Moderated mediation in SPSS using Hayes Process macro (August, 2019) 

Thanks Supreet! Best wishes!

I appreciate it, Steven! Cheers!

I'm glad they were helpful! Thanks for visiting my site!

Thanks Jack! Best wishes!

I appreciate you visiting, Yulin!

Hi Lily, thank you for visiting! I also wanted to let you know that I have some newer videos out on moderated mediation from this spring. You might check out: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-oYMV8yWk_7Y.html ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-1fiNuGtXc6I.html ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-nCh12SHXPo0.html ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-Z-3hzK2fC6g.html Best wishes!

That's very kind to say. I appreciate you visiting and taking the time to comment. Best wishes to you!

Hi Chris. I just saw your post. I've never had occasion to do what you are asking, but I just gave it a try using some data I had open. It looks like you can add in parallel mediators by including multiple mediators in the mediators box and then test for the conditional indirect effects involving those mediators. Best wishes.

If you have more than one independent variable with paths not assumed to be moderated by a variable W, then you can simply add them in as covariates and direct effects of these variables on the DV can be estimated directly. Best wishes!

Mike Crowson How can we do moderated mediation with more than 1 IV? Thanks :)

Hi Renee, thanks for your question. If you have a variable affected by depression, then it sounds like you have a serial mediation model (e.g., X-->M1-->M2-->Y) that you are proposing. Hayes' process macro (template #6; see www.personal.psu.edu/jxb14/M554/specreg/templates.pdf ) allows you to test for serial mediation. However, if you are asking whether you can test for moderation of serial mediation, then I don't believe I've seen a template that addresses that question. Hope this helps!

The conditional indirect effects represent the indirect effect of the IV on the DV at different levels of the moderator variable. Cheers!

Hi Zhu, you stated above that the p-value for the interaction term was >.001 (indicating the moderation effect is not significant). If you are getting a p-value in the output of .001 or less, typically this is taken as an indication of statistical significance. Hope this helps.

Hello. A moderator variable is one that alters or changes the relationship between two other variables. In other words, the relationship between X and Y is conditional depending on the level of variable Z. You can think of a moderator as answering the 'when' question as it pertains to the relationship between X and Y. For example, when the moderator z equals some value, the relationship between X and Y takes on one value. When then moderator z equals some other value, then the relationship between X and Y is something else. A mediator variable is one that theoretically 'transmits' the variation from X to Y. For instance, we might say that the effect of mastery goals (X) on student achievement (Y) is accounted for by deeper cognitive engagement (M). So the proposed causal path would be: mastery goals (X)-->engagement (M)-->achievement. The classic article on the distinction between mediators and moderators is Baron & Kenny (1986). You should be able to download a copy from the web at this address (I quickly googled to locate a copy): www.sesp.org/files/The%20Moderator-Baron.pdf

Hello, do you have that PPT? I am currently developing a model and I used the model 5, but i did not find so much information about how to interpret and write the hypothesis.. I will really apreciate if you have something that can help me. Thanks in advance.

I have just written a similar comment. :D

Hayes has just simply provided numbers for the different models that are available in Process. As far as I can tell, there is nothing special about his numbering. Best wishes!

Process does allow you to include nominal and ordinal variables in various ways. However, how they can be included depends on the model you are running. If you are performing a mediation analysis, then the IV (X) can be nominal or ordinal (if doing this, you would indicate to the program that the variable is categorical using the 'Categorical option). The only categorical outcome (Y) you can use in a mediation model is a binary outcome variable. If you have a binary outcome, Process will recognize it and will run a logistic regression as the second regression in your path model. If you are running a moderated multiple regression then your IV and/or Moderator variable can be categorical. The DV can also by binary (as described above). If it is binary then then regression will be a moderated binary logistic regression. There are other more complicated models that incorporate mediation and moderation and with various 'rules' for what types of variables can be included. In general, the above rules for mediators, moderators, and outcome variables stand. There is NO mechanism to explicitly specify a mediator as categorical. If your mediator is nominal (for instance), the program will run OLS regression for that portion of your mediator model, which will give you incorrect results. If your mediator is ordinal, it will do the same thing (although this might be a more justifiable approach to testing mediation if you have enough scale points; but there is debate on this issue; but there is debate on this point). Hope this helps. Cheers!

I believe the moderator at that point in the demonstration was originally continuous. At the 12:00 minute mark, I was demonstrating the plotting of simple slopes at 3 (arbitrary) levels of the moderator (i.e., at -1sd, mean, and +1sd). If you have a binary moderator, then I would encourage you not to mean center it, but to leave it in it's original metric and then you'd be plotting only 2 simple slopes (one for each level of your moderator). Best wishes!

@@mikecrowson2462 Sir thank you for your well educated answer! Have a great day

​@@mikecrowson2462 Would you mean center when both IV and Moderating variables are dummy variables, mediating variable is continuous and DV is dummy?

There's actually a couple of centering options in Process. Usually, I will mean center the continuous predictor variables to facilitate interpretation. However, technically mean centering is not a requirement. Since moderated mediation involves at least one regression in comprising the full model including an interaction term (capturing the moderated effect), it might be worth reviewing that topic. Here is a video you might check out: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-JZ2fl8exNTE.html

@@mikecrowson2462 thank you Prof for valuable answers

The index of moderated mediation is a test of whether the conditional indirect effects vary across levels of the moderator. It is not a test of the conditional indirect effects themselves (in terms of them being different or not different from zero). So, if the index of moderated mediation is not significant, then you have no evidence of moderated mediation. The fact that the conditional indirect effects are significant is unrelated to whether you have moderated mediation. So given the fact that you have no evidence of moderated mediation, you might re-specify your model as one with only the mediation (without the moderation of the indirect effect). Of course if your original hypothesis was moderation of the mediation, then you should be sure to discuss your original model (moderated mediation) and why you chose to re-specify it as mediation only. By the way, I do have several newer videos on moderated mediation. You might check out: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-oYMV8yWk_7Y.html and ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-1fiNuGtXc6I.html . Cheers!

The null hypothesis when testing mediation (i.e., indirect) effects is the same as the typical null hypothesis in other tests: that is, the population effect is 0 (null). If the null of 0 falls between the lower and upper bound of a confidence interval then you maintain the null that the population indirect effect is 0, whereas if 0 falls outside the lower and upper bound of an interval, you reject the null and infer the population indirect effect is not 0 (i.e., statistically significant). The same logic is applied to the confidence interval associated with the Index of Moderated Mediation (IMM). The null in this case is that the IMM=0 (meaning no evidence of significant moderation of a mediated effect), and if 0 falls within the confidence interval for IMM, then this is interpreted to mean there is no evidence of significant moderation of that mediated effect). If 0 falls outside the confidence interval, then it is an indication that the IMM is considered significant, in which case you interpret this to mean that the indirect effect is conditioned on the moderating variable. Hope this helps! cheers!

Hi there. The confidence intervals are given for the regression coefficients as LLCI and ULCI for upper and lower bound of the confidence interval. The percentile bootstrap applied when bootstrapping is provided as: BootLLCI and BootULCI. The default is 95% confidence intervals. You can change that default on the main menu screen by using the Confidence Intervals drop down. And if you want bootstrap confidence intervals for your regression parameters, just click the little box for bootstrap inference for model coefficients. I hope this helps!

@@mikecrowson2462 thank you so much 😊

I have seen that Hayes models are using one independent variable. I'm curious if I can analyze my model using PROCESS Macro or better using AMOS.

Hi there. I'm pretty sure that you can only test moderation of the mediated effect of only one independent variable at a time. Maybe one day Hayes comes up with a system for testing moderation of multiple paths in a single model. Until then, you probably are better off trying to model using another software. You might look at a software like jamovi (www.jamovi.org/download.html). They have a module in there that would allow you test multiple moderated paths. The only downside is that they don't have tests of an index of moderated mediation like Hayes does. Cheers!

Hi there. Unfortunately, there is no mechanism in Process to print out diagrams. The diagram I created was done using Powerpoint. I believe that the jamovi program might have the capability to draw out a diagram for you, but I haven't spent a whole lot of time on moderated mediation using that program. Cheers!

Hi Madhu, the answer to your question is 'yes'. You can include categorical moderators. I don't really cover that in this video, except where you have a binary moderator. But technically you can include a categorical moderator with more than two levels. But you'll have to use the Multicategorical option in Process. I don't have a video on using it in the context of moderated mediation models. However, I do have a video on using the multicategorical option in the context of moderated regression. This video (ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-IG-FnhvTud4.html) might provide you wish some insight for how to use the option with moderated mediation models. At some point I hope to do something with a multicategorical moderator in moderated mediation. Cheers!

@@mikecrowson2462 sir, actually .. I want to check 3 type of categorical variables as moderator in the model..Is it possible.. to check this and how? Please guide me.

Hi Muhammed, the IMM indexes the change in conditional slopes / effects) across levels of your moderator. The simple slopes tests are tests of conditional effects at specific points (eg 16th, 50th, 84th percentiles) of the moderator. These tests are not a test of change in conditional effects, as indexed by IMM. To get a better picture of how slopes vary across the moderator you might request the Johnson Neyman output. These are still simple slopes, but you get a larger number of slopes conditional on the moderator and you can better see the change in slopes over the moderator.

Dear Mike i will do that analysis. İs there any book or article which explain all principles of moderated mediation analysis? Could you recommend?

@@muhammedsabrisirin5324 You might check out Hayes' book on Conditional Process modeling. That's where I learned everything I know. Cheers! www.amazon.com/Introduction-Mediation-Moderation-Conditional-Analysis/dp/1462549039

Dear Mike i have bought book and looked Fundamentals but i have not seen my question's answers. İf you mind, i would like to send output to you? Could you look at it?

Dear Foad, I'll see what I can do on how to perform the same analyses using more open source software. I've run across a Shiny App version of Process before (but can't seem to relocate it). It's a web-based version of the program created using R (I believe). If I find it again I'll try to do something with it. However, a really great open-source program that I've been encouraging is jamovi (which is a pretty easy menu-based program that is built on R programming). The developers continue to add in new functions all the time. You can download it here: www.jamovi.org/download.html . They also have a community of contributors in terms of videos and other materials. I don't have a video on moderated mediation using jamovi but you CAN perform analyses involving moderated paths in path analysis using the MedMod module (here's a very little something I put together for you to see: drive.google.com/open?id=1zJ1q24SEDuwf84JrOxwZwTPqkIzqTD1f). I do have a video using a separate module under MedMod on moderated regression (in case you are interested): ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-ezwug5d--v8.html (my apologies but the audio will not be as good as this was made before I improved the audio for my videos starting this summer). Finally if you are interested in other R videos (on regression, SEM, etc.) check out the following playlist: ru-vid.com/group/PLITf5JBUH0l6My-aSE4CnROnPS9D6WuI8 . Hope this helps! best wishes and thanks for visiting!