An earlier presentation by the authors at the ESOMAR 2006 Panel Conference identified that a hidden bias exists as a result of respondents belonging to multiple online panels. This presentation provides a remedy to remove this bias when preparing research outputs. The procedure, known as non-parametric modeling using CART, provides an elegant solution to this issue. Other weighting processes are too simplistic to model all possible effects simultaneously.
We are presenting a paper on the representativeness of panellists who are in multiple panels as compared to those in one or a limited number of panels. Our paper assesses the data received from respondents on only one, a few, and multiple panels and compares the same data with information collected by CATI. We have found major differences in the attitudes of those who participate on multiple panels, while the demo-graphic may be similar, as compared with others in only one panel.in some instances the attitudes of those in only one panel are closer to those who were interviewed by CATI than the attitudes of those on multiple panels. Case study material will be presented using a range of questions relating to attitudes and behaviour, to test the hypothesis that people on multiple panels are not representative of the population.the data has been extensively tested for significance and exhibits strong correlation between multiple panels and certain attitudes. Examples of this difference are that people in multiple panels are more likely to be price sensitive than influenced by brands Price is more important to me than brand names. We found that the deviation from CATI results increased proportionally as the number of panels a respondent belongs to increased.We have identified that there are major differences between people who are members of only one panel, and those who are members of multiple panels. these differences are seen in demographics, attitudes and behaviour.