Do authors who publish their peer review get corrected less?
Written against the design standard.
1. The question
Some journals let authors choose whether their peer review reports are published alongside the article. That choice creates two groups within one journal, under one editorial process, with the same reviewers: articles whose review can be read and articles whose review cannot. Whether the first group is corrected or retracted less often has not been measured.
Two mechanisms predict the same direction and cannot be separated here: authors confident in their work opt in, or published review deters sloppiness. The design says so.
2. Prior work, and the search behind the novelty claim
Log key P15.
corpus regex: (transparent peer review|open peer review|published peer review
|peer review reports? (were|are) published)
PubMed: ("transparent peer review"[tiab] OR "open peer review"[tiab]
OR "published peer review reports"[tiab]) AND (correction*[tiab]
OR retraction*[tiab] OR "data sharing"[tiab] OR citation*[tiab] OR quality[tiab])Seventy-four PubMed records. The relevant ones argue for transparent review (PMID 36261430) and survey journal policies and editors’ opinions (PMID 33211009). The search found no within-journal comparison of opted-in and opted-out articles on post-publication outcomes. The author has written on post-publication peer review (PMID 34686787), a different stage of the process.
3. Hypothesis
Within journals where it is optional, articles that publish their peer review file have a lower hazard of correction or retraction in the 36 months after publication.
Null. The hazard ratio, opted in against not, stratified by journal and year, is 1.
What would falsify it. A hazard ratio at or above 1.
4. Design and unit
Retrospective cohort within journal. One article is one unit; follow-up from publication date, censored at 36 months or at the data snapshot (S10).
5. Sample
Research articles in journals where publication of the review file is at the authors’ option, identified by probe PA from journal policy pages and from the presence of both groups in PMC. Articles published before a journal made review optional, or after it made it mandatory, are excluded because the choice did not exist for them.
6. Exposure
Opt-in, detected in JATS as a supplementary material or related-object element whose label or title identifies a peer review file, read from markup rather than text (S2).
7. Outcome
First correction or retraction, from PubMed CommentsCorrections of type ErratumIn or RetractionIn, cross-checked against Crossref update-to relations. Dated by the notice’s publication date (S7). Corrections and retractions are also analyzed separately, because a correction can be the opposite of a failure: an error found and fixed.
8. Validation gate
Opt-in detection is checked by the author against the journal-hosted article page for 100 articles, 50 per group. Below 0.95 sensitivity or specificity, the detection rule changes before the study proceeds; an exposure this simple should be near perfect.
9. Analysis
In R.
coxph(Surv(months, event) ~ opted_in + strata(journal, pub_year), data = d)Preregistered alongside: separate models for corrections and for retractions; a competing-risks analysis treating retraction as competing with correction.
10. Threats
| threat | handling | removed or disclosed | standard |
|---|---|---|---|
| Selection versus deterrence | Not separable in this design; stated as association | Disclosed | |
| A correction may reflect diligence, not error | Corrections and retractions analyzed separately | Disclosed | |
| Opt-in detection error | Gate on 100 articles, near-perfect threshold | Disclosed | S2 |
| Follow-up length differs by publication date | Fixed 36-month window with censoring | Removed | S10 |
| Journal policy changed during the window | Articles outside the optional period excluded | Removed | S5 |
11. Probes
| probe | computes | goes to | what would change the design | status |
|---|---|---|---|---|
| PA | journals with optional transparent review, and their optional periods | 5 | fewer than two journals with both groups makes this a single-journal case study | not run |
| PB | opted-in and opted-out article counts per journal-year, exposure only | 5 | an opt-in share near 0 or 1 leaves no comparison | not run |
| PC | power by simulation at the realized counts and a stated correction rate from the literature, not from this sample | 9 | a detectable hazard ratio above 2 makes the study uninformative | not run |
| PD | the author’s 100 checks of opt-in detection | 8 | below 0.95 the detection rule changes | needs the author |
12. What this cannot show
- Whether publishing review improves articles.
- Anything about journals where transparent review is mandatory or absent.