01 · The Question
Can persuasive interpretation change how strong a result feels?
You reach the Results section and find a small effect, a wide confidence interval, an inconclusive primary outcome, or an association vulnerable to alternative explanations. Then you read the Discussion, and somehow the study begins to sound considerably more impressive.
Nothing in the numerical results has changed. The framing has.
The Discussion is supposed to interpret findings, consider limitations, compare them with existing evidence, explore plausible explanations, and discuss implications. Those are legitimate functions. But interpretation can also make uncertain evidence feel more decisive through selective emphasis, stronger wording, favorable comparisons, speculative mechanisms, or disproportionate attention to secondary findings.
03 · What You Need to Know
How a Discussion can strengthen the impression of weak evidence
First decide what “weak” means in this particular study
A weak result is not simply a result with p >.05. Evidence can be limited in several different ways.
An estimate may be small, imprecise, vulnerable to bias, inconsistent across analyses, dependent on strong assumptions, based on a small sample, confined to an exploratory analysis, or poorly aligned with the study’s primary outcome. An observational association may be reasonably precise but still provide limited support for a causal claim.
Before evaluating the Discussion, therefore, establish what the study actually found before interpretation. Otherwise, you have no independent reference point against which to judge the narrative.
The Discussion is supposed to interpret, not merely repeat
The presence of interpretation is not itself a problem. STROBE describes the Discussion as the section concerned with the validity and meaning of a study and recommends that authors provide a cautious overall interpretation considering the study objectives, limitations, multiplicity of analyses, findings from similar studies, and other relevant evidence.
CONSORT 2025 similarly emphasizes discussion of trial limitations, including potential bias, imprecision, generalizability, and multiplicity of analyses.
A good Discussion therefore adds context. The problem begins when that context systematically makes the evidence look stronger than it did in the Results section.
Watch what happens to uncertainty
Compare the language of the Results with the language of the Discussion.
A Results section might report an estimated difference of 3.0 points with a 95% confidence interval from -0.8 to 6.8. The Discussion might then describe the intervention as “producing meaningful improvements.”
The point estimate has not changed, but the uncertainty around it has largely disappeared from the prose.
Confidence intervals and other indicators of uncertainty should influence interpretation, not merely decorate the Results section. When a range remains compatible with materially different conclusions, the Discussion should not quietly collapse that uncertainty into a single confident narrative.
A non-significant primary result can be reframed around something favorable
One well-studied form of reporting distortion is sometimes called spin. The term has been used for reporting strategies that emphasize apparent benefit or distract from an inconclusive primary outcome.
In a 2010 study of 72 randomized trials with statistically non-significant primary outcomes, Boutron and colleagues identified spin in 43.1% of main-text Discussion sections and 50.0% of main-text Conclusions. The study concerned a particular sample of medical trials published in 2006, so those percentages should not be generalized to research as a whole. They nevertheless demonstrate that interpretive framing can diverge materially from primary results.
One common strategy is to redirect attention toward statistically significant secondary outcomes, within-group changes, subgroup analyses, or other favorable findings.
When this occurs, determine whether the Discussion has effectively changed the study’s question after seeing the results. A particularly interesting secondary finding can deserve discussion, but it should not erase the analytical status of the primary result.
“Promising” can do more work than it appears to
Words such as “promising,” “encouraging,” “notable,” “meaningful,” “important,” and “clinically relevant” may be reasonable descriptions. They are also interpretive judgments.
Suppose a pilot study produces a highly uncertain estimate that favors an intervention. Calling the result “promising” may communicate exactly the intended degree of tentativeness. But if the rest of the Discussion then treats effectiveness as essentially established, the apparently cautious adjective has become a rhetorical bridge to a stronger conclusion.
Do not police individual words mechanically. Read the cumulative message.
Possible mechanisms can gradually become established explanations
Discussions often propose reasons for observed findings. That is scientifically useful. A result can stimulate hypotheses about mechanisms even when the study did not directly test them.
The important distinction is between:
Plausible explanation
A mechanism that could account for the observed result but has not been established by the study.
Supported mechanism
An explanation for which the study design, measurements, and analyses provide relevant evidence.
If engagement was never measured, for example, an improvement in test scores does not demonstrate that the intervention worked because it increased engagement. The Discussion may propose that explanation, but it should remain recognizable as a hypothesis.
Comparison with previous studies can strengthen a fragile narrative
A Discussion appropriately places new findings alongside existing literature. Yet citation can also create rhetorical momentum.
Suppose the present study produces an uncertain estimate, but several earlier studies reported favorable results. The Discussion may devote considerable space to explaining how the new study “supports” that literature. Before accepting that characterization, ask what the present study independently contributes.
Consistency with earlier work can matter. It does not transform an imprecise estimate into a precise one, repair bias in the current design, or turn an inconclusive primary analysis into a conclusive result.
Limitations should affect interpretation, not merely appear in a paragraph
Many papers contain a limitations section because journals and reporting guidelines expect one. Its presence alone does not guarantee that the limitations have meaningfully constrained the authors’ claims.
Imagine a paper acknowledging serious residual confounding and then concluding that the exposure “leads to” the outcome. The limitation has technically been disclosed, but its implications have not been incorporated into the causal language.
STROBE recommends discussing both the direction and magnitude of potential bias where possible and treating limitations as part of interpretation rather than as a ceremonial paragraph before the conclusion.
A useful question is: If I took the stated limitations seriously, would I write the conclusion in the same way?
Statistical significance can be turned into practical importance
A small effect can become rhetorically larger when the Discussion repeatedly emphasizes that it was statistically significant without addressing whether the magnitude matters.
The reverse problem also occurs. An imprecise estimate that does not meet a conventional significance threshold may be described as a “trend” toward benefit, especially when the point estimate favors the preferred interpretation.
Neither statistical significance nor proximity to a threshold supplies substantive importance. Examine whether the Discussion has converted statistical significance into practical importance without an independent substantive justification.
Association can become causation through vocabulary
Results may report that two variables “were associated,” while the Discussion says one variable “influences,” “drives,” “improves,” “reduces,” or “leads to” the other.
That change may be justified in a design supporting causal inference. In other circumstances, it represents a stronger claim than the observed association itself.
Whenever the verbs become stronger as you move through the paper, check whether the evidence has also become stronger. If not, you may be seeing a narrative shift from association to causation.
Read the Discussion against the primary result, not in isolation
A Discussion can be beautifully reasoned sentence by sentence while still producing a disproportionate overall impression.
Return periodically to the primary result. What was the effect estimate? How uncertain was it? Was it the prespecified primary analysis? What limitations matter most? Which alternative explanations remain plausible?
The Discussion should illuminate those facts, not make you forget them.