A data-driven approach to monitoring data collection in an online panel

Résumé 0

Longitudinal or panel surveys suffer from panel attrition which may result in biased estimates. Online panels are no exceptions to this phenomenon, but offer great possibilities in monitoring and managing the data collection phase and response-enhancement features (e.g., reminders), due to real-time availability of paradata. This paper presents a data-driven approach to monitor the data collection phase and to inform the adjustment of response-enhancement features during data collection across online panel waves, which takes into account the characteristics of an ongoing panel wave. For this purpose, we study the evolution of the daily response proportion in each wave of a probability-based online panel. Using multilevel models, we predict the data collection evolution per wave day. In our example, the functional form of the data collection evolution is quintic. The characteristics affecting the shape of the data collection evolution are characteristics of the specific wave day and not of the panel wave itself. In addition, we simulate the monitoring of the daily response proportion of one panel wave and find that the timing of sending reminders could be adjusted after 20 consecutive panel waves to keep the data collection phase efficient. Our results demonstrate the importance of re-evaluating the characteristics of the data collection phase, such as the timing of reminders, across the lifetime of an online panel to keep the fieldwork efficient.