FORMAL MODELLING, COMPUTER SIMULATION AND GROUNDED
FIELD RESEARCH TO CONDUCT ENQUIRY IN MANAGERIAL SCIENCES
Edoardo Mollona, Department of Computer Science, University of Bologna.
INTRODUCTION
With different fortunes and oscillating enthusiasm, computer simulation has supported
theoretical investigation in managerial disciplines since the 60’s. In the attempt to
further corroborate the role of formal modeling and computer simulation in the
repertoire of research strategies available to social scientists, the aim of the present
essay is to sketch out a framework for an enquiry that combines computer simulation
and field-based investigation, this latter a typical research strategy in social sciences.
To begin with, it is important to set up in the front a definition for computer simulation.
Computer simulation has to do with the manipulation of symbols using a computer
code; more specifically, it uses algorithms to derive propositions from the assumptions
that come together in a computer model. A computer model is a formal model in which
‘[...] the implications of the assumptions, that is, the conclusions, are derived by
allowing an electronic digital computer to simulate the processes embodied in the
assumptions’ (Cohen and Cyert 1961: 115).
In this respect, computer models can be regarded as special cases of mathematical
models (Cohen and Cyert 1961) in which conclusions are derived from assumptions by
using a computer simulation rather than a process of analytical solution. On the other
hand, however, computer models not necessarily have to be stated in mathematical and
numerical form (Clarkson and Simon 1960) since they allow manipulation of symbols
that can be words, phrases and sentences. Therefore, computer models make up the
subset of mathematical models that are solved numerically rather than analytically but
not all the computer models are stated in mathematical terms since they may incorporate
not-mathematical symbols. In this respect, Troitzsch suggests that computer simulation
is a third system beside natural language and mathematics (1998: 27).
In principle, computer simulation is just a technologically-aided process of deduction.
Yet, the crude technology can vary strongly from different approaches and, more
importantly, the difference in the adopted technology often unveils profound differences
in the philosophy that lies beneath modeling.
System Dynamics approach is inspired by the idea that behavior of individuals that are
embedded within a social system as determined by the feedback nature of the causal
relationships that characterize the system (Forrester 1958, 1961). In this line, System
Dynamics models reduce aggregate and often puzzling behaviors into underlying
feedback causal structures and, a consequence, typically aggregate agents into a
relatively small number of states, assuming their perfect mixing and homogeneity
(Rahmandad and Sterman 2004).
Taking as an example another modeling and simulation approach, Agent-based
modeling, we observe that underpinning viewpoint is notably different. In Agent-Based
models logic of investigation is typically inclined to show how interaction among
individual decision-making and learning may generate complex aggregate behavior.
Consequently, modeling needs to preserve heterogeneity and individual attributes.
However, independently of the approach adopted and the inspiring philosophy, research
work employing computer simulation has frequently been regarded, in social sciences,
as influenced by an autonomous logic in respect to mainstream research. Too
formalized to effectively capture the complexity of social processes for someone,
leading to results excessively dependent of specific parameters’ calibrations for others.
Simulation studies, however, have a long tradition in organizational research. Going
back to seminal work in the area of the behavioral theory of the firm and organizational
decision theory, some of the most important theoretical pieces are based on a simulation
approach. This is true, for example, for the well known Garbage Can model (Cohen,
March and Olsen, 1972) and for the work leading to the development of The Behavioral
Theory of the Firm (Cyert, Feigenbaum and March 1959; Cyert and March 1963).
In recent times, computer simulations have recuperated terrain in mainstream
management journals and the aim of this essay is to contribute to this process of
legitimization process by laying bare an approach to associate formal modeling,
computer simulation and field study to conduct an enquiry in managerial sciences. The
approach that we describe is conceived of independently of different philosophies
beyond formalization and simulation.
The article is organized as follows; in the next section I consider a sample of recent
works that use simulation and I muse on the differences in the underlying logic of
enquiry. In the following section, I focus on a specific issue: the association of formal
modeling, computer simulation and grounded field research. In the last section of the
chapter I draw some conclusions.
COMPUTER SIMULATION FOR THEORY DEVELOPMENT IN STRATEGY
AND ORGANIZATION
In examining early contributions that adopt a research design that includes formal
modeling and computer simulation, three groups of typical justifications recur to justify
the adoption of this methodology.
First, computer simulation, in comparison to formal analytical approaches, allows
retaining a greater richness of details. Economists that adopted computer simulation
operated in a cultural milieu in which research method, and rhetoric, was erected upon
the rock-hard plinth provided by mathematical modeling. The typical way of proceeding
demanded consequences of modeled assumptions to be deducted by the means of
analytical solution of a mathematical model. The rigor of the approach, however, does
not come without costs since the need to solve analytically a model bounds the
complexity that the model can incorporate. The portrayal, for example, of non linear
relationships among variables introduces in a model a considerable amount of
complexity so to possibly impair its analytical solution.
Under this perspective, we can interpret the candid enthusiasm that permeates writings
of Cohen, for example, who explains that ‘It requires a much more extensive knowledge
of mathematics to obtain an analytical solution to a complex mathematical model than it
does to formulate the model’ and, thus, computer simulation ‘... allows a more flexible
and easy approach and preserves richness of details...’ (Cohen 1960: 535). This
enthusiasm is shared by Orcutt that gives the idea of how powerful computer simulation
appeared to these pioneers as a tool to deal with complex systems:
The use of simulation techniques by the authors of this demographic
study does not, of course, offer any guarantee in itself that they have
produced an acceptable and useful population model. However, by
producing a feasible means of solution it permitted them to introduce
a variety of interactions, variables, nonlinearities and stochastic
considerations into their model which they otherwise would have
been forced to leave out despite strong evidence of their importance.
(Orcutt 1960: 905)
This characteristic rescues the researcher from a typical dilemma. The dilemma requires
a researcher to either abandoning the idea to represent the object of study closely,
thereby accepting costly simplification in order to rigorously generate testable
hypotheses through mathematical analysis, or to preserve complex representations of
the object of study at the cost of producing appreciative theories of behavior that have to
deal with the ambiguity of natural language.
A second motive that is frequently mentioned is that in a computer model the
relationship between assumptions and deducted consequences can be easily manipulated
to account for a variety of changes and amendments in the model structure. The fact that
a computer simulation does not require an analytical solution to derive consequences
from assumptions entails that researchers can explore how modifications in a model’s
structure have an impact on the unfolding behavior of the model without remaining
entrapped into the quandaries of often laborious mathematical analysis. As Cohen
explains:
A further advantage of computer models is the ease of modifying the
assumptions of the theory. When suitable programming languages
become available, relations can be inserted, deleted, or changed in
the model, and only local changes, which can be quickly made, will
be required in the computer program.
Modifications of this kind will have a much smaller effect on the
procedures for simulating a formal model than they would on the
means used for obtaining analytical solutions to the model
(Cohen 1960: 536)
Considering the work done by Hoggatt (1957), for example, Shubik noted that ‘The
number of cases and conditions worked out by Hoggatt would have been unfeasible
without a simulation’ (Shubik 1960: 917). As Cohen and Cyert suggest (1961), the work
of Hoggatt is a good example of how computer simulation may help to revive an old
model (in this case the neoclassical decision model for determining output of firms
given a market price) addressing complex questions that were not practicable with other
techniques of analysis. The easiness in the manipulation of computer models is also
connected to the fact that computer models may be structured in a modular format.
Thus, ‘It is extremely convenient to be able to formulate a complex model in terms of
several component submodels, to deal with each component separately at first, and then
to integrate them into a complete model.’ (Cohen 1961: 45). In similar veins, Gilbert
and Troitzsch (2005) suggest that simulation is more appropriate for formalizing social
science theories than mathematics because programming languages are more expressive
and less abstract than most mathematical techniques and because computer models are
often modular, so that major changes can be made in one part without the need to
change other parts of the program.
Finally, computer simulation allows researchers to generate complex hypotheses of a
system’s behavior that are testable against empirical world. This is because deductions
obtained with computer simulation, besides being as rigorous and reliable as those
obtained through mathematical analysis, may be cast in the form of time series to be
directly compared with observed behaviors. Imagine a theory that predicts, in specified
circumstances, the emergence of a particular behavior over time of a specified variable.
In this case, a verbal description of the behavior has to be compared with empirical
paths of behavior. This verbal description may be ambiguous in comparison to the
string of reported quantities collected over time as appearing into an empirical time
series. On the other hand, computer simulation, to produce hypothesis of behavior,
adopts the same language that is used to collect empirical time series: a string of
quantities reported in specific intervals of time. In this way, computer simulation
improves the capability to generate testable hypotheses of behavior (Meinhart 1966). As
Orcutt explains: computer simulation makes it ‘... possible comparison of generated
results with observed time series and cross sectional data and thus permitted testing of a
sort that would not otherwise have been possible.’ (Orcutt 1960: 905).
THE LOGIC OF ENQUIRY USING COMPUTER SIMULATION
As Cohen and Cyert suggest (1961), computer models are of two types: synthetic and
analytic. In synthetic models, the modeler knows with a high degree of accuracy the
behavior of the component units of the phenomenon under scrutiny. On the other hand,
in analytic models, the behavior of the phenomenon is known and the problem is to
capture the mechanisms that produce the behavior. In this classification, synthetic and
analytic models reveal different underpinning logics of enquiry. While synthetic
computer models are informed by a pure deductive logic, analytic models are
characterized by an inductive logic (Cohen 1961).
To start with, however, a word has to be said to better define what we mean by
inductive or deductive process. More specifically, the associations synthetic/deductive
and analytic/induction may sound not necessarily intuitive.
Deductive process has been acknowledged as a key component of scientific reasoning
since Aristotle. A deductive inference moves from general assumptions to specific
consequences; in this respect, consequences drawn from assumptions have an inferior
degree of universality than their premises. Deductive inferences have two properties;
first, the information embodied in the deducted consequences is more or less explicitly,
included in the assumptions; second, deducted consequences originate necessarily from
assumptions. In other words, if assumptions are correct, deducted consequences must be
correct as well.
On the other hand, inductive processes move from particular instances to general
conclusions. In this respect, in inductive inferences, derived conclusions are not entirely
included in the premises. In other words, the information content in inducted
conclusions is greater that the one crystallized into the premises. Thus, inductive
inferences say something new, or different, in respect to premises; thus, they add
information. This property conceals an hazard because correctness of premises does not
necessarily imply that conclusions are correct as well.
As for the distinction between analytic and synthetic, starting from Kant’s Critique of
Pure Reason (firstly published in 1781), an analytic statement is purely explanatory of
an existing concept and it does not add more information than that already contained
into the concept itself. A classic example reported by Kant regards the statement that
affirms that an entity of matter is extended in the space. The fact that an entity of matter
is extended in the space is already implicit in the definition of entity of matter. It does
not add information regarding the concept entity of matter; rather it provides an
extension, or further explanation, of the concept. On the contrary, a synthetic statement
is extensive because it adds more information than that contained originally in a
concept. For example, the fact that an entity of matter has a weight, explains Kant, is
not included necessarily in the concept of entity of matter (it suffices to think of a state
of absence of gravity) and rather it stems from a synthesis between an original concept
and a quality external to the concept.
Given this distinction between synthetic and analytic statements, Peirce, for example,
put forward a dichotomy between deductive/analytic and inductive/synthetic inferences
(Harshome and Weiss 1931/1935).
Thus, we have to be very careful in interpreting the distinction proposed by Cohen and
Cyert between analytic/inductive and synthetic/deductive, since in their framework the
concept of synthesis pertains to the use of simulation to aggregate local, or partial,
components of a phenomenon, into a global emerging behavior. On the other, analysis
concerns the dissection of behavior of interest into its components, or determinants.
To be clear about the wording we are going to use and to avoid misunderstanding, we
focus on the distinction between computer simulations that adopt a deductive or an
inductive logic of inference and we ignore the dichotomy between analytic and
synthetic computer models. Within this framework, deductive computer models focus
on the specification of a set of mechanisms or processes and explore unfolding
consequences of such specifications whereas inductive computer models move from the
definition of an aggregate behavior and use simulation to test whether candidate
mechanisms or processes are able to determine in vitro, and thus explain, the aggregate
behavior.
We suspect, however, that simulation studies show a much broader variety of
approaches that blend elements of deduction and induction. In addition, in computer
simulations induction and deduction are intertwined in a cyclical process of theoretical
investigation. Induction works when we introduce in a model a casual mechanism that
we deem possibly responsible for an observed behavior. In this case, we run history
backward to reproduce the conditions for the behavior under study to emerge. On the
other hand, once we have found a candidate causal mechanism that we think may
explain observed behavior, we might be interested in understanding how robust is the
relationship between causal structure and the emerging behavior. Additionally, we may
want to understand if the causal mechanism is connected to other possible behaviors. In
other words, we may be interested in the relationship between the causal mechanism, or
a class of similar causal mechanisms, and a class of behavioral phenomena. In both
cases, we can generate a sensitivity analysis by simulating the model with different
calibration of model’s parameters or we can simulate the model with a variety of
modifications in the structure of key causal mechanisms. In this way, we can explore
10
near-histories or hypothetical histories (March, Sproull and Tamuz 1991) in order to
articulate our understanding of a phenomenon. When we run a computer model and we
observe simulated consequences of changes in parameters’ calibration or amendments
in the model's structure, we are embarking into a deductive inference. Thus, deduction
and induction are inseparable in a research design based on computer simulation. We
thus expect differences among simulation studies to be detected in the degree of
accuracy of the description of the elements that compose an aggregate phenomenon or
of the features that characterize the aggregate phenomenon itself. Simulation studies in
which a deductive logic of inference prevails will move from accurate modeling of
components while simulation studies informed by an inductive logic will set forth from
the description of an aggregate behavior.
Nonetheless, maintaining two idealtypes of computer models, deductive and inductive,
seems a good strategy, or at least a safe point of departure, to get the picture of what
logic of enquiry simulation studies have adopted in the field of strategy and
organization. Differently from other typical, qualitative and quantitative, research
strategies that are more legitimized and disciplined, simulation-based research has been
structured in a variety of different guises. Only recently, Davis, Eisenhardt and
Bingham (2007) have convincingly positioned simulation studies among other methods
of enquiry within strategy and organization research developing a roadmap for rigorous
simulation-based research. To carry on this avenue, we apply the two idealtypes to
capture the often subtle differences in the logic underlying simulation studies.
To address typical features of deductive inference in simulation studies, we begin from
the classic Cohen, March and Olsen’s Garbage Can simulation model (1972). The
11
authors do not specify in details a reference mode of behavior to be explained, beyond
the broad idea that they want to address the way in which organized anarchies! embark
in decision-making activity. Rather, the emphasis is on the modeling of the structural
features of decision-making processes in specific types of organizations. The aim is to
develop ‘a behavioral theory of organized anarchy’ (1977: 2). To do so, the authors
develop a model that describes decision making within organized anarchies and
examine ‘...the impact of some aspects of organizational structure on the process of
choice...’ (1972: 2). The structure of the research design encompasses the modeling of
organizational decision-making processes and the analysis of the behavioral
consequences of such modeling.
More specifically, the authors adopt a view of an organization as a garbage can in
which are collected ‘... choices looking for problems, issues and feelings looking for
decision situations in which they might be aired, solutions looking for issues to which
they might be the answer, and decision makers looking for work.’ (1972: 2). Along
these lines, they modeled problems that require a specific amount of energy devoted by
members of the organization to be solved and depicted two matrix structures that
describe organizational features. The first matrix defines the access structure that
associates choices to problems by determining what choice is accessible to what
problem. The second matrix represents the decision structure and associates decision-
makers to choices by establishing what decision maker is eligible to make what choice.
In their experimental design, they portrayed different kind of organizations with
different energy distribution, different problem loads and different organizational
structures. Through simulation experiments, the authors derived emerging decision-
12
making behaviors with typical features. For example, they observed that, depending of
the different assumptions crystallized into the initial calibration of the model,
organizations may show different styles in decision-making and problem-solving.
We define this type of work deductive since the curiosity that triggers the effort of
researchers regards the deduction of typical emerging patterns of organizational
behavior given the description of organizational structures and decision-making
processes.
On the other hand, researchers adopt an inductive inference when they proceed from a
phenomenon, more specifically, from the description of a behavior that unfolds
longitudinally over time, and use computer simulation to select plausible determinants
of the phenomenon among alternative causal mechanisms.
For example, Adner (2002) studied the emergence of disruptive technologies and he set
up his research design by stating at the front the description of the characteristics of the
phenomenon he wanted to investigate. A fter clarifying that his contribution is to explain
the emergence of disruptive technologies, Adner modeled consumers’ individual
preferences and firm technological strategy to obtain mechanisms that are sufficient to
produce the phenomenon.
A similar logic inspires the work of Sastry (1997). Sastry analyzed Tushman and
Romanelli’s verbal theory of punctuated change to demonstrate that the verbal theory
does not contain the necessary causal mechanisms to explain the described behavior.
Sastry conducted a textual analysis of the verbal theory and used qualitative
descriptions of behavior to produce a dynamic behavior to test the theory. Then, she
identified constructs and causal relationships that provided the basis of the formal
model. Once a computer model that formalized key traits of the theory was built, Sastry
13
simulated the model and compared simulated behavior with those crystallized into the
theory. The discrepancy between theoretical and simulated behaviors guided Sastry to
introduce two new mechanisms that were not originally included into the verbal theory
but that proved necessary to produce the behavior purported in the theory. The two
mechanisms are a routine for monitoring organization-environment consistency and a
heuristic that suspends change for a trial period following each reorientation. The work
of Sastry provides the opportunity to speculate further on the features of inductive
simulation research. As we said in the foregoing, typically, inductive inferences bring
about additional information that is not necessarily crystallized into the premises. The
inductive nature of the study of Sastry emerges when we appreciate that in the original
premises of the study, which are captured in Tushman and Romanelli’s verbal theory,
there was not mention or any sort of indication that pointed at, or give a clue about, the
causal mechanisms that Sastry included into the theory ex-post.
To clarify the position taken in this essay, however, when I suggest that inductive
simulations bring in a study information content that is not included in the stated
premises, we are not speaking about empirical information. Computer simulation may
interact with empirical information and help to investigate real instances but do not per
se say anything about the empirical world. What I am suggesting is that given a set of
initial premises, a simulation study has an inductive nature when it facilitates the
enlargement or the modification of this set of premises.
Nevertheless, often computer simulation studies maintain a more or less close
relationship with empirical data. Malerba, Nelson, Orsenigo and Winter (1999), for
example, propose a class of computer models that they define history friendly because
of the adherence of these latter to the empirical realm that is the object of exploration.
14
To elucidate their approach, they focused on an appreciative theory that describes the
pattern of evolution of the computer industry and developed a formal representation of
that theory. Through simulation, they checked the consistency between the appreciative
and the formal version of the theory by examining whether the formal version is able to
reproduce the same stylized facts as described in the appreciative theory. The empirical
information is the pedestal to build the computer model but the contribution of the
simulation study is not one of extending such information. The contribution of the study
rests in its corroborating the relationship between causal mechanisms and emerging
behaviors as observed in the real world.
The interplay of induction and deduction in simulation studies
A consideration is fundamental in order not to misinterpret the distinction between
deductive and inductive simulation studies. In most of the simulation studies in social
sciences, inductive and deductive inferences are intertwined. However, we cannot avoid
noting that the logic by which they are inspired often differs not marginally. For
example, in the mentioned study of Sastry, the logic of enquiry is clearly stated and
hinges upon two elements. First, the author has a clear imagine of the dynamic features
of the behavior she wants to explain. Second, she uses the comparison between
theoretical and simulated behavior as a trigger to import in her modeling candidate
causal mechanisms.
On the other hand, at the other extreme, consider, for example, Cohen, March and
Olsen’s Garbage Can simulation model. The authors described how problems, choices
and people met within an organization but they start their enquiry without a precise idea
about the aggregate decision-making behavior that follows from the premises they
15
designed. The curiosity was exactly to understand what the consequences are of
representing an organization as an organized anarchy and the contribution of the study
is indeed to suggest that organized anarchies maintain a peculiar style in their decision
making behavior. Most of the simulation studies, however, blend the two components.
Beside cases in which the inductive or deductive approaches clearly come into view,
most of studies incorporate both approaches. A simulation study may incorporate a
loosely defined idea of the features of the behavior it is aimed to explain and this idea
guides the modeling of the premises. The deduction of consequences from premises
through computer simulation aids the refinement of the description of the behavior of
interest. On the other hand, the materialization of surprising or counterintuitive
behaviors induces the search for alternative causal mechanisms to modify the original
set of premises.
The diagram in figure 1.1 suggests that induction and deduction are often embedded in a
cyclical process of discovery. Deduction generates repertoires of patterns of behavior
that represent near-histories that proceed from a common deep causal structure. This
exercise contributes to theory building by making available ex-ante falsifiable
hypotheses that connect casual mechanisms to behaviors. Deduction may also create
counterintuitive and surprising behaviors that bring about marginal amendments in the
modeling of the premises or may trigger inductive processes of revisions of modeled
premises. In this case, the discrepancy between expected and simulated behavior is the
incentive to refine, or deeply modify, the modeled set of premises by introducing in the
model new causal mechanisms.
For example, in their study on population ecology and competition among structurally
different populations of organizations, Carroll and Harrison (1994) built a mathematical
16
model, designed a structurally superior population and simulated competition between
two populations (one inferior and one superior). Through the simulation study, they
demonstrated, in vitro, that the dominance of structurally superior populations may not
emerge depending on their timing of entry in the industry. The contribution of this
theoretical falsification is to delineate the hypothesis of historical inefficiency,
according to which the explanation of an observed behavior is history-dependent and
the time in which events happens modify their expected consequences.
Thus, the diagram in figure 1.1 conveys one of the key ideas that inspire this essay.
Technically speaking, a computer simulation cannot be anything different than a
computer-aided process of deduction. This deduction process both unveils not
necessarily intuitive cause-effect relationships that are implicitly hidden in the premises
and assists rigorous articulation of appreciative theories. This facilitates researchers in
producing testable hypotheses. On the other hand, when deducted behaviors do not
match with expectations, this mismatch activates an inductive inference that amends the
original set of premises. In this respect, I suggest that by embedding a computer-based
process of deduction into a richer research perspective provides a powerful environment
to use of computer simulation for theory development.
17
Figure 1.1: Logic of enquiry underpinning computer simulation
DEDUCTIVE INFERENCE
—
H c a: H
{ Premises —C Sens, —> Consequences H
Se
! H
t !
Revision of initial Inclusion of additional Counterintuitive deductions +
set of premises < causal mechanisms “—— Surprise behaviours H
A FRAMEWORK TO INTEGRATE COMPUTER MODELING AND
GROUNDED FIELD RESEARCH
In this section, I outline some ideas to inspire the use of computer modeling and
simulation as a support for theory building associated to grounded field research. This
area of methodological development is under investigated and hopefully the few
directions offered in this essay may indicate a possible avenue to explore.
Among the possible, and slightly different, logics that animate field studies, I take the
viewpoint of grounded theorizing. Glaser and Strauss (1967) described a grounded
approach to systematically discover theory from empirical data in field research.
According to this approach, by comparative analysis, researchers first generate
conceptual categories and the conceptual properties of these latter, then, they create
hypotheses on the relationships among the categories. In their view, researchers need to
start their research into an empirical setting without any previously structured
18
conceptual category’. This approach is, for example, different from the approach based
on explanatory case study which has a recognized tradition in management and has
been thoroughly described by Yin (1994). According to this latter approach, field
research is aimed at answering to how or why questions by eliciting causal links among
variables over time. Interestingly, Yin draws a distinction between the case study
approach, which he describes, and the grounded theorizing as described by Glaser and
Strauss. Yin affirms that the key difference is that in a case study researchers use a
previously developed theory as a template and the design of a case study is tantamount
the conceiving of a theoretical experiment aimed at further articulating the theory.
To assess the contribution of formal modeling and computer simulation to the
enhancement of the quality of grounded field research, I focus on two typical problems:
the internal validity problem and the problem of theoretical saturation of a grounded
study.
The problem of internal validity
Given an explanation that infers a casual relationship between two events, internal
validity is a judgment on the robustness of that causal relationship. Thus, a potential
threat to internal validity is the existence of spurious effects. When a researcher makes
an inference, and connects an event to an earlier occurrence, a spurious effect intervenes
if the appearance of the event object of observation is connected instead to another
unobserved occurrence. Y in describes three techniques to improve internal validity of a
case-study: pattern-matching, explanation-building and time-series analysis (1994: 35,
106-118).
19
Patter matching implies the comparison between the predicted and the actual behavior
of a variable. When empirically observed results match those predicted by a theory, the
case study represents an experiment that corroborates propositions embedded in the
theory. On the other hand, if patterns do not match, theory has to be questioned. The
more articulated is the predicted pattern of dependent variables, the more demanding is
the test of pattem matching and the stronger is the test of theoretical propositions.
For example, if a prediction involves not one pattern but a variety of patterns for a
variety of dependent variables, matching of those pattems allows for strong causal
inferences. Pattern matching also includes independent variables. Researchers may
articulate rival explanations that imply different causal mechanisms, different
independent variables and different, mutually exclusive, unfolding patterns for
independent and dependent variables. The matching between one specific predicted
pattem and the observed empirical behavior supports selection among rival
explanations. When a complete explanation for the phenomenon under analysis does not
exist at the beginning of the study, Yin suggests that the pattern-matching procedure
gives the way to a more sophisticated protocol that is named explanation-building.
Explanation building consists of an iterative process though which researchers gradually
build an explanation by making initial theoretical statements and predictions, comparing
predictions with available empirical patterns and revising statements.
Finally, pattern matching can be applied on time series of variables rather than to a
chain of events chronologically linked. This kind of analysis is named time series
analysis.
At the core of the three techniques, is the problem of understanding how a causal
structure is able to explain observed patterns of behavior. The key theme here is the
20
ability of a researcher to enact and maintain a dialogue between theoretical behaviors, as
predicted by an explanation, and observed empirical patterns.
The problem of theoretical saturation
In grounded research, Glaser and Strauss consider research as a process in which a
researcher often starts her analysis with a ‘partial framework of local concepts (1967:
45). She has a general idea of the concepts and the processes that will be parts of her
theoretical investigation but she ignores the relevancy of these concepts and processes
neither she knows whether additional concepts will emerge in the course of her study or
if some of the concepts initially selected will result irrelevant.
In this research process, initial collection of data needs to be followed by further
collections. This practice of analyzing collected data and deciding what data to collect
next is defined theoretical sampling (Glaser and Strauss, 1967: 45) and is an emergent
process that cannot be planned in advance since it guided by gaps emerging in a
analysis. The objective of theoretical sampling is to fully develop a repertoire of
concepts to be used in the theory being developed and to articulate the properties of
each concept.
When theoretical sampling is sufficient and can be stopped?
Glaser and Strauss suggest that theoretical sampling continue until theoretical
saturation is reached (1967: 61). The attainment of theoretical saturation implies that a
researcher cannot find additional data to develop further properties of conceptual
categories. The aim of theoretical saturation is to maximize the variety of empirical data
21
connected to each conceptual category in order to achieve adequate richness of
properties.
The problem of theoretical saturation entails deciding when the search for further data
collection can be stopped and how to select a research site for the collection.
A framework to integrate computer modeling and grounded field research
In the following, I delineate a research design in which computer modeling and
simulation, and field research are associated to support theorizing. The research design
proceeds in a sequence of steps in which a researcher begins by theorizing from an
exploratory field study, translates this preliminary theorizing into a formal model and
through computer simulation both strengthens the causal structure of the theory and
envisions new research sites to locate further field studies that serve as experiments to
consolidate the theory.
Building of a preliminary theoretical framework
In the sketched approach, the point of departure is an exploratory field study (step 1 in
figure 1.2). The field-study entails the grounding of theorizing in a specific research
site. As described by Glaser and Strauss (1967), the field-study leads to the building of a
theoretical framework by defining conceptual categories, conceptual properties of the
categories and hypotheses regarding the causal relationships among categories. The
sketch of the theoretical framework needs to proceed without any ‘... preconceived
theory that dictates, prior to the research, “relevancies” in concepts and hypotheses.’
(Glaser and Strauss 1967: 33).
22
A researcher should select, at this stage, a research site because she is interested in a
specific empirical phenomenon not because the site is appropriate to conduct a
theoretical experiment on an existing theory. Of course, it is naive to propose that a
researcher approaches a research site without any previously crystallized theoretical
lens. It is plausible to suspect that the theoretical background of the researcher, along
with the state of the art of the literature to which she aims at contributing, plays a role in
the sedimentation of a more or less uncovered cognitive filter that steers the attention
towards one or another research site. The intellectual curiosity that illuminates a specific
research site is motivated by the interest for an observed phenomenon and it is likely
that this interest is, at least implicitly, driven by the fact that the phenomenon is an
empirical instance that confirms or disconfimms a prior theory. Hardly can theorizing
totally be disconnected from relevant literature because what captures the attention of a
researcher is the observation that a conceptual category is empirically associated with a
property different from the one expected, that two conceptual categories are empirically
linked by a counterintuitive causal relationship or that an empirical phenomenon
escapes previous conceptualizations.
It is not the purpose of this chapter to dwell into the delicate dispute regarding the
selection of a research site; neither this chapter gives attention to how a researcher
extracts a preliminary theoretical framework from a field study. We simply assume that
a preliminary empirically grounded theoretical framework exists, this latter including a
number of conceptual categories, conceptual properties that characterize the categories
and a number of tentative hypotheses on causal relationships among categories.
23
Figure 1.2: Computer simulation and grounded field research
y
Structural changes
Parameters’ re-calibration
— falsification <——_,
manana DIVERGENT
i Aiea BEHAVIORS
articulation of
behavior
Formal 3, synthetic 4 5. Pattern |
model ~~ experiments matching i
H
2 Categorical i
Theoretical articulation of ¥
model behavior CONVERGENT
n BEHAVIORS
“ '
1
i
Field Finding of new Sensitivity
study “7777 research fields “7777777777 analysis
6
Cycle of internal validation
Cycle of theoretical saturarion -------~=:
Building the computer model
Once the exploratory field study has generated a preliminary theoretical framework, the
second step implies to transform an appreciative theorizing into a set of formal
propositions (step 2). At the end of this second step, a computer model embodies the
preliminary theoretical framework. This is a subtle endeavor that requires the
transformation of conceptual categories into measurable constructs that reflect their
theoretical properties and the formalization of causal link among constructs. Causal
mechanisms included in a computer model may originate from two sources. They may
be formalizations built upon a researcher's interpretation of verbal descriptions
collected during the field study or they may be formalizations that replicate either
existing formal theories or descriptions of processes that already exist in a quantitative
format.
24
Let’s take for example a researcher that conducts a field study to explain the process of
strategy formation in large firms. If a firm makes available to the researcher memos,
blueprints and manuals with already formalized decision-making routines, then the
formalization is likely to adhere more realistically to the empirical setting under
scrutiny. More often, however, the researcher needs to translate verbal description of
operating organizational routines into formal modeling. Furthermore, let’s suppose that
the researcher wants to include in its theorizing the behavior of financial markets that
respond to focal organization's financial performances by allowing credit. In this case,
modeler may take advantage of existing theory of financial markets and include in her
model the formalizations that are provided in the literature to capture expected behavior
of financial markets. In this case, the use of an existing theory does not really violate the
mandate, stated at the beginning, to initiate an exploratory field research without
previous preconceived theory since the theory employed regards the behavior of
financial markets not the object of study of the research, which is the process of strategy
formation. In other words, in this case, the researcher borrows elements from the theory
of financial markets’ behavior to complete the description of the environmental context
in which the object of study - the firm - operates.
In addition, the field study may be helpful to provide researcher with information to be
used for a provisional calibration of model’s parameter. The calibration will be useful in
the next step of the described research protocol that requires the simulation of the
formal model. As Kaplan (1964) suggests, ‘...in all simulation experiments the
fundamental problem is that of “scaling” - that is, the translating of results from a
simulation model to the real world’. The grounding and calibration of a simulation
25
model on a case study facilitates the process of translating the abstract insights of a
formal model into real-world problems.
Synthetic experiments with computer simulation
The fourth step entails the use of the formal model to produce simulation runs that
describe behavioral implications of the causal relationships that originated from
preliminary theorizing. If the theoretical framework that a researcher has built to explain
the observed behaviors is correct, simulation runs tend to replicate observed behaviors.
In this light, computer simulation supports researchers in using data from field studies
to detect fallacies in underpinning logic and to test a theoretical framework. The use of
computer simulation as a tool to derive behavioral consequences from stated
assumptions brings about a number of advantages.
First, in general, computer simulation generates time-series. This may result of some
help when time-series can be compared directly with real-world quantitative figures, for
example financial figures extracted from balance sheets and economic reports. In this
case, the availability of real and simulated time series that are accessible in a similar
quantitative format facilitates pattern matching by assigning to a researcher the
possibility to generate a measure of how predicted events match empirical instances of
those events (Sterman 1984).
Second, computer simulation allows for a rigorous longitudinal articulation of
theoretical behaviors. In other words, the computer-aided process of deduction goes far
beyond the human capability to appreciate the long-term features of the behavior of
selected variables. Thus, computer simulation can support researcher to articulate
hypotheses on behavioral properties of conceptual categories included in a theoretical
26
framework. For example, complex pattems of behavior such as peaks and lowest point,
oscillations with different characteristics and changes in rate of growth or decline may
become parts of a more developed description of a behavioral property.
Third, researchers, by simulating a formal model, can generate what I define a
categorical articulation of theoretical behaviors, that is, they can further extend their
theory by contemporaneously producing behavior of different conceptual categories.
For example, researchers can simulate the interaction of independent and dependent
variables in each time step, along a given time horizon. Using this exercise, researchers
can produce additional hypotheses on the interaction among behavioral properties of a
numbers of conceptual properties contemporaneously.
This longitudinal and categorical articulation of theoretical behavior increases the points
of contacts between the theoretical propositions and the empirical world of the case
study. As Kaplan suggests ‘What counts in the validation of a theory, so far as fitting
the facts are concerned, is the convergence of the data brought to bear upon it [...]’
(1964: 314).
I argue that computer simulation expands the terrain where comparison between theory
and empirical setting takes place by generating a rich theoretical framework that
crystallizes a dense collection of interweaved theoretical behaviors. Thus, convergence
of data and concatenation of events that are necessary to confirm the adherence of
theory to empirical evidence are increasing demanding.
In this respect, computer simulation aids researchers to consider field research as a
difficult experiment where the falsifiability of a theory is easier because fitting the facts
becomes increasingly hard. Of course, on the other hand, had empirically collected facts
27
to fit into a complex web of interweaved simulated behaviors, the experiment would
lead to stronger evidence to confirm propositions contained in the theory.
In figure 1.3, we imagine to start a field study with the objective to explore the increase
in profits empirically observed in the time period comprised between t; and t. A
researcher can build a variety of hypotheses to explain the behavior. These hypotheses
can be formalized into a computer model. Yet, there might be a large collection of
computer models that are able, for different reasons, to produce a behavior similar to the
one observed. However, once we use computer simulation to articulate the behavioral
implications of the model beyond the observed time span ti-t2 (longitudinal articulation)
and for both the profits and other variables such as revenues and market share
(categorical articulation), then the theory of behavior captured in the model becomes
more complex and easier to falsify by further data collections of empirical instances
regarding market share and revenues.
28
Figure 1.3: Integration of empirical and simulated patterns of behavior
1,500,000,000 |..-......,, REVENUES
45% |"
Profits
1,000,000 |"
1 t Time
In this respect, in the example of figure 1.3, longitudinal articulation of behavior of the
variables of interest, that is the generation of an hypothesis of behavior that extends
beyond the originally considered time span t; -tz, and categorical articulation, that is the
generation of hypothetical behavior for a variety of relevant variables, orient further
data collection and increase falsifiability of a theoretical framework. This process of
data collection to falsify formalized theoretical hypotheses narrows down the set of
candidate explanatory models.
Pattern matching
Now we tum to the process that involves the analysis of the matching between
simulated and empirically observed patterns of behavior (step 5 in figure 1.2). In
particular, we investigate this process by looking at two cases. The first case is when the
field study confirms the predictions made through computer simulation. The second
29
case applies when a researcher reports a mismatch between computer-generated
predictions and empirically collected evidences and time-series.
Sensitivity analysis and history-convergent runs
When simulated and historical patterns of behavior match, computer simulation can be
used as a laboratory to conduct sensitivity analysis in order to explore in what
circumstances simulated and empirically observed behaviors diverge (step 6).
Sensitivity analysis entails the analysis of the sensitivity of a simulated behavior to the
change in the calibration of a model’s parameters. Also, changes in parameters’ values
may often imply that parts of a model are deactivated thereby testing the sensitivity of a
model’s behavior to the inclusion or exclusion of specific conceptual categories.
Field cases are retrospective studies. Retrospective studies explain, ex-post how a set of
variables interacted to drive an observed behavior of interest. However, it could become
troublesome to ascertain the extent to which a theoretical explanatory model and the
observed behavior are linked. This difficulty is explained by the fact that retrospective
studies are not particularly efficient in connecting causes and effects (Leonard-Barton
1990).
If, for example, we are aware that two conceptual categories affect the observed
behavior, given the complex web of interactions in which the concepts are embedded, it
might be hard to determine their relative strengths. It might be the case that the
influence of one of these two concepts is insignificant, and could be omitted from the
analysis to satisfy the criterion of parsimony for a good theory (Eisenhardt 1989).
To investigate further the importance of that concept, an experiment could be run to
detect what happens if it is omitted from the model.
30
Thus, sensitivity analysis helps revising a theoretical explanation by suggesting that
specific conceptual categories are not necessary to explain a behavior whereas others
are fundamental since the change in their calibrations produces simulated behavior that
is divergent from the one observed (step 7 in figure 1.2).
In addition, the intentional generation of history-divergent simulation runs orients
further empirical enquiry by indicating new potential research sites. Indeed, in a new
site that resembles the simulation settings that have been adopted in the sensitivity
analysis, a researcher can test whether, given the characteristics of the new site, a
behavior closer to the history-divergent run is observed (step 8 in figure 1.2). For
example, some longitudinal event studies have compared polar cases, that is, cases of
organizations that have shown opposite behaviors in responding to an identical
exogenous stimulus, and have explained the different unfolding of their histories as the
result of different initial conditions (Noda 1994; Noda and Bower 1996).
What we suggest is that using sensitivity analysis to generate history-divergent runs
may be helpful to illuminate the potential of a research site to become a polar case in
which, given a change in some key features of the research context, a behavior
divergent from the one observed in the original field study ensues.
In general, simulation, by connecting a theoretical structure to a variety of possible
emerging, often unexpected, behaviors, activates dormant consequences of a theory,
which were not observed in the original empirical study. This generation of a
distribution of near-histories, or unrealized events, both strengthens the understanding
of causal structures and envisions areas for further empirical investigations. Field
researches conducted in these sites represent further theoretical experiments to reinforce
internal validity of a theory. Thus, computer simulation helps validating a theory by
31
supporting a researcher in demonstrating that a common theoretical engine may explain
a repertoire of different behaviors in different empirical contexts. In this vein, the
coupling of field-study and computer simulation speaks to the problem of learning from
samples of one or fewer as presented by March, Sproull and Tamuz (1991).
In addition, when a theoretical argument includes the mention of specific parameters’
calibration as necessary conditions for a predicted behavior to emerge, change in the
parameters’ calibration can represent a further test of the theory encapsulated in the
computer model. For example, Malerba et al. modified parameters’ calibration in order
to test that model calibrations ‘that are counter to those argued as strongly causal in the
appreciative theory should obtain history-divergent results’ (1999: 35).
Sensitivity analysis and history-divergent runs
Finally, we address the case in which a researcher observes a mismatch between
computer-generated and empirically observed events and time-series. In this case, the
problem is to understand why the behaviors diverge. The idea here is that computer
modeling and simulation provide a theoretical laboratory that is relatively easy to
manipulate in order to investigate the origins of the discrepancy between simulated
predictions and observed behaviors.
In this respect, I agree with Malerba et al. (1999) in suggesting that computer
simulation provide an appropriate terrain to nurture a friendly dialogue between
empirical evidence and theory. When history-divergent simulations appear, researcher
tries to explain where discrepancies come from. Investigators can intervene on the
structure of a computer model or on the calibration of model’s parameters and
32
rigorously deduct whether these interventions narrow down the gap between predicted
and actual behaviors.
Pressures for historical and simulated behaviors to diverge arise in two cases. The first
pressure intervenes when the causal structure of the theory that is captured in the
computer model is isomorphic to the causal relationships at work in a specific empirical
context and the discrepancy is the consequence of flaws in the specifications of
parameters’ calibrations. The second pressure for historical and simulated behaviors to
diverge arises when the causal structure of the theory and the causal relationships at
work in the real world are not isomorphic in some respects.
This may be the result either of the fact that a researcher has not properly formalized a
theoretical argument arising from a field study or the fact that the researcher was not
able to select the key causal mechanisms at work in the case studied.
The first direction to explore is the analysis of sensitivity of model’s behaviors to
change in parameters to check whether simulating the model with a new calibration
improves the match between simulated and observed behaviors (step 9). The fact that
the fit between simulation and empirical data is improved by manipulating a model’s
parameters points at two areas of analysis. First, it may suggest that the model is
characterized by non-linear causal relationships among variables so that slightly
different model's calibrations yield very different emerging behaviors.
Second, the causal structure at work may include positive feedback among variables and
initial calibration of variables has a mounting weight in molding unfolding patterns of
behavior. For example, as reported in Carrol and Harrison (1994), the presence of
positive feedback among variables generates behaviors that unfold in a way that is
33
history-dependent, that is, depends on the research’s calibration of a parameter: time of
entry in the simulated scenario of a population.
Finally, in general, the fact that a computer model produces history-replicating
simulation runs only after implausible values are assigned to parameters casts an
alarming light over the robustness of causal structure of the theory.
The second avenue to explore discrepancy between predictions and observed behaviors
is the analysis of the structure of the model, that is, the causal relationships among
variables that are deemed necessary to produce behaviors of interest. Different
formalizations may exist for specific relationships and including in the model one or the
other may have different behavioral implications. To revise formalization, researchers
need to go back and compare the formal structure of the computer model and the real
processes at work in the case study. This further investigation plays as a catalyst to
define possible amendments to the theory (step 7).
To begin the analysis of the discrepancy between the structure of the computer model
and the structure of the phenomenon under study, those formalizations that are directly
obtained from descriptions sufficiently clear and less questionable are not good
candidates to look for the origins of the discrepancies. Researcher ought to start to
generate alternative formulations for those descriptions that were originally provided in
a verbal form and, thus, to be formalized, required a more dubious and arguable
interpretation. The fairly intuitive idea here is that formalization that required a
researcher’s translation of verbal descriptions into quantitative formulations are more
debatable, more prone to conceal misinterpretation and hence good candidates for the
analysis of history-divergent simulations.
34
However, such an instinctive expedient ought not to veil another potential source of
history-divergent simulation runs that materializes when firms describe processes on the
basis of existing formal procedure whereas everyday activity is grounded on informal
and tacit routines which are different from those crystallized in official manuals and
blueprints.
Theoretical saturation and internal validity
In the iterated process of pattem matching, structural adjustment and theory refinement
that I delineated in the foregoing, field study informs computer model and this latter
puts on the right track empirical research.
Once the dialogue between computer model and field research kicks off the critical
issue is whether a researcher is able to mediate the dialogue by pursing two critical
processes. First, the researcher has to feed the model with the information extracted
from a case study. Second, the researcher needs to understand what information the
observed gap between simulated and historical behavior provides that can be utilized to
both indicate further research sites and to refine underlying theoretical argument.
Associated with a simulation study, the field study is not more a retrospective
photograph of what has happened, but rather becomes a living picture illustrating what
could have happened in different circumstances. Capturing in a simulation model the
rich but static appreciative theorizing grounded on a field study, the researcher can build
a laboratory where simulation experiments are used to pursue two endeavors.
First, analysis of history-divergent runs triggers a cycle of internal validation thereby
missing variables and hidden assumptions are elicited, emerging theory is tested for
internal consistency (Langley 1999).
35
Second, sensitivity analysis on history-convergent prompts a cycle of theoretical
saturation thereby new research sites are selected to refine the set of theoretically
relevant conceptual categories, to capture the properties of these categories and to
describe the nature of causal relationships among conceptual categories.
CONCLUSION
As Montgomery, Wemerfelt and Balakrishnan suggested almost 20 years ago (1989), a
serious problem that may compromise the quality of theory development in strategy and
organization is the looseness and the lack of logical consistency in developing
implications from a set of assumptions where “Small changes in assumptions or
parameters can alter dramatically the implications of a model.” (Montgomery,
Wernerfelt and Balakrishnan 1989: 192)
In this article I proposed that computer modeling and simulation support theory
generation in managerial studies and, in general, in social sciences by contributing to
amend for the critical shortcomings that emerge in theory development when
implications are not rigorously derived from assumptions. In particular, computer
modeling forces a researcher to tease out unambiguously her theoretical argument. A
simulation experiment entails the formalization of a theory. Formalization enhances
simplicity and parsimony and helps to clarify the morphology and to sharpen the
discussion of the theory thereby supporting both its audit trial (Saloner 1994: 170) and
its communication. In this respect, I suggest that formalization and computer simulation
of a theory represent devices that support communication among scholars of different
disciplines.
36
Furthermore, the discourse articulated in this chapter suggests that computer modeling
and simulation offer a helpful tool to enhance quality of field-based theory building.
Field research and simulation studies, although both having strong roots in management
and organization theory research, have not often been used in combination. This essay
contains the sketch for a research protocol that integrates simulation-based research and
field study.
The idea that motivates this attempt is that computer simulation, by producing artificial
time series that are directly comparable with real time series, sets basis for a fruitful
dialogue between the observation of empirical pattems of behavior and the modeling of
theoretical hypotheses. As Cohen and Cyert suggest (1961), this dialogue both
strengthens the theoretical argument and directs field research:
The requirements of a computer model can provide a theoretical
framework for an empirical investigation, and, in retum, the
empirical information is utilized in developing a flow diagram for
the model. Through this process of working back and forth, it is
possible to know when enough empirical information has been
gathered and whether it is of the proper quality.
(Cohen and Cyert 1961: 127)
This dialogue between available empirical data, in the forms of detailed description of
observed behavior, and a theory, or a set of hypotheses, formalized in a computer model
establishes the premises to develop sound theories of behavior.
37
REFERENCES
Adner, R. (2002) ‘When Are Technologies Disruptive? A Demand-Based View of the
Emergence of Competition’, Strategic Management J ournal, 23 (8): 667-88.
Carroll, G.R. and Harrison, J.R. (1994) ‘On the historical efficiency of competition
between organizational populations’, The American Journal of Sociology, 100
(3): 720-49.
Clarkson, G.P.E. and Simon, H.A. (1960) ‘Simulation of Individual and Group
Behavior , The American Economic Review, 50 (5): 920-32.
Cohen, K J. (1960) ‘Simulation of the Firm’, The American Economic Review, 50 (2),
Papers and Proceedings of the Seventy-second Annual Meeting of the American
Economic Association: 534 -40.
Cohen, K J. (1960) Computer Models of the Shoe, Leather, Hide Sequence, Prentice-
Hall: Englewood Cliffs, NJ.
Cohen, KJ. (1961) ‘Two Approaches to Computer Simulation’, The J ournal of the
Academy of Management, 4 (1): 43-9.
Cohen, K. and Cyert, R.M. (1961) ‘Computer Models in Dynamic Economics’, The
Quarterly Journal of Economics, 75 (1): 112-27.
38
Cohen, M.D., March, J.G. and Olsen, H.P. (1972) ‘A garbage can model of
organizational choice’, Administrative Science Quarterly, 17 (1): 1-25.
Cyert, R.M., Feigenbaum, E.A. and March, J.G. (1959) ‘Models in a Behavioral theory
of the firm’, Behavioral Science, 4 (2): 81-95.
Cyert, R.M. and March, J.G. (1963) A Behavioral Theory of the Firm, Prentice-Hall:
Englewood Cliffs, NJ.
Davis, J.P., Eisenhardt, K.M. and Bingham, C.B. (2007) ‘Developing theory through
simulation methods’, Academy of Management Review, 32 (2): 480-99.
Eisenhardt, K. (1989) ‘Building theories from case study research’, Academy of
Management Review, 14 (4): 532-50.
Forrester, J.W. (1958) Industrial Dynamics-A Major Breakthrough for Decision
Makers, Harvard Business Review. 36 (4): 37-66.
Gilbert, N. and Troitzsch, K.G. (eds.) (2005) Simulation for the Social Scientist, Open
University Press.
Glaser, B.G. and Strauss, A.L. (1967) The Discovery of Grounded Theory, Hawthome,
New Y ork, NY: Aldine de Gruyter.
39
Harshome, C. and Weiss, P. (1931/1935) Collected papers of Charles Sanders Peirce,
vol. II: 374, Cambridge, MA: Harvard University Press.
Hoggatt, A.C. (1957) ‘Simulation of the Firm’, Research Paper RC-16, IBM Research
Center, Poughkeepsie, N.Y .
Kaplan, A. (1964) The Conduct of Inquiry, San Francisco, CA: Chandler Publishing
Company.
Langley, A. (1999) ‘Strategies for theorizing from process data’, Academy of
Management Review, 24 (4).
Leonard-Barton, D. (1990) ‘A dual methodology for case studies: synergistic use of a
longitudinal single site with replicated multiple sites’, Organization Science,
1(3): 248-66.
March, J.G., Sproull, L.S. and Tamuz, M. (1991) ‘Learning from a sample of one or
fewer’, Organization Science, 2 (1): 1-13.
Malerba, F., Nelson, R., Orsenigo, L. and Winter, S. (1999) ‘History-friendly' models of
industry evolution: the computer industry’, Industrial and Corporate Change, 8
(1): 3-40.
40
Meinhart, W.A. (1966) ‘Artificial Intelligence, Computer Simulation of Human
Cognitive and Social Processes, and Management Thought’, The Academy of
Management J ournal, 9 (4): 294-307.
Montgomery, C.A., Wemerfelt, B. and Balakrishnan, S. (1989) ‘Strategy Content and
the Research Process: A Critique and Summary’, Strategic Management
Journal, 10 (2): 189-97.
Noda, T. (1994) ‘Intra-organizational Strategy Process and the Evolution of Intra-
industry Firm Diversity: A Comparative Study of Wireless Communications
Business Development in the Seven Bell Regional Holding Companies’,
Doctoral Dissertation, Harvard University Graduate School of Business
Administration.
Noda, T. and Bower, J.L. (1996) ‘Strategy making as iterated processes of resource
allocation’, Strategic Management Journal, vol.17, Special Issue: Evolutionary
Perspectives on Strategy, 159-92.
Orcutt, G.H. (1960) ‘Simulation of Economic Systems’, The American Economic
Review, 50 (5): 893-907.
Rahmandad, H. and Sterman, J. (2004) ‘Heterogeneity and Network Structure in the
Dynamics of Diffusion: Comparing Agent-Based and Differential Equations
41
Models’, Massachusetts Institute of Technology Engineering Systems Division,
Working Paper Series, ESD-WP-2004-05.
Saloner, G. (1994) ‘Game theory and strategic management: contributions, applications,
and limitations’, in R.P. Rumelt, D.E. Schendel and DJ. Teece (eds)
Fundamental Issues in Strategy. A Research Agenda, Boston, Mass.: Harvard
Business School Press.
Shubik, M. (1960) ‘Simulation of the industry and the firm’, The American Economic
Review, 50 (5): 908-19.
Sterman, J.D. (1984) ‘Appropriate summary statistics for evaluating the historical fit of
system dynamics models’, Dynamica, 10: 51-66.
Troitzsch, K.G. (1998) ‘Multilevel process modelling in the social sciences:
Mathematical analysis and computer simulation’, in W.B.G. Liebrand, A.
Nowak and R. Hegselmann (eds.) Computer Modeling of Social Processes,
London, UK: Sage Publications.
Tushman, M. L. and E. Romanelli (1985), ‘Organizational Evolution: A Metamorphosis
Model of Convergence and Reorientation’, in L. L. Cummings, and B. M. Staw
(eds.) Research in Organizational Behavior. Vol. 7, Greenwich, CT: JAI Press:
171-222.
Yin, R.K. (1994) Case Study Research, Thousand Oaks, CA.: Sage Publications.
42
43
