The abstract says 240 participants, Table 1 sums to 228 and the regression implies 214. None of the three has to be wrong, but nowhere does the paper explain where the differences come from — and that is what the reviewer asks about.
The abstract. It nearly always carries the recruitment N, the largest one, and it nearly never gets updated when the analysis changes. It is the figure most often left behind by the rest of the paper.
The participants section. This should be the final analytic sample, with the route from recruitment to it: who came in, who dropped out and why.
The descriptive tables. Each variable can have its own missing values, so the N varies by row. That is correct and it has to be noted.
And the degrees of freedom of each analysis. They are the N that was ACTUALLY analyzed, because they come out of the model that was fitted. Those cannot be dressed up, which makes them the best place to start looking.
With a flow sentence, not a new table. "Of the 240 recruited, 12 did not complete baseline and 14 had missing values on the outcome, leaving an analytic sample of 214." That answers 90 % of a reviewer’s questions on this point.
With the N for each analysis when it varies. A column in the results table, or a footnote, is enough. What does not work is leaving the reader to infer it from the degrees of freedom.
And with a flow diagram if the study is a trial or a follow-up. CONSORT requires one in clinical trials and STROBE recommends it in observational work. When there are several phases it is the only clean way to tell it.
Because selective attrition biases results. If the 26 who dropped out were systematically different — more severe, younger, less adherent — the analytic sample no longer represents the one you started with, and the finding reads differently.
That is the real reason it gets asked: it is not an arithmetic check, it is that the pattern of loss is information about the study’s validity. Which is why the follow-up question is always whether those who left resembled those who stayed.
The sister check, within a single table, is group sizes against the total N, and the one that starts from the degrees of freedom is degrees of freedom against N.
No, it is normal when missingness differs by variable. The error is not saying so. A reviewer is not bothered by a varying N; they are bothered by having to work it out themselves.
The analytic sample, the one that supports the results you are summarizing. If recruitment was larger and the difference matters, both fit in one sentence.
Then the analytic N is the complete one, and you have to say it was imputed, by what method, with how many imputations and which variables entered the imputation model. Without that sentence, an N that suddenly returns to the larger figure looks like a mistake.
There is no well-supported threshold; the 5 % and 20 % that circulate are rules of thumb. What decides is not the amount but the mechanism: it matters more whether cases are missing for a reason related to the outcome.