Punishment despite reasonable doubt: A public goods experiment with uncertainty over contributions

Under a great variety of legally relevant circumstances, people have to decide whether or not to cooperate, when they face an incentive to defect. The law sometimes provides people with sanctioning mechanisms to enforce pro-social behavior. Experimental evidence on voluntary public good provision shows that the option to punish others substantially improves cooperation, even if punishment is costly. However, these studies focus on situations where there is no uncertainty about others' behavior. We investigate punishment in a world with 'reasonable doubt' about others' contributions. Interestingly, people reveal a high willingness to punish even if their information about cooperation rates is inaccurate, or noisy. If there is some non-trivial degree of noise, punishment (1) cannot maintain high contributions and (2) reduces welfare even below the level of a setting without punishment. Our findings suggest that sufficient information accuracy about others' behavior is crucial for he efficiency of sanction mechanisms. If a situation is characterized by low information accuracy, precluding sanctions can be optimal.