Modeling reciprocation and cooperation in primates: evidence for a punishing strategy.

Experiments in which animals strategically interact with one another or search over some controlled domain are becoming common. While these experiments often promise to illuminate sophisticated animal behavior, the analyses brought to bear on these data are often quite coarse. For example, many papers simply tally the number of observations consistent with a behavioral theory. This analysis is simple, but ignores a potentially rich source of information by failing to take into account patterns and systematic variation among observations inconsistent with the theory. Using a new data set generated by cotton-top tamarin monkeys playing a repeated food-exchange game, we apply a maximum-likelihood estimation technique (more commonly used to study human economic behavior) which utilizes much more of the information in these data, and which uncovers unexpectedly sophisticated cooperative behavior from our subjects. Tamarin cooperation remains stable as long as both actors consistently cooperate, but requires at least two consecutive unexpected acts of cooperation to restart cooperation after it has collapsed, a strategy that resembles two-tits for a tat. We conclude by enumerating the benefits of a maximum-likelihood approach in experimental settings such as ours, and suggest other areas in which these techniques may be fruitful.