Immediate Post Mortem

I should not have drunk the Selzer

I was wrong.

On Monday I set out the four scenarios that I had written about in October for the US Presidential election - a big Harris victory, a tight Harris victory, a tight Trump victory and a big Trump victory. When I had initially written about those scenarios I had argued, following the polls, that the election was indeed a coin flip so all were equally likely.

To quote myself talking about people claiming the election was lean Harris or lean Trump back on October 15: “There is simply too much uncertainty right now to make either claim with real confidence.”

But then I played myself. In my final prediction post I argued that the election was lean-Harris, giving the first two scenarios a combined 65% subjective probability to 35% for a Trump victory. And importantly I argued the big Harris victory was most likely and the big Trump victory was least likely.

It’s too early to know for sure if we are in small or big Trump victory territory. But it’s not too early to say my predicted probabilities were wrong. Had I stuck to my more balanced assessment a week earlier - on Substack and on Newsnight - I could have come back and said - see, I told you it was close and anything could happen. That is not, however, where I ended up.

Now this is a very thin reed but the only (and really the only) benefit of having made a more determinative but wrong prediction (Ben of November 4) rather than just waving my hands in the air and saying who knows (Ben of October 15, or even Ben of Newsnight of Oct 31st) is that we can look at why I got it wrong right at the end. Now we don’t have all the data yet and I hope to come back to this in a more rigorous fashion later. But even at this point I can set out a number of reasons I got it wrong.

---

The first and most obvious reason is standard issue wish-casting. If you are going to go from an analytical, polls are too close to call, perspective to forcing yourself to make a prediction - as I did - then you are going to end up following some mix of your gut and your hopes.

I tend to be a glass half-full person and so since I preferred a Harris victory I was probably looking for data that supported that and down-weighting data that undermined it. I am, for my sins, a human. And I would imagine many of you, whatever your preference in the election, were guilty of the same frailty. But then most of you didn’t write a Substack post about it.

---

The second reason was that I got Selzer-pilled. Like so many other people I saw Ann Selzer’s pro-Harris numbers from Iowa, produced by simple random digit dealing and very minimal survey weighting, and I thought she was picking up on something that other, herding, pollsters were missing. I was concerned that by weighting on previous vote, the pollsters were producing results too similar to one another and (though as Andy Eggers noted to me later, this is more tenuous) that this would push results towards 2020’s very close results in the big swing states.

But Trump is going to win Iowa by similar numbers to 2020 and Selzer was just plain wrong. And the pollsters actually look pretty good - all the swing states are going to be pretty close it seems - just in Trump’s favour not Harris’s. Had I trusted the polling averages completely I would have stayed with my initial take that it was too close to know and both sides were equally likely to win.

---

And then there’s the third reason. A belief that Trump had a polling ceiling. That he would find it hard to crack 47% of the electorate, as in 2016 and 2020. But those elections had more votes heading to third party candidates (especially 2016) and RFK Jr dropped out and endorsed Trump. Ultimately, I suspect many of us underplayed the importance of that decision - it marked the start of a decline in Harris’s polling numbers in August / September that was only, it seems temporarily, halted by the Harris / Trump presidential debate.

If Donald Trump does, as seems quite likely, win the popular vote, then he has achieved something no Republican has since 2004 and he has achieved a serious and impressive victory. The American public were not fed up with him. He did not face an inevitable ceiling. He is the legitimate winner of a fair democratic election.

I know that last sentence will annoy / upset some people. But I think Democrats will need to make their peace with this. This was the American people’s choice - not the perverse outcome of American’s arcane electoral college (presuming Trump win’s the popular vote). And it was not a choice made in ignorance. It was one made in full recognition of Trump’s first term, the January 2021 insurrection, and Trump’s behaviour during this campaign.

I do not believe that Trump is a great friend of liberal democracy. I think it’s reasonable to believe he will weaken or erode some components of a well-functioning liberal democracy - whether by rhetorical attacks on the press, dismantling of the administrative state, or the use of agents of the state to round up illegal immigrants or crack down on protests. In my view these would be bad outcomes.

But I also don’t believe that Trump is the ‘end of American democracy’ or that American democracy really was ‘on the ballot’. Checks and balances will remain. I expect the midterms and the next Presidential election to be free and fair. The First Amendment will protect free speech. And so forth.

And it’s pretty clear that the American public did not think Trump would mark the end of democracy either. This was the core theme of Biden’s re-election campaign and was picked up by Harris and basically I think it flopped because people did not believe it was true. Now perhaps they and I are wrong. It is of course possible that the electorate signed its own death warrant. It would not be the first such occasion. As you may note, my record on predictions is, um, sketchy.

---

But this takes us to the fourth reason. People’s memories of the Trump administration were quite favourable. Until COVID broke out at least. And it seems that - as with Boris Johnson for example - the public weren’t inclined to blame Trump for the initial period of COVID chaos in early to mid 2020. For many people, their memories of Trump from early 2017 to early 2020 were strong economic growth, lower taxes, stable inflation and a calm international environment. And those same people looked at high inflation in 2021-22, high levels of border crossings, and a chaotic international environment, and they used a simple heuristic. Things were pretty good under the last guy. Things are less good under these guys. I vote for the last guy.

It is very tough to be an incumbent politician right now. But it’s even tougher when you are up against an opponent who can associate themselves with a better earlier period of office. I think Harris ran a decent campaign. We can question the VP choice, or the media outreach, or the messaging. But ultimately she was defending an incumbent government that was unpopular. Sometimes there’s just very little you can do.

In sum, I had all kinds of gut reasons why the polls might not be picking up Harris properly. But I ignored the most crucial gut reason of all. How were people feeling about the country. And they were not happy.

---

Finally, the fifth reason I got it wrong is that I was unconvinced by the racial depolarisation or ‘racedep’ argument. Though I am still glad I got to explain what it was on Newsnight. This is the idea that racial divides in voting would weaken this election, especially among ethnic minorities. I was skeptical about it. I had looked at polls that focused on black voters alone, or on just Hispanics and I couldn’t find strong evidence that Harris was putting up weaker numbers among blacks or Hispanics. It felt like these surveys were telling us that Harris would perform similarly to Biden.

By contrast, the case for racedep was that if you looked at cross tabs in the national polls, Harris was clearly struggling with ethnic minorities. Well the national polls proved correct. That makes me wonder if there is more social desirability bias in polls that focus on single ethnic groups - that respondents answering a survey focusing on Latinos think there are answers they ‘ought to give’ that they wouldn’t worry about in a more general survey.

---

So to summarise - first lesson is trust the numbers even when they conflict with your priors or preferences. Second lesson is that the pollsters actually know what they are doing with all their adjustments. Third is that Trump just is really popular and most Americans don’t think he is the end of democracy. Fourth is that it is so hard to be an incumbent right now that your gut instinct should always be that the incumbent will lose. And fifth, the racial basis of American voting does appear to be changing in an important way - one that challenges the logic of how the Democrats have been forming their electoral coalition since Barack Obama.

Is there a sixth lesson? That you shouldn’t make predictions because then you will end up having to write a post like this when you are wrong?

I think I am OK with making predictions as long as I do something like this afterwards to interrogate my own mistakes. If my Harris positivity on Monday ruined the election for you then perhaps you will be less sanguine about this decision. In which case, my apologies.

But hopefully most people who read this Substack are interested in thinking about politics. And, beyond everything else, we will have to do a lot of thinking and rethinking about politics in America, the UK, and the rest of the world after this election result. I suspect I will be wrong again about some things, right again about others, but that, emphatically, I will not lack for things to write about.

---

Now wish me luck for talking about the election results in media appearances this week. In particular, I’ll be on Trendy with John Curtice and Rachel Wolf tomorrow. Do listen in, at least for Rachel and John’s wisdom if not mine!