Learning to be selfish? A large-scale longitudinal analysis of Dictator games played on Amazon Mechanical Turk

We investigate the role of experience playing the Dictator Game (DG) on allocations by aggregating data from 37 studies conducted on Amazon Mechanical Turk over a seven-year period. While prior evidence has shown a correlation between experience on Amazon Mechanical Turk and selfishness, it is unclear to what extent this is the result of selection versus learning. Examining a total of 27,266 decisions made by 17,791 unique individuals, our data shows evidence of significant negative effects of both selection and learning. First, people who participated in a greater total number of our studies were more selfish, even in their first game - indicating that people who are more likely to participate in our studies are more selfish. Second, a given individual tends to transfer less money over successive studies - indicating that experience with the DG leads to greater selfishness. These results provide clear evidence of learning even in this nonstrategic social setting.