Voodoo economics

“Erica, I didn’t expect I’d ever catch you in your office again”

Erica Goldsmith is shocked out of her concentration. She looks up to find David de Ridder, co-chair of the NYU Applied Statistics and Probability Group, standing in the doorway, holding an empty coffee mug. Her social skills take a while to re-surface, and for the moment all she can manage is a quizzical expression.

“The starlet of statistics”, David explains, “Oracle of the election. Successfully predicted not just both primaries, but a monthly poll-prediction, consistently within 0.01 RMS. You’re the first statistician ever to hold the nation’s interest for more than half a blogpost.”

“Don’t remind me. This is the first day in two weeks I’ve been allowed to get away from the press.”

David sits down across her desk.

“Without joking. I wanted to say it’s a hell of a thing you did. I hope you realize. Everybody is suddenly on the side of numbers and facts over pundits and gut feelings. It won’t last, but it’s nice to see.”

Erica leans back. She had hoped to focus on her work today. Clearly, coming to the university was a mistake.

“Well the pundits each play their part. They’re all there in the model.”
“How so?”
“They’re all on Twitter, and Facebook. And of course everything they write on CNN, BBC, Al-Jazeera. I mine it all, plug it into the model. Pretty shallow stuff, mostly keywords, n-grams, word2vec. But, it does the trick.”
“So every pundit becomes a factor in your model?”
“Most of them don’t add much. About a handful show significant ability to predict. Nothing like the predictive power of the stock market, or the trending topics, but it adds something.”
“Which ones do best?”
“What you’d expect: Davidson from NBC, Hargreaves from the BBC.”
“I’ve met Hargreaves. He’s a bit of a number cruncher himself, isn’t he?”
“Yeah, most of the ones that score well are basically doing what I’m doing. Except their models are a decade old.”
“Well, you’re one of them now. Better make room for another factor in your model. The world is one pundit richer.”

Erica smiles up at David, who is raising himself up.

“Make my model a factor in its own prediction. What a silly idea.”

“I implemented your suggestion.”
“My what?” David looks up from his newspaper.
“Your suggestion. That I make myself a factor in my model. Or, make the model a factor in itself.”

David struggles to remember what she’s referring to.

“The model! The election is in two weeks, and the press are badgering me for a prediction. So, I updated my model. You’re right, the pundits are important, and I’m one of them now.”
“Yes, I remember …“
“You see, most features, the stocks, the weather, they’re just indicators of a kind of general vibe of the country. But the pundits… they don’t just predict, they sway.”
“They influence the population. Like, who’s that radio guy, the bigot who had the fight with the actor?”
“How should I know… Do you mean Case Madison?”
“Right, when he said Hutchison was going to win, it carried almost no predictive value. He’s an idiot, he has no insight whatsoever. But the model still selected him. Why?”

“A glitch?” David knows it doesn’t matter what he says. His only function at this point is to be Erica’s sounding board.

“No, he sways people. He has a massive audience. And when they hear him predict Hutchison, they see it as an endorsement. Hutchison’s numbers shoot up. With some pundits, what they say could be completely random, but because people listen, it predicts.”
“But surely that doesn’t hold for the serious ones? For Hargreaves?”
“It holds more. It’s a classic bandwagon. People like to vote for a winner, and Hargreaves is a good forecaster. He says Hutchison, people vote Hutchison.”

Erica looks at him. She has a tendency to stop just before the punchline, expecting him to see immediately what she’s just spent a night figuring out.

“So your model has sway too.”
“Boatloads, according to itself. Much more than any other pundit.”
“So… there’s a feedback loop.”
“You could call it that. The model’s outcome depends on its own prediction.”
“You’ve run these experiments?”

Erica nods.

“So? What does it say?” David eyes Erica carefully. He’s beginning to appreciate the magnitude of the situation.

“Well, it’s pretty non-linear. There’s a fixed point for Hutchison with low confidence. If I predict him, but not too confidently, he gets the job. According to the model, anyway. All the others are unstable states. If I say Hutchison with high confidence, it goes to Baker. And if I predict Baker, high or low confidence, she loses. It’s a shame really, I like her.”
“What do you mean?”
“Well, I have to go with the fixed point, right? It’s the only thing that makes sense.”

David takes a deep breath and gathers his thoughts. He takes a sip of coffee, and then another deep breath.

Erica is famously not much aware of politics. It’s a running joke in the group that she would happily tell her models that the building is on fire, without making for the exit. David is a little more aware. He has been following the campaigns with increasing intensity.

“Erica, don’t you realize … what you have here?”

She stares at him.

“What if you told the journalists that your model said Hutchison, high confidence?”
“Well, most likely, I’d be wrong. Baker would get it.”
“Erica. You wouldn’t just be wrong.”

She still doesn’t see.

“Erica, you’d get to pick the next president of the United States.”

Erica frowns at him.