Forensic Likelihood Ratio
Also known as: Bayes Factor in Forensics, Forensic Evidence Weight, LR-Based Forensic Evaluation, Adli Olabilirlik Oranı
The Forensic Likelihood Ratio (LR) is a Bayesian framework for quantifying the weight of forensic evidence relative to two competing propositions — typically the prosecution and defence hypotheses. Formally developed and systematised by Colin Aitken and Franco Taroni in their 2004 Wiley monograph, the LR expresses how much more probable the observed evidence is under one hypothesis than under the other, providing the court with a single, interpretable number that separates the scientist's role from the fact-finder's role.
Key highlights
- Logically coherent: embeds Bayes' theorem, ensuring the scientist's contribution is cleanly separated from the court's prior beliefs.
- Interpretable scale: log10(LR) provides a symmetric, intuitive measure of evidence weight with validated verbal equivalents.
- Applicable across evidence types: DNA, fingerprint, speaker, glass, fibre, document, and digital forensics all employ LR frameworks.
- Avoids the transposition of the conditional: by design it prevents the prosecutor's fallacy of equating P(E|Hp) with P(Hp|E).
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use the Forensic LR when trace evidence (DNA, glass, fibres, fingerprints, handwriting, speaker comparison) must be evaluated under two competing source or activity propositions. The framework requires well-defined hypotheses at a consistent level, a reliable reference population or background model, and careful assignment of prior odds to the court rather than the scientist. It is not appropriate when hypotheses cannot be clearly specified, when reference databases are absent or unrepresentative, or when the court is not prepared to reason probabilistically. Alternatives include verbal categorical opinions or posterior-probability reporting, though these lack the LR framework's logical transparency.
Strengths & limitations
- Logically coherent: embeds Bayes' theorem, ensuring the scientist's contribution is cleanly separated from the court's prior beliefs.
- Interpretable scale: log10(LR) provides a symmetric, intuitive measure of evidence weight with validated verbal equivalents.
- Applicable across evidence types: DNA, fingerprint, speaker, glass, fibre, document, and digital forensics all employ LR frameworks.
- Avoids the transposition of the conditional: by design it prevents the prosecutor's fallacy of equating P(E|Hp) with P(Hp|E).
- Requires a reference population database that may be unavailable, unrepresentative, or too small for rare evidence types.
- Prior odds are theoretically the court's domain, yet in practice scientists may inadvertently influence them through hypothesis framing.
- Score-based LR systems (e.g., speaker recognition) require empirical calibration and validation, adding methodological complexity.
- Courts and jurors frequently misinterpret probabilistic outputs despite verbal scale translations, risking evidence misuse.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
Can the forensic LR be greater than 1 and still favour the defence?
No. By definition, an LR > 1 means the evidence is more probable under Hp than Hd, so it always provides at least some support for the prosecution hypothesis. However, a small LR (e.g., 1.5) represents only marginal support and, when combined with very low prior odds, may still yield posterior odds that favour the defence.
Who should assign the prior odds — the scientist or the court?
The court, not the scientist. The scientist's legitimate contribution is the LR alone, which quantifies only the evidential weight of the trace material. Assigning prior odds requires weighing all other case circumstances — witness accounts, alibi evidence, circumstantial facts — which is the exclusive role of the trier of fact.
What is a 'calibrated' LR and why does it matter?
A calibrated LR is one whose numerical value accurately reflects the true underlying probability ratio, validated against ground-truth datasets using metrics such as the log-likelihood ratio cost (Cllr). Poor calibration means the LR systematically over- or under-states evidence weight, which can mislead courts. Calibration evaluation is mandatory before deploying any score-based LR system operationally.
Sources
- 1.Aitken, C. G. G., & Taroni, F. (2004). Statistics and the Evaluation of Evidence for Forensic Scientists (2nd ed.). Wiley.ISBN 978-0-470-84367-3
You have read it. What now?
Cite this page
ScholarGate. (2026, June 2). Forensic Likelihood Ratio. ScholarGate. https://scholargate.app/forensic-science/forensic-likelihood-ratio