Mixing survey modes appears almost inevitable today. Three important reasons to use a mixed-mode survey design are improving coverage, increasing response rates and reducing costs. However, there are also potential drawbacks, such as, increased administrative and logistic burden, and potential for mode specific measurement error.
I discuss several mixed-mode designs, summarize the empirical evidence for reducing coverage and nonresponse error and then focus on measurement error.
I review three related issues in mixed-mode survey practice: design, diagnosis, and adjustment. The first step in a mixed mode survey should be to design the survey in such a way that it minimizes mode measurement effects. I will give some empirical examples. In the second step, one has to investigate to what extent apparent differences between modes are the result of intended differential selection of respondents to different modes, and hence help to reduce coverage error. This is followed by estimating the (unwanted) mode measurement effect while controlling for the selection effect. If there are unwanted mode effects in the measurements, adjustments are needed in the analysis phase. The emphasis will be on design and diagnosis, if there is time I will discuss some adjustment procedures and illustrate these with examples.
Presented by:
Edith de Leeuw (University of Utrecht)
Date & time:
November 16, 2015 4:00 pm - November 16, 2015 5:30 pm
External seminars home