Hair et al., (2019) state:

“When testing statistical models with latent variables, researchers frequently use regression analysis together with Preacher and Hayes’ (2008) PROCESS method (Hayes, 2018)................. However, PROCESS only allows for the sequential testing of model parts without taking the entire model structure into account. PLS-SEM, on the other hand, considers the entire theoretical structural model in the estimation process, making it superior to regression analysis.

Even more critical, however, is the practice of averaging several indicators that together comprise a measurement model, and using them as input in a PROCESS regression model. PROCESS does not account for measurement error inherent in the items and thus introduces bias in the results, whereas PLS-SEM removes the measurement error and reduces the bias...........................Therefore, researchers should use PLS-SEM to estimate mediating effects among latent variables rather than regression analysis with the PROCESS macro.”

Joseph F. Hair, Marko Sarstedt, Christian M. Ringle, (2019) "Rethinking some of the rethinking of partial least squares", European Journal of Marketing, https://doi.org/10.1108/EJM-10-2018-0665

Methods from spss and amos are different. I suggest you to testing the relationships between constructs using amos since seems you has complex modeling (mediator and moderator). So, you need to study about structural equation modeling. Please bear in mind, SEM has two types which are pls-sem (exploratory) and cbsem (confirmatory). You need to determine the method based on your research design

Thank you Derya Doğan and Asyraf Afthanorhan.

It will be helpful.

@Derya: I have to disagree, PLS-SEM does not account for measurement error per se as it builds linear combinations of indicators to obtain construct scores. So the results are similarly biased as if you would use sum scores and subsequently regression to estimate the structural model. To account for the attenuation bias introduced by measurement errrors, you have to use PLSc in the context of PLS-SEM.

Moreover, it is currently not clear which model PLS-SEM actually estimates if mode A is used, since such a model is not devloped yet. Currently only mode B consistently estimates the composite model, see Dijkstra (2017).

Best regards,

Florian

Best regards,

Florian

The use of PLS-SEM allows simultaneous testing of connexions among variables. Hence, using PLS-SEM, the research hypotheses are simultaneously tested. Also, as opposed to other multivariate techniques, PLS has no reliance upon the normality assumption (distribution-free). This is owes to PLS being non-parametric in nature. Still, for assessment of local model, there should be consideration on assumptions including multicollinearity and goodness of fit indices.

I agree with @Florian Schuberth in regarding to account for the attenuation bias introduced by measurement errors, you have to use PLSc to confirm your results.