This is the check a reviewer with a calculator makes: add the column in Table 1 and compare it with the N in Participants. When they disagree, the difference is people nobody has said anything about.
Missing values on that particular variable. By far the most common and the most legitimate: twelve people did not answer the education question, so that row sums to 108 rather than 120. That is correct, and it has to be stated in a table note or a sentence.
A category that is not shown. "Other", "Prefer not to say", "Not reported". They get dropped because they have few cases, and people disappear. If the category exists, it belongs in the table even with three people in it.
An undeclared subsample. The table describes only those who completed follow-up, or only the experimental arm. A reasonable decision that stops being one when it goes unstated.
Categories that are not exclusive. Ethnicity, diagnosis or comorbidity with multiple response: one person can appear in two rows. The sum exceeds N and that is correct, but then the percentages do not add to 100 and it has to be flagged.
Observations instead of people. In repeated measures it is easy for the table to count measurements while the text counts participants. The N inflates with nothing to warn you.
A transcription error. A 47 that was 4 and 7, an extra digit. It shows up quickly, because the difference is usually a round or very large number.
Because it needs no understanding of the study. An editor can do it during screening in twenty seconds without reading the method, which is why it is among the first things checked in a serious review.
And because it teaches more than the mismatch: if three tables in one paper imply three different Ns, what you have is missing-data handling that was never described. At that point the reviewer’s question is no longer arithmetic.
How to declare missing data and what to say for each analysis is in how to report missing data, and the table that usually carries it in Table 1 in APA 7.
It is the cleanest option: a "Not reported" row with its n and percentage. The alternative is a footnote saying how many are missing on that variable. What does not work is a table that quietly sums to less.
The one you actually use, and say so. If you compute them on respondents, the note must read "percentages of valid cases". If you compute them on the whole sample, the missing cases appear as a category.
That is normal. Then the table carries the n for each variable rather than one N in the header. It costs a column and heads off the question.
Not in itself, but it still has to be explained. What a reviewer reads in a small unexplained mismatch is not fraud, it is carelessness — and that colors the reading of everything else.