So . . . I like music (yeah, kind of like saying "I like food," I know). That is to say I enjoy music - I'm not really one of those "music can change the world" types. My tastes tend to be eclectic (very few genres I don't listen to) and my collection is fairly extensive, I think (not a world record sized one, but a fairly robust 15000 songs). Being a researcher, though, I also have an interest in what makes some music popular and others not. The scientific answer after the jump.
A little while ago I read an article by Malcolm Gladwell about, essentially, the application of science to art. He uses examples in the film and music industries to show how advanced mathematical modelling and computer programming can help predict the likelihood of success of a song or movie. I'm focusing more on the music example, though the article focuses more on the movie one.
When we hear a song, we like to think we evaluate it based on how it makes us feel, or how much we enjoy it, or how it resonates. If this were the case, each song would be independently evaluated by each person, and it would mean that hit songs were "better" songs in the sense that they resonated positively with the most people.
From an artist's perspective, this is what they are trying to do - reach people. Use the words and music to elicit some kind of reaction. There is no science to it, just managing to convey a feeling in three minutes or so. Attempts to change or tailor the artist's output are typically portrayed as profit-minded actions by greedy record execs for marketing ends (see: Spice Girls, N'Sync, etc.). "Real" music fans (ironically, I think most "real" music fans don't reside in the "real" America) see through such attempts and reward artists who have integrity, such as Norah Jones, or Gnarls Barkley, or Johnny Cash, depending on your tastes.
But what if musical taste was predictable, that we just liked the same things over and over again? In Gladwell's article he discusses software that has found that the hit songs from the past several decades all share mathematical commonalities (with regard to factors such as melody, rhythm, tone, etc.). There are sixty or so mathematical "clusters" where hits reside (only about 15 are active at a time), which means that if your song is outside of one of these clusters, it's highly unlikely to be a hit. There just seems to be certain patterns that we like more than others.
We already know that music preference isn't only about the music. Past studies have shown that familiarity has a big impact (we like what we know), which is kind of similar to this but on a micro scale. I think we've all had the experience of hearing a song for the first time and either not liking it or being indifferent, and then repeated exposure makes us like it (current songs like that for me: Love the Way You Lie by Eminem and Rihanna, Breakeven by the Script). Of course, overexposure can wear that way, like with Bryan Adams' Everything I Do, which was super-overplayed in the early 1990's due to it being a) from a hit movie, b) a hit song and c) Canadian, making radio stations fulfill their entire Canadian content requirements with one song.
Three things about the model. First, the computer software can only tell us if a song is in a cluster or not - it can't say how to get there. If a song is outside a cluster, an engineer or producer can start to fiddle with it to try to see how to get it closer to a cluster, but it's trial and error. So we don't know what it is about a song we like, but we know when it's a song that we like. For this reason the artistic types can breathe easy (for now), because we don't know how to build a hit, only to assess whether an existing song has hit potential.
Second, inclusion in a cluster doesn't guarantee a hit, but exclusion nearly guarantees a non-hit. This is kind of like the discussion of ten thousand hours (also from a Gladwell piece) - having the experience doesn't guarantee expertise, but a lack of the experience makes success very difficult.
Third, we might have the most to learn from the songs that defy the model. Is there something about the songs that are in the clusters but don't become hits? What about the hits that don't belong to a cluster? Are these just exceptions, or are there reasons? For example, if Elton John's massive-selling Diana tribute was a hit not in a cluster (which I don't think was the case, as the original version was also a hit) we could attribute it to the tribute component. What about the others?
So if you, like me, complain that music today isn't as good as it used to be, be warned: the math shows that music today is exactly how it used to be.
No comments:
Post a Comment