Conferences

Statistics Netherlands conducts many surveys of the general population. Till a few years ago these were conducted in one data collection mode, some like (the first round of) the Labour Force Survey were CAPI surveys, others were CATI. For both types sample frames were available. The last decades CAPI surveys suffer from ever increasing non-response rates and the quality of CATI surveys is threatened because of growing undercoverage caused by unlisted numbers.

Besides these non-response and undercoverage problems rapidly increasing costs of especially CAPI surveys prompt to look for new ways of collecting data. Single mode web surveys do not qualify because of undercoverage, self selection bias and very low response rates. Theoretically mixing modes of data collection (here defined as using (almost) the same questionnaire in different modes) can solve the undercoverage problems, can raise response rates by approaching respondents in their preferred mode and reduce costs by collecting part of the data in cheap modes.

Several experiments have been conducted by Statistics Netherlands to investigate how to design large-scale surveys that use different data collection modes. A design that is extensively tested is approaching respondents with an introduction letter asking them to log in on a web site where they can fill in a questionnaire. Non-response is then approached by telephone if a telephone number can be found and approached by a field interviewer if no number can be traced. Another mixed mode design uses only two modes: CAPI and CATI. Respondents are approached by phone if they have a listed number. If not, they are approached by CAPI.

Both designs have advantages and disadvantages. The first uses three modes and will show greater mode effects, especially because in one mode no interviewer is present. The absence of an interviewer on the other hand reduces costs enormously. The CATI-CAPI design can be expected to have smaller mode effects, but is more expensive and will become still more expensive if the number of unlisted numbers will continue to increase in future. In our paper we discuss these advantages and disadvantages based on the experiments that were carried out.