Pivot-Point Procedures in Practical Travel Demand Forecasting

For many cities, regions and countries, large-scale model systems have been developed to support the development of transport policy. These models are intended to predict the traffic flows that are likely to result from assumed exogenous developments and transport policies affecting people and businesses in the relevant area. The accuracy of the model is crucial to determining the quality of the information that can be extracted as input to the planning and policy analysis process. A frequent approach to modelling, which can substantially enhance the accuracy of the model, is to formulate the model as predicting changes relative to a base-year situation. Often, base-year traffic flows can be observed rather accurately and the restriction of the model to predicting differences reduces the scope for errors in the modelling – whether they be caused by errors in the model itself or in the inputs to the model – to influence the outputs. Such approaches are called ‘pivot point’ methods, or sometimes incremental models. The approaches have proved themselves beneficial in practical planning situations and now form part of the recommended ‘VaDMA’ (Variable Demand Modelling Advice) guidelines issued by the UK Department for Transport. While the principle of the pivot point is clear, the implementation of the principle in practical model systems can be done in a number of ways and the choice between these can have substantial influence on the model forecasts. In particular modellers need to consider: 1.whether the change predicted by the model should be expressed as an absolute difference or a proportional ratio, or whether a mixed approach is necessary; 2.how to deal with apparently growth in ‘green-field’ situations when applying these approaches; 3,at what level in the model should the pivoting apply, i.e. at the level of mode choice, destination choice, overall travel frequency or combinations of these; 4,whether the pivoting is best undertaken as an operation conducted on a ‘base matrix’ or the model is constructed so that it automatically reproduces the base year situation with base year inputs. The paper reviews the alternative approaches to each of these issues, discussing current practice and attempting to establish the basis on which alternative approaches might be established; in particular, whether pivoting is treated as a correction to a model which is in principle correctly specified but incorporates some error, perhaps from faulty data, or as a partial replacement for a model that handles at best part of the situation. These views of the pivoting lead to different procedures. It goes on to present and justify the approach that the authors have found useful in a number of large-scale modelling studies in The Netherlands, the United Kingdom and elsewhere, pointing out the problems that have led to the calculations that are recommended.