22 Oct Paper on LARGE scale experiments
One of the key questions about experiments on human behavior is that, more often than not, they are carried out with small groups of people. This was one of the main issues we addressed on our project IBSEN, by building a platform that allowed us to do large experiments with a controlled pool of subjects. Now, a very relevant paper resulting from that work is out on Scientific Reports. We had several groups of 100 people and one group of 1000 people play a repeated public goods game. The important point here is that they all played together so after every round we gave them information about the collective behavior of the group. Of course, giving detailed information on everybody’s contribution would be rather silly (who is going to make sense of a list of 1000 contributions), so we either give them the mean contribution, or the number of people with contributions in each of the quintiles of the distribution (ranging from 0 to 10 points). The result is astonishingly different: when people know the mean, they cluster around the mean. When they know the distribution, they polarize contributing either 0 or 10. If you only look at the mean, you would never know! On the other hand, groups of 100 behave more or less the same as the group of 1000, so it seems we reached the large size limit here. But perhaps we could try… 10 000? This LARGE work was done with Maria Pereda, Alberto Antonioni, Ignacio Tamarit, José A. Cuesta, (all of them at GISC/Universidad Carlos III de Madrid, like myself) and Penélope Hernández (at LINEEX, Universitat de València).