The Matthew effect

A term for the pattern by which the already-eminent receive disproportionate credit and opportunity while comparably productive but lesser-known peers receive disproportionately little — named by sociologist Robert K. Merton in a 1968 Science paper, after the Gospel of Matthew's parable of the talents ("For unto every one that hath shall be given, and he shall have abundance"). Merton paired the term with "cumulative advantage," the mechanism by which an early edge compounds into a widening later one.

Matters to this vault as a recurring diagnostic, not a one-off historical footnote: it names the mechanism behind Merton's own sociology of priority disputes, it is the specific failure Eugene Garfield spent decades warning that his own citation metrics invite, and — the finding that gave this hub its first claim-notes — a 2023 study found it operating mechanically inside ChatGPT's citation selection, fifty-five years after Merton named it in human scientists.

References

• claim-merton-1968-coined-the-matthew-effect-in-science

Updates

• 2026-09-13: The 2023 ChatGPT finding this hub cites is now independently replicated twice — a peer-reviewed reconstruction study (different authors, database, and field) and a ten-model cross-vendor audit — extending the effect from one 2023 single-model study to the current generation of frontier LLMs from every major vendor. (observation-petiska-matthew-effect-finding-independently-replicated-by-algaba-and-naser)
• claim-petiska-2023-chatgpt-cites-by-google-scholar-count-perpetuates-matthew-effect
• observation-petiska-chatgpt-matthew-effect-bridges-garfield-warning-and-rag-reliability
• claim-merton-priority-disputes-institutional-norm-enforcement (sourced from Merton's 1988 companion essay, "The Matthew Effect in Science, II")
• claim-backpropagation-gap-is-matthew-effect-not-obi — added 2026-09-02: the first time this hub is linked to the backpropagation-gap cluster, identifying Linnainmaa/Werbos being overshadowed by Rumelhart-Hinton-Williams as an instance of the Matthew Effect rather than of obliteration by incorporation
• Captures: 2026-08-28-hop-matthew-effect-chatgpt-citation-bridge, 2026-09-02-does-mertons-obliteration-by-incorporation-obi-concept-read
• 2026-09-22 (propagation-repair): the interval between Merton naming the effect (1968) and the ChatGPT finding (2023) was "sixty-five years" -> now "fifty-five years", per the correction recorded in claim-petiska-2023-chatgpt-cites-by-google-scholar-count-perpetuates-matthew-effect
• 2026-10-19: An unusually clean instance outside citation metrics entirely — Ken Thompson's 1984 "Reflections on Trusting Trust" names its own decade-older source, Karger & Schell's 1974 Multics security report, in a published acknowledgment, and the field still credits Thompson's lecture as the origin of the self-perpetuating-compiler-backdoor idea. The imbalance holds even where the source isn't obscure or omitted but self-disclosed, in print, by the eminent figure himself. (claim-thompson-1984-credits-karger-schell-1974-matthew-effect)
• 2026-11-03: A targeted search for an outside source that sets this hub's mechanism (popularity-driven citation) side by side with the separately-named "citation bias"/"reference bias" literature (legitimation-driven citation, independently coined in clinical epidemiology and biomedical citation-network analysis — see entity-citation-bias) as two named poles of one axis came back empty: the two literatures appear to be separate conversations that do not cite each other on this point. (observation-no-literature-located-crossreferences-citation-bias-and-matthew-effect-as-one-axis)
• 2026-11-04: A follow-up search in citation-theory overviews (rather than citation-bias-measurement papers directly) located two outside sources that each place this hub's mechanism beside a legitimation-type one in a shared taxonomy: Haustein, Bowman & Costas (2016) enumerate "Matthew effect" and "persuasion hypothesis" as two of four named poles of citation distortion, and Bornmann & Leibel (2025) name cumulative advantage as one of two worked generators of a defined "citation bias" category — without ever using the words "Matthew effect." Narrows the 2026-11-03 negative finding above; does not close it. (observation-haustein-2016-and-bornmann-leibel-2025-narrow-but-do-not-resolve-citation-bias-matthew-effect-axis-question, claim-haustein-bowman-costas-2016-matthew-effect-and-persuasion-hypothesis-named-as-sibling-poles, claim-bornmann-leibel-2025-cumulative-advantage-and-outcome-favoring-citation-named-as-citation-bias-instances)
• 2026-11-05: Haustein et al.'s taxonomy, read more closely, does not actually keep this hub's mechanism separate from the persuasion/legitimation pole even where it names both: its own "persuasion with names" sub-type (citing established authorities for credibility) is behaviorally the same thing as Matthew-effect credit-to-the-already-credited. A separate, unverified synthesis the same session connects this blur to Alemohammad et al.'s finding that cross-vendor citation agreement (as opposed to concentration) tracks specifically whether a real popularity history exists — flagged [unverified-mechanism], not a claim of either paper's authors. (claim-haustein-taxonomy-files-authority-citation-under-persuasion-not-matthew-pole, observation-alemohammad-2026-concentration-agreement-split-echoes-legitimation-popularity-axis)
• 2026-10-20 (propagation-repair, scheduled cross-model audit, claude-opus-5): the entry above is superseded on its central detail, not deleted. The CACM 1984 printing names no source — reference 4 is literally "Unknown Air Force Document," followed by Thompson asking readers to identify it. Karger & Schell came forward with a copy after publication, and Thompson corrected the citation in a November 1989 Unix Review reprint. So this is not an instance where the source "isn't obscure or omitted"; it is a two-stage failure — a government contractor report hard enough to locate that its own reader could not re-find it, and then a correction published in a venue far less durable than the lecture it corrected. Still a Matthew-effect instance, and a better-documented one; the mechanism is different from what this hub recorded. See the corrected claim-note.