How to Write a Discussion Section When Your Results Are Not Significant
You’ve designed your study, and you’ve carried it out only to find that the results did not achieve statistical significance.
So, what do you do now?
Scrap the whole project?
Denounce the scientific pursuit of uncovering the hidden truths of the universe?
Before you do any of that, consider this: Despite the nonsignificant results, your study is very much worth publishing. As an article published in Nurse Author & Editor states:
“Although researchers are often hesitant to report non-significant findings, these results can be just as important to circulate as significant results.”
But the question, then, is “How can I discuss my findings if they aren’t statistically significant?”
The answer to that question can be found in this article, which lists the six most important dos and don’ts to keep in mind when writing about nonsignificant results in your research paper’s discussion section:
Do treat statistically nonsignificant results like they’re important.
Do compare your findings to the literature.
Do offer explanations for unexpected results.
Don’t say that the results would be significant with slight changes to the methodology.
Don’t discuss nearly significant results as though they’re significant.
Don’t mistake nonsignificant results as confirmation that there is no effect.
Let's look at each of these in more detail.
Do treat nonsignificant findings like they’re important.
Just because a finding is not statistically significant, that doesn’t mean it’s not significant to the body of research on your topic.
So, don’t hesitate to draw the reader’s attention to nonsignificant results, which are very commonly seen in published papers, as shown in the following examples:
We did not observe any short-term negative behavioral responses from dolphins to the construction of the bridge, indicating a high tolerance for such disturbances.
We observed no significant difference regarding physiological responses or lane change performance.
We observed no significant between-group difference in the rank-based assessment of left ventricular reverse remodeling at 2 years.
Do compare your findings to the literature, just as you would for a significant finding.
Some authors feel that statistically nonsignificant findings don’t add anything worthwhile to the body of knowledge on the research topic.
But this isn’t true.
In fact, nonsignificant findings are very often discussed alongside significant findings from published studies, as shown in the following example:
Rough-toothed dolphins have not been extensively examined in the cetacean cognition literature. Previous research on cetacean species other than bottlenose dolphins generally confirms the right-eye visual advantage (e.g., Karenina et al., 2016). However, several studies have found a lack of species-level laterality for other species (Platto et al., 2017; Yeater et al., 2014). The rough-toothed dolphins in the present study did not display significant laterality at the group level. When examining eye use at the individual level, there was a large variation in individual preference, consistent with previous studies (Yeater et al., 2017).
As the example shows, even if your findings are not statistically significant, they can still contribute something to the literature.
Do offer explanations for unexpected results.
If you expected to find an effect, correlation, etc., but found none, it is important to offer possible explanations for why nothing was found.
Returning to a previous example, the authors explain why their experiment may have produced a nonsignificant finding:
We did not observe any short-term negative behavioral responses to the construction of the bridge, indicating a high tolerance for such disturbances. This does not mean the environment was optimal or that there were no consequences for the survival and reproduction of individuals, but rather that the dolphins can tolerate sub-optimal conditions to obtain certain resources in areas with such disturbances.
There is, however, an important exception to this point, and it’s the next item on this list. . .
Don’t claim that nonsignificant results would have been significant if the methodology were improved.
As noted in Nurse Author & Editor:
“A . . . frequent mistake when your findings are not what you expected is to say the results would have been significant if the sample was larger or if the instruments had better reliability or validity. You have no way of knowing if that statement is true.”
In short, you should not use a small sample size or other methodological issues to explain an unexpected result.
This isn’t to say that you should ignore the limitations of your study’s methodology. You absolutely can state that overcoming these limitations would provide more accurate or more generalizable results, for example.
But do not propose that the results would be different if the limitations were resolved.
The following example shows how you can mention a limitation with the sample size without suggesting that the results would have been significant with a larger sample size:
One limitation of this study was the small number of subjects, especially rough-toothed dolphins. Additionally, the rough-toothed dolphins in the present study were all stranded, rehabilitated, and deemed non-releasable, meaning that they are potentially, but not definitively, behaviorally and/or physiologically nonrepresentative of their species in some ways. Limited conclusions can be made from this small sample, and it is necessary that further research on this species be conducted to learn more about their lateralization and other aspects of their cognition.
Don’t discuss “almost significant” results as though they are significant.
If your statistical tests indicate that a result was close to reaching statistical significance, that result should not be discussed as though it were significant.
In other words, a result with a p-value of .052 should not be described as statistically significant any more than a result with a p-value of .520 should (assuming a significance level of α = .05).
Some authors are tempted to describe nearly significant findings as though there was an effect, as follows:
The treatment showed a trend toward improving survival (HR = 0.80, 95% CI = 0.63–1.02, p = .06).
But this isn't the best way to report this finding.
Although readers can see from the statistical data that the “trend” is not significant, it’s better to be clear about the nature of the trend to avoid potentially misleading anyone:
Patients receiving the treatment had a lower hazard of death than those in the control group, although the difference was not statistically significant (HR = 0.80, 95% CI = 0.63–1.02, p = .06).
Don't interpret a nonsignificant result as proof that there is no effect.
When discussing your results, it’s important not to interpret nonsignificant findings as evidence of the absence of an effect or correlation, for example.
An article in Royal Society Open Science encourages researchers to “explicitly [consider] the possibility that their nonsignificant finding may reflect a true effect that could not be found.”
In other words, don’t assume that there is no effect just because your study didn’t reveal one.
For example, don't say:
The results show that there is no correlation between laterality and interest in stimuli.
Instead, word your interpretation of a nonsignificant finding so that it only indicates that your study failed to reveal a significant effect:
No significant correlations between laterality and interest in stimuli were found in the present study.
Summary
The most important thing to take away from this post is that nonsignificant results are worth reporting.
However, when interpreting such findings, it’s crucial to follow the best practices and avoid common mistakes:
Treat nonsignificant results like they’re important.
Compare your nonsignificant findings to the literature.
Offer explanations for your nonsignificant results.
Don’t say that the results would be significant with slightly different methods.
Don’t discuss nearly significant results as though they’re significant.
Don’t mistake nonsignificant results for confirmation of no effect.
The examples used in this article were taken (and, in some cases, modified for illustrative purposes) from the following sources:
Goldstein, D., et al. (2016) Two-year outcomes of surgical treatment of severe ischemic mitral regurgitation. The New England Journal of Medicine, 374(4), 344–353. doi: 10.1056/NEJMoa1512913
Le Pendu, Y., et al. (2024). Response of Guiana dolphins to the construction of a bridge in Ilhéus, Northeastern Brazil. PLoS ONE, 19(11), 1–15. https://doi.org/10.1371/journal.pone.0312476
Lilley, M. K., et al. (2020). Laterality of eye use by bottlenose (Tursiops truncatus) and rough-toothed (Steno bredanensis) dolphins while viewing predictable and unpredictable stimuli. International Journal of Comparative Psychology, 33. doi: http://dx.doi.org/10.46867/ijcp.2020.33.03.01
Weidner, F., et al. (2017). Comparing VR and non-VR driving simulations: An experimental user study. 2017 IEEE Virtual Reality (VR), Los Angeles, CA, pp. 281-282, doi: 10.1109/VR.2017.7892286.
