Am trying to examine the effect from my independent variables using mediation effect in the middle (OCBs) toward the dependent variable (employee performance), as I knew before, I can use SEM approach using the statistical package (SmartPLS). Becuase some people telling me that you can't apply SEM approach for your case where other recommend for me to use SEM approach using PLS. Please find my research hypothesis and my conceptual model in the attachment.
reviewer comment:
"If you have a single mediator, then regression is fine. For more complex mediational processes, I would use SEM; since it provides more sophisticated information concerning the significance of the indirect effects, etc" !!!!!!
Dear Ahmed,
SEM is recommended for mediational analysis if you have latent constructs (measured through other observable items). SEM takes account of the measurement error by including the effects of the observable variables simultaneously.
Furthermore, regression is first generation analytical technique, while SEM is second generation technique, which is preferred and recommended tool for estimation of models.
Go ahead with SEM.
Which is best to use, Simple regression or SEM model using PLS?
Agreed with the answer shared by Wali. But SEM have 2 types i.e. Variance-based SEM or PLS-SEM like SmartPLS, PLS Graph, VisualPLS etc & Covariance-based SEM like AMOS, LISREL etc. What is your rationale to use Variance-based SEM / PLS-SEM?
In case you need to know the differences, you can refer to this RG link:
https://www.researchgate.net/post/Can_we_use_structural_equation_modelling_SEM_in_an_exploratory_research
I agree that SEM makes most sense if you have multiple indicators for your variables. Otherwise, use simple regression if all you want to do is assessi the indirect effect of moderating variables. A Sobel test will tell you if an indirect effect is significant.
I thought thew Sobel test was not generally recommended as it has very poor statistical power (e.g., see link).
http://davidakenny.net/cm/mediate.htm
I recommend SEM, but not PLS. If you follow the link in Han Ping Fung's post above, and especially read the contribution by Holger Steinmetz, you will get the explanation as to why PLS is not a good idea. Your reviewer did not mention this, but I guess that he/she did not consider PLS as a viable method in this case (which make you vulnerable for additional reviewer objections if you proceed with PLS).
I understand that you may have the restriction of which software is available for you. However, the best choice would be Mplus, which have a free demo version. This may be sufficiant for your project, since you have few varables (but check yourself at www.statmodel.com). If not, Mplus is not very expensive for academic purposes.
Also, the Mplus staff has recently published a book (Muthén et al. (2016) Regression and Mediational analysis using Mplus) about regression and mediation, which is very recommendable and explains everything in understandable details, . This book also have all the examples available for download at their website (www.statmodel.com). Probably, you would find that you can modify these examples to fit your own project.
Best
Jan Ivanouw
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science
- Materials
- Inorganic Materials
- Ceramics