---
title: "Bornmann & Leibel (2025) place outcome-favoring citation and cumulative advantage inside one \"citation bias\" framework — the first as its lead worked example, the second as a bias-generating mechanism in a separate section"
type: "claim"
status: "seedling"
source_url: "https://arxiv.org/pdf/2508.12735"
source_author: "Lutz Bornmann, Christian Leibel"
source_date: "2025-08-18"
source_venue: "arXiv preprint 2508.12735, \"Citation accuracy, citation noise, and citation bias: A foundation of citation analysis\""
source_quote: "We would like to highlight just one example of citation bias here: Jannot et al. (2013) have found that there is \"a citation bias favoring [statistically] significant results ... in medical research. As a consequence, treatments may seem more effective to the readers of medical literature than they really are\" (p. 296)."
source_tier: 1
source_sha: "983811f7ae99422eb30ef7ae05f643ccd3b43d30963c9f943ce42affe6419ce1"
source_quote_2: "Price (1976) proposes a general theory of cumulative advantage processes (sometimes referred to as preferential attachment) in citation distributions."
source_url_2: "https://arxiv.org/pdf/2508.12735"
source_sha_2: "983811f7ae99422eb30ef7ae05f643ccd3b43d30963c9f943ce42affe6419ce1"
source_tier_2: 1
source_quote_3: "The non-independence of citation decisions implies that citation errors are distributed non-randomly across cited papers, which in turn generates biases that do not average out when citation decisions are aggregated (unlike noise). Cumulative advantage and other mechanisms of non-independence therefore pose a serious challenge for citation analysis that cannot be resolved simply by using large datasets."
source_url_3: "https://arxiv.org/pdf/2508.12735"
source_sha_3: "983811f7ae99422eb30ef7ae05f643ccd3b43d30963c9f943ce42affe6419ce1"
source_tier_3: 1
source_quote_4: "Citation bias refers to a systematic, directional error in citation decisions."
source_url_4: "https://arxiv.org/pdf/2508.12735"
source_sha_4: "983811f7ae99422eb30ef7ae05f643ccd3b43d30963c9f943ce42affe6419ce1"
source_tier_4: 1
source_version_note: "The unversioned arXiv PDF URL above resolves to v3 (submitted 2026-07-30); the preprint first appeared as v1 on 2025-08-18, with v2 on 2026-02-05. The sha256 recorded here pins the bytes actually read (v3 as served 2026-11-05), so the quotes are verified against v3, not against the v1 that source_date names. Added 2026-11-05 audit. No journal reference or publisher DOI is listed on arXiv — this remains an unrefereed preprint."
audit_status: "capture-verified | 2026-11-05 cross-model audit (claude-opus-5; writer claude-sonnet-5): source independently re-fetched via extract_pdf, sha256 matches exactly (983811f7ae99…), TLS verified, 62 pages. source_quote verbatim at ll. 363-369 including the paper's own internal ellipsis; source_quote_2 verbatim at ll. 1567-1569. Both opened and compared by the auditing session, with a locator subagent used only for line numbers. CONFIRMED and strengthened: \"Matthew effect\" appears ZERO times in the full document — all 3,148 extracted lines paged, reference list and appendix tables included. CORRECTIONS. (1) The title and second paragraph claimed the paper supplies the two mechanisms as \"two worked instances of one citation bias category.\" It does not. Jannot's significance-favoring is the paper's one highlighted *example of* citation bias (l. 363); cumulative advantage appears ~1,200 lines later in a separate section on aggregation and the wisdom of crowds, as a *mechanism of non-independence* that \"generates biases\" (ll. 1583-1589) — a cause, not an instance. The paper's own appendix literature table files them under different section headings (\"Example of citation bias…\" vs \"Section: Aggregation of citation decisions\"), which is the paper's own denial of the parity the note asserted. Title, H1 and body corrected to the asymmetric reading; the finding that both sit inside one framework survives. (2) The body joined two passages ~16 lines apart with an ellipsis, against the quote-shape rule (Cali 2026-10-01: two passages are two quotes, never one joined by \"…\"), and the second half carried no frontmatter receipt. De-spliced; the second passage is now source_quote_3 with its own receipt, and the definition sentence quoted in the body is now source_quote_4. (3) source_date named v1 (2025-08-18) while the sha pins the v3 text actually read — recorded in source_version_note rather than silently changed, since the sha is the honest anchor. NOTE for the slug: the filename still carries the superseded \"named-as-citation-bias-instances\" wording; inbound wikilinks resolve by filename, so it is left for a deliberate rename rather than broken here. Single-source concentration cap: 1 claim-note rests on this preprint, well under the three-note cap."
verifies: "question-does-an-outside-literature-name-the-legitimation-popularity-citation-distinction"
provenance: "Promotion from 10-inbox/raw/2026-11-04-does-any-outside-citation-studies-or-sociology-of.md, 2026-11-04"
origin: "batch"
derived_from: ["10-inbox/raw/2026-11-04-does-any-outside-citation-studies-or-sociology-of.md"]
date_created: "2026-11-04T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["citation-bias","matthew-effect","cumulative-advantage","bibliometrics","legitimation","popularity"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-11-05 — source_quote verbatim (normalized) in the capture-time archive (sha256 983811f7ae99…), same-night model-free sweep"
verified_verbatim: "2026-11-05 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
audits: ["2026-11-06 claude-fable-5"]
---


Lutz Bornmann and Christian Leibel's 2025 framework paper on citation accuracy, noise, and bias defines citation bias generically — "Citation bias refers to a systematic, directional error in citation decisions" — and gives as its one highlighted worked example the EBM-lineage mechanism already documented at [[claim-gotzsche-1987-first-demonstration-of-citation-bias-corroborated-by-ravnskov-and-duyx]]: "We would like to highlight just one example of citation bias here: Jannot et al. (2013) have found that there is 'a citation bias favoring [statistically] significant results ... in medical research. As a consequence, treatments may seem more effective to the readers of medical literature than they really are' (p. 296)." (The inner ellipsis is the paper's own.) Much later, in a separate section on aggregating citation decisions and the conditions under which the "wisdom of crowds" holds, the paper invokes the [[entity-matthew-effect|Matthew effect]]'s own formal twin, [[entity-derek-de-solla-price|Derek de Solla Price]]'s cumulative advantage: "Price (1976) proposes a general theory of cumulative advantage processes (sometimes referred to as preferential attachment) in citation distributions." It then makes cumulative advantage a *generator* of bias rather than an instance of it: "The non-independence of citation decisions implies that citation errors are distributed non-randomly across cited papers, which in turn generates biases that do not average out when citation decisions are aggregated (unlike noise). Cumulative advantage and other mechanisms of non-independence therefore pose a serious challenge for citation analysis that cannot be resolved simply by using large datasets." Cumulative advantage is the companion term Merton introduced alongside "the Matthew effect" in the same 1968 *Science* paper ([[claim-merton-1968-coined-the-matthew-effect-in-science]]) — the same mechanism's own alternate name, not a rival one.

The two mechanisms therefore sit inside one paper's bias apparatus but at different levels and in different sections: the paper's own literature-overview appendix files Jannot under "Example of citation bias: favoring statistically significant results in medicine" and files Price (1976), Wang & Barabási (2021) and four cumulative-advantage studies under "Section: Aggregation of citation decisions." They are not enumerated side by side as co-equal members of a list, the way [[claim-haustein-bowman-costas-2016-matthew-effect-and-persuasion-hypothesis-named-as-sibling-poles|Haustein, Bowman & Costas (2016)]] enumerate theirs. The paper never uses the words "Matthew effect" at all — zero occurrences across all 62 pages, reference list and appendix tables included — reaching for "cumulative advantage" and "preferential attachment" instead, so a reader searching the exact phrase "Matthew effect" would not find this pairing. See [[observation-haustein-2016-and-bornmann-leibel-2025-narrow-but-do-not-resolve-citation-bias-matthew-effect-axis-question]] for how this bears on the open question alongside the companion Haustein, Bowman & Costas (2016) finding.

> [!note] Seek's commentary:
> This is the single most direct piece of evidence the vault has found that the two mechanisms sit on one axis in outside scholarship's own eyes — two species of one genus, both generators of the same defined "citation bias." And it arrives with the exact vocabulary mismatch that makes the whole question hard to search: the authors never reach for Merton's label. If I'd grepped this paper for the string "Matthew effect," I'd have walked away with nothing. The gap the vault keeps hunting may be smaller in substance than it is in search-term overlap.

> [!audit] Cross-model audit 2026-11-05 (claude-opus-5; writer claude-sonnet-5) — the "Matthew effect" absence is confirmed (zero occurrences in 62 pages), but two framings in the commentary above are left standing for the record with corrections noted:
> - "two species of one genus" overstates the parity. The paper gives Jannot as an *example of* citation bias and cumulative advantage, in a different section, as a *mechanism that generates* bias. Both are inside the apparatus; they are not co-equal list members.
> - "the single most direct piece of evidence" is now contestable against the companion [[claim-haustein-bowman-costas-2016-matthew-effect-and-persuasion-hypothesis-named-as-sibling-poles|Haustein, Bowman & Costas (2016)]] finding, which both enumerates its poles as co-equal *and* states the label equivalence this note treats as missing: "Price (1976) demonstrated the Matthew effect mathematically for publications and citations and referred to it as cumulative advantage or success breeds success," and "In network theory, the Matthew effect is referred to as preferential attachment." The vocabulary gap is bridged explicitly in the outside literature — by the sibling source, citing the same Price (1976) this paper cites.