US election: predictions

Time to stop hiding behind snide criticism of everyone else, and take the risk of being horribly wrong. Time to put out what I think will happen in November.

The state of the nation

As I’ve previewed a few times, I think the commentariat is in danger of confusing an extremely uncertain race with an extremely close race.

Some things I think are true:

• I think that most of the pollsters are doing their best to get things right, with solid methodology; and so the polling averages are as good as they can be. Rasmussen, AtlasIntel, Trafalgar and perhaps some others aren’t doing their best to get things right - they’re doing another job. But adjusting for this only shifts the average a tiny way towards Harris.

• I think pollsters are making an error in starting to weight by recalled vote - an over-reaction to the failures of 2020 and 2022. It might help, it might not, but I think it’s just adding noise and uncertainty at this point. If this is right, it results in a slightly larger tilt towards Harris, but we’re still in toss-up territory here.

• I think the methodologies will turn out to be systemically wrong in one way or the other to historical levels (around 2ppts or beyond, and more at the state level), and because these errors will tend to turn out to be correlated across states, one side or the other will win by more than 100 EV. I put a very low probability on the errors being uncorrelated.

As a result, my expected outcome looks something like this:

---

Translating into an electoral college expectation

So, what does this mean when we translate to Electoral votes?

Most people would say that we only really need to look at the swing, or “toss-up” states. That Democrats have 226 EV banked, and Republicans 219, and it’s only the remaining 93 that are up for grabs. I don’t quite believe this, but I am willing to accept that the swings that you’d need to get states like Minnesota, Texas, Virginia and Florida to flip outside their historical voting patterns are in the 20%-or-below bracket.

I’d also note that a full-on blow-out landslide scenario is more likely for Harris than Trump, just because both Florida and Texas (70 EV total) fall into the range of a large (5-7ppt) polling error, while Harris has about 31 EV at the same level of risk.

So, back to the swing states: there are seven of these, worth 93 Electoral Votes.

Here are these seven states, and the latest polling averages (via 538 - others give similar values).¹

So, given that a reasonably correlated 2ppts swing in either direction would allow Trump or Harris to sweep the lot, that’s what we should expect. That’s just the kind of error we tend to get in US Presidential elections.² There’s only one possibility in the last six cycles that would even give us a close race, and even then, only if the error broke towards Harris.

So I think we’re about at:

75% - Convincing victory - winner gets 300 EV or more

25% - A close race - winner gets 299 EV or fewer

---

Of for ****’s sake, who’s going to win?

But although all this about the closeness of the race is helpful, and perhaps even relevant to important events, given the potential for close votes to be contested, it’s not what most people want to know.

What most people want to know is who is more likely to win: Harris or Trump, and hang whether it’s in a convincing victory or not.

Up until last week, I would have said (just) Harris - I think the questionable pollster methodology is more likely to be slanted away from her than it is away from Trump. And so a significant error is slightly more likely to lead the actual vote to break towards her than to Trump.

But now early voting numbers being reported by Jon Ralston over in Nevada (the only early voting commentary anyone should pay any attention to³) have shaken this belief. And shaken it hard. If we take his reporting at face value - and I think we should - then Harris is struggling in Nevada and falling well behind in the early voting. That is, so far she is doing significantly worse than the polling in that state suggested. Now, Nevada itself is not a huge issue - it only represents 6 EV. But if we apply the idea that polling errors are likely to be correlated across states, then the consequences are fairly obvious - she is much more likely to be struggling behind her polling in other states too. That’s how conditional probabilities work. And very quickly, Harris looks to be in trouble.

On the other hand, the early voting in Nevada is only a weak signal - it may well just be showing us a realignment post pandemic of how different types of people vote. We will know a little more soon, and a good get-out-the-vote could swing it back on election day, but I think we must take it seriously enough that these numbers significantly shift the overall chances.

Given all this, I think now the chances now look something like this:

• 50% on a convincing victory for Trump

• 25% on a convincing victory for Harris

• 25% on close race, in which polling errors aren’t well-correlated - and we can cut this down the middle.

Overall, this gives a 62%/38% split in favour of Trump. Which is - coincidentally - not far off where the betting markets are.

Now, these are not the probabilities I’d personally want at this stage, and to be clear, they’ve switched considerably in the last few days (since seeing the emerging early voting patterns, in fact).

I also don’t claim to understand anything about how American voters think. In fact, these patterns make me even more sure of my inability to comprehend than I was before. I just think that this picture is close to where you end up objectively, if you take seriously the historical patterns, the most likely dynamics of the polling errors, and what we know about the early voting patterns.

---

UPDATE: 4 November: Jon Ralston in Nevada has gone full reverse ferret. Despite huge differences in favour of Republicans in early voting patterns, he is clear on his final prediction.

So, what does this change? Well, Ralston’s prediction - if we take it as gospel - would represent a 0.5ppt-1.2ppt swing towards Harris versus the polls (depending on who’s polling average you take).

And if I continue to rigorously apply the logic I’ve been applying so far (there will be polling errors, they will be significant, and they will tend swing in the same direction), then this simply reverses that tiny push at the end taking me from one “hump” of the probability distribution to the other. An error of that size is slightly smaller than I’ve been expecting, but if it breaks towards Harris and is correlated across states would be enough for her to get over the line in a fairly convincing manner (especially as the polling averages have also slightly shifted towards her in the last few days).

So, do I want to take this tiny sign from one analyst - Ralston - in a state with a grand total of six Electoral Votes, and have it reverse everything in the last few hours before polls close? Or do I want to stick to my guns and stay where I am?

It’s pretty much a choice between being consistent on methodology, and consistent on answer.

I’ll be consistent on methodology. I still think the overall distribution is bimodal, I still think the errors will be correlated. And I am still extraordinarily swayable by the tiniest signs that the break is going one way or the other, which - honestly - will make these last hours one hell of a rollercoaster. The terrors of a bimodal probability distribution.

So, Uno-Reverso we go:

• 50% on a convincing victory for Harris

• 25% on a convincing victory for Trump

• 25% on close race, in which polling errors aren’t well-correlated - and we can cut this down the middle

And don’t believe anyone who says they are sure.

1. It’s worth noting that with the adjustments I just mentioned on the polling averages, then PA (and potentially NC) flip from a narrow edge for Trump to a narrow edge for Harris. But this doesn’t matter much since I don’t think it is going to happen anyway.

2. Almost certainly because Presidential elections are the ones where low-engagement voters - notoriously hard to reach in polls - actually turn up to vote.

3. Georgia early voting suggests much better news for Democrats. But - as I’ve said already - the historical rule is to ignore everything that isn’t Nevada and not run by Jon Ralston, as simple noise.