Statistical Analysis + Love = AWESOME

Late Valentine's Day update today, folks! This one is is another dump, although most of the juicy bits come from various posts on a single blog site. Here goes!

Let's start with the mainstream media summaries, for our friends with slightly less spare time. My interest was piqued by this NY Times article that discusses data from an online dating site called okcupid, which apparently was started by math peeps. Some of the more interesting results concerned racial preferences, which is the focus of a TIME article on essentially the same subject.

The real interesting stuff is on the blog itself, to be found here. This you really should spend some time on. The two most fascinating ones that I saw were tthis one, on racial preferences, and another on the effect of looks and pictures on your results. The most interesting thing about the looks post is twofold: one of course is that it confirms that looks are way more important for a woman than for a man. That much, as a certain female friend of mine noted, is obvious starting from middle school (I guess obvious to the girls, at least, us guys are a bit slower...).

The more interesting thing I thought was the distribution of attractiveness in guys. The distribution among women was a pretty straightforward Gaussian-looking thing, but for men it's skewed way towards the low end. Apparently women either can't agree on what guys look good, or they just hate all men. I wonder what kind of individual distributions on attractiveness women have. Does everyone just have a gaussian with a really low mean? Girls in the audience, let us know!

Further evidence of this: a friend of mine ran an experiment a while back where he did PCA on male and female faces that were rated for attractiveness by a number of volunteers. There was an obvious and distinct visual difference between the averaged female face constructed from the most attractive faces versus the least attractive faces, but the "most" and "least" attractive averaged male faces were barely distinguishable. The obvious conclusion is that people pretty much agree on good looking girls, but not at all on good looking men. Not surprising from an evolutionary point of view, given that the male reproductive strategy would select for traits indicating health/fertility that could be quickly detected so that they can make quick decisions.

Man, it's amazing when the numbers back up common wisdom! The only caveat is that okcupid's analysis (at least the published stuff) is really fairly simple and basic, with little in the way of controls. Somebody needs to go publish some real papers on this business...