The Basic Bayesian Blunder

Abstract

Direct Inference was distinguished from Inverse Inference early in the development of mathematical statistics. Direct Inference was the form of uncertain inference that took as premise a distribution in a population, and yielded a (probable) conclusion about the composition of a sample from the population. Inverse inference was to take as a premise the composition of a sample, and yield as a conclusion a (probable) conclusion about a distribution in a population. Direct inference seemed problematic. But inverse inference seemed to be needed to obtain the general premised needed for direct inference. Inverse inference proper is based on Bayesian principles. This paper argues that these principles are inconsistent with direct inference. It is concluded that we should hold fast to direct inference, and accept Bayesian procedures only when they can be put into the frame work of direct inference.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 1989
Accession Number
ADA250978

Entities

People

  • Henry E. Kyburg Jr.

Organizations

  • University of Rochester

Tags

Communities of Interest

  • Autonomy

DTIC Thesaurus Topics

  • Abstracts
  • Artificial Intelligence
  • Binomials
  • Classification
  • Data Science
  • Errors
  • Frequency
  • Information Science
  • Intervals
  • Measurement
  • Measuring Instruments
  • Optical Scanning
  • Philosophy
  • Probability
  • Psychology
  • Reasoning
  • Statistics

Readers

  • Regression Analysis.
  • Strategic Security Studies

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference