Hi Mike ! Thank you for this video ! Your explanation was very clear. I am also running a model 6 (PROCESS macro). However, I first wanted to test the direct effect (X --> Y) with a simple regression (it was signficant). Then I used twice the model 4 to test each mediator separately. My direct effect became negative in only one of the two model 4. Finally, when I put all together in model 6, the direct effect (X --> Y) was insignficant. Is there a reasons why this direct effect that was previously signficant became insignficant?
Hi Mike. I need your guidance in analyzing my research model. My model has 4 IVS, 2 Mediators and 1 DV. I want to test (i) direct impact of 4 IVs on the DV (ii) indirect effect of 4 IVs on the DV through M1 (iii) indirect effect of 4 IVs on the DV through M2 (iv) sequential mediation effect for 4 IVs on DV through M1 and M2 Firstly, I want to know which approach would be best to test this model. Should i run all at the same time or analyze every relationship separately by drawing all models separately Secondly, 5 variables in my model have few dimensions and 2 variables dont have dimensions. So in this case should I go for first order cfa/sem or second order cfa/sem?? Waiting eagerly for your kind response
Hi Mike, Thanks for this great video. It's the first I have found that demonstrates how to interpret PROCESS Model 6 output, and your explanation is very clear. I would be grateful for the answer to a question: If one of the indirect paths is statistically significant (95% CI does not include zero), but the total indirect effect is not statistically significant, how do you interpret that? Thank you!
Hi there. Thanks for your post, and interesting question. The only way I could see this happening is if you have some indirect effects being negative and others being positive, and when they are summed to the total indirect effect it sums to zero (or nearly so). (Logically, this is the same thing that could happen when you are testing the total effect of X on Y where you have a single mediator. If the direct effect is one sign and the indirect effect is another sign, it's possible for the total effect can sum to zero.) I would focus my discussion on the specific indirect effects since this is what you are most interested in to begin with. Hope this helps.
Hello, if I have 4/5 independent variables and all which will be mediated by M then how I will do the analysis in SPSS? As one time I can put 1 independent variable only. Will I have to do separately? Additionally, in my model dummy and control varibles will also be included as covariates. So if I have to do separate analysis then which value of covariates will have to be reported? I am using process v3.
Hi Eric, it sounds like your model is a fairly complex mediation model. My strong suggestion is to use an SEM program like AMOS, LISREL, MPLUS... I don't think there's a really easy way of accomplishing what you want to using Process (although I wouldn't say it's necessarily impossible; but require some series gymnastics). The nice thing about using an SEM program is that in addition to testing indirect effects, you can also test for overall model fit.
Hi, mike, thanks for this video! but also could i ask which model would i use if im testing sequential mediation with moderators, looking at the appendices of models. I am unable to find one that is appropriate. if there isn't any, what is my alternative route? I have 1 IV, 5 mediator, and 1 DV.
Hi Mike, so I do not have to account for multicollinearity of anxiety and interest by standardising them? - In other words, does the model 6 already account for that when assessing the effect of mastery, anxiety, and interest on achieve?
Hi there. Thanks for visiting! I'm unclear what you mean by 'accounting for multicollinearity by standardizing them.' The full model that is being tested here is constructed from three separate regression equations - one per endogenous variable in the model. Eq. 1: mastery-->anxiety; Eq. 2: mastery+anxiety-->-->interest; Eq. 3: mastery+anxiety+interest-->achieve. Equation 1: is a simple linear regression that reflects the zero-order relationship between mastery on anxiety. Equation 2 is a multiple linear regression with mastery and anxiety entered as separate predictors. The slope for mastery is controlling for anxiety, and the slope for anxiety controls for mastery. Equation 3 is a multiple linear regression with the mastery, anxiety, and interest as predictors of achievement. The regression slope for each predictor is controlling for the effects of the other predictors in the model. All regression slopes by default are unstandardized. If you request 'standardized coefficients' under the Options tab, then you will also obtain standardized regression slopes, as well as standardized indirect effects in your output. You should check out my newest video on Model 6 here (th-cam.com/video/vFGSa0KfDYA/w-d-xo.html). You can also download a powerpoint as well. Cheers!
May you prepare a video which shows testing this model in AMOS? I'm running a similar model with different variables and in the output of AMOS I cannot find the third indirect effect (the path from mastery to anxiety to interest to achieve). Can you explain here, how can I find this value?
Dr. Crowson, I missed that video. Thank you very much for your response, I finally found the answers for my questions. You are doing a great work here. Thanks a lot.
Sure, it could suggest the possibility of other possible mediators between X and Y; on the other hand, the effect of X on Y could be both direct and indirect. You have to consult your theory to help you out with your model specification. If you can come up with other plausible mediators that might be included in the model between X and Y, then you can test these out using a multiple mediator model.
DEAR VIEWERS: I HAVE RECENTLY UPLOADED A NEWER AND BETTER QUALITY VIDEO COVERING THESE TOPICS ON PROCESS. THERE ARE ADDITIONAL LEARNING RESOURCES AND THE DATA SET PROVIDED AS LINKS UNDERNEATH THE VIDEO DESCRIPTION. GO HERE: th-cam.com/video/vFGSa0KfDYA/w-d-xo.html . I ALSO HAVE UPLOADED ANOTHER VIDEO FOR TESTING MEDIATION USING PROCESS WHEN YOU HAVE A BINARY OUTCOME HERE: th-cam.com/video/vW69r24hPs4/w-d-xo.html
Super helpful! Thank you so much!
Thank you so much!
You're welcome!
Hi Mike ! Thank you for this video ! Your explanation was very clear. I am also running a model 6 (PROCESS macro). However, I first wanted to test the direct effect (X --> Y) with a simple regression (it was signficant). Then I used twice the model 4 to test each mediator separately. My direct effect became negative in only one of the two model 4. Finally, when I put all together in model 6, the direct effect (X --> Y) was insignficant. Is there a reasons why this direct effect that was previously signficant became insignficant?
Hello Lea, I have the same problem as you with data. Have you found a solution?
what assumptions should you check before performing this analysis?
Thanks
8:26 12:02 take a note thanks sir
Hi Mike. I need your guidance in analyzing my research model. My model has 4 IVS, 2 Mediators and 1 DV. I want to test
(i) direct impact of 4 IVs on the DV
(ii) indirect effect of 4 IVs on the DV through M1
(iii) indirect effect of 4 IVs on the DV through M2
(iv) sequential mediation effect for 4 IVs on DV through M1 and M2
Firstly, I want to know which approach would be best to test this model. Should i run all at the same time or analyze every relationship separately by drawing all models separately
Secondly, 5 variables in my model have few dimensions and 2 variables dont have dimensions. So in this case should I go for first order cfa/sem or second order cfa/sem??
Waiting eagerly for your kind response
Hi Mike, Thanks for this great video. It's the first I have found that demonstrates how to interpret PROCESS Model 6 output, and your explanation is very clear. I would be grateful for the answer to a question: If one of the indirect paths is statistically significant (95% CI does not include zero), but the total indirect effect is not statistically significant, how do you interpret that? Thank you!
Hi there. Thanks for your post, and interesting question. The only way I could see this happening is if you have some indirect effects being negative and others being positive, and when they are summed to the total indirect effect it sums to zero (or nearly so). (Logically, this is the same thing that could happen when you are testing the total effect of X on Y where you have a single mediator. If the direct effect is one sign and the indirect effect is another sign, it's possible for the total effect can sum to zero.) I would focus my discussion on the specific indirect effects since this is what you are most interested in to begin with. Hope this helps.
Many thanks, Mike. Much appreciate your prompt and clear response!
Hello, if I have 4/5 independent variables and all which will be mediated by M then how I will do the analysis in SPSS? As one time I can put 1 independent variable only. Will I have to do separately? Additionally, in my model dummy and control varibles will also be included as covariates. So if I have to do separate analysis then which value of covariates will have to be reported? I am using process v3.
Hi Eric, it sounds like your model is a fairly complex mediation model. My strong suggestion is to use an SEM program like AMOS, LISREL, MPLUS... I don't think there's a really easy way of accomplishing what you want to using Process (although I wouldn't say it's necessarily impossible; but require some series gymnastics). The nice thing about using an SEM program is that in addition to testing indirect effects, you can also test for overall model fit.
Hi, mike, thanks for this video! but also could i ask which model would i use if im testing sequential mediation with moderators, looking at the appendices of models. I am unable to find one that is appropriate. if there isn't any, what is my alternative route?
I have 1 IV, 5 mediator, and 1 DV.
any good paper to know how to report the results?
Hi Mike, so I do not have to account for multicollinearity of anxiety and interest by standardising them? - In other words, does the model 6 already account for that when assessing the effect of mastery, anxiety, and interest on achieve?
Hi there. Thanks for visiting! I'm unclear what you mean by 'accounting for multicollinearity by standardizing them.' The full model that is being tested here is constructed from three separate regression equations - one per endogenous variable in the model. Eq. 1: mastery-->anxiety; Eq. 2: mastery+anxiety-->-->interest; Eq. 3: mastery+anxiety+interest-->achieve. Equation 1: is a simple linear regression that reflects the zero-order relationship between mastery on anxiety. Equation 2 is a multiple linear regression with mastery and anxiety entered as separate predictors. The slope for mastery is controlling for anxiety, and the slope for anxiety controls for mastery. Equation 3 is a multiple linear regression with the mastery, anxiety, and interest as predictors of achievement. The regression slope for each predictor is controlling for the effects of the other predictors in the model. All regression slopes by default are unstandardized. If you request 'standardized coefficients' under the Options tab, then you will also obtain standardized regression slopes, as well as standardized indirect effects in your output. You should check out my newest video on Model 6 here (th-cam.com/video/vFGSa0KfDYA/w-d-xo.html). You can also download a powerpoint as well. Cheers!
May you prepare a video which shows testing this model in AMOS? I'm running a similar model with different variables and in the output of AMOS I cannot find the third indirect effect (the path from mastery to anxiety to interest to achieve). Can you explain here, how can I find this value?
Hi there. Try: th-cam.com/video/9jGL45NuVAA/w-d-xo.html
Dr. Crowson, I missed that video. Thank you very much for your response, I finally found the answers for my questions. You are doing a great work here. Thanks a lot.
Can I ad a moderator in model 6? If so, how do I justify this decision/should I justify this decision?
Mike, if one of the indirect paths is significant, but the direct effect also remains significant, does this suggest a missing mediator? Thank you!
Sure, it could suggest the possibility of other possible mediators between X and Y; on the other hand, the effect of X on Y could be both direct and indirect. You have to consult your theory to help you out with your model specification. If you can come up with other plausible mediators that might be included in the model between X and Y, then you can test these out using a multiple mediator model.
Many thanks, Mike. Much appreciated!
DEAR VIEWERS: I HAVE RECENTLY UPLOADED A NEWER AND BETTER QUALITY VIDEO COVERING THESE TOPICS ON PROCESS. THERE ARE ADDITIONAL LEARNING RESOURCES AND THE DATA SET PROVIDED AS LINKS UNDERNEATH THE VIDEO DESCRIPTION. GO HERE: th-cam.com/video/vFGSa0KfDYA/w-d-xo.html . I ALSO HAVE UPLOADED ANOTHER VIDEO FOR TESTING MEDIATION USING PROCESS WHEN YOU HAVE A BINARY OUTCOME HERE: th-cam.com/video/vW69r24hPs4/w-d-xo.html