Why his analysis matters
Prof. Fenton’s research area — Bayesian networks for legal evidence — is the formal mathematical framework that most directly applies to cases like Letby. A Bayesian network models the prior probability of the prosecution hypothesis, the likelihood of observed evidence under each hypothesis, and produces a posterior probability of guilt that incorporates both. It is what a rigorous statistical analysis of the Letby case would produce.
Fenton’s sustained public commentary on the Letby case applies this framework and produces a conclusion that is not supportive of the Crown’s inference. The Bayesian prior on a single nurse committing an unprecedented cluster of fatal air embolism acts is extremely low; the likelihood of observed evidence under natural- cause alternatives is high; the posterior probability of guilt, on Bayesian calculation, does not meet the criminal-law threshold.
Professional background
- Professor of Risk Information Management, Queen Mary University of London; founder of Agena Ltd (Bayesian-network software).
- Author of Risk Assessment and Decision Analysis with Bayesian Networks (with Martin Neil, multiple editions), a standard reference.
- Long-standing engagement with statistical evidence in UK legal cases, including commentary on the Sally Clark, Lucia de Berk, and Letby cases.
The Fenton Bayesian critique of the Letby chart
- Prior probability. The prior probability of an individual committing a specific unprecedented cluster of fatal neonatal-air-embolism acts is extraordinarily low. The absence of an international-comparable cluster constrains the prior.
- Likelihood ratio. The observed evidence (skin signs, shift overlap, collapse timing) has high likelihood under the natural-cause hypothesis (sepsis, thrombosis, NEC, the documented outbreak conditions). The likelihood ratio therefore does not strongly favour the prosecution hypothesis.
- Posterior probability. Low prior multiplied by modest likelihood ratio produces a posterior probability that does not meet the criminal-law threshold of beyond reasonable doubt.
This is not a speculative conclusion; it is what a formal Bayesian analysis of the available evidence produces. Fenton’s public work lays out the calculation in operational detail.
Why Bayesian analysis matters
Criminal-law standard of beyond-reasonable-doubt does not map one-to-one onto a specific Bayesian posterior-probability threshold. But it maps close enough that a Bayesian analysis showing low posterior probability is substantial evidence that the criminal-law threshold is not met. The RSS post-Sally-Clark framework implicitly asks for this kind of analysis; Fenton’s work provides it explicitly.