Dear Reynan,

I think it is not necessary to show the influence of X on Y for different levels of M, if the interaction XM is not significant, unless in an exploratory manner.

I assume you mean centered the predictors for better interpretation (can be done under PROCESS)? If yes, this would for example nicely fit to your presented results. For the mean value of M (0 for centered variables), the influence of X is very close to the result of X alone on Y. The change in b is very small although increasing. It is possible that this is due to increasing CIs the more extreme the values are.

Certainly include all the results.

Would suggest that you produce a graph as well (high, mean, low)\

Importantly, clear statements on what is significant and what is not should be made

Dear Reynan,

I think it is not necessary to show the influence of X on Y for different levels of M, if the interaction XM is not significant, unless in an exploratory manner.

I assume you mean centered the predictors for better interpretation (can be done under PROCESS)? If yes, this would for example nicely fit to your presented results. For the mean value of M (0 for centered variables), the influence of X is very close to the result of X alone on Y. The change in b is very small although increasing. It is possible that this is due to increasing CIs the more extreme the values are.

Thank you Silburn and Rainer.

@ Rainer, yes, Process did the centering. I think you are right too about the small b value. but that's okay, it can be explained by related literature anyway. Thanks again, =)

Thank you Silburn and Rainer.

@ Rainer, yes, Process did the centering. I think you are right too about the small b value. but that's okay, it can be explained by related literature anyway. Thanks again, =)

Reynan,

Rainer's given you excellent advice. You should certainly report the moderation results, but I don't think you need to graph them.

The thing to bear in mind is "not significant" is not equivalent to "is zero in the population." Even with a fairly high p-value, the strictly correct thing to do is retain the interaction in the model to avoid post-hoc mis-specification, but you can justifiably focus on interpretation of the lower-order terms (X->Y) when the moderator is held to (centered) zero.

Pat

Reyan,

some further guidance on how and when to report insignificant interactions in a moderation model:

https://us.sagepub.com/sites/default/files/upm-binaries/47569_ch_15.pdf

"If a test for an interaction is theoretically based or of practical importance, it may be informative to retain the interaction term in the model and report that the interaction was not found to be statistically significant."

"Failure to find statistical significance for an interaction term may occur because of lack of statistical power, unreliability of measures, nonlinear forms of interaction, or sampling error."

Hence if the interaction is theory-based, as a researcher, you would need to speak to the lack of significance found in your study.

"Where an interaction term is not statistically significant and/or not predicted from theory, does not make sense, and/or accounts for a very small proportion of the variance in the regression, the researcher might choose to drop the interaction term from the regression model when the final analysis is reported."

Best,

Silburn

Dear Reynan,

Have a look at:

http://www.unc.edu/~dbauer/manuscripts/preacher-curran-bauer-2006.pdf

and:

http://www.quantpsy.org/interact/hlm2.htm

for an application (scroll down on the page)

cheers,

Pierre