Hi Syamsulang,

SEM generally divided into Variance-based SEM / PLS-SEM (e.g. SmartPLS) and Covariance-based SEM (e.g. AMOS which is an extension module from SPSS).  Variance-based SEM e.g. SmartPLS can be used for exploratory research whereas Covariance-based SEM is meant for confirmatory research / analysis.

Following shows the differences between the 2 SEMs in general:

1) Objective

Variance-based SEM - Focus on construct’s prediction i.e. variance explained

Covariance-based SEM - Focus on covariance i.e. explanation of items’ relationships

2) Theory Treatment

Variance-based SEM - Theory Exploration

Covariance-based SEM - Theory Confirmation i.e. need strong prior theory & established questionnaire

3) Sample Size

Variance-based SEM - Support small sample e.g. < 100

Covariance-based SEM - Need sizeable sample e.g. > = 300

4) Data Type

Variance-based SEM - Support both metric & non-metric data types

Covariance-based SEM - Only support metric data types

5) Construct Type

Variance-based SEM - Support both reflective & formative constructs

Covariance-based SEM - Support only reflective constructs

6) Construct-Item Relationship

Variance-based SEM - Support Construct with single-item

Covariance-based SEM - Support Construct with three-items & above

7) Missing Values

Variance-based SEM - Support data sets with missing values

Covariance-based SEM - Need to address missing values before analysis

8) Existence of Multi-collinearity

Variance-based SEM - Support data sets with multi-collinearity

Covariance-based SEM - Need to address multi-collinearity before analysis

Regards,

Fung

I suppose you speak about AMOS included in Spss (by IBM), for instance, but there are other software for SEM. You can use SEM both for primary or secondary data. The basic idea is having a theoretical idea of how reality behave, and test it with variables, stablishing a theoretical path of behaviour. It makes sense if we speak about multivariable models. There is a big quantity of references. See for instance:

Arbuckle, James L. 2008. Amos 17.0 Users' Guide. Chicago: SPSS Inc.

Bollen, Kenneth A. 1989. Structural Equations with Latent Variables. New York: Wiley.

Maruyama, Geoffrey. 1998. Basics of Structural Equation Modeling. Thousand Oaks: Sage.

Good luck exploring your data!

Thank you Estrella for the very informative answer..;-)

Hi Syamsulang,

SEM generally divided into Variance-based SEM / PLS-SEM (e.g. SmartPLS) and Covariance-based SEM (e.g. AMOS which is an extension module from SPSS).  Variance-based SEM e.g. SmartPLS can be used for exploratory research whereas Covariance-based SEM is meant for confirmatory research / analysis.

Following shows the differences between the 2 SEMs in general:

1) Objective

Variance-based SEM - Focus on construct’s prediction i.e. variance explained

Covariance-based SEM - Focus on covariance i.e. explanation of items’ relationships

2) Theory Treatment

Variance-based SEM - Theory Exploration

Covariance-based SEM - Theory Confirmation i.e. need strong prior theory & established questionnaire

3) Sample Size

Variance-based SEM - Support small sample e.g. < 100

Covariance-based SEM - Need sizeable sample e.g. > = 300

4) Data Type

Variance-based SEM - Support both metric & non-metric data types

Covariance-based SEM - Only support metric data types

5) Construct Type

Variance-based SEM - Support both reflective & formative constructs

Covariance-based SEM - Support only reflective constructs

6) Construct-Item Relationship

Variance-based SEM - Support Construct with single-item

Covariance-based SEM - Support Construct with three-items & above

7) Missing Values

Variance-based SEM - Support data sets with missing values

Covariance-based SEM - Need to address missing values before analysis

8) Existence of Multi-collinearity

Variance-based SEM - Support data sets with multi-collinearity

Covariance-based SEM - Need to address multi-collinearity before analysis

Regards,

Fung

Dear  Syamsulang,

I want to add some points to Mr Fung answer. I wouldn't talk about variance-based (VB) or covariance-based (CB) SEM instead I would differentiate between VB and CB estimators because the underlying model is the same but the way of estimation is a different.

Variance-based estimators, in particular PLS can also be used for confirmatory purpose, since Dijkstra & Henseler (2015) introduced consistent PLS, which corrects for attenuation if  constructs are modeled as common factors.

PLS is known to produce results for small samples sizes and in some cases you can estimate models with less observations than indicators. But you should be very very cautious with the interpretation. You can run a Monte Carlo simuation to figure out the statistical power of your model (Aguirre-Urreta & Rönkkö, 2015).

There exist a bunch of CB estimators which are able to deal with different types of scale, e.g WLSMV for categorical indicators. I'm not sure why VB estimators should deal every kind of scale as it is based on OLS most of the times (e.g. GSCA or PLS). Of course, there are some VB approaches which can deal with ordinal or nominal scaled indicators, e.g. Non-metric PLS (Rusolillio, 2012).

It's not true that CB estimators can only deal with reflective indicators, if you use google I'm sure you find ways to model formative measurement models. But VB is superior if the composite is in an endogenous position of the path model.

Why do you think CB estimators don't support single indicators? See e.g. Hayduk & Littvay (2012).

Which VB estimator supports data sets with missing values, as far as I know there exist different ways to deal with missing values but they also exist for CB methods. 

If you are interested in PLS, I reccomend the following article Henseler et al. (2016).

Best regards,

Florian

Dijkstra, Theo K., and Jörg Henseler. "Consistent partial least squares path modeling." Mis Quarterly (2015).

Aguirre-Urreta, Miguel, and Mikko Rönkkö. "Sample Size Determination and Statistical Power Analysis in PLS Using R: An Annotated Tutorial." Communications of the Association for Information Systems 36.1 (2015): 3.

Russolillo, Giorgio. "Non-metric partial least squares." Electronic Journal of Statistics 6 (2012): 1641-1669.

Hayduk, Leslie A., and Levente Littvay. "Should researchers use single indicators, best indicators, or multiple indicators in structural equation models?." BMC medical research methodology 12.1 (2012): 159.

Henseler, J., Hubona, G., Ray, P. A. (2016). Using PLS path modeling in new technology research: updated guidelines. Industrial Management & Data Systems, 116 (1), 2-20

Thanks for sharing ...

Both CB-SEM like AMOS and PLS-SEM can be used for both primary and secondary data.

"Structural-equation modeling is an extension of factor analysis and is a methodology designed primarily to test substantive theory from empirical data."

https://www.sciencedirect.com/topics/neuroscience/structural-equation-modeling

Structural-equation modeling is a multivariate statistical tool that combines correlation, regression, covariance and causality. they are of two types: covariance based and partial least square PLS based, that is CB-SEM and PLS-SEM......

First of all, i do thank all researchers and scholars for their comments and thoughts. In short, CB-SEM is regarded as a strictly theory-driven. Researchers in this stream tend to confirm the theoretical-assumptions and parameters-accuracy, while researchers in VB-SEM tend to focus primarily on the predictive power of the hypothesized-model (Davcik, 2014). More specifically, researchers through CB-SEM attempt to confirm specific theories (Davcik, 2014).

Some very informative informations.