moderator effect with interaction term

[mario_super]

Dear Community,

I have a question regarding moderator effects with interaction terms.

Is it possible to only consider the moderator effect on the dependent variable without the direct effect of the variable used as the moderator (see left side of the picture)?

For this case, I calculated the interaction terms by myself (product indicator approach) (see left side) and I get significant results.
If I do it the way smartPLS suggests (right side), with the direct effects of the moderator variables, I get totally different and also not significant results.

Moderators: Trust and Competence
Independent variable: Offering MU system
Dependent variable: Perceived Uncertainty
Arrows: Path coefficients and p-values

Looking forward to an answer. Thank you.

Best regards,

Mario

[jmbecker]

The reason why we do not allow to exclude the direct effect is very simple: it is wrong and results are not meaningful. You might get nicely significant results, but you would not test a correct model and thus could also do any other data hack to get significant results.

I think we also discussed this issues several times already in the forum. Please also search for those discussions on more information about the issue.