MINIMIZING ERROR IN AGGREGATE PREDICTIONS FROM DISAGGREGATE MODELS

This paper presents empirical tests of aggregated prediction error on a sample of work-trip mode choices for the San Francisco area and systematic criteria for choosing classification variables. It introduces a more efficient utility scale classification criterion for aggregated prediction. Aggregation error is found to be much larger than previous tests have indicated. The choice of classification variables that produces the smallest error is found to vary with the scale of the prediction aggregates. Level-of-service variables are more important for large aggregates, socioeconomic variables for smaller. Classification of the sample based on the scales of the total utility of the explanatory variables in each alternative is found to be much more efficient in error reduction than classification by individual variables. /Author/

Media Info

  • Media Type: Print
  • Features: References; Tables;
  • Pagination: pp 59-65
  • Monograph Title: Transportation forecasting and travel behavior
  • Serial:

Subject/Index Terms

Filing Info

  • Accession Number: 00195992
  • Record Type: Publication
  • Files: TRIS, TRB
  • Created Date: Sep 15 1979 12:00AM