01 · The Question
Did the study actually establish everything claimed in its conclusion?
The Conclusion is often only a few sentences long, yet those sentences may become the most quoted part of a paper. They can appear in literature reviews, presentations, policy documents, media coverage, and later studies long after the underlying statistical estimates or qualitative evidence have disappeared from view.
That makes a simple question surprisingly important: where did each major claim in the conclusion come from?
A conclusion should synthesize and interpret evidence already established in the study. It can discuss meaning and implications, but it should not quietly introduce an empirical claim, causal relationship, mechanism, generalization, or level of certainty that the preceding results did not adequately support.
03 · What You Need to Know
How to trace a conclusion back to its evidential basis
Break the conclusion into individual claims
A conclusion that appears to contain one broad statement may actually contain several distinct propositions.
Consider: “Our findings demonstrate that frequent use of the platform improves student achievement by increasing engagement and should therefore be incorporated into university teaching.”
That sentence potentially contains separate claims that:
- frequent platform use improves achievement;
- the relationship is causal;
- increased engagement is the mechanism;
- the finding applies sufficiently broadly to university teaching;
- implementation is warranted.
A study might support some of these propositions without supporting all of them. Evaluating the conclusion as one block can hide those differences.
Trace each claim to a result
For each substantive conclusion, ask: Which result supports this?
You should normally be able to locate the relevant evidence in the Results section, tables, figures, or reported qualitative findings. If the claim depends on an analysis, outcome, comparison, theme, or mechanism that never appears there, investigate further.
Sometimes the supporting analysis is located in supplementary material. Sometimes the conclusion summarizes several findings rather than one. But if no evidential basis can be identified, the claim should not acquire credibility merely because it appears at the end of the paper.
This is the practical extension of distinguishing what the data show from what the authors conclude.
Watch for new causal language
One of the clearest forms of unsupported expansion occurs when associational findings become causal conclusions.
The Results may say that higher social-media use “was associated with” lower well-being. The Conclusion may say that social-media use “reduces” well-being.
Those statements are not interchangeable.
A causal claim requires a design and analysis capable of supporting the relevant causal inference under defensible assumptions. Statistical adjustment alone does not automatically remove confounding, reverse causation, selection bias, measurement error, or other alternative explanations.
When the conclusion uses verbs such as “causes,” “leads to,” “improves,” “reduces,” “prevents,” “drives,” or “results in,” compare them with the language used in the Results and determine whether you are seeing an unsupported shift from association to causation.
Check whether a mechanism was actually tested
Conclusions sometimes explain not only that an effect occurred but why.
Suppose an intervention group performed better than a comparison group, and the conclusion says the intervention worked “by increasing learner motivation.” Was motivation measured? Was mediation or another appropriate mechanistic analysis conducted? Was temporal ordering established? Were plausible alternative mechanisms considered?
If not, the mechanism may be a reasonable hypothesis rather than a finding.
Mechanistic speculation is often useful. It simply needs language that preserves its status as speculation or interpretation rather than presenting it as demonstrated fact.
Check whether the conclusion expands the population
Generalization can happen almost invisibly through nouns.
The sample may consist of 180 first-year engineering students at one university, while the conclusion refers to “university students.” A study of nurses at three hospitals may end with a statement about “healthcare professionals.” A survey conducted in one country may conclude something about “teachers” without retaining the geographic boundary.
Broader inference may sometimes be reasonable, but it requires justification. Compare the population named in the conclusion with the population actually sampled and consider whether the study design, context, sampling, and wider evidence support the expansion.
Check whether the conclusion expands the construct
A similar shift can occur between what was measured and what is claimed.
A study measuring scores on a particular test may conclude that an intervention “improves learning.” A self-report intention measure may become “behavior.” Publication counts may become “research quality.” A satisfaction scale may become “effectiveness.”
These broader constructs may be related to the measured variables, but they are not automatically equivalent.
Ask what was actually operationalized and whether the conclusion preserves that measurement boundary.
Practical importance needs more than statistical significance
A statistically significant result does not automatically justify adjectives such as “important,” “substantial,” “meaningful,” or “transformative.” Those terms concern substantive magnitude or consequence.
If the conclusion says an intervention produced a meaningful improvement, determine what criterion makes the observed effect meaningful. Depending on the field, that might involve a minimally important difference, educational benchmark, absolute risk reduction, cost-benefit consideration, stakeholder judgment, or another substantive reference point.
Without such reasoning, the conclusion may be turning statistical significance into practical importance.
An inconclusive result should not become “no effect” without justification
Suppose a study estimates a treatment effect with a confidence interval that includes both no effect and effects large enough to matter. A conclusion stating that the interventions “do not differ” may be stronger than the evidence permits.
Failure to establish a difference in a conventional superiority test is not equivalent to demonstrating equivalence. Appropriate equivalence or non-inferiority questions require designs and analyses suited to those claims.
When the Results communicate uncertainty but the Conclusion announces absence, check for a shift from “no evidence” to “no effect”.
Secondary findings should remain secondary
A conclusion may emphasize a statistically significant subgroup or secondary outcome when the primary analysis was inconclusive.
The secondary result is not necessarily invalid. But its analytical status matters, especially when multiple analyses increase the opportunities for apparently noteworthy findings.
CONSORT 2025 emphasizes attention to multiplicity and distinguishes prespecified from post hoc analyses. When the conclusion appears to revolve around one favorable subgroup, determine whether the authors have replaced the primary result with a subgroup story.
Recommendations require another layer of reasoning
“The intervention changed outcome X” and “institutions should adopt the intervention” are different propositions.
A recommendation may require information about effect magnitude, harms, costs, feasibility, equity, implementation, alternatives, stakeholder priorities, and the wider evidence base. The original study may address some of these issues, all of them, or almost none.
STROBE guidance encourages cautious interpretation in light of limitations, multiplicity, similar studies, and other relevant evidence. It also emphasizes that an individual observational study should generally be viewed as one contribution to the broader literature rather than automatically as a stand-alone basis for inference and action.
A recommendation can therefore be reasonable without being directly “shown by the data.” Your task is to identify the additional reasoning and evidence on which it depends.
The final sentence deserves disproportionate scrutiny
The final sentence is rhetorically powerful. It is also easy to remember and quote.
Read it independently. Then ask what would remain if you removed adjectives, causal verbs, generalizations, and recommendations unsupported by the design.
If the resulting statement differs dramatically from the original, investigate which inferential steps account for the difference.
This is particularly useful after a Discussion has already made an uncertain result sound stronger, because the Conclusion may compress that interpretive framing into an even more definitive claim.
