The dictator game represents a workhorse within experimental economics,
frequently used to test theory and to provide insights into the
prevalence of social preferences. This study explores more closely the
dictator game and the literature's preferred interpretation of its meaning
by collecting data from nearly 200 dictators across treatments that
varied the action set and the origin of endowment. The action set
variation includes choices in which the dictator can "take" money from
the other player. Empirical results question the received interpretation
of dictator game giving: many fewer agents are willing to transfer
money when the action set includes taking. Yet, a result that holds
regardless of action set composition is that agents do not ubiquitously
choose the most selfish outcome. The results have implications for
theoretical models of social preferences, highlight that "institutions"
matter a great deal, and point to useful avenues for future research
using simple dictator games and relevant manipulations.