Off-shift collapses excluded — the missing denominator
The prosecution’s claim: The shift chart shown to the jury plotted 25 events with Letby present at all 25, framed as proof of guilt.
Prosecution claim
The shift chart shown to the jury plotted 25 events with Letby present at all 25, framed as proof of guilt.
Counter-evidence
Operational records for the Countess of Chester Hospital neonatal unit document deteriorations, collapses and unexpected events on shifts where Letby was not on duty across the 2015-2016 cluster period. These off-shift events were excluded from the chart shown to the jury, which was constructed by selecting events at which Letby was present and then plotting her presence as the inferred pattern. The Royal Statistical Society's 2022 report and 2024 statement name this Texas-sharpshooter selection effect specifically: when the same selection method is applied to any other nurse on the unit who works the equivalent volume of unsociable shifts, similar-looking charts can be produced. The statistically meaningful question — whether collapses are more likely on Letby's shifts than on equivalent shifts staffed by other nurses, with the full denominator — was never put to the jury.
Key point: Painting the target around the bullet hole. The chart was constructed by selecting events at which Letby was present, then using her presence as evidence of guilt. The full denominator — collapses on shifts where she was not on duty — was excluded.
What the jury heard
The 25-event chart with Letby's row fully shaded. The Crown's opening invited the inference that coincidence was impossible. The off-shift events that would have changed the inference were not plotted.
What the Panel says
The Panel did not opine on statistics directly but referenced the Royal Statistical Society's published warnings that selection effects vitiate any inference from the chart.
What independent experts add
- Prof. Richard Gill's published critique covers the selection-bias mechanism in detail.
- Prof. John O'Quigley's peer-reviewed analysis using a proportional-hazards / beta-binomial approach produces an 'anomalous presence' probability of roughly 10% when hours-worked is properly controlled.
- triedbystats.com models the chart for any nurse on the unit if the selection were not pre-restricted.