Pivoting from a known base when predicting choices using logit models

Logit models are widely used by transport practitioners for forecasting. It is important that such models make appropriate use of existing independent base data, a process often referred to as “pivoting” and recommended by the UK Department for Transport’s Transport Appraisal Guidance. In the transport context, such base data typically relates to mode and destination shares. The general aim of pivoting is to produce a forecast under changed circumstances while maintaining compatibility with a reliable base position. Various methods for pivoting are available, and the paper investigates three of them in the context of different logit models (MNL, NL and CNL), illustrating them using a simple example of mode and destination choice. For the simplest MNL model, the three methods are essentially equivalent, but they start to diverge as the models become more complex. Although it will not always be the most convenient approach, depending on the software implementation, the “residual disutility” method would seem to be able to deal satisfactorily with all the cases investigated. It is recommended that software be developed to deal with some of the currently less tractable cases.