Economic Theory and Experimental Seminar - Juan Dubra (Universidad de Montevideo)

Title: Overconfidence, Illusion of Control, or Unskilled and Unaware? (joint with Jean-Pierre Benoît and Giorgia Romagnoli)

Abstract: We report the results of five experiments where we address the nature and measurement of Overconfidence. First, we develop model to define and evaluate various possible definition of control, and estimate each one. In second term, we develop a new method to measure beliefs that accounts for the tendency of people to bet on devices where their skill might be involved (a "bias" called Control in the psychology literature). This allows us to improve over the methods traditionally used to measure beliefs' about one's own performance. The influence of Control in the measure of overconfidence in this second experiment is significant and the point estimate is 6%: true average belief of being in the top half in a test is 59%, but when we do account for Control, individuals report 65%.

In three other experiments we address the hypothesis of Kruger and Dunning, "unskilled and unaware". Their hypothesis is that populations exhibit overconfidence because the unskilled are unaware of their lack of skill (eg: the same skills needed to write a grammatically correct sentence are those needed to know that the sentence is correct). We test this hypothesis against the null that the whole population (not just the unskilled) is overconfident, but regression to the mean makes the unskilled look overconfident (eg: the beliefs of the unskilled are a weighted average of the prior --"I am average"-- with the signal --"I did poorly in this test"). Using a variety of tests (on the experiment in Benoît, Dubra and Moore, and in a new experiment) we cannot reject that the whole population is overconfident. To the best of our knowledge, this is the first statistical test of the "unskilled and unaware" hypothesis.

Finally, on a fifth experiment we explore the same issue of unskilled and unaware vs general overconfidence, studying directly the (structurally estimated) signalling structures elicited in the experiment. This also allows us to verify whether each individual is also unaware when he is unskilled, vis a vis those questions where he is skilled. Our estimations show that one cannot reject that the signalling structures of the unskilled are as good as those of the skilled.