Moderated Mediation - Results Table with indirect, direct, total effects --> swapped lines?

[denisedoerfel]

Hello, I am using Jamovi-amm, version 1.6.23.0 to analyse the following: x1, x2 as predictors m as mediator, y as outcome (there are additional 3 covariates, I will neglect here). My model is y ~ m + x1 + x2 + x1*x2 (+ additional covariates). x2 was entered via interaction into the mediator model. (not as moderator in the moderator model).

The results are: Direct effecs on outcome: x1 --> not significant, x2 --> significant , interaction --> not sign. Indirect effects on outcome: x1 --> sign. , x2 --> sign., interaction --> not sign. Direct effects on mediator: x1 --> sign., x2 --> sign. So far everything fits, but the total effects puzzle me: Total effects on outcome: x1 --> sign., x2 --> not sign. (!), interaction --> sign. (!)

Is it plausible that with a highly significant (<.001) indirect effect of x2 and a slighly significant (p=.044) direct effect of x2 on the outcome the total effect of x2 does not become significant with p =.991? Is it possible, that in the results table the lines under total effects are swapped and the results of the interaction are in fact the results of x2?

I'll attach a png with detailed results of my analysis with the acutal labeling of my variables. It's not only a problem of p just not reaching significance, which can be seen in the attached table.
I have also noticed this in similar analyses with other samples and more predictors and one more interaction. With SPSS/PROCESS and lavaan in R the results of the total effects and the direct/indirect effects fit.

It's not only a problem of p just not reaching significance, which can be seen in the attached table.
Thanks for looking into this. Denise

[mcfanda]

Hi, if you are referring to T1_emotional_job_demands when you asked about the plausibility of the results ( highly significant (<.001) indirect effect of x2 and a slighly significant (p=.044) direct effect of x2 on the outcome the total effect of x2 does not become significant with p =.991), the results you posted seem to be correct. The problem is not the significance, but the direction of the effect.

For T1_emotional_job_demands , the indirect effect is -.079, and the direct effect is .079 (both rounded). Since the total effect is the sum of indirect and direct effect, you have Total=-.079+.079=0. Indeed (without rounding), you get -4.36^-4 which is practically 0.

If you want to be sure, run a regression with all variables excluding the mediator. It should confirm a coefficient for T1_emotional_job_demands close to zero and an associated p-value around >.9.

Please let me know how the test went

[denisedoerfel]

Dear Marcello, you are absolutely right, I was so stuck that I didn't consider this. The direction of the effects makes sense and has been hypothesized this way. The regression showed exactly what you predicted, but only if I also leave out the interaction term. Is this correct?

Thanks a lot!!! Denise.

[mcfanda]

if you center the variables involved in the interaction, you'll get the same results also with the interaction in the model

[denisedoerfel]

Ah, I see! Thanks a lot.