Project group B07: Experiments

Conducted Marta Kozakiewicz (former member A01) and Lorenz Götte, B07

In the project we experimentally investigate the mechanism behind misguided learning. It is closely related to our recent work, Götte and Kozakiewicz (2018), in which we corroborate the theory by Heidhues et al. (2018) and show that, under certain conditions, persistent overconfidence can lead individuals to taking suboptimal actions, misreading the output and forming erroneous beliefs about the state of the world. In line with the model's predictions, the learning process of overconfident subjects is "misguided", as it drives the agents away from the correct belief and, since the agents themselves generate observation that lead them astray, one can describe it as "self-defeating".

In the current project we conducted an additional control treatment in which the output is based on a parameter that is not relevant to agent's self-esteem. The results demonstrate that the ego-relevance of the unknown parameter exacerbates overconfident agents' tendency to mislearn. Pooling the data from our main study and the additional treatment, we find that for the overconfident agents the effect of the treatment is large and significant. The effect persists even if we control for the initial bias, and is not significant for underconfident of unbiased agents. Our interpretation of the results is that when the output is based on a parameter that is not important to one's self-esteem, overconfident subjects are more willing to admit that they are wrong and abandon their model of the world. The experiment enabled us to shed a light on the driving forces behind the misguided learning, and better understand the behavior of overconfident individuals. The topic is of great relevance as the accurate assessment of the state of the world is essential in virtually all economic decisions.

Conducted by Lorenz Götte, B07, Hua-Jing Han (former member of B07) and Benson Tsz Kin Leung

Information overload is a prominent issue nowadays, especially given the amount of information available on the Internet. There are concerns that the Internet, or information overload, will drive individuals to cherry pick the information that confirms their beliefs, which leads to biased information processing and polarization. In this paper, we present the results of a well-controlled experiment that, for the first time, study how individuals update their belief with information when they face different levels of information overload. In our experiment, subjects in the treatment and the control condition receives the same signal about an underlying state of the world, but are subject to different levels of difficulty of information processing. Subjects in the treatment condition are subject to a stronger cognitive load and thus a stronger information overload. We say that a stronger information overload drives a stronger confirmation bias if subjects update more with belief-confirming information and less with belief-challenging information in the treatment condition than in the control condition. Indeed, we find that upon receiving belief-challenging information, subjects in the treatment condition update less compared to the subjects in the control condition. On the other hand, upon receiving a belief-confirming signal, subjects' belief updating behaviors do not differ significantly across the two conditions. Thus, keeping the available signals constant, stonger information overload leads to symmetrically more biased processing behavior. The results of our experiment have several implications. First, it sheds light on the debate of whether or not the Internet strengthens biased behavior and promotes ideological polarization. On the other hand, the results of this paper suggest a novel channel driving the confirmation bias. In contrast to the existing preference-related explanations, we show that the bias changes with the informational environment. It brings to light a new concern for informational policies, as they might change the processing behavior of individuals.