Estimating the measurement effects of mixed modes in longitudinal studies: Current practice and issues.
Jul 25, 13:45
It has become increasingly popular to conduct surveys using multiple modes of data collection, such as face to face, Web or telephone. By using a combination of modes it is hoped to decrease non-response and reduce survey costs. Longitudinal surveys have also implemented such mixed-mode approaches. For example, the UKHLS is currently using a sequential Web – Face to Face design for part of the sample.
While mixed modes are exciting they also come with some complications. One of these is measurement error and how this is related to different modes. From previous research we know that different modes can impact measurement error in different ways. For example, respondents might be more honest in a self-administered mode than in an interviewer-administered mode. Trends in sensitive topics can therefore be distorted if mode switches occur. Because of this, concerns have been raised that mixed modes can increase measurement errors or can make the survey results incomparable with others.
In this chapter we discuss the current practices regarding mixed modes and how this could impact measurement error in longitudinal data. We will start by discussing the most common mixed-mode designs used in longitudinal studies together with their strengths and weaknesses related to measurement error. We will use this opportunity to discuss how mixed modes can impact estimates of change. We will also discuss both the designs and statistical methods currently used to estimate mode effects on measurement error in longitudinal studies and highlight their different strengths and weaknesses. Finally, we will illustrate some of these approaches using the UKHLS Innovation Panel.