Graphical models for event history analysis based on local independence (Q2715918)

The author considers graphical models for analysing of dependence structures in event history data. This thesis explores graphical models defined through local independence structures induced by suitable graphs and investigates the properties of the resulting local independence graphs with particular respect to their interpretation and their implications for statistical inference. The material is subdivided into five chapters.NEWLINENEWLINENEWLINEChapter 1 provides necessary background on graph theory, special attention being paid to graph separation. In Chapter 2, two different approaches to modeling of random structures in time are presented, namely conditional independence graphs and causality graphs. The fundamental notion of local independence is introduced in Chapter 3, its general definition being based on the framework of marked point processes. In Chapter 4, the local independence graphs are defined and it is shown which information on local independencies can be read off the graphs. The final chapter gives an overview over the applications of different statistical inference procedures to local independence graphs.