Solved – Index of Moderation Mediation vs Conditional indirect effects at values of the moderator

confidence intervalinteractionmediationspss

I'm analyzing a data set with following variables:

X: Continuous (IV)

Y: Continuous (DV)

M: Continuous (Mediator)

W: Categorical [0, 1] (Moderator)

To run moderated mediation analysis, I'm using Hayes Process macro: Model 8.

Here is the result:

As you can see in the first part, in Condition "0", the CI of indirect effect does not include zero whereas in Condition "1" it does.

However, the CI of moderated mediation index includes zero. Meaning the conditional indirect effect is not significantly different across the two conditions.

How can I make a conclusion here? Can I base my conclusion on the CIs of conditional indirect effects and say that, in condition zero there is evidence for mediation which is not supported in condition 1?

Thanks,

Best Answer

I have your answer. The index value is the difference in the slopes for the indirect effect ie .3791-.3320. So your results are not contradictory. The fact that one indirect effect is significant and the other is not is not what matters. The index and its CI is a formal test as to whether the difference in the indirect effects is significant. Hope this helps. Alastair Anderson , Cairnmillar Melbourne

Related Solutions

Solved – Moderated Mediation with categorical moderator

It has been a while since the OP, but for posterity and future readers, I will point out that the latest vignette for the mediation package on CRAN provides an example of moderated mediation in section 3.2, showing how to obtain pairwise comparisons of indirect effects between levels of Condition. So something like

med.init <- mediate(medmod.fit, outmod.fit, sims = 2,
                    treat = "Spatial", mediator = "TotCorr")
## Compare conditions 1 and 2
test.modmed(med.init, sims = 1000,
            covariates.1 = list(Condition = 1), 
            covariates.2 = list(Condition = 2))
## Compare conditions 1 and 3
test.modmed(med.init, sims = 1000,
            covariates.1 = list(Condition = 1), 
            covariates.2 = list(Condition = 3))
## and so on ...

Solved – PROCESS macro in SPSS – Analysis for multiple IV

I could finally find the answer in the Hayes' book (pp. 196-197):

As discussed in section 6.2, PROCESS can estimate a mediation model with statistical controls as in Figure 6.3, so it follows that it can also estimate a model with multiple X variables. However, in order to estimate the direct and indirect effects of all k X variables in Figure 6.5, PROCESS must be executed k times, each time putting one Xi in the model as X and the remaining k − 1 X variables as covariates. Each time PROCESS is run, the direct and indirect effects of the variable listed as X will be generated. Repeating k − 1 times generates the indirect effects for all k X variables. Mathematically, all resulting paths, direct, and indirect effects will be the same as if they had all been estimated simultaneously (as in a structural equation modeling program). [...] In the preceding code, the random number generator is seeded with a common (and arbitrary) seed using the seed= option so that at each run, the bootstrap confidence intervals will be based on the same set of 10,000 resamples from the data. It makes no difference what seed is used—pick it randomly if you choose as I did in the code above—so long as it is a positive integer.

Best Answer

Related Solutions

Solved – Moderated Mediation with categorical moderator

Solved – PROCESS macro in SPSS – Analysis for multiple IV

Related Question