Prediction, Diagnosis, and Casual Thinking in Forecasting.

Abstract

While forecasting involves forward/predictive thinking, it depends crucially on prior diagnosis for suggesting a model of the phenomenon, for defining 'relevant' variables, and for evaluating forecast accuracy via the model. The nature of diagnostic thinking is examined with respect to these activities. We first consider the difficulties of evaluating forecast accuracy without a causal model of what generates outcomes. We then discuss the development of models by considering how attention is directed to variables via analogy and metaphor as well as by what is unusual or abnormal. The causal relevance of variables is then assessed by reference to probabilistic signs called 'cues to casuality'. These are: temporal order, constant conjunction, contiguity in time and space, number of alternative explanations, similarity, predictive validity, and robustness. The probabilistic nature of the cues is emphasized by discussing the concept of spurious correlation and how causation does not necessarily imply correlation. Implications for improving forecasting are considered with respect to the above issues. (Author)

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 03, 1981
Accession Number
ADA104372

Entities

People

  • Hillel J. Einhorn
  • Robin M. Hogarth

Organizations

  • University of Chicago

Tags

Communities of Interest

  • Biomedical
  • Human Systems

DTIC Thesaurus Topics

  • Air Force
  • Applied Psychology
  • Behavioral Sciences
  • Biomedical Research
  • Business Administration
  • Cognition
  • Commerce
  • Engineering
  • Human Factors Engineering
  • Information Science
  • Judgment
  • Military Research
  • New York
  • Operations Research
  • Psychology
  • Students
  • Systems Engineering

Readers

  • Artificial Intelligence
  • Life Cycle Cost Analysis
  • Theoretical Analysis.

Technology Areas

  • Space