Abstract
How do people make legal judgments based on complex bodies of interrelated evidence? This paper outlines a novel framework for evidential reasoning using causal idioms. These idioms are based on the qualitative graphical component of Bayesian networks, and are tailored to the legal context. They can be combined and reused to model complex bodies of legal evidence. This approach is applied to witness and alibi testimony, and is illustrated with a real legal case. We show how the framework captures critical aspects of witness reliability, and the potential interrelations between witness reliabilities and other hypotheses and evidence. We report a brief empirical study on the interpretation of alibi evidence, and show that people's intuitive inferences fit well with the qualitative aspects of the idiom-based framework.
Acknowledgements
David Lagnado is supported by ESRC grant (RES-062-33-0004). We thank Jamie Tollentino for help with constructing the materials and running the experiment.
Notes
These conditional probabilities do not need to equal 0.5; there could be contexts where an unreliable source is more likely to give a positive rather than a negative report (or vice-versa). The key point is that they are equal: .
A similar approach is adopted by Friedman Citation(1987), also using inference networks.
The current experiment has a relatively small sample size. However, very similar findings have been replicated with larger sample sizes and across several variations of alibi evidence (see Lagnado Citation2011, Citation2012).
We lack the space to discuss other formal approaches to legal argument (Walton Citation2008). For comparison between the BN approach and other theories of argumentation, see Fenton et al. Citation(2012). One of the main differences is that argumentation approaches have typically avoided the use of probability theory and Bayesian reasoning to model uncertainty.