Non-response and data yield: Experiences from Austria and France

Abstract

In many surveys, for example of crime, illness, travel, purchasing etc, the amount of information requested from the respondent is not known apriori to the survey designer. In situations like this, the design of the survey will aim to assure both the largest possible response, but also a significant data yield, i.e. reports of the survey object, for example episodes of illness, number of victimizations, number of trips undertaken, objects bought. The design process has therefore two main objectives and it not clear that one can achieve both at the same time. This paper will report the analysis of two sets of experiments carried out in 1996/97 in Austria and France, which aimed to provide insight into these trade-offs. The surveys were concerned with the amount of long-distance travel undertaken by the population. The experiments in Austria involved eight postal surveys, which varied the temporal orientation of the survey, the reporting period and the amount of detail requested about each movement, while leaving the other elements of the survey unchanged. The protocol involved a detailed telephone interview of those not responding in writing, which captured the essential items from the written survey. The eight experimental surveys in France varied different elements: the sample frame (fresh random sample versus an existing panel of respondents), the data retrieval method (CATI versus postal) and the number of contacts (reporting twelve weeks either in one contact or in three monthly intervals). The paper will report the disaggregate modelling of the response behaviour and of the data yield, as a function of the experimental variables and of the socio-demographics of the respondents. The results indicate that there are substantial interactions and that certain variables increase response, but lower the data yield.