Combining information from different survey samples - a case study with data collected by world wide web and telephone

This paper considers an investigation of the users’ opinion of the web site ODIN, which is the official web site for information from the Norwegian Government and Ministries. A survey were performed by asking people who log on to the ODIN web site to fill out questionnaires on their use of the web site. This could be a potentially very useful way to collect information about the users’ needs, since the sample was selected directly from that subpopulation who uses the web site. However, answering was voluntary, and most users did not answer the questionnaire. The large proportion of possibly informative non-response makes it difficult to draw general conclusions from the observed web data only. Therefore, a telephone survey was carried out in addition, with people randomly selected from the whole population in Norway. The information from the web survey and the telephone survey were combined to give reliable results. A crucial part of the statistical modelling was to take into account the different sampling schemes of the two surveys: In the web survey, people were sampled proportionally to their frequency of using ODIN, whereas people in the telephone survey were sampled with equal probability. The statistical approach was Bayesian, and the BUGS software was used to perform inference.