A limitations section fails a paper two different ways, and they pull in opposite directions. Hedge too much and a reader stops trusting that the study found anything worth acting on. Hedge too little — or skip a limitation a careful reader will notice anyway — and a reviewer flags it as an omission, which reads as either carelessness or an attempt to hide a weakness. This guide covers the common categories of limitation that recur across empirical papers, how to scope the language so a real constraint doesn’t sound like a fatal flaw (or the reverse), and works through one fully annotated, explicitly illustrative example from start to finish.
For the Limitations paragraph’s place within the wider Discussion section — comparison to prior literature, stating implications, structuring the section as a whole — see CASRAI’s Writing the Discussion Section of a Research Paper guide, which this page extends with a dedicated, worked treatment of the limitations paragraph specifically.
Why the Limitations Section Matters
Every empirical study has constraints on what it can and can’t claim — a sample drawn from one setting, a design that can show association but not causation, an instrument that measures a proxy rather than the thing itself. Naming those constraints specifically is a marker of a credible manuscript, not an admission of weak work: reviewers and readers already know no single study is definitive. What damages a paper’s credibility is a limitations section that is vague, boilerplate, or missing a constraint the reviewer would have spotted anyway — not the existence of real limitations.
Widely used reporting checklists build this expectation directly into what counts as a complete paper: both CONSORT (for randomized trials) and STROBE (for observational studies) include a dedicated checklist item requiring authors to discuss the study’s limitations, including sources of potential bias and imprecision. A reviewer working from one of these checklists, or simply from field convention, is actively looking for this section — not hoping the author forgot to include it.
Two Ways Authors Get It Wrong
Over-Hedging
Over-hedging is a limitations section that undercuts the paper’s own contribution — stacking qualifier on qualifier until a reader can no longer tell what the study actually supports. Signs of over-hedging: every sentence in the paragraph starts with a variant of “it is possible that,” limitations are listed with no indication of which ones matter most, or the paragraph reads as a generic disclaimer that could be pasted into any paper in the field regardless of its actual design. Authors sometimes over-hedge defensively, anticipating criticism by pre-conceding everything at once; the effect is the opposite of what they intend — a reviewer reading a paragraph like that has no way to judge how seriously to weigh any individual limitation, and the paper’s genuine contribution gets lost under the qualifiers.
Omission
Omission is the mirror failure: leaving out a limitation that a reviewer familiar with the method or field would flag unprompted — a small sample, a single-site recruitment, a self-report measure with known social-desirability bias, an obvious confound the design didn’t control for. Reviewers who work in a subfield know its common measurement and design constraints; an author who doesn’t name one is not concealing it successfully, only signaling that they either didn’t recognize the constraint or chose not to disclose it. Either reading damages the manuscript’s credibility more than naming the limitation directly would have.
The target sits between these two failure modes: name every limitation a knowledgeable reviewer would independently notice, state plainly what each one does and doesn’t affect, and stop there — don’t generate additional caveats the study doesn’t actually have just to appear thorough.
The Common Categories of Limitation
Most limitations paragraphs draw from a small, recurring set of categories. Not every category applies to every study — naming a category that genuinely doesn’t apply just re-introduces over-hedging.
Sample Size and Generalizability
Who was studied, how many, and drawn from where. A small sample limits statistical power to detect real effects; a sample drawn from a single site, institution, or narrow demographic limits how far the findings can be extended to other populations or settings. The useful version of this limitation states not just that the sample was small or narrow, but what kind of generalization claim it rules out — a single-hospital sample, for example, may still support claims about that patient population while not supporting claims about care models that differ structurally from that hospital’s.
Study Design Constraints
What the study’s design does and doesn’t permit the author to claim. A cross-sectional design can show that two variables are associated at one point in time but cannot establish which came first or that one caused the other; an observational design (as opposed to a randomized experiment) leaves open the possibility that an unmeasured variable explains the association. Naming the design honestly — and naming the specific claim it can’t support — is more useful to the reader than a general statement that the design has “limitations.”
Measurement and Instrument Limitations
Whether the tools used to measure the outcome or exposure captured what the study needed them to capture, and how precisely. This includes instrument reliability and validity, whether a proxy measure stood in for the construct the study actually cared about, and any known measurement error. A validated instrument used outside the population it was validated on, for instance, is a specific, statable limitation — more useful to a reader than a blanket statement that “measurement error may have occurred.”
Potential Confounders
Variables that could plausibly explain some or all of an observed association, but that the study’s design didn’t measure or statistically control for. Naming a specific, plausible confound — and being honest about the direction it could bias the finding — is far more credible than a generic acknowledgment that “other factors may have influenced the results.” If a study did control for a variable statistically, that belongs in Methods, not here; this category is specifically for what wasn’t controlled for.
Self-Report Bias
Where outcomes or exposures were measured by asking participants to report on themselves — surveys, questionnaires, diaries, recall-based interviews — rather than through direct observation or objective measurement. Self-report data is vulnerable to social-desirability bias (participants answering in the way they believe is more acceptable), recall bias (inaccurate memory of past events or behavior), and common-method variance (when both predictor and outcome come from the same self-report source, inflating the apparent association between them). Where a study relies on self-report for a sensitive or normatively-charged construct, naming this limitation specifically, rather than folding it into a generic “measurement limitations” sentence, is usually worth the extra line.
Calibrating the Language: Matching Severity to Actual Impact
The same underlying limitation can be written in a way that either buries it or inflates it. The calibration that avoids both failure modes has three parts: name the limitation specifically, state which claim in the paper it actually constrains (not every claim the paper makes), and stop — don’t follow it with an additional round of hedging language that re-litigates the same point. A limitation stated once, specifically, and connected to its actual consequence for interpretation reads as more credible than the same limitation mentioned twice, once specifically and once as an additional vague caveat.
A Worked Example
The example below is an illustrative composite written for this guide, not a limitations section from any real, identifiable published study. It is built around a plausible, common study design specifically so the categories above can be shown working together in one connected paragraph, the way they’d actually appear in a manuscript, rather than as an isolated list.
Illustrative study setup: a cross-sectional survey of self-reported burnout among clinical staff at a single academic medical center, using a validated burnout inventory, with a response rate of 41%.
“This study has several limitations. First, data were collected at a single academic medical center, and the response rate of 41% raises the possibility of non-response bias if staff experiencing higher burnout were more or less likely to respond than their colleagues; findings may not generalize to community hospital settings or to health systems with different staffing models. Second, the cross-sectional design captures burnout at a single point in time and cannot establish whether the workplace factors examined here precede or follow changes in burnout, nor can it rule out that both are driven by an unmeasured third factor, such as a department-wide staffing shortage during the survey period that this study did not track. Third, burnout was measured entirely by self-report; while the inventory used is validated and widely applied in this literature, self-report measures of a construct like burnout are subject to social-desirability bias and may not align precisely with independent indicators such as absenteeism or turnover. These limitations bear most directly on the generalizability and causal claims of the study; they do not undermine the core finding that self-reported burnout and self-reported workload were correlated within this sample, which is the claim the study is positioned to support.”
What this paragraph is doing, sentence by sentence:
- Sentence 1 names the sample and site limitation together with the specific mechanism of concern (non-response bias) rather than just noting a 41% response rate as an isolated fact.
- Sentence 1 also states the generalizability boundary specifically — community hospitals and different staffing models — rather than a generic “findings may not generalize.”
- Sentence 2 names the design constraint (cross-sectional, no temporal ordering) and a specific plausible confound (a staffing shortage) rather than a generic “other factors may explain this.”
- Sentence 3 names the self-report limitation specifically, credits the instrument’s validity so the sentence doesn’t read as an unwarranted attack on the study’s own methods, and names the specific bias mechanism (social desirability) and a concrete alternative (absenteeism, turnover) rather than a vague reference to “measurement error.”
- Sentence 4 is the calibration step: it states plainly which claims the limitations affect (generalizability, causal claims) and which core claim they do not undermine (the within-sample correlation). This is the sentence that prevents the paragraph from reading as over-hedged — it tells the reader exactly how much weight the rest of the paper’s claims can still bear.
Notice what the paragraph does not do: it does not apologize for the study, it does not introduce hypothetical limitations the design doesn’t actually have (there is no line about, say, publication bias, which doesn’t apply to a single primary study), and it does not repeat the same limitation twice in different words to appear more thorough.
A Quick Checklist Before You Submit
- Does every limitation a reviewer familiar with this method would independently notice appear somewhere in the paragraph?
- Is each limitation stated specifically — naming the mechanism or constraint — rather than as a generic, interchangeable sentence?
- Does the paragraph state which of the paper’s claims each limitation actually affects, rather than leaving the reader to work that out alone?
- Does the paragraph end by stating what the study’s core finding still supports, rather than trailing off after the last caveat?
- Have you removed any limitation that doesn’t genuinely apply to this study’s design, just because it’s common in the field?
Frequently Asked Questions
How long should a limitations section be?
There’s no universal length rule. A single connected paragraph of three to six sentences is typical for most empirical papers; studies with several distinct methodological constraints, or journals that require a dedicated “Limitations” sub-heading, may run longer. Length isn’t the goal — covering every limitation a reviewer would expect, specifically and once each, is.
Should limitations be their own section or part of the Discussion?
Convention and journal instructions for authors vary. Most commonly, limitations are addressed within the Discussion, either as a dedicated paragraph or under an explicit “Limitations” sub-heading; some journals ask for it as a distinct section, and a smaller number place it directly before the Conclusion. Wherever it sits structurally, it should not be folded into Methods, where it can read as an oversight the author didn’t flag rather than a disclosed constraint.
How many limitations should a paper disclose?
As many as genuinely apply and no more. The failure mode in both directions is calibration, not count: omitting a limitation a reviewer would notice anyway damages credibility, and so does padding the list with caveats that don’t meaningfully bear on the study’s actual claims.
Do reviewers penalize papers for having limitations?
Reviewers penalize papers for limitations that are undisclosed, vague, or disproportionate to what the study can support — not for having limitations in the first place. Every published study has them; a reviewer’s job includes checking whether the author recognized and appropriately scoped the ones that matter.
Can limitations be addressed in a cover letter instead of the manuscript?
No. A cover letter is read by the handling editor, not necessarily by every reviewer, and it isn’t part of the published record. Limitations belong in the manuscript itself, where any future reader — not just the reviewers at first submission — can see how the findings should be weighed.
