forum.smartpls.com

I am writing to get your advice regarding the PLS-MGA: PLS Multi-Group Analysis. I could not run PLS-MGA to see if there a significant difference between two groups (Failed and Non-Failed banks).
SmartPLS 3 gives me a message that there is an error occurred during the outside of estimation for latent variables. I tried to solve it but am still getting the same message. I can say that I have some indicators with a extreme collinearity, but I can not remove them because of my financial theory ( financial ratios). Multicollinearity is a major problem when building models based on financial ratios since ratios share a common numerator or denominator.
Your help will be totally appreciated
Best
Mohamed

If in one of the groups the collinearity is perfect (i.e., correlation of 1) it is just not possible to estimate the model with your specification and you may need to drop one of the indicators. Alternatively, you could also try to specify your construct as reflective (Mode A) if it is currently specified as formative (Mode B). There is a paper by Becker, Rigdon, Rai (2013) that explains why using Mode A instead of Mode B makes sense for formative constructs, especially if the indicators are highly collinear.

An alternative problem that sometimes occurs in PLS-MGA due to misspecification of the model are indicators with zero variance for one of the groups. It happens for example, if the grouping variable is part of the model and thus the variable has all entries the same for one of the groups.