De-biasing the Measurement of Conditional Cooperation

Fischbacher, Gachter and Fehr (2001) let subjects condition their contributions in a linear public goods game on the average contribution of their groupmates using the strategy method. About half of their subjects exhibit “conditional cooperation†(CC) in that they contribute more the more the groupmates are assumed to contribute. This finding has been extensively replicated. However, recent studies have found large fractions of conditional cooperators (CCs) even in placebo settings in which we would not expect to see any CC, suggesting that the measure of CC is upwardly-biased. We investigate whether mitigating subject confusion and experimenter demand can eliminate or at least reduce the bias. We introduce several design features to mitigate confusion. To mitigate experimenter demand, we provide participants with “exit options†that allow them to avoid conditioning their contributions on those of their groupmates. We evaluate the extent of the bias by the proportion of subjects classified as CCs in a mirror placebo setting involving a meaningless conditioning variable. When we mitigate confusion but not experimenter demand, more than a quarter of subjects end up classified as CCs in the placebo mirror. When we also mitigate experimenter demand, this proportion drops to a level indistinguishable from random behavior. In a standard setting, mitigating experimenter demand reduces the proportion of CCs by almost 40 percent. We therefore conclude that CC should be measured in the presence of the exit options in order to mitigate experimenter demand.