Abstract (eng)
The thesis generally deals with a parametric estimator of the at-risk-of-poverty rate, assuming lognormally distributed income data. In the first chapter, state-of-the-art nonparametric estimation of that indicator is reviewed and the motivation for developing an alternative estimator is explained. Chapter 2 deals with parametric modelling of income distributions as well as the basic properties and main advantages of the lognormal distribution. In the following three chapters formulae for parametric point and interval estimators and their asymptotic properties are derived, and in chapter 6 the applicability of the large-sample results in the small-sample case are examined in a Monte Carlo study. Simulation is also used in the subsequent chapter to compare the performance of the parametric with the nonparametric estimator applied to real life datasets. The last two chapters in the thesis deal with variance inflation caused by estimating for persons in households as well as by complex sampling designs and nonresponse.
The main finding of the thesis is that usage of the parametric estimator might substantially improve the accuracy of estimates if the sample size is small, whereas it is not recommended if the sample size is large. Concerning medium sample sizes, the performance of the parametric compared to the nonparametric estimator depends on the deviation from the theoretical lognormal assumption in the actual dataset. We investigate two empirical distributions, indicating a better performance of the parametric estimator up to a sample size of about 150 observations in the first dataset and up to a sample size of about 2,000 observations in the second dataset.