German Register Data for Regression Estimation in Survey Sampling - A Study on the German Microcensus Respecting for Data Protection

Modern survey sampling methods generally deal with improving estimators with respect to available prior information. An increasingly important source of prior information is data from registers. In some countries, e.g. Canada, The Netherlands, and the Scandinavian countries, personal identification numbers in registers generally make it possible to improve estimation processes mainly due to the exact match of units between register and estimation variable. In many other countries, this exact matching of information is not possible, e.g. for legal reasons. In the German Micro-census, however, the variable unemployed refers to the register data of the Bundesanstalt für Arbeit in Nuremberg which are not permitted to be matched to each other for legal reasons. This article deals with an improvement of estimators with respect to available auxiliary information within the survey to enable an appropriate use of aggregated register information. A Monte Carlo simulation study will allow for the comparison of the estimators with different information or matching levels. This will be achieved in a practical environment using the data of the German Microcensus which is a 1 % survey sample of the German population. The given example yields recommendations on applying the methodology to similar cases that are influenced by data protection in order to allow for improved estimates in practice.