Can science predict the future? The great science fiction author Isaac Asimov wrote a series of books in which a scientist figures out a way to do just that.
“It was the science of human behavior reduced to mathematical equations,” Asimov wrote in one of the books in his Foundation series. “The individual human being is unpredictable, but the reactions of human mobs … could be treated statistically.”
It’s a great concept for a book series — and a well-reviewed streaming TV show — but in the real world, it falls flat.
It would be comforting to hear from an expert who can read the tea leaves and tell us who will win.
The subject has come up again as we are now just a month out from a hotly contested election that could determine everything from the future of abortion rights to the strength of American democratic institutions. It would be comforting to hear from an expert who can read the tea leaves and tell us who will win.
Demand is so high that multiple models are competing for dominance, and two of the biggest forecasters — historian Allan Lichtman and close poll reader Nate Silver — got into an online feud over the weekend over how to read the “13 keys” that Lichtman uses to make his predictions. (Lichtman says they predict a Kamala Harris win, but Silver argues they actually favor Donald Trump.)
“Allan let’s just say the little tricks you’ve played with the Keys in the past will come back to haunt you! The Keys shall be respected,” posted Silver, in what sounded like something the villain in a children’s movie shouts shortly before an army of elves overtakes the castle.
If you’re not sure about the 13 keys, you could look instead at the Misery Index, a crude measure developed by an adviser to President Lyndon Johnson that combines the unemployment rate and the inflation rate, which will tell you the numbers are looking good for Harris. Or you can check out the more sophisticated Time for Change model, which zeroes in on the popularity of the current president, the state of the economy and the number of terms the incumbent party has been in the White House. That also predicts a narrow Harris win.
Or maybe, like Silver, you favor polls. In that case, you can look at 538’s proprietary mix of polls, economic and demographic data to see that Harris is projected to win 57 out of 100 times (Silver is no longer with 538). If you’re willing to pay, you can see Silver’s own forecast, which is based on his new model. Or you can just look at the state-by-state polls yourself and obsessively play with the maps on 270 to Win while muttering about crucial Waukesha County.
Or perhaps you like the newfangled prediction markets, which ask bettors to put their money where their mouth is. PredictIt has a bet on a Harris win at 56 cents, with Trump as the underdog at 47 cents, while VirtualTout blends PredictIt data with some fancy math to come up with a forecast for a “strong Democratic victory.”
The good news for these forecasters is that if Harris wins, they’ll all get to claim that their models were accurate. And if Trump wins, they can say that, after all, their models showed only a high probability of a Harris win, or they’ll switch to talking about the popular vote.
The problem with all of these models is that they attempt to make iron laws out of the scrap metal of past elections.
The problem with all of these models, however, is that they attempt to make iron laws out of the scrap metal of past elections. We’ve had only 59 presidential elections since the Constitution went into effect — even fewer if you consider only those since the expansion of voting rights, the advent of TV news and the development of sophisticated voter targeting in the 20th century. Statistically speaking, that’s just not a lot of data.
More recently, you can throw in the complications the Electoral College has caused since 2000 by diverging twice from the national popular vote, the increasingly intense partisanship and the closely divided elections, and it’s basically impossible to make definitive predictions. Would your favored political model have predicted an Al Gore or George W. Bush win in 2000? Did it predict Trump or Joe Biden winning by tens of thousands of votes in a handful of swing states?
If your model assumes how voters will respond to a president running for re-election, how does it factor in the president’s dropping out and endorsing his vice president just months before the election? Or a former president’s running for a nonconsecutive second term? What about a candidate who was impeached for trying to overturn the results of the last election?
Most models can be useful to a point. They can give us the general sense of the political landscape or enough information to start make contingency plans for after the election. If Harris wins, maybe the models will make at least some voters less open to bogus arguments that the election was stolen. (I’m not holding my breath on that one.) And some can tell us how well a generic Democratic or Republican candidate might have done in a particular environment, which can help the parties review their strategies.
But ultimately, we aren’t voting for a generic candidate. We have to choose between two specific people, with their own strengths and weaknesses. And sometimes the polls, the economic models and the betting markets can’t measure that.
In Asimov’s series, the scientist’s predictions of the future are accurate for centuries until, through random chance, a man is born with a mutation that allows him to psychically manipulate other people’s emotions, creating fear and devotion in order to rise to power. That proposition — that predictions of the future ultimately cannot account for the unknown or the unexpected — is worth keeping in mind as we head into the November election, especially given that we have a candidate on the ballot who’s adept at creating fear and devotion.
In the end, election forecasts are not bad at telling us how the future will look, as long as it continues to look like the past.