The simple exponential smoothing learning model
- Working Paper
There is an increasing interest in transport in modelling the dynamics of repeated choices, especially in the context of route and departure time choice. Those models adopting microscopic simulation require an assumption about the learning behaviour of the drivers, i.e. about the mechanism which integrates the information obtained from the most recent choice into the expectations of the drivers, which in turn form the basis of the next choice. A number of models have been proposed, but the simple exponential smoothing model: Expectationi = τ Experiencei + ( 1 - τ ) Expectationi -1 has been adopted by a number of authors, some of whom have estimated the value of the smoothing parameter τ using experimental data obtained from a sample of respondents. This paper is using the results from two recent and comparable experiments, one carried out in London and one carried out in Athens, to analyze the distribution of the values of the smoothing parameter calculated at the individual level. The experiments involved the respondents making a series of choices of departure time and route, once between two competing bus routes and once between competing roads. The travel experience resulting from that choice was visualized on screen to convey as vivid an impression of the experience as possible to the respondents. As part of the experiments the respondents were asked to state their travel time expectations three times during the intervals providing data which is not available in this form from previous experiments. The results of the individual level analysis showed that the distribution of the τ-values is mostly trimodal, implying three groups of respondents, one ignoring the past, one ignoring the current experience and one integrating the two elements. The results showed also that the respondents used different τ's over time reflecting their experiences and choices. The implications of these results for the simulation of recurrent choices are twofold. First the simple model is not suitable until it can be shown that the difference between the modelling assumption of an unimodal distribution of τ and the true trimodal distribution does not affect the results. Second there is need to include models of the choice or development of τ into models to recreate the response of the travellers. Show more
PublisherCentre for Transport Studies, Imperial College London and Department of Civil Engineering, National Technical University Athens
Organisational unit03521 - Axhausen, Kay W. / Axhausen, Kay W.
02226 - NSL - Netzwerk Stadt und Landschaft / NSL - Network City and Landscape
02655 - Netzwerk Stadt und Landschaft D-ARCH
MoreShow all metadata