Experimental Evidence on Misguided Learning

We test experimentally the theory of misguided learning formulated by Heidhues et al. (2018). The model predicts the behavior of an agent who has a biased perception of his ability and is learning about an unknown, decision-relevant parameter. We use a novel experimental design to demonstrate that the learning process of an overcon dent agent di ers signi cantly from that of an unbiased agent. In a dynamic setting, the overcon dent agent repeatedly takes suboptimal actions, misinterprets the output and forms erroneous beliefs about the unknown parameter. We provide the rst empirical evidence that giving a biased agent the opportunity to experiment and acquire new information is not only ine ective, but in some cases counterproductive.