Voting on a sharing norm in a dictator game

I conduct an experiment to assess whether majority voting on a nonbinding sharing norm affects subsequent behavior in a dictator game. In a baseline treatment, subjects play a one shot dictator game. In a voting treatment, subjects are first placed behind a 'veil of ignorance' and vote on the amount that those chosen to be dictators 'should' give. The outcome of the vote is referred to as a 'non-binding agreement.' The results show that a norm established in this fashion does not induce more 'fairness' on the part of those subsequently chosen to be dictators. In fact, dictators were significantly more likely to offer nothing under the treatment. I outline a simple model to account for this 'crowding out' effect of a norm that may demand 'too much' of some subjects.