An Associationist View of Biases in Causal and Probabilistic Judgment

1 Introduction During the last few decades, an important debate has taken place in cognitive psychology concerning the rationality of human reasoning. One field in which this debate has been vigorous is that of inductive inference and probabilistic reasoning or, in more general terms, situations in which people make predictions about research has produced mixed results, ranging from some showing that human predictive behaviour under uncertain conditions is basically rational to others demonstrating violations of various principles of rationality. Equally controversial has been research on the psychological processes involved in this kind of reasoning. The debate has been intense and models have been proposed based on very different mechanisms, from those which involve the application of rules and procedures of a quasi-logical nature as the basis of human In spite of this variety, a great deal of this research has been guided by normative analyses of the tasks facing the reasoner. Marr (1982) emphasized the importance of the analysis of the task our cognitive system has to solve in order to correctly characterize the psychological processes involved. Normative theories have been taken as computational descriptions of the goals guiding behaviour. In the case of inferential tasks, normative analyses have frequently borrowed from disciplines such as logic or the mathematical theory of probability. Generally, researchers in this field take theoretical principles from such formal disciplines as the basis of a computational description of the objectives people have to satisfy when solving the task. In short, normative theories have played the role of a heuristic or guide (the heuristic function) in the formulation of psychological theories about how inferences In addition, normative theories have served another purpose (the arbitration function). Any deviations of individuals' inferential behaviour from the prescriptions made by such theories have been viewed as errors or irrationalities; that is, 2 normative theories have served to define what counts as an error (Gigerenzer & The heuristic function rests on a basic assumption, often made implicitly. This assumption is that there exists a degree of agreement between the task demands as characterized by the normative analysis and the demands to which people are sensitive. Note that a good computational description does not entail a perfect agreement between the normative analysis and all the demands people respond to. In fact, such a perfect agreement is virtually impossible because this would presuppose a well-established understanding of the processes involved in making inductive inferences. …