The Role of Theory in Structural

Equation Modelling Research

Presentation for the International Symposium on Applied SEM,
Sarawak, 13 October, 2017
Peter Seddon
 Message for today (2 slides)
1. Structural equation modelling (SEM) is a tool designed to help in producing better
theory, mainly in the social sciences.
Tool = SEM Product of tool use = better theory

e.g.,

https://bothsidesofthetable.com/when-you-re-a-
hammer-everything-looks-like-a-nail-6ebaf6addf64 https://www.homebuilding.co.uk/how-to-choose-a-timber-frame-system/

2. Rigorous use of SEM is important because we dont want to draw unsound

conclusions but pursuit of theory remains the main game.
3. No single study takes us very far along the road to confidence that a theory is valid.
Rather, research is an activity undertaken by a community of scholars, over time.
Todays Arg: The Role of Theory in SEM
1. Theories are products of human intellect/imagination that are Human
intended to explain phenomena of interest. knowledge
2. Valid theories are those that have a high correspondence with
reality. Therefore, theories must be tested (again and again).
3. Support for the validity of theories grows as supportive Theories:
evidence accumulates from many studies. Explanations of
phenomena of
interest.
4. Most theories have been built and/or tested without using SEM, SEM
e.g., Einsteins theory of relativity. Valid theories
5. SEM is most useful for quantitative, social-science research that
uses latent variables and ordinal/Likert-scale type indicators.
6. There are two types of testing: abductive testing for theory
building, and independent-data testing for true theory testing.
7. As with all empirical research, generalizing SEM results requires
detailed logical argument, not more statistics. Theories where SEM has
been/is useful
 0
Plan for this presentation
1 My credentials
2 A researchers life
3 What is a theory?
4 Theories evolve: They are never true or proven true.
5 Two types of testing: Abductive and Independent-data testing
6 What does testing really tell us?
 My Credentials 1
1. I completed my PhD in 1990, aged 43. It was a design science PhD.
2. During 1992-2005, I conducted a series of quantitative studies (Amos,
PLS) (see next slide).
3. From about 2000, I started to use qualitative analysis. Today, I find
interviews with senior managers the most valuable source of data.
4. From 2005-8, I was a Senior Editor for MIS Quarterly (MISQ). Making
accept/reject decisions for MISQ required a quite detailed understanding
of both quantitative and qualitative research.
5. In 2014, I retired as a Professor at The University of Melbourne.
6. For the last few years, post retirement, I have been trying to answer to
my own satisfaction: Where does knowledge come from?
 1996

All reflective indicators for Amos

CB-SEM (above)

Respondents were 100 users of a

departmental accounting system
at The University of Melbourne
 A Researchers Life 2
This entire research process Analyze Collect
is underpinned by ongoing Papers Papers
personal sensemaking. This, 80-90%
in turn, is deeply informed by
the researchers personal Literature of time
knowledge, experience, Reflect Plan
tastes, and interests.

Conceptual
Research
Framework Write Paper
Questions
& Theory

Reflect Empirical Plan

10-20%
Analyze Collect
Data Data
 3

What is a theory?

Answers: next 9 slides

Theories are Only One Type of Knowledge
Human
Procedural knowledge: Ride a bike; drive a car; do book- knowledge
keeping;
Traditional knowledge: how to survive in the desert;
traditional medicine; superstition?? Theories:
Explanations of
Experiential/tacit knowledge: assess safety risk in a fire; phenomena of
how to design a Stradivarius violin; interest.
Common-sense knowledge: drive more carefully on wet
roads;
Scientific/propositional/theoretical knowledge:
Explanations of how and why the world works as it does,
e.g., Massive bodies attract one another.

Aside: Nagel (1979) The Structure of Science, lists four types

of explanation (pp.20-26). I wont explore those here.
 Poppers Worlds 1, 2, and 3,
(in Popper, Unended Quest, 1974, pp.210-230)
World 1 The world of physical objects
World 2 The world of subjective experiences (such as fear, and
thought processes)
World 3 The world of statements; products of the human mind
(p.217), e.g., a fairy story

Theories exist in Poppers World 3.

They are real, but they are not physical objects.
Theories may be about Worlds 1-3.
 Eight Definitions of Theory
Author(s) Definition
Van de Ven Good theory goes beyond establishing empirically observed patterns, that is, it tries to explain
(1989, p.487) what caused them.

Whetten (1989, To summarize thus far: What and How describe; only Why explains. What and How provide a
p.491) framework for interpreting patterns, or discrepancies, in our empirical observations. This is an
important distinction because data, whether qualitative or quantitative, characterize; theory
supplies the explanation for the characteristics.

Gioia and Pitre We broadly define theory as any coherent description or explanation of observed or
(1990, p.517) experienced phenomena. This atypically broad definition is necessary to encompass the wide
scope of theoretical representations found in the alternative paradigms.

Bacharach In more detailed terms, a theory may be viewed as a system of constructs and variables in
(1989, p.498) which the constructs are related to each other by propositions and the variables are related to
each other by hypotheses. The whole system is bounded by the theorists assumptions as
indicated by Figure 1. (next slide)
Bacharachs (1989) Figure 1

We want to
know about this,

but we can
only test here.
SEM may be
useful here
 Eight Definitions of Theory
Author(s) Definition
Sutton and Staw Unfortunately, the literature on theory building can leave a reader more rather than less confused
(1995, p.371) about how to write a paper that contains strong theory (Freese 1980). There is lack of agreement
about whether a model and a theory can be distinguished, whether a typology is properly labeled a
theory or not, and whether falsifiability is a prerequisite for the very existence of a theory. As Merton
(1967: 39) put it:
Like so many words that are bandied about, the word theory threatens to become meaningless.
Because its referents are so diverseincluding everything from minor working hypotheses, through
comprehensive but vague and unordered speculations, to axiomatic systems of thoughtuse of the
word often obscures rather than creates understanding.
Gregor (2006, Thus, the word theory will be used here rather broadly to encompass what might be termed
p.614) elsewhere conjectures, models, frameworks, or body of knowledge.

Suddaby (2014, theory is simply a way of imposing conceptual order on the empirical complexity of the phenomenal
p.407) world. Ultimately, theories reflect, in highly abstract terms, the organization of a disciplines
knowledge base. However, theory does much more than simply abstract and organize knowledge. It
also signals the values upon which that knowledge is built.

Honderich (1995, Hypothetico-deductive method. A theory in science is a general statement (or hypothesis) from which
p. 386). particular inferences may be deduced. Observations can then be seen as confirming or falsifying
the hypothesis.
 My conclusions about definitions of theory
1. There is no single best definition of theory. Many powerful people have
their own definitions, and they are different. This means that when we
use the term theory, we have to define what we mean. (And if someone
uses the term without definition, you have to ask them what they mean.)
2. For me: A theory is a knowledge claim that provides a plausible, well-
justified, and clearly-bounded explanation of how some part of the world
works (more details next slide).
This entire research process Analyze Collect
3. In addition to my Researchers Life is underpinned by ongoing
personal sensemaking. This,
in turn, is deeply informed by
Papers Papers

the researchers personal Literature

diagram, is helpful to think of what knowledge, experience,

tastes, and interests.
Reflect Plan

researchers do as theory building and Research

Questions
Conceptual
Framework
& Theory
Write Paper

theory testing. Reflect Empirical Plan

Analyze Collect
Data Data
 More on my definition of theory
1. Theories are products of human intellect/imagination that are intended to explain how
some part of the world works. They exist in Poppers (1974) World 3.
2. With theories, we seek to answer the question Why?.
3. Theoretical knowledge claims are framed in terms of concepts and relationships
between them, e.g., allergies are due to lack of exposure to bacteria in childhood.
4. Propositions may be logically deduced from theories (without reference to the
empirical world). (Dubin, 1969, Ch.7)
5. Hypotheses are propositions for which measures/indicators have been chosen. Thus,
hypotheses are testable propositions based on a theory. (Dubin, 1969, Ch.9)
6. The validity (or probable truth) of theories is demonstrated by assessing their fit with
(perceived) empirical reality. Such fit assessment is called testing.
7. Theories are never absolutely true; they are always subject to revision. (It is therefore
never valid to talk of empirically proving a theory to be true.)
8. Theories are normally only valid (or believed true) within certain limits, called
boundary conditions (e.g., Newtonian mechanics ceases to be true as relative
velocities of objects approach the speed of light).
 Abstract: Jasmine Yeap, Ramayah, and Yapp (this conference)

As new entrants to the Malaysian market, Japanese household and lifestyle product retail
brands have successfully gained a strong following among consumers of all ages and
ethnicities within a short period of time. This study was conducted to uncover the branding
elements which are meaningful to consumers during their experience with such retail brands
and to ascertain how consumers brand evaluations affect their response, feelings and
behaviour towards these brands.
Using the Service Brand Verdict Model (SBVM) as a basis, data was collected from 662
customers of a large Japanese chain of household and lifestyle product stores using a self-
administered survey.
Results from the PLS-SEM analysis revealed that consumers feelings towards the brand,
value-for-money, servicescape, uncontrolled communications (word-of-mouth and publicity)
and merchandise are key components affecting consumers satisfaction towards the brand.
Satisfaction positively affects consumers attitude as well as attachment to the retailers
brand. Consequently, consumers satisfaction, attitude and attachment lead to consumers
loyalty towards the brand.
In sum, the SBVM has been shown to remain robust and valid when applied to evaluating
more recent retailing concept stores. Furthermore, in building brand loyalty among
consumers, aspiring retailers must ensure that their brands appeal to both the consumers
head (mind; utilitarian needs) and the heart (feelings; emotional needs).
Grace and OCass, Service branding: consumer verdicts on
service brands, J of Retailing and Consumer Services, 2005

(= Behavioural intention)

N=256 retail consumers at a high-

traffic shopping mall (p.131).
PLS analysis.
All seven hypotheses were found
to be significant, at p<0.05.

Omitted variables??
1. Proximity to home
2. Ease of parking
3. Loyalty-card factor
Theories evolve: They are never true, 4
or proven true.
Copernicus heliocentric model of the solar
system (1510)
Newtons laws of motion (1687)
Adam Smiths theory that specialization of
labour results in greater productivity, e.g., in a
pin factory (1776)
The Darwin/Wallace theory of evolution
through natural selection (1859)
Wegeners theory of continental drift (1912)
The resource-based theory of the firm
(Barney, 1991) 500 years ago
Example of an evolving theory: Tectonic-plate Theory
Wegener: Why are the soil and flora so
similar in Brazil and the west coast of Afrika?
Answer: Continental-drift theory (1912)

Google Maps
National Geographic Map
of the Atlantic Ocean Floor,
Marie Tharp, 1968
http://www.nationalgeographic.com
.au/science/how-one-brilliant-
woman-mapped-the-ocean-floors-
secrets.aspx
https://en.wikipedia.org/wiki/Plate_tectonics#/media/File:Oceanic.Stripe.Magnetic.Anomalies.Scheme.svg
https://en.wikipedia.org/wiki/Subduction#/media/File:Subduction-en.svg
Cocus islands
Tectonic plate theory is now regarded as a highly valid theory
because it is consistent with so much empirical evidence.

https://en.wikipedia.org/wiki/Ring_of_Fire#/media/File:Pacific_Ring_of_Fire.svg
Support for theory grows as evidence is accumulated
from many studies yet it may take many years for
new, initially heretical, theories to be accepted
Support for tectonic-plate theory is strong because there is support from so many
different types of evidence, collected using so many different types of method.
Today, the accumulated evidence is strongly suggestive that tectonic plate theory
is true.
Yet for many years (1912 1970) the validity of tectonic plate theory was strongly
resisted by leading geologists (Powell, 2015, pp.67-163).
The leading professors, of the worlds leading universities, were all
wrong!
Similarly, because of the risks of errors of inference from any one empirical test of
a theory, conclusions from single studies should be treated with caution. This is
true even for studies that report very small p-values (e.g., p<0.001).

Powell, J.L., Four Revolutions in the Earth Sciences, From Heresy to Truth, NY: Columbia Uni Press, 2015
 Two Types of Testing: Abductive and 5
Independent-data Testing
Abductive test: A test that uses the same data used to build a theory to test
the theory an abductive test of that theory. This is actually a test of the quality
of the theory builders efforts in theory building, not of the theory. (Abductive
testing is testing as part of theory building.)
Abduction means inference to the best explanation (Lipton 2004)
Independent-data test: An independent-data test means that a theorys
correspondence with reality is assessed using data collected independently of
the theory-building process. This tests the validity of the theory. Some studies
use hold-out samples for this purpose.
Independent-data testing is the gold standard in testing. A double-blind,
randomized, placebo-controlled test in medical science is an independent-
data test.
Cf induction!!!
Example of both Abductive and Independent-data Testing:
Venkatesh et al. (2003) UTAUT model p.447
Data used for
building and
testing UTAUT
(Four applications, all
in a corporate setting.)

N = 54+65+58+38 = 215
Theory Building:
Testing various
models (= theories)
from prior
researchers

***
Synthesis: Venkatesh et al.s (2003) UTAUT model diagram suddenly appears,
22 pages into their 54-page paper. It is then followed by 10 pages explaining
and justifying their theory (pp.447-456), and 5 pages of preliminary testing
(pp.457-61). Preliminary testing is abductive testing.

p.446
UTAUT just pops up, like magic, in Venkatesh et al. (2003), p.446.
It is an example of what I meant by saying that:
Theories are products of human intellect/imagination that are
intended to explain how some part of the world works.
Abductive
Testing of
UTAUT (p.462)
(37 pages into their
54-page paper)
Independent-
data test of
UTAUT (p.465)
(40 pages into their
54-page paper)
Venkatesh et al.s Abductive and Independent-data testing
Since Venkatesh et al. (2003) no doubt modified their model (probably
many times) to fit their data, the SEM-based statistics that they present in
their preliminary test (pp.457-61) are only an abductive test of their
model.
Recall:
Abductive test: A test that uses the same data used to build a theory to
test the theory an abductive test of that theory. It is actually a test of the
quality of the theory builders efforts in theory building, not of the theory.
Independent-data test: An independent-data test means that a theorys
correspondence with reality is assessed using data collected independently
of the theory-building process. This tests the validity of the theory. Some
studies use hold-out samples for this purpose.
 6
What does testing really tell us?
In this last part of this presentation (six slides), I want to argue that
there are some characteristics of any empirical study in the social
sciences (not just SEM-based studies) that limit our ability to draw
sound general conclusions from such studies.
These characteristics are:
need to select a sample;
non-response on surveys;
need to chose indicators to represent theoretical constructs;
what does significant (e.g., p<0.05) mean?
need for logical argument to support generalization claims.
Choice of sample

Data used for

building UTAUT

(Four applications,
all in a corporate
setting.)

N = 54+65+58+38 = 215
 Empirical studies also force researchers to decide how to
measure constructs (variables). seven-point Likert scales
 Do these indicators actually capture the intended meaning
of the theoretical constructs? (validity)

Thus, having to choose:

(a) to collect data from some particular place in the world,
and
(b) to use certain indicators and not others,
again limits our ability to make general claims about the
validity (probably truth) of our theories.
 What does significant mean?
Ronald Fisher (1890-1962) was a researcher at an agricultural research station in the UK.
Researchers there gave different treatments to plants in various plots of land.
Fisher analysed data from samples from various plots and devised p-values for deciding
whether a given treatment had a significant effect on output. In other words, he used
findings from data in his samples to make inferences about all the plants in the various
plots.
In 1925, he published his famous book: Statistical Methods for Research Workers
Researchers today use Fishers p-values as a way of identifying important relationships
in their samples. Although p-values are described as inferential statistics, researchers
today rarely think about the population to which their inferences apply. Today, p-
values are used simply as a way of highlighting relationships in a sample that are unlikely
to be due to chance. You dont have to define the population to compute p-values.
For this reason, Venkatesh et al. (2003) made no attempt to specify the population of
relevance in their study. We all just look for the statistics with stars beside them (as I
have done on the previous slides) and assume the p-values are relevant to whatever
population we think the study is about!
How general are Venkatesh et al.s sample-based claims?
Since so many of the hypothesized relationships in UTAUT were
significant, UTAUT appears to be substantially true in the sample
collected.
It is implied in the paper that the UTAUT theory is also likely to be true
for all users of IT applications, worldwide (= the population of interest?).
It takes a big logical leap to believe this. Really, much more empirical
testing is required before one may have confidence in this claim.
So the role of SEM in the UTAUT paper is to provide dual-sample, single-
type-of-IT-use (in business), specific-measure support for the theory
proposed in the paper.
The analyses provided are an important step down the road to assessing
the validity of the theory, but only a first step down a very long road.
The argument for generalizing from findings
from this dataset goes something like this:
IT users around the world feel and think about technology in much the
same way as the business people in who agreed to participate in Venkatesh
et al.s (2003) surveys.
(Critical realism calls this motivational force a mechanism.)
If this is true, Venkatesh et al.s sample is probably representative of all
business IT users all around the world.
Therefore, UTAUT is probably valid for all IT users around the world.

Each step above involves some big logical leaps of faith (probably x 2).
But this is what we are implicitly doing with most empirical studies.
Note that generalization arguments such as that above are not based on
statistics or p-values. They are just logical arguments.
 Wrapup: Message for today (2 slides)
1. Structural equation modelling (SEM) is a tool designed to help in producing better
theory, mainly in the social sciences.
Tool = SEM Product of tool use = better theory

e.g.,

https://bothsidesofthetable.com/when-you-re-a-
hammer-everything-looks-like-a-nail-6ebaf6addf64 https://www.homebuilding.co.uk/how-to-choose-a-timber-frame-system/

2. Rigorous use of SEM is important because we dont want to draw unsound

conclusions but pursuit of theory remains the main game.
3. No single study takes us very far along the road to confidence that a theory is valid.
Rather, research is an activity undertaken by a community of scholars, over time.
Todays Arg: The Role of Theory in SEM
1. Theories are products of human intellect/imagination that are Human
intended to explain phenomena of interest. knowledge
2. Valid theories are those that have a high correspondence with
reality. Therefore, theories must be tested (again and again).
3. Support for the validity of theories grows as supportive Theories:
evidence accumulates from many studies. Explanations of
phenomena of
interest.
4. Most theories have been built and/or tested without using SEM, SEM
e.g., Einsteins theory of relativity. Valid theories
5. SEM is most useful for quantitative, social-science research that
uses latent variables and ordinal/Likert-scale type indicators.
6. There are two types of testing: abductive testing for theory
building, and independent-data testing for true theory testing.
7. As with all empirical research, generalizing SEM results requires
detailed logical argument, not more statistics. Theories where SEM has
been/is useful