The power of regression analysis by Malcolm Gladwell

As Malcolm Gladwell expressed in a New Yorker article, when it comes to art, most people fall in the Hume camp of reasoning, but a defiant minority always falls in the Kames camp:

“Something has long been argued about art: there is no way of getting beyond one’s own impressions to arrive at some larger, objective truth. There are no rules to art, only the infinite variety of subjective experience. “Beauty is no quality in things themselves,” the eighteenth-century Scottish philosopher David Hume wrote. “It exists merely in the mind which contemplates them; and each mind perceives a different beauty.” Hume might as well have said that nobody knows anything.

But Hume had a Scottish counterpart, Lord Kames, and Lord Kames was equally convinced that traits like beauty, sublimity, and grandeur were indeed reducible to a rational system of rules and precepts. … He genuinely thought that the superiority of Virgil’s hexameters to Horace’s could be demonstrated with Euclidean precision, and for every Hume, it seems, there has always been a Kames—someone arguing that if nobody knows anything it is only because nobody’s looking hard enough.”

I first came across a follower of Kames while studying econometrics at Princeton. My professor, Orley Ashenfelter, wrote a model to predict the quality of wines in Bordeaux based on the temperature and precipitation that year. His model could predict the quality of the wine before it was even on sale and had been tasted by experts like Robert Parker. Needless to say the wine community was outraged. The typical reaction was “How can you reduce our noble art to mere equations? How can you claim to know the quality of the wine without tasting it?” The critics were pointing out “flaws” such as “the model only predicted the exact quality of the wine in one specific year” failing to understand that econometrics is about confidence intervals. The model and the experts only disagreed on one year, 1964, where the experts suggested the wine was not very good, while the model suggested otherwise. Needles to say Professor Ashenfelter’s cellar is full of wine from 1964 🙂

Gladwell’s article covers the travails of a few brave souls as they try to predict hit songs and hit movies. He first describes how a small New York based startup called Platinum Blue analyzed the mathematical relationships among a song’s structural components. The applied their model to thousands of songs and noticed that all the hits came out of a predictable and highly conserved set of mathematical patterns. As McCready, the firm’s founder explains this does not mean that all songs that conform to the pattern will be hits, they still have to “sound right”, but almost certainly songs that fall out of that pattern will not be hits. Interestingly enough, when they first ran their analysis, Platinum Blue was really excited about by the Norah Jones record “Come Away with Me”. It went on to sell twenty million copies and win eight Grammy awards.

Gladwell then describes the story of Dick Copaken, a lawyer and cinephile, who created a company called Epagogix that analyzes screenplay elements predict US box office receipts. Predictably, the industry reception was terrible at first: how could outsiders claim they knew more than the insiders? But after a number of successful tests Epagogix is now working with major studios.

It seems that art is more structured than we thought after all – not that we necessarily want all songs and movies to be blockbusters…

Read the full article at:
http://www.newyorker.com/archive/2006/10/16/061016fa_fact6?currentPage=1

Newsletter

Select list(s):
  • Of course this “art” is highly structured. The predictability of standard movies is mind-numbing and I feel childish when I go to see one because it is so very. In addition Man vs ( self,other man, nature ) are the only plots you ever see in conventional non-docu films. Thus I say to you want to go camping? Don’t forget the camping wig!!!

  • Very interesting to hear that hit songs fall into a common mathematical pattern. Would like to know which songs they predicted would be hits and actually were. Maybe you could use a technology like that to buy and sell song “futures” on a music “exchange”. Taking it one step further, it would be great to finesse that technology so that it could be used to predict what other songs a person would like, given their listening or buying history. I guess there are many sites that attempt to do this, but only by tracking other peoples’ habits or by lumping songs into categories.

  • You’re falling into a logical trap here – the whole debate between Kames and Hume is over the definition of beauty, and you’re starting your argument with the assumption that a majority of people agreeing to call something beautiful is the same as it actually being beautiful, in essence defining beauty as majority rule, or the sum of individual judgments about the beauty of a thing. Which is kind of avoiding the whole argument. I haven’t read Kames, and have read probably too much Hume, so I may be misinterpreting the summary of their arguments you give, but I think the substance of their debate goes deeper than whether you can predict whether a majority of people will perceive something as beautiful.

    Take a property such as an object being red. It is red because it reflects light back at certain wavelengths. We can measure these wavelengths either just by looking with our eyes or with more advanced tools, but this redness exists completely independently of our perception and measurement of it. Even if a colorblind person were looking at the object and failed to distinguish the wavelengths from others emitted by green objects, it would still be red. Even if the entire population of Earth were blind, or colorblind, or involved in a mass hallucination where they saw the object as blue, it would still be emitting those same wavelengths and hence still be “red” (now it’s possible we’re currently involved in a mass hallucination and the wavelengths are actually not being emitted at all, but that’s getting way beyond the point).

    Being “beautiful,” according to Hume, is not the same thing. Hume thinks that beauty is in the eye of the beholder. Whether or not something is perceived as beautiful by a majority of the populace will change over time (e.g., the majority of my generation thinks Disco sucks while the majority of my parents’ generation seemed to like it), by geography (for some reason Europeans love Robbie Williams while he just has not caught on in the States), or even based on the personal state of the individual (certain songs touch me post-breakup that I thought were stupid during the relationship) or the context of what is being observed (Rock and Roll is much more brilliant if you know something about communism and/or rock, I think Picasso’s Las Meninas is absurd unless placed directly next to the Velasquez version, etc.).

    Hume would not argue that there are no trends to what people find beautiful, or dispute that certain characteristics will lead to a larger subset of the population finding something beautiful (as per the regressions), but he would object to the idea that a movie or song cannot be beautiful simply because the majority of the populace does not find it beautiful. He would say that it could still be beautiful for someone at some point, and it’s OK if it’s not beautiful for anyone else – that doesn’t make it not beautiful.

    Kames, on the other hand, seems to be arguing for a more absolute beauty – something either is or is not beautiful just like something is or is not red. If I perceive an object as not beautiful even though it is, it is a problem with my own perception, attributable to my own failings, and not at all a reflection on the object. Roses just are beautiful, period, because of mathematical properties such as symmetry, fractal formulas, etc. If I don’t like them, it’s because I’m perceiving something inaccurately.

    Where do I come out? I think that in the end they’re both right, and it’s a failing of our language that there aren’t two words for beauty, one to capture the individual perception and one to capture these other properties that exist within the object itself (I keep thinking of math things, like the natural log, that are just so beautiful their beauty has to exist outside of perception). Which one is the regression capturing? A combination of both. Through some accident of evolution and environmental conditioning, we as a species tend to agree on beauty at a lot of points, and the regression captures the sum of this agreement, a lot of which is driven by these intrinsic beauty properties. You could also run a regression for any single individual, adding as variables all of the environmental factors at play (Will Stacie like this song at point in time x? Probability is .25 times her age + .075 times (12 – the number of months since her last breakup) – 1.25 if she’s in a good mood. Sum all those up and you get it for the population, and know whether or not you have a hit.
    🙂

  • 1. It’s much more fun for me to frame it as you being wrong

    2. Gladwell (and you by association, although I guess this most recent email disassociates you) is trying to say that the ability to quantify what is popular supports Kames and negates Hume. My point was just that it neither supports nor negates either.

  • Thanks for sending along – very cool additional take. This whole regression analysis of art kind of reminds me of one of my favorite speeches in Rock and Roll, at the end of the first act (I’m sure you with your superhuman memory remember the one I’m talking about) – where he is insisting that we’re all really just wires and beer cans forming computers that produce certain mathematical outputs, and his wife starts yelling at him, asking if the tears he cries at her funeral will be the same sort of output, and he has to confess that yes they will. Whom we fall in love with, what we think is beautiful, all comes down to some algorithm predetermined by the OS code in our heads. Which in the end I believe and think isn’t as tragic as everyone likes to pretend it is.

    The big thing that bugs me about all this is how many people are trying to equate commercial success or fit with the algorithm as the determining factor as to whether or not something is “good.” I think that new forms of art and new patterns are emerging all the time, and they may create new clusters on the Platinum Blue chart, new categories or variables, that are “beautiful” by both the objective and subjective measures. I don’t want to see those surpressed or dismissed as we get too scientific about what we like. And I also don’t want to lose the idea (naive as it may be) that artists create art for themselves as some sort of catharsis, some need independent of us appreciating it, or that if a piece of art touches just one person, it was worthwhile in the creation.