Social media isn't the main force driving most partisan news consumption

They concluded that 17 percent of Americans are "partisan-segregated" by TV news, meaning they get large portion of their total news from sources that lean either left or right. By contrast, only four percent — that's one in 25 — of people who consume news online are partisan-segregated. When the researchers turned their attention to individuals over time, they found that TV viewers who were partisan-segregated in one month were likely to continue consuming similar news diets the following month. In contrast, online news consumers who were partisan-segregated one month were a lot more likely to consume a more varied news diet the following month. 

They say their results should encourage researchers to look more closely at TV, rather than social media or digital news, as a source of polarization. "[P]artisan segregation in TV audiences—whether it is large enough to be considered alarming—is large enough to justify TV news receiving at least the same level of scrutiny as its online counterpart," the authors write.

Interesting Engineering caught up with one of the authors, Daniel Muise, to learn more about this surprising new study's methods, results, and implications.

This interview has been edited for length and clarity.

Interesting Engineering: There's no doubt that Americans have become more culturally and politically divided over the past several years. A lot of commentators have placed most of the blame on social media, but you and your co-authors present evidence that something else is at play. What's actually going on?

Daniel Muise: There's been a lot of focus for years and years on how the internet or social media — these new exciting platforms — must be catalyzing and really exacerbating this issue [of Americans consuming partisan news]. But there are a lot of academics who have done really, really great studies over the years using browsing log data and other types of rigorous methodologies to study this, and they've consistently found ambiguous, non-confirmatory results generally suggesting that the online audience is not actually that polarized in terms of what they consume.

It's actually a more balanced news diet more often, with only a few folks that are really, really on the far right or far left, in terms of what they take in. 

IE: You found that news consumers with highly partisan news diets were watching more television, right? How did you come to that conclusion?

It was missing the elephant in the room, which is that a much more popular method of consuming political content — and all content in general — in the United States is still the television screen. That means cable news.

On an hourly basis, especially, it's just massively larger than online news consumption. So we took a lens onto that. We got a really fantastic data set where we can compare apples to apples using hundreds of thousands of Americans as part of a big sample, with everybody consenting to send their data through the Nielsen Company.

By looking directly at the TV audience and the online audience, breaking down a bunch of different parameters of what counts as partisan bias and what counts as a skewed news diet, we unpacked in a bunch of different ways that TV is maybe where researchers want to focus if they're concerned about partisan audience segregation, as we're calling it.

IE: How did you reach those conclusions?

It all starts from the Nielsen company's data. So, this is comprised of two large panels, each with hundreds of thousands of participants. One set of panelists has their computer's web browser URL tracked every second, and the other has their TV tracked every minute. That gives us the TV channel, the TV program, and some categorization information.

What we do is identify news websites and programs on TV, and we code them for partisan bias. And then we calculate the partisan bias and each panelist's news diet based on the browsing or viewing history. Those news diet calculations actually enable everything else in the paper.

The four things we go after are the share of Americans with partisan-bias new diets each month (in each demographic), the number of months that typical news consumers maintain biased news diets, the overall pattern of news consumption in the United States, and how that pattern is evolving.

IE: How do you determine the partisan bias of a particular channel or site?

We start with the Nielsen company's own categorizations. So they have a big incentive to understand what types of programming there is on, on browsers and on TV. For coding partisan bias, we have this really wonderful set of 20,000 domains that came from Robertson et al. — a really good study that we cite. They used a whole different method to basically create an ordinal ranking of websites according to partisan bias based on audience-based metrics.

We take that list, and we use that to create points of separation between what we might call centrist, moderately right or left, and more further right and left. The beauty of our method is that while we choose those cut points here, based on sort of heuristics and pretty reasonable choices, we, more importantly, provide a framework where other folks can move those thresholds based on what they think is an appropriate cut-off for partisan-slanted or skewed content.

IE: That sounds sort of circular, using information about one group of consumers to define the partisan slant of an outlet and then using that designation to assess the partisanship of another group of consumers. How do you justify that move?

Yes, that's something that we went back and forth on a lot. So what we would love to do and what everybody in this field would love to do — and what we recommend for future research — is a seriously rigorous and ultimately massive effort at finding some kind of content-based or hypothetically objective ranking of partisan bias of all news content in the news sphere. Of course, that would be a pretty massive undertaking.

The audience-based approach is pretty common in our field. What we do to make sure that we're not just being too circular here, to use your word, is we validate it against a few different other audience-based rankings using different methods. The one we use is based on Twitter sharing behavior, we use others that are generated from audiences that are not on Twitter, for example.

There's a very fantastic correlation between a lot of different ways of triangulating this sort of ordinal ranking.

IE: This figure from the paper contains some of your most important results. Can you explain what's going on here?

What we're showing is the share of American adults that are on the partisan right or on the partisan left based on their consumption of news via television or online browsing. So let's break that down by looking at that spike. On the x-axis, we have 48 months from January 2016 all the way to December 2019. On the y-axis, we're showing the percentage of American adults that fall into a given category.

So zooming in right at that highest peak, that was November 2016. Obviously an auspicious time in politics. See that dashed line is right around just a little bit past 13 percent? What that is saying is, in November 2016, 13 percent of American adults consumed the majority of their television news diet from left-leaning political content, in this case, MSNBC and/or CNN. If you look at the next dashed line down, that shows that between 11 percent and 12 percent of U.S. adults were consuming most of their television news diet from Fox or Fox Business. 

IE: Were you surprised to see how important TV seems to be in Americans' partisan news diets?

Well, yes and no. We were familiar with the fact that TV was huge and that it was understudied. That was the impetus for this study. So we weren't necessarily surprised it was like this. We were surprised by the magnitude. 

IE: To what extent does this show that people just spend a lot more time watching team TV as a medium irrespective of this news? 

Yeah, it is reflective of that. And that is a big point we have here, that TV is the top medium for content in terms of minutes consumed. One really, really interesting way of looking at this that I really like is that the focus people have had on say, echo chambers or filter bubbles or some of these pop-ish ideas about the online ecosystem, is that the structure of the system pushes people into consuming similar partisanship to what they already have in their minds.

But if you take that same idea that it's the system — the structure of the system — that pushes people to consume more partisan content, then think about what TV is in terms of how content is recommended to you. If you're on Facebook, there's like a high likelihood that it will notice that you like right-leaning content, and that the algorithm will provide you more of that.

Let's say you are simply left-leaning and you are tuned in MSNBC. First off, it's 30 straight minutes of left-leaning partisan news content. You have to turn the channel away or turn the TV off to get away from it. And, if you sit there and the content ends, it's not up to a recommendation engine to maybe give you more left-leaning content. The next thing up is guaranteed to be, essentially, left-leaning content. It's like the strongest recommendation engine that we have available to us.

So, it makes sense people are worried about structural impacts on news consumption online, but to the extent I care about that, look at TV.