We often observe that people help strangers in real life. Such altruistic behaviour is puzzling for economics and biology because it promotes the welfare of another person at a cost to oneself. It is also difficult to explain it by long-term strategic motivations when the recipient is a stranger who may not be able to return a favour. Real-life evidence does not offer many opportunities to learn about the motivations behind individual altruism, unfortunately.

 In this project, we instead investigate reciprocal altruism in a laboratory experiment with people playing a long-term economic game of indirect reciprocity. This experiment provides sufficiently detailed data about motivations for altruism to facilitate the classification of almost 90% of participants into several theoretically feasible strategies. For this, we apply a recently developed statistical method to our indirect reciprocity experiment. We compare the resulting classification to the classifications of three other existing methods in the literature and demonstrate that it is the closest to a consensus measure. We then show how the other three existing methods ignore a learning strategy that is used by almost half of the subjects in one of our experimental treatments. Finally, we compare the strategy classification to people’s self-reports and show that these are a very unreliable source of data about individual motivations for altruism.