Giving to Varying Numbers of Others

Within a modified N person dictator game, we test the extent to which giving behaviour changes as the number of recipients varies. Using a within-subject design, in an incentivised laboratory experiment, individual-level preference parameters are estimated within five alternative utility functions. Both goodness-of-fit and predictive accuracy of each model are analysed, with the "best" model identified for each individual. The Dirichlet distribution is proposed as a random behavioural model to rationalise noise; with parameters accounting for differential error arising from the complexity of decision problems. Results show that, on average, participants are willing to give more total payoffs to others as the number of players increase, but not maintain average payoffs to others. Extensive heterogeneity is found in individual preferences, with no model "best" fitting all individuals.