You must take into account that they are significant and depending on your objectives considering several possible effects of measurement to determine that the direct effects are not significant.

Hello Jun,

As Cristian Ramos-Vera 's reply states, you must acknowledge that the estimated mediation effects are non-zero. That's a statistical significance issue. The practical significance/import of the effects, given the apparently small magnitude, is another matter altogether, and likely depends on the specific ensemble of variables under investigation, your research question(s), and the audience.

Why the audience? Consider as a simple example, a difference of 1% change in the cost of petrol/gasoline for automobiles. For the individual auto owner, that's not a substantial effect. For the vendor who sells millions of liters/gallons annually, it becomes a noteworthy effect.

Good luck with your work.

Hi Jun,

beyond what David already mentioned, two thoughts:

a) Simply note (and reflect that in the paper) the fundamental difference between a significant null hypothesis test (NHST) which targets the question "is the effect zero or non-zero" and a more fundamental issue about a theory-relevant effect. By totally adhering to NHST, we duck away from these more relevant questions. That is, your indirect effect may be significant (which is, see above, the relevant criterion for your paper and its hypothesis) but practically/theoretically it is irrelevant. This problem gets worse the larger the sample becomes

b) A potential add-on would be to calculate the total effect and how much (in %) of this effect is mediated. This does not completely turn a non-substantial effect into a substantial effect but is something worth mentioning

Best,

Holger

Cristian Ramos-Vera Thank you so much for your reply. It really helps.

Holger Steinmetz Thank you always for your help!

David Morse Thank you so much for your kind reply. It was helpful!