Are Media Echo Chambers As Big As We Think?

We’re constantly told that we’re trapped in media “echo chambers”, that our media diets mirror our political leanings. But what do the data say? Is it possible that a majority of us have a much more moderate media diet than we assume?

A new paper by Andrew Guess, Assistant Professor of Politics at Princeton, provides a completely unique data set that complicates our assumptions about America’s “echo chambers” and media diets.

Listen on Apple Podcasts or wherever you enjoy podcasts.

Transcript

Anthony Fowler:

I am Anthony Fowler.

William Howell:

I'm William Howell.

Wioletta Dziuda:

I'm Wioletta Dziuda, and this is Not Another Politics Podcast.

Anthony Fowler:

We are told frequently that there are two Americas. There's the America in San Francisco that eats the overpriced avocado toast, and they read New York Times, and then there's the America in Kansas where they sip on cheap coffee in diners, and they watch Fox News. If they have to read a paper, they maybe read the Wall Street Journal. The concern is that we have these two Americas, and they don't even get the same news, they don't get the same information, they're not hearing the same arguments. If Donald Trump says something stupid, then that's going to be passed around by all the liberals, and everyone's going to know about that. When Joe Biden says something stupid, that's going to be passed around by all the conservatives, and they're all going to know about that.

When it comes time to actually have an election or make an important decision, the American electorate is just not even up to the task, because the media consumption and the conversations are so partisan that we can't even agree on basic facts. What do we think? Is that a problem? When's the last time you guys had expensive avocado toast versus cheap coffee in a diner?

William Howell:

Oh, they're both delicious from my vantage point.

Wioletta Dziuda:

Yes, we are eating expensive avocado toast with cheap coffee over here in Chicago. Yes, it definitely feels this way, so it would be good to look at the data and see whether that's indeed the case. Will, you talked to someone who tried to figure out whether we are consuming very different news, whether Democrats are reading very different news online than Republicans?

William Howell:

I did. I talked to Andrew Guess who is assistant professor in the Politics Department at Princeton, who has a new paper that's in the American Journal of Political Science called "Almost Everything in Moderation: New Evidence on Americans' Online Media Diets," and tries to speak to precisely this concern by focusing in particular on people's consumption patterns of news, and political news in particular. He looks deeply at a bunch of data that looks at online consumption behaviors and has some really interesting findings to share, so let's give it a listen.

Andy, what a pleasure to meet you and have a chance to talk about your new paper, "Almost Everything in Moderation: New Evidence on Americans' Online Media Diets." Let's talk about the evidence. Much of the evidence that was animating, certainly the popular debate was drawn from survey research. That is you'd ask people about what news do they consume, and then you correlate that with their political attitudes, and you say, "Wow, we see a lot of separation and a high degree of connection between what my ideological priors are and whether or not I'm going to consume predominantly liberal or conservative media outlets," but you're doing something different.

Andrew Guess:

Yeah. My approach is motivated by the shortcomings of survey-based methods of studying these questions as you suggested. If I asked you, "How many hours did you spend watching the news last week, and which channels were you watching, and which news programs were you watching on those channels?" Maybe you would give me a reliable answer, but maybe not. We tend to just overestimate our own accuracy. The data that I rely on circumvent this issue by directly measuring the visits to websites on people's devices, so desktop, laptops, and more available on their mobile devices. The secret sauce of this paper really is being able to combine fine-grained information about what people are actually doing online with individual-level survey data, which tells me is this person liberal or conservative?

What's their gender? What's their education level? All these basic facts about individuals, and then putting those together allowed me to start to put together more comprehensive portraits of people's online media consumption habits. My paper covers two separate samples from 2015 and 2016. The 2015 dataset, it has over six million rows. Compare that to standard survey sample, that might have 1000 or 2000 observations. Each of those observations in the 2015 dataset, corresponds to a web visit. The row will have a domain, it'll have a full URL, and it will have a few other pieces of metadata. It'll have the duration in seconds, which is a rough estimate for how long a person may have spent on that page.

Then there's an identifier that allows me to link it back to an individual from those survey data. Those six plus million visits are actually based on fewer than 1400 people in the US.

William Howell:

Describe that picture for us, the core finding that comes out of the analysis of these data.

Andrew Guess:

I come up with a very simple summary of the ideological slant of people's overall political media diets. Basically, I take all of these web visits, and I distill them down to a single number for each person, and then I simply look at the distribution of these numbers between mostly two groups, Democrats and Republicans. The idea is that if "echo chambers are real," you're going to get Democrats clustered on the left of this hypothetical ideological continuum of media consumption, and Republicans are clustered on the right, and there's not a lot of overlap between the two. Well, I directly compute what this overlap is, and I find so in 2015, that overlap is almost two thirds, so 65% overlap. Then in 2016, it was about 50%.

That to me was evidence that there are substantial reference points in common between people from different political backgrounds.

William Howell:

What's driving the overlap? Is that because people are seeing the same sites, we're all consuming moderate outlets, or is there some differentiation in the moderate media outlets that people are visiting?

Andrew Guess:

Yeah, it's not necessarily that people are visiting the exact same sites, but they do, on average, seem to be visiting the same types of sites more than others. A surprising share of people's visits to politics or news-relevant pages in the data seem to go to large, mainstream portal, or aggregator type sites. AOL.com really pops out as a huge source of traffic. It was a default homepage setting for lots of people on their browsers. It has tons of news content. Same for MSN.com and things like Google and Yahoo. They loom very large in people's consumption behavior, best as I can tell. I think that goes a long way towards explaining what's behind these patterns. It's important to remember that we're political scientists.

Presumably, we're both very interested in politics and probably consume a lot of political media, but most people aren't like that. To the extent that many people encounter any kind of hard news content at all, it's often incidentally or in the course of looking at other kinds of information. These mainstream portals, of which there are many others, can often serve as a hub and a place where people encounter this kind of information.

William Howell:

The moderation is a function of people generally going to moderate sites?

Andrew Guess:

Yeah.

William Howell:

It's not a function of people who, on the left, read BuzzFeed and the New York Times, but every once in a while, check out Fox News, and then you observe that, you calculate some average for them, and boy, they look pretty moderate?

Andrew Guess:

Right.

William Howell:

It's about the primary consumption comes from these more moderate portals?

Andrew Guess:

Yeah, and I looked at that too there's this question of whether maybe this could be driven by people who are sampling. These omnivores who are looking at sites on both the left and the right. I found a few of those people, but it seemed to be pretty rare.

William Howell:

Mm-hmm (affirmative). Let's talk a little bit about how you figure it out, or how you calculated the political slant of each of these sites. How do you know how liberal or conservative CNN is versus Fox News?

Andrew Guess:

Yeah, this is a persistent challenge in research of this kind. The basic method is this. Some Facebook researchers basically took a bunch of data from users on the platform who self-reported their ideology, and then did a basic crowdsourcing method, where they take the average ideology of people who share links on Facebook to different news articles. Basically, once you aggregate it to the source level, you get these kinds of average crowdsourced measures of ideology.

William Howell:

Let's think about this for just a second, we want a measure of the political slant of each media outlet, and the strategy is not to go in and read the content comprehensively, or even sample from it and then judge whether or not it leans right or leans left. Rather, it's to leverage the extent to which the public engages that material—

Andrew Guess:

Yeah.

William Howell:

... and by engages here, for you, it means their willingness to share an article from a particular site. To the extent that we see regularly moderates sharing CNN material, then self-identified moderates sharing CNN material on Facebook, then seeing what we're going to the inference that we're going to make is that then CNN must be reasonably moderate.

Andrew Guess:

That's right. What I'm mainly interested in are relative comparisons. I'm not particularly concerned that we come up with the absolute ideological slant of CNN.com, to the extent that that's something we can even identify, but I am interested in knowing whether BuzzFeed is to the left or to the right of CNN. I'm also interested in the extent to which these estimates correlate with other prominent methods that are out there, which it does very highly. That gives me some confidence that even if it's not perfect, this approach does give this relative ranking, and it also covers an order of magnitude more sources than I would be able to do otherwise.

William Howell:

Without getting too lost in the weeds, I want to raise one last point on this, which I wonder if you've thought about, which is that you think about shares of CNN. I can imagine what a lot of the shares look like. Those shares consist of breaking news, right? A tornado hit, somebody prominent died, that kind of thing, and that draws from lots and lots of people because CNN is the place you go to for breaking news, which would suggest the kind of moderation on their part, because lots and lots people are interested in breaking news, but the news slant, when you think about their editorial page, or when you think about how they analyze the decisions that are made by Donald Trump, for instance, looks very different and then caters to a very different kind of population.

You can imagine similar kinds of things going on with outlets that have sports or entertainment as a way to draw in a broader public that then make them look more moderate than in fact they are, because their news coverage diverges dramatically from.

Andrew Guess:

Yeah, yeah. That's a really good point. Yeah, I acknowledge that I think there's some sites where there's going to be a more plausible set of cases where you might have people across the spectrum who are going to share like with CNN. For what it's worth, CNN gets estimated at negative 0.27 out of negative from negative one to one. It's actually—

William Howell:

Where negative one is really liberal—

Andrew Guess:

It's really liberal.

William Howell:

... and positive one is really conservative.

Andrew Guess:

Right. It's actually pulled a little bit maybe more to the left than you might think if you're imagining CNN as being like smack in the middle, which is actually true of a number of the mainstream news sources in the data. New York Times and ABC, and a lot of these sites do score as quite left of center, and I think it's due, at least in part, to the expressive sharing behavior that you might expect to find on Facebook, which is related to your question. I think probably the most general conclusion that I can make is that, if anything, my approach is using scores that probably exaggerate the partisanship of most websites, with the exception maybe of these kinds of breaking news type sites.

My argument would be that it would make it more difficult for me to find evidence of moderation and overlap in people's online news consumption diets.

William Howell:

The overlap is considerable that you document, but it's not perfect and another thing that you show, and tell me if I've got this right, is that there is a significant portion of people who are self-identified as Republicans, who consume distinctly conservative news, right? That it's not bimodal, but the right tail is pretty thick, and it grows in size from 2015 to 2016 to the extent that you want to make anything of that. Let me try making something of that, then tell me how I'm being completely irresponsible, which is to say that, right-wing media has taken off and taken hold, and among the people who are politically most engaged, they are disproportionately consuming this highly conservative news, and that that's increasingly so.

While you show it between 2015 and 2016, we have every reason to believe it's taken off even more over the course of the Trump presidency. There's lots of things that are packed into that claim, but what you're delivering is not good news. What you're delivering is cause for real concern, because you've documented, even as far back as 2015, this big chunk of highly politically engaged people on the ideological right, who are consuming dramatically more conservative news than are people on the left, and that we have reason to believe that those consumption patterns are only intensifying.

Andrew Guess:

What you're saying is consistent with what I show on the paper, and you're right that there is this asymmetry that comes through at numerous points. I think there are supply-side considerations here that I can't fully disentangle, right? It seems clear from other work that there's no equivalent to Breitbart in terms of its reach and influence on the left. I mean, there are left-wing partisan websites with very opinionated slanted content, but they just don't have the same pull and the same audience as their counterparts on the right.

That is reflected in the news consumption habits of some of the people in my data, but basically, I argue in the paper that, it's a minority of people who, as you suggested, are disproportionately engaged in politics, they consume a lot of news about politics, and these are the people with the most polarized media consumption habits.

William Howell:

Right. We can take comfort or find cause for concern in the different parts of that characterization. What you might say is that there are very many of them in the main, the overlap is considerable on the one hand. On the other hand, you might say, and I guess I would be inclined to say that, "No, it's not just a sizeable chunk, it's in the data on offer, it appears to be increasing, and that these are the politically most engaged folks, and therefore, we can expect to have a disproportionate impact on our politics." To the extent that we worry about then echo chambers, to the extent that they do exist, they're populated by people who are politically informed and politically engaged.

Andrew Guess:

Yes. I try to make that clear, and I try to be explicit about that. In fact, I show that these people vote more. This is nuance to the story that there are people who fit into this popular notion of echo chambers, but they're not representative of most people, even so they might have disproportionate influence on our political system.

Wioletta Dziuda:

Very good interview, Will. Can you just briefly summarize what Andrew does in his paper?

William Howell:

Sure. He gets access to online behavior of a sample of Americans. He's actually able to track on all the clicks that they make. I mean, it's just not self reports. They're actual data on the sites that are visited over a 3-week period in 2015, and then again in 2016. He's able to link those data to survey data that gives him a bunch of information about the age, ideology, and partisanship and so forth of the person who's doing all that clicking. He then collect these other kinds of data about the ideological leanings of different media outlets based upon sharing patterns on Facebook.

Look, if it's all liberals sharing the news for a particular site, then the presumption is that that's a really liberal outlet, and if it's all conservatives sharing news on particular site, well then, we're going to infer that the content of that new site is more conservative. If it's the case that we live in echo chambers and in altogether different realities, at least when it comes to the consumption of news, he suggests we would expect is that there would be effectively a bimodal distribution.

There would be, way off on the left would be all the news consumed by the liberals, and way off on the right would be all the news consumed by the conservatives, but instead, he shows that there's actually a remarkable amount of overlap, roughly two thirds, and then suggests on the basis of that that we're not quite as polarized, or certainly not living in the echo chambers that Cass Sunstein wrote about in 2001.

Anthony Fowler:

Should we talk for a second about what could go wrong potentially? What are the things that might lead us to get the wrong impression? I guess I'll start with the people in the sample know that they're being monitored. They've agreed to participate in the study. One potential concern is that you could imagine that if you know someone's looking over your shoulder and going to know what sites you visited, maybe you don't spend all of your time on hyperpartisan new sites, and you say I want to look like I'm a well-rounded person. If I'm reading a lot of New York Times, I also make sure I read some Wall Street Journal or vice versa. Is that possible here?

Is it possible that people are, when they're not being watched, they're more partisan, but when they're being watched, they want to make sure they look like they're sensible moderates?

Wioletta Dziuda:

I think it's completely possible. I think it would be even strange to assume that knowing that you're part of the study does not change your behavior. There might be also selection into the sample that, if I know that my browsing habits are questionable, they would be questionable to someone who is mainstream, I would probably not agree to install the software. To his credit, Andrew tries to check whether people who selected to download the software are different in many different ways that he can observe from people who didn't.

He seems that he cannot find any real differences between them, especially there's a question in the survey that people answer about how much they care about privacy, and people who opted in care about privacy to the same extent than people who opted out, but for me, that actually raises a red flag instead of calming it down because if you and I care about privacy to the same extent, why would you decide to download this tracking device and why would I decide to say no? Perhaps there's something unobserved, and this unobserved thing is the thing that comes to your mind first is our behavior online. I might be just a little bit more ashamed of my behavior than you are.

Andrew does a good job trying to see whether his sample's bias, but I think it would be unreasonable to believe that we don't have any bias here. My guess would be that we see more overlap than we would had we have data directly from Facebook and had we seen everyone.

Anthony Fowler:

It's worth pointing out that the bias could go in either direction, and we've even when we interviewed Greg Huber about his paper on partisan cheerleading in surveys, the concern was maybe people behave in a more partisan ways when they know they're being studied than they would in normal life. In principle, it could go the other way as well that, "Oh, I know I'm part of this political survey, and I know I'm supposed to be a good liberal. I'm going to make sure I'm not going to read the National Review this week, even though I normally..." Who knows? I mean, it could go the other way.

Wioletta Dziuda:

It could, yes. I think it's less likely, but yeah, you're right, that it could, you could tell stories for that.

William Howell:

I am guessing, Anthony, that this paper you're going to put into your good news file, which is, "You see, people are not only reasonable, they're also pretty moderate, and that the differences between Democrats and Republicans are genuinely overstated, and a lot of the worries about how everything's on fire, and it's on fire because the voters are crazy are exaggerated."

Anthony Fowler:

Yeah. You said it well. I don't even have to be here. You can just construct my argument for me, but yes, absolutely, I think most of the evidence we have suggests that the American public is relatively moderate and given that, it's not all that surprising that Andrew finds that the media diets of these people are relatively moderate as well, that they're not just reading Breitbart all day long, although there is actually this small little tail of people in his dataset that look like they essentially do just read Breitbart, but there aren't very many of those people. I think that's mostly a good thing.

I think that's mostly good thing in the sense that, if we're worried that people were only getting one worldview from the news and we're worried that people just aren't even getting a lot of the relevant facts when it comes to discussing important issues and voting and so forth, this paper assuages those concerns. This paper suggests that, even if it's the case that our media is too partisan, and I think there probably is a fair case to me that our media is too partisan, that when a Democratic governor does something bad, the New York Times is way too slow to point it out. When a Republican governor does something bad, Fox News is way too slow to point that out.

Nevertheless, if you've got a bunch of moderate Americans who are looking at both outlets, they're still going to get pretty good information despite the partisanship of the media outlets.

William Howell:

Okay, so I am committed to my worry, and let me try to resurrect the fear and dread that I have about the state of contemporary American politics by saying a few things, by reference to the paper. One is that the moderation that's documented in the paper is not born of people sampling broadly across the space, which would be great to see, right? I mean, if people were really like, "I look moderate, but it's only because I'm really sampling everything and that on average puts me in the middle." That's one thing. It's in the main...

Anthony Fowler:

Although he does do this analysis where he tries to exclude the aggregators. He tries to exclude the big websites that people just land on by default, for example.

William Howell:

That's right.

Anthony Fowler:

And he still finds this overlap even among the other sites that aren't those.

William Howell:

Yes, yes, but it isn't just about the aggregators, it's about a couple of main portals. It's about everybody's checking out CNN or MSN. There aren't many people who were sampling the extremes. The other thing is that, the right tail is not trivial. It appears to be growing from 2015 to 2016. All of that is before Donald Trump, and I would venture to say that that tail has probably grown even more. Those people consist of the people who are more politically engaged. If you look out in the world, and you say, "My God, I'm worried about things like political violence. I'm worried about things like hate speech," that obviously is not representative of the electorate as a whole, but it has a real effect on certain populations.

Then, within this good news story of his own paper, there's cause for real concern and that doesn't appear to be as great on the left. Did I depress you? Did I bring you back, or are you going to stay and talk?

Anthony Fowler:

No, no, you make some compelling arguments. I think the part of your argument that I find most compelling is the potential concern about political violence. I guess if I was going to push back, you point out that it's true that that tail is there on the right and that hump is not there on the left. It seems as if maybe this is a bigger problem on the right than it is on the left. Although, if I were going to play devil's advocate there, I might say, "Well, maybe the entire media in general has a bit of a left slant." We know that most journalists themselves are left of center. The fact that CNN comes out as somewhat centrist outlet in his scoring could actually just indicate that most of the media is left of center.

Maybe the big hump that we see in the data is actually left of center relative to what the typical American would like to consume.

William Howell:

It's true.

Anthony Fowler:

That's plausible to me. Maybe that's also cause for concern. The fact that everyone's reading the same thing doesn't mean that everyone's reading unbiased, accurate information necessarily.

William Howell:

Yes. An interesting finding though that's embedded in there is that the moderation is not born for the most part of people both reading the Nation and the National Review, which cancel each other out. Rather, it's really about consulting these mainstream portals that they're brought together, not just by reference to their average scores, which consists of very divergent answers, but rather than in the main, they're looking at reasonably moderate media outlets. That said, a concern that I have that he's well aware of, and it's a hard issue is that again, his estimates of the ideology of the media outlets is born of the sharing behavior of the consumers. I mean, it's not a direct measure of the content itself, right?

He doesn't go in, and it's hard to see, how would you do that? Rather, he's inferring the ideology of the material on the basis of how people engage it. Then notice, he's using that to predict or to correlate it with another measure of how they engage it. It's not about sharing behavior, it's about clicks. He's aware of this circularity problem, and he attends to it somewhat, but it's still in play.

Wioletta Dziuda:

Can we unpack it a little bit? When I was thinking about this, I thought that two things might be happening, and tell me if I'm thinking about this in the wrong way. One thing might be that our sharing behavior is actually pretty well correlated with our reading behavior, in which case, he has the circularity problem. He's first measuring the slant, or the ideological position of the media, basically by looking at who reads this kind of media, whether Democrats or Republicans for this kind of media. Then he uses this measure to predict who actually reads this kind of media. In this case, we are not learning much from the analysis. At least we're not learning more from the analysis than what we would learn just by looking at "Do Democrats share different media than Republicans?"

William Howell:

I share that concern.

Anthony Fowler:

Oh, I don't know. I guess I don't know why we're supposing that. Why are we supposing that sharing behavior is the same as reading behavior?

Wioletta Dziuda:

I think there are two possibilities. Either we suppose that there is a strong link, and I would actually think that there's a strong link between what you read and what you share, or we suppose that, and I think this is where Andrew is going, we suppose that actually the link is weak. Now, I read something, and then I share things that are just aligned with my ideology. If the first assumption is true, it seems to me that we are not learning much from the analysis that Andrew later engages in. We could have stopped at just looking at "Do Democrats share different stories from different media than Republicans do?"

Anthony Fowler:

I mean, it could be the case that on average, there are differences between Democrats and Republicans in the media they share that allows us to infer that indeed CNN and MSNBC are to the left of Fox News and Breitbart, etc. On average, we're picking up these patterns, but nevertheless, for any individual partisan, they still get a relatively moderate diet that's a mix of liberal and conservative outlets. Yeah, so why can't both of those things be true at the same time?

Wioletta Dziuda:

You're positing that perhaps the correlation between what I share and what I read is not very strong? Am I understanding this correctly?

Anthony Fowler:

It could be, it could be either of the two things. It could be one that in fact, the sharing is more partisan than the reading. That's possible, and it's easy to think about why that might be the case if we're advertising to the world versus just deciding by ourselves what we're going to read. It could also be the case that maybe the sharing might not be super partisan either, but nevertheless, you can still look at average differences in sharing behavior and infer that CNN is clearly to the left of USA Today, even though plenty of Republican share articles from CNN and plenty of Democrat share articles from USA Today. You see my point, or no?

William Howell:

I think this is a real challenge. I think in one instance, we're inferring the ideology of the outlet on the basis of who shares it, and we're suggesting that you only share things that you believe, right? That's the motivating claim for that measurement strategy. Then in the latter, and since you're trying to look at overlap in media diets when it comes to consumption, the claim is something very different is happening. As an analysis, it only makes sense. There's only something to discover if our sharing habits systematically differ from our consumption habits. We share that which lines up perfectly, but we consume more broadly.

Anthony Fowler:

I don't agree with this. Let me try to explain why I don't agree with this. My interpretation is he's using the sharing just because he wants some measure of the partisan slant of each of these media outlets. There's different ways of doing it, and there are lots of papers that have different things. For newspapers that make endorsements, you could look at their endorsing behavior, you could look at what members of Congress cite when they make speeches in Congress, et cetera. There's lots of different things people have done. This is another way of looking at it is looking at who shares what on Facebook.

He just wants some way of measuring the average slant of these outlets, and he doesn't want to use the same data, the exact same data that he uses for his main analysis to do that, because he wants to avoid the circularity problem there that it might just be the case that if you've got a small sample, it could be the case that there's three Republicans who all read the same skiing site, and that's going to come out as looking like a very Republican site, when it just so happened that there were these three Republicans that all read that same site. He wants to avoid that statistical problem of using the same data he's going to analyze to code the alignment. He just wants some other objective dataset.

William Howell:

But it's still behavior predicting behavior.

Anthony Fowler:

Sure.

William Howell:

I mean, that is... I mean, if you really wanted to break it, you would have an outside independent measure that looks at the content of these things, which we understand why he can't do that, right? But it's still saying, "I'm going to use sharing behavior to explain clicking behavior and look at the relationship between these two things."

Anthony Fowler:

Then the next thing you want to ask is, "Do the actual slants that he gets from Facebook measure match with your expectation more or less? Does it look like he has a reasonably good proxy for the media slant of these outlets?" I think the answer is yes. I mean, if you look at the appendix of the paper, BuzzFeed is on the left and CNN is also pretty far to the left, and MSN is a little bit more to the middle, and the New York Post is a little bit more to the right, and then Fox News is really far to the right, and so forth. I mean, they fit with your expectation. If you just wanted to come up with a reasonably good proxy for the slant of each outlet, this seems fine to me.

Wioletta Dziuda:

Yes. It's not surprising I think that he found that his measure of slant aligns with some outside measures where people just read newspapers and assign the measure because it seems to be the case, and this was probably been a predictable that left-leaning people are sharing left-leaning news and right-leaning people are sharing right-leaning news, and moderates are sharing moderate news, but that's where we can stop. We can stop our analysis, and we already know that on average, we are sharing...

Anthony Fowler:

Well, on average, on average. He's not saying that Republicans never share from the New York Times, or the Democrats never share from the New York Post, but on average.

Wioletta Dziuda:

But on average, and then he could look at how frequently people share, the Republicans share the same news that Democrats, or like a piece from the same outlets, but we don't have to do the next step where we are looking at their browsing behavior because we already are looking at sharing behavior, which is basically, presumably very similar to... but we don't know.

Anthony Fowler:

We don't know.

Wioletta Dziuda:

Let me...

Anthony Fowler:

Yeah, we don't know.

Wioletta Dziuda:

Let me try to think about what would happen if actually this wasn't the case, which is what I'm understanding Andrew is arguing. He's arguing that sharing behavior is more partisan and is different than reading behavior. Initially, I thought, "Well, this is great, then he can do his analysis," but then in a sense, he's answering his question by making this assumption. He doesn't need any analysis, because if it's true that we just read...

Anthony Fowler:

I still don't understand why he has to make that assumption. He doesn't have to assume that sharing behavior is more partisan than reading behavior to use sharing is his measure.

Wioletta Dziuda:

He doesn't make an assumption that it's more partisan. He makes an assumption that it's uncorrelated, or it is not strongly correlated with your reading behavior.

Anthony Fowler:

I don't think he has to make that assumption either. I mean, of course, reading and sharing behavior are correlated with each other, but he doesn't want his results to just be tautological. He doesn't want it to be the case that it just so happens that Republicans read the skiing website more than Democrats and therefore, the skiing website ends up being coded as Republican and now all of a sudden, it looks like Republicans are really in this echo chamber of skiing.

Wioletta Dziuda:

No, because the same problem arises with sharing. If so it happens that only Democrats share the skiing website and not Republicans, he still calls the skiing website as left-leaning.

Anthony Fowler:

Right, but at least he has two independent samples. He has fresh samples of data. He has the sample of sharing on Facebook, and then he has different people where he can monitor their reading behavior, so at least you avoid that kind of...

William Howell:

There is that, there is that. There are two forms of separation, right? It's still not as far as, I think, either Wioletta or I would like, but it's two different samples, and it's sharing as opposed to clicking, but really to pin this down, what you'd really want to do is bring in an expert, who would just read all these news sites and say, "I'll tell you, I've read this stuff.-

Anthony Fowler:

Sure, sure.

William Howell:

... I can tell you how liberal or conservative these things are. I'm not going to infer the content of the material on the basis of the behavior." We have gone down a rabbit hole. As a little plea, I wish we did more of that, frankly, in our discipline. This coding thing happens in all kinds of spaces, wherein, we're so afraid of talking about how liberal or conservative legislators are, or how interest groups are, or newspapers are, that we make inferences about their ideological leanings on the basis of not just their behavior, but other people who engage them, instead of just owning the decision ourselves. That's an editorial aside.

Anthony Fowler:

I think that would be good. I think that would be fine to do it, but I don't see what he's currently done as being as problematic as you guys are pointing out. Have you ever wondered what goes on inside a black hole, or why time only moves in one direction, or what's really so weird about quantum mechanics? Then you should listen to Why This Universe. On this podcast, you'll hear about the strangest and most interesting ideas in physics, broken down by physicists Dan Hooper and Shalma Wegsman. If you want to learn about our universe from the quantum to the cosmic, you don't want to miss Why This Universe, part of the University of Chicago Podcast Network.

Wioletta Dziuda:

Can we talk a little bit about something that's really outside of his analysis, but we have to understand what conclusions to draw from his analysis? I would like to talk a little bit about how we think about how media consumption actually affects outcomes affects people's beliefs and voting behavior. Because somewhere in the background of his analysis is that, if we read different outlets, if I read Fox News and you read CNN, then we are going to end up with completely different worldview. I'm going to be persuaded to vote one way, you're going to persuade to vote another way. While only if we read the same source, then we would actually agree, or we would get somehow closer to each other's positions. How should we think about this?

Anthony Fowler:

Well, certainly it's not the case that if only we read the same news outlets, we would think the same things because, of course, people's beliefs depend not just on the news outlets they read, but also their underlying principles and values, et cetera. That part is surely not right, but I think underlying our concerns about echo chambers, there's some sense that people aren't getting as informed as they could or should be in some ideal world, right? If I only read CNN and I only get the facts that happened to be favorable for Democrats, and bad for Republicans, and I'm not getting the other facts and maybe I would cast a better more informed vote, for example, or just be a better citizen who could better engage in discourse with my fellow Americans if I got all the facts.

Wioletta Dziuda:

No, I don't know actually, I think that's one view. Another view might be that, I already have the facts, that basically all the outlets are giving me the basic facts. It's just I happen to dislike, let's say, Donald Trump, and I like when I read a newspaper where the newspaper uses negative adjectives about Donald Trump, instead of positive adjectives, and that's why I'm reading the left-leaning outlets. They are not affecting me at all. I know exactly what Donald Trump is doing, and people who are reading Fox News exactly know what Donald Trump is doing. We're just reading those different outfits because they give us different level of pleasure, but our views would be exactly the same way.

It would be exactly the same, we would feel exactly the same way about each other if we just started reading the outlets read by the other side.

William Howell:

Yeah, I mean, this is clearly outside of his analysis. How do people engage? What kind of updating occurs, and how does that updating lead to changes in behavior of one sort or another? On the one hand, we can see the problems of being a world in which liberals or conservatives are completely at odds with one another and consuming entirely different media outlets, going to entirely different media outlets. On the other hand, if all they do is consult the same one, it's not clear that that leads to an enhancement of democratic discourse, right? If all we do is watch CNN, that's all that everybody does, is that going to lead to better decisions or a richer politics? That's not clear at all.

Then we end up in this world in which we don't want too much separation, but we don't want them to be perfectly overlapped either. We want to leverage diversity. I don't know what we do with that.

Wioletta Dziuda:

I don't think that's an idea that he's hoping for. I think the idea would be that we do consume different outlets. You'll read something different than I do, but it doesn't correlate very strongly with our political leanings. I think the worry is that if it correlates very strongly with political leanings, then it somehow explains maybe our political leanings. Maybe it's the case that for some reason, I have become convinced that I dislike certain types of policies and certain politicians, and then I'm just stuck in the slope, even though new information arrives that actually would convince me I should vote differently, or have different positions on policy issues.

I think he doesn't hope that we're all going to read the same, we're all going to follow the same point on the distribution, but that it doesn't correlate strongly with...

William Howell:

I think it's in the background of the writings that motivate his empirical work, which is, we once lived in a world in which we all came home at 5:00 and watched Walter Cronkite in the news, and that provided a point of stability, an anchor...

Anthony Fowler:

Speak for yourself, Will, just to be clear.

William Howell:

Okay. Okay, I know, I'm old. Okay, all right. In any case, that that was healthy because we all would then show up at work the next day and talk about what we saw on the same show. Then you see this proliferation of cable news first, and then all kinds of online outlets, and the worry is everybody's scattering to the extremes and that that's a problem, but it's not clear that somehow... I don't know. I mean, maybe it is a problem, but the diversity of the profusion of different media outlets, on the one hand leaves, you could argue, to a greater diversity. It supports new ideas. It may enrich the polity in ways that Walter Cronkite rest his soul couldn't do alone. What's your bottom line on the paper?

Anthony Fowler:

I like the paper. I think this is a really valuable thing for somebody to have done in a really rigorous systematic way. I come away thinking that I learned something from this paper that I didn't know before, and I also come away with that reassuring view about the relative moderation and lack of hyper partisanship in the American electorate. I find the paper to be valuable just as a piece of social science and also useful in terms of understanding our current political climate and the sense in which maybe things are not as bad as you might think if you just read these alarmist accounts of echo chambers.

William Howell:

I agree with all of that. I'll say the challenges and concerns that we raised, I don't think Andy, none of them would catch him by surprise. I'm really looking forward to seeing what he produces next, but what he has done in this particular paper provides a real contribution relative to a world in which we were just relying upon survey, data, and experiments. He's actually looking at actual behavior of news consumption. The one thing I'll hold out, and I guess this is the study I'd like to read, and maybe it's already been done, is to get a sense of not just what people consume, but how they then interact with one another.

That a world in which you and I read radically different things and were in fact consuming qualitatively different kinds of news just doesn't strike me as problematic if we then come together and engage one another. If we come together and engage one another, I actually might prefer that world to a world in which we're both reading exactly the same thing, and then just reaffirming each other. There's actually a normative attraction to the bimodal distribution that he's arguing against, or the extent to which there is diversity in consumption patterns. If there are ways to figure out not just what are people consuming, but what are the networks that hold people together, and what are the flows of information across this network look like, that would be really interesting to see.

Then, of course, all of this is as a function of the conversations we have with the news we consume, we'd like to know more about how that leads people to update their beliefs and their actual behavior.

Wioletta Dziuda:

Yeah, I really think Andrew did a lot of hard work, and he was extremely careful telling us about what decisions he made, and he found this very positive finding that we are not as polarized in our media consumption as we are. I would really like to take it with me and be hopeful, but I think the jury's still out. I think there's a lot of limitations of his analysis that he's aware of that made me think that we can't just sit now on laurels and say that's it. I think the question is still out there.

William Howell:

Thanks for listening to Not Another Politics Podcast.

Wioletta Dziuda:

Our show is a podcast from the Harris School of Public Policy and is produced by Matt Hodapp. Thanks for listening.