Abstract

Panel conditioning is an important source of measurement error unique to panel surveys. It refers to changes in survey responses due to respondents’ participation in a panel study in which they are interviewed multiple times over several waves. Previous research on panel conditioning has shown a relationship between the number of times respondents were interviewed and their responses. However, a major weakness of (and challenge to) the existing research is its inability to distinguish between changes to respondents’ true behaviors and changes in reports of the true behaviors. Thus, the existing studies are heavily reliant on assumptions and models when examining changes over time. This paper studies panel conditioning effects using data from five waves of a large German panel survey on labor market outcomes (PASS). Because administrative data on employment and unemployment are available for nearly all respondents, we can tease apart panel conditioning caused by changes to true values and panel conditioning due to changes in self-report without depending on assumptions. Our results are, as a result, more robust than those provided in earlier studies.