrong, as I did earlier this summer in my predictions about the British election. In that case, I was more positive to the Conservatives than many other predictions but I did underestimate them and overestimated the SNP and Labour. I nailed the Lib Dems and Reform. For whatever that’s worth.

So I am going to go out on a limb again and give you my best guess - and it is of course only a guess - about likely scenarios.

Scenarios?

Regular readers will recall I set out four scenarios for the election a couple of posts back and now I want to set out what I think the probability of each is and roughly what it would look like in terms of states won and the national popular vote. Please note the probabilities are my subjective probabilities, not drawn directly from estimating polling averages and uncertainty. This is coming, for good or ill, from my judgment. Oh, and I will spend a LOT more time discussing my most likely outcome and why I think it’s the case. And that is this…

Harris Wins Big (40%)

Predicted vote share: Harris 51, Trump 47.5

States won: Harris - MI, PA, WI, NV, GA, NC; Trump AZ

I’ll be honest, I wouldn’t have picked this option a few days ago. But I think - well, guess - that this is now the most likely of the four outcomes. So this is my modal outcome and you can call it my prediction and laugh at me when it proves wrong.

It’s not just me being influenced the Ann Selzer poll of Iowa, which to extraordinary surprise had Harris leading Trump by 3 in a state he defeated Biden in by 8. But that helped shore up a suspicion I have had that the way pollsters have adjusted their models this year hasn’t quite worked.

A lot of people have been complaining about pollsters herding, and in Nate Silver’s case, arguing that this level of herding is statistically unlikely. Given the standard margin of errors in a basic poll (remember my figures a few posts back?) we should expect more variation in polling numbers than we have been getting. That suggests shenanigans of some kind. But it doesn’t need shenanigans to get this outcome.

The reason - as Justin Wolfers and Natalie Jackson have both noted, is that Silver’s margin of error calculations only work if you assume that polls aren’t weighted. Weighting is when you adjust each observation in your sample by a survey weight to reflect how ‘important’ that observation should be (so a weight of less than one downweights and a weight of more than one upweights). Most polling firms do a lot of weighting because the sample they get online or over the phone is not a true random sample of the population and they adjust it to look more like the ‘expected’ electorate.

And here is where the trouble can come in. In 2016 pollsters got in trouble because they did not weight enough. People with higher levels of education were more likely to answer the phone / respond to an online poll, and so the samples that pollsters were finding were biased to college-educated voters. Hence they missed Trump’s support among non-college whites. In 2020, that was corrected but a new problem emerged - because Democrats were more likely to stay at home during COVID they were more likely to answer surveys. And so this time pollsters inadvertently got samples with more people who voted Clinton in 2016 than they should have - even weighting by education.

Which means for this cycle, pollsters have all been weighting by education (and other demographics) AND past vote in 2020. As Josh Clinton from Vanderbilt has pointed out, each of these weighting choices (along with weighting by party ID) can have really sizeable effects on your bottom line estimation of Harris vs Trump. In particular, weighting by 2020 vote necessarily makes your estimate of 2024 look more like the outcome in 2020. The herding that is going on is everybody using these 2020 vote weights not pollsters being scared of having a final poll that’s too far from everyone else. The 2020 election was relatively close in swing states and thus so too are 2024 polls.

But you know who doesn’t weight their surveys - Ann Selzer. Selzer in fact still uses random digit dealing (RDD) - which is the old-school method pollsters used to use back in the days where people still picked up their phones. And maybe people do still pick up their phones in Iowa (my seven years living in Minnesota give me warm feelings about Upper Midwesterners).

More grist to this mill came today from Mitchell Polling, a Republican outfit based in Michigan, who acknowledged today that both their sampling schemes were picking up too few women and African Americans (not a weighting issue in this case) and that Harris has a larger lead than they have been picking up.

And finally, Nate Cohn, who oversees the New York Times / Siena polls, which have been bouncier and less herdy than others, has argued that pollsters weighting on 2020 vote may have been keeping the polls artificially close. Now it’s also worth noting that Cohn is still concerned that Trump voters aren’t being picked up by polls properly - so he has hedged in both directions - but we’ll come to that below.

So where does this put us? You shouldn’t read too much into early voting figures, especially those released in the first few days of early voting, because Trump encouraged Republicans to vote early this time while denouncing it harshly in 2020. So partisan behaviour around early voting has changed. This has confused analysts in Nevada - even the great Jon Ralston - as Republicans did start with a huge lead that was largely eroded by the end of early voting.

But one thing we can state with confidence from early voting is that women have been out there voting at much higher rates than men - usually with something like a ten to twelve point turnout lead. In the 2020 election the final turnout gender gap was just four percent. So men will have to get to the polls in massive numbers on election day if Trump is to win. Will they? I suspect not.

Selzer thinks that survey firms have been either under-sampling women or simply underestimating what the final gender gap in turnout will be. And she also argues that abortion rights have been a huge player in this gap and that analysts have not placed sufficient emphasis on this. Now Iowa has had a particularly polarised debate around abortion so this may not travel to the same extent everywhere. But the polling miss in 2022 - where Democrats were under-estimated - was arguably best explained by a post-Roe surge in women voting, and voting for Democrats.

There is also ample evidence that pollsters have had problems surveying black and Hispanic Americans. The cross-tabs for these groups in major polls have often had Trump doing far better than he did in either previous election and indeed better than any Republican president more or less ever (save GWB in 2004 with Hispanics). There is an argument out there - which I am proud to have got onto Newsnight - called racial depolarisation or racedep: that Trump has secret sauce and has reduced the importance of race in voting. Which would indeed help him. But surveys by Chris Towler of African-Americans and surveys by Pew and Univision and Latino Decisions of Hispanics don’t show much if any racedep. And if that’s the case, given that Harris also seems to be doing better than Biden with college-educated whites and rural whites (not dissimilarly to Barack Obama…), then polls are underestimating Harris.

Now my last post was about vibes not polls. And the vibes do indeed appear to have shifted in the last few days from DC getting ready for Trump 2.0 to a gradual acknowledgment that not only could Harris win but she might be the favourite. The ridiculous crypto-led betting markets such as Polymarket have seen a sudden surge towards Harris, hopefully putting the last nail in the coffin that these sites tell you much more than who crypto-bros would like to see in power. Axios and the Washington Post have both reported fights and second-guessing in the Trump camp and confidence in the Harris camp. Now the conventional wisdom is shifting towards a Harris win.

The blame/credit if this happens will probably be put on the Madison Square Garden farrago - not least because polls of Puerto Ricans in Pennsylvania show that this really did cut through and that Trump’s attempt to make the most of Joe Biden’s unfortunate ‘garbage’ remark by riding around in a garbage truck might have confused these voters further - they think he is doubling down on the initial insult to Puerto Rico.

That will get the blame. But maybe we have all just been misled all along about Trump’s likely support relative to Harris. Ann Selzer might have pulled one of the all time great emperor’s new clothes moves, as every pollster now panics that they have been weighting their surveys too much in order to try to capture Trump voters. We will of course never really know what mattered most - vibes or polling mistakes.

Or it could just be that people remembered who Donald Trump is…

So… to the prediction itself. If I’m right it’s Harris leading Trump by 3.5 percent points once all is done and dusted (so a week or so from now) and a comfortable 308 to 230 electoral college win, where she takes every swing state except Arizona. You can see what this looks like in the map at the top of the post.

In this scenario, Harris declares victory by Thursday, maybe even by Wednesday evening. Trump declares he won but in a desultory and unconvincing fashion. And there are law suits and such but it’s basically OK.

Now for the disclaimer - do I want this to happen? Yes. So you should feel free to ignore me, I suppose. But I’ve given my reasons so I think this is more analysis than just cope.

Harris Squeaks It (25%)

Predicted vote share: Harris 50, Trump 48.5

States won: Harris - MI, PA, WI, NV; Trump AZ, GA, NC

This is the Blue Wall holding outcome (plus Nevada because well why not). This is basically Harris winning the states where at the margin she has led more times than Trump or has held a consistent lead. And losing those where the reverse is true.

This is the world of a small but persistent electoral college bias of about 1.5%, which is my feeling for what it will be this time. But a world where we basically trust the polls. National polling has a slight edge for Harris. In the swing states it’s really really tight but my sense from the people at Split Ticket (whose aggregation model I prefer because it doesn’t double-count mad polling by Brazilian pollsters) is that Harris has a very slight edge in those four winning states (by 1.6 points in WI, 1.7 points in MI, 0.4 points in PA and 0.1 points (!) in NV).

This is the case where the polls are basically right. And this is the coin-flip election where Harris wins the coin flip. Should this happen I don’t expect Donald Trump to do anything other than declare victory himself and for the next three months, probably year, of our lives to be election litigation. Fun.

Trump Squeaks It (25%)

Predicted vote share: Harris 49.5, Trump 49

States won: Harris - MI, PA; Trump AZ, GA, NC, WI, NV

This is the other side of the coin flip. Harris wins the popular vote but loses one state in the Blue Wall. Because I am annoying I have it as Wisconsin not Pennsylvania (following Wisconsin being to the right of Pennsylvania in the last two elections). But look it could easily be the reverse - or both going to Trump.

The polls also predict this, in the sense that this outcome is completely consistent with the current set of polls we have. Hence this outcome and the one above are equally likely.

If this happens then it’s likely the case that Trump either manages to get young men to the polls (they haven’t yet but clearly still could) or racial depolarisation is real and we have four billion political science articles about racedep over the next decade.

This outcome is one where Harris maybe tries to relitigate some of the voting if it’s really close but ultimately ends up failing to change the result like Al Gore did in 2000. But more likely she concedes by the weekend and we are in for Trump 2.0 Electric Boogaloo. Maybe the Democrats keep the House and constrain Trump a bit. Maybe they don’t…

Trump Wins Big (10%)

Predicted vote share: Harris 48, Trump 50

States won: Harris; Trump AZ, GA, NC,MI, PA, WI, NV

This is MAGAmania. Trump actually wins the popular vote. He even hits fifty percent of the vote. He wins all the swing states. The pollsters once again fail to pick up Trump voters, despite all their efforts to do so. Trump exiles Ann Selzer to Greenland (which he then tries to buy in exchange for Puerto Rico - it’s happened before).

This almost certainly means a Republican trifecta of Presidency, House and Senate and an untrammelled Trumpist policy program - tariffs, yuuuge tax cuts, deportations of millions of illegal/undocumented migrants, deflouridisation of the water supply…

What those of us outside the US will have to make peace with is that this would be a decision that clearly lay with the American people. Blaming Trump would not be the point. If you don’t like it, then blame the electorate who voted for it.

It would also mean that ‘national conservatism’ had a real spring in its step. It beat all the naysayers. And for the German federal election in 2025 and the French Presidential election in 2027, we would have to take the AfD and RN very very seriously.

This absolutely could happen. I’ll be honest, it’s not my preference. But it very much could happen

But it might not! I only give it a ten percent chance. I think Harris is more likely to win than Trump (65/35). Obviously I may very well be wrong. You may regret having read this post, should you be pro-Harris, for which my future apologies. But as I said, I think there’s some merit in laying out your subjective probabilities for outcomes and your justification, so hopefully we will at least know why I was wrong.

Or maybe I will be right.

I’ve been keeping busy on various broadcast outlets - OK the BBC - over the past week or so. You can hear my Rethink episodes on care and energy, here and here. My trips to Newsnight are linked above. And today I just stepped out of the BBC Newscast studios, where I chatted with Adam Fleming, Lyse Doucet and Nick Bryant, so that should be up soon too.

Political Calculus

Discussion about this post