Nolan Ryan, the backup quarterback, and Donald Trump

Andrew Gelman has an interesting analysis of my response to Nate Silver’s claims about what would happen if Napoleon had a B-52 at the Battle of Waterloo Joe Biden and/or Donald Trump were replaced on their respective presidential tickets before November’s election. ETA: Gelman points out to me that the chances of Napoleon having had a B-52 the Battle of Waterloo can be calculated as zero, while the odds of either Biden or Trump dying or suffering a sufficiently debilitating injury as to require withdrawal are non-trivially higher than zero. Which is a valid point.

Gelman’s post is full of statistical details and should be read in full, and here I want to focus on this part of it:

What about Nate’s other statement: “Trump should drop out! . . . Biden would lose by 7 points, but I agree, the Republican Party and the country would be better served by a different nominee”?

Would replacing Trump by an alternative candidate increase the Republican party’s share of the two-party vote by 3.5 percentage points?

We can’t ever know this one, but there are some ways to think about the question:

– There’s some political science research on the topic. Steven Rosenstone in his classic 1983 book, Forecasting Presidential Elections, estimates that politically moderate nominees do better than those with more extreme views, but with a small effect of around 1 percentage point. When it comes to policy, Trump is pretty much in the center of his party right now, and it seems doubtful that an alternative Republican candidate would be much closer to the center of national politics. A similar analysis goes for Biden. In theory, either Trump or Biden could be replaced by a more centrist candidate who could do better in the election, but that doesn’t seem to be where either party is going right now.

– Trump has some unique negatives. He lost a previous election as an incumbent, he’s just been convicted of a felony, and he’s elderly and speaks incoherently, which is a minus in its own right and also makes it harder for the Republicans to use the age issue against Biden. Would replacing Trump by a younger candidate with less political baggage be gain the party 3.5 percentage points of the vote? I’m inclined to think no, again by analogy to other candidate attributes which, on their own, seemed like potential huge negatives but didn’t seem to have such large impacts on the election outcome. Mitt Romney and Hillary Clinton both performed disappointingly, but I don’t think anyone is saying that Romney’s religion and Clinton’s gender cost them 3.5 percentage points of the vote. Once the candidates are set, voters seem to set aside their concerns about the individual candidate.

What Gelman is emphasizing here is that the idea that either the Democrats or the Republicans would obviously be vastly better off with a different presidential candidate has no statistical support in terms of past elections, and is even less likely to be true given this:

Political polarization just keeps increasing, which leads us to expect less cross-party voting and less short-term impact of the candidate on the party’s vote share. If the effect of changing the nominee was on the order of 1 or 2 percentage points a few decades ago, it’s hard to picture the effect being 3.5 percentage points now.

A couple of related observations:

(1) Gelman’s statement that Trump has some unique negatives is a considerable understatement. Trump has just been convicted of 34 (related) felonies. He was recently punished for sexually assaulting a woman via an award of five million dollars, and then was punished with a further award of $83 million to her for libeling her in the wake of the former verdict. He was impeached for attempting to extort the president of Ukraine in order to harm his main political opponent, and then impeached again for inciting an insurrection. He is personally responsible for half of the impeachments of presidents in the nation’s history. While Biden occasionally stumbles verbally when giving speeches or answering media questions, Trump goes on long rambling bizarre diatribes, that sound far more incoherent than anything else a major presidential candidate has ever inflicted on the public. His closest advisers include a host of convicted felons, and he himself is in the pretrial stages of three separate very serious criminal trials, at both the federal and state level.

Which brings me to Nolan Ryan. Back in the day, Bill James developed various methods for predicting the future of baseball players, that were all based on the principle that the best way to do this was to find the most similar players in the past, and extrapolate the future of the current player by assuming that it would resemble the future performance of the most similar players when they were at the same chronological stage of their career.

This method worked quite well, except it occasionally ran into what could be called the Nolan Ryan problem, which is this: What happens when you have a player whose performance is so off the charts in some respect that there’s really no one he can be compared to, in regard to the relevant predictive metric? For example, how long a starting pitcher was likely to remain effective had been found to have a strong correlation to how high his strikeouts to innings pitched ratio was. But Ryan’s ratio was, for his age, completely off the charts: At age 40 he was averaging 11.5 strikeouts per nine innings, which was totally unprecedented for a player anywhere close to his age. James pointed out that, when dealing with something that had no real precedent, precedent was much less useful for predictive purposes than it normally would be.

I would argue that something similar applies to Trump. Trump’s political history is in so many ways so completely unprecedented in American presidential politics that normal metrics for prediction are of limited usefulness. This doesn’t mean they’re completely useless, of course, but it would seem to call for a certain amount of humility on the part of analysts. By contrast, the only even vaguely analogous fact about Biden is his age, but even here Trump would be older than any other president was in American history at the beginning of his term, including Biden himself, so this “unique” liability is, as Gelman notes, of very limited salience under the circumstances.

(2) To continue these sportsball analogies, it seems clear to me that Nate Silver is has fallen for backup quarterback syndrome, which is the well-known fact that, on any team that isn’t completely dominating its competition, the backup quarterback tends to be the most popular player, because fans can so easily project their fantasies onto that player, since the starting quarterback’s flaws are viewed in real time, while the backup quarterback can bask in the future glory attributed to him by optimism bias.

Thus Joe Biden has an actual record as president, and produces measurable reactions, positive and negative, among voters, while Gretchen Whitmer or Gavin Newsom etc. are blank slates in these regards, allowing Silver to indulge in the pleasant fantasy that they somehow wouldn’t be found immediately wanting by both pundits (including, of course, Silver himself) and the voters themselves, in countless ways. This is perhaps not as true for a replacement for Trump, since Trump is such a bizarre figure that any supposedly normal Republican would be met with hosannas from reactionary centrist pundits such as Silver, but needless to say short of possible epidemiological developments this experiment is not going to be run.