Letter from Philadelphia

Where we are in the current US election and what we can and can't learn from polls

In What’s Wrong With Democracy? this week we have continued our three-part miniseries on the US election. Part one on voting and demographics had interviews with Chris Towler from the Black Voter Project, Kira Sanbonmatsu from the Center for American Women and Politics, and Mark Hugo Lopez from Pew. Part two on leadership has me talking with Lindi Mazibuko (past head of the opposition in South Africa), Lee Drutman from New America, and Darrel West from Brookings. Do listen and then get yourselves excited for my interviews on the current campaign with John Sides from Vanderbilt, Lauren Gambino from the Guardian, and Jim Messina, architect of Barack Obama’s 2012 campaign - out on Wednesday. Oh and I also did an explainer of the Electoral College for Sensemaker. Let me know what I got wrong!

I’m writing this Substack post from the city of brotherly love, Philly, also home of America’s most grotesque sports mascots (Gritty and the Philly Phanatic) and a city hall that is lit up in ectoplasm green like some FU to Ghostbusters having been set in New York. Never change Philly.

Philly is of course also the largest city in Pennsylvania, whose namesake William Penn, stands 36ft tall atop City Hall, from which every night, when lit lime green, he stalks the citizens of this fine city (maybe). And it’s the home of this year’s American Political Science Association (APSA) annual conference, which is why I’m here.

But back to Pennsylvania. Where else to be during a Presidential election year now that Pennsylvania has become the swing state par excellence. Because right now, the game is Pennsylvania - it is the ‘tipping point state’ in every election analysis - the one where the vote is closest and which could tip the election either way.

We are currently in a bit of lull in the Presidential election, at least until the Presidential Debate on Tuesday. And that’s why really nerdy fights have broken out. Over the past week, the merits of poll aggregators - websites that combine various national and state-level polls with various adjustments for ‘house effects’, survey-size etc - have been on the line.

The questions about how to deal with polls are legion. Should aggregators adjust for ‘convention bounces’, the traditional short-lived booms candidates get after the party conventions? Are Republican-leaning pollsters 'flooding the zone’ with low quality polls? Is Kamala Harris’ recent surge just the effect of ‘differential responsiveness’, that is Democrats being more likely to answer the phone to pollsters or to fill in online surveys because they feel more enthusiasm currently?

These are tricky questions. Because we are looking for auguries about a future event. Yes, polls beat tossing bones into piles at the Oracle of Delphi. But fundamentally, they are about estimating something that hasn’t yet happened, where future circumstances are going to be different from how people are feeling today.

There are now a whole array of modellers out there trying to figure out both how to combine the polls and - and this is the important part - what this means about the relative probabilities of Donald Trump or Kamala Harris winning the election.

The best known is of course Nate Silver, once of 538 but now running the Substack Silver Bulletin. Nate, who is the son of political scientist Brian Silver, has long been a bête noire of political scientists that he isn’t related to. And in particular, he has got into a fight with a series of prediction skeptics led by Justin Grimmer of Stanford University. More on that shortly.

But we also have other competing aggregators and predictors. 538 is now owned by ABC and run by G Elliot Morris - they have their own prediction model which mysteriously vanished for a while (largely because it was predicting Joe Biden would win the election comfortably) but is now adjusted and back. The New York Times, under Nate Cohn, also collate polling data as does the new team Split Ticket, whose best known analyst is Lakshya Jain.

The aggregators generally do two things. First off, they aggregate the polls, using different assumptions and weightings, related to the pollster, sample size and date of survey. As you can see from the table below, produced by Lakshya Jain, the various aggregators give pretty similar results.

They all find Harris leading Trump nationally by just over three percent. They all have Wisconsin and Michigan as Harris led by two to three percent. They currently all agree on a slight Republican lead in Arizona and North Carolina, a slight Democratic lead in Nevada and a pretty much tied game in Georgia. And then there is the big kahuna - Pennsylvania - where Harris just leads. Pennsylvania has the most electoral votes (19) of any of these states. And because of the order above in the lean of the states, Pennsylvania is the one that tips the election. You can see this from the map below from 270towin. Michigan and Wisconsin get Harris to 251 electoral votes so she needs another nineteen exactly. Pennsylvania will do the trick - as would any two of Georgia, Arizona, and North Carolina (or I guess Nevada and North Carolina).

Graph taken from 270towin.com

Many regular readers of Political Calculus will know all this. I’m not telling you anything shocking or new here. And the money is following this logic with both Republicans and Democrats pouring as much money as humanly possible into Pennsylvania TV ads. Trust me, I’ve already seen a bajillion from both sides on TVs in bars and restaurants in Philly.

But there’s a more important lesson for election watchers and that’s that in an election this close, every assumption matters. So a series of fights breaking out over election prediction websites are inevitable - if you think the convention bounce should be nil, then maybe Harris is the favourite, if you think it should be three or four percent, maybe she’s behind. If you think polls still underestimate Donald Trump, as they did in 2016 and 2020, then Trump’s edging it. If you think the pollsters have figured this out, or it’s more like the pro-Republican mistakes they made in 2022, then Harris has a slight lead. If you think that Democratic enthusiasm is leading more people to ID themselves as Democrats in polls, then Harris is ahead. If you think polls have ‘too many’ Democrats - as British pollster James Johnson contends - then Trump is ahead.

WE. DON’T. KNOW. And we won’t for at least two months. And perhaps never. Close elections are inherently hard to predict. And personally I don’t buy the kind of argument that Nate Silver is fond of making - that political betting markets, where people have ‘skin in the game’, will provide more insight. First because some of them favour Harris (PredictIt) and others Trump (Polymarket). But second, because most of the bettors don’t have much more information than you or I. And third, I don’t think political betting markets have been especially accurate in the past decade.

In any case I promised more Nate Silver vibes. Well Nate is on a tear. Here are some sample tweets.

‘Partisan dorks’ vs ‘boring academics’ - a gunfight for the ages.

Anyway, what’s got Nate’s goat is a Politico piece by Justin Grimmer that builds off a working paper he wrote with Dean Knox and Sean Westwood. The basic gist of Grimmer et al is that we don’t have enough data to adjudicate among the various prediction models being offered. Should Nate Silver include Republican-aligned pollsters who some claim are ‘flooding the zone’. Should he down-weight them? Should he have had a convention bounce for Harris (as he noted today, removing it moves the election from being >60% Trump to evens again)? What is the role of fundamentals (incumbency, economic growth, etc), which had led the 538 prediction model to favour Joe Biden, even as he was tanking in the polls? Do we even need polls at all to predict elections or can we rely on performance in the primaries or incumbency like Helmut Norpoth’s longstanding ‘Primary Model’?

Grimmer and company say that we won’t ever know. That presidential elections are so rare, and the assumptions so legion, that we lack the data necessary to choose among models. By their reckoning it would take between 56 and several thousand years to ascertain if any of these models are better than any of the others. So from this perspective anyone can choose a model that gives them the outcome they want (Trump fans head to Silver Bulletin, Harris fans to Norpoth), without that being the ‘wrong’ choice.

Now there are some important and limiting assumptions with Grimmer and company’s argument - the most important, as Andy Gelman has noted, being that Grimmer et al use the binary election outcome as the outcome variable, whereas we actually have more data than that, namely the relative vote shares of each candidate, which we could use to adjudicate more effectively among models. So we might not have to strictly wait millennia. Moreover, there’s nothing stopping us from using, shock horror, data from other countries to help calibrate a model. But nor am I hugely convinced by Nate Silver’s standard comeback to criticism, which is to say ‘let’s bet on it’, as if every academic debate can be resolved by laying money on it.

All that said, I am absolutely here for Nate Silver’s drama - the guy knows exactly what buttons to push and just passed 200,000 Substack subscribers, so kudos on that front.

And I am sure Nate Silver can join up with his political science critics, like some Professor X / Magneto team of rivals, against our common enemy - Alan Lichtman, Professor of History at American University, and self-proclaimed sage of Presidential election predictions with his infamous ‘keys’ to the Presidency.

Lichtman, somehow, has managed to convince (trick?) the media to cover his Rube Goldberg prediction technique. I would say model, but it’s not a model in a way that social scientists would understand. It’s a series of factors that he thinks are important, where he uses his judgment to choose the favourite, adds them up and declares a winner. It has roughly the same methodological rigour as the scoring system in Numberwang.

I strongly encourage you - should you have a NYT subscription - to watch the single weirdest video about election prediction I can imagine. For my British audience, imagine Michael Fabricant was a historian racing against several senior citizens around an athletics track.

Now Lichtman claims his keys favour Harris - which to be honest, I do think is very marginally the most likely outcome - but honestly, I wouldn’t read too much into it.

So, what should we be paying attention too for the next few months if we are not placing our trust in Alan Lichtman’s tarot cards, or fretting about small changes in Nate Silver’s predictions?

I have an episode of What’s Wrong with Democracy? coming out next week - you’ll be able to find it here. It’s on campaigns in US presidential elections and I have an amazing set of interviews - with John Sides, Professor at Vanderbilt and the author of three books about the past three Presidential campaigns, Lauren Gambino, the chief US political correspondent for the Guardian, who is currently following the Harris campaign, and Jim Messina, who famously led Barack Obama’s successful re-election campaign in 2012 (as well as David Cameron’s successful one in 2015).

I won’t spoil the lede - please listen! But let me highlight a couple of things that came out of those conversations. First, that a huge amount is already baked in - the economy, border crossings, incumbency - and so the campaigns may not be able to shift the territory that much (at least in normal times). Second, that nonetheless, the decision to step down by Joe Biden and the successful rollout of Harris as his successor is unprecedented and really has shocked the campaign.

And third, and in Jim Messina’s view, most important - it’s all about enthusiasm not polling. That enthusiasm can be measured partly when pollsters ask voters how they are feeling about the candidates. Gallup has found that Democratic enthusiasm is night and day compared to earlier in the year.

Gallup enthusiasm data available here.

Enthusiasm can also be seen in that most American of campaign inputs - money. Kamala Harris’ campaign raised $361m in August alone (allegedly from over three million contributors), whereas Trump raised $130m. And enthusiasm also can be seen among political activists, from people staffing small offices around states up to various party elites.

On polling it’s a tie game, where if the polls are right Harris has the very slightest of edges. On enthusiasm, right now, Harris has a clear advantage. And so IMO, I would make her the favourite. I won’t attach probabilities. I won’t be betting with Nate Silver, but that’s my hunch.

To finish, here are a few things I think each campaign should be thinking, perhaps worrying about, over the next two months.

What should the Harris campaign be worried about?

1. The polls might still not be picking up Trump support. I know I said don’t focus completely on polls. But they do provide a map of the territory on which the election is being fought (that’s why the campaigns are focusing on seven states in particular). But if we are in the world of 2016 or 2020, Trump will beat them again, and Dems are in for yet another night of disappointment. The Conservative Party in the UK recently beat their (rather lousy) polling numbers, by about four points - so there’s recent evidence. Though note that Nigel Farage’s populist right Reform UK - perhaps a better comparison to Trump - slightly undershot their polls.

2. Keeping the enthusiasm going. Harris’ rollout was phenomenal. I never saw it coming. It was professional, effective, funny, coherent. Shocking. It has led to huge amounts of earned media and minimal press critique, except for the criticism that Harris has been avoiding interviews - which, tbh, is the classic move made by candidates who are leading and don’t need to ‘get back in the game’. But it probably won’t last. The media will get bored. They will look for some controversy. Or Trump’s campaign will finally have a consistent attack line. And this will be a crucial moment for the Harris campaign. Can they power through and most importantly, keep up enthusiasm among Democrats during the ‘dark night of the soul’ period?

What should the Trump campaign be worried about

1. Their candidate. Trump cannot keep his mouth shut. Yes, yes, that what people like. Until he becomes the drunk uncle who won’t shut up. A disciplined Trump should have already sewn up this election. Harris has weaknesses as a candidate and Trump could hone in on them. But he can’t seem to figure out how to respond to her. Something about her sets him off internally. And the rambling and insulting could really backfire in the debate on Tuesday. Trump got through the last one with Biden pretty cleanly but largely because Biden’s absolute corpsing in the debate covered up Trump’s pretty bizarre performance. He may not be so lucky this time.

2. Their inability to define Harris. Is she a Communist? Or is she a flip-flopper? Is she a useless lightweight? Or is she behind all the worst outcomes of the Biden Administration? Trump’s campaign - not just the man himself - simply cannot seem to focus on an attack line. They have had weeks, nay months, to figure this out and they still don’t. Seems to me the obvious thing to do would be to run an incumbent-attacking campaign, given the dissatisfaction many Americans have with the economy. But neither the Trump campaign nor DJT can keep to this.

But perhaps this is all moot. For years, political scientists thought debates didn’t matter. And then they suddenly did, divebombing Joe Biden’s chances of remaining President and producing this crazy election we now have. So perhaps they will matter again on Tuesday. Perhaps I’ll have to write a follow-up Substack revising all my advice and predictions in this post. But unlike Nate Silver, I won’t have to worry about picking the wrong percentages or adjusting my model. So, if he wants a bet, I bet my week will be more relaxing.