A Model of the Conjunction Fallacy.

Abstract

Tversky & Kahneman show that both sophisticated and naive people, in many different substantive problems, often judge the conjunction of events to be larger than one of its components (hereafter called a single violation). Furthermore, for some problems, people judge the conjunction as larger than both of its components (a double violation). The purpose of this document is to propose a quantitative model of how people judge the probability of conjunctive events. The advantage of the model is that it makes specific predictions as to when conjunction fallacies of different types will or will not occur. Moreover, the model is naturally extended to deal with conjunctive explanations for events. The conjunction fallacy occurs when the judged probability of a conjunctive event is larger than the probability of one (or more) of its constituents. A model of this phenomenon is proposed in which the judged conjunctive probability is a weighted geometric average of the component probabilities, where the weights reflect the representativeness of the components. The model generalizes to the case where the events involve a casual or correlational link, and, to the use of conjunctive explanations of an event. In all three situations, the model specifies the conditions under which different degrees of the fallacy will or will not occur.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jun 01, 1985
Accession Number
ADA160876

Entities

People

  • H. J. Einhorn

Organizations

  • University of Chicago

Tags

Communities of Interest

  • Biomedical
  • C4I
  • Human Systems
  • Space
  • Weapons Technologies

DTIC Thesaurus Topics

  • Applied Psychology
  • Biomedical Research
  • Classification
  • Computer Science
  • Engineering
  • Human Factors Engineering
  • Industrial Engineering
  • Information Science
  • Massachusetts
  • Military Research
  • New England
  • New Mexico
  • New York
  • Psychology
  • Reasoning
  • Systems Engineering
  • United States

Readers

  • Artificial Intelligence
  • Regression Analysis.