> ******* UCL STATISTICS SEMINAR **************** > > All are welcome > Please see our web page for details of further seminars and how to find > us: > www.ucl.ac.uk/stats/research/journals.html > ** Note change from usual day** Friday 24th November 2000 - 4pm, Room 102, 1-19 Torrington Place, > Department of Statistical Science - University College London. > Speaker: Vanessa Didelez Graphical Models for Event History Data based on Local Independence Abstract: The idee of 'classical' conditional independence graphs is to represent the conditional independencies of a multivariate distribution through a graph. The vertices stand for the variables and missing edges for conditional indpendencies. Such graphical models are by now well established and are applied in various fields of statistics. However, when the variables are dynamic as for instance in survival analysis, the classical graphs do not meet the requirement of the dynamic character of the associations. Local independence graphs have been developed to represent the dependence structure of different events occuring in time (Didelez, 1999). Examples for such data can be found in soci-econometric studies where one is interested in the interplay of events such as unemployment, health status, marital status; or in clinical studies where survival is considered together with time dependent covariates such as specific side effects. In contrast to classical graphical models, these 'new' graphs are based on the concept of local independences which seems to be more suitable for dynamic structures (Aalen, 1987; Schweder, 1970). Here, the vertices stand for the different events and the (directed) edges for local dependence which has some similarity to the notion of Granger causality for time series. The graphical representation has to be interpreted as follows: If there is no directed edge from node A to node B then the intensity for the occurence of event B does not depend on whether event A has occurred previously or not. This can be formalized by properties corresponding to the Markov properties of conditional independence graphs and will be addressed in my talk. Further interesting properties concerning the interpretation of local independence graphs will be illustrated. References: Aalen, O.O., 1987: Dynamic modeling and causality. Scand. Actuar. J., 177-190. Didelez, V. (1999), Local independence graphs for composable Markov processes. Discussion Paper No. 158, SFB 386, University of Munich. Schweder, T. (1970), Composable Markov processes. Journal of Applied Probability, 7, 400-410. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%