Li & Yu (2014) credit Dawid & Skene's 1979 confusion-matrix/EM model as 'the first improvement over majority voting'

RAGmachine-learning-theoryvoting-theorycrowdsourcinghistory-of-statisticsmultiple-discoveryEM-algorithm

RA-RAG's 2025 weighted majority voting (WMV) mechanism extends Li & Yu (2014), "Error Rate Bounds and Iterative Weighted Majority Voting for Crowdsourcing" (arXiv:1411.4086). That paper's own introduction names its ancestor explicitly: "The first improvement over majority voting dates back at least to (Dawid and Skene, 1979). They assumed that each worker is associated with an unknown confusion matrix... a local optimum can be obtained by using the Expectation-Maximization (EM) algorithm." Li & Yu's finite-sample error-rate bounds and iterative WMV procedure are built directly on top of the Dawid-Skene confusion-matrix/EM framework, not invented independently.

This closes one citation hop that the vault's existing note on RA-RAG's citation practice left open: that note established RA-RAG cites Li & Yu but stopped there, without asking what Li & Yu's own crowdsourcing WMV method rests on. The answer is a named, specific 1979 paper, not an unattributed reinvention — a real citation chain rather than a second instance of the vault's more common "same idea, no shared ancestor" pattern.

Hongwei Li and Bin Yu are the authors; A. Philip Dawid and Allan M. Skene are the credited 1979 originators.