How diverse is your social network? How diverse should it be?

Last November, cable TV network USA Network launched a campaign called Characters Unite, designed to call attention to issues surrounding racial, ethnic, religious and cultural diversity in the United States. Like Stephen Colbert, who doesn’t see race, “at USA Network, all we see are characters. Millions of smart, funny, quirky, heroic, shy, glamorous, fierce, stubborn characters, each one with a unique background and perspective.”

In that spirit, Characters Unite has launched a set of laudable initiatives focused on diversity. They’ve sponsored a civil rights documentary hosted by Tom Brokaw, a tour with the wonderful storytelling project The Moth, and a fellowship/awards program.

They’ve also launched an intriguing Facebook game, which is how I found out about the campaign. I’m doing some research on diversity and homogeneity in social networks, inspired in no small part by Andreas Wimmer and Kevin Lewis‘s remarkable work on racial homogeneity in a group of college students who used Facebook. I wondered whether anyone had tried to develop a tool that allowed someone to calculate the diversity or homogeneity of their Facebook friends and compare their findings to their friends. And so I stumbled on Character Unite Social Circle.

It’s unclear when USA Network introduced the game, but let’s just say that it hasn’t yet reached a Farmville level of cultural visibility or popularity – I “liked” the game yesterday and became the 113nd person to do so… and 24 hours later, I’m still #113. That’s too bad, because it’s a clever and provocative game that could provoke some worthwhile questions about how societies think about diversity and what we want social networks to do.

When you begin the game, you’re presented with eight of your Facebook friends, selected at random, and eight questions of the form: “Which friend is Muslim?” Your job is to match the question to one of those eight friends, or “spin again”, choosing another set of friends to find someone who fits the given diversity criterion. At the end of a five minute round, the game presents you with a diversity score, based on how many of questions you were able to answer with an exemplar from your friends.

I’ve played the game a couple of times and was impressed by the variety of questions the game puts forward – I’ve been asked to find friends from different regions of the US (the game is entirely US-centric, and will probably be pretty frustrating for non-US players), friends of different ethnicity, nationality, religion, sexual and political orientation. And despite having over a thousand Facebook friends from around the world, I found myself spinning again and again, wondering when I’d find one of my Hispanic friends so they could serve as the answer to question 7.

You can cheat, of course – despite their best efforts at ad targeting, Facebook doesn’t actually know the religion and sexual orientation of all your friends, so there’s no way they can tell that your friend Tom MacMaster isn’t actually a lesbian muslim. But using the honor system doesn’t change the uncomfortable feeling I had wondering whether one of my few Hindu Facebook friends would turn up in time to improve my diversity score.

Graphic from Facebook’s Data team on maintained relationships

The average Facebook user has 120 friends, and is in close touch with only a subset of them. An analysis from Facebook’s data team (including Lee Byron, Tom Lento, Cameron Marlow and Itamar Rosenn) looks at three ways of distinguishing closer friendships from someone from whom I’ve simply accepted a friend request. Friendships based on reciprocal communication – you post to my wall and I post to yours, for instance – are the most rare. A male Facebook user with 150 friends is in reciprocal communication with 5 on average, a female user with 7. (In general, female Facebook users maintain more relationships than male users do.) One way communication (I post on your wall, but you don’t reciprocate) is more common, and “maintained relationships” – defined as relationships where you’ve clicked on a story in your News feed or visited the other person’s profile twice – are the most common. Still, they’re a small portion of Facebook friendships – a male Facebooker with 500 friends maintains relationships with 39 friends, a female Facebooker with 47.

(There are other ways to determine tie strength on Facebook. The Wimmer and Lewis paper, as well as a broader set of research carried out by my friend Jason Kauffman and his colleagues Marco Gonzalez and Nicholas Christakis with Wimmer and Lewis, uses presence in the same photo, as determined by user-provided metadata, as evidence of an offline friendship manifesting online. Photo co-presence is likely a less common phenomenon than maintained relationship and might be a good indicator of stronger ties.)

The difference between friendships, maintained friendships and reciprocally maintained friendships might help explain one aspect of the Social Circle game I found particularly challenging. I accept friend requests from anyone I can recall meeting in person – it’s a low bar, but manages to keep my friends feed slightly less spammy than it otherwise might be. As a result, there’s lots of people who appear as “friends” who I’ve met once or twice and couldn’t conceivably tell you their religious beliefs or sexual orientation. I found myself wondering whether people with fewer Facebook friends found the process easier because they knew their friends better, though the data team’s paper suggests that there are lots of Facebook friendships that don’t extend beyond a superficial knowledge of the other party. (At least one Facebook friend has included me in a group called “real friends”, evidently distinguishing us from the people she feels obligated to list as friends. For all I know, this is common practice, and she’s the only one out of a thousand Facebook friends to consider me a real friend… )

The objection to the Social Circle game is that it’s trivializing – it encourages you to think of your friends as fulfilling a category requirement rather than considering the nuances of your relationship with them. The mechanism the game uses to encourage you to share your results seems particularly vulnerable to this critique – it asks you to share the game with the friends you used to score points. I can only imagine the conversations that could result from this: “Hey Mohamed, thanks for being my token Muslim friend and scoring me some points!”

But there’s a certain utility to this exercise. One of the hopes of social networks is that they may expose us to opinions and perspectives we might not encounter in our daily existence. Cameron Marlow of Facebook’s research team suggested an interesting thought experiment when we met a few weeks ago. “Are you more likely to hear about an event in a distant part of the world through your social network or through the traditional media?” The obvious answer is that you’re far more likely to hear through media – the job of a newspaper is to present you with news from different parts of the world. But as broadcast media becomes less influential and choice-based digital media more common, it’s quite possible that you might miss an important story because the headline didn’t catch your interest, or because you never searched for information on the topic.

Cameron argues that, over time, we may be more likely to be alerted to these stories through social media than through the subset of professional media we choose to encounter. His argument, in part, is that news stories we receive through our social networks come with an added piece of information – this event happened, and this friend thinks it was important enough to share. Knowing that a friend cares about protests in Syria or earthquake recovery in Japan might help solve “the caring problem“, the difficulty many of us have paying attention to news in parts of the world we know little about and have little connection to.

I’m far from persuaded by Cameron’s hypothesis. Despite the diversity of my social networks on Facebook and Twitter, the 20 minutes I spend listening to NPR on the way to my office are the ones most likely to alert me to stories I otherwise would have missed. But he’s not wrong to point out the value of seeing what your friends care about. The fatal shooting of Oscar Grant received some, but not overwhelming, media attention in the Northeast US. I started paying attention to the story because African-American friends and friends in Oakland were following the case closely – the signal that they cared about the case was an indicator to me that I should pay attention as well.

The utility of your social network in providing this information is a function of its diversity. If you don’t have many Muslim friends on Facebook, you’re less likely to discover that one or more friends is paying close attention to Herman Cain’s (ludicrous, absurd, shameful) remarks about the “dangers” of including Muslim-Americans in a potential presidential cabinet. One possible response to my experience of searching for a Hispanic friend to score points in the game would be to look for more Hispanic friends so I have a better understanding of the issues and concerns of that community. That’s what led Eli Pariser to start adding Republican friends to his Facebook account – the promise that the news he encountered through his feed would have more political diversity. In the process, Pariser discovered that the diversity of your friends isn’t the only factor governing the utility of your social network in presenting diverse information – Facebook’s algorithms emphasized news from the friends he paid most attention to, and Pariser found himself as trapped in a “filter bubble” as before he’d made the effort.

Most people don’t use Facebook the way Pariser did, consciously adjusting his friends list to diversify his information inputs. And I can imagine some people reacting to the Social Circle game with some frustration: “I’m supposed to have Muslim, Hindu, Jewish, Buddhist, gay, straight, southern, northeastern, foreign-born, Republican, independent and Democrat friends? All in the space of 120 people?” This is compounded by the scolding the game gives you if you’re insufficiently diverse: after scoring a 75%, the game told me I’d done well, but still had some cultural divides to bridge, and encouraged me to visit their website for fifty tips, evidently to help me befriend the Asian-American atheist who’d prevented me from scoring 100%.

Graph of racial “saturation” on Facebook, by Cameron Marlow

What’s a reasonable amount of diversity to expect in one’s Facebook friends? Historically, Black and Latino users were underrepresented on Facebook in proportion to their distribution in the American population, and Asian-American and White users were over-represented. (The blog post where Cameron Marlow shares these results includes a helpful footnote explaining the usage of US census categories to organize this data, troublesome as they may be.) As the community has grown, the representation of different races is nearing parity on the service. This helps counter an argument that I don’t know many Latinos on Facebook because there aren’t many Latinos in the userbase – that’s far less true now than it was in 2006. But as a guy who grew up in upstate New York, went to high school in Connecticut and college in western MA, there are fewer Latinos in my social circle than there might be had I grown up in Texas.

Wimmer and Lewis’s paper looks at the friendships between students at an elite college, using data from Facebook to question existing sociological theories about friendship and race. There’s a large body of research in sociology that documents the phenomenon of homophily, the tendency of people to form friendships with people who share a common race, ethnicity, gender, religion, socioeconomic status, etc. Acknowledging that homophily is real and has effects on friendship formation, Wimmer and Lewis examined other factors that help explain friendship patterns: availability, propinquity and balancing effects. Availability is the simple point that you’re unlikely to have many Latino friends if there were few Latinos in the population of possible friends. Propinquity is the idea that you’re more likely to form a friendship if you participate in the same activities – if the people who participate in the activity of your choice are predominantly of one race, you’re more likely to have friends of that race. Balancing effects are effects that emerge from the social “rules” of friendship – if you consider me a friend, it’s likely that I’ll reciprocate the relationship, if only because it’s awkward and uncomfortable otherwise. If I’m friends with you and a third person, there’s an increased chance that you’ll befriend that third person through a process called “triadic closure” – it’s more comfortable, socially, for the three of us to be friends than for me to maintain two, separate and unconnected friendships.

When Wimmer and Lewis look at the friendships at the college they study, they see lots of racial homogeneity in friendship – black students are more likely to be friends with other black students than with white students, Asians with Asians, etc., though the effect is quite small in the population of white students. When they build more complex models that consider factors of availability, balance and propinquity as well as homophily, they’re able to draw some other conclusions. The phenomenon of Asian honophily disappears – the effect comes from stronger effects of ethnic homophily. It’s not that Korean students are more likely to befriend Indian students than African American students… they’re massively more likely to befriend other Korean students, and the Korean/Korean ties end up looking like Asian-American homophily when you consider the broad racial categories of the US census. These effects can get amplified by “balancing” effects – if you’re friendly with a couple of Korean students, reciprocity and triadic closure may quickly alter your social network to the point where you know a lot of Koreans instead of one or two. (These balancing effects are more significant in explaining friendship formation in their data than any other factors.)

But other forms of common ground besides race and ethnicity govern friendship formation as well. Students who come from Hawaii or from Illinois (oddly enough) are extremely likely to for friendships with each other, regardless of race. So are students who study microbiology or applied math. Socioeconomic status matters a great deal – students who’ve attended an elite prep school show a very pronounced tendency to befriend each other.

What’s most interesting to me is that the single factor (beyond the tendency towards reciprocity) that’s most important in explaining friendship formation in Wimmer and Lewis’s model is the dorm you live in. In the population they study, students have been randomly assigned to dorms and assigned roommates in a way that makes it unlikely that students of the same race will be roommates. Despite this social engineering, students who live in the same dorm are likely to be friends and students who are roommates are very likely to become friends.

In other words, who we are friends with is a product of our life experience, as well as our racial, ethnic, religious and class identity. If your Facebook friends are highly diverse, that’s not evidence of moral rectitude or political correctness as much as it the result of a life history that included opportunities to befriend a diverse range of people.

That should make you feel better if you scored a 25% on Social Circle. But it shouldn’t let you off the hook. If Marlow and Pariser are right, who you know helps govern what you know. The holes in your social network that Characters Unite’s game helps you diagnose may point to perspectives and points of view you’re missing out on.

If you get the chance to play Social Circle, I hope you’ll let me know what you thought. And if you’re one of the folks at USA who commissioned or worked on the game, I’d love to hear more about the thinking behind the game and how it’s been received thus far.