The data behind our partisan divide

Client wanted accurate, engaging copy about Stanford University's Matthew Gentzkow's lecture on his research into partisan politics. I attended the conference, read the research, and wrote a feature article to help capture the findings for their web and social media audiences.

Client: Becker Friedman Institute for Research in Economics at the University of Chicago

Stanford University’s Matthew Gentzkow opened his recent Becker Brown Bag Seminar with a vivid image suggestive of the partisan political divisions in the US today. The photo showed two individuals standing on opposing cliffs, shrouded by fog. A deep canyon separated them. Conventional wisdom suggests that Americans have become almost as polarized politically as the two in the photo, he explained. But is it true? If so, can we blame fake news and the Internet?

In his May 16, 2017 lecture, “Polarization, Fake News, and the 2016 Election,” Gentzkow sifted through research on polarization for his student audience. Defining polarization broadly as “systematic divisions” between liberals and conservatives, he reported that evidence is mounting that the nation is indeed becoming more polarized. However, the outsized role commonly attributed to social media, fake news, and the Internet is probably not to blame, at least, not yet. Instead, Gentzkow argued that TV news, a more highly partisan Congress, and deeper factors such as growing income inequality, seem to contribute to the divide to a much greater degree.

To get a richer understanding of the problem, Gentzkow and coauthors first looked at the divide between politicians in Congress, as reflected in their public statements. The researchers used a machine learning algorithm to conceptualize and measure “a partisanship of language.” After identifying distinctly partisan terminology, Gentzkow and colleagues analyzed speeches from the Congressional floor to determine how well you could guess which party a member is in based on their rhetoric. This predictor remained relatively low through much of the nation’s history, then shot up dramatically beginning in the Clinton Administration.

“It’s off the charts,” Gentzkow said. “Something unprecedented is happening.”

Not surprisingly, the data also show that House and Senate members have become much more likely to vote with their party—another indicator of polarization. This trend has risen steadily since the Carter Administration, but it was also high in the early 20th century, Gentzkow noted.

Next Gentzkow reviewed the evidence on polarization among voters. Literature reviews up to 2008 have revealed a steady electorate clustered in the political center. He reports little evidence of increasing division over specific issues (e.g., gun control, abortion), in how voters describe their party affiliation (e.g., Democrat, Republican), or even on how they describe their ideology (e.g., liberal vs. conservative). He also could not find convincing evidence that Americans are relocating geographically along party lines. Gentzkow observed that these trends “might be an optimistic note.”

What is increasing is the negative feelings voters held about members of the opposing party. He found striking evidence that compared to data from 1960, in 2008 people were more likely to judge members of the opposite party harshly, rating them less smart and more selfish. Today 25 percent to 30 percent of Americans say they would be concerned if their child married someone of the opposite party. In 1960, Gentzkow said, “that would (have been) a weird question."

What is Driving Us Apart?

Gentzkow acknowledges that it’s tricky to identify the causes behind this increased hostility, but he doesn’t think social media or the Internet drove it. By 2013, only roughly about 8 percent of all news media was consumed through online sources (all apps, all tablets, all mobile, etc.), confirming that TV is still king in audience and influence.

Moreover, Gentzkow’s research put the number of Americans for whom social media is their most important source of election news at 14 percent— but most users tend to be young. However, the highest levels of polarization are seen in the elderly, and it’s rare that 75-year-olds are reporting strong use of Twitter and Facebook. People over 75 “really, really like” Donald Trump, Gentzkow said, which suggests that it wasn’t fake news or real social media that elected him— although he ran through a few alternatives that could explain a larger role for social media.

Instead, Gentzkow said there are many recent papers using natural experiments that show TV news from cable and mainstream sources most likely has a big impact. TV influence could “plausibly explain quite a bit of the rising polarization” and is consistent with the demographics, unlike social media.

Then what role does the Internet play in today’s polarized climate? Drawing on his research from 2011, Gentzkow demonstrated that most online news consumption concentrates in large centrist websites, rather than on extremist ones. In fact, ideological separation is much lower among Americans online than it is in everyday, face-to-face interactions. To a greater degree, Americans are much more segregated along party lines in their in-person political conversations, as families, and with those they say they trust.

“Media as a whole, including digital media [at least at the time of this paper], are a force pulling away from segregation, whereas the people you actually know are pulling in the opposite direction,” Gentzkow said.

This may already be shifting, though, as Facebook’s influence grows. Gentzkow expressed concern for what may happen to the nation’s centrist views as political news becomes filtered through personal circles of influence on Facebook. He also notes as a concerning trend that the social media giant consistently “down weights” the influence of traditional media outlets in its feeds.

“There are fundamentals in the economics of social media that should make us worry, but so far they are not a driving force,” he said.

The Impact of Fake News

In more recent work with Hunt Alcott, Gentzkow tried to assess the reach and impact of fake news—such as unarguably false stories that suggested Trump was endorsed by the Pope or that Hillary Clinton led a child sex trafficking ring out of a pizza parlor. By identifying fake stories through fact-checking sites, Gentzkow then analyzed them for page views and sharing rates. He concluded the average voter in the 2016 election saw one to five fake stories. The fake news leaned strongly pro-Trump; they found 115 pro-Trump fake stories that were shared on Facebook a total of 30 million times, compared to 41 pro-Clinton fake stories shared a total of 7.6 million times.

What impact these had on the outcome of the election is difficult to gauge. The study found that about 15 percent of survey respondents recalled seeing headlines of fake news stories shared before the election. Interestingly, 14 percent also reported seeing placebo stories the researchers made up—“fake fake news” that never appeared anywhere. This suggests that Americans are inclined to remember or relate to headlines that align with their ideological beliefs.

Drawing on what is known about TV ad impact, Gentzkow estimated roughly that voters would have had to have seen in the magnitude of 20 ads each for a sway in the election. Yet, he cautions, it may have plausibly swayed the outcome, especially if voters really believed what they read.

In wrapping up, Gentzkow confirmed that the conventional wisdom that polarization is rising seems to be true, but TV, not the Internet drives it. He also speculates that the generally hostile language out of Congress may be influencing voters as much as voters influence Congress.

Ultimately, he suspects that polarization is not really about political rhetoric as much as it is about structural challenges.

“It’s about deeper things of the people living in this country, like income equality and inequality in outcomes, where (Americans) are really living very different experiences from one another,” he said.

— Jennifer Roche

Posted: 6/12/17

More work samples