how to interpret some new indexes in Newversion of SmartPLS3

[Man18]

Hi professional and creator of smartPLS the best software ever (my point of view),
I find some modification in your last release of SmartPLS : the model fit icon : Rms theta, ULS, GLS and Chi-square i just know that srmr should be less than .1 at max but the others indices ?
i find for the saturated model : d-ULS 40.580
d_G : n/a
Chi-square : infinite
NFI n/a
rms theta is 0.158
please what does this mean ? i feel that all are not good even this is the firts time i saw them as indexes ..any references ?
thank you all

[jmbecker]

You are raising very good questions. The problem is that they are mostly unanswered in the literature.
Some researchers started to request these model fit indices for PLS. In my opinion, without a solid knowledgebase on their behavior. As researchers are requesting these features, we (SmartPLS) decided to provide them to fulfill the demand.
Yet, we believe that there is much more research necessary to apply them appropriately. Easy access to these indices via SmartPLS might foster the necessary research. Hence, we decided to implement them, even though they are not yet fully understood.
We are preparing some information for the resource section on smartpls.com in which we will try to summarize current findings and our position on using these new indices.

My personal recommendation would be: use them with care; experiment with them; get a feeling on their behavior; but don’t base too much weight on assessing these indices without further research.

One observation regarding your model: You are probably having repeated indicators in your model (e.g., a second-order repeated indicator model). In this case some of the indices are not properly defined (you will get in “infinite” or “n/a” value). Something else that needs further research.

[Man18]

You are raising very good questions. The problem is that they are mostly unanswered in the literature.
Some researchers started to request these model fit indices for PLS. In my opinion, without a solid knowledgebase on their behavior. As researchers are requesting these features, we (SmartPLS) decided to provide them to fulfill the demand.
Yet, we believe that there is much more research necessary to apply them appropriately. Easy access to these indices via SmartPLS might foster the necessary research. Hence, we decided to implement them, even though they are not yet fully understood.
We are preparing some information for the resource section on smartpls.com in which we will try to summarize current findings and our position on using these new indices.

My personal recommendation would be: use them with care; experiment with them; get a feeling on their behavior; but don’t base too much weight on assessing these indices without further research.

One observation regarding your model: You are probably having repeated indicators in your model (e.g., a second-order repeated indicator model). In this case some of the indices are not properly defined (you will get in “infinite” or “n/a” value). Something else that needs further research.

Thank you for your reply Dr,
well you guess it true I have repeated indicator since i have second order construct and using reflective-reflective model. So waiting for your new work on it. thank you all for what you do and please please let's deal more with second order model it's always tricky and no much research on it. finally just a question please about your article on method to use with second order LV you recomended to use the two steps method if the model (2nd order) is formative. well in my case it's not necessary as i understand from your article but if i use it. is it false?

[Hengkov]

I think the PLS community needs to develop its own without adopting GoF index of CB-SEM. It will make "a little confusion" on the results and interpretation.

[m.magalhaes]

SRMR model fit results have different values ...?

I work on some research together with Prof. Dr. José Luis Roldán from Sevilla University, among others. I have been working with SmartPls 3.2.0 till recently I downloaded the latest release 3.2.4.
When I tested such latest version over a certain model that had already been studied with the antecedent 3.2.0 version, I noticed some discrepancies in results gathered between the two releases. Although the regressions and other indicators belonging to constructs are equal, as it was supposed to be, the value of SRMR is different. Unless the process of calculating this indicator had been changed, something is wrong ...

I would be very obliged for a quick reply clarifying such SRMR discrepancy between the two releases, once I'm now stuck with a paper almost ready for publication, which in light of such anomaly may require some review.

Thank you

[m.magalhaes]

Besides SRMR, another problem ...

Another problem with release 3.2.4 is the changing of the number format, by default come with US format, but when one proceeds converting to the european format, it doesn't really changes anything. Such issue affects all calculations giving different results.

Please give Your best attention to such problems, asap

Thank you

[jmbecker]

Yes, we changed the calculation of the SRMR. In the former versions, we reported the SRMR for a factor and a composite model. Now, we report the SRMR for a mixed model. That means each construct is treated as either a factor (reflective) or composite (formative) based on how the measurement model is specified. You should get the same results if you have a fully reflective (factor model) or fully formative (composite model).

You can change your default number format in the preferences.