In conjunction with the above, can someone provide insights on the following:
1. What are all the different names pertaining to VB-SEM and CB-SEM e.g. VB-SEM also called Partial Least Squares, VB-SEM also called Covariance Structure Analysis, Covariance Structure Modeling, Analysis of Covariance Structure etc.?
2. What are the differences between Variance-based SEM (VB-SEM) vs Covariance-based SEM (CB-SEM)?
3. What are the advantages & disadvantages for each VB-SEM & CB-SEM?
4. What are the software (proprietary & OpenSource) to run VB-SEM & CB-SEM?
5. Where to find some best cited academic / journal articles / PowerPoint sharing the differences between VB-SEM vs CB-SEM?
6. Heard some people mentioned that CB-SEM is more superior / rigorous than VB-SEM - is it true? Is there any artifact / researched article support such claim?
Hello Dr.
The topic of choosing vb-sem vs cb-sem is an interesting and very controversial topic in multivariate data analyses. Unfortunately, researchers pass this important step without enough contemplation and choose one of them without enough reasons.
As I know, pals-sem is a specific kind of vb-sem. But it’s the most prevailing kind of vb-sem too.
There are some topic that a person must consider when want to choose between pls-sem and cb-sem:
1- Pls has more exploratory characteristics but cb-sem has more testing and verifying characteristics. So choose pls when your theories isn’t well developed or matured. Because discriminating between exploratory and confirmatory researches is very challenging and isn’t as straightforward as it looks, this criteria needs specific attention.
2- Another topic is considering data distribution. Cb-sem needs the data to be normally distributed. But pls-sem holds no assumption about distributions of datas.
3- Another topic is sample size, cb-sem need a lot more sample size. But pls-sem can converge with smaller sample sizes.
4- Another criteria is that pls-sem handles more complex models, more easily and very more efficiency.
5- Using formative measurement models is more logical and very easier in pls-sem rather than cb-sem. So if you insist on using formative measurement models, you must consider pls-sem as your primary choice.
6- Another criteria is that there is no universal goodness of fit (GOF) in pls-sem. So if you need one, you can’t reach one with pls-sem. Although there is some news in introducing new kind of pls (PLSc) that also has a universal gof (you can talk more with Professor Theo K. Dijkstra about this topic.)
https://www.researchgate.net/profile/Theo_Dijkstra
7- Another criteria is an inherent weakness of pls. There is a bias in estimating model parameters in pls-sem. Although some resources that those biases are subtle (Hair et al in their recent book on pls-sem), other sources do not (Professor Theo K. Dijkstra in his researches on PLSc).
In cb-sem, you can use either of these software: Lisrel, Amos, Mplus, R, etc.
In pls-sem, there is also a wide variety of choices, but I strongly recommend smartPls. Other options are “pls graphs” and “WarpPLS”.
A good reference on not only differences between pls and cb-sem but also on using PLS-SEM in practice is the recent book by Hair et al:
http://www.sagepub.com/books/Book237345
Best Regards
Hello Dr.
The topic of choosing vb-sem vs cb-sem is an interesting and very controversial topic in multivariate data analyses. Unfortunately, researchers pass this important step without enough contemplation and choose one of them without enough reasons.
As I know, pals-sem is a specific kind of vb-sem. But it’s the most prevailing kind of vb-sem too.
There are some topic that a person must consider when want to choose between pls-sem and cb-sem:
1- Pls has more exploratory characteristics but cb-sem has more testing and verifying characteristics. So choose pls when your theories isn’t well developed or matured. Because discriminating between exploratory and confirmatory researches is very challenging and isn’t as straightforward as it looks, this criteria needs specific attention.
2- Another topic is considering data distribution. Cb-sem needs the data to be normally distributed. But pls-sem holds no assumption about distributions of datas.
3- Another topic is sample size, cb-sem need a lot more sample size. But pls-sem can converge with smaller sample sizes.
4- Another criteria is that pls-sem handles more complex models, more easily and very more efficiency.
5- Using formative measurement models is more logical and very easier in pls-sem rather than cb-sem. So if you insist on using formative measurement models, you must consider pls-sem as your primary choice.
6- Another criteria is that there is no universal goodness of fit (GOF) in pls-sem. So if you need one, you can’t reach one with pls-sem. Although there is some news in introducing new kind of pls (PLSc) that also has a universal gof (you can talk more with Professor Theo K. Dijkstra about this topic.)
https://www.researchgate.net/profile/Theo_Dijkstra
7- Another criteria is an inherent weakness of pls. There is a bias in estimating model parameters in pls-sem. Although some resources that those biases are subtle (Hair et al in their recent book on pls-sem), other sources do not (Professor Theo K. Dijkstra in his researches on PLSc).
In cb-sem, you can use either of these software: Lisrel, Amos, Mplus, R, etc.
In pls-sem, there is also a wide variety of choices, but I strongly recommend smartPls. Other options are “pls graphs” and “WarpPLS”.
A good reference on not only differences between pls and cb-sem but also on using PLS-SEM in practice is the recent book by Hair et al:
http://www.sagepub.com/books/Book237345
Best Regards
Extremely Helpful and Adequately Described the difference between CB-SEM and VB or PLS-SEM. You can also read the following articles on more grip over the difference of two as it is essentially and somehow critically important for advance and appropriate data analysis.
Book A Primer on Partial Least Squares Structural Equation Modeli...
Book Partial Least Squares Structural Equation Modeling (PLS-SEM)...
Saeed Zolfaghari.. explanation about the difference between CBSEM and PLSEM is insightful.
The rationale behind this notation is grounded on the following three characteristics:
i) Basic specification of the structural models is similar, although approaches differ in terms of their model development procedure, model specification, theoretical background, estimation and interpretation (cf. Hair et al. 2010).
ii) VBSEM intends to explain variance, i.e. prediction of the construct relationships (Fornell & Bookstein 1982; Hair et al. 2010; Hair et al. 2012); CBSEM is based on the covariance matrices; i.e. this approach tends to explain the relationships between indicators and constructs, and to confirm the theoretical rationale that was specified by a model.
iii) Model parameters differ in two streams. The variance-based SEM is working with component weights that maximize variance, whereas the covariance-based SEM is based on factors that tend to explain covariance in the model (cf. Fornell & Bookstein 1982).
Source: http://bru-unide.iscte.pt/RePEc/pdfs/13-07.pdf
I think PLS-SEM is type of SEM that is usefull when the theory behind the relation ship is not clear. Somehow it is an exploratory analysis to find the relation between latent variables and its indicators. Also its very flexible method under critical condition such as lack of smaple size. However, covariance based-SEM (CB-SEM) are mostly recommended for linear parametric relationships.
It sounds weird to participate in this discussion after it was even concluded five years ago with a very comprehensive answer by Mr Saeed Zolfaghari , however this question by Dr. Han Ping Fung is still very much valid for not only novice but for many experienced researchers as well. Moreover, a head to head comparison posted by Dr. Mohsin Altaf is also really useful for the reader.
Though I have read many relevant articles providing comparison between PLS-SEM and CB-SEM, the first point (repeated below in quotes) mentioned by Mr Saeed Zolfaghari is still under-explained, at least in my understanding about the discussion to choose one method of modeling over and above the other.
"1- Pls has more exploratory characteristics but cb-sem has more testing and verifying characteristics. So choose pls when your theories isn’t well developed or matured. Because discriminating between exploratory and confirmatory researches is very challenging and isn’t as straightforward as it looks, this criteria needs specific attention."
Can anyone explain with examples from basic academic research (not from applied research) that What exactly means by Exploratory Modeling Vs Confirmatory Modeling? And please mention some reputed authors (articles) using CB-SEM with this same logic or reason. BECAUSE this exploratory/confirmatory argument is normally used only by researchers (like me :)) using PLS.
I totally agree with All other points mentioned above by all scholars.
Best Regards
- Badges
- Science method