Introduction to SEM Using the

Partial Least Squares (PLS)

T. Ramayah
School of Management
Universiti Sains Malaysia
http://www.ramayah.com

Structural Equations Modeling

Structural Equations Modeling . . . is a family

of statistical models that seek to explain the
relationships among multiple variables.
It examines the structure of interrelationships
expressed in a series of equations, similar to a
series of multiple regression equations.
These equations depict all of the relationships
among constructs (the dependent and
independent variables) involved in the analysis.
Constructs are unobservable or latent factors
that are represented by multiple variables.
Called 2nd Generation Techniques

Distinguishing Features of SEM

Compared

to 1st Generation Techniques

It takes a confirmatory rather than
exploratory
Traditional methods incapable of either
assessing or correcting for measurement
errors
Traditional methods use observed
variables, SEM can use both unobserved
(latent) and observed variables
Testing in one complete model

Components of Error
Observed

score comprises of 3 components

(Churchill, 1979)
True

score
Random error (ex; caused by the order of items in
the questionnaire or respondent fatigue) (Heeler &
Ray, 1972)
Systematic error such as method variance (ex;
variance attributable to the measurement method
rather than the construct of interest) (Bagozzi et
al., 1991)

SEM

SEM, as a second-generation technique, allows the

simultaneous modeling of relationships among
multiple independent and dependent constructs
(Gefen, Straub, & Boudreau, 2000). Therefore, one
no longer differentiates between dependent and
independent variables but
distinguishes between the exogenous and endogenous
latent variables, the former being variables which are not
explained by the postulated model (i.e. act always as
independent variables) and the latter being variables that are
explained by the relationships contained in the model.
(Diamantopoulos, 1994, pp. 108)

Terms
o Exogenous

constructs are the latent, multi-item

equivalent of independent variables. They use a
variate (linear combination) of measures to
represent the construct, which acts as an
independent variable in the model.

o Multiple measured variables (x) represent the exogenous

constructs.

o Endogenous constructs are the latent, multi-item

equivalent to dependent variables. These constructs

are theoretically determined by factors within the model.

o Multiple measured variables (y) represent the endogenous

constructs.

Indicators
Formative

Reflective

LIFE STRESS

TIMELINESS

X1

X2

X1 = Accommodate last minute request

X2 = Punctuality in meeting deadlines
X3 = Speed of returning phone calls

X1

X3

Indicators must be highly

correlated (Hulland,
1999)

X2

X3

X1 = Job loss
X2 = Divorce
X3 = Recent accident

Indicators can have +, - or

0 correlation (Hulland,
1999)

Two approaches to SEM

Covariance

based

EQS,

AMOS, SEPATH, COSAN, LISREL

and MPLUS

Variance

based

PLS
Smart

PLS, PLS Graph

Why PLS?

Like covariance based structural equation modeling

(CBSEM), PLS is a latent variable modeling technique
that incorporates multiple dependent constructs and
explicitly recognizes measurement error (Karim, 2009)

In general, two applications of PLS are possible (Chin,

1998a): It can either be used for theory confirmation or
theory development. In the latter case, PLS is used to
develop propositions by exploring the relationships
between variables.

Reasons for using PLS

Researchers arguments for choosing PLS as the

statistical means for testing structural equation models
(Urbach & Ahleman, 2010) are as follows:

PLS makes fewer demands regarding sample size than other

methods.
PLS does not require normal-distributed input data.
PLS can be applied to complex structural equation models
with a large number of constructs.
PLS is able to handle both reflective and formative constructs.
PLS is better suited for theory development than for theory
testing.
PLS is especially useful for prediction

Choice

Overall, PLS can be an adequate alternative to CBSEM if the

problem has the following characteristics (Chin 1998b; Chin &
Newsted 1999):

The phenomenon to be investigated is relatively new and

measurement models need to be newly developed,
The structural equation model is complex with a large number of
LVs and indicator variables,
Relationships between the indicators and LVs have to be modeled
in different modes (i.e., formative and reflective measurement
models),3
The conditions relating to sample size, independence, or normal
distribution are not met, and/or
Prediction is more important than parameter estimation.

Choice

Process

Comparison

The 2 Step Approach

A structural

equation modeling process

requires two steps:
1. building and testing a measurement
model, and
2. building and testing a structural model.
The measurement model serves to create a
structural model including paths representing
the hypothesized associations among the
research constructs.

Confirmatory Factor Analysis (CFA)

Confirmatory Factor Analysis (CFA) . . . is similar to EFA in some

respects, but philosophically it is quite different.

With CFA, the researcher must specify both the number of factors
that exist within a set of variables and which factor each variable
will load highly on before results can be computed.

So the technique does not assign variables to factors.

Instead the researcher must be able to make this assignment
before any results can be obtained.

SEM is then applied to test the extent to which a researchers apriori pattern of factor loadings represents the actual data.

Review of and Contrast with Exploratory

Factor Analysis

EFA (exploratory factor analysis) explores the data and provides

the researcher with information about how many factors are
needed to best represent the data. With EFA, all measured
variables are related to every factor by a factor loading estimate.
Simple structure results when each measured variable loads
highly on only one factor and has smaller loadings on other
factors (i.e., loadings < .40).
The distinctive feature of EFA is that the factors are derived from
statistical results, not from theory, and so they can only be
named after the factor analysis is performed.
EFA can be conducted without knowing how many factors really
exist or which variables belong with which constructs. In this
respect, CFA and EFA are not the same.

Measurement Model and Construct Validity

One of the biggest advantages of CFA/SEM is its ability to

assess the construct validity of a proposed measurement
theory. Construct validity . . . is the extent to which a set
of measured items actually reflect the theoretical latent
construct they are designed to measure.

Construct validity is made up of two important components:

1.

Convergent validity three approaches:

2.

Factor loadings.
Variance extracted.
Reliability.

Discriminant validity

Average Variance Extracted (AVE)

2i = squared loadings of indicator i of a latent variable

var(i ) = squared measurement error of indicator i

AVE should exceed 0.5 to suggest adequate

convergent validity (Bagozzi & Yi, 1988).

Composite Reliability (CR)

i = loadings of indicator i of a latent variable

i = measurement error of indicator i
j = flow index across all reflective measurement model

Composite reliability should be 0.7 or higher

to indicate adequate convergence or internal

consistency (Gefen et al., 2000).

Cronbach Alpha ()

N = number of indicators assigned to the factor

2i = variance of indicator i
2t = variance of the sum of all assigned indicators scores
j = flow index across all reflective measurement model

Cronbach alpha values should be 0.7 or

higher to indicate adequate convergence or

internal consistency. (Nunnally, 1978)

Reporting Measurement Model

Model Construct

Measurement Item

Loading

Commitment

COMMIT1

0.686

COMMIT2

0.767

COMMIT3

0.885

COMMIT4

0.751

COMMUN1

0.842

COMMUN2

0.831

COMMUN3

0.829

TRUST1

0.580

TRUST2

0.587

TRUST3

0.948

PERFORM1

0.837

PERFORM2

0.900

PERFORM2

0.853

Communication

Trust

Performance

Cronbach

CRa

AVEb

0.843

0.856

0.601

0.860

0.873

0.696

0.744

0.759

0.527

0.885

0.898

0.747

Presenting Measurement Items (Table)

Model Construct
Commitment

Communication

Trust

Performance

Measurement
Item
COMMIT1

Standardized estimate

t-value

0.686

5.230**

COMMIT2

0.767

6.850**

COMMIT3

0.885

22.860**

COMMIT4

0.751

6.480**

COMMUN1

0.842

20.628**

COMMUN2

0.831

16.354**

COMMUN3

0.829

15.011**

TRUST1

0.580

1.960*

TRUST2

0.587

2.284**

TRUST3

0.948

2.640**

PERFORM1

0.837

16.081**

PERFORM2

0.900

33.456**

PERFORM2

0.853

13.924**

t-values > 1.96* (p< 0.05); t-values > 2.58** (p< 0.01)

Discriminant Validity
The square root of the Average Variance Extracted

(AVE) that exceeds the intercorrelations of the construct

with the other constructs in the model to ensure
discriminant validity (Chin, 2010; Chin 1998b; Fornell &
Larcker, 1981).
Example:

Communality and Redundancy

Construct

R2

Communality

H2

Redundancya

GOF
a

Redundancy j = communality j X R2. The redundancy index measures the quality of the structural model for each
endogenous block, taking into account the measurement model (Tenanhaus et al., 2005).

F2

Communality and Redundancy

and F2 the values should be greater than the threshold

of 0 (Fornell & Cha, 1994)
2

Q-squares Statistics

A The Q-squares statistics measure the predictive relevance of the model by

reproducing the observed values by the model itself and its parameter
estimates. A Q-square greater than 0 means that the model has predictive
relevance; whereas Q-square less than 0 mean that the model lacks
predictive relevance (Fornell & Cha, 1994).
In PLS, two kinds of Q-squares statistics are estimated, that is, crossvalidated (c-v) communality (H2j) and cross-validated redundancy (F2j).
Both statistics are obtained through blindfolding procedure in PLS. Blindfolding
procedure (while estimating Q-squares) ignores a part of the data for a
particular block during parameter estimation (a block of indicators is the set of
measures for a construct). The ignored data part is than estimated using the
estimated parameters, and the procedure is repeated until every data point has
been ignored and estimated. Omission and estimation of data point for the
blindfolded construct depend on the chosen omission distance G (Chin, 1998a).

Q-squares Statistics

A cross-validated communality H2j is obtained if prediction of the omitted data points in the
manifest variables block is made by underlying latent variable (Chin, 1998). In other words,
the cv-communality H2j measures the capacity of the path model to predict the manifest
variables (MVs) directly from their own latent variable (LV) by cross-validation. It uses only
the measurement model.
The prediction of a MV of an endogenous block is carried out using only the MVs of this
block (Tenanhaus et al., 2005). On the other hand, a cross-validated redundancy predicts
the omitted data points by constructs that are predictors of the blindfolded construct in the
PLS model (Chin, 1998a).
In other words, the cv-redundancy F2j measures the capacity of the path model to predict
the endogenous MVs indirectly from a prediction of their own LV using the related
structural relation, by crossvalidation.
The prediction of an MV of an endogenous block j is carried out using all the MVs of the
blocks j* related with the explanatory LVs of the dependent LVj (Tenanhaus et al., 2005).
This index is used for measuring the quality of the path model. In accordance to effect size
(f2), the relative impact of the structural model on the observed measures for latent
dependent variable is evaluated by means of q 2 (Henseler et al., 2009).
The q2 values of 0.02, 0.15, and 0.35 signify small, medium, and large predictive relevance
of certain latent variable, thus explaining the endogenous latent variable under evaluation

Goodness of Fit (GOF)

What

is Goodness of Fit (GOF)?

Goodness-of-fit (GoF) (Tenanhaus et al., 2005) is used to

judge the overall fit of the model.
GoF, which is the geometric mean of the average
communality (outer measurement model) and the average
R2 of endogenous latent variables, represents an index for
validating the PLS model globally, as looking for a
compromise between the performance of the measurement
and the structural model, respectively.
GoF is normed between 0 and 1, where a higher value
represents better path model estimations.

Assessing Goodness of Fit (GOF)?

Assessing

GOF

Goodness of Fit (GOF)

2

R x Average Communality

Global validation of PLS models use these cut-off

values (Wetzels et al. 2009):
GoFsmall = 0.10

GoFmedium = 0.25

GoFlarge = 0.36).

Allows to conclude that the model used has better

explaining power in comparison with the baseline
model

Assessing R2

According to Chin (1998b), R

values for
endogenous latent variables are assessed as
follows:
2

0.67
0.33
0.19

substantial
moderate
weak

Also path coefficients range between 0.20

0.30 along with measures that explain 50% or

more variance is acceptable (Chin, 1998b)

Calculating Effect Size (f2)

Effect size f

is not automatically given in

PLS, we have to do manual calculation
using the formula:
2

According to Cohen (1988), f

small
0.02
medium
0.15
large
0.35

is assessed as:

Presenting the results (Figure)

Presenting the results (Table)

The

t-values are generated using the

bootstrapping with re-samples of 200 (Chin,
1998b)
Relationship

H1
H2
H3
H4

SYS SAT
IQ SAT
SQ SAT
SAT INT

Coefficient
0.23
0.17
0.24
0.38

t-value
2.588**
1.725
2.645**
3.895**

t-values > 1.96* (p< 0.05); t-values > 2.58** (p< 0.01)

Supported
YES
NO
YES
YES

Moderator Effect Assessment

According to Cohen (1988), f

small
0.02
medium
0.15
large
0.35

is assessed as:

Mediator Effect Assessment

Can Use the Sobel (1982) test

Online:
http://www.people.ku.edu/~pre
acher/sobel/sobel.htm

VAF = Variance accounted for

Indirect effect / Total Effect

The indirect effect is VAF is converted

significant at:
0.05 if z > 1.96
0.01 f z > 2.58

into %

http://www.people.ku.edu/~preacher/sobel/sobel.htm

a and b = path coefficient

sa and sb = standard errors

ta and tb = t-values for a and b path coefficients generated from

bootstrapping

References
Hulland, J. (1999). Use of partial least squares (PLS) in strategic management research: A review
of four recent studies. Strategic Management Journal, 20, 195204.
Bagozzi, R. P., Yi, Y., & Philipps, L. W. (1991). Assessing construct validity in organizational
research. Administrative Science Quarterly, 36, 421458.
Heeler, R. M., & Ray, M. L. (1972). Measure validation in marketing. Journal of Marketing
Research, 9, 361370.
Churchill, G. A. J. (1979). A paradigm for developing better measures of marketing constructs.
Journal of Marketing Research, 16, 6473.
Diamantopoulos, A. (1994). Modelling with LISREL: A guide for the uninitiated. Journal of
Marketing Management, 10, 105136.
Gefen, D., Straub, D. W., & Boudreau, M.-C. (2000). Structural equation modeling and regression:
Guidelines for research practice. Communications of the Association for Information Systems, 4,
179.
Chin, W.W. (2000). Partial least squares for researchers: an overview and presentation of recent
advances using the PLS approach. available at: http://disc-nt.cba.uh.edu/chin/icis2000plstalk.pdf

References
Chin, W. W. (2010). How to write up and report PLS analyses. In V. Esposito Vinzi, W. W. Chin,
J. Henseler, & H. Wang (Eds.), Handbook of partial least squares: Concepts, methods and
application (pp. 645689). New York: Springer.
Chin, W.W. (1998a). The partial least squares approach to structural equation modeling. In
G.A. Marcoulides (Ed.), Modern business research methods. Mahwah, NJ: Lawrence Erlbaum
Associates.
Chin, W.W. (1998b). Issues and opinion on structural equation modeling. MIS Quarterly, 22(1),
7-16.
Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable
variables and measurement error. Journal of Marketing Research, 18(1), 39-50.
Tenenhaus, M., Esposito Vinzi, V., Chatelin, Y.-M., & Lauro, C. (2005). PLS path modeling.
Computational Statistics & Data Analysis, 48(1), 159205.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.), Hillsdale,
Lawrence Erlbaum Associates, NJ.
Nunnally, J. C. (1978). Psychometric theory. NewYork: McGraw-Hill.