Sociologist Duncan Watts (whom I interviewed a couple years back) has an article in the New York Times Magazine on the idea of “cumulative advantage” in culture markets—a kind of aesthetic version of Schelling points or stock bubbles that, says Watts, explains why it’s so hard to predict what will become the “next big thing” in music, film, or art.

Watts’ findings come from the Music Lab experiment he runs at Columbia. The idea was to divide volunteers into virtual subgroups and expose them to a bunch of songs by bands they weren’t likely to have heard of before. The volunteers listen to as many tracks as they like, rank them, and have an option to download those they liked. A control group just got the music; other subgroups, however, got to see how many others in their group had downloaded each track, with the songs listed in order from the most to fewest downloads. Unsurprisingly, this created a significant lock-in effect, since people naturally were more apt to sampled and download the tracks ranked highly by initial users. The results weren’t entirely random, but the “hits” in the “social” subgroups tended to be relatively bigger, and the rankings varied wildly from subgroup to subgroup, with only a weak relationship to the putatively more objective ranking of the control group. (I’m assuming the order of the listing in the control group was randomized for each user, or this wouldn’t be terribly significant—that ranking too would be biased by what users saw first.)

Now, a couple quibbles with the experimental design. First, this is patently unlike actual cultural consumption in a variety of ways: Though “social” in some loose sense, it’s obviously not communicative. Users aren’t talking to each other about why they liked what they liked, you don’t have people specializing in music criticism, and so on. People with especially intense preferences aren’t evangelizing their friends—a tepid download counts the same as an enthusiastic one, and within levels of enthusiasm, there’s no way of knowing which voters share your tastes. That might do something to magnify the apparent volatility of user preference. On the other hand, it also means we’ve left out a major source of this kind of snowball effect in cultural tastes, which is the desire to be listening to the same things as your peer group so you’ve all got a kind of common reference point. Also, I’d like to see the results adjusted for the number of times various tracks were listened to by each group, to sort out whether the social effect was primarily on exposure (with subjects allocating scarce sampling time to tracks others had preferred) or if it actually affected people’s opinions of the tracks they did listen to.

Perhaps more importantly, my impression after listening to 15 or so tracks was that these were overwhelmingly pretty generic, mediocre pop or hard rock tunes. I only bothered to download one of them (by a defunct Indiana outfit called Cape Renewal), and I think that was mostly because it reminded me slightly of a much better Weakerthans song. The thing is, the more middling the options, the greater the chance that people with relatively strong and distinct musical preferences aren’t going to be offering much input into the experiment. And that seems bound to increase the socially-determined “randomness” of the results.

All that said, the result is intuitively pretty plausible. Just to pick a random case that’s been on my mind: Arcade Fire. Now, I like this band just fine. Funeral was a great album, and Neon Bible is a good one. But I’ve also never understood why they seem to generate so much more buzz and attention than any number of other comparable bands—why they and not, say, fellow countrymen The New Pornographers, are getting NYT Mag cover stories and selling out live shows in a matter of minutes everywhere they go. Even if they’re genuinely better than any number of similar bands, they’re clearly not that much better, not exponentially better.

So does all this, as Watts hints, undermine the notion that market outcomes reflect public preferences in some deep way? Well, it depends what we mean. At least when it comes to transient pop, it may be that people’s main “preference” is, as it were, de dicto rather than de re: People want to be listening to “whatever their friends are listening to,” rather than the particular artist who satisfies that description at any given time. And from that perspective maybe it is partly arbitrary that Justin Timberlake or Snow Patrol or whomever end up topping the charts at any given time—any number of more obscure artists might have served as well. Whether that amounts to some kind of problem is another story; nobody ever claimed culture markets were exempt from Ecclesiastes. Over the long haul, though, and for some of the reasons I outlined above, I’m skeptical that this experiment models the actual filter mechanisms that select and preserve art and artists.