Can a direct effect be greater than one with plsc?

 PLS Junior User
 Posts: 2
 Joined: Fri Sep 26, 2014 5:58 am
 Real name and title: Jose Ramon Segarra
Can a direct effect be greater than one with plsc?
I obtained all the values of direct effects with pls (these values are between 0 and 1). But if I use plsc, one of them is greater than one (1.10) and there are other direct effect that it is negative (0.26)
 cringle
 SmartPLS Developer
 Posts: 797
 Joined: Tue Sep 20, 2005 9:13 am
 Real name and title: Prof. Dr. Christian M. Ringle
 Location: Hamburg (Germany)
 Contact:
Re: Can a direct effect be greater than one with plsc?
PLSc is a little bit tricky. Only use it if you have a data set that is large enough (e.g., a sample size > 300) and reflective measurment models with many indicators (e.g., more than 3) and high loadings of all indicators (e.g., >0.7 better >0.8). Then, you should not obtain odd outcomes. Otherwise, there always is a possibility to get outcomes as described by you.
Best
CR
Best
CR
Prof. Dr. Christian M. Ringle, Hamburg University of Technology (TUHH), SmartPLS
 Literature on PLSSEM: https://www.smartpls.com/documentation
 Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de
 Literature on PLSSEM: https://www.smartpls.com/documentation
 Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de

 PLS Junior User
 Posts: 2
 Joined: Fri Sep 26, 2014 5:58 am
 Real name and title: Jose Ramon Segarra
Re: Can a direct effect be greater than one with plsc?
Thanks, I think I have the solution with your support.
The problem was in the number of reflective indicators because the sample contains 281 observations. My model has three second order factors (value, brand and relationship) as antecedents of loyalty. Value for example, has three dimensions or first order factors: functional, emotional and social.
When I applied the dimensions from method of the twostep as indicators (Wright 2012: latent variable scores), then PLS estimators were between 0 and 1, but they were very different in PLSC (value was greater than 1, brand was zero and relationship was negative).
However, if I apply all original indicators of the dimensions corresponding to each one of the three second order factors, then PLSC estimators are between 0 and 1 and they are very similar to PLS too.
But now I have other problem: I don't have a second order model with PLSC and some indicators with individual fiability are lower than 0.707.
The problem was in the number of reflective indicators because the sample contains 281 observations. My model has three second order factors (value, brand and relationship) as antecedents of loyalty. Value for example, has three dimensions or first order factors: functional, emotional and social.
When I applied the dimensions from method of the twostep as indicators (Wright 2012: latent variable scores), then PLS estimators were between 0 and 1, but they were very different in PLSC (value was greater than 1, brand was zero and relationship was negative).
However, if I apply all original indicators of the dimensions corresponding to each one of the three second order factors, then PLSC estimators are between 0 and 1 and they are very similar to PLS too.
But now I have other problem: I don't have a second order model with PLSC and some indicators with individual fiability are lower than 0.707.
 cringle
 SmartPLS Developer
 Posts: 797
 Joined: Tue Sep 20, 2005 9:13 am
 Real name and title: Prof. Dr. Christian M. Ringle
 Location: Hamburg (Germany)
 Contact:
Re: Can a direct effect be greater than one with plsc?
Hi
2nd order models do not work with PLSc.
Best
CR
2nd order models do not work with PLSc.
Best
CR
Prof. Dr. Christian M. Ringle, Hamburg University of Technology (TUHH), SmartPLS
 Literature on PLSSEM: https://www.smartpls.com/documentation
 Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de
 Literature on PLSSEM: https://www.smartpls.com/documentation
 Google Scholar: https://scholar.google.de/citations?use ... AAAJ&hl=de

 PLS User
 Posts: 13
 Joined: Mon Jun 15, 2015 10:38 am
 Real name and title: Dr. Linda Plotnick
Re: Can a direct effect be greater than one with plsc?
Will it ever be possible to use PLSc with models that have 2nd order formative constructs?
I have a model for which there are some problems with multicollinearity so I would like to use PLSc but I have 2 formative constructs which are predictors. I want to be able to use Gaskin's 2 step approach for PLS (use latent variable scores) but, although I can do this in the regular PLS, I cannot in PLSc. Is there a solution? I apologize for showing my ignorance in this question.
Thanks,
Linda
I have a model for which there are some problems with multicollinearity so I would like to use PLSc but I have 2 formative constructs which are predictors. I want to be able to use Gaskin's 2 step approach for PLS (use latent variable scores) but, although I can do this in the regular PLS, I cannot in PLSc. Is there a solution? I apologize for showing my ignorance in this question.
Thanks,
Linda

 SmartPLS Developer
 Posts: 879
 Joined: Tue Mar 28, 2006 11:09 am
 Real name and title: Dr. JanMichael Becker
Re: Can a direct effect be greater than one with plsc?
If you want to do a twostep approach, you can use the latent variable scores from the first step of a normal PLS and then use PLSc only for the second step. That should work quite well.
However, I don't understand why PLSc should be advantageous when you have multicollinearity in your model.
However, I don't understand why PLSc should be advantageous when you have multicollinearity in your model.
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 User
 Posts: 13
 Joined: Mon Jun 15, 2015 10:38 am
 Real name and title: Dr. Linda Plotnick
Re: Can a direct effect be greater than one with plsc?
Thank you for your answer.
Perhaps I misunderstood the 2015 Dijkstra and Henseler article (my math is not quite up to the standard needed to thoroughly understand the article), but the authors, in the results section, state "The most important insights referred to small or zero effects under multicollinearity. .... In other words, the inconsistent techniques of PLS and regression evoke an undue number of Type II errors, whereas the consistent techniques do not. A different phenomenon was observed for the zero effect under the multicollinearity condition. Whereas PLSc and the covariancebased SEM techniques (except WLS) maintained acceptible levels of Type I errors, PLS and regression found significant effects in more than 30 percent of the cases for small sample sizes and, in more than 90 percent of the cases, for large samples. ..." (page 309)
Perhaps I misunderstood the 2015 Dijkstra and Henseler article (my math is not quite up to the standard needed to thoroughly understand the article), but the authors, in the results section, state "The most important insights referred to small or zero effects under multicollinearity. .... In other words, the inconsistent techniques of PLS and regression evoke an undue number of Type II errors, whereas the consistent techniques do not. A different phenomenon was observed for the zero effect under the multicollinearity condition. Whereas PLSc and the covariancebased SEM techniques (except WLS) maintained acceptible levels of Type I errors, PLS and regression found significant effects in more than 30 percent of the cases for small sample sizes and, in more than 90 percent of the cases, for large samples. ..." (page 309)