When is a Face Hot or Not?

Why does Johnny Depp look so good in eyeliner? Why is the girl next door rarely ever also a Victoria’s Secret model? If somebody wanted to make Will Ferrell into a model, exactly which bits of his face would need to be tweaked to make it happen?

These are the sorts of questions that keep attractiveness researchers awake at night. But now, there’s a solution.

Published online last week in Psychological Science by face researcher-turned-vision scientist Chris Said at NYU and Alexander Todorov, who happens to be Said’s former Ph.D. adviser, is an amazingly comprehensive new statistical model for facial attractiveness. Yes, hotness can be measured and predicted.

This is not a frivolous paper (although, says Said, it was the most fun of all the data sets he’s ever worked on). For starters, faces are important to humans; in fact, they’re among the most compelling things for us, almost from day one. Add to that, for psychology researchers and the rest of us, attractive faces are also important: among other things, we associate a pretty face with health, intelligence, competence and honesty — and reward them with better jobs, more money, and informally speaking, lots of game.

There are a few long held theories around — good ones, in fact — about what makes a face attractive. For instance, there’s the average hypothesis, in which the average face is an attractive face (true), and the sexual dimorphism hypothesis, in which, for example, a woman’s face is increasingly attractive the more feminine its features (also true much of the time). But here’s the catch: neither theory explains everything (why, for instance, a manly face on a guy isn’t always hot) and when you think about it, the theories are actually incompatible. The most average face is not, by nature, the most anything.

So, he found a way to resolve both issues, wrapping them up into a neat little statistical model.

Said took 40 undergrads (20 male, 20 female) and set them down to rate 4,000 faces. Then he took those faces, measured them in 50 different dimensions, incorporating both shape and shading, then folded in the ratings from the undergrads and built himself a statistical tool. Compared to the two other models (which match up to human raters about 1-30 percent of the time), this one has a solid accuracy hovering around 70 percent. Not too shabby.

If you’re curious, it also found that the average face is a pretty face, but rarely ever a knockout. Contrary to the dimorphism theory, not only can a femme-featured guy be above-average appealing, but (and this goes for women, too), the face becomes seriously attractive with a little extra darkening around the eyes — something the makeup artists of Pirates of the Caribbean probably realized a long time ago.

By the way, in case you’re wondering whether you’re hot, or … well … not, his model is online here. (Way more fun to play around with though, is the software he used to help create the data set. I’m not kidding: check it out.)