Running head: DIVISIVE NORMALIZATION AND FACIAL ATTRACTIVENESS 1 Facial attractiveness choices predicted by divisive normalization

Do people appear more or less attractive depending on the company they keep? I employed normalization models to predict context dependence of facial attractiveness preferences. Divisive normalization – where representation of stimulus intensity is normalized (divided) by concurrent stimulus intensities – predicts that choice preferences between options increase with the range of option values. I manipulated attractiveness range trial-by-trial by varying the attractiveness of undesirable distractor faces, presented simultaneously with two attractive targets. The more unattractive the distractor, the more one of the targets was preferred, suggesting that divisive normalization (a potential canonical computation in the brain) influences social evaluations. We obtained the same result when participants chose the most “average” face, suggesting that divisive normalization is not restricted to value-based decisions (e.g., attractiveness). This new application to social evaluation of a classic theory opens possibilities for predicting social decisions in naturalistic contexts such as advertising or dating.