You are on page 1of 8

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/304629779

STRUCTURAL EQUATION MODELING (SEM): A


STEP BY STEP APPROACH (PART 1)

Working Paper · July 2016


DOI: 10.13140/RG.2.1.1238.6809

CITATIONS READS

0 520

1 author:

Zuraidah Zainol
Universiti Pendidikan Sultan Idris (UPSI)
19 PUBLICATIONS 15 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

customer-brand engagement View project

All content following this page was uploaded by Zuraidah Zainol on 01 July 2016.

The user has requested enhancement of the downloaded file.


STRUCTURAL EQUATION MODELING (SEM): A STEP BY
STEP APPROACH
(PART 1)
By: Zuraidah Zainol (PhD)
Faculty of Management & Economics, Universiti Pendidikan Sultan Idris
zuraidah@fpe.upsi.edu.my
2016

INTRODUCTION

Structural equation modeling (SEM) is no more than another statistical technique that can be

used to analyse data. It is not a new technique for analysing data, just an extension of the

existing statistical techniques. However, even with some statistical background, reading and

understanding SEM’s output in a research report seems very difficult and could lead to a total

frustration. As a matter of fact, SEM procedures are easy to perform and the results are easy

to interpret. However, due to many uncommon statistical terms that are used to describe

almost every aspect of SEM, as if SEM has its own language, explanation of SEM procedures

and the results is just hard to understand. For instance, instead of using familiar terms such

independent and dependent variables, in SEM, the terms such as observed variables, latent

variables, endogenous variables and exogenous variables are used to distinguish the types of

variables. By analogy, learning SEM, is just like learning how to drive an automatic

transmission car. Without any driving experience, it is still easy to master the skill. But once

you already grasp the skills to drive a manual transmission car, mastering the skill to drive

automatic transmission car would be much easier. Hence, with a clear understanding of the

basic concepts of SEM, it would be easy for anyone to conduct SEM analysis and, interpret

and report the results of SEM. Thus, for a person to conduct SEM procedures and report the

results, or only to have an idea of what is explained in SEM-based research report, having a

clear idea of the basic concepts of SEM is a must. Accordingly, this manual was written

1
based on three main objectives, namely to clarify what is SEM, the basic concepts in SEM,

the steps to perform SEM analysis and how to report the results.

WHAT IS SEM?

Structural equation modeling (SEM) is a multivariate technique that allows for simultaneous

analysis of a series of direct or indirect dependence relationships between multiple

independent and dependent variables (Garson, 2012b; Groenland & Stalpers, 2012; Hair,

Black, Babin, & Anderson, 2010; Ho, 2006; Tabachnick & Fidell, 2007). Hence, it is a

technique that can be used to test a simple direct relationship between two variables, a direct

relationship between one or several independent variables and several dependent variables

and even an indirect relationship, with the existence of a mediator. In that regards, SEM is an

advance statistical techniques as it incorporates many of the existing statistical techniques

including factor analysis, path analysis, correlation, analysis of variance (ANOVA) and

multiple regression (Garson, 2012b; Ho, 2006). Due to its capability to test many types of

relationship, including direct and indirect, as well as the mediating and moderating effects,

SEM is considered as a comprehensive analysis to test the hypotheses about the relationship

between observed and latent variables.

SEM APPROACHES

To conduct SEM, there are two approaches that are widely identified in the literature, namely

covariance-based SEM (CB-SEM) and partial least square SEM (PLS-SEM). CB-SEM is

considered more appropriate to conduct SEM whenever confirming the relationships between

latent variables serves as a major research concern, the structural model is low to moderate

complexity, the relationship between indicators and latent variables is modelled in reflective

mode and the sample size used is large (Haenlein & Kaplan, 2004; Hair, Ringle, & Sarstedt,
2011; Henseler, Ringle, & Sinkovics, 2009; Henseler & Sarstedt, 2013; Urbach & Ahlemann,

2010). By contrast, PLS-SEM is deemed appropriate for prediction, particularly when the

proposed model is relatively new with lack of theoretical supports, the structural model is

complex, the measurement model is modelled in reflective and/or formative mode and a

small sample is used (Hair, et al., 2011; Henseler, et al., 2009; Henseler & Sarstedt, 2013;

Urbach & Ahlemann, 2010). Thus, it is apparent that both approaches to SEM are

complementary and neither is superior to the other (Hair, Sarstedt, Ringle, & Mena, 2012).

To perform CB-SEM, software such as Analysis of MOment Structure (AMOS), EQS,

LISREL and Mplus can be used (Garson, 2012b; Hair, et al., 2011; Kline, 2011; Tabachnick

& Fidell, 2007), while for PLS-SEM, SmartPLS, SPSS PLS, PLS Graph and R package

semPLS can be used (Garson, 2012a; Hair, et al., 2011; Wan Mohamad Asyraf, 2013; Wong,

2013).

To decide on which approach to use, three aspects need to be considered, i.e. the research

objective, model set-up and data characteristics (Hair, et al., 2012; Ronkko & Evermann,

2013). If the research aim is to verify the proposed relationships between constructs, which

derived from strong relationship theories, using reflective measurement of latent variables

and a large sample size, CB-SEM deems as better SEM approaches for the study. More

importantly, it has been emphasized that “PLS-SEM should be considered as an alternative

where CB-SEM could not be used for any reason” (Oke, Ogunsami, & Ogunlana, 2012, p.

91). Therefore, whenever the data collected failed to satisfy the CB-SEM assumptions, PLS-

SEM will be considered as alternative approaches. Details regarding the selection of SEM

approach are shown in


Table 1.

4
Table 1: Rules Of Thumb For Selecting PLS-SEM or CB-SEM

Criteria PLS-SEM CB-SEM Source


- Theory testing, confirmation, selection or comparison
Research - Predicting key target constructs or identifying key (Hair, et al., 2011;
of alternative theories
Goals driver constructs Henseler, et al.,
- A theory oriented (prior theory is strong)
- Exploratory, theory building or an extension of an 2009; Henseler &
- Testing the structural relationships (parameter
existing structural theory Sarstedt, 2013)
estimation) between the latent constructs is the
- To test and validate exploratory model in the early
primary concern
stage of the theoretical development
- For studies that intend to explain variances of the
dependent variables and predict the dependent
variable when a priori model does not exist .
- Investigate a relatively new phenomenon and - Established constructs and reflective measurement Chin & Newsted
measurement models need to be newly developed models are available (1999) as cited in
- The relationship between indicators and LV has to be - In a confirmatory study, which structural model is (Urbach &
modelled in different modes – formative and low to moderate complexity Ahlemann, 2010)
reflective - More frequently accepted for rigorous model
- Prediction is more important than parameter validation purposes –established approach with
estimation recognized GOF metric
Model set-up - If formative are part of the structural model - If error terms require additional specification (Hair, et al., 2011)
- If the structural model is complex (many constructs - If the model is small to moderate complexity – e.g. (Hair, et al., 2011;
and indicators) – e.g. 100 constructs and 1000 less than 100 indicators, strong prior theory Henseler, et al.,
indicators; but lack of theoretical supports 2009; Urbach &
Ahlemann, 2010)
- If need to use latent variable scores in subsequent - If research requires a global goodness of fit criterion (Hair, et al., 2011)
analyses - Need to test for measurement model invariance
Data - Low sample size and nonnormal distribution - If the data meet the CBSEM assumptions (Hair, et al., 2011;
characteristics Urbach & Ahlemann,
*more accurate results for studies with a small sample *under normal data, CBSEM provide a slightly more
2010)
(minimum sample of 30 to 100 cases) precise model estimates/accurate result
Sample size: min 200 to 800
- In dealing with the problem of ‘consistency at large’ – - For a large data set (Haenlein & Kaplan,
for large data 2004)

5
WHEN TO USE SEM?

Despite there are two approaches of SEM, this manual will concentrate on CB-SEM rather than

PLS-SEM. It is, therefore, CB-SEM will be referred as SEM hereafter.

SEM is a statistical technique for those researchers that follow the positivist, quantitative and

deductive approach. The basic aim of the study is to test a theory-driven hypothesis, that is either

to confirm the theory, test the modified model or compare competing theoretical models. To

confirm the theory means testing the standard theoretical model in different context in order to

prove its applicability, while to test the modified theory is to test the extended model

(modification of the standard model with either addition and/or deletion of new constructs in the

model). As for comparing competing theoretical models, the aim is to choose the best fit model

by testing several models, either standard and/or modified models. Basically, SEM is used to test

a model comprising a minimum of 2 independent variables (IVs) and more than 1 dependent

variable (DV), but all the variables and data must be either interval or ratio.

TEST YOUR UNDERSTANDING

Can we use SEM to test the following research objective?

1. to develop and verify a scale to measure brand economic and social investment

2. to determine the effect of customer commitment on customer engagement

3. to investigate the effect of green attitude, subjective norm, perceived behavioural control

and green practice consequences toward Malaysian customers’ intention to recycle and

spread positive word-of-mouth


4. to test the mediating role of trust in the effect of satisfaction on customer commitment

5. to assess the moderating effect of relationship duration in the relationship between

customer engagement and its antecedent variables.

View publication stats

You might also like