Top banner


Comments and Corrigenda in Scientific Literature

How self-correcting is the written record of scientific and engineering endeavors?

Joseph Grcar

Because human knowledge is by definition fallible, correction remains a necessity. An understanding of how error correction works in the scientific literature helps rebut fears that the veracity of the literature is declining and may suggest changes to enhance the correcting process.

2013-01MargGrcarFB.jpgClick to Enlarge ImageScholarly publishing has a relatively short history in its present form. It is interesting to note that journal articles have explicitly referenced other articles only since the mid-19th century, thereby enabling citation impact factors—lamentably, some might say. The first large tranche of about 400 journals were founded in the latter part of the 18th century around the time of the American Revolution. Many early journals published articles on any subject, but they restricted submissions to members of a royal academy (which served as a kind of peer review) or to faculty of a university (which traded its journal gratis with other schools, a continuing practice). The modern research university with a comprehensive library of current publications developed in the 19th century. As these universities employed growing numbers of increasingly specialized researchers, they created audiences for the journals that predominate today: addressing a single discipline and drawing submissions from authors not affiliated with the publisher. About 2 million articles were published in the whole 19th century, while roughly that number have been published annually in recent years (see the top graph at right).

Errors Are Inevitable

2013-01MargGrcarFA.jpgClick to Enlarge ImageIt comes as no surprise that the scientific literature contains mistakes, because an integral aspect of scientific research is to cope with errors. The types of possible mistakes vary with the purpose of the work. Thomas Kuhn identified three activities in science per se: establishing facts, reconciling facts with theory and articulating theory. A similar spectrum of work occurs in research into engineering processes or medical treatments. All these activities involve choices that are susceptible to error. For example, measurements are subject to fluctuations, and phenomena generally must be observed indirectly, so the quality of the selected data may be poor, or the chosen transformation of the data to the quantity of interest may be inappropriate.

Click to Enlarge ImageScholarly journals exhibit a variety of responses to errors in research articles (see the table at right). The peer review process undoubtedly detects some errors but journals supply little information about rejection rates and reasons. (Acceptance for publication depends on correctness, relevance and significance. Reviewers offer opinions about the correctness of articles, but they cannot be expected to duplicate the work to be certain.) Authors themselves report the majority of known mistakes and correct them in corrigenda and errata (see the top graph below right). Errors found by readers may appear in formal “comment on” and “reply to” exchanges. Identifying errors in complicated reasoning is itself a type of research that may result in separate articles. For example, the present article disputes the sometimes-pessimistic assessment of scientific error rates by supplying more data and a broader context. Finally, evaluating the merit of data and theories is intrinsic to scientific discourse. In recent decades the annual incidence of all three types of corrections (corrigenda and errata, comments and replies, and other disputatious material) has been nearly constant at about 1.6 percent of full-length journal articles, which at present amounts to almost 25,000 corrective articles per year.

2013-01MargGrcarFC.jpgClick to Enlarge ImageIt is misleading to impugn the scientific literature by focusing on fraudulent research as the Wall Street Journal did in 2011. Forced retractions number only a few hundred annually by the newspapers’s own estimate, which is vanishingly small compared to the tens of thousands of corrections. Some retractions do involve fraud that can be difficult to prove. However, the journal did not explain that many retractions are prompted by simpler chicanery such as plagiarism or unapproved authorship. These transgressions are easier to notice now that documents are available through electronic media, which may account for the recent increase in the still very small numbers of retractions.

Enough Correction?

The main issue is whether the present rate of correction is adequate. This question fundamentally concerns the social traditions of editorial boards and research communities because correction rates vary considerably across subject areas (see the bottom graph on the previous page). In general, the overall rates of corrections are lowest in engineering (0.5 to 1.0 percent of journal articles), somewhat higher in the sciences (1.0–1.5 percent), and highest in health and social sciences and liberal arts (1.5–2.0 percent). Multidisciplinary journals are outliers on the high end with correction rates over 3.0 percent.

The positive impact of editorial policies may be seen in the uniformly high rates of self-correction in subject areas connected to biomedicine (seven areas with 1 percent corrigenda and errata). These fields have the most experience in formulating discipline-wide correction policies. The International Committee of Medical Journal Editors suggests procedures for making each type of correction, and the corrections are consistently indexed by the U.S. National Library of Medicine through the PubMed portal to the MEDLINE database. These policies emphasize that “the editor’s concern should be correcting the literature so the readership can rely on the information published.” Knowledge that journals will routinely publish commentary and corrections may encourage authors to be diligent in correcting their own mistakes lest they be corrected by others.

The view that correction is a normal aspect of scientific publishing may also be a factor in the very high correction rates of multidisciplinary journals. Such journals have traditions of fostering scientific communities through editorials, reviews and sections devoted to commentary from readers. For example, Nature has a correction rate of 5 percent achieved through explicit editorial policies for different types of corrections (addenda, corrigenda, errata, retractions and refutations).

The low rate of correction in engineering literature may be abetted by the manner of publication. Many articles for the engineering sciences appear in conference proceedings (see the bottom graph on the previous page). These articles do undergo peer review, and some of the venues are highly selective. Because the proceedings appear infrequently, they do not include corrigenda or reader comments pertaining to the previous conference. The inability to publish corrections for a large part of the literature may contribute to a low rate of reporting errors for articles that do appear in journals.

In summary, the scientific literature is self-correcting though corrigenda and though reader comments. Corrections of various kinds appear at the rate of one to two per hundred journal articles, compared to which the rate of forced retractions is negligible. The rates of correction vary widely across subject areas and journals. Biomedical fields have the highest rates of self-correction while multidisciplinary journals have by far the highest correction rates and include some of the most widely read publications, which should remove any perceived stigma associated with questioning results and reporting errors. Other editorial boards should be encouraged to emulate these journals by establishing sections for reader commentary and policies for publishing corrections so as to encourage a tradition of vigorous scientific debate and self-correction.


The Scopus database classifies publications by document type, source type and subject area. Document types include article (ar), conference paper (cp), editorial (ed), erratum (er), letter (le), note (no), review (re) and short survey (sh). Source types include book (b), journal (j), book series (k) and conference proceedings (p).

In the graphs that depict the quantity of publications, “journal articles” are document type (ar) and source type (j). “Other journal material” are document type (ed, le, no, re, sh) and source type (j). “Articles in collections” are any document type and source type (b, k, p), as well as document type (cp) and source type (j); the latter criterion finds conference papers published in special or supplementary issues of journals. These three groups comprise 96 percent of all publications in the database for 1990–2010. The balance consists of some technical reports and some articles in trade publications.

In the graphs that depict quantities of corrections, “corrigenda and errata” are document type (er), which as indexed by Scopus also includes the small number of retractions. “Comments and replies” are documents of any type and source whose title contains “comment* on” or “reply to”; the asterisk matches any other full word such as “comments” or “commentary.” “Other refutations” are somewhat difficult to identify; for present purposes they are documents of any type and source whose title, abstract or keywords contain “contradict(s),” “counterexample to,” “disproves(s),” “inconsistent,” “invalidate(s),” or “refute(s).” The overlap among these three groups is 0.2 percent. Source type (j) contains 97 percent of these corrective documents; hence, the percentages are taken with respect to “journal articles” as previously defined.

Scopus also classifies publications into 27 overlapping subject areas. Two of these, mathematics and physics, have been separated in the graphs from computer science and from engineering, respectively, to more faithfully represent the underlying fields.

The Scopus database was reoriented to compete with the more established Web of Science database in the 1990s. (The famous citation impact factors are drawn from the latter database.) The expanded coverage of Scopus may account for the sharp increase in journal articles in 1996. Similarly, a failure to fully classify the larger volume of documents may account for the temporary decrease in corrigenda beginning in the same year.


  • Clark, W. 2006. Academic Charisma and the Origins of the Research University. Chicago, IL: University of Chicago Press.
  • Editor. 2010. A painful remedy. Nature 468:6. doi: 10.1038/468006b
  • Editorial Policy Committee. 2012. CSE’s White Paper on Promoting Integrity in Scientific Journal Publications. Wheat Ridge, CO: Council of Science Editors.
  • Elsevier. 2011. SciVerse Scopus. Web Site.
  • International Committee of Medical Journal Editors. 2010. Uniform Requirements for Manuscripts Submitted to Biomedical Journals. 2010.
  • Naik, G. 2011 Mistakes in scientific studies surge. The Wall Street Journal CCLVIII(34):A1, A12, August 10.
  • Kim, H. J. 2011. The transition from paper to electronic journals. The Serials Librarian 41(1).
  • Kronick, D. A. 1976. A History of Scientific and Technical Periodicals, 2nd Edition. Metuchen, NJ: Scarecrow Press.
  • Kronick, D. A. 1991 Scientific and Technical Periodicals of the Seventeenth and Eighteenth Centuries. Metuchen, NJ: Scarecrow Press.
  • Kuhn, T. S. 1970. The Structure of Scientific Revolutions, 2nd Edition. Chicago, IL: University of Chicago Press.
  • Mackenzie Owen, J. S. 2007. The Scientific Article in the Age of Digitization. Springer.
  • Nature Publishing Group. 2011. Authors and Referees: Policies: Corrections
  • U.S. National Library of Medicine. 2011. Fact Sheet: Errata, Retractions, Partial Retractions .… National Institutes of Health.



Bottom Banner