Assembling an access panel: a study of initial nonresponse and self-selection bias


Internet-based household panels are becoming increasingly popular as an instrument to obtain wide-ranging knowledge about the general population. Since the Internet penetration in the general population is large (83% in the Netherlands) and still growing, such panels are often treated as a random sample of the population, and the results of panel studies are commonly generalized to the whole population. In this chapter, we describe the setting up of an Internet panel and the effects of choosing the Internet as the only source of data collection. A random sample of 10,000 individuals was drawn from the general population and invited to join a panel. From the non-re-sponders a second sample of 2000 was drawn and asked to complete a short question-naire including the demographic variables and 9 questions from the first panel survey on health-care reforms. Results indicate that the panel members differ markedly from non-members on almost all demographic variables, and on the knowledge questions. The application of a logistic regression weighting technique resulted in more skewed data after weighting than before. We conclude that if an Internet panel is not a rep-resentative sample of the population, then no weighting procedure will correct for the selection bias.

In Stoop, I., and Wittenberg, M. (eds.). Access panels and online research, panacea or pitfall? Amsterdam, Aksant Publishers