On Salon, Thomas Bartlett mentions the Music Lab experiment at Columbia:
A research team led by Columbia University sociology professor Duncan Watts is conducting an experiment called Music Lab … an experiment that involves a lot of free MP3 downloads. After signing in and answering a few questions, you’ll find yourself on a page listing 48 songs. If you click on a song, you’ll hear a streaming version of it, along with a prompt to rate how much you like it. After answering, you’re given the option to download it (free and legal) or to move on. I’ve listened to a handful of the songs, all of them by bands I’ve never heard of, and so far haven’t found them particularly good. How the experiment works, and what the Columbia researchers are actually measuring, isn’t explained anywhere on the Web page. (Link, may be premium)
The description of the experiment made me think of Hit Song Science and how the firm that owns it (Polyphonic HMI) breaks down successful songs using computers to examine their underlying structure… I wonder if the experiment is along similar lines?
Another possible approach is discussed by Malcolm Gladwell in Blink: music research firms such as Pick the Hits and the creatively-named Music Research, which measure how a sample audience responds to a song and can predict success with relative accuracy. I’m curious as to which method - surveys vs computer analysis - produce “better” results. The Hit Song Science people claim that it can pick out leftfield candidates, such as Norah Jones.
Of course, what this all presumes is that popular songs are intrinsically good, rather than made popular by publicity…