Path Dependence, Competition, and Succession

in the Dynamics of Scientific Revolution

John D. Sterman • Jason Wittenberg
Sloan School of Management, E53-351, Massachusetts Institute of Technology, Cambridge, Massachusetts 02142
Department of Political Science, Massachusetts Institute of Technology, Cambridge, Massachusetts 02142

Abstract the traditional view science applies universally-accepted

What is the relative importance of structural versus contextual norms of logical inquiry, and scientific development is
forces in the birth and death of scientific theories? We describe seen as the cumulative triumph of an ever more truthful
a formal dynamic model of the birth, evolution, and death of and encompassing understanding of reality. In contrast,
scientific paradigms based on Kuhn’s Structure of Scientific Kuhn (1970, pp. 84–85) argues that new theories replace
Revolutions. The model represents scientific activity as a chang- old ones rather than building on them, revolutionizing
ing set of coupled institutions; a simulated ecology of interact- science’s very image of itself. For Kuhn, scientific de-
ing paradigms in which the creation of new theories is stochas- velopment is fraught with errors, blind alleys, and intense
tic and endogenous. The model captures the sociological
dynamics of paradigms as they compete against one another for
competition among competing worldviews, proceeding
members, solve puzzles, and recognize anomalies. We use sen- “as a succession of tradition-bound periods punctuated by
sitivity tests and regression to examine the role of intrinsic ver- non-cumulative breaks” (Kuhn 1970, p. 208). Of course,
sus contextual factors in determining paradigm success. We find Kuhn can be situated in a long tradition of scholars who
that situational factors attending the birth of a paradigm largely reacted against realist and positivistic views of science,
determine its probability of rising to dominance, while the in- perception, and knowledge, including Paul Feyerabend
trinsic explanatory power of a paradigm is only weakly related (1975), and famously, Ludwik Fleck, whose 1935 theory
to the likelihood of success. For those paradigms surviving the of the social construction of scientific facts was cited by
emergence phase, greater explanatory power is significantly re- Kuhn (1970, p. vii) as “an essay that anticipates many of
lated to longevity. However, the relationship between a para-
my own ideas.” Yet no other formulation captured the
digm’s “strength” and the duration of normal science is also
contingent on the competitive environment during the emer- imagination—or generated the ire—that Structure did.
gence phase. Analysis of the model shows the dynamics of com- Indeed, the idea that individual, social, and historic
petition and succession among paradigms to be conditioned by contingencies play a role in scientific development rival-
many positive feedback loops. These self-reinforcing processes ing or exceeding a theory’s intellectual content has elated
amplify intrinsically unobservable microlevel perturbations in many social scientists and historians as much as it has
the environment—the local conditions of science, society, and infuriated many philosophers and scientists. (The litera-
self faced by the creators of a new theory—until they reach ture is massive. For a representative sample of the debate
macroscopic significance. Such path dependent dynamics are
over the social construction of science, see e.g., Bijker et
the hallmark of self-organizing evolutionary systems. We con-
sider the implications of these results for the rise and fall of al. 1987, Collins 1985, Cushing 1994, Donovan et al.
new ideas in contexts outside the natural sciences such as man- 1988, Gross and Levitt 1994, Martin 1996, Pickering
agement fads. 1984, 1992, Pinch 1986, and Roth and Barrett 1990). For
(Complexity; Simulation; Competition; Sociology of Sci- many social scientists Kuhn’s theory legitimated resis-
ence; Scientific Revolution) tance to the century-old attempt to make the study of
society, politics, and culture more like Newtonian phys-
ics. For others Kuhn’s attempt to historicize scientific en-
deavor was reckless and heretical. Yet whether as proph-
ecy or apostasy, his ideas continue to stimulate vigorous
1. Introduction debate about the evolution of science (e.g., Hoyningen-
The late Thomas Kuhn’s Structure of Scientific Revolu- Huene 1993, Lightman and Gingerich 1992).
tions heralded a radically new conception of science. In The controversy over the nature of scientific progress

1047-7039/99/1003/0322/$05.00
ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 Copyright 䉷 1999, Institute for Operations Research
pp. 322–341 and the Management Sciences
JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

parallels the debate in the social sciences over whether model creates a simulated ecology of interacting para-
organizational change is gradual and evolutionary or ep- digms in which the genesis of new paradigms is stochastic
isodic and revolutionary. Though Kuhn (1970, pp. 208– and endogenous. The model captures the sociological dy-
210) cautioned against the applicability of his model to namics of paradigms as their members formulate and
the social sciences, it is nevertheless widely cited by so- solve “puzzles,” recognize and react to anomalies, and
cial scientists as descriptive of organizational behavior alter their beliefs and behavior. Competition for mem-
and cognitive shifts in contexts far beyond the natural bership and resources is explicit. The model is used to
sciences. Organization theorists argue that the pattern of investigate the relative importance of structural versus
punctuated equilibrium Kuhn finds in the history of sci- contextual factors in determining the fate of new ideas.
ence also characterizes many instances of organizational Although the model is based on Kuhn’s work, we do
change (see Gersick 1991, Tushman and Anderson 1986, not claim to capture his theory fully. Translating the the-
Sastry 1997). Tushman and Romanelli (1985, p. 171) pro- ory from its highly abstract written form into an internally
pose a model of organizational change in which consistent formal model has involved simplifications and
“[o]rganizations evolve through convergent periods the introduction of auxiliary hypotheses (for a discussion
punctuated by reorientations . . . which demark and set and critique see Wittenberg 1992, Sterman 1992,
bearings for the next convergent period.” Gersick (1991) Radzicki 1992, Barlas 1992). Nonetheless, formalization
shows there are many domains and levels of analysis, has advantages. Most discussions of Kuhn’s theory are
from paleontology (Gould 1990, Eldredge and Gould based on ambiguous mental models, and Kuhn’s text it-
1972) to group dynamics, in which change can be char- self is rich with ambiguity, multiple meanings, and im-
acterized by long periods of stasis or gradualism punc- plicit assumptions (Masterman 1970). More importantly,
tuated by sudden upheavals and revolutions. Some argue Kuhn offers no calculus by which one can assess whether
that change is often a more continuous and adaptive pro- the dynamics he describes can be produced by the causal
cess (e.g., Orlikowski 1996), while still others argue that factors he postulates. Formalization helps to surface im-
organizational adaptation is rare, with selection and evo- plicit assumptions so they can be debated and tested (see
lution occurring at the population level (Hannan and Gorman 1992, Rappa and Debackere 1993, Sastry 1997,
Freeman 1989; Van de Ven and Poole 1995 provide a Stewart 1986, Thagard 1968, and Turner 1987 for ex-
survey of change research). amples). Formalization is complementary to the work of
The similarity between the dynamics of science and of
historians, sociologists, and philosophers of science
organizations is not mere coincidence. Scientific activity
working to develop and test theories of scientific change
is not primarily the work of solitary geniuses, but is em-
and institutional upheaval.
bedded in a wide range of organizations, from the small
Finally, the model applies nonlinear dynamics to social
group level of researchers and graduate students in a lab,
phenomena and human behavior. Modern theories of
to departments and universities, to the network of funding
nonlinear, far from equilibrium systems, though emerging
agencies, journal boards, and professional societies that
constitute the “invisible colleges” defining a community in the physical sciences, have great potential to illuminate
of practice (Crane 1972). There are of course differences evolutionary behavior in social, economic, and other hu-
between the institutions of science and organizations in man systems (e.g., Anderson et al. 1988; Arthur 1989,
other domains such as business firms. Yet the institutions 1994; Bruckner et al. 1989, 1990; Ebeling 1991; Lomi
of science are among the most influential in our society, and Larsen 1996). The full potential of these tools in the
deserving of study in their own right. Additionally, the social sciences will be realized, we believe, only when
dynamics of scientific revolution may shed light on or- they are used to develop and test formal models. The
ganizational evolution in general. Why is it that some merely metaphorical use of concepts from nonlinear dy-
scientific paradigms last for centuries while others namics, while provocative, is not sufficient, a point also
quickly wither? How do intellectual, structural, and con- stressed by Carley and Wallace (1995) and Richardson
textual forces interact to shape and constrain the devel- (1996, 1991). In addition, useful models will draw on
opment of new paradigms (and organizations)? What de- experimental and field studies of human behavior to spec-
termines whether a new theory (or organization) survives ify the decision rules governing the behavior of the sim-
its founding and becomes dominant? Do structural or ulated agents (see Carley 1991, 1995; Cyert and March
contingent forces dominate the dynamics of social sys- 1963/1992; Forrester 1961; Hall 1976; Haxholdt et al.
tems? 1995; Lant and Mezias 1992; Morecroft 1985; Nelson
We address these questions with a formal dynamic and Winter 1982; Radzicki and Sterman 1994; and
model of paradigm emergence and competition. The Sterman 1988, 1989 for discussion and examples). Here

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 323

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

we develop a stochastic, nonlinear, disequilibrium, be- crisis may deepen for decades as new theories fail to
havioral model of the evolution of scientific theories, sprout or flower. Kuhn provides little guidance into the
grounded in Kuhn’s theory. As will be seen, the dynamics forces that cause one new idea to triumph and another to
exhibit self-organization and path dependence, two com- fail, or the determinants of the longevity of those new
mon modes of behavior in complex systems. paradigms that survive their founding and become dom-
inant. The central debate has been the relative importance
of intrinsic explanatory power—the “truth” of a new the-
2. A Theory of Paradigm Evolution and ory—versus contingencies external to science such as the
Succession social, political, and cultural context of emergence, or
We assume familiarity with Kuhn’s work and the many even chance factors—the existence of an Einstein, Bohr,
interpretations and alternatives to it (see Lakatos and or Keynes—in conditioning which paradigm candidates
Musgrave 1970 for classic critiques and Hoyningen- flourish and which perish.
Huene 1993 for a comprehensive survey and bibliogra-
phy). A core concept in Kuhn’s theory is the life cycle of 3. The Model
a paradigm. Kuhn describes a sequence of four stages: We construct a multiparadigm model in which existing
emergence, normal science, crisis, and revolution (fol- theories compete for membership and resources and in
lowed by the emergence of a new paradigm). The emer- which the creation of new theories is stochastic and en-
gence phase is characterized by the absence of commonly dogenous. The model can be thought of as a set of inter-
accepted beliefs or standards governing scientific activity. acting agents (the communities loyal to a particular the-
Conflict among paradigm-candidates arises from incom- ory or paradigm). Like other agent-based models (e.g.,
patible metaphysical beliefs and logics of inquiry, as Holland 1995, Weisbuch 1991), the collective system dy-
Kuhn (1970, pp. 13–15) illustrates with the state of elec- namics emerge from the interaction of the individual
trical research before Franklin and his colleagues pro- agents over time. Unlike some agent-based models, the
vided the field with a paradigm. As a theory attracts number of agents is not fixed; new theories are created
nearly every scientist in the field—thereby becoming the with a probability that varies endogenously as conditions
dominant paradigm—normal science begins. Debate over change. Also unlike some agent-based models, the indi-
fundamental assumptions dwindles, and, convinced their vidual paradigms have a rich internal structure represent-
paradigm is the proper way to characterize reality, sci- ing the activities of each community, including the belief
entists proceed to apply it to nature’s puzzles. During structure of the members, recruitment and defection, sci-
normal science, clashes between theory and data are often entific activity such as puzzle solving and anomaly rec-
resolved in favor of theory—it is often presumed that any ognition, and the flows of people and information that
anomalous observations are wrong, or the calculations couple the different paradigms competing against one an-
erroneous, so that further puzzle-solving effort will re- other at any given time.1 In what follows we provide an
solve the anomaly, a behavior Kuhn (1970, p. 81) illus- overview of the model; complete documentation is avail-
trates by citing anomalies facing Newtonian mechanics able from the authors.
involving the speed of sound, the moon’s perigee, and the The heart of the model is the identification of the meta-
precession of the orbit of Mercury. This is observed today physical and epistemological facets of paradigms with
in the debate over the value of the Hubble constant and metaphors, limited representations of reality that generate
the age of the universe (Chaisson 1997). anomaly when pushed too far. Four properties of meta-
Normal science continues until a crisis arises. A para- phor that are also properties of paradigms bear particular
digm can enter crisis when enough unsolved puzzles be- mention. First, metaphor is everywhere. Goodman (1968,
come recognized by practitioners as important anomalies, p. 80) argues that “metaphor permeates all discourse, or-
persuading them that the theory must, after all, be ques- dinary and special, and we should have a hard time find-
tioned. As persistent anomalies accumulate, increasing ing a purely literal paragraph anywhere.” Turbayne
numbers of scientists will devote their time to solving (1970) goes further, suggesting metaphor permeates our
them rather than addressing new puzzles, and some may thought as well as our language. Similarly, Kuhn (1970,
propose radical solutions. A revolution occurs when a p. 113) stresses the priority of paradigms, suspecting that
new paradigm based on such a radical reconceptualiza- “something like a paradigm is prerequisite to perception
tion gains wide acceptance, and science is reconstructed itself.” Second, metaphor involves a “transfer of schema”
from new fundamentals. Obviously the timing, character, from one area of experience to another (Goodman 1968,
and context of each stage differ from case to case. A pp. 71–80). Consider the metaphor “the brain is a com-
dominant paradigm in crisis may quickly be replaced, or puter.” The characteristics of computers are transferred,

324 ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

via the metaphor, to our image of the brain. The metaphor Thus the central dynamic hypothesis of the model
works because the characteristics of computers are well draws on the notion that paradigms are extended meta-
known and carry a constellation of meanings and exam- phors, and that metaphors are not unlimited in their ap-
ples that illuminate certain characteristics of the brain. plicability to reality. Specifically, we assume that the av-
For Kuhn paradigms operate similarly: scientists are erage difficulty of the puzzles faced by the practitioners
taught to transfer familiar models to new puzzles, to of a paradigm increases as the cumulative number of puz-
“grasp the analogy” (p. 189). Third, metaphors filter re- zles they have solved grows. This “paradigm depletion”
ality. Because metaphors are inevitably inexact, as are all represents the idea that each paradigm is a limited model
models, they highlight certain relationships and obscure of reality that may apply well in the domain of phenom-
others. Metaphors focus our attention on particular facts ena it was originally formulated to explain, but will be
and relations while others are pushed into the back- harder and harder to apply as scientists extend it to new
ground. The filtering power of paradigms is central to domains. The formalization of this hypothesis is de-
Kuhn’s theory as well: “In the absence of a paradigm . . . scribed below.
all the facts that could possibly pertain to the development The model creates a simulated ecology of interacting
of a given science are likely to seem equally relevant” (p. paradigms, each representing a community of practition-
15). Finally, metaphors define reality. Max Black (1954– ers; recruitment and defection from that community; and
1955, pp. 284–285) notes that “[i]t would be more illu- the intellectual activities of the members such as formu-
minating in some of these cases to say that the metaphor lating and solving puzzles, recognizing and trying to rec-
creates the similarity than to say that it formulates some oncile anomalies, and conceiving new theories. The
similarity already existing.” Kuhn (1970, p. 111) attrib- model simulates the attitudes and beliefs of the practi-
utes the same power to paradigms: tioners within each paradigm through constructs such as
“confidence in the paradigm” and the time required to
The historian of science may be tempted to exclaim that when perceive unexplained phenomena as anomalies which
paradigms change, the world itself changes with them. Led by
challenge the theory. The major sectors of the model and
a new paradigm, scientists adopt new instruments and look in
the linkages among paradigms are shown in Figure 1; we
new places. . . . [They] see new and different things when look-
ing with familiar instruments in places they have looked before.
will use causal diagrams to illustrate the feedback pro-
Insofar as their only recourse to the world is through what they cesses and stock-and-flow structure of the model
see and do, we may want to say that after a revolution scientists (Richardson and Pugh 1981, Sterman 1985, Weick 1979).
are responding to a different world. Each paradigm has the same internal structure; for clarity
we display only paradigms i and j.
Yet metaphors are imperfect models, and if pushed too
hard crack and fail. Consider the brain-as-computer meta- 3.1. Confidence in the Paradigm
phor. Applying this metaphor might yield statements that The focal point of the model is a construct called “con-
generate insight, motivate theory, or suggest experiments, fidence.” Confidence captures the basic beliefs of prac-
such as “people transfer information from long-term to titioners regarding the epistemological status of their par-
working memory with characteristic seek times.” Even- adigm—is it seen as a provisional model or revealed
tually, however, overextension of the metaphor yields ab- truth? Encompassing logical, cultural, and emotional fac-
surdities such as “brains run Microsoft Excel” or “brains tors, confidence influences how anomalies are perceived,
are composed of silicon semiconductors.” The accumu- how practitioners allocate research effort to different ac-
lation of such anomalous claims undermines the appeal tivities (puzzle solving versus anomaly resolution, for ex-
of a metaphor, and can send it to its grave, disgraced as ample), and recruitment to and defection from the para-
falsehood. Kuhn views the life cycle of paradigms in a digm. It is defined from 0 (absolute conviction the
similar way. The elaboration and extension of a paradigm paradigm is false, nonsensical) through 0.5 (maximum
to new domains can lead to the accumulation of anoma- uncertainty as to its truth) to 1 (absolute conviction the
lies. As an “almost entirely typical” example Kuhn cites paradigm is truth). Pressures leading confidence to
the accumulation of anomalies in Newtonian mechanics, change arise both from within a paradigm and from com-
such as the repeated failure to detect drift through the parisons with other paradigms (Figure 2). Confidence
ether resulting from the effort to provide a Newtonian rises when puzzle-solving progress is high and when
foundation for Maxwell’s theory of electromagnetic ra- anomalies are low. The impact of anomalies and progress
diation. As a result, “Maxwell’s theory, despite its New- is mediated by the level of confidence itself. Extreme lev-
tonian origin, ultimately produced a crisis for the para- els of confidence hinder rapid changes in confidence be-
digm from which it had sprung” (p. 74). cause practitioners, utterly certain of the truth, dismiss

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 325

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

Figure 1 Overview of Model Structure anomalies are less than those of the dominant paradigm,
or if it has greater explanatory power, as measured by
cumulative solved puzzles. Confidence tends to decrease
if the dominant paradigm has fewer anomalies or more
solved puzzles. Practitioners in alternative paradigms as-
sess their paradigms against one another as well as against
the dominant paradigm. Confidence in an alternative par-
adigm tends to decrease (increase) if it has more (fewer)
anomalies or fewer (more) solved puzzles than the most
successful of its competitors.

3.2. Puzzle Solving

The determinants of puzzle solving are shown in Figure
3. Three categories of puzzles are distinguished. Solved
puzzles are puzzles that have already been integrated into
the corpus of theory and data constituting the paradigm.
Anomalies are unsolved puzzles which have come to be
recognized as serious challenges to the theory. The third
category, puzzles under attack, consists of those puzzles
Figure 2 Internal and External Determinants of Confidence that are actively under study, but which have neither been
in a Paradigm solved nor yet recognized as anomalies. Four flows con-
nect the different categories. Most puzzles, once formu-
lated and attacked, will be solved, adding to the cumu-
lative stockpile of knowledge generated by the paradigm.
Such puzzles flow into the class of solved puzzles via the
puzzle-solving rate. But as the intrinsic difficulty of puz-
zles grows, a growing number will resist solution long
enough to be recognized as anomalies. Anomalies may
sometimes be resolved, adding to the stock of solved puz-
zles via the anomaly resolution rate. The shifting balance
between these flows determines the behavior of the sys-
tem.

Figure 3 The Puzzle-Solving Sector

any evidence contrary to their beliefs. Practitioners with

only lukewarm commitment, lacking firm reasons to ac-
cept or reject the paradigm, are far more likely to alter
their beliefs in the face of anomalies.
The external factors affecting confidence encompass
the way in which practitioners in one paradigm view the
accomplishments and claims of other paradigms against
which they may be competing. We distinguish between
the dominant paradigm, defined as the school of thought
that has set the norms of inquiry and commands the al-
legiance of the most practitioners, and alternative para-
digms, the upstart contenders. The confidence of practi-
tioners in a new paradigm tends to increase if its

326 ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

3.3. Puzzle Formulation and Puzzle-Solving Rates finite when the stock of cumulative solved puzzles
The rate at which scientists formulate and solve puzzles reached some finite value, just as no amount of effort can
depends on the number of puzzles under study, the frac- bring any diamonds out of a mine once it is played out.
tion of practitioners involved in puzzle solving, the frac- We make the less restrictive assumption that the puzzle-
tion of their time devoted to puzzle solving, and the av- solving potential of paradigms is infinite, though it rises
erage difficulty of the puzzles (Figure 4). continuously on the margin as solved puzzles accumulate.
The average difficulty of new puzzles depends on how As the difficulty of puzzles grows, puzzle solving may
far the root metaphor defining the paradigm has been ex- slow and more unsolved puzzles may become recognized
tended. As described above, the average difficulty of puz- as anomalies. If the stock of anomalies grows too large,
zles is assumed to rise as the paradigm is applied to phe- the confidence practitioners have in the “truth” or utility
nomena increasingly removed from the original domain of the paradigm may fall. The collapse of confidence is
for which the paradigm was formulated. Specifically, the self-reinforcing: anomalies erode confidence, and falling
average difficulty of new puzzles to be solved, D, rises confidence increases the ability and willingness of prac-
as the number of puzzles the paradigm has solved grows. titioners to perceive the gaps in the theory.
We assume The majority of practitioners will usually be involved
in puzzle solving, while some will be working to resolve
D ⳱ (S/C)c, (1) anomalies and others try to generate alternatives or en-
gage in other activities such as administration or popu-
where S is the cumulative number of solved puzzles. The larization. The distribution of practitioner effort among
nominal solved puzzle reference, C, represents the intrin- these three categories is a function of confidence in the
sic capability of each paradigm, and c is the rate at which paradigm. The higher the confidence, the greater the frac-
difficulty rises with cumulative progress. When c ⬍ 1, tion of practitioners involved in normal science. As con-
the rate at which puzzle difficulty rises with cumulative fidence falls, more practitioners turn their attention to
progress becomes progressively smaller, while c ⬎ 1 in- anomaly resolution or altogether away from the normal
dicates the difficulty of puzzles on the margin rises ever science they increasingly come to doubt.
faster. For parsimony we assume c ⳱ 1. Small values of
the reference capability C mean a paradigm’s intrinsic 3.4. Anomaly Recognition Rate
explanatory power is low—the difficulty of new puzzles Anomaly recognition is a subtle psychological process
rises rapidly as normal science proceeds. Large values (Lightman and Gingerich 1992). Kuhn notes that anom-
indicate a more powerful paradigm, one that could en- alies are not simply experiments that run counter to ex-
compass a wider array of phenomena. Note that our for- pectation, as there are always disagreements between data
mulation differs from that of Masterman (1970), who and theory. Rather, a puzzle becomes recognized as an
viewed paradigms as analogous to nonrenewable re- anomaly when normal science repeatedly fails to resolve
sources, arguing that the domain of applicability for any the differences. Kuhn (1970, p. 82) argues that “One
paradigm is finite, so all attempts to extend it further source of the crisis that confronted Copernicus was the
would yield only anomaly. Her assumption would mean mere length of time during which astronomers had wres-
puzzle-solving difficulty in the model would become in- tled unsuccessfully with the residual discrepancies in
Ptolemy’s system.” Similarly, we assume that the longer
Figure 4 Determinants of the Puzzle-Solving Rate an unsolved puzzle has resisted solution, the greater the
chance it will be recognized as an anomaly. Thus, the
probability a puzzle is recognized as an anomaly rises as
the average difficulty of puzzles rises. However, recog-
nition of anomalies also depends on the degree to which
practitioners are conditioned to see reality as consistent
with their paradigm. Kuhn cites the Bruner-Postman play-
ing card experiments to illustrate how a paradigm con-
ditions perception, concluding “In science, as in the play-
ing card experiment, novelty emerges only with
difficulty, manifested by resistance, against a background
provided by expectation” (1970, p. 62ff). Thus in the
model, the average time required to recognize an unsol-
ved puzzle as an anomaly depends on practitioners’ level

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 327

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

of confidence in the paradigm. High levels of confidence would overturn 30 years of our thinking about strong in-
slow the recognition of anomalies as practitioners’ ex- teractions,” he “as a tenured Nobel laureate, has the ‘lux-
pectations, behaviors, and even perceptions become in- ury’ of continuing the search.” Others caution that “a
creasingly conditioned to be consistent with the para- younger scientist trying to make a reputation would be
digm. Decreases in confidence will cause more of the well-advised to avoid this line of work.” (Nadis 1996, pp.
puzzles under attack to be considered anomalous as prac- 1361–1362).
titioners’ skepticism and doubts grow.
3.6. Practitioner Population
3.5. Anomaly Resolution Rate
The population of practitioners committed to each para-
The rate at which anomalies are resolved depends on the
digm is endogenous, increasing with recruitment and de-
number of practitioners in sanctioned research, the frac-
creasing with retirement of elder scientists and defection
tion of those involved in anomaly resolution, and the av-
of others to competing paradigms. Without loss of gen-
erage difficulty of anomalies (Figure 5). Anomalies are
erality we assume the total population of scientists is con-
assumed to be more difficult to solve than puzzles, and
stant: scientists who leave one paradigm enter another;
as the difficulty of puzzles increases, the difficulty of
and entry of young scientists is balanced by retirement of
anomalies rises as well. The fraction of practitioners in-
the old. The assumption of constant total population sim-
volved in anomaly resolution depends on the balance be-
tween the number of anomalies and the acceptable num- plifies the interpretation of the results but is in no way
ber. The acceptable number of anomalies is the number essential to the main conclusions. Practitioners defect
that can be tolerated without losing confidence in the par- based on their confidence relative to the confidence of
adigm. If the number of anomalies increases, additional those in the dominant paradigm (Figure 6). The greater
practitioners are drawn into anomaly resolution in an at- the (negative) discrepancy between a challenger’s confi-
tempt to solve the major outstanding problems challeng- dence and confidence in the dominant paradigm, the
ing the theory. This negative feedback is comparatively larger the proportion of the challenger’s practitioners that
weak, however: Kuhn argues that most practitioners are will defect. Recruitment is proportional to a paradigm’s
reluctant to work on anomalies, preferring instead the relative attractiveness and its total number of practition-
relative safety and professional rewards of puzzle- ers. The greater a paradigm’s attractiveness, the greater
solving. The belief that anomaly hunting may be hazard- the proportion of defectors from other paradigms it will
ous to your career is widespread among scientists today recruit. Attractiveness is proportional to the number of
and often reinforced in the professional journals. Exam- practitioners since large paradigms are assumed to get
ples abound: a 1996 news article in Science reports Nobel more funding, train more students, and have a larger voice
laureate Martin Perl’s efforts to detect free quarks, a phe-
nomenon counter to the predictions of quantum chro-
modynamics, the long-successful theory of the strong Figure 6 Internal and External Determinants of Practitioner
force pioneered by Murray Gell-Mann and George Zweig Recruitment and Defection
in the 1960s. Though Perl asserts “a positive finding

Figure 5 Determinants of the Anomaly Resolution Rate

328 ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

in tenure and other peer-career decisions than small par- for more than one new paradigm to emerge. In this case
adigms. Attractiveness also depends on the confidence of the newly created paradigms will vie for ascendancy not
the paradigm’s practitioners, capturing the competence of only against the dominant paradigm but against one an-
the members, the capability of their tools, and the excite- other.
ment and enthusiasm flowing from a successful endeavor.
3.7. The Creation of New Paradigms 4. Exploring the Dynamics of Paradigm
We model the creation of a new paradigm as a stochastic Development
event whose probability depends upon the distribution of We begin by simulating the model with fully endogenous
practitioner activities in the currently dominant paradigm. competition among paradigms. The initially dominant
Practitioners may toil in normal science (puzzle solving), theory (Paradigm 1 [P1]) is initialized in the midst of
anomaly resolution (the attempt to reconcile anomalies normal science, and new theories are created stochasti-
with the current paradigm), and other activities (described cally, with a probability depending upon the vitality of
by Kuhn as including philosophical reconsideration of the the dominant paradigm as specified by Equation (2). The
paradigm and other activities not sanctioned by the dom- intrinsic capability of each new paradigm is determined
inant paradigm). In general, each of these activities may by a host of factors including the richness of the theo-
result in the creation of a new paradigm, but the proba- retical constructs emerging from the paradigm’s root
bility that a new paradigm is created as a result of a prac- metaphor and of course the particular genius of the par-
titioner year of effort devoted to each activity may differ. adigm’s creators. Thus, the rate at which puzzle solving
Thus: becomes difficult as solved puzzles accumulate (the par-
P(Creation)t ⳱ 兺i p*N
i i,t, i 僆 {PS, AR, OA}, (2) adigm’s inherent potential, C) is stochastic. Specifically,
C is drawn from a lognormal distribution (truncated such
where that C ⱕ 800). Otherwise all paradigms have identical
P(Creation)t ⳱ probability a new paradigm is created at structure and parameters.
time t; Figures 7a and 7b show the first 1400 years of a rep-
Ni,t ⳱ number of practitioners in the dominant resentative simulation. New paradigms are created sto-
paradigm engaged in activity i at time t; chastically, but the probability of creation is endogenous,
pi ⳱ probability of creating a new paradigm per
practitioner year of effort in activity i;
{PS, AR, OA} ⳱ Activities: Puzzle Solving, Anomaly Res-
Figure 7 A Typical Simulation Showing Competition and
olution, Other Activities, respectively.
Succession among Paradigms: Random Potential
Following Kuhn, we assume pAR ⬎ pOA ⬎ pPS: Normal 7a
Explanatory Power
science is unlikely to produce new paradigms, focused as
it is on solving puzzles within the context of the existing
paradigm. Other activities are more likely to produce a
new paradigm, while effort devoted to anomaly resolution
is most likely to result in the creation of radical new the-
ories (the values of these parameters are small enough
that the overall probability of creating a new paradigm in
any given year is low). In the model, the distribution of
effort among these three activities is endogenous. Thus
the probability that a new paradigm will be created in any
time period is endogenous and will vary as practitioner 7b
effort changes in response to the changing health of the
dominant paradigm.
Once a new paradigm is created, we assume it begins
with a small number of practitioners, a confidence level
of 0.5 (neutral), a very small stock of solved puzzles, and
no initial anomalies. The newly launched paradigm must
then compete for members against the dominant para-
digm. During a period of crisis the probability of creating
a new paradigm may rise and remain high long enough

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 329

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

as specified in Equation (2). Each new paradigm is en- year 580 is matched by an equally rapid drop as its prac-
dowed with a randomly-selected intrinsic explanatory titioners quickly exhaust the limited potential of its un-
power [the parameter C in Equation (1)]. The simulation derlying metaphor, making way for paradigm 14. Figure
yields a succession of dominant paradigms in which the 9 illustrates the details of P14’s life cycle. In the early
initial paradigm gives way to challengers, each of which period (⬇ years 560 to 610), confidence rises dramati-
goes through the typical life cycle as described by Kuhn, cally, since puzzle-solving progress is rapid and anoma-
though with variations in length and timing. Because all lies are low. The paradigm, initially untested, proves ca-
paradigms have identical structure and parameters, all dif- pable of solving puzzles, and thus attracts more
ferences in outcomes are due only to two factors: (1) the practitioners, further boosting confidence.
intrinsic capability with which each is endowed; and (2) The simulation illustrates how multiple positive feed-
the competitive environment (number and state of other back processes cause the self-reinforcing rise of a new
paradigms) at the time of their founding. theory. Figure 10 shows a causal diagram highlighting
What is most interesting is not what the figures display two of the positive feedback loops that cause an initially
but what they conceal. Most new theories face early ex- unorganized and weakly committed group of practitioners
tinction. As evident in Figure 7a, paradigms 2–4, 7, 9– to coalesce into a highly focused paradigm (for clarity
12, 15, and 17–18 never become dominant, illustrating negative loops are not shown). In causal diagrams, arrows
what Kuhn (1970, pp. 136–143) calls the invisibility of indicate the direction of causality. Signs (“Ⳮ” or “ⳮ”)
revolutions, where the linear and cumulative character of at arrow heads indicate the polarity of relationships: a
“Ⳮ” indicates that an increase in the independent variable
normal science portrayed in the textbooks conceals the
causes the dependent variable to increase above what it
contentious character of actual scientific practice. The
would have been, ceteris paribus (and a decrease causes
simulation replicates the “punctuated equilibrium” pat-
a decrease). A “ⳮ” indicates that an increase in the in-
tern described by Kuhn and observed in many other
dependent variable causes the dependent variable to de-
fields, including organizational theory (Gersick 1991,
crease below what it would have been. That is, X → ⳭY
Tushman and Anderson 1986). ⇒ (⳵Y/⳵X) ⬎ 0 and X → ⳮY ⇒ (⳵Y/⳵X) ⬍ 0. Positive
The endogenous forces underlying a paradigm’s evo- loop polarity, denoted by (Ⳮ) in the loop identifier, in-
lution are best illustrated by focusing on the life cycle of dicates a self-reinforcing (positive feedback) process.
a particular paradigm. Figure 8 enlarges that portion of Negative (ⳮ) loop polarity indicates a self-regulating
Figure 7a portraying the life cycle of P14. Around year (negative feedback) process (Richardson and Pugh 1981).
500, paradigm 8 is in the full flower of normal science, Rising confidence and successful puzzle-solving boost
with 100% of the practitioners, a high level of confidence, practitioner confidence, leading to more focused and suc-
and few anomalies. Paradigm candidates 9–12 are, by cessful effort, articulation and improvement of theory and
chance, created during the period of normal science and technique, and still greater success in puzzle solving, fur-
quickly perish. However, the continued success of the ther boosting confidence and attracting still more mem-
dominant theory P8 leads practitioners to apply it to more bers. Rising confidence, skill, and familiarity with the
and more phenomena. Anomalies slowly accumulate as paradigm increasingly condition practitioner perceptions
puzzles gradually become more difficult to solve, even- and expectations, suppressing the recognition of anoma-
tually leading to crisis and a drop in confidence. Para- lies; a low level of anomalies further increases practition-
digms 13 and 14 both arise during the crisis of paradigm ers’ confidence in and commitment to the theory. These
8 (around years 545 and 566, respectively). By chance, and other positive feedbacks (shown in Figure 10) boot-
P13 has very low inherent potential. Its rapid rise around strap paradigm 14 into dominance by around year 625,
its metaphor, method and metaphysics triumphant over
the now-discredited P13.
Figure 8 The Rise and Fall of Paradigm 14 (From Figure 7)
4.1 Normal Science
During the successful period of normal science (approx-
imately years 620 to 830) practitioners focus their efforts
on puzzle solving and are blinded to potential anomalies
by their faith in the paradigm. The probability a new par-
adigm is created falls [see Equation (2)]. In this fashion,
success suppresses the generation of new competitors
which might challenge the dominant paradigm, leading

330 ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

Figure 9 The Life Cycle of Paradigm 14

to further success. Through this self-reinforcing feedback year 750 (Figure 8). Such challengers usually perish in
a successful theory alters its own environment in ways the face of competition with the still successful dominant
that provide further advantage. This important dynamic paradigm. Indeed, P15 vanishes within a few years.
operates through the training of graduate students, which 4.2 Crisis
reproduces the worldview and prejudices of the dominant As Paradigm 14 is elaborated and extended beyond the
theory and socializes them in the accepted canon of prior scope of its root metaphor, puzzles gradually become
work, through the control of institutions via appointments more difficult to solve. Anomalies begin to accumulate.
and tenure, through resource allocation via peer review Confidence begins to fall, slowly, around year 780. As
of grant proposals, and through access to journals via con- anomalies increase, a few practitioners leave puzzle-
trol of editorial boards and the selection of referees. How- solving, eroding progress and decreasing confidence fur-
ever, occasionally a new theory does emerge during pe- ther. Practitioners, increasingly sensitive to the para-
riods of normal science, such as Paradigm 15 just before digm’s limitations, become more apt to see difficult

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 331

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

Figure 10 Some Positive Loops Driving Path-Dependent Be- period to accelerate the collapse of a paradigm which has
havior accumulated sufficient anomalies for confidence to begin
Note: Shows two of the positive loops that cause initially uncommitted falling.
and unorganized practitioners to coalesce into a highly focused par- The simulations raise a number of important questions.
adigm. (Negative loops are not shown.) Why do some paradigms rise to dominance while others
quickly wither? Does the fate of a new paradigm depend
on its intrinsic capability to explain nature or on situa-
tional contingencies surrounding its birth? Does ‘truth’
eventually triumph as better theories defeat inferior ones,
or is timing everything?
There is evidence for both positions in the results. Sup-
porting the view that intrinsic explanatory power is criti-
cal are examples such as Paradigm candidate 13, which
rapidly exhausts its low intrinsic potential and never
achieves dominance. However, intrinsic capability does
not explain the fate of many others. Consider Paradigms
8 and 9 in Figures 7a and 7b, launched around years 199
and 203, respectively. Although they emerge only about
four years apart, during the crisis of Paradigm 5, P8
comes to dominate the field, while P9 eventually perishes.
Here the contingency of outcomes on situational factors
is decisive. Paradigm 8 does not succeed because of a
head start in attracting practitioners: between years 212
puzzles as anomalies, further increasing anomalies and and 215 it actually has the same number as P9. Nor is
decreasing confidence. The positive feedbacks that pre- Paradigm 8’s success a result of superior explanatory
viously caused membership to rise now cause accelerat- power: by chance, P9 is endowed with a potential 13%
ing collapse. By year 850 the paradigm is in crisis. greater than P8. The difference in their destinies lies in
As the number of practitioners engaged in normal sci- their levels of confidence. In the year 212 Paradigm 8,
ence falls, and those seeking to resolve anomalies grows, though equal in size to P9, is slightly more attractive be-
the probability that a new paradigm will be created rises. cause its adherents, having had a 4-year lead over P9 in
Around year 855 a new paradigm is in fact created (P16 solving puzzles, have been able to articulate their para-
in Figure 8). Because the new theory emerges during the digm more coherently and persuasively than their chief
crisis of Paradigm 14, it quickly gains adherents while rivals. The small advantage held by P8 is amplified as
P14 loses members. Confidence and membership in P16 success begets success through the many positive loops
then accelerate sharply through the same positive feed- surrounding the emergence process (Figure 10). Para-
backs which earlier led to the success of P14. The cycle digm 8 eventually dominates science, while Paradigm 9
is completed as Paradigm 14’s confidence and member- slowly fades into obscurity, to be remembered, if at all,
ship eventually fall to 0, while P16 grows to dominate as a blind alley, foolish error, or curiosity.
the field. What was once uncontested “truth” is now seen The simulations illustrate the subtle interplay between
as primitive error. Paradigm 17, created around year 870, endogenous feedback processes and contextual, situa-
is quickly crushed by the now dominant P16. tional factors in determining the dynamics and succession
4.3 Positive Feedback and Path Dependence of paradigms. The basic life cycle of paradigms is deter-
The many positive feedbacks described above create the mined by the recursive, reflexive feedback loop structure
self-organizing dynamic by which uncommitted and un- discussed above. Figure 11 shows some of the positive
organized practitioners coalesce into a highly focused feedback loops that act to differentiate competing para-
paradigm with a productive program of normal science. digms even when they are initially quite similar (the many
Through these feedbacks a successful paradigm alters its negative feedbacks are not shown). These positive feed-
environment by suppressing the creation of competitors backs boost confidence and rapidly generate a focused
and rapidly starving any that do emerge of the resources community from a promising but unexplored new idea.
they would need to succeed. The same feedback pro- They give a paradigm with an initial advantage an edge
cesses operate in the opposite direction during the crisis in recruitment of new members, leading to still greater

332 ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

Figure 11 Some Positive Feedback Loops that Create Path- previously weaker rival. Such random events might in-
Dependent Behavior clude factors related to the theory, such as the announce-
ment of an important experimental result, but can also
include events wholly outside of science, such as the ill-
ness of the candidate’s champion or political upheavals
that disrupt the work of key people. However, as the posi-
tive loops confer greater and greater advantage to one of
the contending theories, the likelihood that particular
events can overcome the advantage of the leader rapidly
diminishes, until the system has effectively “locked in”
to a solution. Once such lock-in has occurred, the domi-
nance of the winning theory is assured (until its own cri-
sis). Yet which particular theory becomes dominant can
be a matter of chance events and small perturbations early
in the emergence phase.
The prevalence of positive feedback processes in the
dynamics means that historical contingencies attending
the creation and early years of a new theory strongly con-
dition their fate. While it is obvious that the creation of
a new theory is intrinsically unpredictable, the simulation
advantage, amplifying small fluctuations in local condi- shows clearly that, once created, the likelihood any given
tions to macroscopic significance, and leading to path de- new paradigm survives its founding and grows to domi-
pendence. Consider Paradigm i in Figure 11. If the num- nance is strongly contingent on the environment into
ber of anomalies and solved puzzles in Paradigm i which it is launched—an environment that in turn de-
compare favorably with the accomplishments of compet- pends on the history of the paradigms preceding it. The
itor paradigms, the confidence of practitioners in i will prevalence of positive feedback processes in paradigm
rise and the confidence of those in its competitors will development means that the evolution of the system as a
fall. The attractiveness of i relative to others grows, thus whole is strongly path-dependent.
strengthening i and weakening its competitors. The net The ability of positive feedback processes to create
flow of practitioners into Paradigm i will increase the gap path-dependent lock-in to particular equilibria from an
in solved puzzles between i and its competitors, causing initially undifferentiated choice set has been amply doc-
the gap in confidence to widen still further. The self- umented in biological, economic, technological, and
reinforcing differentiation continues until one paradigm other systems. Examples beyond the familiar QWERTY
emerges dominant and the others become extinct. These keyboard include the universal left-handed chirality of
same loops are responsible for the resistance of the dom- proteins throughout the plant and animal kingdom, the
inant paradigms to challenges, as high confidence sup- choice of technological standards such as the gauge for a
presses the creation and retards the progress of new the- railroad or the shape of electrical plugs, the designation
of Greenwich as the prime meridian, the length of the
ories. High confidence leads to normal science and low
standard meter in Paris (or the choice of the metric over
anomalies, suppressing the type of inquiry likely to lead
the English system), the dominance of the IBM/Microsoft
to the creation of new paradigms [Equation (2)]. And
Windows architecture for personal computers, and the
should by chance a new theory be created, the high con- growing dominance of the major world languages while
fidence and low anomalies of a dominant paradigm make the languages of small indigenous peoples become ex-
it unlikely a new theory can succeed, even if it has high tinct.
intrinsic explanatory potential. Note that once a dominant Even when all choices are equally attractive ex ante as
paradigm begins to experience depletion of its root meta- in the choice of the length of the standard meter or the
phor, these same loops operate as vicious cycles, accel- shape of electrical plugs, the symmetry is broken by mi-
erating the collapse. croscopic noise or external perturbations. The positive
In the early phase of a competition between two or feedbacks then amplify these small initial differences to
more paradigm candidates, when the differences among macroscopic significance. Once a dominant design has
the competing theories are small, chance events can per- emerged, the costs of switching become prohibitive, so
turb the system sufficiently to shift the advantage to a the equilibrium is self-enforcing, at least until there is an

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 333

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

architectural shift that renders the dominant design ob- Table 1 Ability of LOGIT Model To Predict a Given
solete (Henderson and Clark 1991), as in the replacement Paradigm’s Rise to Dominance
of analog broadcast television by HDTV. The model reduces the error rate in predicting dominance by half
compared to chance.

5. Intrinsic Capability or Historical Predicted

Contingency?
Nondominant Dominant Total
To test the argument above and quantify the roles of in-
trinsic versus contingent factors, we analyzed the pooled
Actual Nondominant 641 35 676
results of 57 2000-year model runs. The only parameters Dominant 135 215 350
varied were the paradigm’s intrinsic explanatory power Total 776 250 1026
and the random number seed affecting the launch of new
paradigms. To eliminate initial transients and end effects, kb ⳱ 0.51; Proportion correct ⳱ 0.83; Sensitivity ⳱ 0.86; Specificity
the first and last five paradigms of each simulation are ⳱ 0.83
eliminated from the analysis. There are 350 dominant par-
adigms and 676 never-dominant paradigms in the sam-
ple.2 The model reduces the error rate by half compared to
We consider a LOGIT model with three explanatory chance.
variables: intrinsic capability (C), the confidence in the The regression results (Table 2) show that all estimated
dominant paradigm at the time the new paradigm is coefficients have the predicted signs. A new paradigm’s
launched (CPdom), and the number of competitor para- chances of success rise with greater intrinsic capability,
digms (not including the dominant paradigm) each new a weaker dominant paradigm, and a smaller number of
paradigm faces when launched. Since the probability of competitors. However, the effect of a paradigm’s intrinsic
success need not depend linearly on the number of com- explanatory potential, C, on its probability of success is
petitors, we treat the number of competitors as a cate- not significant, while the contextual variables are highly
gorical variable. Thus, the dummy variable COMPETi ⳱ significant. In particular, the confidence level of the dom-
1 if the number of competitors equals i at the time each inant paradigm at the time a new contender is created has
paradigm is founded, and zero otherwise, for situations a strong effect on the challenger’s likelihood of success.
of up to four competitors: Similarly, the chances of success fall precipitously as the
Pt(Dom) ⳱ 1/(1 Ⳮ exp(ⳮ(b0 Ⳮ b1C number of competitors rises. The estimated coefficients
illustrate the weak role of intrinsic capability in compar-
4
ison to the contextual factors in determining whether a
Ⳮ b2CPdom
t Ⳮ 兺
i⳱1
wiCOMPETi,t))), (3) paradigm becomes dominant. Table 3 shows how the

where the subscript t indicates that the probability is cal-

culated in the year each new paradigm is created.
Table 2 LOGIT Regression Comparing Intrinsic and
Table 1 shows how well the model predicts successes Contextual Factors in Likelihood of Success
and failures, where an estimated probability greater than
0.5 is interpreted as a prediction that the paradigm be- Variables characterizing the competitive environment have a
comes dominant and estimated probabilities ⬍0.5 are in- strong impact on the likelihood of success while the intrinsic
terpreted as predictions of failure. Overall, 83% of the explanatory power of a paradigm (C) has only a weak effect.
cases are predicted correctly. The sensitivity and speci- Indep. Variable Estimated Coeff. Standard Error t-statistic
ficity of the model are roughly equal (both ⱖ0.83), in-
dicating the model’s error rate is about the same for pre- Constant 5.44* 0.52 10.42
dictions of dominance when the paradigm in fact fails C 6.86e-4 4.34e-4 1.58
versus predictions of failure when the paradigm in fact CPdom ⳮ7.27* 0.55 ⳮ13.19
COMPET1 ⳮ1.43* 0.23 ⳮ6.17
succeeds. The statistic kb ⳱ 1ⳮ ((errors|model)/(er-
COMPET2 ⳮ4.99* 0.52 ⳮ9.54
rors|no model)) measures how much the model improves
COMPET3 ⳮ13.52 50.00 ⳮ0.27
prediction success compared to the chance success rate. COMPET4 ⳮ14.65 147.91 ⳮ.099
In the absence of the model, the best guess is that any
paradigm picked at random fails, since fully two-thirds N ⳱ 1026
of the paradigms in the sample never become dominant. *P ⬍ 0.05

334 ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

Table 3 The Influence of Intrinsic and Contextual Factors Figure 12 Logit Model Results
on the Probability a New Paradigm Becomes The probability a given paradigm rises to dominance as it depends
Dominant on confidence in the dominant paradigm and the number of compet-
itors at the time it is created.
Probability of Dominance
Conditions at Emergence (with 95% confidence interval)

One Competitor 0.18 ⱕ 0.25 ⱕ 0.40

Two Competitors 0.004 ⱕ 0.01 ⱕ 0.05
C ⳱ 100, No Competitors 0.44 ⱕ 0.53 ⱕ 0.65
C ⳱ 800, No Competitors 0.52 ⱕ 0.63 ⱕ 0.79
C ⳱ 100, One Competitor 0.15 ⱕ 0.22 ⱕ 0.38
C ⳱ 800, One Competitor 0.21 ⱕ 0.31 ⱕ 0.57

probability that a given paradigm rises to dominance de-

pends on its intrinsic capability compared to the contin- Pt(Dom) ⳱ 1/(1 Ⳮ exp(ⳮ(5.44 Ⳮ 0.000686Cavg
gent factors. Each row shows the probability a new par-
adigm becomes dominant given the conditions listed and ⳮ 7.27CPdom Ⳮ wi))). (4)
assuming all other explanatory variables take their mean For new paradigms competing only against the dominant
values, along with the 95% confidence interval for the paradigm, the probability of dominance is given by the
probability.3 The table clearly illustrates the relatively curve in the upper right. Curves are also displayed for
weak influence of intrinsic capability (C) on the proba- environments with two and three competitors. The curve
bility of becoming dominant. On average, a paradigm for four competitors has probabilities ⬇ 0. For all but the
launched in the presence of one competitor has only a smallest values of CPdom, the greater the number of com-
25% chance of succeeding. If there are two competitors petitors, the less likely a new paradigm becomes domi-
the probability drops to only 1%. Even if a new paradigm nant. Likewise, the greater the value of CPdom, the less
is endowed with the maximum amount of intrinsic ca- likely a new paradigm is to become dominant. The re-
pability (C ⳱ 800), the consequences for dominance are gression results and Figure 12 show the number of com-
still strongly mediated by the contingent factors at emer- petitors existing at the time a new paradigm is created
gence. Thus, for example, while the probability a new strongly influences its fate. When CPdom is between about
paradigm becomes dominant reaches 0.63 if it faces no 0.1 and 0.6, a new paradigm stands a better than even
competitors, it has only a 0.31 chance of surviving when chance of becoming dominant if it faces a total of two
it faces one additional competitor. Furthermore, increas- competitors or less, and will likely fail if there are three
ing the intrinsic capability of a new paradigm by a factor or more competitors. When CPdom is between about 0.6
of eight boosts the probability of success by only about and 0.8, the new paradigm is more likely than not to be-
ten percentage points. Contingent factors at the time of come dominant if it faces only the dominant paradigm,
likely to fail if it faces two competitors, and almost sure
emergence far outweigh the influence of intrinsic capa-
to die if faces three or more competitors.
bility.
Thus the likelihood that a new paradigm will rise to
The relative importance of intrinsic capability C versus
dominance in the model is overwhelmingly determined
the contextual factors CPdom and the number of compet- by historical contingencies and only weakly influenced
itors COMPETi is further illustrated in Figure 12. Each by its intrinsic explanatory power. The relative impor-
point in the plot represents the probability of dominance tance of inherently unpredictable situational factors is not
of a particular paradigm, as predicted by its intrinsic ca- particularly sensitive to the parameters. Rather it is a con-
pability, the number of competitors it faces at birth (ex- sequence of the many positive feedbacks by which par-
cluding the dominant paradigm), and the confidence of adigms bootstrap themselves from doubt to normal sci-
the dominant paradigm it faces. The smooth curves plot ence (Figure 10).
the predicted probability of dominance as CPdom varies But how do internal and contextual factors interact to
over the [0,1] interval, for each number of competitors determine the longevity of those paradigms that survive
and assuming intrinsic capability takes on its mean value their founding and go on to dominate their field? Do in-
Cavg ⳱ 371.4; that is: trinsically powerful paradigms remain dominant longer

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 335

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

than their weaker counterparts? Here one would expect native. With high confidence, skill, and a productive
that the paradigms with greater explanatory power should agenda to focus research on the puzzle solving of normal
survive longer. Figure 13 shows a paradigm’s longevity science, the new paradigm is poised to realize its intrinsic
as a function of its intrinsic capability only for those par- potential. For these paradigms, longevity follows the
adigms that went on to become dominant. The figure power law scaling with intrinsic capability fairly closely
shows those successful paradigms that emerged when the (L ⬀ C.78, R2 ⳱ .64).
dominant paradigm against which they had to compete Paradigms emerging when confidence in the dominant
was strong (CPdom ⱖ 0.75; N ⳱ 131) and those that paradigm is low face a competitor in crisis. Thus, as new
emerged when the dominant paradigm was weak (CPdom and unproven as the new paradigm is, it nonetheless
ⱕ 0.25; N ⳱ 104).4 As expected, for those paradigms quickly wins new members. The rapid influx of new prac-
surviving their founding, longevity is significantly related titioners means the rate of effort is high. Rapid growth in
activity means the average difficulty of puzzles rises
to intrinsic capability. In both cases, longevity roughly
quickly, increasing the number of unresolved puzzles
follows a power law in capability L ⳱ ␣Cb. Such power
likely to be seen as anomalies. Most important, the influx
law scaling is common in a wide range of dynamical sys- of new practitioners occurs when confidence is low,
tems (Schroeder 1991). meaning basic disagreements about methods, data, and
However, Figure 13 shows that even for successful par- criteria for validity still persist. Without the learning and
adigms, historical contingencies matter greatly to their skill experience afford, without the acculturation and per-
longevity. Those paradigms emerging when the dominant ceptual filters provided by a well-articulated paradigm,
paradigm is very strong (with confidence ⱖ 0.75) actually disagreements and anomalies arise at an alarming rate. If
survive significantly longer than those emerging when enough anomalies accumulate, confidence can fall. Fall-
their principal competitor is weak (confidence ⱕ 0.25). ing confidence causes people to perceive anomalies still
The median longevity for those emerging when CPdom ⱖ more readily, further decreasing confidence. The new par-
0.75 is more than twice as great than that for those emerg- adigm rapidly disintegrates, its high intrinsic potential
ing when CPdom ⱕ 0.25. The differences in outcomes largely unrealized. For paradigms emerging when their
arise from differences in the circumstances attending the principal competitor is weak, longevity scales with in-
birth of these successful theories. Paradigms emerging trinsic capability only as L ⬀ C.66, and the variance of
when confidence in the dominant paradigm is relatively longevity around the best fit is much greater.
high face strong competition. Most scientists are still sat- The results show the strong role of contingent, histori-
isfied with the dominant paradigm, so the rate of recruit- cal factors even for those paradigms that become domi-
ment to the new paradigm is relatively slow. During this nant. As expected, the probability of surviving the found-
time, however, the few adherents of the new paradigm ing period and becoming dominant is negatively related
are able to solidify the foundations of their theory and to the intensity of the competitive environment. However,
develop skill with their tools and techniques. Anomalies counter to what one might expect, the more intense the
remain low as practitioners solve the relatively easy puz- competition, the longer the expected life of the successful
theories. There are two reasons. First, strong selective
zles for which their paradigm is well suited. Their con-
pressures during the emergence phase ensure that only
fidence rises. By the time the crisis of the dominant par-
those paradigm candidates with high intrinsic capability
adigm deepens and its members become disaffected, the
can survive. When selection pressure is weak, some par-
initial adherents of the new theory will have articulated adigm candidates with low intrinsic potential can become
it well enough to provide an attractive and viable alter- dominant. Second, and even more insidiously, when com-
petition is weak many paradigm candidates with high in-
Figure 13 Relationship between Longevity (L) of Successful trinsic potential die young as they grow too rapidly, over-
Paradigms and Intrinsic Explanatory Power (C) extending themselves before their members develop
Note: Paradigms that never become dominant are not shown. enough skill, understanding, and confidence to prevent
the accumulation of anomalies. Historical contingencies
not only determine which paradigms succeed but also
how long those that succeed may thrive.

6. Discussion: Guru Dynamics and

Management Fads
The dynamics generated by the model resemble the life
cycle of intellectual fads. Often a promising new idea

336 ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

rapidly becomes fashionable through excessive optimism, Sastry and Coen (1998) discuss positive feedbacks in or-
aggressive marketing, media hype, and popularization by ganizations. Of course formal models of innovation dif-
gurus. Many times the rapid influx of poorly trained prac- fusion as a process driven by positive feedback go back
titioners, or the lack of established protocols and methods, at least to Bass (1969), and conceptual models of such
causes expectations to outrun achievements, leading to a positive feedback processes can be traced to Myrdal’s
backlash and disaffection. Such fads are commonplace, (1944) “principle of cumulative causation,” Merton’s
especially in (quack) medicine and most particularly in (1948) theory of the self-fulfilling prophecy, and J. S.
the world of business, where “new paradigms” are rou- Mill’s (1848) theory of speculative bubbles (see
tinely touted in the pages of popular journals of manage- Richardson 1991 for the history of feedback theories in
ment, only to be displaced in the next issue by what many the social sciences). More recent work discusses the dif-
business people have cynically come to call the next “fla- ferences between the diffusion of ideas and of technolo-
vor of the month” (see Abrahamson 1996). No doubt gies, and the role of social networks and other factors in
many such fads have no intrinsic merit (in our terms, conditioning the strength of the positive loops driving
intrinsic capability C is low) so their rapid demise is the adoption, e.g., Rogers (1995), Valente (1995), and Kaufer
desired and rational outcome (similar to the fate of the and Carley (1993).
low potential Paradigm 13 in Figure 8). However, too The wide range of positive feedbacks identified above
many of these fads achieve broad acceptance and lead to can lead to the swift and broad diffusion of an idea with
large expenditures, only to suffer a backlash when they little intrinsic merit because the negative feedbacks that
fail to live up to their promise. might reveal that the tools don’t work operate with very
The theory developed here helps explain how this oc- long delays compared to the positive loops generating the
curs. Typically, a guru proposes a new theory, tool, or growth. In science there are often long delays between
process promising to address persistent problems facing the initial success of a theory and the execution and in-
businesses (that is, a new paradigm claiming to solve the terpretation of experiments that can test it. In the world
anomalies that have undermined the old paradigm.) The of social action, the delays are often even longer. Rigor-
early adopters of the guru’s method spread the word and ous follow up studies to assess the effectiveness of a new
initiate some projects. Even in cases where the ideas of management tool are notoriously difficult because of the
the guru have little merit, the energy and enthusiasm a inability to conduct controlled experiments in social sys-
team can bring to bear on a problem, coupled with Haw- tems, the essential participation of human beings in the
thorne and placebo effects and the existence of “low interventions, and the ambiguity of outcomes. The com-
hanging fruit” will often lead to some successes, both real bination of strong positive feedbacks promoting the
and apparent. Proponents rapidly attribute these successes growth of new management ideas and slow, weak nega-
to the use of the guru’s ideas. Positive word of mouth tive feedbacks revealing which are wheat and which chaff
then leads to additional adoption of the guru’s ideas. (Of predisposes the world of management to a succession of
course, failures are covered up and explained away; as in highly touted “new paradigms,” each shining brilliantly
science there is the occasional fraud as well.) Media at- for a few brief years only to be discarded once the neg-
tention further spreads the word about the apparent suc- ative feedbacks of follow-up evaluation lead to disaffec-
cesses, further boosting the credibility and prestige of the tion and the advent of a new guru with a new, more at-
guru and stimulating additional adoption. tractive theory. The same positive feedbacks can also lead
As people become increasingly convinced that the to inflated expectations, insufficient practitioner skill,
guru’s ideas work, they are less and less likely to seek or overly broad scope of application, and inadequate time to
attend to disconfirming evidence. Management gurus and resolve anomalies, causing some theories with high in-
their followers, like many scientists, develop strong per- trinsic capability to be abandoned too soon, as seen in the
sonal, professional, and financial stakes in the success of simulations.
their theories, and are tempted to selectively present fa- As discussed, our model shows that the likelihood a
vorable and suppress unfavorable data, just as scientists new theory will be created and gain significant popularity
grow increasingly unable to recognize anomalies as their is endogenous, rising as confidence in existing theories
familiarity with and confidence in their paradigm grows. falls. Thus we would predict a higher incidence of man-
Positive feedback processes dominate the dynamics, lead- agement fads during times of economic and social stress,
ing to rapid adoption of those new ideas lucky enough to when confidence in existing institutions and their moti-
gain a sufficient initial following. Hirshleifer (1995) and vating ideologies falters. Indeed, the rise of management
Bikhchandani et al. (1992) present similar models of fads fads has coincided with the slow growth, downsizing,
caused by positive feedbacks, and Sastry (1998) and globalization, rapid technical change, and other pressures

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 337

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

of the past few decades. These stresses constitute the thorny problems, thus identifying potential anomalies.
anomalies eroding confidence in existing organizational Bibliometric techniques could be used to determine how
structures and political ideologies. At the same time, our long a research problem (“puzzle”) has gone unsolved
model predicts that low confidence in existing institutions and gauge the number of researchers working on it, to
increases both the number of new theories lacking intrin- yield a measure of the difficulty of puzzles. Donovan et
sic merit that gain significant popularity and the number al. (1988), and Jacobsen and Bronson (1995) discuss the
of high potential ideas that die young as a result of the practical difficulties involved in such empirical tests;
skill dilution and insufficient learning caused by rapid Rappa and Debackere (1993) use survey and bibliometric
growth. Though economic stress may stimulate manage- tools to shed light on the demographics and attitudes of
ment innovation, it also increases the probability busi- scientists in several fields, illustrating how the constructs
nesses will both embrace useless theories and prema- in the model might be measured.
turely discard potentially useful ones.
The results of our model suggest that the long-term
success of new theories can be enhanced by slowing the 7. Conclusion
positive feedback processes, such as word of mouth, mar- Before turning to the conclusions, we pause to consider
keting, media hype, and extravagant claims of efficacy by the limitations of the model. All models (formal or oth-
which new theories can grow, and strengthening the pro- erwise) are inevitably less than the world their authors
cesses of theory articulation and testing, which can en- seek to portray. We agree with Cartwright (1983, p. 153)
hance learning and puzzle-solving capability. that models “are a work of fiction.” Of course the model
So-called “chaos” or “complexity theory” itself pro- is not comprehensive, nor does it capture all the subtleties
vides a recent example. The practical value of nonlinear of Kuhn’s theory. Rather, we seek to demonstrate that it
dynamics has repeatedly been demonstrated in physics is both desirable and possible to portray in a formal model
and the life sciences (see Chin et al. 1996, Costantino et the causal hypotheses embodied in written theories of sci-
al. 1997, and Sturis et al. 1991 for recent examples). entific endeavor and test whether they can generate the
However, rapid growth, fed by successful popularization dynamics as those authors see them. The process of for-
(e.g., Gleick 1987, Waldrop 1992) and ill-advised claims malizing such hypotheses helps to identify inconsisten-
for the universality of “complexity” as a “new paradigm” cies, implicit assumptions, glosses, and errors in the men-
for the reconstruction of the social as well as natural sci- tal simulations authors necessarily perform to infer the
ences have already led to a backlash (for example, dynamics of science from their theories of its structure.
Horgan 1995). Developing the full potential of complex- Such an endeavor is worthwhile as a complement to his-
ity theory, especially in the social sciences, requires more torical and sociological studies. Complete documentation
rigorous theory development and fewer popular articles of the model is available; we invite others to replicate,
extolling the virtues of the “new paradigm”, more studies critique, revise, and extend the model to test views of
testing the new theories and fewer anecdotal claims of scientific development different from ours.
efficacy, greater development of tools tailored for partic- The simulations suggest an important role for situa-
ular contexts, and fewer claims of universality. Without tional contingencies in the evolution of science. We find
such rigor, social scientists face the danger that, despite that the fate of a particular new theory or paradigm is
its high potential, “complexity theory” will soon be dis- strongly conditioned by the circumstances surrounding its
carded, perhaps prematurely, as yet another unfortunate creation, and only weakly influenced by its explanatory
case of physics envy. power or logical force (at least for theories above a min-
Testing our theory against real-world examples such as imum threshold of explanatory power). Environmental
the emergence of complexity theory poses daunting but conditions at the time a new theory is created, such as the
not insurmountable challenges. Testing the model empir- morale and confidence of practitioners in the old para-
ically requires measuring model constructs such as “con- digm and the number of contending alternative new the-
fidence,” “anomaly,” and “average difficulty of puzzles.” ories, powerfully determine whether a new theory will
Confidence might be measured through surveys or inter- rise to dominance or quickly perish. In particular, the sim-
views with relevant researchers, asking them to rate their ulations show new theories with great explanatory power
degree of belief in the theory. Content analysis of publi- frequently fail to attract a critical mass of adherents, while
cations in the field would also reveal the strength and weaker ones often triumph. The frequent eclipse of the
universality of the claims made by key practitioners, in- strong by the weak is not a pathological outcome, but
dicating their confidence level. Content analysis might rather a normal consequence of scientific activity as we
also be used to analyze critical reactions to particularly have modeled it.

338 ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

The interplay between intrinsic explanatory potential . . . the world is not invented or constructed. . . . [It] has been
and historical contingency is quite subtle. A paradigm’s experientially given, in part to the new inhabitants directly, and
inherent potential—its logical force and power to explain in part indirectly, by inheritance, embodying the experience of
nature—does influence its future development: of those their forebears. As such, it is entirely solid: not in the least
respectful of an observer’s wishes and desires; quite capable of
paradigms surviving their youth, those with high intrinsic
providing decisive evidence against invented hypotheses which
capability do remain dominant longer, on average, than fail to match its behavior. Creatures born into it must take it as
those that are weaker. But the impact of intrinsic capa- they find it. They can, of course, interact with it, altering both
bility on the longevity of any given paradigm is mediated it and themselves in the process, and the populated world thus
by the competitive conditions in the emergence period. altered is the one that will be found in place by the generation
In particular, weak competitive environments make it which follows.
more likely a new paradigm will rise to dominance, but
Through these feedbacks the world we inhabit is made;
can condemn even powerful paradigms to early deaths as
it is a world of nonlinear, disequilibrium dynamics in
they are extended too far and too fast, generating anom-
which, as Kuhn (1990, p. 12) says, “small changes . . .
alies and prematurely destroying confidence. On the other
can have large-scale effects.”
hand, though competition reduces the likelihood of sur-
vival, competition gives those that do survive time to Acknowledgments
bootstrap themselves into normal science, insulating them The authors thank Kathleen Carley, David Edge, Erik Mosekilde,
against mere disconfirmation, and ensuring they persist Michael Rappa, Anjali Sastry, Frank Sulloway, the referees, and sem-
until the anomalies ultimately causing revolution, in inar participants at MIT for helpful comments.
Kuhn’s words, “penetrate existing knowledge to the core” Endnotes
(Kuhn, 1970, p. 65). 1
Sterman (1985) provides a formal model of Kuhn’s theory represent-
Most important, however, competition does not serve ing the life cycle of a single paradigm; full documentation is provided.
to weed out the weak paradigms so the strong may grow. In this paper we extend the original model to allow for explicit com-
On the contrary, competition decimates the strong and petition among different theories.
2
weak alike—we found that intrinsic capability has but a In most of the simulations, intrinsic capability, C, for each paradigm
weak effect on survival. The mortality rate for paradigms was drawn randomly from a lognormal distribution truncated such that
seems to depend almost entirely on the environmental C ⱕ 800. In some runs all paradigms had identical intrinsic capabilities,
with C ⳱ 200, 300, or 400, to further reduce the variance and isolate
conditions surrounding their birth. This is a sobering re-
the role of historical contingencies. These restrictions do not affect the
sult, since we can never know the microlevel contingen-
model’s qualitative behavior.
cies of history that can prove decisive; here favoring an 3
To compute the 95% confidence intervals we drew 1,000 simulated
intrinsically weak paradigm, there killing an intrinsically parameter estimates from a multivariate normal distribution defined by
strong theory.5 These characteristics of the competition the estimated coefficients and variance–covariance matrix, sorted the
among paradigms are consequences of the powerful posi- resulting estimates of the probability of dominance, and extracted the
tive feedback processes operating within and among par- probabilities from the 25th and 975th values (see King et al. 1998).
4
adigms. These positive loops can amplify microscopic We omit from Figure 13 those paradigms emerging when the confi-
perturbations in the environment—the local conditions of dence level of the dominant paradigm was between 0.25 and 0.75 to
science, society, and self faced by the creators of a new simplify the presentation; including the full sample does not signifi-
theory—until they reach macroscopic significance. Such cantly alter the power law scaling or the result that longevity depends
on contingencies including the confidence of the principal competitor.
dynamics are the hallmark of path dependent evolution- 5
See Gould (1990) for a similar view applied to the evolution of life.
ary systems.
Contemplating the reflexive feedbacks between people References
and the world, Kuhn (1990, p. 7), in “The road since Abrahamson, E. 1996. Management fashion. Acad. Management Rev.
Structure,” captured the essence of path-dependence in 21(1) 254–285.
evolutionary systems, arguing that “scientific develop- Anderson, P., K. Arrow, D. Pines, eds. 1988. The Economy as an
ment must be seen as a process driven from behind, not Evolving Complex System. Addison-Wesley, Reading, MA.
Arthur, B. 1989. Competing technologies, increasing returns, and lock-
pulled from ahead—as evolution from, rather than evo-
in by historical events. Econom. J. 99 116–131.
lution towards.” Yet while acknowledging the role of the ———. 1994. Increasing Returns and Path Dependence in the Econ-
biological, cognitive, and social in the evolution of sci- omy. University of Michigan Press, Ann Arbor.
ence, Kuhn (1990, p. 10) argues forcefully that path- Bass, F. 1969. A new product growth model for consumer durables.
dependence does not mean the course of scientific devel- Management Sci. 15 215–227.
opment is entirely arbitrary or reality merely a social Barlas, Y. 1992. Comments on “On the very idea of a system dynamics
construction: model of Kuhnian science” System Dynam. Rev. 8(1) 43–47.

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 339

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

Bijker, W. E., T. P. Hughes, T. J. Pinch, eds. 1987. The Social Con- Hall, R. 1976. The system pathology of an organization: The rise and
struction of Technological Systems: New Directions in the Soci- fall of the old Saturday Evening Post. Admin. Sci. Quart. 33 1–
ology and History of Technology. MIT Press, Cambridge, MA. 23.
Bikhchandani, S., D. Hirshleifer, I. Welch. 1992. A theory of fads, Hannan, M. and J. Freeman. 1989. Organizational Ecology. Harvard
fashion, custom, and cultural change as informational cascades. University Press, Cambridge, MA.
J. Political Econom. 100 (5) 992–1026. Haxholdt, C., C. Kampmann, E. Mosekilde, J. D. Sterman. 1995. Mode
Black, M. 1954–1955. Metaphor. Proc. Aristotelian Soc. 55 284–285. locking and entrainment of endogenous economic cycles. System
Bruckner, E., W. Ebeling, A. Scharnhorst. 1989. Stochastic dynamics Dynam. Rev. 11(3) 177–198.
of instabilities in evolutionary systems. System Dynam. Rev. 5(2) Henderson, R., K. Clark. 1991. Architectural innovation: The reconfig-
176–191. uration of existing technologies and the failure of established
———. 1990. The application of evolution models in scientometrics. firms. Admin. Sci. Quart. 35(1).
Scientometrics 18 (1–2) 21–41. Hirshleifer, D. 1995. The blind leading the blind: Social influence, fads,
Carley, K. 1991. A theory of group stability. Amer. Sociological Rev. and informational cascades. M. Tomassi, K. Ierulli, eds. The New
56 331–354. Economics of Human Behavior. Cambridge University Press,
———. 1995. Computational and mathematical organization theory: Cambridge. 188–215.
Perspective and directions. Comput. Math. Organization Theory Holland, J. 1995. Hidden Order: How Adaptation Builds Complexity.
1(1), 39–56. Addison-Wesley, Reading, MA.
———, W. Wallace. 1995. Editorial. Comput. Math. Organ. Theory Horgan, J. 1995. From complexity to perplexity. Sci. Amer. (June) 104–
1(1) 5–7. 109.
Cartwright, N. 1983. How the Laws of Physics Lie. Clarendon Press, Hoyningen-Huene, P. 1993. Reconstructing Scientific Revolutions:
Oxford. Thomas S. Kuhn’s Philosophy of Science. University of Chicago
Chaisson, E. 1997. Cosmic age controversy is overstated. Sci. 276 (16 Press, Chicago.
May) 1089–1090. Jacobsen, C., R. Bronson. 1995. Computer simulations and empirical
Chin, G., L. Senesac, W. Blass, J. Hillman. 1996. Stabilizing lead-salt testing of sociological theory. Sociological Methods and Res.
diode lasers: Understanding and controlling chaotic frequency 23(4) Sage, Beverly Hills. 479–506.
emissions. Sci. 274 (29 November) 1498–1501. Kaufer, D. S., K. M. Carley. 1993. Communication at a Distance: The
Collins, H. 1985. Changing Order: Replication and Induction in Sci- Influence of Print on Sociocultural Organization and Change.
entific Practice. Sage Publications, Beverly Hills. Lawrence Erlbaum Associates, Hillsdale, NJ.
Costantino, R., R. Desharnais, J. Cusing, D. Dennis. 1997. Chaotic King, G., M. Tomz, J. Wittenberg. 1998. Making the most of statistical
dynamics in an insect population. Sci. 275 (17 January) 389–391. analyses: Improving interpretation and presentation. Working pa-
Crane, D. 1972. Invisible Colleges: Diffusion of Knowledge in Scien- per. Department of Government, Harvard University.
tific Communities. University of Chicago Press, Chicago. Kuhn, T. S. 1970. The Structure of Scientific Revolutions, 2nd ed. Uni-
Cushing, J. 1994. Quantum Mechanics: Historical Contingency and versity of Chicago Press, Chicago.
the Copenhagen Hegemony. University of Chicago Press, Chi-
———. 1990. The road since structure. PSA 2(7) 3–13.
cago.
Lakatos, I., A. Musgrave. 1970. Criticism and the Growth of Knowl-
Cyert, R., J. March. 1963/1992. A Behavioral Theory of the Firm, 2nd
edge. Cambridge University Press, Cambridge.
ed. Blackwell, Cambridge, MA.
Lant, T., S. Mezias. 1992. An organizational learning model of con-
Donovan, A., L. Laudan, R. Laudan, eds. 1988. Scrutinizing Science.
vergence and reorientation. Organ. Sci. 3(1) 47–71.
Kluwer Academic Publishers, Dordrecht, The Netherlands.
Lightman, A., O. Gingerich. 1992. When do anomalies begin? Sci. 255
Ebeling, W. 1991. Stochastic models of competition processes in non-
(7 February) 690–695.
physical systems. Syst. Anal. Model. Simul. 8 (1) 3–17.
Lomi, A., E. Larsen. 1996. Interacting locally and evolving globally:
Eldredge, N., S. Gould. 1972. Schopf, T., ed. Models in Paleobiology.
Freeman, San Francisco. 82–115. A computational approach to the dynamics of organizational pop-
Feyerabend, P. 1975. Against Method. New Left Books, London. ulations. Acad. Management J. 39(4) 1287–1321.
Forrester, J. 1961. Industrial Dynamics. MIT Press, Cambridge MA. Martin, B. 1996. Social construction of an “attack on science.” Soc.
Gersick, C. 1991. Revolutionary change theories: A multilevel explo- Stud. Sci. 26 161–173.
ration of the punctuated equilibrium paradigm. Acad. Manage- Masterman, M. 1970. The nature of a paradigm. I. Lakatos, A.
ment Rev. 16 (1) 10–36. Musgrave, eds. Criticism and the Growth of Knowledge. Cam-
Gleick, J. 1987. Chaos: Making a New Science. Viking, New York. bridge University Press, Cambridge.
Goodman, N. 1968. Languages of Art. Bobbs-Merrill, Indianapolis. Merton, R. 1948. The self-fulfilling prophecy. Social Theory and Social
Gorman, M. 1992. Simulating Science: Heuristics, Mental Models and Structure. Free Press, New York.
Technoscientific Thinking. Indiana University Press, Blooming- Mill, J. S. 1848. Principles of Political Economy. J. W. Parker, London.
ton. Morecroft, J. 1985. Rationality in the analysis of behavioral simulation
Gould, S. J. 1990. Wonderful Life. W. W. Norton, New York. models. Management Sci. 31(7) 900–916.
Gross, P., N. Levitt. 1994. Higher Superstition: The Academic Left and Myrdal, G. 1944. An American Dilemma. Harper, New York.
Its Quarrels with Science. Johns Hopkins University Press, Bal- Nelson, R., S. Winter. 1982. An Evolutionary Theory of Economic
timore, MD. Change. Belknap, Harvard University, Cambridge, MA.

340 ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999

JOHN D. STERMAN AND JASON WITTENBERG Path Dependence

Nadis, S. 1996. Oil drop trap is set for lone quark. Sci. 271 (8 March) Sterman, J. D. 1985. The growth of knowledge: Testing a theory of
1361–1362. scientific revolutions with a formal model. Tech. Forecasting and
Orlikowski, W. J. 1996. Improvising organizational transformation Soc. Change 28 (2) 93–122.
over time: A situated change perspective. Inform. Systems Res. ———. 1988. Deterministic chaos in models of human behavior:
7(1) 63–92. Methodological issues and experimental results. System Dynam.
Pickering, A. 1984. Constructing Quarks: A Sociological History of Rev. 4 148–178.
Particle Physics. University of Chicago Press, Chicago. ———. 1989. Deterministic chaos in an experimental economic sys-
———, ed. 1992. Science as Practice and Culture. University of Chi- tem. J. Econom. Behavior and Organ. 12 1–28.
cago Press, Chicago. ———. 1992. Response to “On the very idea of a system dynamics
Pinch, T. 1986. Confronting Nature: The Sociology of Solar-Neutrino model of Kuhnian science.” System Dynam. Rev. 8 (1) 35–42.
Detection. D. Reidel, Dordrecht, Holland. Stewart, J. 1986. Drifting continents and colliding interests: A quan-
Radzicki, M. 1992. Reflections on “On the very idea of a system dy- titative application of the interests perspective. Soc. Stud. Sci. 16
namics model of Kuhnian science.” System Dynam. Rev. 8 (1) 261–279.
49–53. Sturis, J., K. S. Polonsky, E. Mosekilde, E. Van Cauter. 1991. Com-
———, J. D. Sterman. 1994. Evolutionary economics and system dy- puter model for mechanisms underlying ultradian oscillations of
namics. R. England. ed. Evolutionary Concepts in Contemporary insulin and glucose. Amer. J. Physiology 260 (Endocrinol. Metab.
Economics. University of Michigan Press, Ann Arbor. 61–89. 23), E801–E809.
Rappa, M., K. Debackere. 1993. Youth and scientific innovation: The Thagard, P. 1968. Computational Philosophy of Science. MIT Press,
role of young scientists in the development of a new field. Mi- Cambridge, MA.
nerva 31 (1) 1–20. Turner, R. S. 1987. Paradigms and productivity: The case of physio-
Richardson, G. 1991. Feedback Thought in Social Science and Systems logical optics, 1840–94. Soc. Stud. Sci. 17 35–68.
Theory. University of Pennsylvania Press, Philadelphia. Turbayne, C. M. 1970. The Myth of Metaphor, rev. ed. University of
———. 1996. Problems for the future of system dynamics. System
South Carolina Press, Columbia, SC.
Dynam. Rev. 12(2) 141–157.
Tushman, M., P. Anderson. 1986. Technological discontinuities and
———, A. Pugh. 1981. Introduction to System Dynamics Modeling
organizational environments. Admin. Sci. Quart. 31 439–465.
with DYNAMO. Productivity Press, Portland, OR.
———, E. Romanelli. 1985. Organizational evolution: A metamor-
Rogers, E. M. 1995. Diffusion of Innovations. The Free Press, New
phosis model of convergence and reorientation. Res. Organiza-
York.
tional Behavior 7 171–222.
Roth, P., R. Barrett. 1990. Deconstructing quarks. Soc. Stud. Sci. 20
Valente, T. W. 1995. Network Models of the Diffusion of Innovations.
579–632.
Sastry, M. A. 1997. Problems and paradoxes in a model of punctuated Hampton Press, Cresskill, N.J.
organizational change. Admin. Sci. Quart. 42(2) 237–275. Van de Ven, A., M. Poole. 1995. Explaining development and change
———. 1998. Archetypal self-reinforcing structures in organizations: in organizations. Acad. Management Rev. 20(3) 510–540.
A system dynamics perspective of cognitive, social, and institu- Waldrop, M. 1992. Complexity: The Emerging Science at the Edge of
tional processes. Proceedings of the 1998 International System Order and Chaos. Simon & Schuster, New York.
Dynamics Conference, Quebec City, Quebec. Weick, K. E. 1979. The Social Psychology of Organizing, 2nd ed. Ran-
———, C. Coen. 1998. Tracing the hand of the past: Towards an dom House, New York.
integrated theory of how history matters in organizations. Work- Weisbuch, G. 1991. Complex Systems Dynamics: An Introduction to
ing paper, University of Michigan Business School. Automata Networks. Addison Wesley, Reading, MA.
Schroeder, M. 1991. Fractals, Chaos, Power Laws. W. H. Freeman, Wittenberg, J. 1992. On the very idea of a system dynamics model of
New York. Kuhnian science. System Dynam. Rev. 8 (1) 21–33.
Accepted by Kathleen M. Carley; received June 27, 1997. This paper has been with the authors for one revision.

ORGANIZATION SCIENCE /Vol. 10, No. 3, May–June 1999 341