The European Social Survey, now in its fourth round, insists on face-to-face interviewing in all countries to encourage equivalence of data. But with response rates under increasing pressure and fieldwork costs increasing, alternative data collection modes (including mixed modes) are now under active consideration.

Any change in mode, whether to another single mode or a mixed mode design, may have an impact not only on cross-national comparability but on the time series. As a result the ESS is undertaking a programme of research into mixed mode data collection in order to ensure that, if a change in data collection mode is made, the effects are known and minimised. The first phases of research examined the extent and cause of differential measurement error across modes. The focus then turned to the feasibility of conducting the ESS, an hour-long questionnaire, by telephone, and the effect of length of interview on response rates and data quality.

The latest piece of research investigates the feasibility of using mixed mode data collection on the ESS, a general population survey. The type of mixed mode design being considered in this experiment is one where different members of the same sample answer the same questions in different modes. The research focuses in particular on the response rates, costs and data quality of two different mixed mode designs; one sequential and one concurrent, compared with the main face-to-face survey. The concurrent design offers respondents the choice of mode (face-to-face or telephone interview or web self-completion) while in the sequential design respondents are offered the cheapest mode (web self-completion) first, followed by telephone and then face-to-face interviews. These designs offer the possibility of reducing (or at least better managing) costs, increasing response rates and reducing nonresponse bias.

This paper will describe the experimental design and discuss the practical implications of a possible future switch to a mixed mode design, such as questionnaire adaptation, available sample frames, and contacting sample members. The paper will then present some initial results, including response rate comparisons.