forum.smartpls.com

Hello everybody! I am currently using the PLS program. Unfortunately, I now have a big problem at structure model level. I hope someone can help me.

The constructs as well as the model have passed all quality tests. Unfortunately I noticed now that a construct has negative influence on model plane (original sample). This would mean in concrete terms that the product knowledge of the employees has negative impact on the company's success. This makes no sense at all. If I only look at the construct (product knowledge), it is highly significant and positive. Multicollinearity is neither present on construct nor model level. What can be the reason for such a result? Can you please help me? I am very greatful for any advice.

Best regards!

If the correlation between the latent variables is positive, but the path coefficient negative, then it is likely a collinearity issue on the model level.

In general, I would always check …
- if the direction of the construct is correct (if you have negative weights you may also have a reversed orientation of the latent variable scores).
- if you have missing values in your dataset and if they are correctly treated.
- if you have extreme responses (outliers).

jmbecker wrote:If the correlation between the latent variables is positive, but the path coefficient negative, then it is likely a collinearity issue on the model level.

In general, I would always check …
- if the direction of the construct is correct (if you have negative weights you may also have a reversed orientation of the latent variable scores).
- if you have missing values in your dataset and if they are correctly treated.
- if you have extreme responses (outliers).

Dear Dr. Jan-Michael Becker,

In my study, I have one negative path coefficient due to outliers. I already used FIMIX-PLS to find two subgroups in my data.
However. is there any citation that helps me explain the negative path coefficient caused by outliers?

Thank you very much Dr.

I'm not sure if there is a PLS reference, but there should be quite many standard statistical texts that discuss the effect of outliers on regression or correlation coefficients, which would also apply to a PLS estimation.