We study how subjects in an experiment use different forms of public information about their opponents' past behavior. In the absence of public information, subjects appear to use rather detailed statistics summarizing their private experiences. If they have additional public information, they make use of this information even if it is less precise than their own private statistics - except for very high stakes. Making public information more precise has two consequences: It is also used when the stakes are very high and it reduces the number of subjects who ignore any information - public and private. That is, precise public information crowds in the use of own information. Finally, our results shed some light on unravelling in centipede games.