Hello,
I am a new user of PLS Equations Modelling. I have an existing model :
* 1 second order formative variable "COCREAT F." made of 2 first order formative variables "COPROD F" / "VALUE IN USE F."
* each of the first or second order formative variable is reflected in 3 reflective indicators COCREAT R. / VALUE IN USE R. / COPRODU R.)
* 3 reflective dependant variables (Valeur clients, Perf., Satisfact.)
I would like to calculate direct vs indirect effects, and for this I try to add direct links between first order formative variables and the 3 dependent variables.
The problem is : when I am adding those links, this leads to some coefficients becoming superior to 1, for instance 1.356 or 1.125. This is not normal, as path coefficients should be standardized figures.
Can you help me ?
COCREAT F. COCREAT. R. COPROD F. COPROD R. PERF. SATISFACT. VALUE IN USE F. VALUE IN USE R. VALEUR CLIENTS
COCREAT F. 0.293 0.710 1.376 1.376
COCREAT R.
COPROD F. 0.193 0.447 0.234 0.099 0.091
COPROD R.
PERF
SATISFACT.
VALUE IN USE F. 0.871 0.060 0.591 0.268 0.596
VALUE IN USE R.
Valeur clients
Path coefficients > 1 HELP

 SmartPLS Developer
 Posts: 879
 Joined: Tue Mar 28, 2006 11:09 am
 Real name and title: Dr. JanMichael Becker
Re: Path coefficients > 1 HELP
Path coefficients like standardized regression coefficients can be larger than 1.
Unlike a correlation coefficient they are not bound between 1 and 1.
Yet, you are right that values outside this range usually give rise to some concerns, especially about multicollinearity problems. The most likely reason for such large values are severe collinearity problems that should be avoided.
See also
https://www.researchgate.net/post/Can_t ... 106e48e770
In your case, this is not a surprise as you have collinearity by design. You regress secondorder and firstorder constructs onto the same dependent variable, but the firstorder construct is part of the secondorder construct. In fact, if you regress all firstorder plus secondorder (which is the combination of all your firstorder constructs) you should have nearly perfect collinearity (because the secondorder construct is nearly perfectly predictable by your firstorder construct). Hence, such a model is not feasible. You should also see that by having extreme VIF values.
Unlike a correlation coefficient they are not bound between 1 and 1.
Yet, you are right that values outside this range usually give rise to some concerns, especially about multicollinearity problems. The most likely reason for such large values are severe collinearity problems that should be avoided.
See also
https://www.researchgate.net/post/Can_t ... 106e48e770
In your case, this is not a surprise as you have collinearity by design. You regress secondorder and firstorder constructs onto the same dependent variable, but the firstorder construct is part of the secondorder construct. In fact, if you regress all firstorder plus secondorder (which is the combination of all your firstorder constructs) you should have nearly perfect collinearity (because the secondorder construct is nearly perfectly predictable by your firstorder construct). Hence, such a model is not feasible. You should also see that by having extreme VIF values.
Dr. JanMichael Becker, University of Cologne, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Ja ... v=hdr_xprf
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Researchgate: https://www.researchgate.net/profile/Ja ... v=hdr_xprf
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de

 PLS Junior User
 Posts: 5
 Joined: Wed Jul 18, 2018 3:12 pm
 Real name and title: Carole Charbonnel
Re: Path coefficients > 1 HELP
Hi JanMichael,
thanks a lot for your reply, this was quite useful to understand the problem. The question now is : how can I deal with that ?
I use an existing formative index which as been developed formerly as Second order formative / formative measure :
12 indicators composing first order A / 11 indicators composing first order B / and second order C composed of A+B = 23 indicators.
I know that formative / formative measurmement models exist : in that case how can I deal with collinearity ? Do I have to integrate other independant variables ? Do I have to add more reflective measures ?
Thanks again for your help,
Kind regards
Carole
thanks a lot for your reply, this was quite useful to understand the problem. The question now is : how can I deal with that ?
I use an existing formative index which as been developed formerly as Second order formative / formative measure :
12 indicators composing first order A / 11 indicators composing first order B / and second order C composed of A+B = 23 indicators.
I know that formative / formative measurmement models exist : in that case how can I deal with collinearity ? Do I have to integrate other independant variables ? Do I have to add more reflective measures ?
Thanks again for your help,
Kind regards
Carole

 SmartPLS Developer
 Posts: 879
 Joined: Tue Mar 28, 2006 11:09 am
 Real name and title: Dr. JanMichael Becker
Re: Path coefficients > 1 HELP
The problem are the direct links from the firstorder constructs to the dependent variables if you include at the same time your secondorder construct. You have to decide: Either secondorder construct or firstorder constructs. You cannot have an effect of all three on your DVs.
Dr. JanMichael Becker, University of Cologne, SmartPLS Developer
Researchgate: https://www.researchgate.net/profile/Ja ... v=hdr_xprf
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de
Researchgate: https://www.researchgate.net/profile/Ja ... v=hdr_xprf
GoogleScholar: http://scholar.google.de/citations?user ... AAAJ&hl=de

 PLS Junior User
 Posts: 5
 Joined: Wed Jul 18, 2018 3:12 pm
 Real name and title: Carole Charbonnel
Re: Path coefficients > 1 HELP
Hi,
this is quite clear. Thanks a lot for your help
Kind regards
this is quite clear. Thanks a lot for your help
Kind regards