When 'neutral' evidence still has probative value (with implications from the Barry George Case)

Area of Science:

• Forensic Science
• Probability Theory
• Legal Evidence Evaluation

Background:

• The likelihood ratio (LR) is widely used in forensic science to assess the probative value of evidence.
• Current interpretation often assumes LRs are universally applicable and intuitively interpretable.
• The distinction between source-level and offense-level hypotheses is known but its impact on LR interpretation is not fully understood.

Purpose of the Study:

• To critically examine the assumptions and limitations of using likelihood ratios (LRs) in evaluating forensic evidence.
• To demonstrate how incorrect hypothesis formulation can lead to erroneous conclusions about evidence probative value.
• To clarify the relationship between source-level and offense-level LRs and their impact on determining guilt.

Main Methods:

• Bayes' theorem was used to illustrate the conditions under which LR interpretation is valid.
• Bayesian networks were employed to model and analyze the impact of different hypothesis types on LR calculations.
• Case studies, including the Barry George case involving firearm discharge residue (FDR) evidence, were analyzed.

Main Results:

• The simple intuitive interpretation of LR (greater than one, less than one, equal to one) is only valid for mutually exclusive and exhaustive hypotheses.
• Using non-mutually exclusive hypotheses leads to erroneous conclusions about the probative value of forensic evidence.
• LRs derived from source-level hypotheses generally do not directly inform the LR of offense-level hypotheses, which are critical for determining guilt.

Conclusions:

• The evaluation of forensic evidence's probative value must strictly adhere to the use of mutually exclusive and exhaustive hypotheses.
• Only the LR associated with offense-level hypotheses accurately reflects the evidence's impact on the ultimate question of guilt.
• Evidence with an LR of one is not necessarily neutral and can still possess significant probative value, contrary to common assumptions.