I don't see this mentioned in the other comments, but there is almost certainly more going on here than meets the eye. From http://www.okcupid.com/faaaq http:/

I don't see this mentioned in the other comments, but there is almost certainly more going on here than meets the eye. From http://www.okcupid.com/faaaq http://www.okcupid.com/faaaq - people get to explicitly filter potential matches according to the answers those matches have provided to these questions. It is not a black box recommender system - people can mandate explicit rejection criteria. This creates an incentive to lie on the questions. It would make sense for users to try and make their profile statistically close to what they think the [type of people they would like to date] specify. For example, Bob wants to go out with a 'good girl' (lets call her 'Alice'). Bob reasons that Alice, being a 'good girl' would probably want to go out with a guy who has never cheated; so Bob will neglect to mention his affair with Eve on his profile. I would guess that users, even if they don't initially realise the strategic situation they are in, will quickly learn it socially and through feedback, and adjust their profiles accordingly. The end result being that the answers given on the website, that provide the data being visualised here, should be treated with caution. I imagine that certain questions such as 'number of sexual partners' comes with large disincentive to lie, due to social pressure. As such, the conclusions here should be taken with a grain of salt - and that's even before discussing selection bias amoung the users of a particular social dating site. Such services are adopted as they spread through the underlying social network; it is well established that social networks are heavily clustered, and rarely give representative samples until they get very large. Its still very interesting stuff, but we should be careful about how strongly we interpret it.