Wednesday, July 6, 2011

Can you make your friends' friends fat by gaining weight?

Dave Johns at Slate has written a rather good article about an important debate in the social sciences. The article starts by describing some of the cooler networks research of James Fowler and Nicholas Christakis, which examines the "contagion" of personal behavior through our social networks. Fowler and Christakis, for example, note that if you're overweight, your friends will tend to be overweight, as will their friends. Same thing if you're divorced; you will tend to have divorced friends, and those friends will tend to have divorced friends. Same thing if you vote. Or if you're happy. Or lonely. And so on. Now, it's one thing to note these correlations, but Fowler and Christakis go on to suggest causality, such that your decision to lose weight can affect your friends, your friends' friends, and other people you don't even know. It's a cool concept, and it was initially embraced by the media (James even went on Colbert), but now there's a blowback within academic and journalistic circles, which Johns details.

I don't wish to weigh in on the statistical debate -- those with far better qualifications than mine are doing that just fine. But I think it worth mentioning that this whole debate is dealing with a topic that is absolutely essential to social network research and to a lot of other areas in the social sciences. The topic is homophily, which is simply a way of saying that birds of a feather tend to flock together. It is an easy enough concept to grasp, but it's very, very difficult to deal with it in actual research.

Let's say that your decision to lose weight actually affects those around you. That's not too hard to believe; we can be inspired by people we know. So there's an effect. But how do we measure that effect separate from people's tendency to hang out with other people who are like them? That is, people who are likely to try to lose weight will tend to be friends with other people who have the same interest. That's not influence, it's homophily. How do we measure the influence on top of the homophily?

Similarly, Democrats tend to be friends with other Democrats. That's not because people are deeply political (for the most part, they're not), and it's not that Democrats are convincing their friends to become Democrats, although that may happen on the margins. It's just that if you're a Democrat, you're probably hanging out in places where Democrats tend to hang out. You probably live in a city rather than a suburb (which is where the Republicans are hanging out with other Republicans), you probably live in a walkable neighborhood and frequent the types of bars and restaurants that exist in such neighborhoods, you probably hold the kind of job that Democrats tend to hold, etc. You're not intentionally selecting a Democratic lifestyle, nor are you necessarily trying to turn your friends Democratic. You just pick the lifestyle you're comfortable with, and it turns out that most others who pick that lifestyle tend to share your political beliefs. Lo and behold, the population from which you pick your friends tends to be filled with Democrats.

If that's the case, then how do we measure social influence? It's really not easy. But it's probably the biggest question we're dealing with right now in networks studies.

No comments: