Rereading your manuscript one more time does not help much. You know it by heart, and the brain corrects what it reads: you see what you meant to write, not what you wrote. That is why the errors a reviewer finds in ten minutes have gone unseen for weeks by the person who wrote the text.
What works is changing reader. Instead of rereading as the author, you read three times as three different people: the editor who decides in five minutes whether the paper goes on, the methodological reviewer looking for reasons why not, and the reader who wants to repeat what you did. And you make a separate pass for the numbers and another for the references, which hide the most errors.
This guide gives a five-pass method with an order, one question per pass and a rough time. It takes an afternoon or two and is the best-paying preparation before you send a paper to a journal.
If you can, do not review on the same day you finish writing. Two or three days of distance are enough for the text to feel foreign and for you to start seeing what is missing. If you have a tight deadline and cannot wait, at least change the format: print the manuscript, switch to another font and size, or read it aloud. Anything that breaks the automatic flow of reading.
Also prepare what you will need: the journal’s author instructions to hand, the reporting guideline that fits your design (CONSORT for trials, STROBE for observational studies, PRISMA for systematic reviews, APA’s JARS for psychology) and the original output of your analyses, so you can check the figures.
The editor decides within minutes whether the paper goes on, looking at four things: the title, the abstract, the first page and the tables. Do exactly that and nothing else. Read the title and ask whether it says what was done and with whom. Read the whole abstract: are the gap you fill, the design, the sample with its size, the main results with their magnitude and an implication that is not decorative all there? Look at the tables without reading the text: is it clear what each one compares?
The test is simple. If someone who does not know your work cannot say in a minute what you did and what you found, the editor will not be able to either. What you fix here is worth more than anything in the later passes, because it is what always gets read.
Now read the method and results with an adversarial attitude: your job is to find reasons the paper should not be published. It is uncomfortable, and it is what the reviewer will do. Make a list of the following questions and answer them in writing, with the page where the answer is:
Every question without a clear answer in the manuscript is a comment you will receive. It does not all need new analyses: often one sentence is enough. But you have to have seen it before the reviewer does. If you want to go deeper on any, see our guide to the statistical errors a reviewer spots first.
It is the dullest pass and the one that finds the most errors. Open the original output of the analyses beside the manuscript and check, number by number:
Recomputing p values and hunting impossible means by hand is heavy and error-prone. There are guides for each case, such as the one on p values that do not match the test statistic and the one on the GRIM test, and tools that do it over the whole text. Even if you automate it, do one manual pass first so you know what is being checked.
References are where errors pile up unseen, because nobody reads them closely. Check that every in-text citation is in the list and every list entry is cited in the text; that years, volumes and pages are right; and that each work really exists, with the authors you say. A frequent error is composing a reference from memory or from an abstract and getting a co-author or a year wrong.
A reference that cannot be found in any bibliographic database is an alarm for an editor, even when it is an oversight. We have a guide on checking the references in your paper, and there are tools that match each entry against several databases automatically.
So far you have done the work. The last pass is someone else’s, and the one that costs most. The ideal is a colleague in your area who took no part in the study, with a specific instruction: “tell me what you would change to reject it”. An instruction like that produces more than “what do you think?”, which invites politeness.
If there is nobody to read it, or you want a first reading before bothering anyone, an automated review can serve as a first outside opinion. Previum simulates a journal’s: a simulated editor and two simulated reviewers read the manuscript and return the objections on method, analysis and reporting, plus recomputing p values and verifying references. It does not replace a colleague with judgement about your field, but it gives you a different reading, in minutes, and the Basic version is free with an account.
If you want to schedule it, this fits into two afternoons of two hours. First afternoon: passes 1 and 2 (the editor and the methodological reviewer), with the changes that come out. Second afternoon, after a night’s rest: passes 3 and 4 (numbers and references), and 5 with whoever reads it for you. Do not change the order: passes 1 and 2 can alter the text, and there is no point checking numbers you will then rewrite.
Before you send, also go through the journal’s submission checklist and our pre-submission checklist, which covers the administrative side (statements, cover letter, files).
With this method, about two hours per full pass and an afternoon or two in all. It is little compared with the months a review takes, and it is what most reduces avoidable objections.
Not entirely, which is why the method changes reader in each pass and asks for an outside reading at the end. Time distance and a change of format help, but they do not replace another person.
It depends on the design: CONSORT for randomised trials, STROBE for observational studies, PRISMA for systematic reviews and meta-analyses, and APA’s JARS for psychology research. Many journals require them in their instructions.
It is useful as one more reading, from another angle and on things a person misses, such as recomputing every p value or checking every reference. It does not replace your supervisor’s reading, who knows your field and your study. They complement each other.
The title and the abstract, because that is what the editor reads first and sometimes all. If it is not clear there what you did, with whom and what you found, everything else weighs less.
Yes, at least the number and reference passes, because each correction can open a new gap, such as a figure that changes in the text but not in the abstract. You do not need to start over: repeating the check on what you touched is enough.