Abstract
Is it possible to measure the extent of altruism in a controlled laboratory environment? Recent attempts to measure altruism towards other players or charities suffer from two potential confounds. One is the fact that the act of giving is typically correlated with extracting surplus from the experimenter. Thus, what could appear to be altruistic acts might just be a desire to make sure that the experimenter ends up with as little as possible. Or, on the other hand, revealed altruism might be understating true altruism if subjects value money being left on the (experimenter’s) table. The other general confound is that subject preferences with respect to altruism are likely to be heterogeneous across subjects. This can confound inferences because the sample sizes typically employed in experiments are insufficient to ensure that randomization to treatment will correct for differences in treatment behavior due solely to differences in sample composition. We illustrate each of these points with simple laboratory experiments that derive from popular recent designs. We find that there is a significant effect from each confound, and that the results are surprising. Depending on the context, revealed altruism might overstate or understate true altruism if one fails to correct for “as if” acts of altruism being negatively correlated with experimenter surplus. Furthermore, unconditional measures of altruism can be qualitatively misleading, due to the unusually variable nature of the propensity to undertake acts of altruism. Fortunately, this variable propensity appears to be associated with readily observable characteristics of individuals, so that one can correct for differences in sample composition across treatments. We conclude that altruism does exist, but that it cannot be identified independently of the circumstances of the task.