Your video are Awesome, Mike I'm doing a parallel mediation using Hayes 2018 model 4 what other statistical tests would I use for the inclusion of covariates/confounding dependent variables
Hello. I have a question. When I interpret the mediation analysis, should I built a table of the direct and indirect effects and bootstrapping alongwith to add in research article or should I just make a model and add values on regression lines to represent in article? I am confused here thanks
Hello Mr.Mike, thank you for the video! I had a question, is it possible to using non-linear variable (example GDP squared) in mediation analysis? Thank you
Thank you so much for your video. It really helps me to have more understanding of the mediating model. Usually, I did testing the mediating by following the baron and Kenny steps, and it takes much time to do the regression analysis. And now I can do the mediating test wit so easily. I still try to understand and really need your PowerPoint, but unfortunately, I did not find it yet. would you please send me the link? thank you Prof. Mike Crowson
Hi there. Thanks for visiting! I have the link to the Powerpoint underneath the video description, along with the data. But here are the links for your convenience: powerpoint: drive.google.com/file/d/1zTs36CfEEApkZTONjIjWtlOT9tbj0_KR/view data: drive.google.com/file/d/1SRKpaJMQ-LkhM0DCbtFeBoRGS2akgTy0/view I hope this helps! Cheers!
Hello Dr. Growson, I have a question about the direction of mediation: if the effects of a positive mediator (e.g., interest) and a negative mediator (e.g., anxiety) are both significant in a parallel mediation, could that possibly make the total effect insignificant, because the positive and negative effects cancelled each other out?
Hi Mike, thanks for the video! I have longitudinal data. The same group of people answered to a questionnaire in four points in time- one time per month, for fours months. I wanted to test some models. Specifically, I wanted to test a mediation model (model 4 in the macro Process from Hayes). My question is: Should I test this mediation model for point 1, then for point 2, then for point 3 and then point 4? And look at the results and see if the model is significant in each stage? Or, should I do an average of the 4 points for each variable that exists in the model and test once? (That would be like: I do an average seperately for the predictor, outcome and mediator for months 1,2,3, and 4 and then I just test once in the macro where I put these averaged variables). Thank you! And sorry if it is too basic.
Dear Mike! Thank you for this amazing video. Is there a way to run a model featuring both serial and parallel mediators? If not, what would be the approach in that case?
Great video. I have a question. In reference to the parallel mediation model, is there a way to determine the if one specific indirect effect accounts for more of the mediation than the other specific indirect effect?
Thank you for the video. I read somewhere that if IVs direct effect on DV is still significant even with the mediator then it means that the mediator only partially mediates the relationship. In this case you should do the Sobel test to check and if you see that the direct effect of IV on DV is lower with the mediator only after that you can say that the mediator actually mediates the relationship. Is this correct or did I misunderstand something?
Thank you for the explanation and the data so I can practice more. Using the Model 4, can we also have three mediators and will it be the same process? I have the Hayes' book and I noticed there is a Model 4 B2 in the appendix as a matrix. Will it work the same?
Hey mike. I have a simple question. How can i run model 14 with 2 mediators in Hayes, when the moderator is ONLY between first meditator and dependent variable. I want to run whole model at once. Should i put both mediators at once and moderator in (moderator W) box?? Please answer this
Respected Sir I am writing a paper on mediation analysis in which I have 7 IDV 1 Mediating Variable and 1 DV. If we r conducting Mediation in Amos using Bootstrapping so should we run all the regression equation of (IDV> Med> DV) Or seperately for each IDV. As the direct result of IDV on DV in a multiple regression scenario is different from single regression. Kindly suggest the suitable mode. Or should we go for Mediation through Process. Looking forward to hearing from you
Hi Andrew, basically you can include your control variables in the Covariates box. The program will do the rest by treating them as additional IV's in your models [even though you may consider them 'Controls', they are mathematically treated as predictors in the constituent regression models]. As a finer point: If you have a multicategorical Control variable, you will need to recode it (through dummy or effect coding) first. Whereas Process will allow you to specify your X and moderator variables (W or Z) as multicategorical and will do the dummy coding for you, I don't believe you have that option when working with including this type of variable as a covariate in your models. Best wishes!
@@mikecrowson2462 Thanks for the detailed reply! I think I don't have a multicategorical variable, the control variables are all data collected from surveys with 3-point or 5-point Likert scales. If there's a multiple items(questions) with 5-point scales question for one control variables, I basically have to find the mean and then just put it in the Covariates box? Thanks again for the reply, and please kindly correct me if there are mistakes. I love your channel and will share it to more people:) Looking forward to see more videos made by you.
@@mikecrowson2462 Hello! One more extra question! If my control variable included "Gender", does this belong to the multicategorical control variable you mentioned? Should I recode it into dummy with "0" and "1"?What else should I be aware of?
Hi. Process is not really set up to test mediation where you have two or more independent variables' effects being mediated on a DV. An alternative solution is to test mediation of your two IV's through another program. A really easy one to use that would allow you to do this is jamovi (www.jamovi.org/download.html). If you install the MedMod module and then go under the GLM mediation model tab, you can specify and run your analysis. Another option is performing path analysis using an SEM program. Hope this helps!
Thank you for explaining! 🙂 I have a question. What if I have multiple independent variables that I want to test on the same mediator and dependent variable? Do I have to do the Process macro one by one for each of them? Thanks!
Hi there. Yes, to test mediation involving each independent variable you'd have to go one by one, switching out each X variable and including the remaining predictors as covariates. For each model, you'd be controlling for the remaining predictors while testing mediation for a specific predictor. The only other option is using an SEM software (e.g., AMOS). Also, just an FYI, there is an open source software that would allow you to test for indirect effects using OLS regressions in a fuller model (instead of having to piece-meal) things together as I'm suggesting. You can download jamovi here: www.jamovi.org/download.html . There is a module you can add via the library called GLM Mediation Model which will allow you to specify the type of model you are asking about. I have a brief demo at about 10:50 of this video here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-texa0x3zKaM.html . It's a simple demo but may give you a flavor of what I'm talking about. There's also more info here: jamovi-amm.github.io/glm.html I really need to work up a fuller demo of this. Hope all this helps. Cheers!
The constant is the intercept for a regression model. It is the expected value on y for a case with value of 0 on all predictors. Usually the significance test that is given for the intercept is not of substantive interest, and rarely if ever reported
Hi there. The parallel mediation model is comprised of separate OLS regressions, one for each endogenous variable in the model. [An endogenous variable is one that is predicted by others in the model]. So for you to test assumptions associated with each regression making up the model, you would have to run separate least squares regressions using the standard regression menu in SPSS. From there, you can generate diagnostics for each model, including identification of influential cases, evidence of heteroskedasticity, normality, etc. I will say that if you are performing your analysis using Process and you go under the Options tab, you can click on the dropdown menu for Heteroskedasticity-consistent inference and this will give you the option of selecting an HC estimator for heteroskedasticity-consistent standard errors (that would be generated for the separate regression models making up the full mediation model.) I generally opt for HC3 or HC4; this would effectively adjust the standard errors in the separate regression models for heteroskedasticity, making the t-test results more trustworthy. The nice thing about these estimators is that if there is no heteroskedasticity, there there's no adjustment to the standard errors. If there is heteroskedasticity, then the adjustments are made. I hope this is helpful to you!
Hi Anup, it supports categorical IV's (binary, ordered categorical, and multinomial) and binary DV's (the latter relying on binary logistic regression). It is not useful with categorical mediators, however. Bummer! Hope this helps!
Hi Nasir. No, I don't have a video of that. However, if you run the model twice - one with the first IV as X and the second as a Covariate (place in covariates box), and one reversing the placement of the two IV's as X and covariate - then you can obtain tests of indirect effects for each IV on the DV. You'll notice if you do this that the regression parameters (i.e., regression coefficients, multiple R, and R-square) are the same in both analyses, and that is because the prediction of M from X is controlling for the covariate; and the prediction of Y from X and M also controls for the covariate. So just reversing the placement of the IV's as X and mediator across two separate analyses will get you calculated indirect effects (which technically you could compute directly from the coefficients in the regression models) AND the bootstrap test results for the indirect effects for each IV. Hope this helps!
Mike Crowson Nice video. Just want to ask. How I want to run process with 3 IVs, 1 mediating and 1 DV? Should I run three different regression separately?
@@rabiatuladawiyah4440 If you are asking about generalizing from Nasir's situation to the case with three IV's, then you could just run the three models separately (substituting variables into the covariates and IV box in the same way I described above). You could then obtain the indirect effects and tests in the same way to get the specific indirect effects.
Mike Crowson yes..i already done that. I also checking dan compare the unstandardized value for each IVs, the result when using spss n process macro are the same. Thank you so much for your explainations
Hi there, I don't know why the link went dead. Nothing's actually changed and was working previously. Anyway, I've gone back and "refreshed" the links so they should work now. Cheers!
@@erinlobrien1 Hi Erin, my apologies. This really doesn't make much sense to me. But I think they should be working now. Please let me know if you still have problems accessing the files. Cheers.
Hello. One option for testing mediation models with multiple x's using the dropdown menu is to add them in as covariates. What will happen is that each regression model that is generated will include the covariates as well as X (in the case of the first regression predicting M) and M (in the second equation predicting Y). You won't get the tests of indirect effects for the covariates. However, if re-run your analysis substituting a covariate for X and add X as covariate, then the regression equations should be the same in terms of the parameter estimates, but you can obtain the bootstrap confidence interval for the indirect effect as previously obtained with X. Do this over and over and you should be able to obtain indirect effects and tests for each 'X' in the model. I believe there may be another way using syntax with covariates. You can find this discussion in the Appendix in Hayes book. Hope this helps!
Hi there, the link and data are under the video description. Here's the link for the Powerpoint you'll find there: drive.google.com/file/d/1zTs36CfEEApkZTONjIjWtlOT9tbj0_KR/ best wishes.