The representativity of a panel for the general population

Presenter: Annamaria Bianchi, University of Bergamo, Italy

Author: Annamaria Bianchi

Co-author(s): Silvia Biffignandi

The validity of the estimates in longitudinal panels depends on the extent to which the sample remains representative of the target population over time. Several elements may potentially undermine the representativity of panels and be causes of bias in the estimates: initial non-response during the recruitment phase, non-response to subsequent surveys, and attrition. Moreover, decreasing response rates are reported for household panels in the last years. It is therefore essential to study the representativity of panels at different steps and to monitor their representativity in different stages of their construction and in the course of their life. Several indicators have been proposed in the literature. The most widely used one is the response rate. However, theory and empirical evidence show that it has a limited relation to nonresponse bias. More reliable indicators are based on the use of auxiliary variables.

In this talk, we first revise different concepts of representativity in the framework of panel studies. Next, we propose an original approach to study representativity of a household panel for the general population based on analyses carried out at different steps and the use of R-indicators. The proposed approach is applied to the first four Waves of Understanding Society. The analysis considers demographic variables, paradata collected by interviewers in the field and ‘psychographic’ variables known to be related to survey participation (political interest, personality traits, health). At each step it is found that attrition slightly affects representativity. More difficult subgroups over time are identified. The information provided could be useful to improve data collection in the next stages, the design of advance and reminder letters, interviewer training, and paradata collection.