Correcting the Published Record: A Guide for Authors

Publication is not the end of responsibility, and this guide shows when a correction, an expression of concern, or a retraction is the honest response.

Explainer · 10 min read · Directive Publications Editorial Team

Common tools journals use

• Correction — fix an error that does not invalidate the main findings
• Expression of concern — notify readers while investigation continues
• Retraction — findings or process are unreliable or unethical

Correction. Some journals still split this into an erratum for a publisher's mistake and a corrigendum for an author's; many now use a single label. Either way it is for an error that is contained. The data stand, the analysis stands, the conclusion stands, and something around them is wrong: a transposed figure panel, an axis labelled in the wrong units, a missing funding statement, a collaborator's name misspelled, a reference pointing at the wrong paper, a conflict of interest that should have been declared. The notice is published as its own item, usually with its own DOI, and it is linked from the article page from then on. That linkage is the point: the correction travels with the article, so a reader arriving in five years sees both.

Expression of concern. This is the one authors have usually never heard of before they need it. It is not a verdict. It is a flag the journal raises while something is being resolved — original data cannot be located, an institutional inquiry is under way, co-authors disagree about what happened, or the editors have credible concerns they cannot yet settle. Its job is to protect readers during an interval that can be long. Two things follow from that. It should eventually be resolved by a correction, a retraction, or a notice recording that the concerns were not substantiated, so ask the editor whether that resolution will be published. And it is not an accusation to fight; arguing that a notice makes you look bad, when the underlying question is genuinely open, only lengthens the interval.

Retraction. This is for cases where the findings or the process behind them cannot be relied on: an analysis error large enough to reverse the conclusion, results that cannot be reproduced from the source records, image duplication, plagiarism, fabricated data, missing ethics approval for work that required it, duplicate publication. Retraction is not erasure. The article normally stays visible and readable, marked as retracted, with a notice explaining what happened and who agreed to it, because people have already cited it and need to know what they cited. A retraction you request yourself reads very differently from one imposed after you resisted.

The rule that decides between them. Ask a reader-facing question rather than a reputation-facing one: would a competent reader who relies on this paper act differently once they know? If no, it is a correction. If yes, and you already know the answer, it is a retraction — and it is better as yours than as someone else's discovery. If yes, but the answer is not yet established, that is the expression-of-concern zone: report it and let the journal and your institution establish the facts. How much the outcome costs you is not part of the test.

A worked example. You find that a figure's axis is labelled in milligrams per litre when the values are micrograms per litre, while the text, the statistics and the conclusion all used the correct numbers. That is a correction, and a quick one. Now change one detail: the wrong units were used in the analysis itself, and the effect you reported disappears once they are fixed. Same typo, different remedy — that is a retraction.

Follow journal policy and COPE-aligned practice: COPE Core Practices Explained

Author playbook

1. Document the issue factually
2. Contact the journal/editor in writing
3. Inform your institution when required
4. Cooperate with investigation timelines
5. Do not silently “replace” PDFs on personal sites without the official notice

Document the issue factually. Before you write to anybody, write it down for yourself. One page: what the article says, what is actually the case, the evidence for that (the file, the notebook page, the script, the raw instrument export, with dates), how the problem came to light, which parts of the paper it touches, and — just as important — which parts it does not. Keep interpretation and blame out of it. "The group labels were reversed in the merge step of the analysis script" is usable by an editor; "there seems to have been a mix-up during a very busy period" is not. Then freeze the evidence: copy the raw files and the exact script version to storage nobody will tidy. Investigations turn on whether the underlying records still exist, and cleaning up a shared drive mid-inquiry looks far worse than the original error.

Contact the journal or editor in writing. Email the editorial office or the handling editor from your institutional address, with the DOI and the word correction or concern in the subject line. Open the first paragraph with the full article title, the DOI, and the problem in a single sentence. Then set out what is affected and what is not, say what you propose — correction, expression of concern, or retraction — and attach a marked-up copy or a short document showing the corrected figure, table or text. Close by naming every co-author, confirming they have seen the letter, and copying them in. If a co-author cannot be reached, say so plainly rather than letting silence imply agreement. Conversations do not count: if you speak to an editor by phone or at a conference, send an email the same day summarising what was said. The dated written record is what later distinguishes "the authors reported this themselves" from a claim nobody can check.

Inform your institution when required. Many institutions require notification through a research integrity officer or an equivalent office, and many journals will ask whether the institution has been told. Find out which applies to you before you write to the journal, not after. If you are unsure, one email to that office asking whether this needs formal disclosure costs you nothing and settles it. Where co-authors sit at other institutions, tell them promptly so they can meet their own obligations, and where the work is funded and the error affects reported outcomes, check whether the funder's terms require notification too.

Cooperate with investigation timelines. Journals and institutions set timeframes, and the single most damaging thing you can do is go quiet inside one. Answer within the window even when the honest answer is "we need three more weeks, and here is what we are doing in them". At the outset, ask the editor what the process is, roughly how long it takes, and exactly what they need from you; keep every message in one thread and every file in one folder. Silence is read as obstruction, and a journal that cannot get answers from the authors will eventually publish a notice without their input. If your co-authors do not all agree with what you are proposing, tell the editor that the authors are not unanimous and let the editor manage it.

Never silently replace files. The published notice is what corrects the record; replacing a file corrects only that copy. Once the notice is out, close the loop on every other copy you control — add an update note and a link on the preprint record, ask your repository to attach the notice to its deposit, fix the entry on your lab site and in your CV — but always by pointing at the notice, never by quietly substituting a clean version.

What not to do

• Quietly edit a posted preprint and pretend the VoR never had the error
• Blame a junior author alone for a lab-wide failure of oversight
• Delay until someone else reports it

Quietly editing a posted preprint. Preprint servers keep version histories and posting dates, so the edit is visible and datable to anyone who looks, including the person who noticed the problem in the first place. A silent swap makes an honest mistake look like concealment, and concealment is what turns a routine correction into a misconduct question. Post an update instead: state what was wrong, link to the version of record and to the published notice, and leave the earlier version where it is.

Blaming a junior author alone. Sometimes a trainee did make the mistake. That is still not the whole story, because the supervision, the checking and the sign-off belonged to the senior authors. Notices are permanent and searchable, and a named student carries one into every job application for the next decade. Describe what happened to the data, not who deserves the blame. If individual responsibility genuinely has to be established, that is the job of a formal institutional process, not of a notice you drafted while upset.

Delaying until someone else reports it. Every month a wrong result stays uncorrected, more people cite it, build on it, or design a study around it, and the cost of the eventual fix grows. Delay also changes the frame completely: an author who reports their own error is dealing with an error, while an author who is found out is dealing with an error plus a question about why they said nothing. If you are holding back because you are not yet certain, that is exactly what the letter is for — report what you know and state what you are still checking.

Situations that catch people out

A reader emails claiming an error and you think they are wrong. Check properly before replying, and go back to the raw data rather than re-reading the paper. Then answer within days. If they are right, thank them and start the playbook. If they are not, explain concretely which step they misread. Ignoring the message usually just means the next one goes to the editor instead.

A co-author refuses to agree to a retraction. Unanimity is preferable, not required. Write to the editor with your own account, state clearly which authors agree and which do not, and let the editor decide how to proceed. Notices can and do record that not all authors agreed.

The corresponding author has left and cannot be reached. Any author can write to the journal. Say who you are, why you are the one writing, and what you know about the corresponding author's last contact details. Do not let an unreachable colleague become the reason nothing happens.

You cannot reconstruct the underlying data. "We can no longer locate the original records" is itself a reportable fact, and you should say it in those words. Rebuilding something plausible from summary tables and presenting it as the original data is a far more serious problem than the missing file ever was.

The problem is a missing disclosure rather than a data error. A conflict of interest, a funding source, or an ethics approval left out of the paper is still a correction, and it is one of the fastest to arrange. Do it the day you notice. These are the easiest notices to get published and the worst ones to be caught having ignored.

Bottom line

Match the remedy to the problem, report it yourself, keep it in writing, and let the published notice carry the fix. Every author who has been through it will tell you the same thing: the correction was smaller than the dread that preceded it, and the version that costs you something is the one somebody else had to start.

Academy resources: Directive Academy · Contact paths via Directive Publications