The Cycles of Theory Building in Management Research

The Cycles of Theory Building in Management Research






Paul R. Carlile
School Of Management
Boston University
Boston, MA 02215
carlile@bu.edu


Clayton M. Christensen
Harvard Business School
Boston, MA 02163
cchristensen@hbs.edu










January 6 2005

Version 6.0

---

The Cycles of Theory Building in Management Research

Theory thus become instruments, not answers to enigmas, in which we can
rest. We don’t lie back upon them, we move forward, and, on occasion, make
nature over again by their aid. (William James, 1907: 46)

Some scholars of organization and strategy expend significant energy disparaging and
defending various research methods. Debates about deductive versus inductive theory-building
and the objectivity of information from field observation versus that of large-sample numerical
data are dichotomies that surface frequently in our lives and those of our students. Despite this
focus, some of the most respected members of our research profession (i.e., Simon (1976), Solow
(1985), Hambrick (1994), Staw and Sutton (1995), and Hayes (2002)) have continued to express
concerns that the collective efforts of business academics have produced a paucity of theory that
is intellectually rigorous, practically useful, and able to stand the tests of time and changing
circumstances.
The purpose of this paper is to outline a process of theory building that links questions
about data, methods and theory. We hope that this model can provide a common language about
the research process that helps scholars of management better understand the roles of different
types of data and research, and thereby to build more effectively on each other’s work. Our unit of
analysis is at two levels: the individual research project and the iterative cycles of theory building
in which a researchers attempt to build upon each other’s work. The model synthesizes and
augments other studies of how communities of scholars cumulatively build valid and reliable
theory.1 It has normative and pedagogical implications for how we conduct research, evaluate the
work of others, train our doctoral students, and design our courses.
While many feel comfortable in their understanding of these perspectives, it has been our
observation that those who have written about the research process and those who think they
understand and practice it proficiently do not yet share even a common language. The same
words are applied to very different phenomena and processes, and the same phenomena can be
called by many different words. Papers published in reputable journals often violate rudimentary
rules for generating cumulatively improving, reliable and valid theory. While recognizing that
research progress is hard to achieve at a collective level, we assert here that if scholars and
practitioners of management shared and utilized a sound understanding of the process by which
theory is built, we could be much more productive in doing research that doesn’t just get
published, but meets the standards of rigorous scholarship and helps managers know what actions
will lead to the results they seek, given the circumstances in which they find themselves.
Our purpose in this paper is not to praise or criticize other scholars’ work as good theory or
bad theory: almost every published piece of research has its unique strengths and shortcomings.
We will cite examples of other scholars’ research in this paper, but we do so only to illustrate how
the theory-building process works. We hope that the model described here might constitute a
template and a common language that other scholars might use to reconstruct using how bodies of
understanding have accumulated in their own fields.

1

---

In the first of the four sections of this paper, we describe a three-step process by which
researchers build theory that is at first descriptive, and ultimately normative. Second, we discuss
the role that discoveries of anomalies play in the building of better theory, and third, we describe
how those who build, evaluate and utilize theories can tell whether they can trust a theory –
whether it is valid and applies to the situation in which they find themselves. Finally, we suggest
how scholars can engage in course research – to design student courses in ways that help faculty
researchers build better theory.
I. The Theory Building Process
The building of theory occurs in two major stages – the descriptive stage and the normative
stage. Within each of these stages, theory builders proceed through three steps. The the theory-
building process iterates through these three steps again and again. In the past, management
researchers have quite carelessly applied the term theory to research activities that pertain to only
one of these steps. Terms such “utility theory” in economics, and “contingency theory” in
organization design, for example, actually refer only to an individual step in the theory-building
process in their respective fields. We propose that it is more useful to think of the term “theory” as
a body of understanding that researchers build cumulatively as they work through each of the
three steps in the descriptive and normative stages. In many ways, the term “theory” might better be
framed as a verb, as much as it is a noun – because the body of understanding is continuously
changing as scholars who follow this process work to improve it.
The Building of Descriptive Theory
The descriptive stage of theory building is a preliminary stage because researchers
generally must pass through it in order to develop more advanced normative theory. The three
steps that researchers who are building descriptive theory utilize are observation, categorization,
and association.
Step 1: Observation
In the first step researchers observe phenomena and carefully describe and measure what
they see. Careful observation, documentation and measurement of the phenomena in words and
numbers is important at this stage because if subsequent researchers cannot agree upon the
descriptions of phenomena, then improving theory will prove difficult. Early management studies
such as The Functions of the Executive (Barnard, 1939) and Harvard Business School cases
written in the 1940s and 50s were primarily descriptive work of this genre – and was very valuable.
This stage of research is depicted in Figure 1 as the base of a pyramid because it is a necessary
foundation for the work that follows. The phenomena being explored in this stage include not just
things such as people, organizations and technologies, but processes as well. These observations
can be done anywhere along the continuum from analysis of huge databases on the one end, to
field-based, ethnographic observation on the other.
Without insightful description to subsequently build upon, researchers can find themselves
optimizing misleading concepts. As an example: For years, many scholars of inventory policy
and supply chain systems used the tools of operations research to derive ever-more-sophisticated
optimization algorithms for inventory replenishment. Most were based on an assumption that

2

---

managers know what their levels of inventory are. Ananth Raman’s pathbreaking research of the
phenomena, however, obviated much of this work when he showed that most firms’ computerized
inventory records were broadly inaccurate – even when they used state-of-the-art automated
tracking systems (Raman 199X). He and his colleagues have carefully described how inventory
replenishment systems work, and what variables affect the accuracy of those processes. Having
laid this foundation, supply chain scholars have now begun to build a body of theories and
policies that reflect the real and different situations that managers and companies face.
Researchers in this step often develop what we term constructs. Constructs are
abstractions that help us rise above the messy detail to understand the essence of what the
phenomena are and how they operate. Joseph Bower’s Managing the Resource Allocation Process
(1970) is an example of this. His constructs of impetus and context, explaining how momentum
builds behind certain investment proposals and fails to coalesce behind others, have helped a
generation of policy and strategy researchers understand how strategic investment decisions get
made. Economists’ concepts of “utility” and “transactions costs” are constructs – abstractions
developed to help us understand a class of phenomena they have observed. We would not label
the constructs of utility and transactions cost as theories, however. They are part of theories –
building blocks upon which bodies of understanding about consumer behavior and organizational
interaction have been built.
Step 2: Classification
With the phenomena observed and described, researchers in the second stage then classify
the phenomena into categories. In the descriptive stage of theory building, the classification
schemes that scholars propose typically are defined by the attributes of the phenomena.
Diversified vs. focused firms, and vertically integrated vs. specialist firms are categorization
examples from the study of strategy. Publicly traded vs. privately held companies is a
categorization scheme often used in research on financial performance. Such categorization
schemes attempt to simplify and organize the world in ways that highlight possibly consequential
relationships between the phenomena and the outcomes of interest.
Management researchers often refer to these descriptive categorization schemes as
frameworks or typologies. Burgelman & Sayles (1986), for example, built upon Bower’s (1970)
construct of context by identifying two different types of context – organizational and strategic.
Step 3: Defining Relationships
In the third step, researchers explore the association between the category-defining
attributes and the outcomes observed. In the stage of descriptive theory building, researchers
recognize and make explicit what differences in attributes, and differences in the magnitude of
those attributes, correlate most strongly with the patterns in the outcomes of interest. Techniques
such as regression analysis typically are useful in defining these correlations. Often we refer to
the output of studies at this step as models.
Descriptive theory that quantifies the degree of correlation between the category-defining
attributes of the phenomena and the outcomes of interest are generally able to make probabilistic
statements of association representing average tendencies. For example, Hutton, Miller and

3

---

Skinner (2000) have examined how stock prices respond to earnings announcements. They coded
types of words and phrases in the statements as explanatory variables in a regression equation,
with the ensuing change in equity price as the dependent variable. This analysis enabled the
researchers then to assert that, on average across the entire sample of companies and
announcements, delivering earnings announcements in a particular way would lead to the most
favorable (or least unfavorable) reaction in stock price. Research such as this is important
descriptive theory. However, at this point it can only assert on average what attributes are
associated with the best results. A specific manager of a specific company cannot yet know
whether following that average formula will lead to the hoped-for outcome in her specific
situation. The ability to know what actions will lead to desired results for a specific company in a
specific situation awaits the development of normative theory in this field, as we will show below.
How Theory is Improved within the Descriptive Stage
When researchers move from the bottom to the top of the pyramid in these three steps –
observation, categorization and association, and in so doing give us constructs, frameworks and
models – they have followed the inductive portion of the theory building process. Researchers can
then get busy improving these theories by cycling from the top down to the bottom of this
pyramid in the deductive portion of the cycle – seeking to “test” the hypotheses that had been
inductively formulated. This most often is done by exploring whether the same correlations exist
between attributes and outcomes in a different set of data than the data from which the
hypothesized relationships were induced. When scholars test a theory on a new data set (whether
the data are numbers in a computer, or are field observations taken in a new context), they
sometimes find that the attributes of the phenomena in the new data do indeed correlate with the
outcomes as predicted. When this happens, this “test” confirms that the theory is of use under the
conditions or circumstances observed.2 However, researchers who stop at this point simply return
the model to its place atop the pyramid tested but unimproved.
It is only when an anomaly is identified – an outcome for which the theory can’t account –
that an opportunity to improve theory occurs. As Figure 1 suggests, discovery of an anomaly
gives researchers the opportunity to revisit the foundation layers in the theory pyramid – to define
and measure the phenomena more precisely and less ambiguously, or to cut it into alternative
categories – so that the anomaly and the prior associations of attributes and outcomes can all be
explained. In the study of how technological innovation affects the fortunes of leading firms, for
example, an early attribute-based categorization scheme was radical vs. incremental innovation.
The statements of association that were built upon it concluded that the leading established firms
on average do well when faced with incremental innovation, but they stumble in the face of
radical change. But there were anomalies to this generalization – established firms that
successfully implemented radical technology change. To account for these anomalies, Tushman
& Anderson (1986) offered a different categorization scheme, competency-enhancing vs.
competency-destroying technological changes. This scheme resolved many of the anomalies to
the prior scheme, but subsequent researchers uncovered new ones for which the Tushman-
Anderson scheme could not account. Henderson & Clark’s (1990) categories of modular vs.
architectural innovations; Christensen’s (1997) categories of sustaining vs. disruptive
technologies; and Gilbert’s (2001) threat-vs.-opportunity framing each uncovered and resolved
anomalies for which the work of prior scholars could not account. This body of understanding
has improved and become remarkably useful to practitioners and subsequent scholars (Adner,

4

---

2003; Daneels, 2005) because these scholars followed the process in a disciplined way. They
articulated theories that could be falsified – that could yield anomalies. Subsequent scholars then
uncovered what these anomalies were, and resolved them by slicing the phenomena in different
ways and articulating new associations between the category-defining attributes and the outcome
of interest.
Figure 1
The Process of Building Theory
In
Statements
duc
C
of association
t
o
i
n
v
fi
e
Predict
rm p
ctive process
(models)
roce
edu
ss
D
Categorization based
upon attributes of phenomena
(frameworks & typologies)
Anomaly
Observe, describe & measure
the phenomena
(constructs)

In contrast to many debates about the virtues of deductive and inductive methods, this
suggests that these are two sides to the same pyramid. Every complete lap around the theory-
building pyramid consists of an inductive side and a deductive side. Theory building efforts stall
when researchers drop the baton and declare victory having run only half of a lap around the
theory pyramid.3
Descriptive theory-building efforts typically categorize by the attributes of the phenomena
because attributes are easiest to observe and measure. Likewise, correlations between attributes
and outcomes are easiest to hypothesize and quantify through techniques such as regression
analysis. Kuhn (1962) observed that confusion and contradiction typically are the norm during
descriptive theory-building. This phase is often characterized by a plethora of categorization
schemes because the phenomena generally have many different attributes. The sequence of
studies of technology change cited above is an illustration of such a plethora. Often, in this phase,
no model is irrefutably superior: Each seems able to explain anomalies to other models, but
suffers from anomalies to its own.
The Transition from Descriptive to Normative Theory
The confusion and contradiction that often accompany descriptive theory become resolved
when careful researchers – often through detailed empirical and ethnographic observation – move

5

---

beyond statements of correlation to define what causes the outcome of interest. As depicted in
Figure 2, they leap across to the top of the pyramid of normative theory, whose capstone is a
statement of what causes the outcome of interest, not just what is correlated with it. Their
understanding of causality enables researchers to assert what actions managers ought to take, in
order to get the results they need. For reasons noted below, normative theory has much greater
predictive power than descriptive theory does.4
Figure 2:
The Transition from Descriptive Theory to Normative Theory
Conf
Predict
irm
Careful field-based research
Statement
of causality
Categorization of the
circumstances in which we
ctive process
edu
might find ourselves
I
D
nduct
Anomaly
Preliminary
ive
C
Observe, describe &

statements of
o
p
nfi
r
r
Predict
m
o
correlation
ce
measure the phenomena
ss
Normative Theory
Categorization by the
ctive process attributes of the phenomena
edu
D
Inductiv
Anomaly
e proces
Observe, describe & measure the phenomena
s
Descriptive Theory

Normative theory, like its descriptive predecessor, still needs to be improved – and
researchers do this by following the same steps that were used in the descriptive stage.
Hypothesizing that their statement of causality is correct, they cycle deductively to the bottom of
the pyramid to test the causal statement: If we observe these actions being taken, these should be
the outcomes that we observe. When they encounter an anomaly, they then delve back into the
lower levels of the pyramid. Sometimes they can resolve anomalies by developing more accurate,
less ambiguous ways to define and measure the phenomena. Often they account for the
anomalies by revisiting the categorization stage. Rather than using schemes based on attributes of
the phenomena, however, in building normative theory researchers categorize the different
situations or circumstances in which managers might find themselves. They do this by asking,
when they encounter an anomaly, “What was it about the situation in which those managers found
themselves, that caused the causal mechanism to yield a different result? By asking this question
as they cycle up and down the pyramid of normative theory, anomaly-seeking researchers will
ultimately define a relatively complete set of the situations or circumstances in which managers
might find themselves when pursuing the outcomes of interest.5 This allows researchers to make

6

---

contingent statements of causality – to show how and why the casual mechanism results in a
different outcome, in the different situations. A normative theory that is built upon well-
researched categories of circumstances can help a manager predict accurately what actions will
and will not lead to the desired result, given the circumstance in which she finds herself. The
relatively accurate, circumstance-contingent predictability of normative theory enables managers
to know, in other words, what they ought to do.6
The history of research into manned flight is a good way to visualize how this transition
from descriptive to normative theory occurs, and how it is valuable. During the middle ages,
would-be aviators did their equivalent of best-practices research and statistical analysis. They
observed the many animals that could fly well, and compared them with animals that could not.
The vast majority of the successful fliers had wings with feathers on them; and almost all of those
that couldn’t fly had neither of these attributes. This was quintessential descriptive theory. Pesky
outliers like ostriches had feathered wings but couldn’t fly; bats had wings without feathers and
were very good at it; and flying squirrels had neither and got by. But the R2 was so high that
aviators of the time copied the seemingly salient characteristics of the successful fliers in the
belief that if they copied the characteristics of the “best practices’ fliers, they could fly, too. So
they fabricated wings, glued feathers on them, jumped off cathedral spires, and flapped hard. It
never worked. For centuries they sought to fly by trying harder – assuming that the prior aviators
had failed because they had bad wing designs; hadn’t bulked up their muscles enough; or hadn’t
flapped hard enough. There were substantial disagreements about the categorization scheme, too –
which of the birds’ attributes truly enabled flight, and which didn’t. For example, Roger Bacon
wrote an influential paper asserting that the differentiating characteristic was birds’ hollow bones
(Clegg, 2003). Because man had solid bones, Bacon reasoned, we could never fly. He then
proposed several designs of machines that could flap their wings with sufficient power to
overcome the disadvantage of solid human bones. But it still never worked. Armed only with the
correlative statements of descriptive theory, aviators kept killing themselves.
Then through his careful study of fluid mechanics Daniel Bernoulli identified a shape that
we call an airfoil – a shape that, when it cuts through air, creates a mechanism that we call lift.
Understanding this causal mechanism, Bernoulli’s Principle, made flight possible. But it was not
yet predictable. In the language of this paper, the theory predicted that aviators would fly
successfully when they built machines with airfoils to harness lift. But while they sometimes flew
successfully, occasionally they did not. Crashes were anomalies that Bernoulli’s theory could not
explain. Discovery of these anomalies, however, allowed the researchers to revisit the
categorization scheme. But this time, instead of slicing up the world by the attributes of the good
and bad fliers, researchers categorized the world by circumstance – asking the question, “What was
it about the circumstance that the aviator found himself in that caused the crash?” This then
enabled them to improve equipment and techniques and articulate circumstance-contingent
statements of causality: “This is how you should normally fly the plane. But when you get in this
situation, you need to fly it differently in order to get the desired outcome. And when you get in
this other situation, don’t even try to fly. It is impossible.”
When their careful studies of anomalies allowed researchers to identify the set of
circumstances in which aviators might find themselves, and then modified the equipment or
developed piloting techniques that were appropriate to each circumstance, manned flight became
not only possible, but much more predictable. This is how this body of understanding about

7

---

human flight transitioned from descriptive to normative theory. It was the discovery of the
fundamental causal mechanism that made flight possible. And it was the categorization of the
salient circumstances that enabled flight to be made more predictable.
The world of managers is unlikely ever to become perfectly predictable, of course.
Managers, like pilots, will likely continue to find themselves in never-before-encountered
situations for which adequate rules and equipment have not yet been created. Complicated human
factors and group dynamics also militate against perfect predictability. But even here, the work
of scholars of group dynamics such as Richard Hackman (198x) has done much to help us
understand what behaviors lead to what results and why – and how the result might differ by
circumstance.7
On the Importance of the Categorization Step
Several prominent scholars have examined the improvement in predictability that
accompanies the transition from the attribute-based categorization of descriptive theory, to the
circumstance-based categorization of normative theory. Consider, for example, the term
“Contingency Theory” – a concept born of Lawrence & Lorsch’s (1967) seminal work. They showed
that the best way to organize a company depended upon the circumstances in which the company
was operating. In our language, contingency is not a theory per se. Rather, contingency is the
categorization scheme, and is a crucial element of every normative theory. Rarely do we find
one-size-fits-all answers to every company’s problem.
Glaser and Strauss’s (1967) treatise on “grounded theory” actually is a book about
categorization. Their term substantive theory corresponds to the attribute-bounded categories of
descriptive theory. And their concept of formal theory matches our definition of normative theory
that employs categories of circumstance.
Management fads often are created when a researcher studies a few successful companies,
finds that they share certain characteristics, concludes that he has seen enough, and then skips the
categorization step entirely by writing a book asserting that if all managers would imbue their
companies with the characteristics of these successful companies, they would be similarly
successful. When managers then apply the formula, some generally find that it doesn’t work in
their companies. This casts a pall on the idea. Some faddish theories aren’t uniformly bad. It’s
just that their authors were so eager for their theory to apply to everyone that they never took the
care to to figure out the circumstances in which their statement of causality would lead to success,
and when it would not. Efforts to identify “the best practices of successful companies” almost
uniformly suffer from this problem.8
Unfortunately, it is not just authors-for-profit of management books that contribute to the
problem of publishing theory whose applicability is not guided by categorization. Reading a
typical scholarly management journal today can be depressing – because the vast majority of
published papers devote few of their column inches to categorization. When the existence of
different categories is noted, often they are “handled” with dummy variables or by omitting the
outliers – as if maximizing R2, rather than getting the categories clearly characterized, is the
hallmark of a good theory.

8

---

Other well-intentioned academics unwittingly contribute to the problem by articulating
tight “boundary conditions” outside of which they claim nothing. Delimiting the applicability of a
theory to the specific time, place, industry and/or companies from which the conclusions were
drawn in the first place is a mutation of one of the cardinal sins of research – sampling on the
dependent variable. In order to be useful to managers and to future scholars, researchers need to
help managers understand the circumstance that they are in. Almost always, this requires that
they also be told about the circumstances that they are not in. In other words, the process of
getting the categories right is an ever-challenging but always important step in theory building.
Some management researchers are convinced that human-laden social systems are so
multi-faceted and dynamic that meaningful simplification into a few categories of circumstance is
impossible. They assert that managers’ world is so complex that there are an infinite number of
situations in which they might find themselves. Indeed, this seemingly infinite complexity of
categories very nearly characterizes the descriptive theory phase in some fields. But normative
theory generally is not so confusing. Researchers in the normative theory phase resolve confusion
by abstracting up from the detail to define a few categories – typically two to four – that comprise
salient circumstances.
As noted above, Thomas Kuhn (1962) discussed in detail the transition from descriptive to
normative theory in his study of the emergence of scientific paradigms. He described a
preliminary period of confusion and debate in theory building, which is an era of descriptive
theory. His description of the emergence of a paradigm corresponds to the transition to normative
theory described above. Even when a normative theory achieves the status of a broadly believed
paradigm, it continues to be improved through the process of discovering anomalies, as we
describe above.9 Indeed, the emergence of new phenomena – which probably happens more
frequently in competitive, organizational and social systems than in the natural sciences – ensures
anomaly-seeking researchers of management that there probably will always be additional
productive laps up and down the theory pyramid that they can run.
Finding the Salient Boundaries of Circumstance-based Categories
If circumstance-defined categorization is so critical to the building of normative theory,
how do researchers decide what boundaries best define the categories, and what potential
definitions of boundaries are not salient to accurate prediction and understanding? Returning to
our account of aviation research, the boundaries that defined the salient categories of
circumstance are determined by the necessity to pilot the plane differently. If a different
circumstance does not require different methods of piloting, then it is not a meaningful category.
We propose that the same principle defines the salience of category boundaries in management
theory. If managers find themselves in a circumstance where they must change actions or
organization in order to achieve the outcome of interest, then they have crossed a salient boundary
between categories.
The Value of Anomalies in Building Better Theory
As indicated before, when researchers in both the descriptive and normative stages cycle
down from the top of the pyramid using statements of association or causality to predict what they
will see at the foundation, they often observe something that the theory did not lead them to

9

---