Never Use PROCESS Model 14 for Moderated Mediation

[@keziaandrianihasimanana1478]

Thank you very much. As a student from Madagascar, our access to research papers are limited so your video is really helpful.

[@wafaasowan8220]

Thanks for the useful video. I ran model 14 and 15, and the results were very close. There is a moderated mediation, as I understand it. I would like to know how I should report my results. Do I need to report that I conducted model 15 or 14? If I conducted model 15, the reviewer may criticize interaction 1 as not significant, so why didn't I conduct model 14? What is the best way to report it?

[@RegorzStatistik]

I would only use model 15 and justify that by stating that model 14 can lead to biased results if, unexpectedly, the c'-path is moderated, too.

[@wafaasowan8220]

@@RegorzStatistik thank you, I will cite your article

[@AOverload]

Interesting, but what are the implications for statistical power? I'm not sure how to compute power for any of the moderated mediation models (a great idea for a video if you know), but since we're adding an extra interaction term, the use of these alternative models will require a larger sample size, no?

[@RegorzStatistik]

In principle, yes. But as long as there is no moderation of the c'-path in the population, adding this one additional interaction to the model does not decrease the power in a meaningful way. As an example, you could look at the simulation results of the video. For the model where in the population there was no interaction for the c'-path, the CI for the index of moderated mediation based on model 14 was [0.072, 0.142] whereas based on model 15 with the same data it was [0.069, 0.144], only slightly wider.

[@Masterfully1]

Research already has a massive underestimation of the sample size needed to test interaction effects. Rule of thumb is that you need between 2-4 times the sample size to detect an interaction effect than you do a main effect. I don't think this is appreciated by most researchers, who continue to churn out underpowered studies and when significant interactions are found, they'll trend towards overestimations of the interaction.

[@willsonlincon1697]

Where is the dataset?

[@RegorzStatistik]

I have uploaded a file with the 3 datasets to my webpage. The link is in the text version of this tutorial (linked in the video discription) since a direct link to the data file in the video description didn't work.