For the European Census Round 2010 the methodology for the census in Switzerland has been changed towards a register-based census. In order to obtain population values not available from registers, an additional sample, the Structural Survey, is gained. The population register contains mainly demographic variables which have low predictive power for many variables of interest for instance unemployement or mother tongue. Besides the population register further registers exist. However, these may be matched only on aggregate level, e.g. due to disclosure reasons. The inclusion of this aggregate register information with partly very high predictive power for the variables of interest may improve small area estimates by far.
Within a large design based Monte Carlo simulation study different small area estimation approaches to incorporate aggregate covariates will be assessed and compared. The data used is the full Swiss Census 2000 data set. The focus lies on the estimation of population counts and means. The studied variables are extremly distributed along the areas, such as very low proportion in most areas, or very high proportion in many areas and no counts in others. This work is done within the research project “Simulation of the Structural Survey” for the Swiss Federal Statistical Office.
Keywords: Small area estimation; Aggregated register covariates; Binomial model; Swiss Structural Survey
Biography: Mr. Jan Pablo Burgard is a research and teaching assistant and PhD Student at the Economic and Social Statistics Department at the University of Trier. His main research interest lies in small area statistics for censuses. He is currently engaged in several projects related to small area statistics including the project “Simulation of the Structural Survey” aiming on the evaluation of different small area estimation methods for the application in the Swiss Structural Survey.