Partial registration plate surveys are a commonly used tool of both the transport planner and those in research, being used - for example-to estimate a (mean) origin- destination matrix and/ or to study journey time variability. Collecting only a part of (rather than the whole) registration number, whilst succeeding in reducing the risk of incorrectly recorded data, introduces uncertainty into the problem of subsequently 'matching' data from different observation points. Having reviewed existing techniques for such analysis, the paper briefly discusses a number of new algorithms (1), which are characterised by their use of data on times at which vehicles pass the observation points to improve the accuracy of the matching process, within a statistical framework. Two of the new methods - each of which is able simultaneously to analyze data from any number of origins and destinations - are described further in relation to an example. A number of other issues are then discussed, such as modifying the matching process to allow for observer error; correcting for possible bias caused by vehicles passing one observation point inside and one outside the survey period; partial registration numbers occurring with different frequencies (both their estimation and incorporation in the matching stage); checking model assumptions; the role of prior information; and modelling dynamic effects. The discussion throughout is non-technical, with the focus on intuitive reasoning and on the needs of the practitioner.
Abstract