In this paper we discuss the recent paradigm change in survey research. Where survey research so far was based on probability sampling, nowadays many surveys are based on self-selected access panels. This means that the statistical basis for generalization from the sample to the population has gone. Some alternative procedures, like weighting based on propensity scores, have been suggested to correct for possible biases. However, the empirical studies done to compare the results of access panels and probability samples are so far inconclusive. Sometimes large differences are found and it was not possible to correct for the differences, while at other times such differences do not surface.

Information from surveys plays a role in several different political processes such as voting, participation in elections, agenda setting, opinion forming about issues; in general survey play a role in any process which involves a reaction to the perceived general political orientation of the public.

Given that one can not say at this moment whether an access panel provides proper information about characteristics of the population, one runs the risk that political processes are influenced by incorrect information based on access panels.

This leads us to the conclusion for the moment to avoid access panels and rely on probability samples or panels based on a probability sample. Furthermore, we conclude that further research is necessary to study the reasons for participation in panels and possible correction methods for panel bias such as (propensity score) weighting, imputation, likelihood-based estimation, etc

.