Process Macro (in SPSS) by default uses bootstrapping, but in SEM analysis, I have not used bootstrapping. How do I justify using bootstrapping only to test moderation effects?
You will find useful information to answer your question by following these links:
- Hayes et al., "The analysis of mechanisms and their contingencies: PROCESS versus structural equation modeling", 2017 - https://fr.scribd.com/document/443977329/PROCESS-vs-SEM
- Igartua et al., "Mediation, Moderation, and Conditional Process Analysis: Concepts, Computations, and Some Common Confusions", 2021 - https://diarium.usal.es/jigartua/files/2012/07/Igartua-Hayes-TSJP-2021-Mediation-Moderation-Conditional-Process-Analysis.pdf
- Hayes, Andrew F. (2013). Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach. New York, NY: The Guilford Press - https://www.academia.edu/10584768/Sample_Chapter_Introduction_to_Mediation_Moderation_and_Conditional_Process_Analysis_A_Regression_Based_Approach
The first paper cites the second one. Excerpt from the conclusion of the first paper:
"[...] The greater flexibility of SEM, both in terms of model specification and handling missing data, as well as its ability to account for random measurement error when estimating relevant effects involving latent variables all make it an attractive choice. But that comes at the price of greater effort and programming skill required to calculate relevant statistics and methods of inference that PROCESS does automatically and painlessly. For models that are based entirely on observed variables, investigators can rest assured that it generally makes no difference which is used, as the results will be substantively identical. The choice, in that case, is inconsequential."
Rainer Duesing I will do that if there’s an additional advantage to using bootstrapping in SEM? Can you please explain what are the additional benefits to be gained by using bootstrapping in SEM? Thanks.
Assuming that your sample represents the population well and is large enough:
1) Bootstrapping does not assume any distribution of the error and is therefore more robust and widely applicable
2) Bootstrap Confidence Intervals may not be symmetrical, as they are with t or z distribution constructed intervals and may therefore show a more realistic distribution of your parameter of interest
3) if all assumptions are met (and I do not know if you checked that), standard CIs and bootstrap CIs should come to very similar conclusions, therefore, there is no harm in using bootstrap CIs
4) last but not least: you would be more consistent with your analyses strategy. As a reviewer, it would be suspicious to me when you change your methods (and why didn't you do the moderation in SEM in the first place but switched to PROCESS is another question I have. Since PROCESS allows only manifest variables, this should be not an issue in any SEM program [interaction of latent variables is a whole other topic....])
P.S.: I do not know your SEM model, but I assume some form of mediation will be included. If this is true, bootrapping is recommended for the indirect effect, since it can be shown that the standard error of the product term is skewed, which makes it more suitable to a bootstrap appoach, instead of normal theory approach.
Rainer Duesing No, I don’t have a mediator in my model, only moderators.
So, then I am puzzled for what did you use SEM? Can you show your model?
I used SEM for covariance-based hypotheses testing of latent constructs having observed variables. Structural equation analysis was used for main hypotheses testing and Process Macro was used for moderation analysis. Hope this helps. Thanks!
No this does not help. So you have latent variables which only have direct paths between each other (or covariances). And you have only manifest variables for your moderation? What is the difference between "main hypotheses" and moderation (where the latter can also be a main hypothesis...)? How are the latent variables related to the moderation part? Why didn't you use the SEM for your moderation, which would reduce to a simple path analysis with an interaction term? It would be helpful to see the actual model.
- Badges
- Science topic