Response Quality in Panel Surveys: Mechanisms and Consequences of Panel Conditioning

Abstract

Panel surveys are well-established data infrastructures in the social sciences, enabling researchers to study individual-level dynamics of stability and change over time. However, the value of these insights depends on the quality of the underlying panel data. One potential threat to panel data quality is panel conditioning, that is, changes in respondents’ attitudes and behaviors, or in the way in which they are reported, caused by repeated interviewing itself. Despite growing methodological attention, questions remain regarding the magnitude, mechanisms, and consequences of panel conditioning. This dissertation investigates panel conditioning and its implications for panel data quality, focusing on three potential mechanisms: (a) increased awareness and reflection through repeated exposure to survey topics, potentially leading to more reliable and stable attitudes (i.e., the cognitive stimulus hypothesis), (b) changes in survey satisficing, and (c) changes in socially desirable responding with repeated interviewing. To disentangle panel conditioning effects from other sources of bias and true change over time, the dissertation uses quasi-experimental and experimental research designs. It primarily draws on data from a longitudinal survey experiment fielded in a probability-based panel and a nonprobability online panel in Germany. The experiment manipulated the frequency with which respondents received identical questions (i.e., conditioning frequency) across six panel waves. Results show that repeated exposure to questions on abortion increased the reliability of reported abortion attitudes and the stability of latent abortion attitudes, consistent with the cognitive stimulus hypothesis. Survey satisficing, by contrast, remained largely unaffected by repeated interviewing. Although selected satisficing response behaviors occurred at substantial levels across panel waves, they did not systematically increase or decrease with repeated survey participation or repeated exposure to identical questions. One exception was speeding, which was more pronounced among experienced panel respondents, suggesting lower engagement and reduced response effort with repeated interviewing. Finally, contrary to the assumption that repeated interviewing reduces social desirability bias, the findings show no general reduction in answer refusal or socially desirable responding to sensitive questions. Instead, changes varied across questions involving different sensitivity concerns, with both increases and decreases in socially desirable responding. Overall, this dissertation shows that panel conditioning effects are highly outcome- and context-specific and do not follow a single, uniform pattern. The findings underline the importance of considering panel conditioning when evaluating panel data quality and interpreting longitudinal changes in respondents’ (reported) attitudes and behaviors.