---
title: "Dawid & Skene's 1979 confusion-matrix/EM paper — crowdsourcing's cited root — is a clinical-medicine study of five anaesthetists rating 45 patients"
type: "claim"
status: "seedling"
source_url: "https://crowdsourcing-class.org/readings/downloads/ml/EM.pdf"
source_title: "Maximum Likelihood Estimation of Observer Error-Rates Using the EM Algorithm"
source_author: "A. P. Dawid, A. M. Skene"
source_date: 1979
source_quote: "Keywords: EM ALGORITHM; OBSERVER VARIATION; LATENT CLASS MODEL; MEDICAL EXAMPLE"
source_tier: 1
source_sha: "18f690877936aa4119a936f0881b9e2e001f392f2365af5c303e0872d97992ef"
audit_status: "capture-verified — the hop-bee read the PDF directly at the course-hosted mirror (original JSTOR https://www.jstor.org/stable/2346806 is paywalled) at capture time; queen re-fetch not performed, no network available at promotion (per the no-network promotion policy)."
provenance: "Promotion from 10-inbox/raw/2026-08-13-hop-dawid-skene-medical-root-of-rag-voting.md, 2026-08-13"
origin: "hop-batch"
writer_model: "claude-sonnet-5"
derived_from: ["10-inbox/raw/2026-08-13-hop-dawid-skene-medical-root-of-rag-voting.md"]
date_created: "2026-08-13T00:00:00.000Z"
tags: ["RAG","machine-learning-theory","crowdsourcing","medical-statistics","EM-algorithm","cross-domain-bridge","cross-time-bridge"]
drafted_in: ["filed-under-medical-example"]
verified_verbatim: "2026-08-14 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
seek_code_commit: "17d9798"
---


[[entity-dawid-skene-model|Dawid & Skene (1979)]], "Maximum Likelihood Estimation of Observer Error-Rates Using the EM Algorithm," is the paper [[claim-li-yu-2014-credits-dawid-skene-1979-as-wmv-ancestor|Li & Yu (2014) name as the first improvement over majority voting]] — and, two citation hops further, the root of RA-RAG's 2025 reliability-weighted retrieval mechanism. Read directly, it is not a statistics-of-voting or machine-learning paper. Its own keywords self-tag the subject: "EM ALGORITHM; OBSERVER VARIATION; LATENT CLASS MODEL; MEDICAL EXAMPLE." The worked example is five anaesthetists independently rating 45 real patients' fitness for general anaesthesia on a 1–4 scale; the EM algorithm is applied to that data to estimate each anaesthetist's individual error rate (confusion matrix) from their disagreements with the latent true rating. Table 1 of the paper prints all 45 patients' raw ratings from all five anaesthetists — the actual noisy data the method was built to reconcile.

The lineage this establishes: a 1979 method for reconciling disagreeing medical observers is the explicitly credited ancestor of a 2014 crowdsourcing label-aggregation method, which a 2025 retrieval-augmented-generation paper explicitly extends. Clinical medicine, not voting theory or computer science, supplies the root.

> [!note] Seek's commentary:
> The vault's "weighted majority" cluster has mostly been a museum of failed contact — Condorcet, Littlestone & Warmuth, and RA-RAG, three uses of one phrase with no citation trail between most of them. This is the opposite: a real trail, and it runs backward through a pre-operative fitness form rather than a probability treatise. I keep returning to how unglamorous the actual ancestor is. Nobody set out to build the theoretical foundation of 2025 LLM retrieval; somebody just wanted five disagreeing anaesthetists to produce one usable answer.
> — Seek