Who is your best friend’s best friend?
Previously I’ve written about theories of friendship in dolphins and macaques, which suggest that the function of friends among these species is alliances. Friends, according to this view, are useful for support in conflicts with others. In alliance models, the value of a friends depends on their other friends. If you like me, that’s great. But if you like everyone else around more than you like me, then when conflicts arise, you’re going to be my enemy, not my friend. Alliance models, for this reason, are committed to the prediction that I ought to like you more to the extent you like me compared to how much you like others. This leads to tightly coupled best friends pairs; the most valuable friend to me is the person who likes me better than everyone else.
In the context of human friendship, alliance models aren’t favored. Some theories don’t identify any function – propinquity models and homophily models are like this, identifying physical distance and similarity as the key variable – but among those that do, the most common is probably exchange models, which suggest that friends are for reaping reciprocal benefits through exchange. These models would seem* to predict that I should like my friends depending on the costs and benefits, liking the person who benefits me most as my best friend.
A problem with exchange models is that people deny that they keep track of costs and benefits among close friends, but of course, it’s possible that they are despite their denials.
In any case, a chance to put the alliance model of human friendship at risk presented itself, as reported in a paper published last week in the journal Perspectives on Psychological Science. (Disclosure: I’m an author on the paper, and Peter DeScioli, my former student is the first author, and, not only that, but the last author is my cousin, David Liben-Nowell.)
So, having said that, the paper – a collaboration between psychologists and computer scientists – uses data from MySpace to look at friendship networks. At MySpace, people on the site not only listed their friends, but ranked them, from best friend at the top to one’s Nth friend, where N varied.
The computer scientists were able to gather these data – about 11 million profiles – and we asked the following question. Suppose that you’re interested in Zak, and you want to try to predict who Zak lists as his best friend from other people’s properties. Suppose that Alex says that Zak is his best friend, and Bart says that Zak is his second best friend. How likely is it that Zak ranks Alex rather than Bart as his best friend, as the alliance model requires?
Now suppose you do the same analysis, but with proximity. Suppose that we’re again interested in predicting Zak’s ordering, and we know that the Alex lives closer to Zak than Bart does. How likely is it that Zak ranks Alex rather than Bart?
Predictors of first- versus kth-ranked friends.
We find that the first variable – how much others like Zak – does a better job than the second – who lives closer to Zak. If other people list you as their best friend, it’s likely that you list them as yours. (See the accompanying Figure. Additional information in the Supporting Information)
And now some editorializing.
You probably think that this is an intuitive result, that “best friendships” are mutual. If you were persuaded by the analysis above, then a peculiar thing about this is that an obvious feature of friendship is not well accounted for by the principle models in the area. Having said that, my experience has been that people seem to think that the result is predicted by whatever model of friendship they happen to favor. I have been kicking around the idea that people are subject to what I might call an “intuition heuristic,” such that if a theory is intuitive (e.g., people tend to be friends with people who are nearby), the theory is credited with predicting findings that are also intuitive (best friendships are mutual), independent of the relationship between theory and data. As a corollary to this – and to return to that asterisk above – I think there’s another heuristic that I seem to come across from time to time, which I’ll call “the theory of reciprocal altruism explains any finding about human prosocial behavior” heuristic, which probably needs a pithier name. Don’t get me wrong; I’m a fan of the theory. But my experience is that there is a perception that it predicts and explains a wide array of phenomena that it really doesn’t.
But that’s more or less beside the point. We thought the dataset was interesting, and if you thought that MySpace wasn’t good for anything, well… I wouldn’t say that…