forum.smartpls.com

Hi everyone,
I would like to use the rho_A coefficient as defined in Dijkstra and Henseler(2015) and implemented in PLS3.2.
In a small model with 4 reflexive constructs, I have both rho_A and sqrt(rho_A) in the tables as implemented by PLS3.2. The above reference does not give an idea of the threshold under which a rho_A indicates that a LV is not reliable. And what use does sqrt(Rho_A) have?

One key LV which is causing me headaches has 8 items, a Cronbach alpha of .705, a composite reliability of 0=.403 and a rho_A of .647. How do I analyze such results? Pruning the items only makes things worse (at least regarding the alpha).

All path coefficients leading from it are t-statistics <1.96, meaning that it does not have any effect on the other LV.
BTW, all the other LV have nice and healthy reliability and discriminating coefficients.

Any help appreciated.
Thanks.

Hi Xavier,

That I remember the cut-off value of coefficient Dijkstra-Henseler Rho_A must be > 0.7 as Composite Reliability.
If you do not want to remove the item, you may have to give a little justification there.

You also need to see the value GoF with SRMR and NFI, I think you are a bit of a bad fit model.

Best regards,

Hi Latan,
can you provide a reference that explains how to calculate the NFI for PLS models?

Hi Latan,
Thank you for the reply. I have looked at the Dijkstra Henseler reference and they don't mention a cut-off value for Rho_A. If I look at the SRMR given by SmartPLS3.2.3, I get a SRMR for common factor of 0.079 and SRMR for the composite model of 0.053. Are these good values?

Would you mind giving a reference on how to evaluate these criteria for model fit? Also, you speak of NFI. I thought that NFI was purely for SEM-ML models, ie, LISREL AMOS type models.

Can you please substantiate how such an index can be calculated in PLS?

Thank you for your time and patience.

Regards.
Xavier

Hi everyone,

Does anybody know what a Rho_A value above 1 means and how to interpret it?

Best regards,

Lisa

A rho_A value above 1 is strange and should not occur. You may revert to Cronbach's alpha in that case.

It may also indicate that you do not have a valid factor model. A composite model might be better in your case. You should re-consider and re-evaluate your measurement approach.

Dear,
what if rho_a (for all indicators) is higher than CR?
Cronbach's are for all indicators are smaller than rho_A.