We live in an uncertain world where the consequences of our actions are not always predictable. Therefore, the decisions to invest in the stock market, to undergo a medical treatment, or to go to court depend on our assessment of the chances that the market will go up, the treatment will be successful, or the court will decide in our favor. Because in general we do not have objective methods for computing the probabilities of such events, we must rely on human judgment as the major instrument for assessing uncertainty. Hence, the question of how people evaluate evidence and assess uncertainty is highly relevant to many aspects of our lives, from the diagnosis of a patient to the evaluation of expert judgment. An extensive body of research on judgment under uncertainty indicates that intuitive judgments of both laypeople and experts are often at variance with accepted normative principles of probability and statistics (see, e.g., Arkes & Hammond, 1986; Camerer, 1993; Dawes, 1988; Kahneman, Slovic, & Tversky, 1982). These findings have commonly been attributed to cognitive limitations, and explained in terms of judgmental heuristics or simplifying strategies. This proposal presents a new approach to subjective probability based on the notion of evidential support. It gives rise to a formal representation that is compatible with heuristic process-based accounts, and encompasses a wide range of phenomena within a unified theoretical framework. The proposal is divided into four parts. Part 1 presents a new model of belief, called support theory, in which the judged probability of an event depends on the specificity of its description. In particular, judged probability is increased by unpacking the focal hypothesis and decreased by unpacking the alternative hypothesis. Part 2 describes a series of experiments designed to test the major predictions of support theory, including the subadditivity of probability assessments, and the difference between probablity and frequency judgments. Part 3 extend s the theory to the analysis of conditional probability and evidential support. In this account, judgments of conditional probability depend not only on the co-occurrence of the relevant events; it also depends on their correlation. This prediction will be tested in both knowledge-based and data-based judgments. Part 4 addresses the problem of ambiguity or vagueness, and provides a method for assessing the imprecision of belief in terms of upper and lower probability judgments. These will be compared to standard measures of reliability, and their implications to decision making will be explored. Theoretical and practical implications of the present approach to the representation and the elicitation of belief are discussed throughout.
