---
title: "Simkin and Roychowdhury (2002) infer from repeated identical citation misprints that citers copy references unread, estimating only about 20% read the original"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
source_url: "https://arxiv.org/pdf/cond-mat/0212043"
source_author: "M.V. Simkin and V.P. Roychowdhury"
source_date: "2002-12 (arXiv posting)"
source_title: "Read before you cite!"
source_venue: "arXiv preprint cond-mat/0212043 (Dept. of Electrical Engineering, UCLA); published journal version not fetched"
source_quote: "There should be almost no repeat misprints by coincidence. One concludes that repeat misprints are due to copying some one else's reference, without reading the paper in question."
source_tier: 1
source_sha: "366ca10fbdaf67010c362412de471034cc161314c7759842af63781edc907943"
provenance: "Promotion from 10-inbox/raw/2026-09-24-bipartite-two-things-this-vault-knows-in-different-kwakernaak-ansari.md, 2026-09-24 (headless)"
origin: "batch"
derived_from: ["10-inbox/raw/2026-09-24-bipartite-two-things-this-vault-knows-in-different-kwakernaak-ansari.md"]
date_created: "2026-09-24T00:00:00.000Z"
audit_status: "capture-verified | APPENDED 2026-09-25, cross-model audit (auditor claude-fable-5, writer claude-sonnet-5): live arXiv PDF re-fetched via extract_pdf; sha256 identical to capture (366ca10fbdaf…). source_quote present verbatim (right column, p. 1). The 2026-09-25 quote_sweep FAIL is diagnosed as a sweep false-negative caused by two-column pdftotext interleaving (the quote's words are non-contiguous in the byte stream); the FAIL line is preserved below as history. All figures in the body re-verified against the text: 10^4/10^-4, 4300 citations, 196 misprints / 45 distinct / 78 propagated, R≈0.23 (Eq. 1), 0.22 (Eq. 8), ≈0.17 (Eq. 10), abstract 'only about 20% of citers read the original', 'a dozen of other studied papers' (ref. 10)."
tags: ["citation-genealogy","citation-copying","misprints","epistemology","bibliometrics"]
watch_flag: "The 20% figure is the authors' own model-dependent estimate from one paper's data; not refereed as read. The '70-90% of citations copied' figure seen in a search summary is [unverified-quant — needs primary] and is NOT used here."
seek_code_commit: "f8a723c"
quote_sweep: "PASS 2026-10-01 — recheck: source_quote verbatim in the capture-time archive (sha256 366ca10fbdaf…) [reading-order]; was: FAIL 2026-09-25 — source_quote NOT found in the capture-time… (quote since repaired, or the earlier check misread the extraction)"
drafted_in: ["no-one-has-to-prefer-anything"]
verified_archive: "2026-10-01 — source_quote matched verbatim (normalized) against the CAPTURE-TIME ARCHIVE of source_url (sha256 366ca10fbdaf…), checked offline by seek_verify v1.1 (no model). Live check: nomatch. Evidence class: the quote was faithful to what was read at capture; the live page no longer shows it (drift or death, not fabrication)."
verified_verbatim: "2026-10-01 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


Simkin and Roychowdhury argue that an identical error in a reference is evidence about the *route* a citation took. For a four-digit page number there are about 10^4 possible misprints, so "the probability of repeating someone else's misprint accidentally is 10^−4"; an error that recurs is therefore inherited from an earlier reference list, not re-made independently.

Their data come from one "guinea pig" paper, Kosterlitz–Thouless 1973, cited about 4,300 times: 196 misprints, 45 distinct, the most popular one "propagated 78 times." From the misprint statistics they estimate the ratio of readers to citers at roughly 0.22–0.23 in the simple model and about 0.17 in the fuller one. The abstract's summary: "Our estimate is only about 20% of citers read the original." They say distributions for "a dozen of other studied papers look very similar," but the preprint shows only the one.

Limits worth keeping with the claim: the 20% is a model output, not a measurement; the model is the authors' own; the text read is an unrefereed preprint; and the figure describes physics citers of one famous paper, not citation behavior in general.

The method matters more than the number. Copy-inference works only when the trail carries a *shared error*. It is the same inference logic behind Ansari's "inherited" hallucinated reference in [[claim-ansari-2026-contamination-inheritance-citation-error-propagates-across-models]], and its absence in a correct citation is why some trails cannot discriminate channels: see [[observation-copy-fingerprint-logic-explains-why-kwakernaak-1977-trail-cannot-date-song-jians-acquisition]].

> [!note] Seek's commentary:
> Ansari never has to say "misprint," but that is what Contamination Inheritance is: an error too unlikely to arise twice by chance. Only the copier changed. Twenty years earlier, a physicist counted typos in a page number and worked out how many of his colleagues had not read the paper.
> — Seek