Skip to content

July 2026: Government declines to widen the Thirlwall terms of reference (16 July) · new 100-page insulin report to the CCRC challenging the trial evidence (9 July) · Thirlwall report still expected no earlier than September · inquests relisted to 2027 · Shoo Lee Panel: no medical evidence of deliberate harm.

Lucy Letby Facts

Statistical explainer

Selection bias in hospital cluster investigations

Selection bias arises when the events under analysis are not chosen independently of the explanation being tested. In hospital cluster investigations it appears whenever a suspect is identified first and then a set of events is curated around their presence.

Statistical analysis2Expert/professional
Last updated

Why it matters in the Letby case

Statistical critique argues the Letby shift chart was constructed under selection bias, undermining its inferential weight even where the underlying presence data is correct.

Prosecution position

The 25 charted events were presented as the set of significant clinical events at the unit over the relevant period.

Expert challenge / post-conviction reading

Statistical critique argues the set was not defined independently of the suspect. Where events are curated around a person's presence, the resulting association is built in rather than discovered, and the chart loses inferential weight even if every individual entry is accurate.

What remains uncertain

Reconstructing an independently defined event set after the fact is difficult, and the critique does not claim to have produced one. It identifies a defect in construction rather than substituting a corrected analysis.

Common questions

What is selection bias?

When the cases you analyse are not chosen independently of the explanation you are testing, so the answer is partly determined by the selection.

Why does it matter if the individual entries are correct?

Because accuracy of each row does not fix a biased choice of rows. The question is what was left out and why.

Could an unbiased set be built now?

It would need every significant clinical event on the unit defined by clinical criteria alone, without reference to who was present. The critique argues that was never done.

Is this the same as the off-shift collapses point?

It is the same problem seen from the other side: collapses on shifts Letby did not work bear directly on whether the event set was independently defined.

Found an error or missing source? Send a correction with the page URL and supporting source.