Which conditions should data meet for PLS analysis?
I believe, normality is not a requirement (even though desirable).
I have an impression that in most publications relying on PLS for model testing, the issue of whether the data fulfill any conditions at all is not addressed.
But, I find it difficult to believe that PLS works with any data.
Which conditions should data meet for PLS analysis?
Which conditions should data meet for PLS analysis?
>Which conditions should data meet for PLS analysis?
I wonder if someone might have a hint where to find this info.
Is it the case that multicollinearity could be a problem?
I wonder if someone might have a hint where to find this info.
Is it the case that multicollinearity could be a problem?
- Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi,
1) PLS-PM was named “soft modeling” because the algorithm is not based in the assumption of the normality of the distribution.
The significances are computed by bootstrap (non parametric way).
2) The PLS algorithm is based on the OLS regression for this reason:
- data should be numeric (interval, Likert, dummy)
- the problems with multicollinearity also is a problem in the structural model and in the formative measurement model (see http://www.angrad.org.br/_resources/_ci ... le_465.pdf)
Best regards,
Bido
1) PLS-PM was named “soft modeling” because the algorithm is not based in the assumption of the normality of the distribution.
The significances are computed by bootstrap (non parametric way).
2) The PLS algorithm is based on the OLS regression for this reason:
- data should be numeric (interval, Likert, dummy)
- the problems with multicollinearity also is a problem in the structural model and in the formative measurement model (see http://www.angrad.org.br/_resources/_ci ... le_465.pdf)
Best regards,
Bido
Hi,
I read many articles on PLS-PM where it is stated, that one can use different measurement scales within one model. However in the study of Hair, Joe F., Sarstedt, Marko, Ringle, Christian M. I read, that it is inappropriate to use a binary endogenous variable because for approximation in the PLS-PM algorithm coefficient are estimated by OLS, which assumes continuous data scores.
My question is: May I use a binary coded variable as my endogenous variable?
Thanks in advance for any help
(study: An assessment of the use of partial least squares structural equation modeling in marketing research,Journal of the Academy of Marketing Science, 2011)
I read many articles on PLS-PM where it is stated, that one can use different measurement scales within one model. However in the study of Hair, Joe F., Sarstedt, Marko, Ringle, Christian M. I read, that it is inappropriate to use a binary endogenous variable because for approximation in the PLS-PM algorithm coefficient are estimated by OLS, which assumes continuous data scores.
My question is: May I use a binary coded variable as my endogenous variable?
Thanks in advance for any help
(study: An assessment of the use of partial least squares structural equation modeling in marketing research,Journal of the Academy of Marketing Science, 2011)