Almost every paper has a limitations paragraph and almost all of them say the same: the sample was small, the design was cross-sectional, the data were self-reported. These statements are true and useless, because they do not say what changes in your conclusions. The reviewer reads them, sees they are boilerplate, and keeps looking for the ones you did not write.
A good limitations section does two seemingly opposite things at once: it protects you, because if you name the problem the reviewer no longer discovers it as a flaw, and it makes you look sturdier, because it shows you understand where what you can claim ends. The hard part is keeping it from turning into either an apology or a disclaimer list.
This guide gives you a four-step formula for each limitation, concrete examples for those that appear most in psychology and health, and a list of mistakes that make the paragraph work against you.
A limitation is a feature of the design, the sample, the measurement or the analysis that restricts what can be concluded. It is not a confession that the study is badly done: every study has them, and no study can avoid them all. What marks a serious paper is not having no limitations but naming them, saying what consequences they have and not claiming beyond them.
It helps to tell it apart from two things it is confused with. A delimitation is a deliberate decision about scope (studying only adolescents, say), explained in the method, and it is not a defect. And an execution error (a questionnaire administered wrongly, data lost through a fault) is not a limitation but a problem, to be corrected or declared separately.
For each important limitation write four things, in this order. Sometimes they fit in two sentences; sometimes they take a paragraph.
Before: “A limitation of the study is the use of a convenience sample, which limits the generalisability of the results.”
After: “Participants were recruited by convenience from two public universities and score above the population mean on educational attainment. This probably reduces variability in the study variable and so may lead us to underestimate the association. We therefore read the coefficients as a conservative estimate and do not extend them to the general or clinical population. Replication with a stratified sample would be the next step.”
The second version says what was done, what bias it produces, what effect it has on the conclusion and how it would be resolved. Four steps in four sentences. It is longer and much harder to criticise.
These five appear in most psychology and health manuscripts. Each links to the guide that develops it.
Say who gets in and who does not, and in which direction that can move the association you estimate. If participants are younger, better educated or more motivated than the population, variability is likely lower and associations weaker, or answers more socially desirable. The full guide is on the page about the convenience sample as a limitation.
Acknowledge that you cannot establish temporal order or rule out reverse causation, and adjust the verbs of the whole paper to that limit: association, not effect. Name at least one plausible alternative explanation, such as a third variable, and say how a longitudinal design could rule it out. The guide on causal claims from cross-sectional designs develops it.
Responses can be shaped by social desirability, memory or mood, and if both variables in your model come from the same questionnaire the association can be inflated by common-method variance. Say whether you did anything (a single-factor test, a marker variable, measures from different sources) and which way the likely bias runs.
It is not enough to say “the sample was small”. Say which minimum effect your sample could detect with the analysis you used and what a non-significant result means in that context: absence of evidence, not evidence of absence. If power was low, accompany null results with confidence intervals. There is a guide on underpowered studies.
Say how many participants were lost, whether the pattern of loss may be related to the study variable, and what you did (complete-case analysis, imputation). If those who drop out are the most affected, your results probably underestimate the effect, and it is worth saying so.
In the discussion, normally in the second-to-last paragraph, before implications and future directions. Do not put them at the end, because the last paragraph is what the reader takes away, and it should not be a list of problems. Do not scatter them through the text unordered either: group them, with the most important first.
As for space, give more to two or three limitations that really affect your conclusions than to a list of seven that fade away. Many journals also ask for the study’s strengths, and it is worth separating them: each in its own paragraph, so neither dilutes the other.
There are five a reviewer sees at once:
The quickest way is to reread your method and results with the question “what would a reviewer say about this?”, and check that each answer is in your paragraph. It is the methodological-reviewer pass described in the guide on how to review your own paper before submitting.
If you want a second opinion, Previum returns the objections a simulated journal review would raise, and among them are the limitations you have not written that a reviewer would have pointed out. It is the way to learn what is missing from the paragraph while you can still add it.
Paste your limitations, ideally with the discussion paragraph around them, and you will see the first objection a reviewer would raise, with the exact sentence from your text and what it would take to fix it. It looks at what this guide proposes: whether you admit the problem that really threatens your conclusions or only the harmless ones, whether you explain which way each limitation pushes the result, and whether your conclusions then respect what you have just conceded.
It returns one objection, the one that weighs most. It needs at least 350 characters and what you paste is not stored. If deciding something would require seeing the method, it says so separately instead of turning it into an objection.
The ones that genuinely affect your conclusions, usually between two and five. A long list of minor limitations dilutes the important ones. It is better to develop three well than to list eight without consequences.
No, the opposite. A reviewer who finds a limitation you did not mention on their own flags it as a flaw; the same limitation named by you, with its consequences, reads as rigour. What does hurt is a limitation that invalidates your conclusion and that you do not acknowledge.
In the discussion, normally in the second-to-last paragraph, before implications and future directions. Some journals ask for a separate section with that heading; check the author instructions.
A delimitation is a deliberate decision about the scope of the study (studying only a specific population) and is explained in the method. A limitation is a feature of the design, sample or analysis that restricts what can be concluded, and is explained in the discussion.
Only if you say which design would resolve them. “Future studies should use larger samples” adds nothing; “a longitudinal design with three waves would allow the temporal order to be tested” does. Be specific.
Many journals ask for it and it is good practice, but in a separate paragraph and not to offset the limitations. Each has its own space.