The data quality of access panels can be threatened by self-selection processes into the panel and mode/response effects. Self-selection is likely to lead to biased sample estimates, while mode effects and mode-specific response effects preclude any generalisation of outcomes produced by one survey mode to another. To study both types of effects, we built up a large access panel for the adult population of Germany using probability sampling for the recruitment of people by phone (landline and cell phones). Possible access panel modes are landline, cell phone and Internet. The project is part of the Priority Programme 1292 on ‘Survey Methodology’ (see www.survey-methodology.de)

The present paper is focused on an analysis of panel participation. In a first step, the propensity to response to the initial survey request (participation in the recruitment interview) is modelled using an extensive set of paradata and mixed-effects logistic regression analysis, controlling for interviewer effects. Then latent variable modelling is used to estimate determinants of follow-up cooperation using an extensive set of paradata, metadata, and survey items as well as the response propensity modelled at step 1. There are two major dependent variables: a) the readiness to participate in the access panel as expressed by the respondent in the concluding part of the recruitment interview. b) Subsequent panel participation when the target persons were re-contacted some weeks later for realizing the actual panel-initializing interviews. As in step 1, here again multilevel analyses were carried out to control for interviewer effects in attempts to realize these panel-initializing interviews. The model of panel participation differentiates between participation in the online and telephone track of the panel.