Logit estimation of conditional cooperation in a repeated public goods experiment

A conditional cooperator in a public goods game wants to match his partners' expected contribution. We investigate theoretically and empirically whether (and to what extent) conditional cooperation can explain how individual contributions evolve in a repeated two-person public goods experiment using a perfect strangers design. To identify a random utility model including non-pecuniary preferences we elicit participants' beliefs. Our econometric results show that the distribution of preferences in the population can be captured by a latent-class mixed logit specification with three subpopulations, and that 55% of participants can be regarded as conditional cooperators. Thus, the decline in average contribution levels may be attributed to the presence of conditional cooperators who have to revise their expectations about the others' behavior.