There are various statistical method which are available for social science researchers but which technique will

be appropriate for their research is the big challenge. When research is based on covariance, CB-SEM approach

is used and when it is based on total variance, then PLS-SEM is an appropriate approach. This paper tries to

capture the attention of the researchers who face problems when to use CB-SEM and when to use PLS-SEM.

With the help of this paper, the effort is made to clearly define that CB-SEM is a parametric approach and PLS-

SEM is non parametric approach. In case of PLS-SEM, two measurement models are considered namely

measurement model(outer model) and structural model(Inner Model). In case of PLS-SEM, internal consistency

reliability is checked with the help of two namely Cronbach’s alpha and Composite reliability and there are other

ways of checking reliability and validity such as Composite reliability, Discriminant validity, HTMT and overall

SEM uses exploratory factor analysis and structural path analysis for evaluating both measurement and

structural models simultaneously (Lee et al., 2011). SEM is a very powerful tool that explains the total variance

There are two methods that are available for the researchers namely CB-SEM(Joreskog, 1978, 1993) and PLS-

CB-SEM is a covariance based SEM whereas PLS SEM is partial least squares. It is crucial for the researchers

to understand the difference between CB-SEM and PLS-SEM while deciding which approach researchers want

The purpose of this paper is to demonstrate the difference between two approaches which are used by the

researchers for their research. For this, we will understand the differences between these two approaches by

testing them into the same theoretical model and data and to check how the results differ empirically and

The statistical objectives of both the methods are different. CB-SEM estimates model parameters that minimizes

the difference between observed sample covariance and the covariance estimated the theoretical model is

confirmed(Hair et al., 2012b) whereas the statistical objective of PLS-SEM is to maximize the variance

covariance that could be used to predict the dependent variable. On the other side, in case of composite model,

all types of variance whether common, specific and even error variance from the exogeneous variables are

Due to the random error included in the composite model and indeterminacy in case of common factor model

will fit the model equally well; Grice (2001) and Steiger (1979)]. Both the approaches only produce the

approximations of the conceptual variables that conceptual variables or constructs seek to represent (Rigdon et

al., 2017). Both the approaches play a vital role in the field of research and no one can say that common factor

The structure of the outer model is completely different depends on the type of measurement. If the constructs

are measured with formative indicators then arrow will go from indicators to construct(Sarstedt et al., 2016) and

if the measurement is reflective then arrow will head from construct to indicators.(Sarstedt et al., 2016). If the

construct and its indicators are measured wrongly then the overall results will be bias and output will be incorrect.

So, it plays a vital role that which measurement type researchers using in their research area whether it is

As you may see in the above model, Y1 represents reflective measurement and Y2 represents formative model

whereas Y3 is an endogenous variable. In case of Y1, arrow goes from Y1 to three indicators which indicates

that construct reflects the indicators and in case of Y2 aas arrows goes from three indicator to Y2, it becomes

Since Y1 and Y2 are the exogenous variables and Y3 is an endogenous one. When we try to establish the

relationship between construct and its indicators, then it is called the measurement model or outer model and

reliability and validity, researchers should not depend on the face validity, they must check reliability and validity

measures and for internal reliability measurement, traditionally, Cronbach’s alpha is used but Cronbach

suggested that researcher should not only relay upon him for measuring internal consistence reliability but they

also use different approach of internal consistency reliability and for this Cronbach suggested Composite

Reliability. Composite reliability is recommended as more appropriate as it considers the indicators’ differential

weights (Chin, 1998; Dijkstra and Henseler, 2015), whereas Cronbach’s alpha weights the indicators equally

In measurement model, outer loading of the indicators are calculated so that AVE(average variance extracted)

can be easily calculated from each construct. The outer loadings of each indicator must exceed 0.708 because

the square of the loading of each indicator indicates that atleast 50% variance in the indicator is included in the

take additional steps where constructs are assessed based on their statistical significance and size of the indicator

Discriminant validity indicates that a construct is empirically unique and different from the other construct in

Discriminant validity means that each construct captures the unique phenomenon which is not represented by

For assessing the discriminant validity, the common approach is the Fornell-Larcker criterion(1981) that

compares the AVE(shared variance within) of the constructs to the squared correlation between the

constructs(shared variance between). For PLS-SEM(Variance based SEM, a more precise measure of

discriminant validity is Heterotrait-Monotrait ratio of correlations(HTMT) which was recently

Whatever SEM method researcher uses, structural relationships in both the methods are evaluated by the size

and the significance of the beta coefficients. In case of PLS-SEM, the structural model also considers the model’s

predictive capabilities which are known as the coefficient of determination (R2 value) which measures the

model’s in-sample predictive power(Hair et al., 2017c, 2017d) whereas in case of CB based SEM, goodness of

GOF(goodness of fit) is measured by the Chi-Square test that further indicates the difference between observed

covariance and the estimated covariance. Besides using Chi-Square statistics for GOF, there are other means also

for assessing GOF under CB-SEM approach. Researchers can calculate CFI, GFI and RMSEA also. But in

size is large means more than 100, then CB-SEM is suitable but it is notion that PLS-SEM will be applicable for

Both the methods play a vital role in the field of research while doing path analysis. It is upto the research