Partial Least square- Structural Equation Modeling

Chapter 1: AN INTRODUCTION TO STRUCTURAL EQUATION MODELING

LEARNING OUTCOMES
1. Understand the meaning of structural equation modeling (SEM) and its relationship to
multivariate data analysis.
2. Describe the basic considerations in applying multivariate data analysis.
3. Comprehend the basic concepts of partial least squares structural equation modeling
(PLS-SEM).
4. Explain the differences between covariance-based structural equation modeling (CB-
SEM) and PLS-SEM and when to use each.

 Social science researchers have been using statistical analysis tools for many years to
extend their ability to develop, explore, and confirm research findings. [p.1]
 Application of first-generation statistical methods, such as factor analysis and
regression analysis, dominated the research landscape through the 1980s. But since
the early 1990s, second-generation methods have expanded rapidly [p.1]
 one of the emerging second-generation tools, referred to as partial least squares
structural equation modeling (PLS-SEM) [p.2]

WHAT IS STRUCTURAL EQUATION MODELING?

 Statistical analysis has been an essential tool for social science researchers for more
than a century. [p.2]
 To comprehend more complex relationships associated with current research directions
in the social science disciplines, it is increasingly necessary to apply more
sophisticated multivariate data analysis methods. [p.2]
 Multivariate analysis involves the application of statistical methods that
simultaneously analyze multiple variables. [p.2]
 The variables typically represent measurements associated with individuals,
companies, events, activities, situations, and so forth. The measurements are often
obtained from surveys or observations that are used to collect primary data, but they
may also be obtained from databases consisting of secondary data. [p.2]
 Exhibit 1.1 displays some of the major types of statistical methods associated with
multivariate data analysis. [p.2]

 The statistical methods often used by social scientists are typically called first-
generation techniques (Fornell, 1982, 1987). [p.3]
 These techniques, shown in the upper part of Exhibit 1.1, include regression-based
approaches, such as multiple regression, logistic regression, and analysis of variance,
but also techniques, such as exploratory and confirmatory factor analysis, cluster
analysis, and multidimensional scaling. When applied to a research question, these
methods can be used to either confirm a priori established theories or identify data
patterns and relationships. Specifically, they are confirmatory when testing the
hypotheses of existing theories and concepts, and exploratory when they search for
patterns in the data in case there is no or only little prior knowledge on how the
variables are related. [p.3]
 It is important to note that the distinction between confirmatory and exploratory is not
always as clear-cut as it seems.
For example, when running a regression analysis, researchers usually select the
dependent and independent variables based on established theories and concepts.
The goal of the regression analysis is then to test these theories and concepts.
However, the technique can also be used to explore whether additional independent
variables prove valuable for extending the concept being tested. The findings
typically focus first on which independent variables are statistically significant
predictors of the single dependent variable (more confirmatory) and then on which
2
 independent variables are, relatively speaking, better predictors of the dependent
variable (more exploratory). [p.3]
 First-generation techniques have been widely applied by social science researchers,
and they have significantly shaped the way we see the world today. In particular,
methods such as multiple regression, logistic regression, and analysis of variance
have been used to empirically test relationships among variables. [p.3]
 what is common to these techniques is that they share three limitations, namely [p.3]
(1) the postulation of a simple model structure,
multiple regression analysis and its extensions postulate a simple model
structure involving one layer of dependent and independent variables. Causal
chains such as “A leads to B leads to C” or more complex nomological
networks involving a great number of intervening variables can only be
estimated piecewise rather than simultaneously, which can have severe
consequences for the results’ quality (Sarstedt, Hair, Nitzl, Ringle, & Howard,
2020).
(2) the assumption that all variables can be considered observable,
regression-type methods are restricted to processing observable variables,
such as age or sales (in units or dollars). Theoretical concepts, which are
“abstract, unobservable properties or attributes of a social unit or entity”
(Bagozzi & Philipps, 1982, p. 465), can only be considered after prior stand-
alone validation by means of, for example, a confirmatory factor analysis.
(3) the conjecture that all variables are measured without error (Haenlein
& Kaplan, 2004).
One has to bear in mind that each observation of the real world is
accompanied by a certain measurement error, which can be systematic or
random (Chapter 4). First-generation techniques are, strictly speaking, only
applicable when there is neither systematic, nor random error. This situation
is, however, rarely encountered in reality, particularly when the aim is to
estimate relationships among measures of theoretical concepts.
 To overcome these limitations, researchers have increasingly been turning to
second-generation techniques. These methods, referred to as structural equation

3
 modeling (SEM), enable researchers to simultaneously model and estimate complex
relationships among multiple dependent and independent variables. [p. 4]
 The concepts under consideration are typically unobservable and measured
indirectly by multiple indicator variables. In estimating the relationships, SEM
accounts for measurement error in observed variables. As a result, the method
obtains a more precise measurement of the theoretical concepts of interest (Cole &
Preacher, 2014). [p. 4]
 There are two types of SEM methods: covariance-based structural equation
modeling (CB-SEM) and partial least squares structural equation modeling (PLS-
SEM; also called PLS path modeling)
 CB-SEM is primarily used to confirm (or reject) theories (i.e., a set of systematic
relationships between multiple variables that can be tested empirically). It does this
by determining how well a proposed theoretical model can estimate the covariance
matrix for a sample data set.
 In contrast, PLS has been introduced as a “causal-predictive” approach to SEM
(Jöreskog & Wold, 1982, p. 270), which focuses on explaining the variance in the
model’s dependent variables (Chin et al., 2020). PLS-SEM is evolving rapidly as a
statistical modeling technique.1

CONSIDERATIONS IN USING STRUCTURAL EQUATION MODELING

 Depending on the underlying research question and the empirical data available,
researchers must select an appropriate multivariate analysis method. [p. 6]
 several considerations are necessary in deciding to use multivariate analysis,
particularly SEM. Among the most important are the following five elements:
composite variables, (2) measurement, (3) measurement scales, (4) coding, and (5)
data distributions.

1
Note: In recent years, a growing number of follow-up textbooks (e.g., Garson, 2016;
Henseler, 2020; Ramayah, Cheah, Chuah, Ting, & Memon, 2016; Wong, 2019) and
edited books on the method (e.g., Avkiran & Ringle, 2018; Esposito Vinzi, Chin,
Henseler, & Wang, 2010; Latan & Noonan, 2017) have been published, which
helped to further popularize PLS-SEM. [p. 6]

4
 1. Composite Variables
 A composite variable (also referred to as a variate) is a linear combination of several
variables that are chosen based on the research problem at hand (Hair, Black,
Babin, & Anderson, 2019).
 The process for combining the variables involves calculating a set of weights,
multiplying the weights (e.g., w1 and w2 ) times the associated data observations for
the variables (e.g., x1 and x2 ), and summing them.
 The mathematical formula for this linear combination with five vari- ables is shown
as follows (note that the composite value can be calculated for any number of
variables):
Composite value = w1 · x1 + w2 · x2 + . . . + w5 · x5 ,
where x stands for the individual variables and w represents the weights. All x vari-
ables (e.g., questions in a questionnaire) have responses from many respondents
that can be arranged in a data matrix.

2. Measurement
 Measurement is a fundamental concept in conducting social science research
 Measurement is the process of assigning numbers to a variable based on a set of
rules (Hair, Page, & Brunsveld, 2020). The rules are used to assign the numbers to
the variable in a way that accurately represent the variable.
 With some variables, the rules are easy to follow, while with other variables, the
rules are much more difficult to apply. For example, if the variable is gender, then it
is easy to assign a 1 for females and a 0 for males. Similarly, if the variable is age or
height, it is again easy to assign a number. But what if the variable is satisfaction or
trust? Measurement in these situations is much more difficult because the
phenomenon that is supposed to be measured is abstract, complex, and not directly
observable. We therefore talk about the measurement of latent variables or
constructs
 We cannot directly measure abstract concepts such as satisfaction or trust.
However, we can measure indicators of what we have agreed to call satisfaction or
trust, for example, in a brand, product, or company. Specifically, when concepts are
difficult to measure, one approach is to measure them indirectly by using a set of
directly observable and measurable indicators (also called items or manifest
variables). Each indicator represents a single separate aspect of a larger abstract
concept.

5
For example, if the concept is restaurant satisfaction, then the several indicators that could
be used to measure this might be the following:
1. The taste of the food was excellent. 2. The speed of service met my expectations. 3.
The waitstaff was very knowledgeable about the menu items. 4. The background
music in the restaurant was pleasant. 5. The meal was a good value compared with
the price.
By combining several indicators to form a scale, we can indirectly measure the overall
concept of restaurant satisfaction.

Measurement scales

Coding

Data distribution

6
Research design: Qualitative, quantitative, and mixed methods
approaches
By John W Creswell. -3rd ed.

Chapter 5: The introduction

After having decided on a qualitative, quantitative, or mixed methods approach and after
conducting a preliminary literature review and deciding on a format for a proposal, the next
step in the process is to design or plan the study. The process of organizing and writing out
ideas begins, starting with designing an introduction to a proposal.

Five components of writing a good introduction: (a) establishing the problem leading to the
study, (b) reviewing the literature about the problem, (c) identifying deficiencies in the
literature about the problem, (d) targeting an audience and noting the significance of the
problem for this audience, and (e) identifying the purpose of the proposed study. These
components comprise a social science deficiency model of writing an introduction
because a major component of the introduction is to set forth the deficiencies in past
research.

The Importance of Introductions

An introduction is the first passage in a journal article, dissertation, or scholarly research
study. It sets the stage for the entire study. As Wilkinson (1991) mentions,

The introduction is the part of the paper that provides readers with the background
information for the research reported in the paper. Its purpose is to establish a framework
for the research, so that readers can understand how it is related to other research. (p. 96)

The introduction establishes the issue or concern leading to the research by conveying
information about a problem. Because it is the initial passage in a study or proposal, special
care must be given to writing it. The introduction needs to create reader interest in the
topic, establish the problem that leads to the study, place the study within the larger context
of the scholarly literature, and reach out to a specific audience. All of this is achieved in a
concise section of a few pages. Because of the messages they must convey and the
limited space allowed, introductions are challenging to write and understand.

7
Model for Introduction- Deficiency model
The deficiencies model of an introduction is a general template for writing a good
introduction. It is a popular approach used in the social sciences, and once its structure is
elucidated, the reader will find it appearing repeatedly in many published research studies.
It consists of five parts, and a separate paragraph can be devoted to each part, for an
introduction of about two pages in length:
1. The research problem
2. Studies that have addressed the problem
3. Deficiencies in the studies
4. The significance of the study for particular audiences
5. The purpose statement

An Illustration
Before a review of each part, here is an excellent example from a quantitative study
published by Terenzini, Cabrera, Colbeck, Bjorklund, and Parente (2001) in The Journal
of Higher Education and titled “Racial and Ethnic Diversity in the Classroom” (reprinted
with permission). Following each major section in the introduction, I briefly highlight the
component being addressed.

Since passage of the Civil Rights Act of 1964 and the Higher Education Act of 1965,
America’s colleges and universities have struggled to increase the racial and ethnic
diversity of their students and faculty members, and “affirmative action” has become the
policy-of-choice to achieve that heterogeneity. [Authors state the narrative hook.] These
policies, however, are now at the center of an intense national debate. The current legal
foundation for affirmative action policies rests on the 1978 Regents of the University of
California v. Bakke case, in which Justice William Powell argued that race could be
considered among the factors on which admissions decisions were based. More recently,
however, the U.S. Court of Appeals for the Fifth Circuit, in the 1996 Hopwood v. State of
Texas case, found Powell’s argument wanting. Court decisions turning affirmative action
policies aside have been accompanied by state referenda, legislation, and related actions
banning or sharply reducing race-sensitive admissions or hiring in California, Florida,
Louisiana, Maine, Massachusetts, Michigan, Mississippi, New Hampshire, Rhode Island,
and Puerto Rico (Healy, 1998a, 1998b, 1999).

In response, educators and others have advanced educational arguments supporting

affirmative action, claiming that a diverse student body is more educationally effective
than a
more homogeneous one. Harvard University President Neil Rudenstine claims that the
“fundamental rationale for student diversity in higher education [is] its educational value”
(Rudenstine, 1999, p. 1). Lee Bollinger, Rudenstine’s counterpart at the University of
Michigan, has asserted, “A classroom that does not have a significant representation
from members of different races produces an impoverished discussion” (Schmidt, 1998,
8
p. A32). These two presidents are not alone in their beliefs. A statement published by the
Association of American Universities and endorsed by the presidents of 62 research
universities stated: “We speak first and foremost as educators. We believe that our
students benefit significantly from education that takes place within a diverse setting”
(“On the Importance of Diversity in University Admissions,” The New York Times, April
24, 1997, p. A27). [Authors identify the research problem.]

Studies of the impact of diversity on student educational outcomes tend to approach the
ways students encounter “diversity” in any of three ways. A small group of studies treat
students’ contacts with “diversity” largely as a function of the numerical or proportional
racial/ethnic or gender mix of students on a campus (e.g., Chang, 1996, 1999a; Kanter,
1977; Sax, 1996) . . . . A second considerably larger set of studies take some modicum of
structural diversity as a given and operationalizes students’ encounters with diversity
using the frequency or nature of their reported interactions with peers who are
racially/ethnically different from themselves . . . . A third set of studies examines
institutionally structured and purposeful programmatic efforts to help students engage
racial/ethnic and/or gender “diversity” in the form of both ideas and people.

These various approaches have been used to examine the effects of diversity on a broad
array of student educational outcomes. The evidence is almost uniformly consistent in
indicating that students in a racial/ethnically or gender-diverse community, or engaged in
a diversity-related activity, reap a wide array of positive educational benefits. [Authors
mention studies that have addressed the problem.]

Only a relative handful of studies (e.g., Chang, 1996, 1999a; Sax, 1996) have specifically
examined whether the racial/ethnic or gender composition of the students on a campus,
in an academic major, or in a classroom (i.e., structural diversity) has the educational
benefits claimed . . . . Whether the degree of racial diversity of a campus or classroom
has a direct effect on learning outcomes, however, remains an open question.
[Deficiencies in the studies are noted.]

The scarcity of information on the educational benefits of the structural diversity on a

campus or in its classrooms is regrettable because it is the sort of evidence the courts
appear to be requiring if they are to support race-sensitive admissions policies.
[Importance of the study for an audience mentioned.]

This study attempted to contribute to the knowledge base by exploring the influence of
structural diversity in the classroom on students’ development of academic and
intellectual skills. . . . This study examines both the direct effect of classroom diversity on
academic/intellectual outcomes and whether any effects of classroom diversity may be
moderated by the extent to which active and collaborative instructional approaches are
used in the course. [Purpose of the study is identified.] (pp. 510-512, reprinted by
9
 permission of The Journal of Higher Education)

Research Problem
A research problem is a problem or issue that leads to the need for a study. It can originate
from many potential sources. It might spring from an experience researchers have had in
their personal lives or workplaces. It may come from an extensive debate that has
appeared in the literature. It might develop from policy debates in government or among top
executives. The sources of research problems are often multiple. Identifying and stating the
research problem that underlies a study is not easy. Unfortunately, too many authors do not
clearly identify the research problem, leaving the reader to decide the importance of the
issue. When the problem is not clear, it is difficult to understand the significance of the
research. Furthermore, the research problem is often confused with the research questions
— those questions that the investigator would like answered in order to understand or
explain the problem.

The authors announce a problem and they justify why it needs to be studied. The type of
problem presented in an introduction will vary depending on the approach. In a quantitative
project, the problem is best addressed by understanding what factors or variables influence
an outcome. Example, researcher may need to understand the high divorce rate among
married couples (a problem) and examine whether financial issues contribute to divorce.

in quantitative introductions, researchers sometimes advance a theory to test, and they will
incorporate substantial reviews of the literature to identify research questions that need to
be answered. A quantitative introduction may be written from the impersonal point of view
and in the past tense, to convey objectivity.

In the Terenzini et al. (2001) article, the first sentence accomplishes both primary
objectives for an introduction: piquing interest in the study and conveying a distinct
research problem or issue. What effect did this sentence have? Would it entice a reader to
read on? Was it pitched at a level so that a wide audience could understand it? These
questions are important for opening sentences, and they are called a narrative hook, a term
drawn from English composition, meaning words that serve to draw, engage, or hook the
reader into the study.

Example of narrative hook - Quantitative experimental study

“There is a large body of literature that studies the cartographic line (a recent summary
article is Buttenfield 1985), and generalization of cartographic lines (McMaster 1987).”
(Carstensen, 1989, p. 181)

10
11