Conservatism in Complex Probabilistic Inference

Abstract

Subjects were presented with data, described as the simulated output of a computerized radar system, consisting of dots that could fall in any one of twelve sectors. They were told that the process generating the data might be in any one of four mutually exclusive states. Displays showed for each state how likely it was that each dot would fall in each sector; an auxiliary display showed the prior probabilities of each of the four states. Subjects were required to estimate posterior probabilities of each state after each datum; comparison of these estimates with the correct values calculated from Bayes' theorem provided the dependent variables.