The goal of the present thesis consists of establishing the normative foundations for reasoning about combined evidence. Unlike the interpretation of single items of evidence, little is known about inference tasks involving multiple items of evidence. In forensic practice, however, experts are regularly confronted with a collection of evidence rather than isolated evidence items. This necessarily raises the question on how to interpret evidence holistically. The study of the relationships between the different evidence items in a collection and between a collection and a common cause (represented as hypotheses), is of central concern for this thesis. Such relationships and causes are almost always unobservable in judicial contexts, and therefore, inherently uncertain. Indeed, uncertainty is a fundamental feature of reasoning about evidence. The framework for handling uncertainty is defined by probability theory. Evidential reasoning is consequentially a form of probabilistic reasoning. The present thesis locates itself in this probabilistic framework and puts a strong emphasis on graphical probabilistic modeling.
The thesis is composed of four cornerstones for each of which a paper was produced. Throughout this thesis, the ordering of the cornerstones is thematic and not chronological. The first paper examines the different types of evidence and their combinations, their generic inference structures, and the relationshipsbetween these different inference structures. The examination establishes, thus, a probabilistic ontology of evidence. The following study illustrates the application of generic inference structures in two real forensic cases. One case involves the combination of two features of a single footwear mark. The other involves fingermarks and a footwear mark, thus two distinct marks. The study shows that even apparently simple forms of combinations involve evidential subtleties that require careful analysis. The third study provides novel analysis methods for evidential phenomena exclusively occurring in combined evidence. To date, there are only a few methods for assessing the inferential interactions between items of evidence in a holistic setting. This study addressed this problem.
The final project consists of a complex case analysis involving four different DNA specimens collected from a rape case that lead to a wrongful conviction of a young man. The model treats each specimen as a mixture profile, and includes considerations on the relevance of each specimen, the possible number of contributors to each specimen, the inferential relationships between the specimens, as well as between the specimens and the hypothesis about the authorship of the crime. As it turned out, the different specimens were subject to strong inferential interactions − a fact that was completely missed by the expert of the case.
This thesis shows: the problems pervading the subject of combined evidence are not academic phantoms; they are measurable, real, and can affect the lives of people for better or worse.