Sunteți pe pagina 1din 11

A CONCEPTUAL UNDERSTANDING OF REQUIREMENTS

FOR THEORY-BUILDING RESEARCH: GUIDELINES FOR


SCIENTIFIC THEORY BUILDING
JOHN G. WACKER
Arizona State University

Business academics have focused their attention on empirical investigation of


programs effect on organizational competitive performance. These studies pri-
marily emphasize theory building. With the many denitions of theory, aca-
demics are not certain whethertheir researchpapers meet the specic
requirements for theory development required by the academic eld of the
philosophy of science. Certainly, supply chain academics generally believe that
their academic articles fulll the requirements of theory building. Although
many of these articles do have elements of theory, more focus is needed on the
specic requirements of theory to assure that academic research is good theory
building. The primary purpose of this research paper is to logically develop a set
of guidelines to assist empirical researchers to assure that their studies fulll the
requirements of good theory based upon traditional scientic theory building.
By fullling the requirements of good theory, researchers will develop studies
that will have a lasting impact on their academic eld. To achieve a lasting im-
pact on an academic eld, it is necessary to follow a logical plan. This article
provides a plan for logical guidelines for developing an understanding of how
and why good theory building is achieved. This article logically develops a
formal conceptual denition of theory along with its related properties to un-
derstand these guidelines. Next, it analyzes the requirements of theory, good
theory, and their properties. These guidelines are included in the existing phi-
losophy of science publications. However, this article consolidates these sources
and logically explains why these guidelines are needed. In the conclusion, the
guidelines are summarized to serve as a summary checklist for supply chain
researchers to use for ensuring their articles will be recognized as a contribution
to the academic eld. So in that sense, this article does not develop a revolu-
tionary new insight into theory-building empirical articles, but rather integrates
diverse traditional philosophy of science requirements into a much simpler set
of guidelines. Through logical development of these guidelines, researchers will
understand the structure of theory and how to ensure their studies can be
modied to have a lasting impact on the eld of supply chain management.

Keywords:supplychainmanagement;empiricaltheorytesting

Like all invited papers and invited notes, the original version of this INTRODUCTION
manuscript underwent a double-blind review process.
Over the last several decades, there has been a dramatic
Acknowledgment: The author wishes to thank many of his close
friends who have commented and discussed these materials over
increase in empirical studies testing theory (Swamidass
many years. The author especially wishes to thank Lawrence Fred- 1991). The use of statistical techniques has been an im-
portant contribution to understanding the wide variety of
endall (Clemson University) for his careful reading of this study and
many corrections that were important for the development of the practices/programs effects on international competi-
study. Additionally, the author wishes to thank his many close ad- tiveness. Most of these articles focus on the testing and
visors. Most notable of these advisors are R. Kenneth Teas (Iowa
development of business theory. In general, academics
State University), Shelby Hunt (Texas Tech University), George Mar-
icoulides (California State University-Fullerton) and Daniel Samsonhave a basic understanding of what theory is. However,
(University of Melbourne). Naturally, any errors, omissions, logicthis understanding is not precise enough to explain
faults, and so forth are entirely the fault of the author. exactly how and why theory building is driven by specic

July2008 5
JournalofSupplyChainManagement

requirements. Without this understanding of theory re- topics is based in the formal study of the theory of the-
quirements and how it relates to empirical research, itories: is the philosophy of science. Somewhat unfortu-
difcult to assess if a statistical set of relationships nately,
are or the philosophy of science literature is not easy to
are not a formal theory. This issue was addressed in understand and even more difcult to apply. Yet under-
marketing in books by Shelby Hunt (1991), in manage- standing the basic properties of theory and good theory
ment in the AcademyofManagementReview s special issue is essential for the conceptualization of theory and the
(1989) and several subsequent articles, and in operations statistical tests of theory. Consequently, one purpose of
management in the Journal of Operations Managements this article is to translate the philosophy of science lan-
(1998) special issue on theory. Although these studies guage into specic, concise guidelines for testing theory.
provided insightful understanding of what constitutes aIn order to achieve this goal, this study claries the
theory-building article, they did not provide a concise, denitions of what is meant by theory conceptualization
logical plan for authors to ensure their articles follow and demonstrates how theory can be evaluated. There
classical guidelines for developing theory-building re-are three different levels of detail for examining theory.
search. This article presents the logic behind theory- These are Theory, Good Theory, and Guidelines for
building articles and why specic properties are necessary Good Theory.Theory is used to determine if any set of
for the article to be considered an important contribution relationships (hypotheses) are really a theory or just
to the academic eld. So in one sense, the article is lay not conjectures: Do these relationships have denitions,
presenting new astounding conclusions but rather is domains, and predictions? Good theory determines
presenting an integrated framework for understanding if a theory can be adequately tested. And guidelines
how to develop theory-building articles. are used to apply specic requirements to elements of
Although outside of academia there is not a wide- theory.
spread appreciation of theory (see Shubik 1987), to Although it is possible to give statistical examples of
academics the need for theory is paramount to under- how each of the guidelines affect empirical estimates, the
standing how theories can be applied (Poole and Van length
de of the discussion of those examples require: (A) a
Ven 1989; Van De Ven 1989; Hunt 1991; Klein, Tosi, and complete literature justication, (B) a development of
Canella 1999, and numerous others). Consequently, the denitions, (C) an internally consistent conceptual
there is a need to increase the precision of theory- model, (D) a specication of the sample, (E) a complete
building procedures for developing theories that are statistical explanation, and (F) a complete substantive
important contributions to the academic eld. Yet, there explanation. In short, each one of the guidelines would
is a need to clearly understand what a theory precisely require
is a separate full-length paper to empirically dem-
and what properties are essential for making a theoryonstrate how statistics support or question any theory.
useful to the pragmatic managerial world. Consequently, this study presents logical insights to these
The literature on the philosophy of science is well over problems so interested readers may apply them to their
2,500 years old. Many of the critical thoughts have own research. This paper will leave it to future research to
evolved over those millennia. Upon reading the philos- demonstrate how the guidelines for good theory devel-
ophy of science literature, it is difcult to discern to opment lead to specic statistical tests.
whom the original ideas and thoughts should be attrib- The remainder of this paper is organized as follows:
uted. Many philosophers of science did not cite theFirst, it develops denitions of theory and good theory.
original authors of these ideas and thoughts. Conse-Then it gives good theory formation guidelines based
quently, attributing thoughts to the original authors isupon the properties of good theory.
not possible. This article apologizes for any omission of
citations that may have addressed these theory-building
issues. WHAT IS THEORY?
To begin, there is a difference between a lay under- The foundation of understanding of all theory is based
standing and an academic understanding of precisely on the philosophy of science where theory is developed
what comprises a theory (most recently seen in Popper following strict rules of logic. In the philosophy of sci-
1957; Kaplan 1964; Bunge 1967; Hempel 1970; Hunt ence, the same rules apply for all theory in all scientic
1991; and numerous others). The purpose of this studyelds. is The classical philosophical properties of theory
to address key issues for developing theory. There are two to determine exactly what a theory is. In order to
apply
important issues: (a) theorys conceptualization and understand these guidelines, a formal conceptual de-
(b) theorys effect on conducting empirical tests. Becausenition of theory is necessary. By denition, a theory is an
the conceptualization of theory drives statistical estima-explained set of conceptual relationships. To determine if
tion, it should precede the statistical tests. a theory is actually a theory, there have to be measures for
To understand theory conceptualization, it is necessary what is meant by an explained set of conceptual rela-
to understand precisely what theory is and to understand tionships. These measures are the called properties of
precisely what good theory is. The literature on these theories and they are used for determining if any set of

6 Volume44,Number3
AConceptualUnderstandingofRequirementsforTheory-BuildingResearch

FIGURE1
TheRelationshipBetweenTheoryandItsProperties

Properties of All
Theory
Who and What
Denitions

Denition of theory:
When and Where
It is an explained
set
Domain
of conceptual
relationships.
How and Why:
Relationships

Should, Could, Would


Predictions

conceptual relationships is actually a theory (Bunge these properties of theory can be met and the theory can
1967). still not be what philosophers of science call a good
Figure 1 illustrates the relationship between the formal
theory. Good theory has additional restrictions that
conceptual denition of theory and its properties. The indicate faults in any theory that may indicate a poor
key terms in the formal conceptual denition of theorytheory. Precisely what constitutes a good theory is the
are an explained set of conceptual relationships. To next
be sections topic.
explained, a theory must answer all the common ques-
tions: Who? What? When? Where? How? Why? and
precisely what Would, Should, or Could Happen? (if the ANY THEORY VS. A GOOD THEORY
previous questionsare answered.)These questions Philosophers of science differentiate any theory from a
comprise four basic properties of the formal conceptual good theory. Good theory has more restrictive prop-
denition of theory: Denitions (Who? and What?), erties that further limit any theory from being a good
Domain (When? and Where?), Relationships (How? and theory. To understand these properties, the formal con-
Why?), and Predictions (Would? Should? and Could?). ceptual denition of good theory must be understood.
Further, denitions explain the terms used in the theory,A good theory is a fully explained set of conceptual
the domain is when and where the theory applies, rela- relationships used for empirical testing. The rst key
tionships explain how and why the relationships exist,terms are fully explained. Fully explained means that
and what should, could, and would is predicted by thenot just any set of conceptual relationships will serve as
theory (Bacharach 1989; Whetten 1989). Without any good theory because many theories are not fully ex-
one of these properties, any conjecture, inference, sup- plained. The properties of good theory are more fully
position, hypothesis, or set of hypotheses, is just not aexplained in Bunge (1967), Hunt (1991), Wacker (1998)
theory. Calling any set of relationships a theory does not
and the aforementioned special issuesAcademy
of of
make it theory. Management Review (1989) and Journal of Operations
There is an important pragmatic point concerning Management (1998).
theory: should any academic article be published that Figure 2 illustrates the relationships of all theorys
does not have any theory in it? The answer to that properties and the restrictions on those properties for
question can be phrased another way. Should an aca- the theory to be good theory. These are denitions
demic article be published that does not answer the tra- (conservatism, uniqueness, and parsimony), domain
ditional questions? Who? What? When? Where? How? (generalizability, abstraction), relationships (fecundity,
Why? and what Would? Should? Could happen? If so, internal consistency, statistical parsimony), and predic-
which questions should not be answered? It is difculttions
to (refutability). To review these properties of good
imagine any practical application that does not requiretheory, this study will review them one by one. However,
answering all those questions. It is even more difcultalthough
to the properties of good theory are associated
imagine any pragmatic manager who would not want with specic properties of theory, in many instances these
those questions answered before they apply any tech-good theory properties may be associated with other
nique (practice, program, etc.) to their organization. Yet,
theory properties as well. For simplicitys sake, each

July2008 7
JournalofSupplyChainManagement

FIGURE2
TheRelationshipofGoodTheorytoItsProperties

Properties of All Good Theory s


Theory Properties

 Conservatism
 Theory Parsimony
Denitions  Uniqueness

Denition of good  Generalizability


Domain  Abstractness
theory:
A fully explained set of
conceptual relationships  Fecundity
used for empirical
investigations. Relationships  Internal Consistency
 Statistical Parsimony
 Substantive vs. Statistical
Signicance

Predictions  Falsiability

good theory property will only be addressed here asprecise enough for scientic purposes. This articial
limitations on a theory property. language requires all theory to precisely dene its terms
An important caveat! A good theory may not be a (see Bollen 1989; Bunge 1967; Hunt 1991; Teas and
true theory. It may be that all the above good theory
Palan 1997; and many others). Unfortunately in many
criteria are met and the good theory is just plain
academic
wrong. articles, many researchers use ill-dened and/or
Yet, a wrong good theory will be identied early. On assume previously ill-dened concepts are good de-
the other hand bad theory can never be disproved nitions. The previous denitions are precise enough
because there are so many conceptual loopholes assumption
that leads to ill-dened concepts being assumed
would never allow the theory to be disproved. The point to be adequate. Using ill-dened concepts prevents the-
is that science advances through good theory and badory from being good theory. All concepts need to be
theory is a major challenge in all scientic elds. precisely dened inside of each specic theory. The
In summary to this section, all four properties of theory
properties of denitions that need to be considered are
are needed for a set of relationships to be a theory and conservatism, denition parsimonyand uniqueness.
additional properties are needed to assure that the Good
theorytheory has good denitions that are conservative,
is a good theory. Good theory is needed for theparsimonious
ad- (short), and unique. A good denition is
vancement of science. The next four sections elaborate
formally on dened as a concise, clear verbal expression of a
each property of good theory. unique concept used for empirical testing.
The rst question is: do business academics dene their
concepts? Unfortunately, academics usually assume that
everyone implicitly understands conceptual denitions
GOOD THEORYS PROPERTIES PROVIDE
because they are widely used. However, one of the few
GUIDELINES FOR SCIENTIFIC THEORY academic articles that ever studied the use of denitions
BUILDING found that the majority of articles do not formally dene
The above discussion provides the basis for a set oftheir concepts. The study by Burgess, Singh and Koroglu
guiding principles for authors to use in their research. (2005) found that out of 100 articles only 42 dened
These principles will be called guidelines for good the-their terms. In short, only 42 of 100 articles precisely
ory building. Each property of theory (denition, domain, knew what they were measuring, testing or discussing.
relationships, and predictions) has specic challenges
Howthat can a researcher precisely measure something that is
need more explanation to provide more clarity for these
imprecisely dened? A concepts measure is only as good
guidelines. The next four sections are explanations of theseas its formal conceptual denition (Bunge 1967 and
guidelines so they can be implemented. numerous others). Yet there is an additional problem
when researchers utilize previously published denitions
Denitions: Science Can Only Advance as Rapidlythat have not been analyzed for conceptual problems. To
as the Language that Expresses Its Concepts avoid these problems, good denitions have restrictive
It has long been recognized that all theory must de- properties that limit how to dene terms (see Bunge
velop an articial language. That is, lay language is not 1967; Teas and Palan 1997; Wacker 2004).

8 Volume44,Number3
AConceptualUnderstandingofRequirementsforTheory-BuildingResearch

A Denition of Conservatism. One frustrationfor lead to innumerable measures that authors may use to
academicsoccurs in articlesthat use new, poorly empirically test theory, causing innumerable contradictory
dened terms that are seeminglyvery similar (or results depending upon the selection of measures. The
identical) to earlier concepts. It is a violation of the quest for better measures detracts from the real problem:
good denitions property to use new terms without goodformal conceptualdenitions (Bunge 1967).
carefullydistinguishingthem from similar, existing Worse, it leads to very confused measurements. Con-
terms. This inclusion violates the conservation property sequently, researchers should search for better formal
and in many cases, may violate the uniqueness property conceptual denitions rather than measures or metrics.
of good denitions. Although this property may sound The fallacy of seeking properties to clarify bad denitions
as if it restricts concepts and theory to the past, this has long been known in the philosophy of science (Bunge
property actually serves the very important purpose of1967; Hunt 1991; Teas and Palan 1997 and going back to
assuring that concepts are not just redened and called Plato in the parable of the cave).
different names (Bunge 1967; Teas and Palan 1997). Many conceptsin businessseeminglyhave an
DenitionofUniqueness. One very common difculty important meaning but cannot be adequately dened.
with denitions is that many denitions are not unique. In general, for these ill-dened terms there is a heavy
That is, they include terms that are used in other emphasis on developing measures for the ill-dened
conceptual denitions. When this happens, there is a concept. In many cases, the term is not claried and is
tautological relationship between the two concepts that included in some form of measurement instrument. This
cause the two concepts to be statistically signicantlyinclusion causes the ill-dened conceptto be
related. This problem has been noted as the overlappingstatisticallysignicantbut practicallyunimportant.
denition problem or concept stretching (OsigwehMore about this problem will be discussed below when
and Chimizie 1989). In short, academic authors shoulddiscussing statistical versus substantive signicance.
be very careful not to use terms found in other con- Measuring what is not clearly dened is an exercise
ceptual denitions. in academicfrustration. Goodformal conceptual
Denition Parsimony. Long denitions carry many denitions have very few measures. Even with measures
problems. Most of these denitions seem to include developed from good formal conceptual denitions, all
everything to avoid being non-unique. In some elds, measures are not created equal. Each measure has two
this is very common in such denitions as total qualitycritical dimensions: the reason for using it and the degree
management,lean manufacturing,totally integrated of necessity for the formal conceptual denition. The
supply chain and even big just-in-time. Many of thesecomplete explanation of these requirements is beyond
denitions confuse the relationships property of theory the intention of this paper. However, for the purposes
with the denition property of theory (for a more here, the property (measure) of a concept should be
complete explanation see Bunge 1967; Teas and Palandirectly derived from the formal conceptual denition
1997; Wacker 2004). Denitions should precede the (called the interpretive property). For a more complete
theorys relationship property because researchers must explanation please see Wacker (2004).
understand precisely what the concept is before they The underlying principle of this discussion is based
can state how it is related to other concepts. A relatedupon an important academic tenet: science can only
difculty with many denitions is that some publi- advance as fast as there is a language to express it
cations include other broad concepts in their de- (Bunge 1967 and other earlier authors). Too frequently,
nitions thereby making the denition even broader as science advances, academics readily accept ambiguous
(concept stretching). In sum, shorter denitions are (severaldistinct properties)or vague(innumerable
preferred to longer denitions to prevent concepts from properties) denitions without trying to rene them.
being confused. Without this renement, the science cannot progress.
Good Measures of Concepts: You Cannot Precisely All conceptualtheory is derivedfrom the existing
Measure What You Cannot Precisely Dene. There is a literature. It is important that the theory derived from
common misunderstanding among academicsthat the literature drives the conceptual model. Frequently,
seekingspecicempirical properties(characteristics, there are competing conceptual models. For business
metrics, traits, measures, operational denitions, etc.)theories,
is it is important to review the literature beginning
the best way to clarify concepts before they have devel- with the formal conceptual denitions to determine
oped formal conceptualdenitions.This empirical which of the competing theories has the better good
search is illogical. How can any concept be measured,formal conceptual denitions. If one theory does not
if the concept is not clear? Bunge (1967) and others havedene its terms or does not evaluate existing denitions
noted that the measure of a concept is dened as an while a second does and offers better formal conceptual
informal denition (in the philosophy of science, denitions, then the second theory should be considered
measuresare also known as accidentaldenitions superior to the rst. Without good formal conceptual
and in the businessliteraturethey are known as denitions, it would be difcult for any theory to be
operational denitions). Ambiguous and vague terms good theory.

July2008 9
JournalofSupplyChainManagement

One major problem in business is the use of the term heterogeneous because every sample point is gathered
constructs. Although it is not the focus of this article, at aa different time and place. Consequently, authors
comment on the term construct is necessary. One should justify the sample they are planning to use to
common denition of a construct is a conceptdevel- test a specic theory. It should be apparent that the
oped for describing relationships among phenomena, sample or should be based upon the theorys domain. A
a theoretical denitionin which concepts are denedfrequent criticism of an empirical study is that the sample
(Vogt 1999). Vogts denition is ambiguous, because usedit to test a theory is biased (a biased sample is dened
points to at least two conceptual denitions in the as theapriori belief that a sample is not representative of
phrases: (1) describing relationships among phenom-the theorys population). It is always possible for any
ena,and (2) or a theoreticaldenition.Vogts critic to state that the sample was not representative of
denition of construct is neither clear, nor precise, nor the population. So, authors should carefully account for
unique. Consequently, it is not a good denition,sample demographics to assure that when and where the
because when an academic uses the term construct sample is collected is well specied (Mitchell and James
they can mean either a relationship or a denition. Recall 2001).
that a formal conceptual denition must lead to a unique
Which demographics (characteristics) should the re-
concept. This property of good denitions is contrary
searcher expect to account for variations in the expected
to some of the current literature that use constructs for
relationships? Here are some typical questions that
multidimensional concepts (note that the denition
should be included in the discussion of the chosen
of the term construct can be made clearer by making
sample: These characteristics can be classied as: (1)
constructinto two denitions: abstractconcept
Countries, (2) Industries, (3) Respondents Position, and
construct and a theory relationship construct).
With that stated, many times it is necessary to use (4) Respondents Demographics. Compounding all these
multidimensional terms to explain broad concepts factors,for there are innumerable differences between these
ease of explanation. datum characteristics. Because the number of character-
In sum, great care should be taken when dening istics is larger than sample size, it would be not be pos-
concepts used in theory building. For theory building, sible for any statistical technique to control for all the
all conceptsshould have good formal conceptual differences. It would be relatively easy for any reader/re-
denitions. All formal conceptual denitions should be viewer to summarily reject any empirical study on the
grounds
unique, conservative, and short. Unfortunately, there is a that the data were heterogeneous, leading to all
natural predispositionto accept traditional terms empirical articles being rejected. One resolution for this
without evaluating them for precision. Without these conundrum is the existence of empirical evidence that
renements, theory and science cannot advance. suggests that specic characteristics are related to im-
portant conceptual variables. If there is no empirical ev-
idence in the academic literature, then the sample cannot
Domain of Theory: When and Where Did the Event be biased and should not be questioned. The require-
Happen? ment of empirical literature to support a readers claim
A theorys domain is when and where a theory is to be that the data were too heterogeneous to draw conclu-
applied. It is a critical property of good theory because sions is known as the previousliteratureconvention
. If there
it limits when and where the theory applies. However, is no previous literature to suggest some characteristic is
the more important theories can be applied at more related to the theorys important variables, then the
times and in more places than less important theories,sample is considered unbiased.
other things being equal. There are two related properties
of good theory: generalizability and abstractness. For Any Statistical Result Can Be Explained
this study, a theorys generalizability is dened as the The theorys relationship property has four key re-
degree to which a theory can be applied to existing quirements for good theory: Fecundity, Internal Con-
populations. The wider the existing populations where sistency, Substantive versus Statistical Signicance, and
the theory applies, the more generalizable the theory Statistical
is. Parsimony.
On the other hand, abstraction is dened as a theorys Relationships Fecundity Property. The fecundity of a
application being void of time and space requirements. good theory is interpreted to mean a new theory may
Put another way, an abstract theory can be applied over explain the current phenomena but also may offer new
all times and all places. Abstraction at its highestareas
level isto research. A new theory that offers new areas to
called a grand theory and is considered the ideal goal explore of is considered superior to existing theory that
theory because the theory applies to all times and allexplains and explores fewer issues. Other things being
places (Osigweh and Chimizie 1989). equal, if the new theory offers more new areas to explore,
The choice of sample is dictated by the theorys do- then it is preferred to the existing theory. However, as a
main. Reviewers of empirical articles uniformly challenge caution it is important that the new theory explains the
the researchers choice of sample. The sample is always phenomenon as clearly as the existing theory. In short,

10 Volume44,Number3
AConceptualUnderstandingofRequirementsforTheory-BuildingResearch

the fecundity property of relationships serves the purposeabstract conceptual world is just an abstruse understand-
of gaining new, wider research areas to explore. ing that is open to any interpretation. This interpretation
New theories integrateexisting theories.This may lead non-academics to believe that their efforts
integration is important for theory development. Fromshould be highly focused on statistical methods and
a strict philosophy of science perspective, theory builds not on conceptualunderstanding.Yet when the
upon previous theory (Bunge 1967; Hunt 1991 going statistical results are explained, the explanation provides
back to Plato). Theory building comes from integratingan insight into the application of the results to specic
existing theories into a larger framework and higher levelorganizations and thus provides important implications
theory. By building upon previous theory,good theory for pragmatic managers.
integrates more and more concepts into a larger theory Relationships Parsimony: Explain More Theory and
that can explain many events. This integration is an Report Less Statistics. It may seem obvious that the
important property for the advancement of science. purpose of any empirical article is to test and to
RelationshipsInternal ConsistencyProperty.The explain a theory. Yet when reading articles,it is
internal consistency of good theory is interpreted toapparent that a hefty portion of the page count is spent
mean that the theory is logically consistent. The internalon explaining statistics rather than completely explaining
consistency property is very difcult to demonstrate. At thea model and how the empirical results of the theory
very simplistic level, it is a graphical model showing therelate to the overall business eld. One result of this over-
relationships between all variables. At the more advanced emphasis on statistics is the lack of truly integrated
theories. Empirical results tend to stand alone without
level, the researcher will illustrate that the theory gives
very specic mathematical relationships between all integration into a larger body of knowledge that is based
relevant variables from extant business specic liter- upon existing theory. From a scientic philosophy of
ature. In some academic elds, symbolic logic gives the science perspective, the empirical investigation should
deductive reasoning for the relationships but more typi- focus on specic theory development and on the use of
cally mathematics is the preferred method for internalspecic statistical techniques to test that theory.
consistency (Moorthy 1993). Recently there have been many newly developed sta-
Although the internal consistency property seems like tistical techniques to test theory. Sometimes researchers
frequently
it would be relatively easy to demonstrate, in practice it is focus on these abstruse statistical techniques.
quite difcult. Certainly,statisticscannot give the The reasons may be: (1) It is easier to report statistics
conceptuallycorrect theory and its related model. than it is to develop a full conceptual model before the
Consider a simple model of just 10 variables. There are data are gathered; (2) sophisticated statistical analyses
3,628,800 possible models. Ata5the 5% level, there are can easily disguise faulty conceptual analyses by report-
(181,440) statistically signicant models by mere chance. ing specic advanced statistical techniques; and (3) many
So only the abstract conceptual world can sort out which academics are better trained in using statistics than in
ones of these models really make logical sense. One of understandingthe underlying traditional statistical
the oldest maxims in statistics and econometrics is Any mathematical difculties and the relationship of these
difculties to explain the theory.
statistical result can be explained. Unfortunate as it may
seem, any one of those models could be explained with Many conceptually developed empirical articles in
some reasoning (convoluted or not). Without logical business need to have sophisticated statistical tech-
integration into the larger body of business knowledge, niques. Yet some articles overemphasize the importance
results become statistical artifacts that could have hap- of the statistical ndings. Because academic studies are
pened by mere chance. Only through conceptual devel- restricted in their length, the presentation and discussion
opment can statistical estimates point to an integrated of sophisticated statistical techniques are not without
understanding among many related concepts. Regardless trade-offs. Unfortunately, sometimes what is lost is the
of what is found during statistical modeling it can be complete examination and explanation of the theory that
explained . . . many times with the tragic result of beingis being tested. It would be difcult to imagine any
subject to random chance. academic wishing to read an article that did not carry a
complete explanation of how and why a theory is
For the last 20 or so years statistics have been a central
driving force in many highly ranked academic journals.important to the pragmatic managerial world.
To some academics, statistical estimation is the only way Academics should heed the words of the leading
an article can be called a study, although this belief statistical scholars of the APA task force (Wilkinson
ignores the denition of a study (study is usually et al., 1999, p. 10) on reporting of statistical results:
dened as the pursuit of knowledge, by reading, obser-
vation, or research). The basic purpose of research is theThe enormous variety of modern quantitative
complete understanding of phenomena. Sometimes this methods leaves researchers with the nontrivial task
complete understanding can be enhanced by under- of matching analysis and design to the research
standing conceptualization of the statistical results. To question. Although complex designs and state-of-
non-academics, statistics are the real world while the the-art methods are sometimes necessary to address

July2008 11
JournalofSupplyChainManagement

research questions effectively, simpler classical including or excluding a variable that is a clear violation
approaches often can provide elegant and sufcient for good theory building. In short, there is a critical
answers to important questions. Do not choosedifference
an between statistical signicance and substantive
analytic method to impress your readers or to signicance. Statistics are not the determining factor if a
deect criticism. If the assumptions and strengthconcept
of is theoretically important. The conclusion is that
a simpler method are reasonable for your data andonly the conceptually important variables should be
research problem, use it. Occams razor applies to included in any estimating procedure.
methods as well as to theories. In short, there are many wrong variables in a statistical
estimatethat are statisticallysignicant but are
meaningless because they have occurred due to chance.
The APA task force indicates that scholars must explain
Also more unfortunate is that there are many important
the results in the conceptual relationship ofapriori
the
conceptual variables that are excluded due to statistical
model rather than trying to impress the reader/reviewer
insignicance from the sample. Only a substantively
with non-meaningful statistics. Just because the statistics
signicantvariable is important while statistically
software program output provides impressive statistics, it
signicant variables may not be substantively signicant.
does not mean that authors should report them. This
The old econometrics maxim still applies today to
view was also supported historically by others (see Fisher
most empirical estimates: if a variable is conceptually
1935 and others).
important enough to be included in the estimate, it is
Statisticalvs.SubstantiveSignicance:StatisticsAreNot important enough to leave in the estimate. In short, do
aSubstituteforConceptualization.There is a disquieting not drop conceptually important variables just because
trend in empirical research that has not been adequately they are statistically insignicant.
addressed in the business literature and not practiced in
many empirical articles. That trend is the trade-off be-
tween statistical signicance and substantive signicance.
The statistics and the abstract theory should go hand-in- Predictions
hand to understand the underlying causes of empirical The prediction property of good theory is refutability
results. Researchers should remember that statistics are (falsiability). Sir Karl Popper has pointed to this prop-
mere facts and do not relate to theory without concep-erty as being a critical aspect of good theory. This
tualization. It does not seem possible to make predic- property states that the more unlikelya prediction is, the
tions without conceptualization. better the theory is. A theory that predicts a very likely
Unfortunately, some articles seem to confuse the event is not considered a good theory. Sometimes
conceptual model with the statistical estimating model. when reading a conclusion of an empirical article, a
This confusion is a major conceptual aw in explanation reader may be left with an empty feeling that there is
because the statistical model does not integrate theory, it
absolutely nothing in the article that is not obvious. If the
merely tests it. This problem was not as severe 3040 reader asks the question: How likely are the articles
years ago when econometric books such as Johnstonconclusions to occur? Frequently, the answer would be:
(1972) clearly differentiated between the conceptual
It would be revolutionary if the conclusion did not
model and the statistical model. Currently, there is a re-
occur. The concern with obvious conclusions violates
emphasis on this problem of confusing the conceptual
the falsiability criterion for good theory. On the other
model with the statistical model. McCloskey and Ziliak
hand, if the conclusion is logical and offers new insights
(1996) have examined this difculty in the economics
into conceptual relationships, then the reader is left with
literature. McCloskey and Ziliak examined the difference
the feeling that these ndings are very important because
between the statistical model and the substantive model.
They then differentiated statistical signicance (con- the relationships were not obvious.
dence level) and substantive signicance (conceptual Frequently, academics during empirical investigations
signicance). Their key point is that there is a large nd a new statistical relationship between variables
difference between the conceptual model (developed(concepts). These relationships frequently surprise the
from the theory) and the statistical model (developed researcher and are important for future studies. However,
to test the theorys conceptual model). They warned these
thatrelationships were not predicted and should not be
portrayed
this confusion causes an over-emphasis on the statistical as being predicted. The explanation of any re-
sult
results and attributes less importance to the theory being that was not predicted has a very special name in the
tested. The important conclusion is that there are many philosophy of science: it iscalling
in the conventional
statistical results that notsubstantive
are results but are stratagem . It is considered a major logic research aw in
merely statistics that could have occurred by mere the philosophy of science. With that said, researchers
chance. During their survey of the economics literature, should attempt to explain unexpected results from the
McCloskey and Ziliak found that 70% of the papers did literature. Explaining unexpected results is only a prob-
not differentiate between statistical signicance and lem sub-
when the researcher states that the results were pre-
stantive signicance. Additionally, 53% used statisticsdicted.for Predictions must be predictions and not what was

12 Volume44,Number3
AConceptualUnderstandingofRequirementsforTheory-BuildingResearch

FIGURE3
TheDerivationoftheGuidelinesforGoodTheory

Good Theorys Guidelines for Good


All Theories Additional Theory Building
Properties Properties

Conservatism Conservatism:
Parsimony Denition Parsimony:
Denitions Uniqueness Uniqueness:

Generalizability Generalizability:
Denition of good Domain Abstractness Abstraction:
theory: Specic times and places to test theory
A fully explained set of
conceptual relationships Fecundity:
Internal Consistency
used for empirical Internal consistency:
Relationships Fecundity
investigations. Relationship Parsimony:
Parsimony
Simplest Explanations Statistical Parsimony:
Substantive vs. Statistical Signicance
Predictions Falsiability Falsiability:

TABLEI
ASummaryofGuidelinesforEvaluatingAllEmpiricalTheoryBuilding

1. Istheproposedtheoryreallya theory?Doesthetheoryexplicitlystateallfourpropertiesoftheory
(denitions,domain,relationships,andpredictions)?
2. Iftheproposedtheoryisactuallyatheory,howgoodofatheoryisit?Whereareitsweaknesses:
precisionindenitions,specieddomains,logicallyconsistentrelationships,falsiablepredictions?
3. Howgoodarethedenitions?
a.Doestheauthor(s)deneallconceptsorpresentdenitionstoclearlyunderstandwhattheyare
proposing?
b.Doestheauthor(s)eschewfromrenaminganexistingconceptwithoutclearlydifferentiatingwhyit
isnecessary?
c.Arethedenitionsconcise(shortaspossible)?
d.Dothedenitionsleadtoauniqueconceptoraretheyambiguousorvague?
e.Dothedenitionspasstheguidelinesforgooddenitions?
4. Howcarefullyisthetheorysdomainspecied?
a.Doesthestudyusetheliteraturetoindicatewherethepreviousstudiesdomainsweregathered:
(1)Countries,(2)Industries,(3)RespondentsPosition,and(4)RelevantRespondentsDemographics?
b.Doesthestudyspecifywhenandwherethetheorycanbeappliedandtested?
5. Fortheorysrelationshipsempiricaltests:
a.Doesthetheoryoffernewareastoexplore(fecundity)?
b.Aretherelationshipsgraphically/mathematicallyexaminedandexplainedtobeinternallyconsistent?
c.Arealltherelationshipsfromtheextantliteratureincludedintheexplanation?
d.Doestheauthor(s)usethesimpleststatisticspossibletodeveloptheirtestsandeschewfrompresenting
statisticsthatarenotgermanetothediscussionsuggestions(Wilkinsonandet.al.1999taskforce)?
e.Doestheauthor(s)differentiatebetweensubstantivesignicanceandstatisticalsignicanceMcCloskey
andZiliak(1996)?
i.Dotheauthorsdropvariablesduetostatisticalinsignicance?
ii.Dotheauthorsincludevariablesduetostatisticalsignicanceratherthansubstantivesignicance?
ii.Aretherelationshipsconceptuallyexplainedversusstatisticallyexplained?
6. Forpredictions:
a.Dotheconclusionsoffernewinsightsandexplanationsforthetheory?
b.Aretheconclusionspredictedfromtheextantliteratureavoidingcallingintheconventionalstratagem?

July2008 13
JournalofSupplyChainManagement

found during empirical research. It is easy to forget that


improve future theory-building research in supply chain
without the apriori models all statistical results are management.
merely artifacts because they are subject to random as-
sociation.

BIBLIOGRAPHY
Guidelines to Evaluate Theory Bacharach, S.B. Organizational Theories: Some Criteria
The purpose of this section is to summarize guidelines for Evaluation, Academy of Management Review ,
(14:4), 1989, pp. 496-515.
for evaluating articles as to their importance to theory
Bollen, K.A. StructuralEquationswithLatentVariables , John
(for this article a theory-building guideline is dened as Wiley & Sons, New York, 1989.
an authoritative, prescribed direction for conducting Bunge, M.Scientic Research I: The Search for System ,
empirical research). Whether the reader is a beginning Springer-Verlag, New York, 1967.
academic student or the most recognized scholar in the Burgess, K., P.J. Singh and R. Koroglu. Supply chain
eld, it is important to be extremely critical of all ac- management: A structured literature review and
ademic literature using logical reasoning (this article is implications for future research,International
not exempt from this evaluation). The above discus- Journal of Operations and Production Management ,
sions give an outline of a logical method for evaluating (26:7), 2006, pp. 703-729.
the contribution to the academic literature and to Fisher, R.A.The Design of Experiments , Oliver & Boyd,
improve the signicance of the readers own studies. Edinburgh, Scotland, 1935.
Figure 3 gives the relationships between all theory andHempel, C.G. Methods of Concept Formation
the guidelines for good theory building for academic in Science.In Otto Neurath, Rudolf Carnap
and Charles Morris (Eds.), Formations of the Unity
articles.
of Science , University of Chicago Press, Chicago,
It should be noted that regardless of the research 1970.
method, all empirical research should follow these guide- Hunt, S.D.ModernMarketingTheory:CriticalIssuesinthe
lines (for different research methods and data gathering Philosophyof Marketing Science , Southwestern
methods see Meredith, Raturi, Amoako-Gyampah and Publishing Co, Cincinnati, OH, 1991.
Kaplan 1989; Wacker 1998, pp. 378380). The guide- Johnston, J.Econometrics Methods , 2nd ed., New York,
lines for good theory building are summarizedin McGraw-Hill, 1972.
Kaplan, A. The Conduct of Inquiry: Methodology for
Table I. These guidelines follow the outline of this paper:
Theory and Good Theory, Formal Conceptual Deni- Behavioral Sciences , Chandler Publishing Company,
tions, Theory Domain, Theory Relationships,and San Francisco,CA, 1964.
Predictions. As such, it provides a framework for theory Klein, K.J., H. Tosi and A.A. Cannella Jr. Multilevel
development. Theory Building: Benets, Barriers, and New Devel-
opments, The Academy of Management,(Apr Review
24), 1999, p. 2.
McCloskey, D. and S. Ziliak. The Standard Error of
CONCLUSION Regression, Journal of Economic Literature ,(34:1),
The general conclusion is that most academic articles 1996, pp. 97-114.
might relatively easily be refocused to develop good Meredith, J.R., A. Raturi, K. Amoako-Gyampah and B.
theory. The purpose of this article is not to criticize ex- Kaplan. Alternative Research Paradigms in Opera-
isting business literature, but to offer suggestions on tions,
JournalofOperationsManagement , (8:4), 1989,
how to improve the theory. The articles purpose was to pp. 297-326.
provide a frameworkfor improving theory-building Mitchell, T.R and L.R. James. Building Better Theory:
to raise theory buildings importance to the academic Time and the Specication of when Things Happen.
eld. Academy of Management, The Academy of Man-
There are numerous areas of future study to illustrate agementReview , (26:4), 2001, pp. 530-547.
many of the points addressed in this studyspecically, Moorthy, K.S. TheoreticalModeling in Marketing,
JournalofMarketing , (57:April), 1993, pp. 92-106.
how theorys conceptualization affects specic advanced
Osigweh, C.A.B. and A.B. Chimizie. Concept Fallibility
statistical methods used for theory testing. Most impor- in Organizational Science, Academy of Management
tant would be the statistical demonstration of how de- Review , (14:4), 1989, pp. 579-594.
nitions affect concept measurementor how and whyPoole, M.S. and A.H. Van De Ven. Using Paradox to
some samples are biased and why others are not. Also Build Management and Organizational Theories,
helpful would be relationship estimation procedures to Academy of Management Review , (14:4), 1989, pp.
suggest specic estimation procedures for addressing 562-578.
specicacademicissues.In short, this article was Popper, Sir K.Philosophy of Science: A Personal Report.
written not as a criticism of current or past research In BritishPhilosophyInMid-Century , C.A.Mace(Ed.),
but with the goal of providing helpful suggestions to 1957.

14 Volume44,Number3
AConceptualUnderstandingofRequirementsforTheory-BuildingResearch

Shubik, M.What Is an Application and When Is TheoryWilkinson, L. et al. Task Force on Statistical Inference,
aWaste of Time?, ManagementScience , Statistical Methods in Psychology Journals,
American
(33:December), 1987, pp. 1511-1522. Psychologist
, (54:3), 1999, pp. 594-604.
Swamidass, P.M. Empirical Science: New Frontier in
Operations Management, Academy of Management
Review, (18:4), 1991, pp. 793-814.
Teas, R.K. and K.M. Palan. The Realms of Scientic
John G. Wacker (Ph.D., Wayne State University,
Meaning Framework for Constructing Theoretically
Meaningful Nominal Denitions of Marketing Detroit, MI) is a visiting professor of supply chain
Concepts, Journal of Marketing
, (2:61), 1997, pp. management at Arizona State University in Tempe, Ari-
52-68. zona. He has published 45 journal articles
Journal
in of
Van De Ven, A.H. Nothing Is Quite So Practical as a Operations Management, Decision Sciences
, International
Good Theory,Academy of Management Review , Journal of Production Research
, Journal of Marketing Re-
(14:4), 1989, pp. 486-489. search, International Journal of Production Economics
and
Vogt, W.P. Dictionary of Statistics & Methodology
,Sage, numerous other journals. His research has covered a
Thousand Oaks, CA, 1999. wide variety of topics such as the use of theory for sta-
Wacker, J.G. A Denition of Theory: Research tistical methods, manufacturing implementation and
Guidelines for Different Theory-BuildingRe- forecasting. He has been on the Editorial Review Board
search Methods in Operations Management, for JournalofOperationsManagement for the last 20 years.
Journal of Operations Management, (16), 1998, pp.
He acted as President of the Global Manufacturing Re-
361-385.
search Group (www.gmrg.org). He has taught in over 16
Wacker, J.G.A Theory of Formal Conceptual Denitions:
Developing Theory-Building Measurement Instru- international MBA and PhD programs for operations and
ments,Journal of Operations Management ,(22), supply chain management including Chinese University
2004, pp. 629-650. of Hong Kong (1 year) and Troy State Europe program (1
Whetten, D.A. What Constitutes a Theoretical Contri-year), University of Melbourne, MacQuarie University,
bution?,Academy of Management Review ,(14:4), Tatung Institute of Technology, and Hangzhou Institute
1989, pp. 490-495. of Electrical Engineering.

July2008 15

S-ar putea să vă placă și