Abstract

Surveys based on random samples with a clearly defined sampling frame and calculable inclusion probabilities are increasingly faced with decreasing or constantly low response rates and increasing survey costs. In contrast, there is a renaissance of non-random samples, e.g. based on surveys with online access panels or river samples, which can generate high case numbers at a comparatively low cost. However, the problem with non-random samples is that unbiased estimation strategies are associated with (strong) model assumptions that cannot be tested in practice without additional information. Non-random samples are also suspected of being heavily affected by data quality problems such as high measurement errors and duplicates due to multiple measurements of the same elements.