nucleation

LIT-724: The Communication Structure of Epistemic Communities

2 min read

Properties6
StatusActive — read 2026-10-05 (NOTE-583): read in full from the author's posted copy. Worth reading as the origin of the "Zollman effect". In a Bala–Goyal bandit model, where agents learn only about the method they pursue, simulated communities with fewer communication links converged on the better method more often and more slowly than fully connected ones. The cause was density, not the centrality of a hub. A simulation study of small networks (3–10 agents) at one payoff gap. LIT-733 shows the effect is confined to hard inquiry.
Tagsepistemology, network-science, philosophy-of-science
Extended byLIT-735 — The Epistemic Benefit of Transient Diversity
Corrected byLIT-733 — In Epistemic Networks, Is Less Really More?
Cited as “paper” byNOTE-583 — The Communication Structure of Epistemic Communities
Cited as “source” byTHEORY-143 — When a group learns only about the options its members pursue and good evidence is scarce, sparser communication makes it more likely to settle on the better option, at a large cost in speed; when evidence is plentiful, sparsity only slows learning

Graph View

Kevin J. S. Zollman (2007), Philosophy of Science 74(5):574–587 — DOI-10.1086/525605

Key takeaways

  • The learning problem. Agents choose between a well-understood action, whose payoff is the same whatever the truth, and a new one that may be better or worse. They learn only from the outcomes of what they and their neighbours actually do, and update by Bayes. If everyone takes the old action, no informative evidence arrives again (§2).
  • Bala & Goyal’s result. On an infinite line convergence to the truth is certain. Adding a “royal family” connected to everyone makes failure possible, because its misleading results reach everybody at once (§2).
  • Finite simulations. For 3–12 agents, cycles succeeded more often than wheels, and wheels more often than complete graphs, over 10,000 runs per size. Complete graphs converged much faster (§3.1, Figures 2–3).
  • Density, not centrality. Across every network of up to six agents, lower density and lower clustering predicted success. Degree variance did not, so the hub’s centrality is not what does the harm (§3.2).
  • A trade-off. The most reliable graphs were minimally connected; the fastest were complete or nearly so (Figures 5–6). Which matters more “will depend on the circumstance” (§3.2).
  • The robustness claim. Footnote 7 says the ordering of the three networks holds “for any setting of the parameters”. LIT-733 shows that it does not.

Standing in the record

Filed on 2026-10-05 at the owner’s request: the founding result of the “research on epistemic networks” an essay’s §8.2 cites for the claim that more communication does not always improve collective judgment. The essay’s wording, “complicates any simple claim”, is as strong as the evidence allows. Rosenstock et al. (LIT-733) found that the effect holds only where inquiry is hard, and the account THEORY-143 is stated with that limit.

Backlinks