OKCupid has an excellent blog post that finds that women on its dating site of a given attractiveness level receive more messages if they have a wider dispersion of attractiveness ratings around the mean. Where attractiveness ratings are broken into quintiles, a regression analysis yields the following expression for the expected number of messages:
where m1 is the number* of bottom-quintile rankings and m5 is the number* of top quintile rankings. The coefficients on m1 and m5 are positive — in other words, if people do not think you are extremely attractive, it is better if they think you are extremely ugly.
The authors suggest a game-theoretical explanation for this: men know who the wider-dispersion women are, and focus on those whom they find attractive but others do not, because it will maximise their chances of success. But I am not sure that this explanation is necessary. This pattern is just what you would expect if distinctiveness is an important component of attractiveness. That idea chimes with the discussions that men have in pubs — a breast man will appreciate unusually large breasts, for example, while a leg man might find them unappealing. Clustering at the extremes of the attractiveness ratings should indicate women who have something special — at least, special to those who will appreciate it.
This makes me think of the old days of having a day job, and discussions about corporate strategy. I have always thought it is better to decide what you do, do it well, and let any potential customers who don’t like it take their money elsewhere. You might think this is “uncommercial.” But this situation is analogous to the dating situation described above. Some people might hate what you do, but that will only mean that you stand out from the crowd as far as the people who like your approach are concerned. In a vast modern economy, you can even turn off a majority of people, provided that a solid minority loves what you do.
* I assume — it is not clear whether it is a number or a proportion, but a number seems more likely.