Nonresponse

is a major challenge to the quality of survey data. Especially for panel

surveys, the ever decreasing number of participants in the course of the panel

and the possible nonresponse bias associated with it is an unsolved problem.

The project consists of two parts: "Understanding Nonresponse" and

"Counteracting Nonresponse Bias".  

In the

first part, studying the GESIS panel, we will examine the extent to which

characteristics of the person and the survey influence participation in the following

wave. In particular, we analyze whether different characteristics affect

nonresponse at different points in the panel life-cycle, whether the direction

or magnitude of the effects change and whether the characteristics depend on

each other. Due to the large number of (potentially correlated) characteristics,

we use machine learning methods for our analysis. The use of double machine

learning methods allows the unbiased estimation of causal relationships. These

are estimated to be dynamic and heterogeneous. Suitable methods are (further)

developed and published in the form of R and Python packages. We will examine

to what extent the results from the GESIS Panel can be transferred to the

German Internet Panel.

In the

second part of the project, wetest the potential of various interventions

(e.g., additional monetary incentive, emailing an additional thank-you letter between

the waves, emailing summaries of study results between the waves) in chancing

response behavior. We will develop a targeted design to allocate the

interventions to the respondents in such a way that the differences in

participation behavior between respondents are reduced as much as possible (by

minimizing the variance of the predicted response propensities).