As the angry and uncertain future of a Trump presidency settles over America, Trump supporters are jubilantly proclaiming that they knew all along that their candidate would win, while chastened liberals are wondering how they could possibly have missed what was about to happen.
Much of the shock derives from pre-election predictions that almost universally gave Hillary Clinton high odds of winning. On the morning of Nov 8, for example, the New York Times proclaimed: “Hillary Clinton has an 85% chance to win.” Nate Silver’s fivethirtyeight was somewhat less confident, giving Clinton just a 71.4% chance, but other sites were even more so: David Rothschild’s PredictWise gave Clinton an 89% chance, while Sam Wang at the Princeton Election Consortium put it at 99%.
How could all these smart people have been so wrong? Silver has just published a thoughtful post-mortem of the pre-election predictions, in which he identifies several potential sources of error. First, while the national polls were no less accurate than in previous elections, some critical state-level polls did underestimate Trump’s support, in part because they made incorrect assumptions about likely voters, and in part because pro-Trump voters may have been reluctant to respond to poll requests or to declare their support. Second, the modelers also seem to have made some mistakes, failing to account for correlations between states, or to update their models in response to new information, or to properly account for undecided voters. Finally, both the modelers and traditional journalists may have been biased by their own experience of living within overwhelming democratic-leaning echo chambers.
It’s a fascinating analysis, but in focusing so much on errors it inadvertently perpetuates a common assumption in discussions about “what we got wrong” — namely that all along Trump was going to win, and the only reason we didn’t know it sooner is that we weren’t paying attention to the right information.
One reason for the ubiquity of this assumption is that seems self-evident: after all, Trump won, so surely it must be the case that he was always going to win, whether we knew it or not. Nor is it unique to this particular election: the same after-the-fact sense of inevitability shows up in explanations for almost every important societal event: What caused the 2008 financial crisis? Why did Facebook succeed while MySpace (remember MySpace?) failed? Why is Harry Potter the most successful book series of all time? In all these cases, we argue about the reasons, but there is rarely any argument about the inevitability of the outcome. Whether we like or not, whether we knew it or not, we feel completely confident that it was going to happen, because it did happen!
And yet, this assumption of inevitability is probably wrong. Quite to the contrary, just because something happens doesn’t mean it was inevitable, or even likely.
To illustrate, imagine rolling a ten-sided die. Assuming it’s fair, the probability of rolling a 0 on any given roll is 10%. So, if I bet you $100 that I can roll a 0 then by definition you are nine times as likely to win as I am. Even worse, if we were to play this game enough times, I am mathematically guaranteed of losing money to you — roughly $800 for every ten games we play. Only a fool would volunteer to play a game that amounts essentially to handing money over to one’s opponent. But if we played the game only once, it’s possible that I would win. Even better, if lots of people played this game just once, then in roughly ten percent of cases, Fred would bet Jane $100 that he could roll a zero and then he would roll a zero.
But if you’re Jane, what should you conclude? One possible explanation is that Fred cheated by fixing the die so that his odds were actually much higher than 10%. The other possibility is that Fred simply got lucky. In effect, all the “what we got wrong” explanations are proposing some version of the first explanation: that the odds of a Trump victory were much higher than what almost everybody thought they were. But in real life, just as in games of chance, it is necessarily the case that unlikely things sometimes happen.
One domain in which this claim doesn’t seem so odd is in sports. When the Upshot made its prediction on the morning of Nov 8, for example, they tried to help readers put it in context by comparing Clinton’s odds of winning with those facing an NFL goal-kicker kicking from the 37-yard line: both were about 85%. And, as Football fans know, kickers sometimes miss those kicks. For example, in final minute of the Minnesota Vikings’ 2015 wild-card game against the Seattle Seahawks, Vikings kicker Blair Walsh missed a 27-yard field goal, causing his team to lose 10–9 and knocking them out of the playoffs. The chances of Walsh missing that kick were even lower than those of Hilary losing the election, and everyone (including the Seahawks) was appropriately stunned when he did, in fact, miss it. But watching the replay you wouldn’t conclude that the Seahawks were somehow destined to win and everyone was deluded in thinking otherwise — you’d just think that they caught an extraordinary lucky break.
Nevertheless, as Silver points out, predictions like this one were widely interpreted to mean that Hillary was not merely likely to win, but certain to; hence the blowback against the models when she didn’t. The football analogy, in other words, didn’t seem to help anyone appreciate the underlying uncertainty of the election.
What is it about elections — not to mention other kinds of societal outcomes like financial crises, disruptive companies, and bestselling authors — that makes it so hard to appreciate the importance of randomness in determining outcomes?
No doubt part of the problem is that our intuition for probability works best for events such as die rolls and field goals that are independent and repeated many times under essentially identical conditions. With a ten-sided die, after all, you can roll it a bunch of times and find that, sure enough, it rolls a zero only about one time in ten on average (although interestingly this doesn’t stop people from believing they can influence the outcome when they are betting on it). And football fans have seen enough field goal attempts to understand that sometimes things go awry for no particular reason. In contrast, presidential elections are rare and unique events that are anything but independent of each other, or any number of other events in the world. One can’t run an election multiple times and see how frequently each candidate wins; thus talking about the probability of a candidate winning doesn’t make the same kind of intuitive sense as it does when talking about field goals.
But there’s another reason, which is that deterministic thinking is seductively plausible. Let’s assume for a moment that on the day of the election, everyone who voted knew they were going to vote, and also knew which candidate they would vote for. Now let’s also assume that it’s possible to measure public opinion with near-perfect accuracy and near-zero latency. It’s actually not possible to do this with existing polling methods, but one can still imagine that had such a poll been conducted that morning, one could have predicted with close to 100% confidence that Trump would win. In other words, as of the morning of Nov 8, it could well have been inevitable that Trump was going to win, and the only reason we didn’t know it was because we lacked sufficiently accurate information.
With this piece of hypothetical reasoning in mind, we can now start winding back the clock. If it’s true that Trump’s victory was inevitable as of the morning of Nov 8, then it’s also probably true that it was very close to inevitable the previous evening, by which stage the overwhelming majority of voters had probably already made up their minds. And if that’s true then it was probably mostly true the day before, and the day before that. One can keep playing this game all the way back to Trump’s announcement of his candidacy. If only we’d really understood what was going on in America — the disillusionment of the white working class, the cynicism of the electorate with politics as usual, the failing appeal of neoliberalism, and so on — we would have known what would happen. Journalists, in fact, have wasted no time coming up with explanations of precisely this sort.
The problem with this reasoning is that as we move backwards in time from the morning of Nov 8, the volume of information that we would have needed to know grows exponentially, as does the difficulty of the inferences we would have had to make. Eleven days earlier, for example, when FBI director James Comey decided to publish his infamous letter, many wondered if it would have an impact on the election. But to know for sure what that impact would be, one would have needed a near-perfect model of how the media would respond to the news, as well as how undecided voters would react to the coverage.
As we go back further, the list of potentially relevant events — the Access Hollywood tape, Clinton’s “deplorables” remark, the text of her Goldman speech, the leaked summary of Trump’s Tax return, his fight with the Gold Star family, the leaked DNC emails about Bernie Sanders, and so on — rapidly increases, along with the number of potential ramifications. Add to this that Trump’s victory was ultimately decided by a tiny fraction of voters — a mere 80,000, or 0.05% of the 138 million who voted, spread out over three states — and that 90 million other citizens who could have voted didn’t, and the notion that any outcome was determined in advance, let alone predictable, starts to sound implausible in the extreme.
So what explains the persistent appeal of deterministic thinking? One answer is that people prefer to think that the outcomes they care about are predetermined because they associate certainty with meaning (and randomness with the absence of meaning). Many religions, for example, are fundamentally deterministic in nature: God is omnipotent and everything is part of his plan. Even when that plan seems hopelessly unfair, inscrutable, or even pointless, it is still preferable to assume that there must be some plan, because the alternative — that there is no plan at all — seems even worse. In a similar vein, no matter how shocked and upset Trump’s opponents might be about his victory, they would still rather believe that it was an inevitable outcome of which they were simply unaware than accept that it was a random fluctuation in history.
Such beliefs are perfectly understandable, but that doesn’t mean they’re correct. Nor does it mean that the assumption of inevitability is the only way to imbue outcomes with meaning. Determinism, after all, is a statement about how the world works. Meaning is something that we create in our own minds. There is no logical contradiction between accepting that outcomes are random and continuing to believe that they have meaning. If you get bumped off of your intended flight and up end up sitting next to the person who will later become your spouse, that event will of course be extremely meaningful to you. But its meaning will derive from all the actions that you and your partner took that day and thereafter; from all the time, energy, and love that you invested in one another. The actual meeting itself didn’t have to be part of anyone else’s plan for your lives together to be part of yours. You make the meaning, not the universe.
Accepting that meaning is under your control can also be empowering, at least in part because you can choose not to confer it. If you are the victim of bad luck, for example, you don’t have to make it worse by inferring that the universe is against you, or that all along you were no good and now you’re finally getting what you deserved. Instead you can simply acknowledge the bad outcome, deal with the consequences, and get back to work creating other opportunities when hopefully your luck will be better.
The same is true for this election. As tempting as it is to view the outcome as preordained, it’s equally reasonable to view it as giant, cosmic fluke. If you’re happy that Trump got elected, lucky you. The Seahawks fans who cheered deliriously when Walsh’s kick drifted inexplicably to the left of the posts were happy too — and also lucky. Hopefully you can appreciate your good fortune and extend some graciousness to the losing side. And if you’re like the Minnesota Vikings, staring helplessly as your dream season veers off into a nightmare, don’t give up either. It’s hard to see your team lose, especially when you were so sure they would win, but you don’t have to make it worse by engaging in endless recriminations about all the things it now seems they should have done. As consequential as this election may one day seem, there was nothing inevitable about it, nor is there anything inevitable about the future.
Happy Inauguration Day everyone! It’s time to get to work.
Duncan Watts is a principal researcher at Microsoft, AD White Professor-at-Large at Cornell University, and Author of Everything is Obvious (Once You Know the Answer): How Common Sense Fails Us (Crown Business, 2011)