nucleation

THEORY-146: In Holman and Bruner's bandit model of honest Bayesian researchers whose beliefs funding cannot touch, who differ in method and whose influence and training of newcomers follow productivity, industry funding of those whose methods already favour its drug lowers, in the runs their figures report, the share of the community on the better treatment, and has no effect without methodological diversity; at the one setting reported, an independent funder that counts industry-funded trials toward its merit threshold lowers that share further, and one that excludes them raises it

3 min read

Properties7
StatusActive
Version2
Tagsphilosophy-of-science, epistemology, network-science
SourceLIT-744 — Experimentation by Industrial Selection
Bears onQUESTION-047 — Can a community of individually sound inquirers settle on a belief none of them reasoned their way to, as an organized whole can settle on a course none of its members intends, and which community-level mechanisms set it?, QUESTION-315 — Does industrial selection, industry funding of researchers whose methods already favour its product, produce industry-favourable consensus in real research fields through those researchers' productivity and their training of newcomers, rather than through funded researchers changing their methods or results, and does an independent funder that discounts industry-funded work counteract it?
SupportsCLAIM-276 — A shared interpretation's coherence, its fidelity to evidence and the plurality of live alternatives are separate quantities, and a population can raise the first while lowering the other two, so a collective interpretation's stability is no evidence of its truth, CLAIM-404 — An organization can be governed from outside without any of its decisions being made for it, through the evidence its own procedures receive or through the criteria by which they weigh evidence; the two routes are distinct and each is possible without the other, so informational and evaluative independence do not imply each other; and where the organization's role is to evaluate another that affects its evidence or resources, the interpretation–authority loop can close between the two institutions, so that formal independence of the evaluator's procedure does not make its evaluation independent
Cited as “extends” byTHEORY-1174 — Industry can bias a scientific community toward a worse treatment without corrupting any individual, by selectively funding researchers whose methods already favour its product, provided researchers differ in method and both influence and the training of newcomers follow productivity; an independent funder that discounts industry-funded work counteracts this, and one that funds on productivity alone makes it worse

Graph View

Source

Holman & Bruner (2017), LIT-744, read in NOTE-574. The authors’ 2015 paper on intransigently biased agents, LIT-751, is unread.

What was actually shown

Holman and Bruner (LIT-744) give a historical case and a model.

  • The case. In the antiarrhythmic drug disaster, industry funded researchers who already used a fast surrogate endpoint and stopped funding one who found deadly side effects. The funded researchers’ views did not change; their productivity made them influential and their method the default.
  • The model. Twenty agents in Zollman’s bandit model, honest Bayesians whose beliefs are unaffected by funding. They differ in productivity (50–100 trials a round) and in methodological bias (measured success drawn around the true rate with variance σ²). Each round an agent leaves with probability 0.02, and the newcomer copies a method in proportion to its holder’s productivity. The true rates are pA = .5 for no treatment and pB = .45 for the drug. Industry adds F trials a round to agents whose measured success for the drug exceeds .45 by more than T.
  • Results. Methodological diversity alone lowers the share converging on the better action. Industry funding lowers it further, more as F grows. With no diversity, industry has no effect. An independent funder using a productivity threshold that counts industry-funded trials lowers reliability; one that excludes them raises it (read from the figures: about 56% with no independent funding, about 48–49% meritocratic, about 77–81% selective).

What this does not say

  • Not that conflicts of interest are harmless. The authors set them aside “while not disregarding” them.
  • Not robust across parameters. A few settings are reported, without run counts or the network used.
  • Not robust across networks, as shown. That the qualitative results held on “both dense and sparse communication networks” rests on a single footnote, with no data.
  • Not that researchers never adapt. In the model no agent changes its method, so the account says nothing about researchers who adjust to what funders reward.

Connections

  • The Zollman effect (THEORY-143). If the footnote holds, the sparse communication that protects a community there does not protect it here, because industry acts on who is productive and who trains newcomers, not on who talks to whom.
  • Transient diversity (THEORY-148). Diversity of pursuit is what protects a learning group there; diversity of method is what a funder exploits here, and it lowers reliability even without one. The two accounts concern different diversities, which the record notes as a tension to keep in view rather than a contradiction.
  • Longino (THEORY-151). The case had venues for criticism and regulatory oversight and was steered anyway, through unequal influence gained from funding. That bears on her norm of tempered equality of intellectual authority. The reading is the record’s.
  • The owner’s essay. §8.2’s epistemic networks can be steered from outside without any member being captured; the remedy proposed is a community-level rule about how merit is counted.

Note of 2026-10-30: the dropped clause

The old title said an independent funder counteracts industrial selection “only if it discounts industry-funded work”. That was imprecise. Holman and Bruner compare two funder policies at one parameter setting (Figure 2) and assert that a merit-based funder worsens the bias and one that discounts industry-funded work counteracts it (NOTE-574, C4). They assert no necessity, and no other policy is tested. The “only if” is not kept as a proposal (ADR-047, kind 3). The rest of the old title, which the paper asserts, is kept as the Proposed reading THEORY-1174, with the funder clause restated without the “only if”.

Backlinks