Trump, Biden & Occam: Make Elections Boring Again!

The election is over. So is the Electoral College vote. So is Congressional approval of the Electoral College vote count. And so are 60-some lawsuits. But after all that, millions of Americans still believe that the 2020 presidential election was stolen—and so far, five people have died as a result of this belief. Should the “#stopthesteal” claim be taken seriously?

Explaining what needs explanation

Viewing the election from a partisan perspective is easy: the candidate I preferred won (or lost) the official election count; therefore, more or less by definition, the election was honest (or was stolen). Unfortunately, we don’t learn much from approaches like this, other than the political preferences of whoever we’re talking to. If we want to actually learn something, we need to put down our MAGA hats (or whatever kind of headgear Biden voters wear) and put on our social-science mortarboards.

OK, now that we’re all academic, how do we proceed?

The guiding principle for our thinking should be Occam’s Razor: the principle that, given some phenomenon we want to explain, we should prefer the simplest possible explanation—the one that requires us to make the fewest assumptions. For example, a year and a half ago I broke my knee riding an electric scooter in Tel Aviv. A complex explanation is that invisible demons magically hid the bump that brought my scooter (but not me) to a sudden halt; the simple explanation is that I was stupid. See? Simple!

So when we look at the results of the election, we want to find the simplest explanation of these results as they’ve been presented to us. The normal course of events—that is, the simplest process that would lead to any given election result—is that a campaign took place, people voted, people administered the election, the system functioned more or less the way it was designed to function, and at the end the guy who got the most votes won the election.

If we think that something other than that happened, we necessarily believe that the process was, in some way, more complicated than the simple explanation—somebody intervened to gimmick the results, such that the final tally does not accurately reflect the good-faith votes that were cast.

When things are simple, the results we get should bear some relationship to our reasonable expectations. Few events are as intensely forecast as US presidential elections; there is a whole industry of “scientific” (we hope) political pollsters, and in turn there are a number of outfits that collect and analyze poll results to try to make predictions more reliable than any single poll would produce.[1]

There is also a lot we can learn from the history of past elections, and of course there’s everything we know—much of it admittedly rather subjective—about the parties, candidates, and circumstances of the current election. A simple explanation of an election result should have some relation to all this information: yes, surprises happen, but really big surprises don’t happen all that often, and when they do it’s legitimate to try to figure out why.

States and snakes

US presidential elections, for better or for worse, are not decided on the basis of the national popular vote—that would be too simple. Instead, each state, in essence, holds a separate all-or-nothing election (with the minor exceptions of Maine and Nebraska, which can split their electoral votes); and the combined state results (with less populous states having a disproportionate influence) lead to the national result. So when we talk about fraud in the 2020 presidential election, we need to be more specific: What state(s), exactly, did Joe Biden win that he wasn’t predicted to win?

Let’s start by looking at the expectations of the polling aggregators—for example, the “snake chart” on fivethirtyeight.com (scroll about halfway down the page). Comparing the “snake” to the election results, we see that every single state that Biden won was one where he was leading in the pre-election polling averages—and two states (Florida and North Carolina) ultimately went for Trump despite the fact that they were polling as “leaning Democratic”.

When a candidate wins a state that s/he was considered likely to win, that’s simple—we don’t need an elaborate explanation, and in fact we should actively avoid elaborate explanations, because Occam. (To be more precise: because they are less likely to be true than the simple explanation that leads to the same result.) Unless we can find some states where Biden won unexpectedly, there is no point in looking for a “stolen” election; the results that were reported were exactly what the pre-election polls predicted, except that Trump actually did a little better than the polls showed.

As he did in 2016, Trump performed better in the actual 2020 result than he did in the polls. The late polls, on average, forecast an 8% nationwide Biden lead, while the actual result was a 4.5% Biden victory; and, as mentioned, Trump took North Carolina and Florida. But in 2016, Trump was running pretty close to Hillary Clinton in late polling, within a “normal polling error” of around 4% (the kind of accuracy we’ve normally seen in presidential-election polling in recent decades); in 2020, he would have needed a polling error almost twice normal to win the national popular vote.

Further, the polling predictions make sense in context. Donald Trump never had a net-positive approval rating after his first couple of weeks in office; he barely won election in 2016 against a terribly weak candidate who ran a poor campaign; he notoriously played to his base rather than trying to expand his appeal to the broader electorate; his conduct in office alienated a lot of suburban voters who had been hoping for something a bit more traditionally “presidential”; and he was running for reelection in the middle of a pandemic and the economic and social disruption it brought—not exactly incumbent-friendly timing.

Nor has recent history been particularly kind to Republicans in presidential elections. Since George H. W. Bush beat Michael Dukakis (a weak candidate who ran a poor campaign) in 1988, Democratic candidates have won the popular vote in every election except for George W. Bush’s reelection in 2004. While the Republican candidate won in 2000 and 2016, both elections were “squeakers” where very small margins in a few states (it came down to just Florida and a few hundred votes in 2000) overcame a national popular-vote defeat. The Republicans have a structural advantage to the extent that low-population Republican-leaning states control a disproportionate share of the Electoral College vote; but on the other hand demographic trends have been slowly making it harder for Republicans to win the national vote. Republican presidential victories are still very possible—but they’re not to be taken for granted, and a Republican “landslide” nowadays is rather unlikely.

So to the extent that we believe that polling organizations and aggregators (and history) are making a good-faith attempt to be accurate, we can say that the simple explanation for Joe Biden’s 2020 victory was that he actually won: what happened was very close to the consensus polling prediction, and was consistent with recent election history.[2] If we don’t believe that polling organizations and aggregators (and history!) are reasonably honest and unbiased, that’s a different issue—and a different blog post.

Of course, had Joe Biden mysteriously won North Dakota or Kentucky or Oklahoma, I would be quite happy to search for evidence to account for such a strange result—anything that anomalous would legitimately lead us to question our simple explanations. But what is strange about any of the actual results, on a national or on a state level, in the 2020 presidential election?

Conclusion: Occam voted Democratic this time

If a Biden victory was consistent with what was known about voter preferences before the 2020 election, both on a national and on a state level, there is no anomalous result that needs any kind of voter fraud or other malfeasance to explain it—we simply don’t need to jump through that hoop.

Anyone—from Donald Trump on down to ordinary individuals—who claims that there was a “stolen Trump landslide” (sacred or otherwise) needs to show some actual evidence that such a “landslide”, or indeed any Trump victory, was in fact reasonable to expect. I have yet to see a single voter or politician show such evidence. I have asked many times: What state was Trump supposed to win that he didn’t win? To date, nobody has given me an answer. All the anecdotal information in the world about “spikes”, “dumps”, or other supposedly nefarious goings-on are unconvincing if they don’t have some kind of anomalous result to explain.

As we saw last week, talking about “stolen” elections—when the election actually played out very much the way the polls predicted—is actively harmful speech; it has already led to violence and death, and is likely to lead to more in the coming weeks, and perhaps months or even years. We need to turn our temperatures down, and let Occam reassure us that the world is sometimes a simpler place than we think it is.

—

[1] My “go to” poll aggregator is FiveThirtyEight.com—which, among other things, was the only major aggregator that gave Donald Trump a legitimate chance of winning in 2016.

[2] One pollster, the Trafalgar Group, did predict a Trump victory—but while they correctly predicted the 2016 final result, they are not taken seriously by the rest of the industry, largely because they refuse to reveal their methodology. Social science, like natural science, is supposed to be based on openness in methodology to produce reproducible results; and by that standard Trafalgar has yet to establish itself as a “scientific poll”.