Skip to main content

Election Outcome: In Nate Silver We Trust

How Bayesian statistics is changing the nature of political commentary

Nate Silver at a 2009 conference.  Image by Randy Stewart from Wikimedia Commons.
Nate Silver at a 2009 conference. Image by Randy Stewart from Wikimedia Commons.

The votes are in, and one big winner Tuesday night was Bayesian statistics. 

The day after the national election, statisticians were sharing in the glow of the praise heaped on New York Times columnist Nate Silver, who hit 50 out of 50 states in the presidential campaign.  While facing considerable criticism from skeptics, Silver proved that the public had a deep desire for well written quantitative electoral analysis that cut through the traditional intuitive assessments of the punditry.

Read More

There is a certain pride that goes with this," said Alan Gelfand, James. B. Duke Professor of Statistics and Decision Sciences.  "We in the field often get criticisms along the lines of 'liars, damn liars and statisticians.' 

"Statistics is not a glamour profession. Happily, we don't need to be in the New York Times or on NPR much.  But we take pride in how we are advancing science. Once in a while when something comes along and reveals that doing something intelligently with probabilities is better than seat-of-the-pants intuition, we take pride in that."

It hasn't always been this way. Polling controversies marred the 2000 and 2004 presidential elections.  But writing in his Fivethirtyeight blog, Silver represented a new wave, at least in the popular media, in statistical analysis, one based on the synthesis of a variety of information sources, all of which were weighted differently based on history, current trending patterns and other factors.

This is the essence of Bayesian statistics, which Gelfand says provides statisticians with a methodology for assigning probabilities to future events given what you have already observed. (For more, see accompanying story.)

"Bayesian analysis allows you to synthesize all sources of information in a coherent way. The sources of information in this case are state-level polls, and national polls," said Gelfand, who chairs Duke's department of statistical science.  "What Nate did was take a collection of these data sources and integrate them.  But to integrate them, he had to evaluate how good each was, and that's where it gets somewhat sophisticated."

Silver isn't alone in this work, Gelfand noted.  Sam Wang of the Princeton Election Consortium also used Bayesian analysis and hit 50 out of 50 states in the presidential election and even correctly predicted one close Senate race that Silver missed. 

Silver and Wang's algorithms repeatedly pointed toward the likelihood of an Obama victory, when national polls showed it a toss-up or even had Romney ahead.  Several Romney supporters accused Silver of bias, but what also drew the ire of many is that Silver never made predictions but only assigned probabilities.  If Romney had won, Silver could claim he wasn't wrong, only that the less probable occurrence had played out.

This critical distinction points to a tension between statisticians and political commentators, Gelfand said.  Where pundits often base analysis on qualitative readings of intangibles such as "momentum" or "ground game," Silver presented an alternative language based on hard data run through rigorous analysis and presented with precision, caution and care.

"I also think, and this is my own bias, the media didn't like what Nate and Sam Wang were offering because it diminishes their story," Gelfand said.  "They like to make hay about the uncertainty, and if someone takes that uncertainty away, they're left with nothing to spin."

Silver was originally known for work on baseball statistics, and some have made the connection between his success and Moneyball, the popular book and film that described Oakland A's General Manager Billy Beane's use of data-driven sabermetrics and his battles with coaches and scouts wedded to traditional approaches of player analysis.

Jerry Reiter, a Duke associate professor of statistical science, noted that both examples "wonderfully illustrate the benefits and power of quantitative analysis --this is a very healthy thing."

But the differences are illustrative, he said.  Reiter said that while pundits "are paid to be polemical and partisan," the most rational of them have long accepted "the benefits of election polling."

"Every high-profile election campaign in the modern era relies extensively on scientific surveying and quantitative analysis," Reiter said.  "However, in the pre-Moneyball era, the scouts did not value quantitative modeling at all, which gave Oakland an opportunity for arbitrage that they effectively used."

Silver's popularity might encourage statisticians to dust off their manuscripts.  His book sales shot up 800 percent in the 24 hours after the election, ranking second on the Amazon best seller list only to the "Diary of the Wimpy Kid" juggernaut. 

"Anything we can do to encourage public literacy about statistics is a good thing," Gelfand said.  "With the advent of inexpensive computation, we've lost a public sense of numeracy and an understanding of magnitudes.  Any time we can get people to think and talk quantitatively about an important issue is a benefit, not just as a matter of pride for statisticians, but for the good of the public discourse."

So if statistics can change baseball and political punditry, what's next? Gelfand, whose expertise is the statistical study of environmental processes, hopes you'll see more statisticians involved in the public conversation on climate change and environmental exposure.

"I think the ability to bring sophisticated modeling to environmental issues will again present benefits.  Al Gore tried a little bit and maybe people still remember the hockey stick graph on global warming. There are challenges, but more and more, this is an important direction for public conversation."