Good morning/afternoon/evening everybody.
As the title said, i am having a confusion in my discussion writing. The result of of my data analysis lead to a non-significant result. My question, what to write in discussion with the said result.
For information, i am doing a moderation study, and the result are:
- There is no correlation between my predictor toward my criterion variable
- There is strong correlation (significant) between my moderator variable and my criterion variable
- There is no correlation between my interaction variable (predictor times moderator) and my criterion variable
Hope by giving these information, somebody could understand better and help me to work on my discussion
Thank you hope, somebody reply and have a nice day
it is a difficult question, particularly when it is hard to get negative results published. I would discuss whether you think the absence of association is real or whether it is a sample size issue (trying to detect a small effect with a small sample size will often result in no association)-
Most importantly, remember that no association does not mean your project has failed negative results can be extremely important. but fidning out whether it is negative because there truly is no association or because your study design didn't allow you to detect the association is important- and if the answer is the former, I would discuss the implications for the field
The answer depends on how the study was designed. Was the power specified? If so, then you can claim that there is no relevant effect with a confidence given by the power. If not, the results are actually of no use for a publication. They will tell you that the effort you took is not sufficient, so you may decide if you increase the effort or change the topic - but this is generally nothing a journal is interested in.
My basic recommendation is that you report that "there is not a significant relationship between...at the .0x level of significance". However, it is always permissible to report what you did find. Also, if you had a relatively small sample size and the effect size was adequate, you can calculate the number of subjects that would be required to attain significance and report that. If that value was within reason, you could recommend that the research be repeated with a larger sample size to determine to determine if your result was a design error (too small a sample size) or the relationship is truly non-significant.
James, you write "...relationship is truly non-significant". This is an interesting formulation. Here you seem to mix several logically different things. The "significance" of a result, as meant in the original post, is a statistical significance = the likelihood to get "more extreme test statistics" than the one actually observed. Your "level of significance" is a cut-off to control a type-I error at a specified rate. The "true significance of a relationship", as you meantioned it in your sentence I cited above seems to aim at some "truth", what is a logically completely different concept and is not related to the statistical inference. No test ever tells you something about a "truth". Tests tell you something about your data, based on assumptions and hypotheses. This may convince you to *believe* in the (un-)reasonability or (un-)usefulness or (un-) helpfulness of a hypothesis you specified. It has nothing to do with a "truth" (unless "truth" is defined as the collective agreement on something by all [relevant?] people).
Hypotheses are neither true nor false. They are models or assumptions in our heads. We need to specify hypotheses to use probability calculus, since any propability value is neccessarily conditional of some context (a part of which we call a [statistical] hypothesis).
Inductive behaviour, as it is made possible through the application of hypothesis tests, is not about the "truth" of any of the tested hypotheses, but about the sensible control of expected costs and benefits of actions that are decided to be taken based on the data we observe.
- Badges
- Science topic