Using Discrete Choice Estimation to Develop the Long Distance Model for Appraising Two Highway Schemes in Scotland.

Using Discrete Choice Estimation to Develop the Long Distance Model for Appraising Two Highway Schemes in Scotland.


Peter Davidson, Peter Davidson Consultancy, Stephen Cragg, Transport Scotland and SIAS, Collins Teye, Peter Davidson Consultancy


Developing a long distance model for Scotland using discrete choice to estimate mode and destination choice utility function coefficients including logsum and scale parameters so as to dove-tail with the Transport Model for Scotland.


Long distance travel is qualitatively different from normal travel being characterised by a different purpose mix, values of time and behavioural choices. Special attention to long distance travel (ie longer than 100km) is required for schemes for which long distance travel forms the major component of the traffic and Transport Scotland (TS) have two such schemes. They were 160km of the A9 Perth to Inverness and 150km of the A96 Aberdeen to Inverness. For both of these roads, very little traffic was generated (or attracted) to places in-between - most of it travelling from end-to-end or places beyond. These routes are the two key strategic routes to northwest Scotland and of keen importance to the economic coherence of Scotland. The Transport Model for Scotland (TMfS) is TS's main strategic modelling and appraisal tool so, in order to appraise these two major schemes the SIAS consortium was commissioned to build a special long distance scheme model which is the subject of this paper.

Due to timing restrictions, existing data sources had to be used which proved difficult because (as in most normal origin-destination surveys), there were few long distance trips. The long distance model implementation had to be fitted-in to the existing TMfS model framework which covered trip generation, mode, destination, time period and park & ride choice for mainly short distance. The time frame for this development was an exceptionally short three months which placed difficult demands on its achievability.

This paper describes the discrete choice logit model estimation used to estimate nested mode and destination choice coefficients. Destination choice was nested within mode choice and the estimation including the coefficients of the destination choice utility function as well as the nesting (or logsum) coefficient for mode choice. In order to fit the long distance model into (short distance) TMfS, the utility functions had to be scaled using discrete choice estimation to match the short distance parts of the trip matrices. TMfS operates as an incremental model whereby the demand model pivots on the 'observed' trip matrices which had already been validated. The estimated long distance model was implemented within the TMfS run-time framework and its pivots re-calibrated ready for passing over to the A9/ A96 scheme consultants.

The areas of innovation in this paper include the estimation of long distance mode and destination choice model coefficients, logsum coefficients and scale parameters so as to dove-tail with the short distance part of TMfS their model implementation so as to solve a practical long distance scheme appraisal problem.


Association for European Transport