Make Humans Randomize

This paper presents results from an experiment studying a two-person 4x4 pure coordination game. We seek to identify a labeling of actions that induces subjects to select all options with the same probability. Such a display of actions must be free from salient properties that might be used by participants to coordinate. Testing 23 different sets of labels, we identify two sets that produce a distribution of subjects’ choices which approximate the uniform distribution quite well. Our design can be used in studies intending to compare the behavior of subjects who play against a random mechanism with that of participants who play against human counterparts.