Rejected: was it the English or was it the numbers?

You have three things open at once: the editor's letter, a quote from an editing service that has just landed in your inbox, and the calendar, because the journal gives you thirty days. The letter mentions the English in the first paragraph. It also mentions, four comments further down and in different words, that the statistical approach needs better justification. The decision in front of you is not a literary one: it is where the money and the three remaining weeks go, with the added discomfort that if you pick the wrong category you will not find out until the second letter arrives. What follows is how to separate the two using what is already in your inbox.

The sentence that decides nothing

«The English requires improvement by a native speaker» is the cheapest line an editor can write and, above all, the safest. It commits nobody to a methodological judgment, it commits the journal to no claim about your design, and nobody is going to argue back. So it appears equally in letters where the English really is the problem and in letters where the English is a detail: writing it costs the same either way, which is precisely why it carries so little information.

That the sentence is handed out unevenly has been measured. Amano and colleagues (2023) surveyed 908 researchers across eight countries in the environmental sciences: among non-native speakers from countries of moderate English proficiency, 38.1% had had a paper rejected because of the English writing, against 14.4% of native speakers. At revision the gap widens: 42.5% of those non-native speakers report being asked to improve their English often, most of the time or always, against 3.4% of native speakers. It is not a psychology sample and the percentages should not be transplanted to your field, but the pattern does travel: the sentence gets written far more readily when the author's surname and affiliation make it plausible.

Hence the practical consequence. If your letter mentions the English, that mention carries a noise component you cannot estimate, so do not use it as a diagnosis. Use it for what it is: a box that will have to be ticked sooner or later, probably for less money than you think, and that tells you nothing about why you were rejected. The real reasons behind a first-quartile rejection are elsewhere and I have gone through them in why Q1 journals reject you.

The pen test

Print the manuscript and the letter. Go comment by comment and ask yourself one question: can I close this with the paper in front of me and a pen, without opening any file that contains data? If yes, it is language. If answering means opening SPSS, R, the dataset or the collection log, it is analysis. It is a crude test and that is why it works: it does not depend on how the comment is phrased, it depends on what work is needed to close it.

The test disambiguates exactly the cases that looked ambiguous. «The Results section is difficult to follow» does not say which kind of problem it is, but the repair does: if what is needed is reordering paragraphs, signposting each block and cutting the three sentences that repeat the table, it is language and the pen reaches. If what is needed is effect sizes with their intervals that appear nowhere in the manuscript, it is analysis and the pen does not come close.

There is a third category worth separating because it is cheap and gets confused with the other two: the administrative. The ethics committee approval number, the conflict of interest statement, the link to the supplementary material, the data availability statement, the reference formatting. You close them by writing two lines you already know how to write, they cost nothing, and they routinely take up a good third of the letter. Take them out of the count before you decide anything or you will overestimate the work ahead of you.

What a genuine language problem sounds like

A genuine language comment points at a place. «Lines 112 to 118 are difficult to parse», «the tense shifts between past and present within the same paragraph of the Methods», «the abstract contains three sentences of more than forty words». The reviewer tripped at a specific point and can tell you which one. That is the fingerprint and it is fairly reliable: language is flagged by line.

Numbers are not flagged that way. An analysis problem almost never lives on a line; it lives in the distance between two places that should match and do not. The N in the Methods and the degrees of freedom in Table 2. The effect size you write in the text and the F it is supposed to come from. The conclusion in the abstract and the design you describe on page three. When a comment needs to name two locations in the manuscript to make itself understood, no language editor is going to fix it, however good.

There is a genuine middle ground and it is better recognized than denied: structure. «The Discussion repeats the Results instead of interpreting them» or «the research question does not appear until page five» are problems of textual architecture rather than grammar, and the higher substantive editing tiers really do cover them, because that is exactly what they advertise. That is money well spent. What no editing tier covers is the missing interpretation being missing because the analysis does not support it.

What a numbers problem sounds like (and why it sounds like style)

Reviewers are trained to be polite, and politeness in review has a fixed grammar: unclear, not clear, requires justification, the authors should clarify. Those four formulas almost never mean «badly written». They mean «this is not here, and I am telling you without accusing you of anything». It helps to keep the translations to hand.

«It is unclear how the sample size was determined.» There is no power analysis. You are not being asked to explain it better: you are being asked for a calculation you never ran. «The authors should clarify whether corrections for multiple comparisons were applied.» Somebody counted the tests in your Table 2, got to thirty, and none of them is corrected. «Please explain the discrepancy between the N reported in the Methods and the degrees of freedom in Table 3.» You have missing data you did not declare. «The reported effect sizes appear inconsistent with the test statistics.» Somebody redid your arithmetic from the F you published and got something else. «The choice of parametric tests for these outcomes requires justification.» They have seen a three-point ordinal variable going into an ANOVA.

The most misread of the lot is «the conclusions are not fully supported by the data presented». It sounds like you got grandiose and toning it down will fix it. It almost never is. It means the claim exceeds the design: a cross-sectional study written in causal language, a single-university sample with conclusions about the general population, a non-significant difference described as a trend. Toning it down is half the solution; the other half is accepting there is a result you cannot sell, and that is decided by looking at the numbers, not at the sentences.

One shortcut that saves arguments: when the comment names a statistic, an assumption, a sample size or a design type, it is a numbers comment even if the whole sentence is wrapped in the language of writing. «The presentation of the regression results is confusing», on a regression table with no standard errors in it, is not a presentation problem.

Count the comments before you sign the quote

Number every comment in the letter, the editor's included, and give each one a letter: L for language, N for numbers, A for administrative. Add them up. The split gives you the decision far better than the general impression the letter left, which tends to be dominated by the comment that hurt most rather than the one that takes most work to close.

Then read the editor's own paragraph, the one that belongs to no reviewer. That is where you see what the journal has kept out of everything it was told. If the editor picks up a numbers comment, even a single one, even with eight language comments from the reviewers, that is the one to close properly, because it is the one they will reread when the manuscript comes back.

Two cases decide by themselves. If there is even one comment from the «inconsistent» or «discrepancy» family, the editing quote waits: those do not close without redoing the arithmetic, and redoing the arithmetic almost always moves text. And if it was a desk rejection with no reviewers, the mention of the English is rarely the real reason; in those letters the underlying reason is usually fit with the journal's scope, and nobody has looked at your numbers yet.

What you are actually buying when you buy English editing

The prices are public and worth looking at. Elsevier's language editing services start at $95 for Standard, $170 for Express and $204 for Plus; all three work on the English of the manuscript, and it is the Plus tier that adds journal formatting, reference editing, a cover letter and twelve months of support. Enago publishes a table for 3,000 words: Copy Editing from $174 to $225 depending on turnaround, Substantive Editing from $252 to $432, and Top Impact Scientific Editing at $600 for six days or $900 for four. Editage advertises $0.0425 per word for Advanced Editing, $0.065 for Premium and $0.2 for Scientific Editing Pro. All checked in September 2026; prices change, and Editage shows its own with a discount already applied.

That money buys something real. A good substantive editor fixes logic and structure, not just grammar, and in a manuscript written in a hurry that is worth far more than it sounds. What it does not buy is stated in the supplier's own catalog, which is the strongest argument available: Enago sells the statistical work separately, in another section of its site and on another tariff, $638 for Statistical Verification at seven days, $765 for Statistical Recommendation at eight and $1,360 for a full Statistical Analysis at fifteen. If the company selling you the editing separates the two categories on its own site, the buyer should not merge them.

The same goes for the AI language tools, and they say so too. Paperpal advertises plagiarism checking, grammar checking, paraphrasing, and research and citation, and describes its pre-submission checker as meeting the language and formatting standards of more than 1,500 journals. Writefull advertises language feedback in Word and Overleaf, paraphrasing, copyediting and LaTeX generation. They are good at what they do and they earn their place in the workflow. Neither advertises recomputing a p value from its degrees of freedom, because that is not what they are for.

That is the other thing, and it is a different category rather than a better version of the same one. Previum, which is the tool I built, recomputes every p from the degrees of freedom in code, checks the effect sizes against the F or the t you published yourself, and verifies the references one by one against eight bibliographic databases. It is not peer review: it is a self-assessment of your own manuscript before you submit it, and everything it returns is in the example report. The honest comparison is not which one is better, it is that they correct different things and the editor's letter tells you which one you need this time.

It is usually both, and the order matters

The normal case is not one or the other: it is a letter with four language comments, three numbers comments and five administrative ones. And then the order matters more than the choice, because fixing the numbers rewrites text. Redo an analysis and the Results change, half the Discussion changes, and sometimes a sentence in the abstract changes with them. Paying per word to polish paragraphs you will delete in a fortnight is the quietest way there is to waste money: nothing breaks, no warning fires, you simply paid twice for the same passage.

The order that works is the boring one. First redo whatever arithmetic has to be redone and see how much text it moves. Second rewrite the affected sections yourself, in your own English, without worrying about how it reads. Third send the final text out for editing, which is by then shorter and therefore cheaper. Fourth write the response letter, which is the first thing the editor will read and where a good part of the second decision is settled; how to answer a statistical reviewer point by point is broken down in how to respond to reviewers on the statistics.

And if the budget only stretches to one thing, it stretches to the numbers. Imperfect English in a paper whose analyses add up gets fixed in production, or by a colleague, or accepted as it stands, because the content got through. An analysis that does not add up inside an immaculately written paper is fixed at no later stage: it is either pulled before submission or published with the error inside it.

Frequently asked questions

The letter says reject and the only stated reason is the English. Do I edit it and send it back to the same journal?

Check the label first: reject and reject and resubmit are different things and the second is an invitation. If it is a plain reject, send it elsewhere after editing, but do the count on your own paper first. It is rare for a journal to reject on language alone with nothing underneath, and you would rather find that yourself than have the second editor find it with the whole wait to go through again.

My co-author is a native speaker and we still get asked for English editing. What is going on?

It is usually one of two things. Either the comment is about structure rather than grammar, which is a real and different problem that nobody fixes simply by being a native speaker; or it is the template sentence people write when something was hard to read and they cannot say why. Run the pen test over the other comments: if none of them points at a specific line, the language was not the problem.

Is an editing certificate worth anything?

It is worth what it says. It shows the editor that a professional went through the text, and at some journals it stops the manuscript being returned on language grounds before anyone reviews it, which is not nothing. It certifies nothing about the analyses and no editor reads it that way. It is a cheap, useful formality, not an argument.

Will an editing service warn me if the statistics are wrong?

Sometimes, at the higher tiers, if the error is visible from the text (an impossible p value, a conclusion the design cannot deliver) and if the assigned editor works in your area. But it is not what you are buying and not what is guaranteed, which is why Enago, for one, sells statistical review as a separate service at a separate price. If you want that check, buy it as what it is.

How many numbers comments does it take before the letter counts as a numbers letter?

One, if it belongs to the inconsistent or discrepancy family. Those do not get weighed against anything: either the numbers add up or they do not, and until you redo the arithmetic you have no idea how much text moves. The rest do get weighed against the balance of the letter.

The reviewer is asking for an analysis I cannot run. Does that count as a numbers problem?

It does, and it is the most expensive case of all when it is resolved late, because the temptation is to answer with a sentence instead of with the analysis and it shows. What can be negotiated, what cannot, and how each answer is written is in when a reviewer asks for an analysis you do not know how to run.

Keep the test: if the comment closes with paper and a pen it is language; if you have to open the data it is analysis; and neither one gets fixed by buying the other. Do the count before you ask for a quote, start with the numbers, leave the English editing for the final text and write the response letter last. If the numbers column fills up and you are not sure where to start, that is exactly what my statistical consulting is for.