Generating an Artificial German Census Population

This technical report details the simulation of an artificial German census population. The dataset was created to estimate the performance of a register census in Germany, planned for 2031, which is based on linking various registers using record linkage methods. The simulation uses 2011 German census data and contains 80,206,946 geolocated records at the 100x100m grid-cell level. The primary focus was to simulate quasi-identifiers (QIDs), such as names and addresses, used for linkage, and their dependencies, which increase linkage difficulty. A novel method is used to allocate the population to 3,177,723 grid cells, which fulfils various regional constraints at the grid cell and municipality level. This base population is enhanced by other data sources to simulate further QIDs, such as full name, date of birth, place of birth, and nationality. Since obtaining data on QIDs in Germany turned out to be the major obstacle, several commercial data sets are used. Finally, we describe the data corruption models used to generate the simulated census registers, including data entry and coverage errors.