Estimating aggression from emotionally neutral faces: which facial cues are diagnostic?

The facial width-to-height ratio, a size-independent sexually dimorphic property of the human face, is correlated with aggressive behaviour in men. Furthermore, observers' estimates of aggression from emotionally neutral faces are accurate and are highly correlated with the facial width-to-height ratio. We investigated whether observers use the facial width-to-height ratio to estimate propensity for aggression. In experiments 1a-1c, estimates of aggression remained accurate when faces were blurred or cropped, manipulations that reduce featural cues but maintain the facial width-to-height ratio. Accuracy decreased when faces were scrambled, a manipulation that retains featural information but disrupts the facial width-to-height ratio. In experiment 2, computer-modeling software identified eight facial metrics that correlated with estimates of aggression; regression analyses revealed that the facial width-to-height ratio was the only metric that uniquely predicted these estimates. In experiment 3, we used a computer-generated set of faces varying in perceived threat (Oosterhof and Todorov, 2008 Proceedings of the National Academy of Sciences of the USA 105 11087-11092) and found that as emotionally neutral faces became more 'threatening', the facial width-to-height ratio increased. Together, these experiments suggest that the facial width-to-height ratio is an honest signal of propensity for aggressive behaviour.