Task information and memory aids in the learning of probabilistic inference tasks

Sammanfattning: A trend in current decision research is a growing recognition of the "contingent" nature of decision making. Decision behavior is not invariant across normatively equivalent tasks and the use of normative, or non-normative strategies is contingent upon the conditions prevalent in the decision situation. An important research task is therefore to determine what these conditions are. The present thesis is concerned with this problem as it manifests itself in the ability to learn in probabilistic inference tasks. A traditional finding in the study of such learning is that people develop strategies which differ in fundamental ways from the statistical regression strategies required for optimal performance. The purpose of the present thesis was to investigate under what conditions people are likely to change this behavior and develop optimal strategies. Specifically, the aim was to assess how learning of optimal strategies was related to (1) information about the statistical nature of the task, (2) information about the appropriate approach to use in such tasks and (3) access to memory aids. This was accomplished in four studies.The results show that information about the statistical nature of the task is not sufficient for optimal learning. It is necessary to also inform about the appropriate approach. Memory aids are not needed to learn the optimal prediction rule but they are needed to fit and use this rule.The results were interpreted to mean that statistical regression is not part of the layman's repertoire of statistical intuitions.