With PROCESS you can run a moderated analysis with a 2 x 2 between-subjects design, and even it is possible to add another moderating variable, as Z variable. This is model 3.
Although Process does not present strong model fit indices, a few things can be done to sort of guard against biases and spurious results. For example, using the Cribari-Neto heteroskedasticity consistent standard error and covariance matrix estimator (HC4) can help reduce biases due to violation of the homogeneity assumption. Bootstrapping can also help with issues of skewness. Using the natural log transformation of the data also helps with skewness issues. With moderation, checking mean-centering of the products also helps control multicollinearity. You also get the RMSE from the MSE which can be helpful when compared with the data range.
Very good point. There are additional tests to help against bias. You need to take extra steps to perform those using PROCESS. As for skewness, bootstrapping helps and if you know the data is skewed you could even transform the data. As for mean centering, Andrew Hayes and I had a conversation over dinner one night about it. He believes that there is no reason to mean center data with moderation. I am still in the other boat that mean centering will help with multicollinearity issues. He has a mean centering option in the macro but in his latest version of his book he says mean centering is unnecessary. Andrew and I respectfully disagree on this point.
@@joelcollier9387 Thanks for the response, and for presenting Andrew Hayes' position on mean centering. I've personally experimented with mean centering in SEM, and I lean more toward your position. But then maybe the observations were due to my data set and would not probably apply in other data. Yes, I'd agree that, at least, a comprehensive common method bias test would have to be performed and addressed for before progressing to PROCESS. Thank you so much.
Hi! Thank you for this, it's very helpful! For a moderated mediation model, I've been told Process is a better fit when the moderator is a multilevel categorical variable. SEM is better suited for continuous moderator. Is this true?
Thank you so much for this informative video! You mention that in PROCESS we can use relatively small samples. Would you mind explaining how we would run a power analysis for a moderated mediation model (e.g., Hayes' model 21)? I've read that we would need to run Monte Carlo simulations to get our needed sample size; however, I'm having some trouble with finding model 21-specific syntax to do this. Any advice would be much appreciated!
To be honest, I have never run a monte carlo simulation to determine power in PROCESS. My method (for good or for bad) is I go to a power calculator and it determines my sample size to determine the desired power. If it is an experiment, I shoot for a cell size of each treatment of 50. That is more than enough power to find the effect. Hope this helps.
@@matthewjohnson705 Here is a quick and easy reference if you are looking to see if you have enough power based on your sample and degrees of freedom in your model: www.sciencedirect.com/science/article/abs/pii/S0148296301003010
There are no preset models in PROCESS to perform analysis with higher order constructs. Especially if it is formative, you are much better off going the SEM route.
Can we apply the principles of PROCESS for constructing structural equation models (SEMs) in simple models that involve one independent variable (IV) and one dependent variable (DV) in the context of moderated mediation models?
Yes, you can. With one DV the analysis will be exactly the same. PROCESS is quicker to use because of the preformed models. If the preformed models do not line up with your model, you are better off using SEM
thank you so much for your video explanation. It was very helpful. Could you please give reference of some papers that talk about advantages of process?
Andrew Hayes's book is your best reference for the advantages of using Process: Hayes, A. F. (2018). Introduction to mediation, moderation, and conditional process analysis. (2nd Ed.). New York: The Guilford Press. There is another article titled "The analysis of mechanisms and their contingencies: Process vs Structural equation modeling. In the Australian Marketing Journal Vol 25 (1) 76-81 by Hayes, Montoyoa, and Rockwood
@@joelcollier9387 Hayes is an excellent reference. By the way, your succinct analysis of Process versus SEM was well-done! Clear and practical. I enjoyed it!
Yes and No....it depends on the software you are using. AMOS/Lisreal/EQS will not let you use dichotomous DVs; PLS will let you use dichotomous dvs but it uses a work around in the analysis....and PLS does not use the covariance matrix as input which has numerous problems.
With PROCESS you can run a moderated analysis with a 2 x 2 between-subjects design, and even it is possible to add another moderating variable, as Z variable. This is model 3.
Although Process does not present strong model fit indices, a few things can be done to sort of guard against biases and spurious results. For example, using the Cribari-Neto heteroskedasticity consistent standard error and covariance matrix estimator (HC4) can help reduce biases due to violation of the homogeneity assumption. Bootstrapping can also help with issues of skewness. Using the natural log transformation of the data also helps with skewness issues. With moderation, checking mean-centering of the products also helps control multicollinearity. You also get the RMSE from the MSE which can be helpful when compared with the data range.
Very good point. There are additional tests to help against bias. You need to take extra steps to perform those using PROCESS. As for skewness, bootstrapping helps and if you know the data is skewed you could even transform the data. As for mean centering, Andrew Hayes and I had a conversation over dinner one night about it. He believes that there is no reason to mean center data with moderation. I am still in the other boat that mean centering will help with multicollinearity issues. He has a mean centering option in the macro but in his latest version of his book he says mean centering is unnecessary. Andrew and I respectfully disagree on this point.
@@joelcollier9387 Thanks for the response, and for presenting Andrew Hayes' position on mean centering. I've personally experimented with mean centering in SEM, and I lean more toward your position. But then maybe the observations were due to my data set and would not probably apply in other data.
Yes, I'd agree that, at least, a comprehensive common method bias test would have to be performed and addressed for before progressing to PROCESS. Thank you so much.
Hi! Thank you for this, it's very helpful! For a moderated mediation model, I've been told Process is a better fit when the moderator is a multilevel categorical variable. SEM is better suited for continuous moderator. Is this true?
Thank you so much for this informative video! You mention that in PROCESS we can use relatively small samples. Would you mind explaining how we would run a power analysis for a moderated mediation model (e.g., Hayes' model 21)? I've read that we would need to run Monte Carlo simulations to get our needed sample size; however, I'm having some trouble with finding model 21-specific syntax to do this. Any advice would be much appreciated!
To be honest, I have never run a monte carlo simulation to determine power in PROCESS. My method (for good or for bad) is I go to a power calculator and it determines my sample size to determine the desired power. If it is an experiment, I shoot for a cell size of each treatment of 50. That is more than enough power to find the effect. Hope this helps.
@@joelcollier9387 Can you provide a citation of a study where you calculated power this way when using PROCESS? Thanks!
@@matthewjohnson705 Here is a quick and easy reference if you are looking to see if you have enough power based on your sample and degrees of freedom in your model: www.sciencedirect.com/science/article/abs/pii/S0148296301003010
Thankyou for the useful content! Can you please clarify if higher /second order constructs can be used in PROCESS ?
There are no preset models in PROCESS to perform analysis with higher order constructs. Especially if it is formative, you are much better off going the SEM route.
Can we apply the principles of PROCESS for constructing structural equation models (SEMs) in simple models that involve one independent variable (IV) and one dependent variable (DV) in the context of moderated mediation models?
Yes, you can. With one DV the analysis will be exactly the same. PROCESS is quicker to use because of the preformed models. If the preformed models do not line up with your model, you are better off using SEM
thank you so much for your video explanation. It was very helpful. Could you please give reference of some papers that talk about advantages of process?
Andrew Hayes's book is your best reference for the advantages of using Process:
Hayes, A. F. (2018). Introduction to mediation, moderation, and conditional process analysis. (2nd Ed.). New York: The Guilford Press.
There is another article titled "The analysis of mechanisms and their contingencies: Process vs Structural equation modeling. In the Australian Marketing Journal Vol 25 (1) 76-81 by Hayes, Montoyoa, and Rockwood
@@joelcollier9387 thank you very much! Your response was very helpful! Wish you all the best and God bless you :)
@@joelcollier9387 Hayes is an excellent reference. By the way, your succinct analysis of Process versus SEM was well-done! Clear and practical. I enjoyed it!
You can perform SEM with dichotomous DV
Yes and No....it depends on the software you are using. AMOS/Lisreal/EQS will not let you use dichotomous DVs; PLS will let you use dichotomous dvs but it uses a work around in the analysis....and PLS does not use the covariance matrix as input which has numerous problems.
You can use use R with a the lavaan and the WLS estimate