Two completed reports are in the file. Reviewer 1 recommends acceptance with minor revision in four sentences. Reviewer 2 recommends rejection over eleven hundred words, most of it about the sampling and the missing denominators.
The obvious move is to send for a third opinion and let the majority decide. But a third report does not adjudicate the first two; it adds a third opinion to a file that already holds two.
Editors are told repeatedly that the decision is theirs and not the reviewers'. Less is written about doing it when the reports point in opposite directions, when one is unusable, or when the decisive objection is one the positive reviewer never raised. Every review line below is illustrative; real report text is confidential to the review process, so none is reproduced here.
A split is not a tie to be broken
Peer review here is double-blind, and at least two independent expert reviewers are sought for each research manuscript, so one reader's judgement is never the whole assessment. Reviewers advise; the handling editor decides, and COPE's guidance is consistent about where accountability sits. Counting recommendations turns advice into votes, and a vote weighs four unexplained sentences the same as eleven hundred reasoned ones.
Classify the disagreement before you act
There are three kinds, settled differently.
- Fact. The reports disagree about what the manuscript contains.
- Standard. Both read the same thing and disagree about whether it is adequate.
- Scope. The dispute is whether the journal publishes this kind of work at all.
The reconciliation worksheet
Complete it before drafting the decision: one line per point, one page.
| Substantive point | Raised by | Agreement | Type | Decisive? |
|---|---|---|---|---|
| No ethics approval statement (illustrative) | R2 (R1: present) | Contradicted | Fact | No — it is in Methods 2.1 |
| Denominators absent from Table 3 (illustrative) | R2 | One only | Standard | Yes |
| Reporting guideline not followed (illustrative) | R1, R2 | Both | Standard | No — fixable at revision |
- Part 1 — extract. Put every substantive point from both reports into one list, one line each. Mark each as raised by both, raised by one only, or contradicted. A point raised by one reviewer only is not weaker for that: a reviewer who never mentioned the sampling has not endorsed it.
- Part 2 — classify. Label every disputed point FACT, STANDARD or SCOPE. A fact is settleable by reading the manuscript. A standard is your call, made against the journal's requirements and the relevant reporting guideline, not either reviewer's preference. Scope is an editorial question, never a review question.
- Part 3 — test for decisiveness. Mark each point DECISIVE or NOT DECISIVE by the honest test: if it were resolved tomorrow, would the outcome change? One DECISIVE point is enough on its own; a NOT DECISIVE point is not a reason to reject by itself, though enough of them together can mean the manuscript is not ready.
- Part 4 — write the residual question. In one sentence, state what a third reviewer would have to answer that the file does not answer already. If you cannot write it, a third reviewer will not help and you should decide.
Part 4, model sentence: "The question neither report settles is whether the mixed-effects model in Methods 2.4 accounts for the clustering by site described in Methods 2.2 — Reviewer 1 does not address it, and Reviewer 2 asserts it does not, without saying why."
A fact dispute is checkable, so check it yourself
Go to the manuscript and settle it. Where a reviewer missed something present, or relied on something absent, that point leaves the file: it is not weighed, it is removed. Record where you found it, or your letter repeats a criticism the manuscript already answers.
A standards dispute is yours, because the standard is the journal's
Two competent reviewers can read the same analysis and disagree about sufficiency, each applying a threshold of their own. Yours is the published one: what the author guidelines require, what the peer review policy describes, what the relevant reporting guideline in the EQUATOR library asks for. "Reviewer 2 was stricter" is not a reason; "the guideline requires per-group denominators, which Table 3 does not give" is one.
A scope dispute should have been settled before the manuscript went out
If the objection is that the journal should not publish this kind of study, that is an editorial judgement, and the moment for it was before the reports were commissioned. Reviewers can flag scope; they cannot decide it. When it surfaces late, own it rather than laundering it through a reviewer's methodological complaint.
Sort out the unusable report before you weigh the split
An unusable report is not half of a split. Triage first.
| What you have | Action | Boundary |
|---|---|---|
| Three lines, no substantive comment | Do not count it towards the reports sought; seek a replacement | Brevity is not the test — a short report naming a real fault is usable |
| Abusive or personal | Do not send it as it stands; remove the abusive passages or withhold it | Any substantive criticism inside it must still reach the author |
| Evidently assessing a different manuscript or version | Set it aside; check which version was sent; seek a replacement | Do not salvage the parts that fit this paper |
| Names or guesses the author, or would reveal the reviewer to the author | Redact the identifying passage before the report is sent on; raise the breach with the reviewer; decide whether the report can stand | Redact the identification, not the criticism |
| Appears machine-generated, or cites references that do not resolve | Do not count it; check the references yourself | Treat it as unverifiable, not as a proven accusation |
On the last row, see the ethical use of AI in peer review. Nothing here assumes you can detect machine-written text: the test is whether the claims check out and the citations exist.
A report may be withheld and abusive passages removed, but substantive criticism may never be quietly dropped because the editor disagrees with it. An editor who suppresses an inconvenient objection has broken the process. Record why any report was set aside, and tell the reviewer when one is edited or withheld.
When a third reviewer genuinely helps
A third reviewer helps when the file holds a specific unanswered technical question that a particular expert can answer — the statistical model, the assay, the coding frame. Send that question, not the manuscript alone.
"We have two completed reports and are not seeking a third overall assessment. We would value your view on one point only: whether the mixed-effects model in Methods 2.4 accounts for the clustering by site described in Methods 2.2."
The letter that overrules a positive review
A rejection made over a positive report invites an appeal for the wrong reason: an author who reads the two reports side by side may assume a procedural error.
Thank you for submitting "[Title]". After considering both reviewer reports and reading the manuscript myself, I am not able to accept it for publication.
The decision turns on one point, raised by Reviewer 2: Table 3 reports proportions without per-group denominators, so the subgroup claim in section 3.4 cannot be assessed. The reported effect cannot be verified from the data as given.
I have read Reviewer 1's report carefully. It recommends acceptance with minor revision and raises no concern about the analysis, but it does not address the denominators and so does not resolve the question above. The decision is mine rather than a tally of two recommendations.
[Either:] I would consider a new submission with the full denominator data and reanalysis; it would be reviewed afresh. [Or:] I do not think this can be resolved within the present study, and I would not encourage resubmission.
Appeals go to the Editor-in-Chief, whose decision is final. An author who can see the decisive point either fixes it or goes elsewhere, instead of spending weeks appealing about the wrong thing. Ten minutes on that third paragraph is the cheapest step in the process.
What the author is entitled to see
Authors should receive the substance of the reports the decision rests on. Editing a report passes on the rest of it; withholding one removes it, so its substantive points must reach the author in your letter instead. Removing "the authors are evidently incompetent" changes nothing about the science; removing "the control group is not described" changes everything, because the author can no longer answer it.
Record what you decided, and why, at the time
File the worksheet with the decision and add four lines: the decisive point, the type of dispute, whether a report was withheld or edited, and whether a third reviewer was sought or ruled out. It takes a minute while the reasoning is fresh, and it is gone by the time an appeal arrives.
How review is run here is set out in the Peer Review Policy and the Reviewer Guidelines. See also the handling editor's role, running fair and ethical peer review, and writing a first review.