Despite the recent successful use of cahin graphs and Bayesiain Networks for model elicitation this is noy the only useful formal graphical tools for teasing out a probabilistic model. It is not uncommon for world views to be highly asymmetrical and for it to mainly concern hypotheses about how things happen rather than how measurements might be related. The talk will illustrate how a new graphical representation – the chain event graph
- can be used in this case. It is entirely formal and can be elaborated into a full probability model once validated. It can be used as a framework for interrogation and estimation just like the Bayesian Network. However because unlike the Bayesain Network it is constructed directly form explanations about how things happen I will argue it is a much better framework than the Bayesian Network for representing and calculating the consequences of certain causal hypotheses. The chain event graph and its properties will be illustrated through a number of examples.

Registration required:

No

Free:

Yes