The hierarchical logit model is the cornerstone of many real life applications of variable demand modelling around the world. It can be used to model a variety of traveller choices, including mode, destination, time period and frequency. The incremental hierarchical logit model (IHL) is the form recommended in recent WebTAG guidance published by the UK Department forTransport. This is intended for use by practitioners in the UK who are undertaking modelling as part of the appraisal of a transport scheme. The work described in this paper concerns two key recommendations in the guidance, namely: demand modelling should be undertaken with trip matrices in production-attraction (PA) format, rather than origin-destination (OD); and that time period choice, if modelled, should appear above destination choice in the hierarchy. Neither of these on their own is particularly onerous,but taken together they cause considerable difficulty. It turns out that it is not possible to have a trip-based PA demand model that meets the above criteria that also takes proper account of outbound and return travel costs, and ensures consistency between outbound and return modes and destinations. One solution would be to use tour-based modelling. This approach has been adopted in many large scale models around the world, including thePRISM model of the West Midlands region in the UK. A drawback of this is the amount of data required to be able to build up a robust picture of thetours undertaken in the model study area. Many of the applications based on the WebTAG guidance will not have sufficient data available. The approach developed by the project team and described in this paper is to supplement the local data with that collected as part of the UK National Travel Survey (NTS). NTS is an annual survey using household travel diaries and provides information on tours for the households in the sample. This is combined with local data on the number of trips taking place in each time period to estimate local tour-making behaviour. In other countries equivalent data is frequently available. One way of looking at this approach is to consider a matrix of time period choice. The different rows represent the different time periods in which outbound travel can take place and the columns represent time periods for return travel. For tour-based modelling it is necessary to know the number of trips in each cell of this matrix, i.e. the number of trips in each combination of outbound and return time periods. If local data is limited typically only the row and column totals of this matrix will be known. However, NTS provides seed values for the individual cells to which iterative proportional fitting (i.e. a Furness or Fratar procedure) is then applied to meet the local row and column totals. The result is a best estimate of local tour-making. For the covering abstract see ITRD E137145.
Abstract