Copyright © 2020 Statswork. All rights reserved 1

Critique of a Published Latent Variable or SEM Study

Dr. Nancy Agens, Head,

Technical Operations, Statswork

info@statswork.com

I. INTRODUCTION

Structural equation modelling (SEM)

becomes a major statistical technique in

examining complex research problems in

marketing and international business. In

most research, the SEM uses covariance

based modelling and later few researchers

argued to use the partial least square

approach for SEM. In this blog, a critical

review of SEM technique presented in

Richter et al (2014) is discussed with

application to business sector.

Six journals related to the business

management and marketing have been

considered and the articles related to SEM

has been scrutinized for this purpose. After

the classification of methods used, it is

found that 379 articles used covariance

based SEM and 45 used partial least

square based SEM. Researchers often

interested in finding the same results by

using these both methods of Structural

equation modelling. However, the

consistency of the partial least square

method or the development of new

algorithm may satisfies this task fully and

yield same result as in covariance based

structural equation modelling.

Generally, the partial least square SEM is

useful to handle complex models and

provide better prediction with no demand

of the data. Thus, this article clearly

reviewed the methodology adopted either

CB SEM or PLS SEM and the purpose of

using the SEM model for better

understanding.

Lets look at few important factors which

differentiate the covariance based and

partial least square SEM in modelling

purpose.

The covariance based SEM has a strong

theoretical background and it estimates the

model by minimizing the covariance

matrix of the theoretical model and the

model based on empirical covariance

matrix of the data. Further, it is used to

identify the extent of empirical fit towards

the theoretical model.

However, the partial least square SEM is a

discovery oriented approach, that is,

without having a prior model and testing

the same, PLS SEM acts as Predictive

Analysis from the latent variable score. In

addition. PLS SEM is suitable for

modelling complex business problems. In

covariance based SEM, the complexity of

the model influence the goodness-of-fit

statistics. For example, consider a chi-

square test statistic, then if the

complexity of the model or the number of

parameters increases then the chi-square

value will get decreased. Hence, the result

will be either the correct model or the

highly fitted model because of the

complexity of the problem.

In the case of PLS SEM, the number of

parameters is not a problem (complexity)

until the sample size is sufficient. Also,

PLS SEM provides more appropriate

prediction than the maximum likelihood

estimation in CB SEM (Reinartz et al.

2009). Hence, it is important to decide

which approach is useful for the analysis

while carrying out the research. The

following table explains the number of

articles used covariance based and partial

least square SEM for the review purpose.

Copyright © 2020 Statswork. All rights reserved 2

Table 1: Number of Articles used Covariance Based and Partial Least Square

In addition, the review went deeper and

found that how many articles used

measurement model and the structural

model or the both for the analysis and a

proper justification of using the same.

From the results, it is found that only few

researchers justified the use of CB SEM is

that to test the theory using statistical

hypothesis testing and others are not and in

the case of PLS SEM most of the

researchers justified the usage of the

proposed model for analysis. Thus, it is

concluded that the PLS SEM might be a

better choice for conducting an analysis for

business and management. Furthermore,

the assumptions and the multicollinearity

factors in the data has been statistically

reviewed and provided a basic guidelines

for using the PLS structural equation

model and the tolerance level for various

factors such as VIF, reliability, validity,

etc.

In conclusion, the studies considered for

understanding the better method for

business problem found that the PLS SEM

is the best methodology than CB SEM

because often the business industry wants

a predictive model to enhance their

business standard or in investments. Thus,

PLS SEM satisfies the needs and works

well for predicting complex problems even

for the small sample sizes.

Further, it is advised to make a critical

assessment of the methodology or an

analytical approach before making a

business decisions. If the objective is to

develop the Theoretical Framework, then

the PLS SEM is appropriate and the

characteristics such as sample size,

assumptions of the distribution, the type of

measurement should be considered as

secondary one.

To sum up, SEM approach provides a

better understanding of the complex

problems in the field of business and

marketing and allow us to use various

modelling approaches. In addition to it,

PLS SEM acts as an major tool for the

exploratory analysis and it outperformed

CB SEM in many cases. A proper

sampling methodology should be adopted

for the analysis purpose and sample size

and its measurement type is also plays a

major role in the inference. Further, there

Copyright © 2020 Statswork. All rights reserved 3

has been few researchers that provide new

algorithm to show both SEM models

performs equally well. But, using a proper

tool makes the inference valid than using

the new method which results equal in

both CB and PLS SEM. Also, there should

be chance of lack of robustness in using

the partial least square method than the

maximum likelihood method. The

researchers should take care of that issue

because if the data contains influential or

the multi collinearity is present in the data

then the results may lead to invalid

conclusion. Thus, usage of PLS SEM

becomes a valid methodology to

understand the relationship between the

international business and the marketing

strategies. In this blog, I have listed out

few critics of the latent variable models

with an application to the international

marketing and business. However, this

may not be the case if you take other field

of research. Thus, a proper guideline

should be considered before processing

any Structural Equation Modeling.

REFERENCES

1. N F Richter et al (2016). A critical look at the use of

SEM in business research. International Marketing

Review. 33, 376-404.

2. Bentler, P.M. and Huang, W. (2014), “On

components, latent variables, PLS and simple

methods: reactions to Ridgon’s rethinking of PLS”,

Long Range Planning, Vol. 47 No. 3, pp. 138-145

3. Hair, J.F., Ringle, C.M. and Sarstedt, M. (2012),

“Partial least squares: the better approach to

structural equation modeling?”, Long Range

Planning, Vol. 45 Nos 5-6, pp. 312-319.

4. Michael O. Killian et al (2019). A Systematic

Review of Latent Variable Mixture Modeling

Research in Social Work Journals. LVMM

Systematic Review, 1-36.

5. Joseph F. Hair et al (2019). When to use and how to

report the results of PLS-SEM. EBR, 31, 1-24.

6. Khan, Gohar F., Marko Sarstedt, Wen-Lung Shiau,

Joseph F. Hair, Christian M. Ringle, and Martin

Fritze (2018). Methodological research on partial

least squares structural equation modeling (PLS-

SEM): An analysis based on social network

approaches. Internet Research, forthcoming