Nate Silver, the founder and editor in chief of the statistics-based news site FiveThirtyEight, joined Columbia professors Tuesday night on a panel to discuss data literacy, the value of polling, and the future of news media following the 2016 presidential election.
Silver has a reputation as one of the most reliable statisticians in the country, having accurately predicted the presidential election results of 49 out of 50 states in 2008 and all 50 states in 2012. But, along with many other well-known polling outlets and media organizations, FiveThirtyEight incorrectly favored Hillary Clinton to defeat President-elect Donald Trump.
The panel, sponsored by the University’s Data Science Institute, featured Silver, Director of Columbia’s Tow Center for Digital Journalism Emily Bell, and Wallace S. Sayre professor of government Robert Shapiro. Professor of public affairs and political science Ester Fuchs moderated the discussion.
Over 800 students remained on a waitlist for the event, where panelists began by discussing what went wrong in polls in the weeks leading up to the election. Silver noted that Trump did not beat his national nor swing state polling by an unprecedented amount, and cited overconfidence in Clinton’s chances as a main contributor to the national shock after Trump’s victory.
“From our vantage point, the question is not why were our polls so wrong, because Trump won only very narrowly, but why do people think the race was so certain when the polls, especially toward the end of the race, showed a very competitive race?” Silver said.
Silver described a bias toward “cosmopolitanism,” noting that reporters and the public were too reliant on Clinton’s coalition, which was consolidated to a few key states.
“The argument that you saw being made over and over again was that Clinton has a diverse coalition, and therefore maybe Trump can win the popular vote by running up the margin in the South, but he can’t win the Electoral College, when actually the opposite was true,” Silver said.
The panel also discussed the lack of data literacy among the public and news media. FiveThirtyEight’s final polling model gave Clinton a 71 percent chance of winning, and Silver said that most people did not take into account that Donald Trump had a 30 percent chance of victory in this model.
“You actually are predicting that the trailing candidate will win three out of 10 times,” Silver said. “People will say, ‘Oh, Nate’s just hedging his bets …’ That’s not true, you’re saying Trump’s polling is good enough that a candidate with that position will win three out of 10 times.”
Panelists went on to comment on the role of journalism and social media in the 2016 election cycle. Bell said that mainstream media had much less control over what news reached consumers than in past elections.
“The difference between 2012 and 2016 is you have a very different ecosystem in terms of news consumers,” Bell said. “We had social media in 2006 to 2010—in 2012, it was definitely a force, in 2016, it had gone from something that was generally adopted by the coastal elites to a global phenomenon which was not just adopted by the candidates, but so thoroughly understood by elements of the campaigns that they were way ahead of the traditional media.”
Bell also noted that polls may have been emphasized too much by the media during the election because polling updates often draw more viewers and readers.
“News organizations place not just a lot of faith, but also a lot of investment in polling relationships and how you then represent that data,” Bell said. “If you want to see your figures go off the charts, you update with polling data.”
Shapiro cited inaccurate and insufficient state polling as a key reason that people were misled by statistics going into election night. He also said that the news media were not focusing on the right coverage, preventing the public from being more prepared for the election’s outcome.
“The media are consumed with avoiding making what’s called a type II error—they want to avoid missing the big story,” Shapiro said. “And here, the big story that was decided was the election of the first woman president of the United States, and that became an important theme of the election, and to the extent that [the media were not] mindful of this issue, of this need for accurate coverage.”
Near the end of the panel, Silver discussed the need for a wider variety of viewpoints in news coverage.
“I think in particular about data science careers [we need to be] finding ways to increase the diversity of people in data sciences and journalism, including some biases that are more subtle,” Silver said. “Maybe people that come from different parts of the country, maybe the next newsroom start-ups should be in Kansas City, and not New York and Washington.”