Attractive or Aggressive? A Face Recognition and Machine Learning Approach for Estimating Returns to Visual Appearance

A growing literature documents the presence of appearance premia in labor markets. We analyze appearance premia in a high-profile, high-pay setting: head football coaches at bigtime college sports programs. These employees face job tasks involving repeated interpersonal interaction on multiple fronts and also act as the “face†of their program. We estimate the attractiveness of each employee using a neural network approach, a pre-trained Convolutional Neural Network fine tuned for this application. This approach can eliminate biases induced by volunteer evaluators and limited numbers of photos. We also use this approach to estimate the perceived aggressiveness of each employee based on observable facial features. Aggressiveness can be detected from facial characteristics and may be a trait preferred by managers and customers in this market. Results show clear evidence of a salary premium for less attractive employees. No beauty premium exists in this market. We also find evidence of an aggressiveness premium, as well as evidence of higher attendance at games coached by less attractive and more aggressive appearing coaches, supporting customer based preferences for the premia. We also provide a methodological contribution by incorporating face recognition and computer vision analysis to evaluate employee appearance.