In order to perform SEM or PLS SEM one needs to do CFA first and then mediation analysis, whereas in Barron & Kenny, there is no CFA involved, why?
Hello Pathak,
Barron & Kenny's seminal article (1986) did not make a distinction of manifest vs. latent variables, so all of their examples presume one is working with manifest or observed variables.
If your data set involves latent variables, and the purported relationships are among the latent variables, then you'll either have to generate a measurement model (e.g., via EFA), impose a measurement model (e.g., via CFA/specified model constraints), or generate and save factor scores before evaluating hypothesized effects such as moderation or mediation. (Note that the latter method will treat the latent variable scores as if they are error-free, which isn't very realistic in my experience.)
This isn't a restriction of PLS-SEM or any other SEM (whether covariance based or not) program, it's just the consequence of shifting your attention from manifest/observed variables to latent variables.
Good luck with your work.
Barron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social-psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182. doi:10.1037/0022-3514.51.6.1173
Mediation analysis is a statistical technique that examines how a third variable, called a mediator, influences the relationship between an independent variable and a dependent variable. Barron & Kenny approach and PLS SEM mediation analysis are two methods for conducting mediation analysis, but they have some differences. Barron & Kenny approach is based on multiple regression models and requires several assumptions, such as linearity, normality, and homoscedasticity. PLS SEM mediation analysis is based on structural equation modeling and does not require these assumptions. PLS SEM mediation analysis can also handle multiple mediators, complex causal structures, and latent variables, while Barron & Kenny approach is limited to simple mediation models with observed variables.