I think this analysis is way off. Given two girls with similar average attractiveness, the one with the higher stdev (wider distribution, whatever) is going to

I think this analysis is way off. Given two girls with similar average attractiveness, the one with the higher stdev (wider distribution, whatever) is going to have more "5" votes than the girl with a low-variance average. This is probably completely driving the trend. Guys are most-likely to message you if they think you're a 5. So, given girls with equal average attractiveness, the one with the wider distribution will have more 5's and get more messages. EDIT: After looking more-closely, their formula basically agrees with what I've said: the number of 5's is weighted most-strongly (0.9 versus the second-strongest of 0.5). Also their R^2 value is pathetic.