All the President’s Myths

The White House from Washington, DC, Public domain, via Wikimedia Commons

As I was working on this post, the Supreme Court smacked down the absurd Texas lawsuit. But I still think it’s worth posting just to go through the farce that was perpetrated on the courts, conservatives and ultimately the American people. On Thursday, Em eviscerated the legal arguments in Texas’ quixotic and now deflated lawsuit against democracy, I thought I’d take a moment — OK, quite a few moments — to eviscerate the statistical arguments that littered the legal briefs like fleas on a dog. Numbers can lie with a straight face in a way that people can’t. We believe numbers. They convey an authority that seems objective. But numbers can also reflect the biases and deceptions of those who produce them. And the biases and deceptions being used the Stop The Steal crowd are profound, deep and destructive.

Let’s get started.

Bellwethers

One of the most insidious statistical sins in our modern world is the concept of p-hacking. P-hacking is a practice where you pore over some data, looking for any correlation to jump out just by sheer random chance. The bigger the data set, the more likely it is that you will find something … anything. You can then proclaim — spuriously — that your correlation is extremely unlikely to happen and therefore must be real. But it’s not. It’s simply looking through thousands of Rorschach blots until you find the one that looks like a dragon. It’s rolling ten dice enough times that, eventually, they all come up as sixes.

A perfect illustration is the chocolate weight loss hoax in which a scientist did this deliberately: feeding one group chocolate and then measuring everything he could until he found some way in which the chocolate-eating group was better than the control group. If you look at 20 different metrics, the odds are really good that you’ll find a 1-in-20 thing happening just by random luck. And the press went all-in on his supposed discovery until he told them what he’d done.

There is a cottage industry in proclaiming that elections are easy to predict. Oh, I don’t mean people analyzing the polls. I mean people who run models that try to predict elections based on the economy, crime rates, prior election results and the astrological signs of the candidates. You’ll occasionally hear that Professor X at University Y has a model that correctly “predicts” 20 of the last 21 elections. I put “predicts” in quotes because you cannot, technically speaking, predict the past.

Now, those models are not useless. Nor are they pure p-hacking. They can reveal the kind of things that sway voters and give us expectations of what might happen in any given year. They can be useful as a benchmark to judge candidate performance: Donald Trump in 2016, for example, massively underperformed in these models. But their predictive power is extremely limited. Every year, some of those predictors crash and burn when they try their hand at predicting an election that hasn’t happened yet. Because ultimately, many of them dabble in at least a bit of p-hacking, combing through correlation after correlation until they find those that work, at least for the elections we’ve already had. So they can miss something important that didn’t correlate because of random chance or jump on something meaningless that appeared to be important just by random chance.

One of my favorites in this genre was a tongue-in-cheek one called the Redskins Rule. According to this rule, if the Redskins won the last home game before the election, the incumbent party would win. If they lost, the challenging party would win. Those of you of a mathematical note will have already seen the qualifying condition that it has to be a home game. With 32 NFL teams and two different venues, you can easily find something that correlates with election results if you look hard enough. And, indeed the Redskins Rule correctly “predicted” every election from 1936 to 2000.

But the Redskins rule was just p-hacking. If it reflected anything, it reflected that there was a brief period of time when our politics was stable, and the Redskins didn’t stink on ice. But it failed in 2004. Then it failed again in 2012. Then again in 2016. Then again in 2020. ¹

A slightly less-dumb version of the Redskins rule has been repeatedly touted by the President and appeared in the brief he filed in support of the Texas lawsuit. The argument goeth thusly: no Presidential candidate has won both Ohio and Florida and subsequently lost the election. Therefore … something. I would actually think this was more of a self-own than anything else, because the President is basically admitting he blew a winnable election. But let’s take this at face value.

First of all, it’s not true. Nixon won Ohio and Florida in 1960 and lost the election.² But even taking Ohio-Florida as a bellwether, this doesn’t mean as much as he thinks it does. Our recent elections have been close. But this was not always the case. Landslide elections used to be a more lot more common and thus winning Ohio and Florida simply reflected that the candidate won most of the states.

Since Florida became a state in 1845, the elections have gone thusly:

• In 16 elections, Florida and Ohio split (1848, 1856, 1860, 1864, every election from 1880 to 1908, 1920, 1924, 1944, 1992).
• In 16 elections, Florida and Ohio went together in a landslide, defined as winning more than 70% of the Electoral Votes (1852, 1868, 1872, 1912, 1928, 1932, 1936, 1940, 1952, 1956, 1964, 1972, 1980, 1984, 1988, 1996).
• In 10 elections that were close, Florida and Ohio went to the winner (1876, 1916, 1948, 1968, 1976, every election between 2000 and 2016).
• In 2 elections that were close, Florida and Ohio went to the loser (1960, 2020).

So, Trump’s argument here is that of the 11 prior close elections in which Florida and Ohio did not split, the man who took both won ten times. This is … an interesting bit of historical trivia. But it’s basically an argument that, in a close election, winning big states is a good idea.

We already knew that.

The Ohio-Florida bellwether is basically p-hacking with a very slim historical backing. We have 50 states. We have over a thousand combinations of two states. So just by luck, you can find some state or some combination of states that always appears to “predict” the Presidential winner. From 1904 to 2004, Missouri went with all but one Presidential winner. But that wasn’t because of Missouri Mojo; it was because Missouri was a contested state with lots of electoral votes. And it just happened to almost always align with the electoral winner. Other states have enjoyed their time as bellwethers: Pennsylvania missed only one election between 1800 and 1880; Illinois went with all but three winners between 1852 and 1996; Nevada has only missed three times since 1904; New Mexico only three times since 1912. But these weren’t really meaningful. Over any period of time, one or two states will always match the electoral winner just by random chance. And that chance is better in a state with lots of electoral votes or one that isn’t entirely in one party’s pocket.

Florida and Ohio have been a reliable bellwether since 1928. The Presidential winner has won Florida all but three times and Ohio all but three times. And only twice has he failed to win both. But this is not magic. This is not some law of nature. This is simply the way politics has played out. Bellwethers rise and bellwethers fall. And it is equally valid for Biden to point out that since New Mexico became a state, only one President (Carter) has won the election after losing both it and Nevada.

The claim that winning Florida and Ohio guarantees the Presidency is only slightly more valid than my claim that the Falcons will win if I don’t watch the first quarter. It’s an mostly appeal to superstition. It’s somewhat interesting if you are fascinated by the ebb and flow of American politics. But it’s not predictive. It’s not meaningful in the sense that, if the bellwether fails, that indicates something “suspicious” about an election.

We’ve been in a long era where Ohio and Florida were close enough to fight for and big enough prizes that winning them gave one a huge advantage in the electoral college. That’s of historical interest, I suppose. You’d much rather win both states with their delicious prize of 47 electoral votes than lose them. But past performance is no guarantee of future results. And it’s possible that their status as a bellwether is coming to an end.

Just as Missouri’s did.

You’re Not Supposed to Eat Wax Apples

“Maths,” photo by Sean MacEntee

One of the more common tactics of the pro-Trump side, one that found its way into numerous supporting briefs, is making bogus comparisons to claim chicanery. For example, Rudy has claimed many times that Pennsylvania counted 700,000 more absentee ballots than it sent out. But that claim is based on a comparison of the number of ballots mailed out in the primary to those counted in the general. These are not the same things.

Another claim is that far fewer absentee ballots were rejected in 2020 (a few tenths of a percent) than were rejected in 2018 or 2016 (a few percent). But this is also garbage. The few tenths of a percent figure represent those rejected because of signature mismatches while the few percent represents those rejected for any reason, including being mailed in late. It is as though you were comparing the number of interceptions thrown by one quarterback to the number of incompletions thrown by another. You’re not measuring the same things. When you do compare signature rejections to signature rejections and overall rejections to overall rejections, the numbers are very similar.

In Which Carl Sagan Says, “Whoa, Dude!”

Perhaps the most outlandish claim in the Texas brief came to us courtesy of Charles J. Cicchetti, Ph.D. In an analysis cited prominently, he claimed that the chance of Biden winning the election in Georgia, Wisconsin, Michigan and Pennsylvania based on the results Tuesday evening were 1 in a quadrillion. And Biden’s odds of winning those four states based on the 2016 election are 1 in a hundred quadrillion. His argument was presented in the Appendix here.

It’s hard to explain just how absurd this argument was. Basically, he took the 2016 election and calculated the likelihood that the same voting patterns would produce a different result. The method he uses is not unusual and can be used to make statistical comparisons under certain circumstances. For example, this semester, many universities gave exams in a hybrid mode — in-person for some students and online for others. One could compare the scores of those two groups of students to see if they were drawn from the same population of students. This might be a useful way to determine if online students were cheating on their exams. For example, if they scored better overall or fewer students got bad scores, that would show up in this kind of analysis. But it only works because the null hypothesis — that is, the assumption that you are testing — is that any random group of students drawn from the same class should score about the same on the same exam. It wouldn’t work if you were comparing, say, student from two completely different classes taking two different exams.

Another example where it might be useful: in 1996, Barry Bonds hit .308; in 1997, he hit .294. A similar but different method could try to determine if he was basically the same hitter but just got a little unlucky in 1997. But you wouldn’t use this method to compare Barry Bonds in 1997 to Mark McGwire in 1998 because they are, demonstrably, not the same person.

Comparing the 2016 election to the 2020 election in this way is absolutely completely useless because, and I don’t want to get too technical here, 2016 and 2020 were two different elections. We expect them to be different. The null hypothesis is not that the 2020 voters would cast their ballots in exactly the same way as the 2016 voters did; it’s that they would not. In fact, getting the exact same result in two different elections would be more suspicious, not less.

What’s more, this analysis implied that any change in an election is statistically impossible. If you argue that Biden winning Pennsylvania is suspicious because Trump won it in 2016, the same technique indicates that Trump shouldn’t have won it in 2016 because Obama won it in 2012. In fact, Trump’s win is more suspicious because Pennsylvania shifted by six points in 2016 but only one point in 2020.

Actually, no, it’s even dumber. Because the same analysis would indicate that every single state was stolen. In 2016, Trump won Florida by a little over a point; in 2020, he won it by 4 points. Using the Cicchetti method, that is a one a quadrillion chance. So, clearly, the Republicans stole Florida.

The analysis of early returns compared to final results was equally flawed. In fact, as Pennsylvania AG Josh Shapiro pointed out in his rebuttal, Texas’ brief rebutted its own argument. The analysis assumed that the people whose votes were counted on election day and those counted in the ensuing days were the same group of voters. But they weren’t. The Texas brief admitted this. It complained about it repeatedly. The absentee vote was overwhelmingly Democrat.

As I said, this kind of analysis can be useful. If you compared Barry Bonds’ 1999 to his 2001, you would find that he was a very different hitter. This is one of the arguments people use to claim that he took steroids. But that analysis works because you are comparing Barry Bonds to … Barry Bonds. The analysis in the Texas brief was like concluding that Barry Bonds cheated by comparing his career to that of Biff Pocoroba.³

And, In the End, The Rage You Feel, Is Equal To, The Rage You Make

Ultimately, just as Em argued that the legal arguments were garbage, the statical arguments were garbage too. This was not a document filed by a group of people trying to actually contest an election. It was a primal scream from a defeated President combined with a pledge of loyalty from the more spineless members of his party combined with a desperate hope that enough SCOTUS judges would be bedazzled enough or partisan enough or both to overturn an election or at least have a close enough vote that the Trumpists claim the Supreme Court stole the election. It’s no surprise that it was bounced as fast as it was.

As such, the arguments presented within were not arguments designed to prove anything because they couldn’t. Multiple recounts and innumerable court cases have shown that there is no there there. When people make bad arguments it’s often because good arguments are not available. No, these were arguments designed to muddy the waters. They were shit thrown at the wall in the hope that something would stick. Anyone who has dealt with conspiracy theories — and the idea that Biden stole the election is a conspiracy theory, nothing more — recognizes the tactic: make a blizzard of wild claims so that you always have one more to fall back on. I spent years going in circles on this with 9/11 conspiracy theories.

“What about the Pentagon?”
“They found pieces of the plane around the Pentagon.”
“Well, steel doesn’t melt at that temperature.”
“No, but it gets really really soft.”
“Well, there were debris and bodies in Indian Lake!”
“No bodies. And the debris could easily have been blown into a place only a mile away.”
“But what about the Pentagon?”
“…”

Conspiracy theories are a classic example of what Bill James called a “bullshit dump”; a way of reconciling what we know to be true with what we want to be true. Anyone who watched the Trump movement for the last four year noticed the absolute confidence his supporters had that not only was he right but that he would easily win re-election in a landslide. He lost. And so, conspiracy theories have sprung up so that people can reconcile what they know to be true (Trump lost) with what they wish were true (Trump won re-election in a landslide; eat dirt, libs!).

Conspiracy theories are dangerous and frustrating. But they are also profitable for those who peddle them. Seen in that light, these whack-a-mole games are not designed to actually reveal a conspiracy; they’re designed to sustain the theory. They are designed to keep the books selling, the speaker fees coming and the podcasts … uh … podding?

And in that sense … the Texas brief was a success. It didn’t change the election. I don’t think anyone involved — except perhaps the deluded naked emperor himself — thought it would. But it helped raise millions of dollars to pay off Donald Trump’s campaign debt and give him a massive political slush fund. It riled up his base so that they will regard any acts by the incoming President as invalid.⁴ And it will be used to push states like Georgia into making it harder to vote and almost impossible to vote absentee.

Because ultimately, as glib as it is to say so, the real complaint of the Trumpists was not that the media stole the election. Or that Biden stole the election. Or that big city Democrats stole the election⁵. Or that Dominion did. Or that the Courts did.

Their complaint was and is that the voters stole the election. All 80 million of those who punched their card for Biden stole an election that Trump was entitled to win because Democrats are just so awful and he’s so awesome.

And I expect that the ultimate goal here will be to make it so that those damned voters can never steal another election again.

1. This has created the “revised” rule where, if the popular vote winner doesn’t win the Electoral College, the rule is flipped in the next election. The weaselly stupidity of this I leave an exercise to the reader.
2. The President has since updated this argument to say that no President has won Florida, Ohio and Iowa and still lost but that’s also not true because 1960 still happened.
3. No insult intended to the former Brave, one of my favorite players as a kid, who passed away earlier this year at the too young age of 66.
4. Some hotheads … OK, Allen West … are talking about secession in the wake of the SCOTUS smack-down.
5. I mean … have you met Democrats?