PREDICTING BEST WITH IMPERFECT DATA

THE EFFECT OF ERRORS AND THEIR PROPAGATION IN MODELS FOR PREDICTION IS DISCUSSED AND SUGGESTIONS MADE FOR STRATEGIES FOR THE SELECTION AND CONSTRUCTION OF MODELS WHICH ARE INTENDED FOR APPLIED WORK. TWO TYPES OF ERROR ARE DISTINGUISHED: ERROR OF MEASUREMENT AND ERROR OF SPECIFICATION. A QUANTITATIVE MODEL PUTS TOGETHER VARIOUS NUMBERS OBTAINED BY MEASUREMENT, AND COMBINES THEM THROUGH ALGEBRAIC OPERATIONS. A WELL-KNOWN FORMULA EXISTS FOR ESTIMATING THE ERROR IN THE OUTPUT WHICH RESULTS FROM THE PROPAGATION OF ERRORS IN THE INPUT. THIS FORMULA IS PRESENTED AND DISCUSSED. SIMPLE AND COMPLEX MODELS ARE DISCUSSED AND EXAMPLES GIVEN. A STATISTICAL ARGUMENT IS ADVANCED FOR A DISTINCTION BETWEEN MODELS FOR FUNDAMENTAL RESEARCH AND MODELS FOR APPLIED WORK. IF THERE IS MERIT TO THE STATISTICAL ARGUMENT, IT FOLLOWS THAT WE SHOULD HAVE RESEARCH GROUPS IN UNIVERSITIES AND OTHER CENTERS WORKING ON COMPLEX MODELS, WHILE OPERATIONAL AGENCIES WOULD BE WORKING WITH SIMPLIER AND SAFER MODELS. A TECHNIQUE IS SUGGESTED FOR ESTIMATING THE VALUE OF IMPROVEMENTS IN DATA.

  • Supplemental Notes:
    • Vol 34, No 4, PP 248-255, 2 FIG
  • Authors:
    • ALONSO, W
  • Publication Date: 1968-7

Media Info

Subject/Index Terms

Filing Info

  • Accession Number: 00240519
  • Record Type: Publication
  • Files: TRIS
  • Created Date: Aug 28 1994 12:00AM