| Mixtures are used for modelling the statistical distribution of many types of variables, e.g. the duration of unemployment (economy) or the expression of genes (biology). The components of the mixture correspond with subgroups in the measured population, which however often are only partly observable, e.g. in the personal and environmental characteristics causing long-term unemployment of somebody. Mathematical statistics develops methods to still estimate these components from the available data. Bayesian methods have become popular during the last decade because of their flexibility and the availability of standard computer algorithms. However, these methods require the choice of so-called a-priori distributions. This research aims at determining which a-priori distributions provide good and worse results. Emphasis is on a-priori distributions which exclude as little as possible mixture distributions, so-called non- or semiparametric distributions. |
