Dawid-Skene model

A 1979 latent-class statistical model, introduced by A. Philip Dawid and Allan M. Skene to reconcile disagreeing raters — worked example: five anaesthetists rating 45 real patients' fitness for general anaesthesia — by estimating each rater's own confusion matrix (error rate) jointly with the unobserved true labels, via the Expectation-Maximization algorithm. Matters to this vault as the traced root of a real, 46-year citation lineage: Li & Yu (2014) name it explicitly as "the first improvement over majority voting" and build their crowdsourcing weighted-majority-voting method on it, and RA-RAG (2025) explicitly extends Li & Yu for LLM retrieval — clinical medicine, not voting theory or computer science, at the root of a mechanism now used to weight sources in retrieval-augmented generation.

References

• 2026-08-13: Li & Yu (2014) credit the model as crowdsourcing WMV's direct ancestor; the model's own paper is a clinical-medicine study, not a machine-learning or voting-theory source. (claim-li-yu-2014-credits-dawid-skene-1979-as-wmv-ancestor, claim-dawid-skene-1979-worked-example-is-anaesthetist-fitness-ratings)
• 2026-08-13: The model's own 1979 paper reportedly proposes reliability-weighted observer voting — RA-RAG's specific 2025 mechanism — though the exact wording is not yet quote-verified. (claim-dawid-skene-1979-proposed-reliability-weighted-observer-voting-unverified)