Two surveys on a rail line provide the basis for deciding the most cost-effective method of drawing inferences of a given accuracy. It is rare to be able to sample randomly from exactly the population of travellers one wishes to draw inferences about. The degree of accuracy with which inferences can be made increases with the sample size, and hence the cost of the sample. However, there is often more than one way to sample, and it may be advantageous to accept some imperfections in the sampling process in order to get a lower unit sampling cost and hence the advantages of a larger sample size. This will depend on whether the imperfections can easily be corrected for. These issues do not appear to be widely taken into account when travel surveys are planned. We have formed recommendations for better practice having undertaken and studied two surveys of travellers on the local rail service between hull and Scarborough. One survey was a cheap self-completion questionnaire whilst the other was on-train interviewer-conducted, using a range of modern data capture devices. The former was subject to suspected large non-response bias, whilst the latter was difficult to gross-up in the absence of complete OD information. The raw data from the two surveys provided very different pictures of the users of the service, whilst a careful grossing-up provided compatible results in respect of most - but not all - characteristics.(a) for the covering abstract of the seminar see IRRD 816404.
Samenvatting