Вы находитесь на странице: 1из 10


Vol. 22, No. 5, SeptemberOctober 2011, pp. 13121321

issn1047-7039 eissn1526-5455 11 2205 1312
Ive Got a Theory PaperDo You?: Conceptual,
Empirical, and Theoretical Contributions to Knowledge in
the Organizational Sciences
Zur Shapira
Department of Management, Stern School of Business, New York University, New York, New York 10012
s the eld of managements devotion to theory too much of a good thing? [Hambrick, D. C. 2007. The eld of
managements devotion to theory: Too much of a good thing? Acad. Management J. 50(6) 13461352]. In his paper,
Hambrick criticizes the practice employed by many journals in the management eld that requires that papers submitted for
publication make a strong theoretical contribution. I argue that part of the problem is caused by the misunderstanding and
misuse of the term theory. To clarify the status of theory, I review three modes of research formulation in the organizational
sciences: theories, models, and conceptual frameworks. Language plays an important role in scientic research. I therefore
discuss two research languages that are used in research in management that appear to be the farthest apart: mathematics,
which is the language of precision; and narratives, which is the language that provides rich data. I provide a discussion of
the use of mathematics in theory development and the use of narratives in research development. The two languages and
three modes of research formulation are needed for contribution to knowledge, which should be the main goal of research
in organization science.
Key words: theory; model; mathematics; narratives
History: Published online in Articles in Advance April 29, 2011.
There is nothing so practical as a good theory.
(Lewin 1952, p. 169)
In a provocative article, Hambrick (2007) laments the
management elds devotion to theory. He criticizes the
practice employed by many journals in the eld that
requires that papers submitted for publication make a
strong theoretical contribution. He points at a confusing
state of affairs in the management eld where an author
who conducts valuable empirical research meets a bar-
rier upon submitting a paper to a journal. Often, authors
are told that because there is no contribution to theory
in their paper, it has to be rejected. As Hambrick (2007)
notes, this practice does not help the science of organi-
zations and can hinder its progress (see also Corley and
Gioia 2011). Although I am familiar with this practice
of journals requiring contribution to theory, it appears
that the word theory is misused in the management
eld, and this may be part of the reason journals tend to
reject manuscripts that otherwise would not be rejected.
I share his concern but nevertheless think that the seed of
the problem lies in an erroneous understanding of what
a theory is.
A theory signies the highest level of inquiry in sci-
ence. It is a formulation of the relationships among the
core elements of a system of variables that ideally is
arrived at after overcoming multiple hurdles and sev-
eral stages of renement and empirical testing. Unfor-
tunately, I have come to realize that the term theory
paper is often used in the management eld merely to
describe a nonempirical paper. I have often heard a man-
agement scholar saying that he or she has a theory paper,
but in many such papers there was no theory, which
came on top of having no data. This, along with state-
ments about contribution to theory being a necessary
condition for publication in top management journals,
creates real confusion as to what the term theory actu-
ally means in the management eld. In the rst part of
this piece, I attempt to clarify the importance of theory
by discussing different forms of research formulation
(conceptual frameworks, models, and theories) that pro-
vide a wide range of approaches to formalizing research
in an increasing level of rigor.
Research, whether empirical or theoretical, is
expressed in a language for description and commu-
nication. The second part of the paper is devoted to
two major languages that are often used in management
research: mathematics and narratives. I describe each
language and show how using each can lead to scien-
tic development. These two languages are at the two
extremes of formality, where mathematics is a language
that describes ideas in a very precise way but at times at
the expense of the richness of a domain. Narrative is a
form of research language that provides rich descriptions
but often at the expense of precision. I end this paper by
Shapira: Conceptual, Empirical, and Theoretical Contributions to Knowledge in the Organizational Sciences
Organization Science 22(5), pp. 13121321, 2011 INFORMS 1313
arguing that the goal of scientic research in manage-
ment should be contribution to knowledge that is based
on a combination of conceptual, theoretical, and empiri-
cal work. Research in management often starts by identi-
fying questions that are observed in the eld where nar-
ratives provide the description of the phenomena under
investigation. When possible, a conceptual framework
emerges that may lead to model development and ulti-
mately to theory construction. The latter often requires
the use of a more formal language such as mathematics.
Theories, Models, and Conceptual
A Theory: A Coarse Denition and a Few Examples
The scope of this paper does not allow a thorough dis-
cussion of what a theory is, and this presentation is
brief and incomplete. Yet this discussion is needed so
that one can know whether he has a theory or not, as
well as to motivate the comparison to other avenues
of knowledge creation (models and conceptual frame-
works) and to discuss how empirical research is linked
to theory development. A theory is commonly dened as
an analytic structure or system that attempts to explain
a particular set of empirical phenomena. Theories dif-
fer in depth and scope; there are theories that attempt
to cover many phenomena, such as Einsteins attempt
to develop a general theory that would unite gravitation
with quantum mechanics. In the social sciences, Simon
devoted the latter part of his life to developing a gen-
eral theory of problem solving. In contrast, an exam-
ple of a more specic theory in the social sciences is
Tverskys (1972) elimination by aspects, which is a
very elegant theory of choice that makes clear assump-
tions and derives specic predictions, several of which
were tested in different choice contexts (see Fader and
McAlister 1990). Prospect theory (Kahneman and Tver-
sky 1979) is another example of a theory of choice that
has had an enormous inuence in the social sciences.
Without attempting to survey the rich literature on the
essence of theory in the philosophy of science (see, e.g.,
Lakatos 1978, Popper 1959), I highlight a few general
aspects of the notion of theory.
(1) A theory is constructed to provide a coherent
explanation of a set of observed phenomena.
(2) Theories make assumptions and, based on them,
draw logical derivations. Those derivations lead to spe-
cic predictions regarding the subject matter with which
the theory deals.
(3) A theory should be formulated in a way that
makes it clear how it can be refuted or falsied.
(4) The ultimate test of a theory is achieved by com-
paring its predictions to reality. Thus, a theorys predic-
tions are subject to a false/true test.
Of the many theories in the natural sciences, perhaps
the most famous is the relativity theory in physics. Many
think of Einsteins theory of relativity as one of the
greatest achievements in scientic inquiry. Its publica-
tion (both the special and the general versions) changed
an entire way of thinking and conducting research in
physics and in the sciences in general. Relativity theory
was not just a theorem about a unique relation between
two variables; rather, it dened a four-dimensional space
that led to a new way of thinking about the relation
between time and space. It stated that there is an upper
limit to speed, which is the speed of light and is constant.
Based on its assumptions, several testable hypotheses
and predictions were derived. One of these derivations,
E = MC
, is perhaps the most well-known equation in
science at large. General relativity theory predicts that
light bends when it travels in the neighborhood of mas-
sive objects such as the sun, a phenomenon known as
light deection. This prediction was tested by measuring
the change in position of stars on the celestial sphere
as they passed near the sun. Eddington and his collab-
orators performed the measurement in 1911 during a
total solar eclipse simultaneously in the cities of Sobral,
Cear, Brazil and So Tom and Prncipe on the west
coast of Africa (e.g., Dyson et al. 1920). The measure-
ments conrmed Einsteins predictions and corroborated
his theory.
An example of an inuential theory in the social
sciences is game theory (see Von Neumann and
Morgenstern 1944). Its inuence increased over the
years, and it became a major building block of modern
economics. The formal ways in which game theory deals
with competition, cooperation, coordination, matching,
and more led Varian to state, Indeed, most economic
behavior can be viewed as a special case of game the-
ory (1992, p. 259).
I think that part of the confusion among management
researchers concerning the need for and evaluation of
theoretical work is caused by the use of the term the-
ory in a very loose sense. For example, in the social sci-
ences utility theory is a term used to describe rational
choice in general and is not a theory in the narrow sense
of the word. A more peculiar misnomer is organiza-
tion theory, which actually describes a level of analysis
rather than a theory per se.
What role have theories played in the development
of the organization sciences, and what role should they
play going forward? The eld of organizations actually
emerged with empirical and observational studies (see,
e.g., Taylor 1911), and it was mainly Cyert, March, and
Simon who transformed the body of observations into a
theoretical framework. The books Administrative Behav-
ior (Simon 1947), Organizations (March and Simon
1958), and A Behavioral Theory of the Firm (Cyert and
March 1963) have had a tremendous effect on the eld.
The Carnegie approach pushed the eld of organizations
Shapira: Conceptual, Empirical, and Theoretical Contributions to Knowledge in the Organizational Sciences
1314 Organization Science 22(5), pp. 13121321, 2011 INFORMS
into the realm of science by making assumptions, deriva-
tions, and predictions and by using mathematics and for-
mal language to describe the relations among variables.
The above discussion of theory ts nicely the Carnegie
School tradition, where theories, models, and conceptual
and empirical work led to a major development in the
organization sciences.
A Model
Models are important tools in scientic inquiry. A model
implies a formulation that
(1) derives predictions based on clearly specied
assumptions, and
(2) is precise and falsiable.
The major differences between a theory and a model
are the rst and fourth points described previously as the
criteria for theory. That is, a model does not necessar-
ily need to provide an explanation of the phenomenon
it deals with and does not need to make a claim about
truth. Therefore, a test of a model is not one of a
true/false type but rather a kind of a usefulness
test. For example, Regnier and Harr (2006) developed a
model for predicting hurricane landfall based on histori-
cal data on tropical cyclone tracks and data derived from
existing forecasts to estimate the likelihood of landfall
at a particular location. The goal of the model was to
help local decision makers by evaluating the trade-off
between lead time and forecast accuracy, estimating the
value of waiting for improving forecasts to reduce the
frequency of false alarms. Such a model can be eval-
uated in terms of saving life and damage as well as
cost of operations. The model can prove useful for a
decision maker who needs to decide on evacuation. It
is not intended to explain the way hurricanes develop,
from which the location of landfall can also be theo-
retically derived; this is an endeavor that is much more
Another example relates to the heliocentric model
of the planets movements. Kepler modied the exist-
ing model, and his adjusted model, which was based
on extensive data on the location of the planets, did a
very good job of predicting planetary motions. How-
ever, Kepler did not provide a clear explanation of what
was behind his equations. Indeed, his model lost its
eminence when Newton developed his theory of grav-
itation (see Livio 2009). Keplers equations, however,
could be derived from Newtons theory. Hence, looking
at Keplers equations from a Newtonian perspective pro-
vides an example of a theoretical modelthat is, a
model derived from a theory rather than from data.
Examples about the use of models based on data
rather than a theory are found in the social sciences as
well. One such arena has been the explanation and pre-
diction of the effects of star actors on the nancial suc-
cess of movies. The data-based models by Ravid (1999)
and Elberse (2007) provide sophisticated analyses for
the prediction of future movies success. Lampel and
Shamsie (2000) propose a wider theoretical background
that examines how information asymmetries inuence
competitive dynamics in the lm industry to explain the
determinants of as well as predict the success of new
lms. Another example is depicted by recent develop-
ments in imaging (functional magnetic resonance imag-
ine) that do not go beyond correlational analysis, and
there is no theory at the moment that can explain the
complex processes in the brain that cause blood to ow
from one area to another. However, as more data get col-
lected, better predictions will be made about the associ-
ation between responses to different questions and activ-
ity in different domains of the brain.
Examples of useful models abound in the natural sci-
ences and are very common in economics. One area
that has developed over the years at Carnegie and other
places is the use of computer simulations to study pro-
cesses of thinking (Newell and Simon 1972). Realizing
that the use of mathematics for arriving at closed-form
solutions is rather restrictive in studying phenomena
such as human and social behavior, many researchers
built simulation models to get better insights about
social and organizational phenomena (see Burton and
Obel 2011).
Models are precise, especially if they are formulated
in mathematical terms. At times, however, researchers
approach new domains that do not allow them to
use precise symbols to describe the phenomena they
are studying. In such cases, researchers try to build
conceptual frameworks, which may not be as specic as
models but may provide a general system of organizing
the observations.
Conceptual Frameworks
Theories and models differ from conceptual frameworks
in that they make testable predictions. Work of this type
may not be possible in, say, an initial stage of scientic
inquiry in a new domain. At such a stage, scientists may
seek to develop frameworks that help organize empirical
observations by using coherent and meaningful frame-
works. Such frameworks allow scholars to make sense
of the eld and understand its boundaries, major nd-
ings, and challenges. Thus, in comparison with the four
criteria for theory and the two for a model described
above, the criteria for a conceptual framework are that it
(1) provides a structure to organize observations, and
(2) describes the structure in a clear and precise
Research in the biological and life sciences often uses
classications and categorizations as a main research
method. Paleontologists use such methods to create
order among many different phenomena. The original
study of the evolution of species (Darwin 1859) started
with a collection of empirical data and developed a more
comprehensive framework to account for the data. This
Shapira: Conceptual, Empirical, and Theoretical Contributions to Knowledge in the Organizational Sciences
Organization Science 22(5), pp. 13121321, 2011 INFORMS 1315
framework had an immense inuence on thinking in
both the biological as well as the social sciences. It
serves as an example of a domain that was developed in
an inductive manner into a comprehensive framework.
A recent treatise by Dawkins (2009) argues that the gen-
eral framework of evolution is superior to many other
perspectives on the development of life that are based
on nonscientic assumptions.
Organizational change can serve as an example of
a conceptual framework in organizational behavior and
theory. This framework helps describe the many forces
that operate in organizational settings, some of which
facilitate change while others do not. Lewins (1952)
eld forces analysis does a great job of providing a
schematic representation of such forces. There are other
treatments of organizational change, and some models
of change have been developed. However, although the
framework may be useful in analyzing cases of change
in organizations, it has not developed into a coherent
set of assumptions, derivations, and predictions as game
theory has, for example.
A conceptual framework does not necessarily make
strong assumptions the way a theory does, and it may not
be as tightly structured as a mathematical or a computa-
tional model. Yet a good conceptual framework may lead
to new insights and may open new avenues of thinking
on particular phenomena. Its ultimate test, so to speak, is
whether it leads to a better organizing of the major issues
in a particular domain of inquiry. Such organization can
enhance our understanding and may eventually lead to
developing models for prediction and ultimately to the-
ories that explain the nature of the domain of inquiry.
The Role of Language in Scientic Progress
and Theory Development
Researchers need to communicate with each other about
their ideas, conjectures, and ndings. To communicate,
they need to use a common language that they and their
community understand. There are different languages
that can be differentiated by the degree to which they
are precise on one hand and rich on another. Usually,
the richer a description, the less precise it is, and vice
versa. Consider, for example, research in strategic man-
agement that attempts to analyze variations in perfor-
mance among rms. Authors of such papers can say, for
example, that variable X affects performance so that the
higher X is, the higher the performance is. Such a state-
ment is not as precise as writing performance =2X. The
quantitative expression is more precise than the verbal
statement. Of course, the researcher can say it in words:
As X increases, performance increases in a double-fold
manner; this is precise but clumsy. The mathematical
expression is more parsimonious and provides a better t
with Poppers (1959) criteria for scientic expressions.
At the other end lies a situation where a researcher
is observing a group discussion in a foreign country
in a language that he or she does not understand. To
communicate the essence of his or her observations, the
researcher will need to use a narrative format that will be
relatively rich but may not be precise. Verbal theories are
much more ambiguous than mathematically formulated
theories (Harris 1976), but in some situations richness
may be a better way to describe the research context
than a more precise language.
In constructing theories, models, and conceptual
frameworks, a researcher can use different languages
such as mathematics, simulations, and graphical tools,
as well as verbal description and narratives. According
to classical treatment by philosophers of science (Pop-
per 1959), a theory has to be parsimonious; that is, if
two theories are offered for explaining the same phe-
nomenon, and do so with a similar degree of success,
the one that is more concise and shorter is thought to
dominate the other. Ultimately, this view almost man-
dates that the language of science be mathematics and
that theories should be formulated with mathematical
tools. Such an argument can be valid in situations where
the domain of investigation is mature enough to allow
precision in theorizing. When a domain of inquiry is
still in its nascent stage, the use of mathematical tools
may be premature, and the development of knowledge
at such a stage may benet more from the use of nar-
rative and other less formal tools. When such a domain
develops further, mathematics can be used to help sort
out good theories from weaker ones. Researchers should
use a language that matches the stage of the problem
they are studying. Descriptive narratives should be used
in the rst stage of a eld study (along with the col-
lection of hard facts/data) to get the perspectives of the
participants on the phenomenon. As the research project
makes progress and certain patterns emerge from the
data, models can be developed using formal language
such as mathematics.
The discussion in the rst section follows arguments
made by the psychologist and philosopher of science
Meehl (1967), who subscribed to the Popperian tradition
and argued that science makes progress in a cumulative
manner. He claimed that in an advanced eld of science
such as physics, theories make point predictions about
parameter values. Theory development progresses by
attempting to surmount hurdles that are increasingly dif-
cult. To do this, theories need to be formulated mathe-
matically, and tests of a theory should be framed as tests
of specic point predictions. If a theory passes a test
namely, that a point prediction has been supportedthat
point becomes the null hypothesis in future tests. That
point prediction is contrasted against an even more dif-
cult point prediction arrived at by developing the the-
ory further. This approach can be contrasted with the
testing of no-difference null hypotheses. Hypotheses
of this type can be rejected by merely increasing the
sample size. Thus, although signicant differences can
Shapira: Conceptual, Empirical, and Theoretical Contributions to Knowledge in the Organizational Sciences
1316 Organization Science 22(5), pp. 13121321, 2011 INFORMS
be found, many times these differences are meaning-
less, and the value of engaging in testing no-difference
hypotheses for accumulating knowledge remains dubi-
ous. That is, meaningfulness is often sacriced for sig-
nicance, whether it is applied in comparing two groups
in an experiment or in testing the difference of a coef-
cient from zero in regression analysis.
Mathematics: The Language of Precision in
Scientic Inquiry
The process of knowledge accumulation in management
research is interrupted at times when multiple theories
and models within the same or related domains coexist,
even though they make different predictions regarding
the same behavior. This raises the question of how we
can know which theory is correct or how to test and
advance either theory or both. In this section I suggest a
way of testing the predictions of such theories and show
that the language of mathematics provides great help in
this process. To illustrate my perspective, I have chosen
Lockes goal-setting model and Atkinsons (1957) the-
ory of achievement motivation. Locke (1968, p. 167) has
asserted that studies based on goal setting atly con-
tradict his [Atkinsons] theory. Before turning to how
empirical research can address this contention, I briey
describe both approaches.
Both Atkinson and Locke make predictions about
where an individuals performance reaches its maximum.
Atkinson drew on expectancy theory and focused on task
difculty, whereas Lockes model describes motivation
to perform based on intentions and goals. A main dif-
ference between the two is their prediction of the point
where performance reaches its maximum. Atkinsons
mathematical presentation of achievement motivation
theory provides a point prediction of 0.50 for the prob-
ability of success on a task that high-need achievers
would choose; they would eventually reach their best
performance there. The theory is represented by the fol-
lowing formula:
T =(M
) P
(1 P
), (1)
where T is the tendency to engage in achievement-
related tasks; M
and M
are the motives for success
and avoidance of failure, respectively; P
is the probabil-
ity of success on the task; and (1 P
) is the incentive
value of success.
Lockes (1968) goal-setting model predicts that if peo-
ple accept goals assigned to them, then their perfor-
mance would be at the maximum on difcult goals. It
can be stated differently as follows: the harder the goal,
the better the performance is. Because the goal-setting
model does not provide a point prediction, it is difcult
to compare the predictions of this theory with those of
Atkinsons achievement motivation theory.
Comparing the Theories Predictions. One way to
compare these theories predictions is by testing the fol-
lowing hypotheses:
: Performance on achievement tasks reaches its
maximum at P
: Performance on achievement tasks reaches its
its maximum at P
where P
is the probability of success on the task. The
null hypothesis (H
) represents Atkinsons prediction,
and H
is consistent with Lockes prediction that perfor-
mance is higher on more difcult goals (tasks).
Although this approach is familiar, there are some
issues that should be considered. According to Meehl
(1967), in comparing two theories whose predicted dis-
tributions have means of M
and M
, the following prob-
lems arise.
Problem 1. Point null hypotheses (e.g., H
: M
, or H
: M
=0) are always (quasi) false as sig-
nicant differences can be detected only due to a large
sample size.
Problem 2. A directional null hypothesis (e.g.,
: M
- M
) does not generate a theoretically expected
distribution because it is not precise (Cohen 1988; Meehl
1967, p. 105). Employing the directional null hypothe-
sis creates no hurdle for a theory to surmount, which
is a major criterion for theory development. In other
words, the prior probability of rejecting H
in favor of
as formulated in (2) approaches 1/2. Put differently,
because H
is always (quasi) false, properly run experi-
ments are going to provide data that would support H
in about half of the cases, regardless of whether it has
any merit.
In an alternative approach to comparing the theories
predictions, one could also treat Lockes model as pre-
dicting that peoples performance would be best when
the probability of success is, say, only P
=0.30. In such
a case, the following hypotheses could be tested:
: Performance on achievement tasks reaches
its maximum at P
: Performance on achievement tasks reaches
its maximum at P
Equation (3) is superior to testing the theories with a
directional null hypothesis as described in (2). Yet the
predictions described in (2) cannot provide support for
Lockes model over Atkinsons, or vice versa. I realize
that there may be people who subscribe to the goal-
setting approach who may not accept the idea that it can
be described by a hypothesis of the type H
: P
Following Meehl (1967), I recommend an alternative
procedure in which the predictions of the two theories
are compared using a third theory that was developed
in the same domain of inquiry. This approach enables a
researcher to generate point predictions that are consis-
tent with but not generated by the original theory.
Shapira: Conceptual, Empirical, and Theoretical Contributions to Knowledge in the Organizational Sciences
Organization Science 22(5), pp. 13121321, 2011 INFORMS 1317
Comparing the Two Theories Using Point Predictions.
In the absence of alternative point predictions for at least
one of the two theories, theories can be compared using
a third model, like the one I developed (Shapira 1976,
1989). The model, developed in the area of intrinsic
motivation, describes choice behavior in an expectancy
theory framework. Specically, Shapira (1976) proposed
an intrinsic valence function that relates in an acceler-
ated manner to task difculty. Mathematically, intrinsic
valence (denoted I
) is described by
=c(1 P
, (4)
where P
is the probability of successful performance on
a given task ], c is some constant that depends on the
unit of measurement, and k is an individual differences
parameter. Assuming an expectancy-valence framework,
the intrinsic force toward choosing a particular task ] is
represented by
c(1 P
, (5)
where P
is the probability of success on the specic
task ]; E
is a family of functions that depends on
and k, and it gets its maximum at P
= 1,(k + 1)
(see Figure 1).
Equation (5) proposes a possible reconciliation of
Atkinsons and Lockes formulations. In terms of the
predicted choice of a task, Atkinsons formulation is a
special case of Equation (5) (when c = M
, and
k =1). Because most studies in the achievement theory
paradigm showed that subjects choose tasks whose prob-
abilities of success range from 0.25 to 0.40, the aver-
age value of k in those studies was between 1.5 and 2.
Lockes model can be interpreted (for this example) as
implying a particular value of k, which can be derived
from Equation (5). The family of functions described
Figure 1 Motivation as a Function of Task Difculty
(k=2) F
(k=1, Atkinsons model)
Probability of success


0.20 0.25 0.33 0.50 0.90 0.75
Source. Reprinted from Organization Behavior and Human Deci-
sion Processes, Vol. 44, Zur Shapira, Task choice and assigned
goals as determinants of task motivation and performance,
141165, 1989, with permission from Elsevier.
in Equation (5) all rise to (different) maximum levels
and then decline as a function of the probability of suc-
cess. Thus, one way to describe Lockes model in the
above formulation is to suggest that very difcult goals
may not be accepted. This would lead to a predicted
maximal performance on levels that are more difcult
than Atkinsons model predicts, though not extremely
Mathematics as a language can help in the cumula-
tive development of a scientic domain by sorting out
the better theories and providing them with increasingly
difcult hurdles to surmount. However, several areas in
management are not developed to the degree that they
can be subject to such a process. Before developing the-
ories, knowledge should be accumulated based on the
data and the development of a conceptual framework to
organize the cumulated knowledge. At the nascent level
of inquiry, other methods such as ethnographic research
and the use of narratives would be more appropriate
for scientic development. In the next section I provide
arguments and a short example in favor of using narra-
tives in management research.
Narrative: The Rich Language of Scientic Inquiry
The perspective on empirical research presented in the
section that dealt with the use of mathematics is not
intended to diminish the importance of narrative anal-
ysis in building a body of knowledge in management
research. In fact, efforts need to be made to ensure
that the precision of mathematics does not become the
standard by which all empirical work is judged. Grice
(1989), for instance, proposed maxims about how con-
versations should be conducted, such as for quality and
quantity, and suggested that when people respond to
each other in a conversation their replies should be
brief, transparent, relevant, and truthful. In a sense, these
are quite similar to the requirements I proposed at the
beginning of this paper for theory formulation: theory
should be coherent, logical, falsiable, and subjected to
a true/false test. Extant work and my own experience
suggest that the richness of narrative-based investiga-
tions (see, e.g., Argyres 1999, Feldman and Orlikowski
2011, Garud et al. 2011, Zbaracki et al. 2004) plays an
important role in knowledge accumulation in situations
where the relevant data can be expressed only in a ver-
bal format. Narrative analysis means a very orderly and
precise analysis of the use of words, a process whose
intent is to extract meaning from text.
An advantage of narratives in ethnographic research is
the ability to richly convey the study context. To provide
a general argument for the role of narratives, I compare
Bruners (1990) discussion of the search for meaning
with Simons (1981) search for structure in hierarchical
systems. Bruner (1990) provides a compelling analysis
of the importance of the narrative approach in studying
human thinking. Bruner focuses on the study of the mind
Shapira: Conceptual, Empirical, and Theoretical Contributions to Knowledge in the Organizational Sciences
1318 Organization Science 22(5), pp. 13121321, 2011 INFORMS
and claims that it should concentrate on meaning mak-
ing, which he conceives as the construction of meaning
as it is embedded in culture. He argues that the cogni-
tive revolution has shifted from meaning to information
and from construction to processing and that these are
profoundly different matters (1990, p. 4). He adds that
computing became the model of the mind, and in place
of the concept of meaning, there emerged the concept of
computability. Bruners assertion is akin to saying that,
in focusing on the technical aspects of information pro-
cessing, the study of the mind moved to the analysis
of its hardware at the expense of its software. I believe
that both aspects are important for the development of
management research as a whole.
It is intriguing to compare Bruners discussion to
Simons (1981) treatise about the sciences of the arti-
cial. In such comparison we see the two aspects, for-
mal modeling and narratives, that contribute to theory
development. Both authors are cognitive scientists, but
they take different perspectives on the study of think-
ing and the implied role for narrative-based research.
Simons framework is about the hierarchical structure
of systems, be they natural or social. He attempts to
provide a general structure to which hierarchical sys-
tems can be mapped or reduced. One of Simons main
concepts is modularity, which he uses to describe the
differences between two watchmen, Hora and Tempus,
who assemble watches but get interrupted by phone calls
from their clients. Hora works in a more modular fash-
ion and is able to minimize the effects of interruptions
on his work in comparison with Tempus. In Simons
words (1981, p. 200),
The watches Hora made were no less complex than those
of Tempus. But he had designed them so that he could
put together subassemblies of about ten elements each.
Ten of these subassemblies, again, could be put together
into a larger subassembly; and a system of ten of the
latter subassemblies constituted the whole watch. Hence,
when Hora had to put down a partly assembled watch to
answer the phone, he lost only a small part of his work,
and he assembled his watches in only a fraction of the
man-hours it took Tempus.
From this narrative, Simon builds a simple formal model
in which he assumes a certain probability that each of
the watchmen can complete a watch without interrup-
tion and shows that it will take Tempus much longer to
assemble a watch than it does Hora.
Simons discussion provides an impressive description
and analysis of the way a hierarchical system can be bro-
ken into modular parts and hence become more efcient
and amenable to formal modeling. But some additional
questions can be raised, such as the manner in which the
two watchmen respond to phone calls, who those phone
calls come from (there are differences among clients), as
well as each watchmans identity in pursuing his work.
A management researcher who is interested in study-
ing such questions can do so by using narratives rather
than formal modeling. The point of the above discussion
is not only to contrast formal modeling and narrative
approaches but to suggest that the two perspectives can
be combined when an attempt is made to study a related
phenomenon like managerial work in the eld, an exam-
ple of which follows below.
A Tale of a Research Project on Managerial Work. I
used to teach a graduate course on motivation and pro-
ductivity, where I took my students to different locations
such as factories and emergency rooms. I was intrigued,
while visiting a textile factory for some years, by the
question of how foremen supervised two (or more)
assembly lines and managed to deal with employee
absence as well as to plan the work schedule for the
following week. I observed and interviewed the fore-
men and listened to their narratives when they explained
the way they handled these issues. I took notes of those
interviews, and years later I talked about them with my
colleague; we then embarked on a project that culmi-
nated in a mathematical model that described managerial
work in the face of interruptions (Seshadri and Shapira
2001). The seeds of the model were based on the way
foremen improvised when some employees were absent;
the foremen shifted employees between two lines, for
example. These attempts looked like trial and error,
but even though they appeared to be random, the fore-
men experimented with such moves so that the outcome
appeared to be small improvements on some trials and
small declines on others. It suggested to me at the time
that this trial and error could be modeled by a ran-
dom walk, where at each point in time a unit is equally
likely to take a step either up or down. When one speeds
the (theoretical) process to smaller and smaller steps in
smaller and smaller units of time, one arrives at Brown-
ian motion.
Granted, the Brownian motion model, Seshadri and
I developed to describe the effects of interruptions was
completely foreign to the foremen I observed. Yet had
I told them about the intuitive results of the simulations
we ran while testing the model, they would have most
likely agreed with them, even though simulation model-
ing was not their area of expertise. One may ask why I
needed to develop the model. The reason is that it pro-
vided me with a useful way to describe the phenomenon
and to lay down the basis for further developments in
the study of how interactions among tasks and interrup-
tions affect the attention of managers and their work.
In contrast, the model would have been rather sterile
had I not spent the time observing, interviewing, and
learning from the foremen narratives. March (2010) pro-
poses a similar point when he discusses the interplay of
models and stories. He argues that both play a role in
learning from experience, especially if the experience is
Shapira: Conceptual, Empirical, and Theoretical Contributions to Knowledge in the Organizational Sciences
Organization Science 22(5), pp. 13121321, 2011 INFORMS 1319
This paper raised two thoughts: rst, that research
formulation can take on the forms of conceptual
frameworks, models, and theories; and second, that there
are different languages of research. I presented two of
them, mathematics and narratives, which vary in their
precision and richness. I believe that these ideas can
help resolve part of the problems that Hambrick (2007)
nicely identied. I discuss a few issues that underlie
these ideas.
First, I see organization science as a eld where
knowledge is progressing in a cumulative manner. In
that respect, the different modes of research formulation
I addressed can describe sequential progress in research
in a new domain. An example for the beginning of such
a research project is provided in Bechkys (2011) work,
which focuses on scientists working in forensic labs.
This is a nascent project, and Bechkys work is ethno-
graphic and uses narratives as a main source of data.
At its inception there was no need to develop a formal
theory. Later, based on the ndings she has collected,
she can move to develop a conceptual framework, to be
followed by a model (not necessarily mathematical) and
ultimately a theory. This is a cumulative process, and the
contribution to knowledge may happen at every stage of
this extensive project.
Second, scientic theories improve by being subjected
to falsication and disproof. In this respect, one can see
the benet of using mathematics, because embedded in
it is proof (or disproof) by negation. Any theorem or a
statement about relations among objects that is made in
a general way can be disproved with one contradictory
example. The advantage in the parsimony of such a pro-
cess is clear. If organization theories are formulated in
way that falsication is a clear option, we will have a
much better pace of theoretical development.
Third, at times I have heard researchers in manage-
ment claiming that they proved their theory with empir-
ical ndings. Unfortunately, this is not possible. Proofs
are possible only in a tightly structured system of rules,
such as in logic or mathematics. Any empirical nd-
ings in any scientic eld face the unhappy future where
contradictory nding can emerge. No ndings can be
proved in the face of an uncertain future. We often con-
fuse corroboration with proof. Theories become stronger
if they surmount more difcult hurdles, and as Meehl
(1967) pointed out, this should increase our condence
in them. However, more corroboration is not proof. More
than that, as Wason (1968) observed, human observers
display a tendency known as the conrmation bias,
where we look for more positive evidence in support
of our claims, ideas, or conjectures. Unfortunately, this
is neither efcient nor a logical process. Knowledge
progresses much faster by disproof (Platt 1964), so in
addition to the idea of opening theories for falsication,
we should be critical of studies that we conduct, whether
empirical or conceptual, and ask ourselves if collecting
more data is warranted and evaluate the data against the
logical mechanism of disproof. There are more and more
meta-analytic papers that evaluate cumulative research.
They are very valuable, so before engaging in a new
research project in a domain where meta-analysis has
been conducted, one should begin by asking oneself if
generating additional data in this domain also means cre-
ating new knowledge.
Fourth, theory construction should not be confused
with quantitative methods for data analysis. Researchers
who use advanced statistical methods in econometrics
may think that the above arguments about the use of
mathematics are achieved by using such methods. The
level of sophistication in data analysis in research in
the organization sciences is very high. However, there
is a fundamental difference between theory construction
using mathematical tools and data analysis that also uses
such tools. In both cases, the tools can use the same
mathematical symbols, but the sophisticated data analy-
sis does not substitute for poor model building. In the
spirit of Meehls (1967) arguments, we can apply a very
sophisticated regression analysis and nd that a certain
coefcient is signicantly different from zero, but the
question is whether there is meaningfulness beyond the
signicance that was veried.
We often use more powerful and more sophisti-
cated statistical analysis to identify differences among
groups. Take, for example, two large groups of people
(5,000 each) differing by, say, gender or education. I
can guarantee that you will nd a signicant difference
in the average height between the two groups, even if
the difference would be as miniscule as 0.5 mm. This
is an example of a signicant but meaningless result. In
contrast, as Meehl (1967) argued, if we have a math-
ematical model with a parameter (such as the one in
Equation (5)), we can subject the parameter value to
increasing scrutiny. For example, if a value of k = 2.5
was found in one study, one can compare it against an
alternative value of, say, k =2.8 based on some theoret-
ical argument. This way, the mathematically formulated
model allows us to make progress in theory develop-
ment by subjecting new parameter values to a test. The
Newton gravity constant, for example, did not emerge by
merely comparing the average speed of different groups
of falling objects; it was derived from a theoretical for-
mula that combined the mass of and distance between
two bodies. There are many constants in the natural sci-
ences, and most of them were arrived at by theoretical
development combined with precise measurement. There
are not many constants in the organizational sciences,
perhaps because dealing with human and social variables
is more complex. The important goal of theory develop-
ment is to nd meaningfulness, not merely signicance.
It is the former that leads to scientic progress, with the
help of the latter (see also Shaver 2008).
Shapira: Conceptual, Empirical, and Theoretical Contributions to Knowledge in the Organizational Sciences
1320 Organization Science 22(5), pp. 13121321, 2011 INFORMS
Fifth, even though the current paper emphasizes the
idea that theory construction is highly facilitated by the
use of mathematical tools, the importance of languages
other than mathematics is emphasized as well. An anec-
dote serves to illustrate this point. The composer Ludwig
Van Beethoven was once asked by a critic what ideas he
tried to express in composing a certain musical piece.
Beethoven responded by saying that had he been able
to write down the ideas in words, he would not have
composed the piece. Not every research idea or nding
can be expressed in mathematical tools, and insisting on
this can be achieved at the expense of an increased loss
of meaning.
Finally, going back to the problem identied by
Hambrick (2007), I personally see theory construction as
the highest level of scientic inquiry. However, the orga-
nization science eld progresses in a cumulative man-
ner. We need more rigorous empirical research, newer
conceptual frameworks, models, andat the endbetter
theories. The practice of requiring contribution to theory
from every submitted paper hinders rather than facili-
tates progress. As I pointed out, that part of the prob-
lem arises from a misunderstanding of the term theory,
which I hold in high regard. We need to look for research
that uses appropriate language and provides high-quality
empirical studies, conceptual frameworks, models, and
theories, as long as it contributes to our knowledge in
the organization eld.
This paper benetted from the comments of Linda Argote,
Elizabeth Boyle, Gary Dushnitsky, Joseph Lampel, and partic-
ipants at the WIP seminar at New York University. The author
is also indebted to the late Ralph Alexander and to John Miller
for introducing the author to the writings of Paul Meehl. All
remaining errors in the manuscript are the authors.
The sociologist Guttman developed a meticulous methodol-
ogy for developing a theory based on initial observations and
some organizing principles, a process that he called facet anal-
ysis (Guttman 1954). For an application of facet analysis to
theory construction in organizational behavior, see Shapira and
Zevulun (1979).
Argyres, N. S. 1999. The impact of information technology on coor-
dination: Evidence from the B-2 Stealth bomber. Organ. Sci.
10(2) 162180.
Atkinson, J. W. 1957. Motivational determinants of risk-taking behav-
ior. Psych. Rev. 64(6) 359372.
Bechky, B. 2011. Science and scrutiny: The work and culture of
a crime lab. Working paper, University of California, Davis,
Bruner, J. 1990. Acts of Meaning. Harvard University Press, Cam-
bridge, MA.
Burton, R. M., B. Obel. 2011. Computational modeling for what-is,
what-might-be, and what-should-be studiesAnd triangulation.
Organ. Sci. 22(5) 11951202.
Cohen, J. 1988. Statistical Power Analysis for the Behavioral Sci-
ences, 2nd ed. Lawrence Erlbaum Associates, Hillsdale, NJ.
Corley, K. G., D. A. Gioia. 2011. Building theory about theory build-
ing: What constitutes a theoretical contribution? Acad. Manage-
ment Rev. 36(1) 1232.
Cyert, R. M., J. G. March. 1963. A Behavioral Theory of the Firm.
Prentice Hall, Englewood Cliffs, NJ.
Darwin, C. 1859. On the Origin of Species. John Murray, London.
[Reprinted 1964, Harvard University Press, Cambridge, MA.]
Dawkins, R. 2009. The Greatest Show on Earth: The Evidence for
Evolution. Free Press, New York.
Dyson, F. W., A. S. Eddington, C. Davidson. 1920. A determination
of the deection of light by the suns gravitational eld, from
observations made at the total eclipse of May 29, 1919. Philos.
Trans. Roy. Soc. London 220A 291333.
Elberse, A. 2007. The power of stars: Do star actors drive the success
of movies? J. Marketing 71(October) 102120.
Fader, P. S., L. McAlister. 1990. An elimination by aspects model
of consumer response to promotion calibrated on UPC scanner
data. J. Marketing Res. 27(3) 322332.
Feldman, M. S., W. J. Orlikowski. 2011. Theorizing practice and prac-
ticing theory. Organ. Sci. 22(5) 12401253.
Garud, R., R. L. M. Dunbar, C. A. Bartel. 2011. Dealing with unusual
experiences: A narrative perspective on organizational learning.
Organ. Sci. 22(3) 587601.
Grice, H. 1989. Studies in the Way of Words. Harvard University
Press, Cambridge, MA.
Guttman, L. 1954. An outline of a new methodology in social
research. Public Opinion Quart. 18(4) 395404.
Hambrick, D. C. 2007. The eld of managements devotion to the-
ory: Too much of a good thing? Acad. Management J. 50(6)
Harris, R. J. 1976. The uncertain connection between verbal theories
and research hypotheses in social psychology. J. Experiment.
Soc. Psych. 12(2) 210219.
Kahneman, D., A. Tversky. 1979. Prospect theory: Analysis of deci-
sion under risk. Econometrica 47(2) 263291.
Lakatos, I. 1978. Mathematics, Science and Epistemology. Cambridge
University Press, New York.
Lampel, J., J. Shamsie. 2000. Critical push: Strategies for creating
momentum in the motion picture industry. J. Management 26(2)
Lewin, K. 1952. Field Theory in Social Sciences. Tavistock, London.
Livio, M. 2009. Is God a Mathematician? Simon & Schuster,
New York.
Locke, E. A. 1968. Toward a theory of task motivation and perfor-
mance. Organ. Behav. Human Performance 3(2) 157189.
March, J. G. 2010. The Ambiguities of Experience. Cornell University
Press, Ithaca, NY.
March, J. G., H. Simon. 1958. Organizations. John Wiley & Sons,
New York.
Meehl, P. E. 1967. Theory-testing in psychology and physics: A
methodological paradox. Philos. Sci. 34(2) 103115.
Newell, A., H. Simon. 1972. Human Problem Solving. Prentice Hall,
Upper Saddle River, NJ.
Platt, J. R. 1964. Strong inference. Science 146(3642) 347353.
Popper, K. R. 1959. The Logic of Scientic Discovery. Basic Books,
New York.
Shapira: Conceptual, Empirical, and Theoretical Contributions to Knowledge in the Organizational Sciences
Organization Science 22(5), pp. 13121321, 2011 INFORMS 1321
Ravid, S. A. 1999. Information, blockbusters, and stars: A study of
the lm industry. J. Bus. 72(4) 463492.
Regnier, E., P. A. Harr. 2006. A dynamic decision model applied to
hurricane landfall. Weather Forecasting 21(5) 764780.
Seshadri, S., Z. Shapira. 2001. Managerial allocation of time and
effort: The effects of interruptions. Management Sci. 47(5)
Shapira, Z. 1976. Expectancy determinants of intrinsically motivated
behavior. J. Personality Soc. Psych. 34(6) 12351244.
Shapira, Z. 1989. Task choice and assigned goals as determinants of
task motivation and performance. Organ. Behav. Human Deci-
sion Processes 44(2) 141165.
Shapira, Z., E. Zevulun. 1979. On the use of facet analysis in
organizational behavior research: Some conceptual considera-
tions and an example. Organ. Behav. Human Performance 23(3)
Shaver, J. M. 2008. Organizational signicance. Strategic Organ. 6(2)
Simon, H. A. 1947. Administrative Behavior. Macmillan, New York.
Simon, H. A. 1981. The Sciences of the Articial, 2nd ed. MIT Press,
Cambridge, MA.
Taylor, F. W. 1911. The Principles of Scientic Management. Harper,
New York.
Tversky, A. 1972. Elimination by aspects: A theory of choice. Psych.
Rev. 79(4) 143151.
Varian, H. R. 1992. Intermediate Microeconomics. Norton, New York.
Von Neumann, J., O. Morgenstern. 1944. Theory of Games and Eco-
nomic Behavior. Princeton University Press, Princeton, NJ.
Wason, P. C. 1968. Reasoning about a rule. Quart. J. Experiment.
Psych. 20(3) 273281.
Zbaracki, M., M. Ritson, D. Levy, S. Dutta, M. Bergen. 2004. Man-
agerial and customer costs of price adjustment: Direct evidence
from industrial markets. Rev. Econom. Statist. 86(2) 514533.
Zur Shapira is the William R. Berkley Professor of Entre-
preneurship and Professor of Management at the Stern School
of Business, New York University. He is a student of managerial
cognition, risk taking, and organizational decision-making.
In this article, Ive Got a Theory PaperDo You?: Conceputal, Empirical, and Theoretical Contributions to
Knowledge in the Organizational Sciences by Zur Shapira (rst published in Articles in Advance, April 29,
2011, Organization Science, DOI: 10.1287/orsc.1100.0636), Lewin (1952) was misquoted. The quote has been
corrected to read as follows: There is nothing so practical as a good theory (p. 169).