Epsilon Theory PDF Download (paid subscribers only): The Grifters, Chapter 3 – Election Prediction
That’s Nate Silver, founder and face of election “modeler” FiveThirtyEight, performing his traditional “Awkshually, we weren’t wrong” dance after mangling yet another national election.
Haha. No, that’s a falsehood, as the fact checkers would say. That claim was made with no evidence, as an ABC News reporter would say.
In truth, this is a picture of Nate Silver speaking at the “ABC Leadership Breakfast” during Advertising Week XII. Of course Advertising Week uses the same numbering system as the SuperBowl ™. That would be 2015 in normie text, about a year prior to FiveThirtyEight’s mangling of the prior national election.
You will only see Nate Silver on ABC News and other ABC media properties and events, because FiveThirtyEight is a wholly-owned subsidiary of ABC News.
ABC News, of course, is a wholly-owned subsidiary of The Walt Disney Corporation.
Hold that thought.
That’s Fivey Fox, the FiveThirtyEight cartoon mascot, who is happy to guide you through the genius-level mathematics and super-science that “powers” FiveThirtyEight’s election models. You may have also seen Fivey Fox on ABC News programming, as part of a weekly animated cartoon segment broadcast over the past nine months to “inform” viewers about “how the election actually works”.
For all you FiveThirtyEight and ABC News viewers, I’d guess that most of you find Fivey Fox and the cartoon infographics pretty cringey. I’d guess that most of you believe, however, that these animated cartoons are not aimed at you, but at “low-information” viewers who are not easily capable of understanding how the election actually works, and certainly not capable of understanding the genius-level mathematics and super-science behind FiveThirtyEight’s election models. I’d guess that most of you believe that yes, Fivey Fox is a little silly, but it’s necessary to speak in cartoon language in order to communicate with all those Fox-watching and Trump-voting dullards out there.
Ask not for whom the cartoon tolls. It tolls for thee.
Fivey Fox and his cartoon friends on ABC News do not exist to “educate” the great unwashed, any more than ESPN programming exists for people who don’t watch sports. Fivey Fox exists to engage YOU, the politically-aware ABC News/FiveThirtyEight viewer.
So does “Nate Silver”.
I put his name in quotation marks because of course a real life Nate Silver exists. But the “Nate Silver” that you see at the ABC Leadership Breakfast or that you hear PhD-splaining every four years that “modeling isn’t polling” is just as much a cartoon – just as much a constructed abstraction of an abstraction in service to narrative ends – as Fivey Fox.
The disheveled look, the stark black eyeglass frames … “Nate Silver” looks exactly the way it needs to look to optimize your engagement with it. Not to like “Nate Silver”. Not to dislike “Nate Silver”. To engage with “Nate Silver”.
For the ABC News/FiveThirtyEight viewers who like the election prediction made by “Nate Silver” and Fivey Fox, this will be a mirror engagement – yes! this Genius Expert ™ agrees with me! Science and Mathematics agree with me! And it’s so obvious that even a child could understand! Ah, sweet dopamine!
For everyone on the other side of the election prediction made by “Nate Silver” and Fivey Fox, this will be a rage engagement – no! this Idiot Egghead ™ has lost all credibility! The polls are clearly not capturing Factor XYZ, and it is enraging to be told otherwise as if I were a child! Ah, sweet norepinephrine!
There’s nothing accidental about any of this.
Three mega-corporations in the world today truly understand the primacy of engagement: Google, Apple and Disney. Other mega-corporations have successfully adopted this principle over time, but Google, Apple and Disney built their empires on the primacy of engagement, on how their products or services make you feel. It’s the foundation of Google’s internet search algorithms. It’s the foundation of Apple’s product design. It’s the foundation of Disney’s media content.
Of the three, the Covid pandemic has hit Disney the hardest. Parks are shut down. Movies aren’t being made. As for television, sports programming is getting killed and overall ad spend is down. The only potential bright spot is that this is an election year, where $11 billion will be spent on political ads, and where maintaining engagement with its news programming has never been more important for Disney.
How do you get more engagement with your news programming? How do you trigger more neurotransmitter brain chemicals in your ABC News audience?
By creating “news” that can be transformed into an entertaining/enraging game.
By transforming a singular Election Day event into a months-long spectator sport, complete with plays and scores and announcers and cheering/anxious fans.
That’s what election modeling does. That’s why public polling and election modeling exist. Polls to create the “news”, election models to create the score, Fivey Fox and “Nate Silver” to announce the game. All to create engagement with a diversified media corporation.
That’s why Disney acquired FiveThirtyEight. That’s why they originally had it within ESPN and then transferred it to ABC News. That’s why they created the cartoon characters of Fivey Fox and “Nate Silver”.
No one understands how to create and sell a spectator sport better than Disney.
Here’s the kicker. This spectator sport that Disney/ABC News/FiveThirtyEight has created around Election Day has very little connection with the election itself. The “scores” and the “announcing” and the game itself are a totally distinct thing from the process and dynamic and the outcome of our most important political institution.
And they know it. And yet they sell their game over and over again as if it were the real thing.
That’s what makes it a grift.
In a nutshell, the FiveThirtyEight prediction model is designed around thousands of simulations of statewide results (based on statewide polls and a hypothesized probability distribution on state level results) that are then mapped against the Electoral College. These thousands of simulations of possible statewide results create a probabilistic distribution on the Electoral College outcome, and whatever percentage of outcomes are on the good side of 269 Electoral College votes for a candidate is the answer for the point-in-time odds of that candidate winning.
FiveThirtyEight went into Election Day 2020 assigning Joe Biden a 90% chance of winning, which was even more divorced from election reality than their 2016 “prediction” that Hillary Clinton had a 72% chance of winning. There is zero alpha … zero useful information … in a model that predicts an election outcome with near certainty when in truth that outcome hinges on a few tens of thousands of votes out of 150 million votes cast.
To use a spectator sports analogy, FiveThirtyEight set the 2020 betting odds for this “football game” with Joe Biden as a massive favorite, say 20 points. He won by 1 point. In 2016, FiveThirtyEight had Hillary Clinton as a somewhat less massive favorite, say 15 points. She lost by 1 point. There’s nothing “robust” about these predictions, as “Nate Silver” is currently claiming. These predictions are disasters. FiveThirtyEight would be laughed out of Vegas for setting odds like this.
The FiveThirtyEight model failed in both 2016 and 2020 – and will fail again in 2024 – for the same two reasons.
First, the prediction model failure in 2016 and 2020 is NOT just a garbage-in-garbage-out problem with the polls that serve as model inputs, as the current “F#ck you, we did a good job“ non-apology tour of “Nate Silver’ would have it.
In fact, the Disney/ABC/FiveThirtyEight business model is in large part responsible for creating the bad polls.
Both polling and responding to polls have become political acts. There is a panopticon effect here, where both pollsters and the polled know that their behavior is being observed. Not in the sense of an enemies list or being personally identified, but observed nonetheless by a massive hidden audience watching the very public playing field of the election spectator sport. And in true panopticon fashion, the polled begin to see themselves as members of a team competing in this election spectator sport, as active political participants through their poll response.
This has an enormous – and predictable – impact on poll response behavior. It’s not that members of the Out group (in this case Trump voters) are “shy”, it’s that both In group and Out group members see themselves as players in a game. Because they are! And when you see yourself as a player in a game, you … play the game. You act strategically. You agree or refuse to participate in a poll for strategic reasons. You answer the questions one way or another for strategic reasons. It’s not that you’re lying in your answers, although of course some people do, it’s that you’re considering both your poll answers and your poll participation within the larger context of this election spectator sport that you know your answers will be used to support.
Everyone knows that everyone knows this is how polls are used today, that you are part of a larger political game that is distinct from the actual act of voting. This is the common knowledge of polling today, and as a result, no one provides “straight”, i.e. non-strategic, poll responses today. No one.
And Nate Silver knows it.
Hell, he and his Disney bosses created this game that uses polls as the “play” that happens on the field! They know exactly how the meaning of polling has changed, how polls are no longer an independent signifier of voter intentions, but are the output of strategic gameplay that is only tangentially connected with Election Day.
FiveThirtyEight depends on bad polling data as the play-by-play action in their election spectator sport.
Bad polls are necessary for this lucrative grift to continue.
So they will.
Second, the FiveThirtyEight prediction model itself is a category error, created and designed to promote a spectator sport business model with hundreds of point-in-time odds (the “score” of the game) over the months-long course of this made-up game, NOT to predict the outcome of a real-life, singular rare event.
To use an online poker analogy, the model is designed as an “engine” for a game where you can play poker all the time, as often as you like. It’s designed for you to engage with Fivey Fox and “Nate Silver” every day if you can stomach it, checking in constantly to see if the “odds” have changed with some new state poll and a rerunning of the simulations. That’s not a bad thing. The math of this game isn’t wrong.
Or as fellow cartoon Jessica Rabbit would say, “I’m not bad. I’m just drawn that way.”
But the way the prediction model was drawn … the way it generates a new probabilistic “score” of this constructed game every time a new state poll comes out … is NOT representative of the experienced odds of the single election event. At all. The math IS wrong when it comes to understanding the odds of who is actually going to win the actual election.
Why? Because Silver’s run-ten-thousand-simulations methodology masks the volatility and the uncertainty hiding in the statewide polls. I’m not talking about the uncertainty of a poll with a big margin of error. The methodology can handle that fine. I’m not talking about the uncertainty of a poll that says a statewide race is 50/50. That’s not a problem, either.
I’m talking about the uncertainty of a poll that doesn’t MEAN what you thought it means, where – in the lingo – error in the poll is not randomly distributed, where – for example – you have a pandemic changing actual voting behavior in a systematic way, but not changing poll response behavior in a systematic way, where – as Nate Silver understands perfectly well – you have poll respondents acting strategically in a systematic way.
If you have THAT kind of uncertainty in your statewide polls, then the FiveThirtyEight prediction model will not catch these errors. No, no … the simulation methodology will MAGNIFY these errors.
The result? FiveThirtyEight has no idea what the real score of the actual election game might be.
All of econometric and statistical analysis – ALL OF IT – exists to give you an answer to two and only two questions:
- What’s your best guess?
- How sure are you?
The FiveThirtyEight election model gives you an answer to the first question. That’s what a model – ANY model – does.
The fatal flaw with the FiveThirtyEight model is that they have no answer to the second question. Or rather, there is an answer – NOT VERY SURE AT ALL – but the Disney/ABC News/FiveThirtyEight business model does not allow them to report that answer. Because if they gave this truthful answer, then we would all ask a third question: Why the hell are we playing this game?
And that’s the question that blows up the entire grift.
A spectator sport must have a score at all times. That’s what it means to have a spectator sport. It’s perfectly fine if you say that the score is tied. In fact, that’s a really good thing for audience engagement. But what you cannot say is that you don’t know what the score is. What you cannot say is this:
“Yeah, I think Biden is ahead, but there’s a lot of uncertainty embedded within my model. Maybe Biden is way ahead and maybe the score is tied, I really couldn’t tell you. But I’m pretty sure that Trump is not way ahead.”
Nate Silver knows that’s the truth of the 2020 election and the FiveThirtyEight prediction model.
“Nate Silver” can never admit it.
Is there a better way to understand the truth of an election? YES.
There’s a toolkit for understanding how to play singular or rare events, where the consequences of being wrong or overconfident or just unlucky are far more impactful than repeated-play events. Jimmie Savage, the smartest statistician you’ve never heard of, called it decision theory. This toolkit is used in military decisions. This toolkit should be used in our Covid decisions. For more, read Once in a Lifetime.
There’s a toolkit for understanding the consequences of a polarized electorate and how that polarization changes both a politician’s behavior and voter response behaviors, including voter response behaviors to pollsters. This is the toolkit of strategic interaction. This is the toolkit of game theory. For more, read Things Fall Apart.
There’s a toolkit – which we are at the forefront of developing – for understanding the structure of narrative and how it impacts social markets. Like investing. Like voting. We call it the Narrative Machine. For more, read Inception.
I think these are the three toolkits required to understand the statistical truth of modern politics. Is there a scalable, billion dollar business model to be created around these toolkits, the way Disney has created a scalable, billion dollar business model around the game-ification of Monte Carlo election simulations? Nah. Not a chance. But that’s the thing about truth, statistical or otherwise. It rarely makes you really rich, but it always gives you a life worth living.
And it never turns you into a cartoon.
Epsilon Theory PDF Download (paid subscribers only): The Grifters, Chapter 3 – Election Prediction