Spatial Interaction Models Using Aggregated Information.

Despite their principal use as predictors of response to changing conditions, most travel demand models are estimated from data collected in single time periods. The fact of high mutual correlation between explanatory variables is perhaps the single greatest problem for the analyst, and has led to increasing use of disaggregate approaches, and a widespread recognition of the need for the 'temporal validation' of the models. Ideally, of course, such models would be estimated from data sets spanning different time periods; in practice the only historic information that is usually available is highly aggregate and incomplete.

For example, the calibration of a conventional mode-splitldistribution model requires the collection of a sufficient set of zone-to-zone flows by each mode to identify the model parameters; such information will not normally exist for other time periods. However, it may happen that for previous years ticketing records and automatic traffic counts do provide information about certain specific groups of these flows. Such information is potentially useful, and may even allow a time dimension to be incorporated in the model specification.

The second major use of conventional models is to Snfill' incomplete data to establish a current pattern of demand; here too estimates of aggregate quantities often exist, and could be used.

In addition to such directly observed groups as flows on roads or revenue collected, there may be independent estimates of other aggregates which can also be used in the model fitting process, or in forecasts. The most familiar of these would be trip-end estimates calculated on the basis of socio-economic characteristics of households in each zone. More recently it has been suggested that the average amounts of time and money allocated to travel, so called 'travel budgets', can also be estimated directly and used to inform more detailed forecasts of travel patterns.

The aims of this paper are as follows; firstly (in Section 2), it will present some recent evidence for the nature of 'travel budgets', and for the implications that the stability of these budgets has for the form of travel demand models. Secondly (in Section 3), it will outline an approach by which all of the forms of data described above may be used during the model fitting, together with a corresponding algorithm which applies to the most common form of spatial interaction model, the logit model.