[Zurück]

V. Puchner:
"Evaluierung von Methoden des Statistischen Matchings und ausgewählte SAE Methoden mit Anwendung auf Mikrozensus und EU-SILC";
Betreuer/in(nen): M. Templ; E107 Institut für Statistik und Wahrscheinlichkeitstheorie, 2013; Abschlussprüfung: 11/2013.

Due to the fact that poverty estimations on regional level on basis of EU-SILC samples is not of adequate accuracy, the quality of the estimations should be improved by additionally incorporating Micro Census data for estimation. In comparison to EU-SILC, the Micro Census survey data consists of more observations. However, income is not questionaired but necessary to estimate poverty. The aim is to find the "best" method for the estimation of poverty in terms of small bias and small variance. Therefor an artificial "close-to-reality" population is simulated in order to know the "true" parameter values. To make an assessment of the quality, considering the respective sample designs, EU-SILC and Micro Census samples are drawn repeatedly. Variables of interest are imputed into the Micro Census data sets with the help of the EU-SILC samples through regression models including selected unit-level small area methods and statistical matching methods and poverty indicators are estimated in the following. The bias and variance for the direct estimator and the several methods are evaluated and compared. The variance is desired to be reduced by the larger sample size of the Micro Census. In conclusion, the result is that it doesn't exist only one method performing by far best in terms of bias and variance among the used models. Concerning the bias, most often the statistical matching methods perform better than the regression methods, but regarding the variances, the regression models do a better job in general. In terms of the average mean squared error of states the direct estimator performs best, followed by logistic (mixed) regression models and all the statistical matching methods.