Bayesian interpretation of discrete class characteristics.

Area of Science:

• Forensic Science
• Probability Theory
• Statistical Inference

Background:

• Bayesian interpretation of forensic evidence frequently relies on the likelihood ratio (LR).
• A large LR is conventionally interpreted as favoring the prosecution hypothesis (H_P) over the defense hypothesis (H_D).
• The LR's role is to update prior odds, not to solely determine the posterior odds.

Purpose of the Study:

• To critically evaluate the interpretation of the likelihood ratio (LR) in forensic science.
• To demonstrate the limitations of LR when the prior odds ratio is neglected.
• To establish criteria for the effective use of multiple, independent characteristics in forensic evidence evaluation.

Main Methods:

• Analysis of the mathematical relationship between prior odds, likelihood ratio, and posterior odds in Bayesian inference.
• Derivation of a condition under which the LR favors the prosecution hypothesis.
• Examination of how multiple independent characteristics, such as DNA loci, can overcome limitations of single characteristics.

Main Results:

• The posterior odds ratio favors the prosecution hypothesis (H_P) only when the LR is at least as large as the number of equally likely potential sources of evidence.
• Using a single, discrete class characteristic severely limits the probative value of evidence for the prosecution.
• Combining multiple independent characteristics significantly enhances the value of forensic evidence.

Conclusions:

• The posterior odds, not just the LR, are crucial for evaluating forensic evidence strength.
• Frequentist interpretations are inadequate for measuring forensic evidence strength due to their focus on the LR's denominator.
• A criterion for determining the necessary number of independent characteristics is presented for robust evidence evaluation.