Social learning with partial and aggregate information: Experimental evidence

In our information cascade experiments, we study social learning in decision-making situations in which decisions “not to do” are unobservable. Subjects, in sequence, choose whether to invest or not, without knowing their position. They observe a private signal and the number of investments made by their predecessors, but not how many predecessors have chosen not to invest. We find that down cascades, in which agents neglect the signal and do not invest, occur, in contrast with the equilibrium predictions. Up cascades, in which agents invest independently of the signal, occur, but less than in equilibrium.