Data yield as a function of survey design: Analysis of some long-distance surveys

Abstract

The travel survey designer has to balance the wish for a comprehensive set of data items with the willingness of respondents to participate in a survey of a given duration and complexity and if participating, the number of journeys, the respondents will describe. This interaction between work load, response and data yield has been the subject of only a few published studies in spite of ist fundamental importance (Axhausen et al., 1997; Ampt and Polak, 1996). A recent series of surveys of long-distance travel behaviour provides the opportunity to analyse this issue in the context of a multi-national study. These pilot surveys were undertaken for the project „Methods of European Surveys of Travel Behaviour“ funded by the EU 4th Framework Programme. y The MEST project aims to develop a benchmark survey of long-distance travel behaviour, which can be applied across Europe with questions understood in all member states of the European Union, the prerequisite to the cost-efficient collection of data. A series of three waves of pilot surveys in various countries is the main tool to achieve this aim. For the first time in the transport literature the experiments undertaken here varied the workload of the respondent by varying <br/><br/> -the number of items to be reported about any long-distance journey <br/> -the duration of the survey period<br/> -the temporal orientation of the survey<br/> -the survey instruments (CATI vs. self-completion mail back) <br/> -the layout of the written instrument (one page for each part of the journey vs. one column) <br/> -the survey unit (trips vs. stages). <br/><br/> The surveys were undertaken in Sweden, the UK, France and Portugal. The survey protocols included both the combination of CATI with mail-back non-response surveys, as well as the more usual mail-back survey with CATI-non-response surveys. The paper will present how the data yield, i.e. the number of journeys, trips and stages reported depend on the complexity of survey, the survey protocol and the willingness of the repondents to participate (measured as the difference between reponse in the initial approach and the non-response approach). The modelling framework will be a system of Negative binomial/Poisson - models to describe the data yields at the various levels of analysis and of probit-models to describe the willingness to participate. The results of these surveys will be compared with the results obtained from an earlier Austrian experiment, which had for example shown that data yield at the level of the number of journeys is not affected by the survey complexity, but that it is affected at the level of the number of stages per journey.