Procedures for the derivation of standard errors for parameters and fitted values are considered with reference to a particular formulation of the gravity model. Methods based on maximum likelihood estimation are found to have certain disadvantages associated with the large number of parameters typically involved and poor goodness-of-fit, probably due to the omission of pertinent factors from the model. Alternative empirical procedures are described, with particular attention given to a method based on the use of the Monte Carlo simulation. Two small example data sets are used to compare three procedures, one based on maximum likelihood estimation, a second on a approximation to maximum likelihood estimation under the assumption of a Poisson error structure, and the third on Monte Carlo simulation. The first example data set is designed to contain Poisson variates and the second to contain non-Poisson variates with a variance-to-mean ratio larger than one. For the covering abstract of the symposium see TRIS 452544. (Author/TRRL)

Media Info

  • Features: Figures; References; Tables;
  • Pagination: p. 65-90

Subject/Index Terms

Filing Info

  • Accession Number: 00452549
  • Record Type: Publication
  • Source Agency: Transport Research Laboratory
  • Files: ITRD, TRIS
  • Created Date: Jul 31 1986 12:00AM