Hi Prof Debi,

SEM generally divided into Variance-based SEM / PLS-SEM (e.g. SmartPLS) and Covariance-based SEM (e.g. AMOS).  Like what Wan had stated 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 Debi,

I have not used it myself but I know that MPlus provides exploratory structural equation model (ESEM) where instead of Confirmatory factor analysis, Exploratory factor analysis may be employed to construct the latent variables and subsequently analyze the existence of structural relationship.

Hope this helps.

Best

Areti

If you interested to apply exploratory sem, you may apply PLS-SEM because i think this approach meet your need.

Dear Prof. Debi,

Although SEM is primarily used for confirmation of hypotheses, however it can also be used for exploratory purpose, for either or both  measurement and structural model. If the sole purpose of the exploration is to find out the factor structure, the ideal way is to start with EFA. However, there are no hard and fast rules, EFA can be pursued in confirmatory mode as well as CFA can be pursued in exploratory mode. If you wish to use SEM for exploration then you should have some supporting theory though they might be ambiguous. If you do not have so then it is clearly advisable to use EFA. If you know well the nuances of EFA, in fact it will provide you with very comprehensive information.    

I hope this helps.

rgds

Saidatt

Hi Prof Debi,

SEM generally divided into Variance-based SEM / PLS-SEM (e.g. SmartPLS) and Covariance-based SEM (e.g. AMOS).  Like what Wan had stated 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

Hi,

you /he may take a look on the work of the TETRAD group (Scheines et al.) who

invented the TETRAD software (can be freely downloaded):

Scheines, Richard, Spirtes, Peter, Glymour, Clark, Meek, Christopher, & Richardson, Thomas. (1998). The TETRAD project: Constraint based aids to causal model specification. Multivariate Behavioral Research, 33(1), 65-117.

http://www.hss.cmu.edu/philosophy/glymour/scheines1998.pdf

Landsheer, J.A. (2010). The specification of causal models with Tetrad IV: a review. Structural Equation Modeling, 17(4), 703-711.

http://www.phil.cmu.edu/projects/tetrad_download/

http://www.phil.cmu.edu/projects/tetrad_download/files/manual.pdf

http://www.phil.cmu.edu/projects/tetrad_download/files/new_manual.pdf

Having said that, I my point of view, throughing SEM and PLS into the same basket is an error. PLS is *not a kind of SEM* its ...well PLS. The core of SEM ist the specification and test of *causal restrictions*. This makes testing possible and valuable and provides some causal evidence (of course dependend of the validity of the model). PLS does not provide such testing. It belongs to the prediction camp - which is absolutely ok - but implies that PLS belongs to a different scientific universe.

Best,

Holger

P.S. (FOLLOW UP). I forgot the purpose of TETRAD: It is a software that delivers all possible causal models for a set of variables that confirm with a certain covariance structure (but no factor models). Hence it is an equivalent for exploratory factor analsis on the structural level.

Dear Prof Debi:

Social media being an evolving field, perhaps it may prove very useful especially if the goal is continued innovation!  

A recent (Alioni and Martini, 2013) survey of 500 Italian medium- and high-tech companies (by the Discontinuous Innovation Lab) tested the relations between search practices, exploration activities and firm performance using structural equation modelling.

The authors argued that the evidence shows that higher levels of search practices lead to higher levels of exploration activities and this, in its turn, can affect firm innovation performance!

• Aloini, D., & Martini, A. (2013). Exploring the exploratory search for innovation: a structural equation modelling test for practices and performance. International Journal Of Technology Management, 61(1), 23-46.

Hope this helps!

Nadeem

Dear Prof Debi:

In SEM there two models involved namely Measurement Model and Structural Model. Since your student is not doing any hypothesis testing, I guess he/she only intends to validate the Measurement Model for certain constructs in the development. If it is so, then Amos is appropriate since it has many indicators for Validity and Reliability. 

Yes you can, PLS-SEM Partial Least Square designed to explore the relationships. there are two tools:

1. Warp PLS

2. Smart PLS

Regards

Yes, the student can use structural equation modelling (SEM) in an exploratory research which can be qualitative/quantitative. However, if the student just wants to explore a 'phenomenon' he/she can use qualitative methods, then forget about SEM etc quantitative tools. If he/she wants to further the exploratory research to develop model(s) & test hypothesis/hypothses, then he/she can choose SEM/PLS-SEM/other quantitative tools/software depending on the  research design & type of data to be collected e.g ordinal/interval/continuous.

SmartPLS is for non-parametric tests. It still needs (better) a strong prior theory for exploratory research for model building and theory testing for various reasons. It can validate the Measurement Model for constructs  and has other features e.g Factor loading, Cronbach's Alpha, CFA etc. Measurement items also have to be tested/checked for collinearity as well. Sample size can be >100.