Nate Silver, math hero
In the closing weeks of the presidential election, prominent pundits complained that not everyone was following their narrative of a “toss-up” race. Who was resisting the wisdom of the “chattering classes”? Mainly polling analysts who preferred empirical data to impassioned opinions. People like Nate Silver, of the widely read FiveThirtyEight blog at the New York Times, noted the persistent lead of President Obama in polls in battleground states and judged the president an overwhelming favorite to win re-election.
As the pundits were talking about Mitt Romney’s momentum after the first debate, and asking whether or not Hurricane Sandy would slow it, the statisticians – including Silver, the folks at Talking Points Memo’s PollTracker, and
Sam Wang – kept revising the president’s chances upward. At FiveThirtyEight.com, Silver’s models showed a clear trend in favor of the Democratic nominee. In the days after the first debate, the likelihood of an Obama victory was about 60 percent. By the end of the week before the election, it was at 90 percent. This did not sit particularly well with the folks whose paychecks depend on the horse race aspect of elections. Silver was denounced as “an ideologue” and “a joke” because he stuck with his quantitative analysis.
A couple of weeks before the attacks on Silver reached a fever pitch, there was another controversy on the right about polling methods. Unhappy that so many polls showed Obama ahead, based on the pollsters’ models of what the electorate would look like, conservative commentators vowed to “unskew” the polls. Not surprisingly, their version of the polls had Romney with a big lead.
The outcome of the election – the president won the popular vote by about three percentage points and the electoral vote 332-206 – once again confirmed the value of a social scientific research method: get reliable and valid data from professional pollsters, develop regression models based on previous elections, run simulations and make predictions.
What does the “pundits versus probability” controversy tell us? I teach survey research methods and statistics, so I have a keen interest in this question. But even for non-specialists, I think the controversy is important because of what it reveals about the way many Americans view technical expertise, like statistics, and the status of professionals with these skills in contemporary society.
Our public discourse does not give much attention to social scientific thinking. The national media, broadcast and print, includes a few “policy wonk” types. Economists have something of a platform – in part, I think, because people think of them as resembling businessmen (and, yes, even today those are mostly men). There is enough prestige for science in our public sphere that pseudoscientists, such as creationists, sometimes adopt the language of science, talking about hypotheses and data, to give their unscientific designs more attention. This just makes it harder to recognize scientific arguments in public discourse.
Part of the problem is that quantitative analysis requires specialized knowledge. Most American adults don’t have this expertise. Indeed, most college graduates aren’t trained in statistics and many have not encountered regression modeling in their course of studies. Without this knowledge, what Silver and the “math nerds” were doing might seem like fraud or magic, depending on whether or not you liked the predicted outcome.
Probability can be confusing. As I tell my students every semester when we study probability in statistics classes, we are used to dealing with uncertainty in our day-to-day lives. Our minds have developed lots of habits to deal with it. These mental shortcuts, or heuristics, are sometimes at odds with the laws of probability. In other words, we think we understand probability better than we do. (Timothy J. Lawson’s Everyday Statistical Reasoning describes this well for interested non-statisticians.)
Many pundits believed they were being careful with their math when they looked at a string of swing state polls that showed Obama ahead, but by small margins. In most of these polls, Obama’s advantage is within the survey’s margin of error, they thought. Therefore the race is a toss-up: both candidates are equally likely to win. Silver, however, noted that while the margin in these polls was not large, it was almost always in Obama’s favor. That, too, is data. “A toss-up race isn’t likely to produce 19 leads for one candidate and one for the other,” Silver wrote on November 3, 2012.
Silver, who began his career analyzing baseball statistics, used a sports analogy to explain why Obama’s odds of winning were rising steeply as the election drew near, even though the president had not opened up a wide lead. In football, Silver said, a small lead late in a game is harder to overcome than a small lead early in the game – there are simply fewer opportunities to make a comeback. The closer to the end, the more likely the leader is to win. So a close game does not automatically mean it is a toss-up contest.
Sports and politics have some things in common.
Sports fanatics often willfully ignore empirical facts in order to hang on to their belief that their team will win. Partisans do the same. Sports journalists, however, have become more sophisticated about statistics than those who cover politics.
This isn’t just about innumeracy. There are some contemporary instances where people recognize that they don’t have specialized knowledge and respect the professionals who do. But when it comes to matters involving politics, people often prefer to disregard an analysis that conflicts with their hopes or ideology. The professionals who do poll modeling value accuracy. They want models that work; it is part of doing the job well. They may have strong partisan feelings, but their professional standing depends on their technical knowledge, not their ideological intensity. For pundits, it’s often the opposite.
A Good Education
I think we can use this recent triumph of math over punditry as an opportunity to engage with each other and our students on the issue of expertise. Part of what a college education is about is gaining specialized knowledge; we want our students to respect the value of that knowledge. None of us, of course, can be expert in everything, so good research skills are also essential to being a critical thinker: when you don’t have a relevant kind of expertise, you need to know how to consult the views of those who do. Using the election as an example of the value of technical expertise – acquiring it and valuing it in others – can convey that lesson.
In short, this discussion is an opportunity to socialize our students to be confident in their use of knowledge. A college degree isn’t just a credential that helps in the labor market; it also offers membership in a community of professionals who respect rational thought and prefer inquiry to zealotry. We need more of that in contemporary America.