Using the Statistical Fitting of Trip Distribution Models to Examine the Validity of Some Common Modelling Assumptions

Using the Statistical Fitting of Trip Distribution Models to Examine the Validity of Some Common Modelling Assumptions


P Emmerson, TRL, UK


This paper looks at some practical uses of a statistical approach to fitting of trip distribution models. In particular it considers what the approach can tell us about the validity of common types of distribution models.


TRL has been developing strategic transport models for a number of regions in the United Kingdom over the past two years: in Essex in South-east England and in the regions around Glasgow and Leeds. One of the main data inputs has been the 2001 Census Journey to Work data. The characteristics of this data set which is split by mode and available at a very fine spatial detail, have been altered by a series of data processes which aimed to preserved confidentiality. Consequently they have given the data-set some degree of statistical uncertainty despite its nominal 100 percent sample.
One of the uses to which the data-set has been put has been to derive trip distribution functions for our transport models. Recent work by John Shrewsbury on using statistical estimation techniques (Generalised Linear Models) on the Wellington (NZ) travel survey data has proved an inspiration to using such a technique to produce 'best-fit' trip distribution models for our transport models.
Whilst the paper will provide an overview of the theory and application of this technique to fitting models to commuting patterns, its main focus is on what the technique can help show in relation to the statistical validity of commonly used trip distribution functions within the UK. In particular, it focuses on the relative fit of exponential and power functions and their combination (Tanner or Gamma function) and it also looks at their spatial variations over the modelled area. Such variations have been found in previous research but the paper will look particularly at the common technique of dividing the model area into sub-areas and fitting functions separately to each. This approach can ensure that the scale of trips crossing from one area to the next matches that observed and that the observed mean trip cost matches but there can be serious discrepancies in the trip cost distributions. The paper presents the relative statistical fit for a variety of such models and discusses the implications of the findings for fitting trip distribution models for practical use.


Association for European Transport