Please don’t fool yourselves into believing that the secret to Trump’s success lay in some grand, secret data innovation.
In the aftermath of any presidential campaign, there is a natural inclination to rewrite history to make sense of the outcome. That’s a bad habit. It leads us to impart meaning to the wrong things and take away the wrong lessons. We ought to avoid it.
So let me take issue with Josh Green and Sasha Issenberg’s latest article on the Trump data operation. Green and Issenberg have revisited the Trump data team—Cambridge Analytica in particular—and discovered that they “picked up disturbances others weren’t seeing.” They forecasted a different electorate, and their forecast turned out to be right. Ergo, they must have known something that the rest of us didn’t.
It would be cold comfort if this narrative was true. It would mean that the Republicans had made some secret leap forward in data science. It would mean Democrats could study those efforts, learn from them, and do better next time.
But I remain pretty damn skeptical, and you should to.
Here’s what Trump’s data team actually said to Green and Issenberg:
Inside his campaign, Trump’s analysts became convinced that even their own models didn’t sufficiently account for the strength of these voters. “In the last week before the election, we undertook a big exercise to reweight all of our polling, because we thought that who [pollsters] were sampling from was the wrong idea of who the electorate was going to turn out to be this cycle,” says Matt Oczkowski, the head of product at London firm Cambridge Analytica and team leader on Trump’s campaign. “If he was going to win this election, it was going to be because of a Brexit-style mentality and a different demographic trend than other people were seeing.(Emphasis added.)
“Trump’s team chose to focus on this electorate, partly because it was the only possible path for them,” Green and Issenberg concluded.
“If he was going to win.”
That’s not forecasting. It’s wishcasting. Cambridge Analytica fiddled with their model until they came up with an answer that would appeal to the client. They had one positive data point—rural whites casting record numbers of early votes—and then they extrapolated until they arrived at a conclusion that they could sell.
Wishcasting isn’t good data science. Wishcasting isn’t good any type of science. And if Democrats decide the lesson from this election is that they need to study and replicate Trump’s data innovations, it’s going to lead them down a winding, dead-end path.
There are a lot of reasons why the Hillary Clinton campaign lost this week. Anyone who thinks they know the answer is kidding you and (probably) kidding themselves. It’s going to take a long time for us to collective sort through what just happened.
But in the meantime, I think it’s important for us not to reconstruct recent history to arrive at comfortable, fruitless conclusions. And even if we can’t explain why this did happen, we can at least engage in process-of-elimination and reject potential explanations that we know to be false.
Donald Trump didn’t have a hidden advantage. His team wasn’t a bunch of savants who saw the truth of things while everyone else was confused. Please don’t fool yourselves into believing that the secret to Trump’s success lay in some grand, secret data innovation.