Nate Silver: ‘Prediction is a really important tool, it’s not a game’
Nate Silver made a name for himself with his uncannily accurate predictions of baseball scores and US election results. But some things – from earthquakes to terrorism – even he can’t predict
It takes a brave man to bet against the New York Times’ polling guru Nate Silver. The founder of the FiveThirtyEight blog – named after the number of voters in the US Electoral College, which elects the president – correctly called not just the overall result, but also the result of 49 of America’s 50 states in the 2008 election.
But in the run-up to the 2012 US election, which numerous American pundits insisted was too close to call, Silver was challenged by swathes of America’s right. Why was he saying the election wasn’t even close? Why did he say Obama had a 90% chance to win an election everyone else agreed was on a knife-edge?
Eventually, Silver got so tired of being called out, he offered NBC pundit Joe Scarborough a bet: if Obama won, Scarborough should donate $2,000 to charity. If Romney won, Silver would do the same.
Silver got a rap on the knuckles from the NYT’s public editor – and Scarborough didn’t accept the bet – but in the end, he was vindicated. Having got 49 states right the last time, he swept all 50 in 2012.
The fight was characteristic of Silver, drawing on a combination of belief in the strength of his stats, disdain for the US political media’s establishment, and a fondness for taking a gamble. It’s a franchise he’s extended through his book, The Signal and the Noise, into a look at prediction and punditry itself, across many more fields.
Calling the US election correctly, Silver says from the comfortable lobby of a London hotel, was less tough – and less worthy of adulation – than the US media made out. His triumph looked bigger, essentially, because most US pundits made such awful predictions. “I began FiveThirtyEight partly as a hobby, but partly because I thought news media coverage of election campaigns was pretty terrible in the US,” he says.
“US general elections are very predictable. The whole thing about having a very stable two-party system is part of it – 90% of the electorate has made up its mind, and so 40 out of 50 states are locked in.”
Silver says about half of the rest vote based on the economy – which is easy to fold into a prediction – and so only 5% of US voters need tracking, and polls help out there. But US outlets tend not to get it right because it’s not really in their interests. “The problem with the election is that you have a very long cycle. It lasts 18 months to two years, if you consider the primary. But most of the time nothing of any significance happens. It’s very hard for journalists to write that story, I think.”
He has a little more sympathy for the Romney campaign, who, reports suggest, on the eve of the election genuinely believed they were set for a win. Their mistake, he says, was putting their pollster in front of the public, rather than as an internal “reality check”. When you’re losing at election time, he says, there’s not much you can say.
“If you’re a pollster and your campaign is losing in August, you can say, ‘Well, we’re down, but we’re going to win the day, and it demonstrates why we have this new message,'” he says. “Whereas if you’re down in November … you’re kinda fucked, right?”
What happened to Silver during and after the US election was far less predictable than the result itself: he’s the new king of stats, doing book tours and conferences across the world, which in turn are spawning articles concerned about the “cult of Nate” – the rise of the unchallengable data wizard. Even on the Pyramid of the Sun in Teotihuacan, Mexico, he got recognised, he says. “It’s a bad omen.”
“I would like to live for a couple of days in the alternate reality where Mitt Romney wins, and just see how things plummet,” he jokes. “In some ways, when people are, like, ‘Whatever Nate does is gospel’, it’s a symptom of the kind of disease of not being self-critical and not thinking through things.”
What he’s really trying to do, he says, is show that prediction is hard, and experts are not always to be trusted.
“I don’t want people to be viewing me as this guy who predicts things,” he says, uncomfortable. “Prediction is this really important tool, and it’s essential to science, to see whether subjective reality matches up with the objective world. But it’s not really a carnival show game. People have been asking me to predict Kate Middleton stuff over here, and things like that. It’s not really the idea, right? This stuff’s hard.”
Predicting how long Silver might stay in the political game might also be a difficult task: he’s never done anything for more than four years, and he’s already four years into FiveThirtyEight.
“The basic plan I have is that I’ll probably trudge through this next election cycle, to 2016, then after that I’ll maybe go out and do something very, very different,” he says. “Because I do have a history of getting bored with things after a while and wanting to move on. I worked at KPMG for four years, then for four years did the baseball/poker phase, then four years at FiveThirtyEight, so I don’t know.
“When you’re 35 years old, and you know that probably – not for sure – but you know probably you’re at the peak of your recognisability and your earning potential in the short term, that’s a little strange, I think,” he reflects. “But it’s a good problem to have, relative to the problems that most people have to contend with.”
The King of Stats predicts …
We asked Nate Silver to gauge how predictable different things in life were: from politics to cricket, from terrorism to sexual orientation. Here’s how he scored ten different areas, on a prediction scale of 0 to 10
US presidential elections: 8.5“I’m lucky that I’m American,” Silver says. “It’s probably the easiest system to predict as far as elections go.”
US elections are predictable, he says, for a few reasons. First, there is masses of polling data: there are regular nationwide polls and also, unlike in the UK, polls for each individual state.
“If we’d only had national poll data, we’d still have had Obama ahead, but with much less confidence,” he says. “We’d have had him winning by half a point instead of 2.5 points.”
Congress (or at least its elections) isn’t so easy, getting a score of about 7.
UK general elections: 5.5-6Brits just aren’t as predictable as Americans, Silver says. In his efforts to call the UK’s general election in 2010, he predicted around 100 seats for the Lib Dems; they got 62.
“The model got Cleggmania, I think,” he says. The UK’s three-party (now, with Ukip, potentially four-party) system means there are more moving parts to track, and less data to use.
Still, Silver isn’t totally admitting defeat. As elections are fair, and polling is reasonably accurate, he says a degree of prediction is possible – and hasn’t ruled out trying again in 2015.
Earthquakes: 2Predicting a specific earthquake is a fool’s game: nothing works. But on the broader scale, there’s a lot we can tell.
“We know the long-term rates of earthquakes: every X years you’ll get Y earthquakes in area Z,” says Silver. “[But] actually predicting time-specific earthquakes – there’s no way.”
That didn’t stop Italian authorities prosecuting six scientists and a public official for saying there was no risk of an earthquake in 2009, shortly before one happened. Silver thinks the prosecution was “ridiculous”, but officials could have been more careful: small earthquakes can represent an increase in risk in the short term – “maybe 1 in 100 chance instead of the usual 1 in 50,000″. Still, he says, communicating that to the public is a tough task.
Asteroids: 8.5You’ve got a much better chance of seeing an asteroid coming than an earthquake – even though no one predicted the meteor shower over Russia earlier this year, the really big stuff we should see coming.
“On the solar system scale, there’s a high level of predictability in astronomy,” Silver concludes. “An asteroid large enough to destroy civilisation, you can probably detect in advance.”
What’s trickier to guess, of course, is just because we see an asteroid coming 10 years in advance, whether we’ll be able to do anything to stop it.
Baseball: 8-9Silver made his name building a system to forecast player performance in baseball, called PECOTA. So, unsurprisingly, he thinks the game is pretty predictable.
In fact, he thinks the systems are so good now that about 95% of what’s predictable (rather than random) is already captured. But baseball has its quirks. “It’s not intrinsically all that predictable,” he says. “A soccer team like Man United loses only twice or so a season, whereas even the best team in baseball loses about a third of the time.”
Cricket: 8-9Cricket isn’t so subject to frenetic statistical analysis as its (sort-of) US counterpart, baseball, but there’s no reason it shouldn’t be, Silver reckons.
Yes, pitches vary greatly – but so do baseball stadiums. Bowlers try all sorts of different throws – but so do pitchers. Baseball may have been more subject to slide rules and stat tricks thanks to there being far more big money in that sport, but cricket could be broken down.
“It shouldn’t really be any different at all,” Silver shrugs. “I’m sure there’s some gamblers who’ve tried already.”
World population in 100 years: 7Forecasters from Malthus to Paul Ehrlich have got population predictions drastically wrong – but Silver thinks they can reach a reasonable degree of accuracy.
UN forecasts, he says, are pretty good. Looking five years ahead, the odds of being right are extremely high. Looking further, there are more ways to be wrong. For one, Silver says with a degree of understatement, “to have some mass catastrophe is one downside risk,” but getting the population of the next century right is really about calling India and China right: if they don’t act as expected, all bets are off.
Sexual orientation: 6“I don’t want to sound politically incorrect,” Silver says, “but it’s easy to have gaydar for people who are acting in a certain stereotypical way, deliberately or not. It’s easy to identify people who fit the stereotype. But how easy is it to identify people that don’t?”
He thinks it’s actually relatively easy: studies have suggested people have a good chance of picking up sexual orientation when shown pictures of strangers. It comes down to non-verbal cues.
“I’m generally not the kind of guy who says intuition is awesome, but there’s a lot of subtle information we transmit,” he says. “In New York I’m a very aggressive walker, we all jaywalk all the time, but you never run into anyone, even though you’re kind of looking at your cellphone, you have all these cues that you develop to negotiate how to walk down a very busy street.
“But here [in London], they’re off, because people have a different left-right sensibility, or just the manners are a little different, I’ll almost run into people all the time. So there’s a lot of non-verbal information people are picking up on.”
Spotting terrorists: 3Spotting terrorists through huge data sifts, profiling and other methods is a high priority for governments on both sides of the Atlantic – but one which might not be all that effective.
“I’m not a big fan of the war on terror and everything it justified, but I do think it’s remarkable that events like the Boston bombings haven’t occurred more often,” Silver says.
“Where there are so many potential threats and thin reeds of information, it’s amazing so many plots are prevented. It’s difficult because you’re talking about the behaviour of a very small number of individuals.”
Silver likens terrorist attacks to earthquakes: using events of different magnitudes, it’s possible to forecast how many bigger attacks there might be over a long period – but forecasting any particular attack is far tougher.
Shopping (what people will buy): 7If there’s one group with an incentive to get predictions right, it’s shops – and luckily for them, whether online or off, they’ve got a stack of data to do it.
There’s a reason shopping habits bear that name, Silver says: people generally buy similar stuff time and again, and retailers pick up on that and are trying to take advantage of it.
“Macy’s is testing 50,000 different versions of its catalogue, targeted at different customers,” he says. “That’s a business where you can go further than you can in other domains.”
You are a target market – and the targeting is getting ever more precise.