“Time Series Modelling in Repeatedly Conducted Sample Surveys”

To put it in a nutshell, the subject matter of this PhD thesis is improving accuracy and comparability of official statistical figures. Statistics are usually compiled based on (random) samples. The bigger the sample size, the closer to the true value the estimate (e.g., the average) is expected to be. Sometimes, the sample size is very small, such that the estimate becomes useless (think of an estimated unemployment rate of 3%±4% at a 95% confidence level). Apart from that, estimates become incomparable before and after the survey redesign. Time series models can be used to solve both problems without resorting to (expensive) additional interviewing. The gained efficiency may allow us to reduce the sample size twice or even thrice.

Key words: hierarchical Bayesian approach; Kalman filter; multilevel models; small area estimation; state space models; variance reduction.