When to consider Model Fit values in PLS?

[HansDampf3000]

Hello,

I am a little bit confused when it comes to Model Fit values. On the one hand, the documentation https://www.smartpls.com/documentation/fit "warns" to use these measures; on the other hand, "Using PLS Path Modeling in New Technology Research: Updated Guidelines by Jörg Henseler · Geoffrey Hubona · Pauline Ash Ray" says that these measures should be considered.
I used an existing motivation framework with 3 exogenous formative constructs (5-6 indicators) and a single endogenous reflective construct (2 indicators).
The question was: how do the formative indicators/the constructs influence the behavior expressed by the reflective construct.

The values for the Measurement Models and Structural Model are all okay; however, do I have to consider the model fit measures? My SRMR is 0,083 - is this value okay or do i (if one should consider these model fit values at all) have to look at all the other values, too?

I am looking forward to hearing from you!

Kind regards,

HansDampf3000

[jmbecker]

The problem with these model fit values in PLS is that there is only little research on their efficacy. In a CB-SEM context, a SRMR of <0.08 would be considered as good, but there is little evidence whether this threshold also applies to PLS.
Generally, I think fit values could be useful to detect some modelling problems, but a general problem of fit measures (also in CB-SEM) is that people tend to take them too serious. Especially, if the values deviate only marginally from the recommended thresholds.
In addition, these fit values show deviations of your model from an idealized theoretical model. Whether this idealized theoretical model is useful in practice does often not matter. In addition, “misfit” does not necessarily mean that the model is wrong. Jöreskog (the inventor of CB-SEM) has stated that model misfit “… is an indication that more information can be extracted from the data.” (1969, p. 201)
Reality is often more complex than our simple models that we estimate and there is always room to improve the model an extract more information from the data that was previously not considered. Whether this renders all previous research irrelevant is a question that everybody should answer for themselves.

[HansDampf3000]

Dear Dr Becker,

Once again I want to thank you for your fast and insightful help!
I really appreciate the help you, your colleagues and all the pros in this forum provide to newibes like me!

Kind regards

HansDampf3000