Thanks for your video and explanation! Why is it better to use model 8 instead of testing interaction effects directly? What is the advantage of model 8 compared to the test of Region of Significance and Proportion of Interaction and the test of crossover point?
I use model 8 when I want to test a conditional indirect effect because model 8 (or in general the moderated mediation models in PROCESS) lets me do that directly.
this was very helpful, thank you! I have some questions if you may be able to help: My mediator significantly predicts my outcome, but no other pathways were significant(p > 0.05). 1. I am using a categorical moderator (2 levels, 0 = male 1 = female) The moderated mediation index is not significant but the conditional indirect effects are as follows : 0.0000 -0.0257 0.0466 -0.1128 0.0751 1.0000 0.0924 0.0478 0.0088 0.1980 so the conditional direct effect for females looks significant.. why then might the pathway not be moderated? 2. The bootstrap results for my outcome variable also say that my moderator and mediator both significantly predict the outcome, though this is not the case for the first output for my outcome variable (bio sex marginally above sig level, p= 0.06). Which result do I report? if you have the time to help, I would really appreciate it (I'm new to statistics!) thanks again :)
1. A difference in significance of two effects does not necessarily imply a significant difference between those effects. It is possible that one effect is significant, another effect is not significant, but they are not significantly different from each other. 2. If the normal results and the bootstrap results contradict each other I would trust the bootstrap results (because if the regression assumptions are met then the results should be the same - if the results differ then that is an indication for a violation of assumptions and in that case only the bootstrap results are trustworthy).
Thanks! Is it still meaningful to explain the main effects (IV/moderator to DV in your video) when the interaction effect is significant in Model 8 and 10?
The interpretation of main effects is highly problematic in all models with significant interactions. I think this applies to moderated mediations, too. Normally, in models with significant interactions the 'main effects' are not true main effects but only effects conditional on the moderator (or IV for the 'main effect' of the moderator) having a value of zero. With continous variables that are mean-centered the values of zero corresponds to average values of the variable, so in this case you could get to a meaningful interpretation (but I haven't really thought about it in the context of a moderated mediation before, only with simple moderations; however I think you might transfer that to the case of a moderated mediation).
Thank you for your tutorial! Could you please help me with one question? I have a multi-categorical IV with 4 categories, a moderator with 3 (low, average, high) categories and continuous DV. Can I still use Process macro model 1 if I indicate that X and W are multicategorical? Thank you in advance for your help.
Thank you so much for your sharing! Meanwhile, do you know how to do the sensitivity analysis and find the effect size for moderated mediation? Thank you so much!
Thank you for your video! However, I still do not fully understand whether I can use model8 if I also want to investigate the moderated effect of X on M. For example, I do have two hypothesis, a) concerns the moderated mediation effect and b) the moderated effect of X on M only. Hence, the question is as follows: Can I use Model 8 to test both hypothesis or do I have to conduct a second analysis with (e.g., Model 1) to test for the moderated effect of X on M? Thank you in advance.
I think the model for the a-path for model 8 is a simple moderation, yes. If you want to be sure, just run both models, 1 and 8. I believe the results for the a-path of model 8 will mirror the results of model 1.
Thank you so much for this video! I have a question on understanding and interpreting results from model 8. I have significant interaction effect (between X and W) on both the a and c path; also significant main effect of X on a and c path (X predicting M, X predicting Y); M is also significantly predicting Y (significant b path). But the indirect effects on different levels of W are not significant, nor is the index of moderated mediation. So it is apparent that there is no mediation or moderated mediation here. But can I still report the significant relationships in path a, b, and c? What might be some explanations to the non-significant mediation while both a and b paths are significant? Would like to know if you have any suggestions. Thank you so much!!
I think you can report what you have found, in your case the significant effects and the nonsignificant index and conditional indirect effects. I haven't hat such a situation yet. Maybe one explanation could be that the b-path is moderated by W, too. So I would try fitting PROCESS model 59 (= W moderating a, b, and c') to investigate that possibility.
@@RegorzStatistik Thank you so much! I will definitely explore the analysis a bit more. Truly appreciate your excellent videos and explanation! Hope you are doing well!
Hi, danke für das Video, sehr anschaulich und einfach erklärt! Mein Problem ist nur, das jetzt alles in einer schönen und anschaulichen Tabelle richtig darzustellen. Und vor allem das alles in Worten wissenschaftlich richtig wiederzugeben. Gibt es denn ein Video oder eine Studie, in der ich sehen kann, wie ich das alles richtig mache?
Ich weiß nicht, ob sich da schon ein einheitlicher Standard etabliert hat, wie man das am besten darstellt - ich habe dafür zumindest kein Muster. Ich würde empfehlen, dass Sie in Ihrem Fachgebiet nach Studien suchen, die mit PROCESS model 8 gearbeitet haben, und mich daran orientieren.
Thanks a lot for this explanation! I have a short question: I plan to include three moderators instead of only one. How should I best proceed here? Should I conduct three analyses, each time with a different moderator and the other two moderating variables as controls?
Thank you so much for your quick response,@@RegorzStatistik . One last question on this: would it make a difference regarding the comparability of results if I chose a PROCESS model which handles 2 moderators and including only the third variable as a covariate (and then doing this twice)? As an alternative to do it three times as proposed initially.
@@ninico436 If you have 3 moderators and are interested in the interplay between the moderators, too, I would recommend using path modeling instead of PROCESS.
Thanks for your kind explanation! I was wondering whether the mediated moderation in repeated measures(within-subject design) from model 8 can also be tested. Thanks.
Hi thanks for the video. For my analysis, I find a non-significant index of moderated mediation as well as a non-significant interaction term on the indirect effect path (path a). However, the interaction term on the direct effect (path c prime) is significant and since I am using dummy coding for the moderator, the direct effect of one level is significant while the other is not. How would you interpret that? Thanks!
@@RegorzStatistik I see, thank you. My initial thought was that the direct effect includes the mediator otherwise it would be total effect. Am I wrong to think that?
The total effect includes the indirect effect via the mediator. The direct effect doesn't. (e.g. in a simple mediation model the total effect would be c = c' + ab )
Hi, eine Frage hätte ich, wenn ich so vorgehe, wie Sie im Video also auch das Model 8 verwende und "Bootstrap inference for model coefficents" auswähle und auch die anderen Optionen so wie Sie im Video wähle, muss ich dann dennoch die Annahmen für Regression per Hand checken, bevor ich die moderierte Mediation ausführen darf?
Die Prüfung der Annahme Normalverteilung und Homoskedastizität entfällt dann, weil man mit Bootstrapping und robusten Standardfehlern Verfahren einsetzt, die gegenüber Verletzungen dieser beiden Annahmen robust sind. Die anderen Annahmen müsste man dann ggf. noch prüfen, wenn man es genau nimmt. Skaleneigenschaften und Unkorreliertheit der Residuen ergibt sich normalerweise aus der Untersuchungsmethode. Was ich auf jeden Fall prüfen würde, ist die Linearitätsannahme (mit Streudiagrammen zwischen jeweiligen Prädiktoren und Kriterien für die verschiedenen Teilmodelle). Ob ich die Voraussetzung der Abwesenheit starker Multikollinearität (Prüfung mit VIF oder Toleranz) noch prüfen würde, weiß ich nicht, aber wenn man es ganz gründlich machen möchte, würde auch das noch dazu gehören. Hinsichtlich von Ausreißern ist m.E. Bootstrapping relativ robuster als sonstige Verfahren, weil es dort dann ja auch Bootstrapping-Stichproben ohne diese Ausreißer gibt, insofern bin ich mir auch nicht sicher, ob ich den Aufwand zusätzlicher Ausreißerprüfung betreiben würde. Aber am Ende sind alles Regressionen, mit den üblichen Regressionsvoraussetzungen.
@@RegorzStatistik Danke! Eine meiner Hypothesen war dass es einen signifikanten DE gibt und darauf bauend eine weitere Hypothese, dass ebendieser Effekt moderiert wird. Wie kann man das genau interpretieren?
@@smndyskDen signifikanten direkten Effekt kann man aus Model 8 nicht ohne weiteres ablesen, für so eine Hypothese bräuchte man dann eher Model 4 oder Model 7.
How to REPORT a MODERATED MEDIATION (Tutorial):
ruclips.net/video/DfpnnXceTi4/видео.html
This was incredibly helpful, thank you so much!
Thank you very much for a good explanation.
It was really helpful.
Thanks for your video and explanation! Why is it better to use model 8 instead of testing interaction effects directly? What is the advantage of model 8 compared to the test of Region of Significance and Proportion of Interaction and the test of crossover point?
I use model 8 when I want to test a conditional indirect effect because model 8 (or in general the moderated mediation models in PROCESS) lets me do that directly.
this was very helpful, thank you! I have some questions if you may be able to help: My mediator significantly predicts my outcome, but no other pathways were significant(p > 0.05). 1. I am using a categorical moderator (2 levels, 0 = male 1 = female) The moderated mediation index is not significant but the conditional indirect effects are as follows :
0.0000 -0.0257 0.0466 -0.1128 0.0751
1.0000 0.0924 0.0478 0.0088 0.1980
so the conditional direct effect for females looks significant.. why then might the pathway not be moderated?
2. The bootstrap results for my outcome variable also say that my moderator and mediator both significantly predict the outcome, though this is not the case for the first output for my outcome variable (bio sex marginally above sig level, p= 0.06). Which result do I report?
if you have the time to help, I would really appreciate it (I'm new to statistics!)
thanks again :)
1. A difference in significance of two effects does not necessarily imply a significant difference between those effects. It is possible that one effect is significant, another effect is not significant, but they are not significantly different from each other.
2. If the normal results and the bootstrap results contradict each other I would trust the bootstrap results (because if the regression assumptions are met then the results should be the same - if the results differ then that is an indication for a violation of assumptions and in that case only the bootstrap results are trustworthy).
Thanks! Is it still meaningful to explain the main effects (IV/moderator to DV in your video) when the interaction effect is significant in Model 8 and 10?
The interpretation of main effects is highly problematic in all models with significant interactions. I think this applies to moderated mediations, too.
Normally, in models with significant interactions the 'main effects' are not true main effects but only effects conditional on the moderator (or IV for the 'main effect' of the moderator) having a value of zero.
With continous variables that are mean-centered the values of zero corresponds to average values of the variable, so in this case you could get to a meaningful interpretation (but I haven't really thought about it in the context of a moderated mediation before, only with simple moderations; however I think you might transfer that to the case of a moderated mediation).
@@RegorzStatistik Yes mean-center is a way to provide meaningful interpretations. Thanks!
Thank you for your tutorial! Could you please help me with one question? I have a multi-categorical IV with 4 categories, a moderator with 3 (low, average, high) categories and continuous DV. Can I still use Process macro model 1 if I indicate that X and W are multicategorical? Thank you in advance for your help.
Thank you so much for your sharing!
Meanwhile, do you know how to do the sensitivity analysis and find the effect size for moderated mediation?
Thank you so much!
Unfortunately not.
@@RegorzStatistik Okay, thank you!
Thank you for your video! However, I still do not fully understand whether I can use model8 if I also want to investigate the moderated effect of X on M. For example, I do have two hypothesis, a) concerns the moderated mediation effect and b) the moderated effect of X on M only. Hence, the question is as follows: Can I use Model 8 to test both hypothesis or do I have to conduct a second analysis with (e.g., Model 1) to test for the moderated effect of X on M?
Thank you in advance.
I think the model for the a-path for model 8 is a simple moderation, yes. If you want to be sure, just run both models, 1 and 8. I believe the results for the a-path of model 8 will mirror the results of model 1.
@@RegorzStatistik thanks a bunch!
Thank you so much for this video! I have a question on understanding and interpreting results from model 8. I have significant interaction effect (between X and W) on both the a and c path; also significant main effect of X on a and c path (X predicting M, X predicting Y); M is also significantly predicting Y (significant b path). But the indirect effects on different levels of W are not significant, nor is the index of moderated mediation. So it is apparent that there is no mediation or moderated mediation here. But can I still report the significant relationships in path a, b, and c? What might be some explanations to the non-significant mediation while both a and b paths are significant? Would like to know if you have any suggestions. Thank you so much!!
I think you can report what you have found, in your case the significant effects and the nonsignificant index and conditional indirect effects.
I haven't hat such a situation yet. Maybe one explanation could be that the b-path is moderated by W, too. So I would try fitting PROCESS model 59 (= W moderating a, b, and c') to investigate that possibility.
@@RegorzStatistik Thank you so much! I will definitely explore the analysis a bit more. Truly appreciate your excellent videos and explanation! Hope you are doing well!
Hi, danke für das Video, sehr anschaulich und einfach erklärt!
Mein Problem ist nur, das jetzt alles in einer schönen und anschaulichen Tabelle richtig darzustellen. Und vor allem das alles in Worten wissenschaftlich richtig wiederzugeben. Gibt es denn ein Video oder eine Studie, in der ich sehen kann, wie ich das alles richtig mache?
Ich weiß nicht, ob sich da schon ein einheitlicher Standard etabliert hat, wie man das am besten darstellt - ich habe dafür zumindest kein Muster. Ich würde empfehlen, dass Sie in Ihrem Fachgebiet nach Studien suchen, die mit PROCESS model 8 gearbeitet haben, und mich daran orientieren.
Thanks a lot for this explanation! I have a short question: I plan to include three moderators instead of only one. How should I best proceed here? Should I conduct three analyses, each time with a different moderator and the other two moderating variables as controls?
Since PROCESS can handle only 1 or 2 moderators I think it makes sense to proceed the way you have written.
Thank you!@@RegorzStatistik
Thank you so much for your quick response,@@RegorzStatistik . One last question on this: would it make a difference regarding the comparability of results if I chose a PROCESS model which handles 2 moderators and including only the third variable as a covariate (and then doing this twice)? As an alternative to do it three times as proposed initially.
@@ninico436 If you have 3 moderators and are interested in the interplay between the moderators, too, I would recommend using path modeling instead of PROCESS.
Okay, thank you so much!@@RegorzStatistik
Thanks for your kind explanation! I was wondering whether the mediated moderation in repeated measures(within-subject design) from model 8 can also be tested. Thanks.
I think you would have to use some advanced modelling approaches, e.g. multilevel modeling, for that.
Hi thanks for the video. For my analysis, I find a non-significant index of moderated mediation as well as a non-significant interaction term on the indirect effect path (path a). However, the interaction term on the direct effect (path c prime) is significant and since I am using dummy coding for the moderator, the direct effect of one level is significant while the other is not. How would you interpret that? Thanks!
I think in this case I would run PROCESS model 5 (mediation in which only the direct effect c' is moderated) and interpret its results.
@@RegorzStatistik I see, thank you. My initial thought was that the direct effect includes the mediator otherwise it would be total effect. Am I wrong to think that?
The total effect includes the indirect effect via the mediator. The direct effect doesn't. (e.g. in a simple mediation model the total effect would be c = c' + ab )
Hi, eine Frage hätte ich, wenn ich so vorgehe, wie Sie im Video also auch das Model 8 verwende und "Bootstrap inference for model coefficents" auswähle und auch die anderen Optionen so wie Sie im Video wähle, muss ich dann dennoch die Annahmen für Regression per Hand checken, bevor ich die moderierte Mediation ausführen darf?
Die Prüfung der Annahme Normalverteilung und Homoskedastizität entfällt dann, weil man mit Bootstrapping und robusten Standardfehlern Verfahren einsetzt, die gegenüber Verletzungen dieser beiden Annahmen robust sind. Die anderen Annahmen müsste man dann ggf. noch prüfen, wenn man es genau nimmt. Skaleneigenschaften und Unkorreliertheit der Residuen ergibt sich normalerweise aus der Untersuchungsmethode. Was ich auf jeden Fall prüfen würde, ist die Linearitätsannahme (mit Streudiagrammen zwischen jeweiligen Prädiktoren und Kriterien für die verschiedenen Teilmodelle).
Ob ich die Voraussetzung der Abwesenheit starker Multikollinearität (Prüfung mit VIF oder Toleranz) noch prüfen würde, weiß ich nicht, aber wenn man es ganz gründlich machen möchte, würde auch das noch dazu gehören. Hinsichtlich von Ausreißern ist m.E. Bootstrapping relativ robuster als sonstige Verfahren, weil es dort dann ja auch Bootstrapping-Stichproben ohne diese Ausreißer gibt, insofern bin ich mir auch nicht sicher, ob ich den Aufwand zusätzlicher Ausreißerprüfung betreiben würde. Aber am Ende sind alles Regressionen, mit den üblichen Regressionsvoraussetzungen.
Danke für das Video! Wie muss ich vorgehen, wenn der direkte Effekt nicht signifikant ist, aber die Moderation davon schon?
Das einfach berichten. Entscheidend ist ja bei der moderierten Mediation eher der indirekte Effekt.
@@RegorzStatistik Danke! Eine meiner Hypothesen war dass es einen signifikanten DE gibt und darauf bauend eine weitere Hypothese, dass ebendieser Effekt moderiert wird. Wie kann man das genau interpretieren?
@@smndyskDen signifikanten direkten Effekt kann man aus Model 8 nicht ohne weiteres ablesen, für so eine Hypothese bräuchte man dann eher Model 4 oder Model 7.
what happens if the direct effect mediation didn't work?
Do you mean the *moderation* for the direct effect? If so: That depends whether you have a hypothesis about the direct effect.