In my study,I use the reflective indicators. Now I want to construct the Important-Performance analysis map(as to a latent variable). If I use outer weights of reflective indicators as "Important" and scores of reflective indicators as"performance".Is that OK?
thanks a million !
Important-Performance analysis
-
Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi,
The “importance-performance” map or “priority map” is a interpretable tool if the variables used are independents, even when we think that the predictors are independent, sometimes the multicollinearity is a problem and the priority map will be different from one sample to other. One beta will increase or decrease, but the only explanation is the multicollinearity, it is not an “importance” matter.
Cohen et al. (2003, 421) give us an example when there is suppression (beta with contrary signal of the correlation) even where the VIF = 5,26.
Then if multicollinearity is a issue when the predictors are “independent”, the things will be worst when we expect that the indicators will be correlated (reflective).
I think that you should not use this approach.
COHEN, J.; COHEN, P.; WEST, S. G.; AIKEN, L. S. Applied multiple regression/correlation analysis for the behavioral sciences. 3rd ed. New Jersey: Lawrence Erlbaum Associates, Publishers, 2003.
Best regards,
Bido
The “importance-performance” map or “priority map” is a interpretable tool if the variables used are independents, even when we think that the predictors are independent, sometimes the multicollinearity is a problem and the priority map will be different from one sample to other. One beta will increase or decrease, but the only explanation is the multicollinearity, it is not an “importance” matter.
Cohen et al. (2003, 421) give us an example when there is suppression (beta with contrary signal of the correlation) even where the VIF = 5,26.
Then if multicollinearity is a issue when the predictors are “independent”, the things will be worst when we expect that the indicators will be correlated (reflective).
I think that you should not use this approach.
COHEN, J.; COHEN, P.; WEST, S. G.; AIKEN, L. S. Applied multiple regression/correlation analysis for the behavioral sciences. 3rd ed. New Jersey: Lawrence Erlbaum Associates, Publishers, 2003.
Best regards,
Bido
-
Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
If you use the scores from the factor analysis (computed with orthogonal rotation, like Varimax) in the SmartPLS, the LVs will be independents (no multicollinearity), but if you use the indicators in the SmartPLS, we will have some correlation between LVs (see the SmartPLS report).
But this is different from what you have said before. Here we are talking about mapping with path coefficients (LVs instead indicators).
Formative indicators with insignificant weights is common when we have multicollinearity between them.
Best regards,
Bido
But this is different from what you have said before. Here we are talking about mapping with path coefficients (LVs instead indicators).
Formative indicators with insignificant weights is common when we have multicollinearity between them.
Best regards,
Bido
Thanks for your help,dear Bido!
Now I want to ask a question about formative indicators!
For one LV , I use the factor analysis to make its formative indicators be independent ( no multicollinearity,VIF =1 or Condition Index=1).
Then I put the the scores from the factor analysis (computed with orthogonal rotation, like Varimax) in the SmartPLS to calculate the overall Path model. Is that OK?
Thanks a million!
Now I want to ask a question about formative indicators!
For one LV , I use the factor analysis to make its formative indicators be independent ( no multicollinearity,VIF =1 or Condition Index=1).
Then I put the the scores from the factor analysis (computed with orthogonal rotation, like Varimax) in the SmartPLS to calculate the overall Path model. Is that OK?
Thanks a million!
-
Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
It looks that you got the idea,
Just clarifying the words:
- indicators = variables in your dataset
- when we run the factor analysis, the indicators will be grouped in factors (like reflective LV), not “formative indicators”.
- These factors will have independent scores when we saved them after a Varimax rotation.
- Now we could import the scores to SmartPLS and used them as formative indicator of a LV or predictors of a LV, in the both cases without multicollinearity.
Best regards,
Bido
Just clarifying the words:
- indicators = variables in your dataset
- when we run the factor analysis, the indicators will be grouped in factors (like reflective LV), not “formative indicators”.
- These factors will have independent scores when we saved them after a Varimax rotation.
- Now we could import the scores to SmartPLS and used them as formative indicator of a LV or predictors of a LV, in the both cases without multicollinearity.
Best regards,
Bido
-
Diogenes
- PLS Super-Expert
- Posts: 899
- Joined: Sat Oct 15, 2005 5:13 pm
- Real name and title:
- Location: São Paulo - BRAZIL
- Contact:
Hi,
I got this idea from the principal components regression (Cohen et al., 2003, p.428-429), that it is in a multiple regression context, it is not PLS-PM.
A colleague of mine had used this approach in your thesis (in Portuguese):
http://www.teses.usp.br/teses/disponive ... /pt-br.php
I have not others references.
COHEN, J.; COHEN, P.; WEST, S. G.; AIKEN, L. S. Applied multiple regression/correlation analysis for the behavioral sciences. 3rd ed. New Jersey: Lawrence Erlbaum Associates, Publishers, 2003.
Best regards,
Bido
I got this idea from the principal components regression (Cohen et al., 2003, p.428-429), that it is in a multiple regression context, it is not PLS-PM.
A colleague of mine had used this approach in your thesis (in Portuguese):
http://www.teses.usp.br/teses/disponive ... /pt-br.php
I have not others references.
COHEN, J.; COHEN, P.; WEST, S. G.; AIKEN, L. S. Applied multiple regression/correlation analysis for the behavioral sciences. 3rd ed. New Jersey: Lawrence Erlbaum Associates, Publishers, 2003.
Best regards,
Bido
Thanks a million,Dear Bido!
You gave me great help, and I have read the paper of your colleague written . I think its very good even though some of words I couldn't understand.It gives some confidence to do like my said previously.
Thanks ! I sincerely thank you from the bottom of my heart!
Best regards,
Frank wang
You gave me great help, and I have read the paper of your colleague written . I think its very good even though some of words I couldn't understand.It gives some confidence to do like my said previously.
Thanks ! I sincerely thank you from the bottom of my heart!
Best regards,
Frank wang