Bayesian Networks for Probabilistic Inference and Decision Analysis in Forensic Science, 2nd Edition

Chapter 13Qualitative, sensitivity and conflict analyses

Bayesian networks are presented and proposed in this book as a general framework for the representation and sensible evaluation of uncertainty in knowledge. It has been pointed out that the graphical structure is amongst a network's most robust constituents. It conveys, on a rather general level of detail, a transparent and clear idea of the factors considered as well as their assumed relationships. When information is available on the strength of the relationships amongst the specified variables, models may be considered at a numerical level and used for inference, that is, a quantitative revision of personal beliefs based on new information.

However, it is sometimes argued or felt that the quantitative specification is too problematic a task for the use of Bayesian networks in forensic science applications. Even though data may be available from various sources such as literature, surveys, databases, or elicited from experts, forensic scientists may be reluctant to provide particular numerical values for all components of interest in the model. One of the reasons for this is that forensically relevant items and traces usually come into existence under particular configurations of real-world circumstances. This may make it difficult to perceive given cases as instances of a series of other cases that occurred under comparable circumstances. This is also one of the reasons why notions such as ‘frequency’ or ‘relative frequency’ ...