Ideatrotter

Harvard researchers Kevin Lewis, Marco Gonzalez and Jason Kaufman published a paper called Social selection and peer influence in an online social network, and it seems to suggest that peers have a smaller influence on what we like than people may think:

Bob Brown via NetworkWorld

Using the Facebook data from a group of more than a thousand college students at one college, the researchers found that students whose music and movie tastes were similar were more likely to become friends or influence the formation of new friends, though book tastes were less of a factor in either case (maybe it would be different for older people, once the book club years kick in?).  The fact that music and movies tend to be more social activities probably has a bearing on their influence on friendships, the researchers write.  They found tastes in classical and jazz music were more likely to get passed along through friendships than tastes in indie/alternative music, where the aficionado of such music might be the sort to be the token indie/alt music lover in a group.

Devin Coldewey doesn’t buy this at all:

[…] the study is also clearly flawed in ways that those versed in social graphs are likely to easily perceive. Pulling useful data from social networks is like catching lightning in a bottle, and I wonder whether the findings may in fact be, as the study attempts to avoid, “a spurious consequence of alternative social processes.”

The central source of data for the study, in fact, doesn’t strike me as solid. Tracking the interests of college kids is a sketchy endeavor in and of itself, but tracking it via their Facebook favorites (i.e. what shows on your profile, not what you post about or share) seems unreliable.

After all, not only does everyone use the network in their own way, but the network itself has changed. Putting Wilco in your favorites is a different act from liking Wilco’s Facebook page, their official band site, or posting their latest video. Gauging someone’s interest in a movie or band by the favorites factor alone is inadequate. Their findings are essentially that taste doesn’t diffuse the way you might expect. But while the data support this, nothing supports the data.

Flattening huge sets of data and removing potentially conflative or distracting connections (“disentangling,” to use the researchers’ well-chosen word) is the bane of social research, and with a limited window on a huge field of data, like that these researchers had, it’s especially hard.

Who among these people was a supernode? What were their Twitter counts? What was the most common unit of interest? How many total posts, how many total favorite changes, how many total friends? The process of disentanglement only gets harder and harder, and the amount of indispensable data grows. The researchers have used advanced statistical techniques, but the data they were interpreting doesn’t seem to be at all complete.

The study does establish something that I think we perhaps understand is true already: you befriend people because of your overlaps in taste, but it’s rare that your existing friends change the tastes you already have. This is as much true out in the “real” world as it is online.

Coldewey is a bit off kilter with his general pronouncements about the difficulty of pulling factual information from social netwroks: they have been shown in many studies, for decades, to be immensely important predictors of health, happiness, trust, and a long list of other factors.

Still, I have to agree that since the results are so counterintuitive, it might be important to segregate friends from influencers. My hunch is that influence follows the power laws, and so unless you find the people that have super levels of influence — and see what strange gravity disturbances they cause — you might not think that there is anything going on at all.

(via stoweboyd)