01 · The Question
What Do You Report When the Data Do Not Support What You Predicted?
You developed a hypothesis, designed the study, collected the data, and ran the planned analysis. The result does not support your prediction.
Perhaps the effect is not statistically significant. Perhaps the estimated relationship is much smaller than expected. It may even point in the opposite direction.
Can you focus instead on the analyses that worked? Or does the unsupported hypothesis still belong in the paper?
If the hypothesis and corresponding analysis are central to the study, the answer is generally straightforward: an unfavorable result does not stop being part of the research merely because it complicates the story.
03 · What You Need to Know
An Unsupported Hypothesis Is Still a Research Result
A Hypothesis Is a Proposition to Test, Not a Conclusion to Defend
A hypothesis establishes an expectation that can be evaluated against evidence. It is not a promise that the study must ultimately confirm.
If the evidence fails to support the prediction, that outcome remains informative about the research question. The appropriate scientific response is to report what was observed, quantify the relevant uncertainty, consider plausible explanations, and revise the strength of the claim accordingly.
Suppressing an unfavorable result reverses this relationship. Instead of allowing evidence to evaluate the hypothesis, the hypothesis begins determining which evidence readers are allowed to see.
“Not Supported” Is Not Necessarily the Same as “Proven False”
Suppose you hypothesize that an intervention improves achievement and the estimated difference produces p =.18.
It would usually be too strong to conclude simply that “the intervention has no effect.” A nonsignificant test does not automatically demonstrate equivalence or establish that the true effect is exactly zero.
The result may reflect a genuinely small or absent effect. It may also be compatible with a range of effects because the estimate is imprecise. Sample size, measurement quality, study design, model assumptions, and the width of the confidence interval all affect what can reasonably be inferred.
The hypothesis was not supported
The study did not provide the evidence required to support the specified prediction under the analysis performed.
The hypothesis was proven false
A much stronger claim that often goes beyond what an ordinary nonsignificant result can establish.
This distinction is particularly important when null-hypothesis significance testing is used. Reporting an unsupported prediction honestly does not require overinterpreting the null result in the opposite direction.
Report Estimates and Uncertainty, Not Just “Significant” or “Not Significant”
A result is more informative when readers can see the magnitude and precision of the estimate rather than only its relationship to a threshold.
For example, “there was no significant effect, p >.05” leaves many possibilities unresolved. A small, precisely estimated difference and a large but extremely uncertain difference might both receive the same nonsignificant label while supporting quite different interpretations.
Where appropriate for the design and analysis, report effect estimates, confidence intervals, exact p -values, descriptive statistics, and other information readers need to evaluate the finding. The specific reporting requirements depend on the statistical method and disciplinary conventions.
Prespecified Primary Results Are Especially Important
If a study was designed around a primary outcome or hypothesis, that result should not vanish because it disappoints.
For medical journal reporting, the current ICMJE Recommendations state that results should include data on all primary and secondary outcomes identified in the Methods. ICMJE also recommends that studies with statistically nonsignificant or inconclusive findings be submitted for publication or otherwise made publicly available because they can still contribute to cumulative evidence.
For randomized trials specifically, CONSORT 2025 requires identification of prespecified primary and secondary outcomes. The primary outcome is the one prespecified as most important to relevant stakeholders and should be explicitly identified in the report.
These are specific reporting frameworks rather than universal rules governing every discipline. The broader principle is nevertheless important: results should not acquire or lose visibility simply according to whether they support the researchers' expectations.
Contrary Results Can Be More Informative Than a Simple Null
Sometimes the data do not merely fail to support a hypothesis. They suggest a pattern inconsistent with it.
Suppose you predicted a positive association between two variables, but the estimate is negative. That deserves accurate reporting even if the finding is uncomfortable theoretically.
You should still avoid constructing an elaborate post hoc explanation and presenting it as though it had been predicted. An unexpected direction can motivate new theory and future research, but the chronology matters. Developing an explanation after seeing the result is legitimate exploration; rewriting that explanation as an original prediction raises the problem of HARKing .
One Unsupported Hypothesis Does Not Necessarily Invalidate the Entire Study
Studies often evaluate several questions. A primary hypothesis may be unsupported while another prespecified hypothesis receives stronger evidence.
Reporting the complete pattern does not require declaring the entire study a “failure.” That language can itself encourage selective reporting by treating statistically favorable results as successful research and unfavorable results as wasted research.
The more useful question is what each analysis contributes to the research question.
A study can produce theoretically useful evidence by narrowing plausible effect sizes, challenging an assumption, identifying measurement limitations, or showing that an expected relationship did not emerge under particular conditions. Whether those conclusions are warranted depends on the design and evidence, not on whether p crossed.05.
You Do Not Need to Give Every Result Equal Prominence
Complete reporting is not the same as flat reporting.
A primary outcome may deserve more space than a peripheral secondary outcome. A theoretically central hypothesis may warrant substantial discussion, while a routine robustness analysis can appear in a table or supplementary material. Some exploratory analyses may not belong in the paper at all.
The relevant question is whether editorial prioritization changes the substantive impression readers receive.
If a result central to the study is hidden because it is unfavorable while an incidental favorable result dominates the manuscript, the issue is no longer simply concise writing. It can become selective outcome reporting .
Do Not Replace the Unsupported Hypothesis With One That Fits the Results
Suppose your predicted relationship is unsupported, but another unexpected association appears. You can report the unexpected result and develop a new hypothesis from it.
What you should not do is quietly remove the original hypothesis and write the manuscript as though the newly discovered relationship had been predicted all along.
This converts an exploratory finding into an apparently confirmatory one. The evidence has not become stronger. Only the narrative has become cleaner.
Watch Out
Do not interpret “report unsupported hypotheses” as “declare the null hypothesis true whenever p >.05.” Honest reporting requires both directions of restraint: do not hide unfavorable evidence, and do not claim that an inconclusive result establishes absence of an effect.
04 · A Practical Example
What to Write When Your Main Hypothesis Is Not Supported
Hypothetical Example
A Study Predicting Better Academic Performance
A researcher hypothesizes that students using a new learning platform will obtain higher examination scores than students using the existing platform. Examination performance is the study's prespecified primary outcome. Engagement is a secondary outcome.
Primary result
The estimated difference in examination scores is small, and its confidence interval includes effects in both directions. The planned test does not provide clear evidence supporting the predicted improvement.
Secondary result
Students using the new platform report higher engagement, and that comparison is statistically significant.
Tempting response
The researcher minimizes the examination result and rewrites the paper around engagement, concluding that the platform was effective.
Better response
The researcher reports that the primary hypothesis concerning examination performance was not supported, presents the estimate and uncertainty, and separately reports the engagement finding according to its prespecified secondary status.
The second version is not a weaker research paper merely because the findings are mixed. It is a more accurate representation of the study.
The researcher can discuss why examination performance may not have changed, what the confidence interval permits, why engagement might nevertheless be interesting, and whether further research should examine that secondary finding. What the researcher cannot legitimately conclude from these results alone is that the intervention improved academic performance.
06 · What This Means for You
How to Handle an Unsupported Hypothesis in Your Manuscript
Your job is not to rescue the hypothesis. It is to determine what the evidence permits you to say about it.
A simple decision framework
If the hypothesis was prespecified and central to the study
Report the corresponding result regardless of whether it supports the prediction.
If the result is statistically nonsignificant
Report the estimate and uncertainty where appropriate and avoid treating nonsignificance alone as proof of no effect.
If the observed direction contradicts your prediction
Report the direction accurately and explore explanations without pretending those explanations were predicted beforehand.
If another outcome supports the broader theory
Report it according to its actual prespecified or exploratory status rather than using it to replace the unsuccessful primary result.
If manuscript space is limited
Prioritize central results in the main text and use tables, supplementary materials, repositories, or linked reports where appropriate rather than selecting by favorability.
The Discussion should then match the Results. ICMJE recommends linking conclusions to the study's goals while avoiding claims unsupported by the data and distinguishing statistical significance from practical or clinical significance.
Sometimes the most defensible conclusion really is, “The study did not provide clear evidence supporting our hypothesis.” That sentence may be less dramatic than the one you hoped to write. It is also how research is supposed to constrain what you claim.
07 · A Quick Checklist
Before Reporting a Hypothesis That Was Not Supported
Check that you have:
Reported the result if it addresses a central prespecified hypothesis or outcome.
Reported appropriate estimates and uncertainty rather than relying only on a significant/nonsignificant label.
Avoided interpreting p >.05 automatically as proof that no effect exists.
Resisted changing outcomes, exclusions, covariates, or models merely to obtain support for the hypothesis.
Kept unexpected supportive findings in their correct secondary or exploratory context.
Explained plausible interpretations without rewriting post hoc explanations as prior predictions.
Made the abstract and conclusion reflect the overall evidence rather than only the most favorable finding.
Checked the reporting guideline and journal requirements applicable to your study design.
11 · Cite this Guide
How to Cite This Guide
This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.
Recommended (Field Guide)
APA
MLA
Chicago
Copy Citation
