The Partisan Internet and the Wider World

Is the internet making us more partisan?

This is one of the most persistent debates in the study of cyberspace. Cass Sunstein, legal scholar, author and now Administrator of the White House Office of Information and Regulatory Affairs, wrote Republic.com in 2002, which made the case that the Internet would lead us to cocoon ourselves in information that confirmed our suspicions, opinions and beliefs. He’s developed the argument further in a Republic.com 2.0, Infotopia (my review here) and Going to Extremes.

There are (at least) three pieces to Sunstein’s argument. First, he asserts that the internet makes it easier to surround yourself with comfortable information – and predicts that Americans are headed rapidly towards a cocooned future. Second, he posits that deliberation with like-minded individuals tends to drive us towards more extreme positions than we originally held. Third, he suggests that we could make small changes in our media environment to “nudge” ourselves towards more diverse, inclusive views of the world. (Some parts of his argument are better developed than others – I think his case that like-minded deliberation tends to polarize us is quite convincing, while I haven’t found his nudges all that compelling.)

Is Sunstein right that we’re becoming more isolated? There’s a whole cottage industry in academia dedicated to confirming or challenging his assertions. Adamic and Glance’s “Divided They Blog” saw little interlinking between bloggers on the left and right as they discussed the 2004 US presidential elections. Eszter Hargittai looked closely at the links between bloggers in “Cross-ideological discussions among conservative and liberal bloggers” and discovered that when bloggers link across ideological lines, it’s often to create “straw man” versions of their arguments. Henry Farrell, Eric Lawrence and John Sides present evidence that suggests that blog readers are as politically polarized as members of Congress, a group that’s far more partisan than an average group of Americans. In the other corner, John Horrigan, Paul Resnick and Kelly Garrett found evidence that internet users had broader exposure to political arguments than non-users. (I offer my critique of the Horrigan, Resnick and Garrett paper here, relying on the Farrell et al paper for counterarguments. Dr. Garrett was good enough to engage with me in the comment thread, and her points are helpful in understanding the issues at hand and the difficulty of answering these questions.)

So here’s the latest paper designed to settle the debate… or perhaps throw more fuel on the fire. In “Ideological Segregation Online and Offline“, a working paper for the National Bureau of Economics Research from Matthew Gentzkow and Jesse Shapiro (both of UChicago’s Booth School of Business”, the authors make the case that “ideological segregation on the Internet is low in absolute terms” and that there is no evidence that this segregation is increasing.

The paper, published about a month ago, attracted widespread attention in part because New York Times columnist David Brooks used the paper to present a downright cyberutopian take on the internet and ideological segregation: “If this study is correct, the Internet will not produce a cocooned public square, but a free-wheeling multilayered Mad Max public square. The study also suggests that if there is increased polarization (and there is), it’s probably not the Internet that’s causing it.”

Slate magazine took the Brooks column and declared “we aren’t nearly as isolated as we think”… and built a tool to prove it. Visit their site, click a button and they’ll access your browser cookies to see what sites you’ve looked at and calculate your ideological isolation index. (This tells us at least as much about the sorry state of web security and privacy as it actually does about partisanship.)

I’m very interested in this issue, so I read the Gentzkow and Shapiro paper. And then I read it again, because I wasn’t sure I was reading the same paper that Brooks and others were reading.

Here’s what Gentzkow and Shapiro did – working with internet marketing firm comScore, they looked at the behavior of comScore panelists during 2009, when they visited sites that the company classifies as “General News” or “Politics”. That gave them data on the comparative popularity of each site. They used another set of data from comScore where panelists are asked their political preferences (ranging from very liberal to very conservative) and were able to calculate the ideological distribution for readers of 119 large sites in the news and politics set. Using this, they are able to estimate that 98% of viewers of rushlimbaugh.com identify as conservative, while only 19% of viewers of moveon.org do.

While that seems to support Sunstein’s contention of a polarized media, the authors discover that most readers spend time on a variety of sites, and spend a lot of time on sites that have a less polarized audience – general news sites like aolnews.com, Yahoo News, MSNBC or CNN. When you look at the span of sites the average conservative user views, the audience of those sites is 60.6% conservative – that’s an audience similar to the audience for usatoday.com. Across the span of sites the average liberal visits, the audience for those sites is 53.1% conservative. The “isolation index” – the difference between those figures – is 7.5, a figure the authors tell us is “small in absolute terms”.

It’s that characterization that I find confusing: small compared to what? The authors go on to calculate isolation indices for other media, using data from the Mediamark Research and Intelligence Survey, which asks 51,000 respondents about their media consumption and political orientation. They calculate similar isolation indices for other media, using data from the 2007 and 2008 Mediamark Research and Intelligence Survey of the American Consumer – again, they measure the percentage of conservative viewership for CNN, Fox News, MSNBC, CNBC and Bloomberg cable news, then calculate the mean audiences for the sites that liberals view versus those that conservatives view. The isolation index is 3.3 in cable news, implying that liberals and conservatives have pretty similar media diets… and that ideological isolation is at least twice as high online (“in absolute term”, to use the language of the authors) as in cable news, which many commentators have decried for its partisan tone.

Indeed, the internet shows more ideological isolation than any other media the study considers – local newspapers, national magazines, broadcast television, cable television – other than a set of national newspapers, which includes only three: USA Today, the New York Times and the Wall Street Journal. Ideological isolation between readers of those three papers is 10.4, higher than the figure the authors calculate for visitors to internet news sites. (Ideological isolation online would be higher than in national newspapers if we exclude the largest two sites – AOL News and Yahoo News – from the set. Remove these very popular sites and ideological isolation online expands to 11.3. You might choose to remove these because they – as well as MSN – are often set as the default homepages for browsers provided by internet service providers. If we’re trying to measure the isolation of media people choose, we might remove this from the set… and the authors do so in section 7.3 to calculate this higher figure.)

So how do Gentzkow and Shapiro assert that ideological segregation is “small in absolute terms”? Well, it is… in comparison to how ideologically isolated people perceive themselves to be in their day to day interactions. They use the “Number Known” module of the 2006 General Social Survey to estimate the ideological diversity of people’s neighbors co-workers, acquaintances through social organizations, family and “people you trust”. Basically, they’re asking a respondent to characterize how many liberals or conservative she knows in these different groups and calculating ideological isolation from these perceptions.

Turns out that we perceive ourselves to be surrounded by folks who agree with us. In particular, we report a high ideological isolation index with our trusted friends (30.3), families (24.3), neighborhoods (18.7) and coworkers (16.8). Of course, it’s not clear that what we’re perceiving is actually reality – using data from the MRI survey, the authors calculate ideological isolation by county (5.9, lower than the calculated online isolation) and zip code (9.4, higher than online ideological isolation, lower than the national newspaper figure.) The disparity between neighborhoods and zipcodes illustrates the difficulty of using these two different methodologies – people perceive their neighbors as being quite ideologically similar (18.7 ideological isolation), but examining folks in their zipcode – as measured by their actual political ideology – leads to a low isolation index (5.9) which implies fairly high ideological diversity.

What these results mean is that the authors are able to present an attractive chart – Figure 2 – where the internet is smack dab in the middle of a distribution which moves from low ideological isolation (most media), moderate isolation (actual geographic distribution of political opinion, internet and national newspapers) and high isolation (percieved ideology of friends, family and coworkers). Does this mean that the internet isn’t especially ideologically isolated. Or are we comparing apples to oranges?

I think the comparison between ideological isolation in media and in face to face encounters is more like comparing apples and hedgehogs. They’re thoroughly different types of interactions and we should have very different expectations for diversity and ideological isolation in each set. The media I consume damn well better be more diverse than the community I live in. That’s what media is supposed to do – give me a broader view than I’m able to get from friends, family and coworkers. It’s okay that there aren’t any Thai people in my rural American town of 3,000, but if there are no Thai protests in my newspaper, there’s something wrong. Whether you live in Lanesboro, Los Angeles or Lagos, you need news media that’s more diverse – in terms of geography, ideology, national origin, religion, language – than your immediate surroundings so you can understand an interconnected world.

Media provides another critical social function beyond informing us about a diverse range of people and issues – it gives us common ground on which we can have conversations. It’s become commonplace in media criticism to remember a past in which Americans watched one of three broadcast networks and where we could expect to discuss what we saw with coworkers the next day. With hundreds of cable channels, we lose that cultural common ground, they posit. The results here seem to suggest that we’re even more fragmented and sorted in online spaces than we are as cable viewers. If you’re talking with someone at the water cooler about the article you both read on the BBC’s website, you can be pretty assured you’re both liberal (the site’s visitors are 72% liberal)… and given that BBC’s site has a daily US audience of 472,000 – reported by comScore – you’re not too likely to find a third coworker who’s read the same story.

Gentzkow and Shapiro do a thorough job of disassembling a straw man – a straw Sunstein, to be specific. They are able to demonstrate that Sunstein’s strongest assertions – “people restrict themselves to their own point of view – liberals watching and reading mostly or only liberals; moderates, moderates; conservatives, conservatives; Neo-Nazis, Neo-Nazis” – are demonstrably false. The right reads Bill O’Reilly, the left reads Think Progress, but we all check Yahoo News. This isn’t necessarily because we’re seeking a common ground for understanding – it might just be because we’re all looking for sports scores.

The authors write: “One could imagine a news site that presented the Neo-Nazi perspective on all of the day’s events: first hand Neo-Nazi reports from a hurricane in Florida, a Neo-Nazi perspective on the Superbowl, and so forth. But such a site does not exist, to our knowledge, likely because the Neo-Nazi audience is too small to make such an investment worthwhile, and the preferences of Neo-Nazis for many stories are not actually all that different from those of the average consumer.”

So, yes – Sunstein overstates his case when he predicts that we’ll reach a future where we’re fully ideologically cocooned – the broad reach of general interest news aggregators ensures that they’ll have an audience with a diverse range of conservative and liberal readers. But I’d argue that Sunstein is right in spirit, especially as we move down the long tail of attention distribution. The isolation index in news sites ranked 1-10 and 11-20 by audience is 6.2 and 5.8 respectively. But the index for sites ranked 50 or lower is 21.3, implying very large differences between liberal and conservative audiences. As the authors note: “The most conservative sites are billoreilly.com, rushlimbaugh.com, and glennbeck.com, all personal sites of conservative radio or television hosts. We estimate these sites’ visitors to be more than 98 percent conservative.”

In other words, there are the sort of highly partisan websites Sunstein worries about. It’s just that their audiences are much smaller than the audience of aggregation sites… in part because Bill O’Reilly and Rachel Maddow don’t yet cover all the news we’d like to encounter. The good news is that some of the people who access these highly partisan sites also read quite widely – the authors discovered that readers of Stormfront (a white supremicist website) were twice as likely to read The New York Times as the average user in their set. That, in turn, raises the question: “Are the Stormfront readers engaging with the New York Times content in the hopes of broadening their worldview, or are they merely keeping a watchful eye on the enemy?”

So here’s my summary of the Gentzkow and Shapiro paper… a slightly different set of conclusions than those put forward by Brooks and others:

– Internet news sites are highly ideologically isolated when compared to other sets of media – local newspapers, broadcast and cable television and magazines. The only media studied that are more ideologically isolated than internet news sites are the major national newspapers, which are widely perceived to be partisan.

– While a small number of popular internet sites have low ideological isolation, smaller internet sites tend to be highly polarized, showing much higher ideological isolation than any other media in the study.

– As the state, ideological isolation online doesn’t appear to be increasing over time… but that’s likely because new internet users are more likely to gravitate to the large, centrist sites like Yahoo and AOL News. Early adopters of the internet included many extremists, who saw the possibility of connecting with the like-minded online. Whether this pattern towards the center continues now that internet adoption is slowing is unclear.

I’ll offer two possible inferences that likely differ from Brooks’s position as well:

– Internet news sites and national newspapers show roughly as high ideological segregation as real-world communities (zip code, county statistics) do. This is disturbing, as we expect media to provide us with a common ground for conversations and to provide us with information from outside of our geographic, ideological, national and other biases. Instead, this finding suggests that we’re sorting ourselves towards partisan media much as Bill Bishop asserts we’re sorting ourselves in physical space… remarkable, as it’s much easier visit a website representing a different community than it is to move your household to another zipcode.

– As media becomes more personalized – through voting sites like Digg and Reddit, as well as through less formal mechanisms, like being forwarded links from friends via email or Facebook – we should expect ideological isolation to increase. That’s because we’re likely to have higher degrees of ideological isolation within our various peer groups than in society as a whole, and our friends are likely to display a selection bias in the news they pass to us.

Perhaps this is why I found Paul Carr’s reaction to this paper so satisfying. Carr reacts to the optimistic framing of the Gentzkow and Shapiro paper – and to the larger idea of a post-racial United States – somewhat bluntly:

The trend, according to champions of Internet diversity, is clear: the Internet makes us less fearful of people with different ideologies, backgrounds or skin colours to our own. And this, of course, is A Good Thing. In just a generation, laws like that passed in Arizona or opinions like that expressed to Prime Minister Brown in Rochdale will be a thing of the past and, thanks to social media, we’ll all live together in perfect harmony. Ebony and Ivory, etc etc etc. In fact, as far as I can tell, there’s just one problem with that vision of Christmas yet to come…

It’s total horseshit.

His argument is in part from anecdote – his surprise at a hotel worker’s strike in San Francisco, which he hadn’t heard about as it was discussed mostly in Spanish-language media. But his strongest example has some hard data behind it. Carr observes, “Twitter feels like one of the whitest sites in the world to me: full as it is with self-important middle-class hipster kids retweeting New York Times stories and the fact that they’re having sushi for lunch.” But a recent report from Edison Research, who surveyed US consumers about their awareness and use of Twitter (landline and mobile phone sampling, 1,753 interviews), found that 25% of American twitter users are African-American… roughly twice the percentage of African-Americans found in the US population. The survey found heavy Hispanic usage as well, and noted that only 51% of survey respondents identified as “white”.

This was a surprise to Carr because his Twitter – the people he follows – are self-important white kids. Watch the trending topics and you’ll end up with a different picture – the most popular hashtags are often started by African American teens, using the space in a very different, more conversational manner than many white users. danah boyd talked about this phenomenon in her talk at SXSW 2010 about social media and inclusion: “As one black user explained to me, joking around on Twitter is a lot like an extension of ‘yo mamma’ culture; it’s a place to blow off steam with humor that is sometimes vulgar in nature.”

If that’s not your experience of Twitter, it’s a function of who you follow. Like most of us, Carr’s network includes a lot of people he shares demographic similarities with. (The phenomenon is called “homophily”, and I’ve already written too much about it. Here’s one of the key papers in the field, if you’re interested in reviewing the academic literature…) It wouldn’t be surprising to discover that the majority of the people he follows share his political views. And as tools like Twitter become more important for sharing news, it’s more likely that Carr will get little news from the African-American and Latino communities and more that’s consonant with his politics.

Gentzkow and Shapiro’s work suggests that we shouldn’t worry too much about Twitter as homophily trap. After all, many more internet users get their information from fairly centrist news aggregation sites than from Twitter or Facebook. But that may be changing. Jane Buckingham, the founder of media research organization, the Intelligence Group, quoted a college student she interviewed as saying, “If the news is that important, it will find me.” Brian Stelter, writing in the New York Times two years back, saw that statement as descriptive of a generation of media users who rely more on “social filters” than on mainstream media. If Stelter’s right, then the reassurance Gentzkow and Shapiro offer – for what it’s worth – may not be valid for very long.

---

I have friends on the right and the left of the US political spectrum who are skeptical of Sunstein’s arguments about the internet and partisanship. What those friends have in common is a passionate and abiding love of the internet. I share that love, and I believe that the internet, overall, has made my life (and that of many of the people I know), more global, connected and diverse than it would have been in a pre-digital age.

The reason to take serious, critical looks at scholarship about the internet and diversity – ideological and otherwise – is that we should be passionately committed to ensuring that the internet helps make the world wider, not smaller. If research shows evidence of increasing isolation online, it’s a call to action to make sure we’re building tools and infrastructures that make it possible to connect beyond our existing circles of friends and to ensure that we’re getting the information we need to live in this wide and connected world.