What Should You Do When Different Versions of the Same Paper Contain Different Information?

What If Two Copies of the Same Paper Do Not Actually Say the Same Thing?

You find a preprint containing one sample size, an accepted manuscript containing another, and the published article reporting a slightly different analysis. Or perhaps the numerical results match but a paragraph has changed, a table has disappeared, an outcome has been renamed, or the final conclusion is more cautious than the earlier one.

These are not necessarily errors. Scholarly articles can pass through several identifiable states before and after publication, and changes may occur through peer review, author revision, production, correction, or updating. NISO's journal-version framework was created precisely because identifying these different states matters for scholarly communication.

Once the versions disagree, however, you should stop treating them as interchangeable copies. Your task is to identify what each document is, determine where the difference entered the publication history, and decide which version supports the particular claim you need to make.

Identify the Versions First, Then Investigate the Difference

Version Differences Are Part of the Publication Record

First establish which versions you actually have

Two files with the same title are not necessarily the same publication state. Scholarly publishing distinguishes several stages. Crossref's versioning guidance, for example, identifies drafts, preprints, pending-publication manuscripts, advance online publications, author accepted manuscripts, versions of record, and updated versions. NISO's established terminology similarly distinguishes an author's original, submitted manuscript, accepted manuscript, proof, version of record, corrected version of record, and enhanced version of record.

Importantly, the file format does not identify the version. Both a preprint and a version of record can be PDFs. Conversely, an article's version of record may be available in HTML as well as PDF.

Differences can be cosmetic, editorial, or substantive

Do not treat every discrepancy as equally important. A changed font, new page numbers, altered reference style, or repositioned figure may have no effect on the study's meaning. A changed sample size, outcome definition, coefficient, exclusion criterion, statistical model, figure, or conclusion is a different matter.

Your response should be proportional to the difference. There is little value in conducting a forensic investigation because a table moved from page 14 to page 9. A changed denominator in the primary analysis deserves considerably more attention.

Peer review can legitimately produce substantive changes

A preprint records the work at an earlier stage. Reviewers and editors may identify methodological weaknesses, request additional analyses, challenge interpretations, or ask authors to clarify or modify the manuscript. Authors may also discover errors themselves.

Therefore, a difference between a preprint and an accepted or published article does not automatically indicate misconduct or unreliability. It may simply document the normal development of the work through peer review and revision.

What matters is the nature of the change and, when relevant, whether the available publication history explains it.

Post-acceptance changes also occur

An accepted manuscript is not necessarily frozen into the final published article word for word. Copyediting and proof correction occur after acceptance in many publishing workflows. Publisher definitions make this explicit. IOP, for example, distinguishes its accepted manuscript from the final published version and from the version of record.

If your accessible copy is an AAM and the final version is inaccessible, you can often still use the manuscript, but the level of reliance should reflect what information you need.

The published article itself can later change

Publication does not always end version history. Crossref describes updated states that may involve supplementary data, corrections, or retractions and emphasizes version control for traceability and identifiability.

This means that “I have the published PDF” is not always the end of the inquiry. A locally saved PDF downloaded before a correction can differ from the current publisher record. Check the DOI landing page when a discrepancy matters.

Look for formal correction mechanisms

If the publisher has corrected the article, the change may be documented through an erratum, corrigendum, correction notice, retraction and republication, updated version, or another mechanism depending on journal practice. Follow links from the current article record and DOI rather than relying only on a file saved elsewhere.

Crossref's versioning guidance recommends linking updates such as corrections and retractions to the relevant scholarly object. Such relationships can help establish which document represents the later state of the publication record.

Compare the disputed information directly

When the difference matters, create a simple version comparison. Record the version, date, source, and exact element that differs. Do not rely on memory after reading two similar manuscripts. Academic déjà vu is not a version-control method.

Do not create a hybrid version without realizing it

One particularly easy mistake is to read the Methods from an accepted manuscript, extract the results from the published article, use a supplementary table from another repository, and then discuss everything as though it came from one internally consistent document.

Sometimes using multiple versions is necessary and defensible. If so, record which version supplied which information and verify that the versions can legitimately be combined for your purpose. This matters especially in systematic reviews, meta-analyses, replication studies, and research-integrity assessments.

What if the earlier version contains information removed from the final article?

The earlier text remains evidence that the earlier version contained that information. It does not automatically mean the information should be attributed to the final published article.

This distinction is particularly important when studying changes in research reporting. If a preprint reported an outcome that disappears from the final publication, you can describe the difference between the versions if you have verified it. You should not cite the final article as though it contains the removed outcome.

What if you cannot access the final version?

Use the publisher's metadata page to establish what you can about the final record, then search for a legitimate accessible copy. An institutional repository may provide the accepted manuscript, and other authorized routes may provide the published version.

If you have only the citation or metadata but no full text, the usual strategies for locating a paper when the citation is available but the full text is not still apply. Do not infer final-version content solely from an earlier manuscript.

When the Preprint and Published Article Report Different Results

The important response is not to average 0.42 and 0.31, choose whichever result better fits the rest of your review, or assume that one is a typo. The versions provide evidence of a change. Your job is to understand enough of that change to use the evidence consistently.

Common Mistakes When Paper Versions Disagree

Treat Version Differences as Evidence That Needs Provenance

The more consequential the information, the stronger the need for provenance. A punctuation change rarely warrants investigation. A changed primary outcome may. Your retrieval and verification effort should reflect that difference rather than treating every variation as equally important.

If resolving one discrepancy begins consuming disproportionate time, the practical question becomes how much effort the source justifies. Version control is essential; archaeological excavation is sometimes optional.

How to Reconcile Conflicting Versions of a Paper

Questions About Different Versions of the Same Paper

When Versions Disagree, Provenance Matters More Than Convenience

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.