732A45 Statistical Evidence Evaluation
This course is about forensic statistics, in particular statistical evaluation of forensic evidence. The word "forensic" stems from the latin word "forum" that stands for square, i.e. an open place in a town. In the ancient Roman empire justice was administered at the main square of a town (cf. "Forum Romanum" in Rome, Italy), and part of this process was to discuss what evidence there was that either supported the prosecutor's case of the defendant's case.
Forensic science stands for a conglomerate of different scientific disciplines gathered together to secure, analyse and iterpret (technical) evidence to be used in the legal process of particular cases. It does not always have to be related to a crime. For instance, finding the identity of a dead person that is a victim of a disaster is a legal process, since it has to be legally stated who that person is (due to various consequences of their death). Another non-criminal forensic case may be to investigate if an individual should be entitled to keep his driving licence in spite of having contracted a serious disease. Most forensic casework is however related to a committed crime and comprises a vast variety of analysis methods ranging from professional visual inspection to chemical analyses with a nuclear magnetic resonance camera.
Forensic evidence evaluation is the part of forensic science that deals with interpretation of the findings. Assume for instance that a shoemark has been secured at a crime scene and that it has been compared with a shoe owned by a person suspected to be the perpetrator. Such a comparison may reveal a lot of similarities (matching sole patterns, matching in size, matching in some details due to wearing of the shoe, etc.). It may be assumed that if enough similarities are found the case should be stated as clear, but that is very seldom the situation. On the contrary, found similarities may not at all be extremely rare and it is very important to sort out the rarities. Moreover, should we yet come to the conclusion that the similarities are almost uniqe (they cannot possibly be absolutely uniqe since it is practically impossible to overview all shoes in the entire world) we still haven't proved that the shoe actually did leave the mark. What if the shoe was found in Kuala Lumpur and the mark was found in New Orleans, and there was hardly time to transport the shoe to Kuala Lumpur from the time-point the mark was made until the time-point when the shoe was discovered?
In the course we shall take up various types of forensic evidence and the kinds of measurements/observations that can be made on these. We shall introduce and explore a statistical model for evaluatiing the strength of the findings with respect to propositions (hypothesis) that can be proposed as explanations to the evidence material obtained. Essential for this evaluation is the concept of Bayesian hypothesis testing which will play a key role throughout the course. The methods explored are not limited to evaluation of forensic evidence. On the contrary, Bayesian hypothesis testing is a very general concept for handling uncertainty in observational data with respect to different explanations to why they occured. In the course we will show the generality of the methods and also take up their conjuction with decision making. Much of this work can be done in terms of graphical modelling, which is a convenient tool both for developers and practitionners in pursuing statistical evaluation and interpretation.
The course content comprises:
- Probabilistic reasoning and likelihood theory
- Bayesian hypothesis testing
- Bayesian belief networks (BBN)
- Evaluation of forensic evidence in a hierarchy of propositions (source level, activity level, crime level)
- Reporting evidentiary strength
- Elements of statistical decision theory and influence diagrams
- Different types of forensic evidence
- Sampling rules for decision making
- Application of BBNs and influence diagram in other disciplines
- Qualitative probability models and sensitivity analysis
The teaching comprises lectures, seminars, and computer exercises. The lectures are devoted to presentations of theories, concepts, and methods. The seminars comprise student presentations and discussions of assignments. Computer exercises in which the students have access to supervision provide practical experience of data analysis.
The course will be examined through
- Assignments encompassing both theoretical and computer-based exercises
- One final oral examination
There is a variety of books that can be used to cover major parts of the contents of the course, but there is no particular text that covers it all. For this purpose a main book has been chosen, which will be supplemented by a number of scientific papers. The main textbook will be
- Taroni F., Aitken C., Garbolino P., Biedermann A. (2006). Bayesian Networks and Probabilistic Inference in Forensic Science. Chichester: John Wiley & Sons Ltd. ISBN 9780470091739
- Taroni F., Bozza S., Garbolino P., Biedermann A., Aitken C. (2010). Data Analysis in Forensic Science A Bayesian Decision Perspective. Chichester: Wiley-Blackwell. ISBN 9780470998359
Course responsible and tutor
Anders Nordgaard. E-mail: Anders.Nordgaard@liu.se. Phone: +46 10 562 8013.
To be announced.
Page responsible: Anders Nordgaard
Last updated: 2012-06-20