Document Type


Publication Date



We present experimental evidence consistent with statistical discrimination in a public good and group formation game. We find that behavior is correlated with race and, to a lesser extent, gender, and people use race to predict behavior when no other information is available. When information on behavior is provided, people disregard personal characteristics completely. These characteristics are also disregarded when individual behavior is induced to break the correlation between characteristics and behavior. That is, people disregard race and gender even when observed behavior is unusual but relevant to payoffs. Finally, our experiments show that sorting into groups has dramatic implications on cooperation. Overall payoffs are higher when sorting is possible than when groups are randomly assigned. This only occurs, however, when personal characteristics are known. Higher payoffs are attained at the cost of an equitable distribution.


To learn more about the Andrew Young School of Policy Studies and ExCEN Working Papers Series, visit and