My Results Are Not Significant. Now What?

Return to the estimate

A result above a chosen p-value threshold does not establish that an effect is absent. Read the estimated effect and its interval in the units that matter for the research question.

In a hypothetical example, a difference of 2 units with a 95% interval from −1 to 5 leaves several interpretations compatible with the data and model. If a 3-unit difference would matter clinically, the interval has not ruled it out. That is different from a narrow interval around zero that excludes differences of practical interest.

The meaningful threshold itself needs justification. Choosing it after seeing the result can make the discussion misleading.

Check the analysis without searching for significance

Verify coding, denominators, model assumptions, and alignment with the prespecified plan. Correct genuine errors and document the correction. Do not cycle through outcomes, cutoffs, covariate sets, or exclusions until a desired result appears.

Additional exploratory analyses can be useful when clearly labeled and interpreted cautiously. They should not be rewritten as if they were the original hypothesis.

Write what the study supports

A useful discussion distinguishes the direction and magnitude of the estimate, its precision, the limitations of measurement and design, and what remains unresolved. Avoid “a trend toward significance” as a substitute for reporting the estimate and uncertainty.

Statistical uncertainty is only one part of interpretation. Selection bias, confounding, missing outcomes, and limited applicability may matter even when an interval is narrow.

Try the Results Interpretation Playground. Keep the effect fixed and vary its standard error to see how precision changes the conclusion. The example is hypothetical and uses a normal approximation.

Download an interpretation worksheet, or request an Understand Your Results Consultation.

Further reading: Cochrane on confidence intervals and interpretation.