integrating modernist and postmodernist perspectives on

INTEGRATING MODERNIST ANDPOSTMODERNIST PERSPECTIVES ON

ORGANIZATIONS: A COMPLEXITYSCIENCE BRIDGE

MAX BOISOTESADE

BILL MCKELVEYUniversity of California, Los Angeles

Competition between modernism and postmodernism has not been fruitful, and man-agement researchers are divided in their preference, thereby undermining the legit-imacy of truth claims in the field as a whole. Drawing on Ashby’s Law of RequisiteVariety, on complexity science, and in particular on power-law-distributed phenom-ena, we show how the order-seeking regime of the modernists and the richness-seeking regime of the postmodernists draw on different ontological assumptions thatcan be integrated within a single overarching framework.

The study of social systems such as organiza-tions has long been caught between two con-flicting bases of legitimacy. On the one hand,we have positivism—a set of procedures for cre-ating valid knowledge expressing a modernistoutlook that originated in the eighteenth centuryEnlightenment project. Positivism presumes areal, relatively stable, and objectively givenworld, populated by phenomena that can be ra-tionally known and rationally analyzed by inde-pendent observers. Such phenomena can be de-composed into observation protocols resting onsense data and predictively related to eachother through stable laws integrated via a math-ematical syntax (Benacerraf & Putnam, 1964;Lakatos, 1976). Positivism promotes the modern-ist agenda: the understanding, manipulation,and control of predominantly physical phenom-ena for beneficial social ends. In contemporarysocial sciences, neoclassical economics re-mains positivism’s foremost exemplar (Colan-der, 2006; Friedman, 1953; Lawson, 1997; Mi-rowski, 1989).

On the other hand, we have postmodern-ism—a movement that emerged in the late 1960sto challenge the basic tenets of modernism andits epistemological ally, positivism. Whereas inmodernism the focus is on a phenomenal worlddirectly and unproblematically observed anddescribed by a disinterested actor who remainsexternal to what is being observed, the postmod-

ernist strategy problematizes the relationship ofactors to observed phenomena by having lan-guage mediate it. Thus, instead of a single di-rect relationship between an external world, W,and an observer, O, we now have two relation-ships: (1) between an external world, W, and adescriptive language, L, and (2) between L andan observer, O. Language is a human resourcethat places the relationship between W and O ina social context where divergent interests(Habermas, 1972) and social power (Foucault,1969) come into play. These shape language andlinguistic usage and, by implication, the regionsof the phenomenal world to which they giveaccess. Language, the postmodernists argue, isnot a neutral observation tool. It shapes obser-vations in ways that reflect the ontological as-sumptions of a particular community of observ-ers (Berger & Luckmann, 1966; Kuhn, 1962).Postmodernism, initially a literary movement,emerged in response to the linguistic turn inphilosophy. Its claim that “everything is text”(Derrida, 1978) highlights the mediating role oflanguage linking observers to their worlds (Lyo-tard, 1984; Rorty, 1980).

Organization theory has been pulled in oppo-site directions by modernist and postmodernistontologies. Organizational scholars, thus, arecaught between two conflicting bases of legiti-macy, with little overall consensus on what con-stitutes valid truth claims. Practitioners have,

! Academy of Management Review2010, Vol. 35, No. 3, 415–433.

415Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyrightholder’s express written permission. Users may print, download, or email articles for individual use only.

then, little reason to act on the research findingsof academics in open disagreement about theirdiscipline’s foundations. Absent any faith inwhat the positivists are measuring—as they jug-gle with sample sizes, normal distributions,means, variance, probabilities, and statisticalsignificance—managers will settle for grippingcorporate yarns that gain traction from vividand compelling narratives readily rememberedand retold. A good story loads on the dependentvariable with gay abandon, leveraging “sam-ples of one” (March, Sproull, & Tamuz, 1991) intouniversal managerial truths. It constitutes ameme (Blackmore, 1999; Dawkins, 1976) thatpropagates owing to its plausibility, its internalcoherence, and its alignment with the experi-ence of its intended audience, rather than anyobjective probability that it might be true.

If, following the Chicago School Pragmatists(Dewey, 1925; James, 1907), we take knowledge toconsist of actionable beliefs, we can view mod-ernism as attempting to substantiate these be-liefs according to rationally derived principlesand rules. Postmodernism challenges this strat-egy as suppressing voices that fail to fit therationalist straitjacket (Calas & Smircich, 1999).While it stabilizes and delineates our differentidentities, modernism also limits our inherentcomplexity and potentiality (Deleuze & Guattari,1984). Order and organization are thus transientachievements based on an infinite rather than alimited set of possibilities, the products of whatDeleuze and Guattari call “chaosmosis” (Carter& Jackson, 2004). The proliferation of uncon-strained beliefs, however, makes them vulnera-ble to biases. Which ones, then, form a legiti-mate basis for action? Does the need to actsuggest that we should accept modernist con-straints while recognizing them to be contin-gent? Or should we abandon these as essen-tially arbitrary and, following Feyerabend (1975:296), argue that “anything goes?” If so, can or-ganizational research still call itself a science-based “discipline”?

We offer a third alternative that draws on sev-eral well-known complexity principles to inte-grate the ordered world of modernists and themore “chaotic” world of postmodernists. Weposit that the conjunction of adaptive tension—the gap between the variety internally availableto a system and that which confronts it exter-nally (McKelvey, 2001, 2008)—connectivity, andinterdependency in social phenomena reflects

these principles and challenges the dominantassumption that social events are independentof each other and identically distributed (i.i.d.)so as to yield a normal distribution. Such a“Gaussian” default assumption underpins anatomistic ontology, one that takes the world asconstituted by a collection of objects. Manyevents connected under tension, however, areoften distributed according to a power law, asillustrated in Figure 1, which shows two Paretodistributions on the left and their equivalentpower-law distributions on the right. A power-law distribution is a Pareto distribution de-picted on a log-log scale. Other (less extreme)skew distributions, reflecting the different waysthat phenomena interact, are also possible. Herewe focus on rank/frequency power laws.

In the upper left of Figure 2, we show a styl-ized representation of the myriad small out-comes—such as the approximately 16,000 Cali-fornian quakes that go unnoticed each year, orthe 17 million ma & pa stores that didn’t becomeWalmarts—that econometricians usually treatas i.i.d. and summarize with a normal distribu-tion.1 Toward the lower right of the figure, incontrast, we see the increasingly high-ranked,very rare, extreme outcomes that defy predic-tion—that is, earthquakes, floods, bankruptcies,stock market crashes, giant firms (Microsoft,Walmart, etc.), and giant cities.

The complex causal connections that, undertension, generate power-law distributions donot allow us to distinguish ex ante what is us-able information from what is noise. Any one ofthe tiny events located in the upper-left region ofFigure 2 could initiate a causal chain reaction,generating an extreme outcome located in thelower-right region of the figure. The Gaussiandefault assumption is therefore easy to make.Figure 2, however, underpins a connectionist on-tology that takes the world’s fundamental con-stituents to be relationships. In contrast to nor-mal distributions, power-law distributions havelong tails, potentially infinite variance, unstablemeans, and unstable confidence intervals (An-driani & McKelvey, 2007). If Gaussian thinkingtakes extreme events to be outliers—too differ-ent from other events in the sample to be enter-

1 “Robustness” techniques (Greene, 2002) translate skewdistributions into normal ones—that is, by making the x axisa log scale so as to produce a log-normal distribution.

416 JulyAcademy of Management Review

tained as probable and, thus, to form part of thedistribution being studied—power laws incor-porate outliers as a significant part of the dis-

tribution and therefore meriting attention. Evenif they cannot make them probable—unstablemeans and potentially infinite variances pre-vent it—power laws signify the existence ofscale-free phenomena worthy of our consider-ation. Their scalability—that is, the causal dy-namics stemming from multiplicative subunitinteractions to produce similar outcomes at mul-tiple hierarchical levels (e.g., network organiza-tions such as the Internet)—renders them plau-sible.

Numerous complexity researchers (Andriani &McKelvey, 2007, 2009; Newman, 2005; West &Deering, 1995) have found power-law distribu-tions to be ubiquitous in social no less than innatural systems. They have captured social phe-nomena ranging from the large number of sta-tistically similar entities located in one tail ofthe distribution to the N ! 1 extreme outcomes

FIGURE 1From Pareto to Power-Law Distributions: Two Examples—Pareto on Left, Power Law on Righta

a Reproduced from Glaser (2009).

FIGURE 2Stylized Power-Law Distribution

Log of event size

Logof

eventfrequency

Gaussianworld

Mean

Paretian world

Power law negative slope

2010 417Boisot and McKelvey

best studied by hermeneutics methods in theother. We argue that modernists and postmod-ernists have each got hold of one tail of a dis-tribution in which extreme outcomes are notrandom outliers as interpreted by Gaussiansbut, rather, the product of tension and connec-tivity effects. These shift a distribution from ani.i.d.-based normal distribution to a power-lawdistribution. While other distributions are, ofcourse, possible, omitting these from the discus-sion does not affect the thrust of our argument.Understanding what drives the distribution ofsocial phenomena at different levels of organi-zation allows us to integrate the seemingly op-posed modernist and postmodernist epistemol-ogies into a unitary representation. Locatingthem both along a single causal continuum en-hances the epistemic legitimacy of each in theeyes of the other.

The structure of our article is as follows. First,we briefly provide working definitions of themodernist and postmodernist positions. To showwhere they differ, we present these as idealiza-tions, hoping that readers will see beyond theresulting simplifications. Following this, wedraw on Ashby’s concept of requisite variety tooffer a complexity perspective on the challengesof adaptation. We argue that such a perspectiveilluminates the modernist/postmodernist de-bate. We then apply our analysis to organiza-tions and explore its implications for organiza-tional research. We end with a conclusion.

MODERNISM VERSUS POSTMODERNISM

Defining Modernism

Modern science is one of the fruits of the En-lightenment’s modernist project. Insofar as thesocial sciences promote the understanding anduse of science to improve modern society, theyalso pursue a modernist agenda (Israel, 2001).While positing the epistemological and moralunity of mankind (Hollinger, 1994), the modernistproject “assumes that human beings are auton-omous subjects, whose interests and desires aretransparent to themselves and independentfrom the interests and desires of others” (Calas& Smircich, 1999: 653). Bacon and Descartes areconsidered to be the main proponents of this“atomistic” ontology (Hollinger, 1994).

Modernism sought knowledge outside reli-gious revelation; Baconian science argued for

the empirical rather than the faith-based justifi-cation of truth claims. Truth arose from a corre-spondence between a claim and empirically ob-served facts, rather than divinely sanctionedrevelations transmitted through sacred—and,hence, unmodifiable—texts. This required therepeatability or replicability of facts and therejection of one-shot events such as miracles.Objectivity, however, could only be fullyachieved by an independent and decontexual-ized observer endowed with a god’s eye view—a“view from nowhere” (Shapin & Schaffer, 1985).

If modernism constituted a world view, therise of positivism at the end of the nineteenthcentury provided it with a methodology. ErnstMach’s rebellion against Hegelian idealismgave rise in 1907 to the Vienna Circle, a group ofphysicists and mathematicians whose dreamwas the attainment of absolute verified truth(“verificationism”) based on a rigid correspon-dence (“correspondence theory”) between opera-tional measures and theory terms (Suppe, 1977).Modernism and its methodological hand-maiden, positivism, have long underpinned theepistemic legitimacy of the natural sciences. Be-ing essentially concerned with what Reichen-bach (1938) called “the context of justification,”however—justification being one of Plato’s pre-requisites for genuine knowledge—modernismand positivism showed little interest in whatReichenbach called “the context of discovery.”If, for Bacon, genuine knowledge yielded predic-tion and control, and hence a basis for action,these forms of justification would separate sci-ence from superstition, alchemy, religion, andfaith-based revealed truth. Into this world ofapodictic certainties, Reichenbach introducedthe idea that probabilistic thinking offered amore realistic basis for justification. WithBrown’s “Brownian Motion” in 1827 (Ford, 1992),Boltzmann’s statistical mechanics of 1877 (Boltz-mann, 1887), Gibbs’s statistical actuarial tablesfor the insurance industry in 1902 (Gibbs, 1902),and Fisher’s statistics of 1916 (Fisher, 1918), ashift occurred, endorsed by Reichenbach (1938),from exact to probabilistic representations.

In a noisy world the structures underpinningthe replicability of independent events are cap-tured statistically by the mean. In the case ofnormal distributions, the variance could oftenconveniently be treated as mere noise—some-thing to be got rid of rather than explored. Overtime, the normality of a distribution became the


default assumption—the taken for granted sig-nature of a universal reality that yielded stable,manipulable objects. Gaussian statistics, thestatistics of the normal distribution now widelyapplied in the social sciences (e.g., Greene,2002), delivers stable means, finite variances,and independent data points (Andriani & Mc-Kelvey, 2007; Taleb, 2007). The social sciences,epitomized by neoclassical economics, thus cre-ated for themselves the stable and (mostly) com-putationally tractable social objects that hadbeen the focus of Newtonian physics (Colander,2006; Friedman, 1953; Mirowski, 1989), while atthe same time eschewing the more complex,messy interactive and dynamic social processescharacterizing human social behavior.

Gaussian-inspired statistical truths artifi-cially structure the world so as to achieve sig-nificant reductions in complexity, a demultipli-cation of explanatory entities, and a consequentreduction in the required degrees of freedom—that is, the number, n, of observed events thatare free to vary minus the number of necessaryrelations, r, obtainable from these observations(Walker, 1940). In line with Occam’s razor—theexplanans should always be more compact thanthe explanandum—they achieve compressibilityand parsimony (Hempel, 1965). In modern cos-mology the search for a theory of everythingillustrates this concern with compressibility andparsimony (Guth, 1997; Weinberg, 1992). In addi-tion to its parsimony, a theory’s worth is alsobased on its predictive power. Predictability assuch, however, does not always require under-standing (Bridgeman, 1936). As Feynman fa-mously pointed out, despite its remarkablepredictive achievements, “No one really under-stands quantum mechanics” (1967: 129).

The modernist approach has not gone unchal-lenged. Unlike the physical sciences, the socialsciences have to deal with the fact that althoughthe people they study are subject to physicalforces, they act primarily on the basis of repre-sentations and interpretations of the world thatmake meaning central to explanations of theirbehavior. The inability of the physical sciencesto deal with the vexing question of meaning ledto the rejection of the modernists’ stance as awhole by many social scientists. After all, what,exactly, constitutes “replicability” when dealingwith a complex social or organizational phe-nomenon? In what respect might two complexsocial outcomes be sufficiently similar to justify

a claim of replicability? And how robust is theconcept of intersubjective objectivity—modern-ism’s substitute for the god’s eye view—giventhe social distribution of power, influence, andbias (Foucault, 1969; Shapin & Schaffer, 1985)? Ifquestions like these suggest an unbridgeablegulf between the natural and the social sci-ences, sociologists of science go further, point-ing out that in the natural sciences no less thanin the social sciences, problems of interpreta-tion, meaning, status, and power effectively con-taminate all claims to objectivity (Callon, 1986;Golinski, 1998; Latour, 1988). No student researchassistant in any physics or biology laboratorylong remains unaware of what results the pro-fessor wants to see!

Defining Postmodernism

Alvesson and Deetz see modernism as

the instrumentalization of people and naturethrough the use of scientific-technical knowledge(modeled after positivism and other “rational”ways of developing safe, robust knowledge) toaccomplish predictable results measured by pro-ductivity and technical problem-solving leadingto the “good” economic and social life, primarilydefined by accumulation of wealth by productioninvestors and consumption by consumers (1996:194).

Postmodernists hold that such scientificknowledge, shaped by local historical and cul-tural contexts, represents one story among many(Calas & Smircich, 1999)—a social constructionserving the ideological agenda of powerfulelites (Koertge, 1998). The postmodern perspec-tive challenges the Enlightenment project by in-troducing a radical subjectivity and the exerciseof power as irreducible constraints on our ac-cess to an objective world (Foucault, 1975). Topostmodernists, the world—especially the so-cial world—is not objectively given. It is kalei-doscopic and unstable, and its constituent com-ponents are elusive. The stability that weimpute to it and from which we derive laws andtheories is partly shaped by our interaction withother observers. Postmodernists therefore dis-trust the modernist’s summary Gaussian de-scriptions and the confident narratives theseproduce (Lyotard, 1984).

Postmodernist epistemology is profligaterather than parsimonious. By entertaining mul-tiple representations of phenomena (“voices”) as


equally valid alternatives, postmodernists shunwhat they see as the exclusions and repressionsunderpinning the modernists’ claims to singularobjective representations. Postmodernists seek“infinite conversations” undistorted by powerconsiderations (Derrida, 1978; Foucault, 1975;Rorty, 1989). Their emphasis on “playfulness” isdesigned to counter a desire to control every-thing and the despair at not being able to do so.Life in all its richness and messiness is moreimportant to postmodernists than the impover-ished conceptions of it found in psychology, eco-nomics, and other positivist-leaning social sci-ences. Expressed as a statistical strategy,postmodernists invite us to focus on the richpromises latently present in the variance ratherthan on an impoverished mean. They believethat either the social sciences accommodate thetheses of postmodernity or they become irrele-vant.

Postmodernism is a broad church that accom-modates a multiplicity of views—not alwaysharmoniously (Jenks, 1992). It challenges mod-ernism’s unitary vision of science and society,deconstructing the modernist object of study, re-vealing the fragility of the assumptions under-pinning its stability, and greeting modernism’smetanarratives with incredulity. Lyotard (1984)would replace these with petit recits—modestnarratives—which, like Merton’s theories of themiddle range (Merton, 1949), would be of limitedspatiotemporal reach. Yet while these mightpromote awareness and reflexivity, they rendertheorizing elusive (Calas & Smircich, 1999)since, trapped as they are in local Wittgenstein-ian language games, there is no basis for choos-ing between competing representations: mean-ing now becomes undecidable. In fact,modernism overreaches precisely when its all-encompassing metanarratives—Marxist, Parso-nian, and so forth—encounter Lyotard’s localpetit recits. Being embedded and contextual, thelatter, far from scaling up into metanarratives,constantly challenge the former’s relevance andvalidity.

In contrast to the natural sciences, postmod-ernism massively increases the variety of phe-nomena that social scientists are required todeal with. These can be viewed as manifesta-tions of complexity at work; they point to higherlevels of interaction and interdependenceamong phenomena and to the irreversible ef-fects of time and path dependency. In effect,

postmodernism is a theory of social complexity(Cilliers, 1998). Assumptions of independenceamong phenomena are here challenged by theoperation of dense feedback loops—both posi-tive and negative—generated as much by howintentional agents construe events (Dennett,1989) as by physical causal links among them.Given complex interdependencies, focusing ex-clusively on the mean of a distribution becomesdysfunctional and misleading since its variancenow contains much of the relevant information;it is more than just noise.

Given complex interdependencies withindensely connected causal networks, how do weproceed? The connectionist ontology implicitlyunderpinning postmodernism massively in-creases the number of plausible patterns needingcausal analysis and interpretation. For postmod-ernists, however, computational conveniencedoes not constitute an epistemic justification fora reductionist stance, so Occam’s razor is oflittle use; the complexity must be absorbed andlived with rather than reduced (Boisot & Child,1999). Postmodernists are interested in unpre-dictable and emergent phenomena rather thanpredictable regularities—in process rather thanstructure. Their methodological preference is forqualitative case-based research. In any trade-off between understanding and prediction, un-derstanding should take precedence.

Postmodernism itself, however—the pursuit of“infinite conversations”—has also come underfire. The relativism resulting when one theory isdeemed as good as another and equal airtime isgiven to all (Hollis, 1982), or when paradigmscannot be reconciled (Kuhn, 1962), makes it im-possible to compare, evaluate, and select fromcompeting alternatives. The primacy postmod-ernism accords to the chaotic nuances gener-ated by the swaying of individual “trees” at theexpense of patterns discernible in the “forest”effectively paralyzes theory choice, thus under-mining justification and practitioner relevance.Yet without a timely and “justifiable” consen-sus, productive social action becomes impossi-ble. This poses a challenge to management in-quiry interested in both valid truth claims andactionable outcomes. How can it contribute topractical action if (1) truth claims cannot be dis-entangled from the “situated” interests that giverise to them; (2) truth claims are framed in in-commensurable languages; (3) competing alter-natives are incommensurable across observers;


and (4) any convergence achieved across alter-native truth claims reflects the influence of sta-tus, power, repression, and coercion?

In what follows we argue that modernism andpostmodernism are not so much competing al-ternatives as alternative moments in a singledynamic process of human adaptation to bothnatural and social phenomena.

ADAPTING TO COMPLEXITY

In the human case, adaptation is about how torespond intelligently to the threats and opportu-nities embedded in the variety of natural andsocial phenomena confronting us as a species.To repeat, variety is often the surface manifes-tation of complexity at work. Since what distin-guishes modernists from postmodernists is howthey approach this variety, we take it as thestarting point for our discussion. In biology theissue is often framed in evolutionary terms—alternative framings are possible (Dooley & Vande Ven, 1999). We draw on the biological ap-proach and apply it to the human and organiza-tional realms.

The Law of Requisite Variety

Ashby’s Law of Requisite Variety states that“ONLY VARIETY CAN DESTROY VARIETY” (1956: 207). Thelaw holds that for a biological or social entity tobe adaptive, the variety of its internal ordermust match the variety imposed by environmen-tal constraints. We treat variety as a proxy forcomplexity (McKelvey & Boisot, 2009). Gell-Mann(1994) holds that emergent complexity is a func-tion of the variety present in phenomena. Wher-ever the variety externally imposed on an adap-tive biological or social system exceeds thatinternal to the system, there emerges an adap-tive tension (McKelvey 2001, 2008) within it thatfills the gap between what the environment re-quires of the system to ensure its integrity orsurvival and what it can actually deliver at agiven moment.

Although Ashby’s law tells us nothing aboutthe nature of the external complexity a systemmust respond to, the fact that systems such asourselves adapt and survive suggests thatwithin a certain range such complexity must bemanageable. Not all of it will be relevant to thesystem’s survival. Gell-Mann (1994) distin-guishes between a “crude complexity” indistin-

guishable from randomness residing in phe-nomena and an “effective complexity” residingin the regularities underpinning their structure.By focusing on effective complexity, a systemcan respond in selective and discriminatingways to the massive variety it confronts (McK-elvey & Boisot, 2009).

The Ashby Space

We explore the difference between crude andeffective complexity in a diagram (Figure 3) thatwe label the Ashby Space (Boisot & McKelvey,2007). The vertical axis measures the variety ofexternal stimuli that register with an agent; thehorizontal axis measures the variety of re-sponses generated by that agent. The diagonalindicates where the variety of responsesmatches that of incoming stimuli and is there-fore adaptive. Above the diagonal, the variety ofthe responses fails to match that of incomingstimuli; below it, the variety of responses is ex-cessive relative to what is adaptive and wastesenergetic resources. We now partition the verti-cal axis of the Ashby Space into different re-gimes: chaotic, complex, and ordered. We couldalso partition the horizontal axis, but we do notneed to do so. In the chaotic regime incomingstimuli exhibit no obviously discernible regular-ities; in the complex regime they exhibit some,even if these still have to be teased out; in theordered regime one can subordinate all the va-riety encountered in incoming stimuli to someordering principle—for example, algorithmic

FIGURE 3The Ashby Space


compression—as when, for example, the se-quence a,b,a,b,a,b,a,b,a,b,a,b,a,b,a,b,a,b,a,b canbe reduced to 10(a,b).

Using this diagram, we offer a complexity-driven interpretation of Ashby’s law. The verti-cal arrow going down from A to C describes acognitive process of variety reduction that aimsto filter out crude complexity and focus on effec-tive complexity. This, according to modernists,requires interpretation and selection—that is,algorithmic compression. If successful, it re-duces the variety calling for responses andgradually moves them into the ordered regime.A postmodernist distrusts such moves, arguingthat what constitutes effective complexity lies inthe eye of the agent and that where adaptiveresponses need to be collective, the modernist’sreductionist strategy is often coercive ratherthan cognitive in nature.

Although stimuli can appear at any pointalong the vertical axis, the horizontal arrow pro-ceeding from point A to point B offers the clear-est illustration of a postmodernist implementa-tion of Ashby’s law. Since postmodernists herefind themselves in the chaotic regime, their de-fault assumption is that there will be no effec-tive complexity to be teased out—that is, norobust underlying structure. All is crude com-plexity. Lacking any agreed upon basis for in-terpreting the stimuli—that is, for reducing theirvariety by traveling down the space prior toformulating a response—postmodernists allowthe variety of responses to expand until itmatches that of incoming stimuli. They, thus, arewilling to remain in the chaotic regime until“Nature shows her hand.” In sum, in contrast tomodernists, whatever the regime they find them-selves in, the postmodernists’ default preferenceis to move horizontally across rather than verti-cally down the Ashby Space. Yet althoughAshby himself does not distinguish betweencrude and effective complexity—between vari-ety that should be treated as noise by the agentand variety that has relevance for it—in at-tempting to accommodate all variety and refus-ing to be selective, the postmodernists’ responseis likely to be costly in terms of energy expendedand may well overshoot point B on the diagonalwhere adaptation is achieved.

While there will be many situations in whichagents confront regimes that are either whollychaotic or wholly ordered, many managementand organizational research challenges arise in

the complexity regime, where both effective andcrude complexity operate. Intelligent agents inthis regime initially move down from point Atoward point C but are led to turn right towardpoint D when they encounter irreducible uncer-tainty. Organizational researchers entering thisregion of the Ashby Space need to toleratehigher levels of epistemic variety than mod-ernists but must then be willing to select fromit. In effect, they must become evolutionaryepistemologists, progressing toward a higherprobability of truth by slowly weeding out in-ferior theories (Hahlweg & Hooker, 1989; Mc-Kelvey, 1999; Radnitzky & Bartley, 1987). Theymust, however, defer to postmodernist sensi-bilities by making their selection more forgiv-ing than modernists would wish, but then de-vise effective procedures to home in on themost promising interpretive schemata. Wherea collective interpretation is possible, some ofthese can gradually be moved into the orderedregime.

Phase Transitions and Scalability

Our discussion so far has centered on the re-spective responses of modernists and postmod-ernists to stimuli appearing high on the verticalaxis of Figure 3, where the world will be expe-rienced as chaotic. But what determines whereon the vertical scale stimuli will actually ap-pear? Complexity science studies elements ininterdependency—and increasingly in livingsystems (Gell-Mann, 2002). Absent such connec-tivity, one has atomistic aggregations to whichi.i.d. assumptions apply. Complexity increaseswith the number of interacting elements and thedensity and nonlinearity of the interdependentoutcomes (Holland, 1988, 2002). Beyond certainthresholds, complexity can lead to phase transi-tions toward either emergent order—that is, dis-sipative structures that maintain themselves inexistence by continuously importing free energyfrom their environment and exporting bound en-ergy back into it (Nicolis & Prigogine, 1989)—orgreater chaos (Kauffman, 1993; Kaye, 1993).

Some scholars study interdependenciesamong heterogeneous agents—these couldrange from nucleotides to individual human be-ings and to organized collectivities of these—operating at what was early on called the “edgeof chaos” but is now seen as a region of emer-gent complexity between the “edge of order”


and the edge of chaos. For Nicolis and Prigogine(1989), the edge of order is the “1st critical val-ue”—a level of energy sufficient to cause phasetransitions in many physical phenomena (aswhen the level of heat in a teapot causes arolling boil). This region of complexity, varyingin size (and separating ordered from chaotic re-gimes), is what Kauffman (1993) labels the melt-ing zone. When, through the amplification offeedback, connectivity-enabled interdependen-cies reach a specific intensity, they can triggerphase transitions from one of the regimes ofFigure 3 to another. In some systems new orderemerges from such phase transitions when ex-isting structures come to be dominated by un-stable modes that become order parametersfor a new regime; Haken (1983) describes theseas becoming enslaved. His “slaving principle”constitutes a disruption of equilibrium (sym-metry breaking; Mainzer, 2007/2004) that re-flects choices made by agents within the sys-tem.

In social systems such choices may reflect theexercise of power by those in a position to re-duce critical uncertainties within the system(Crozier, 1964). Often, the connections them-selves are established and amplified when thesystem is put under adaptive tension—is forcedacross the edge of order into the melting zone.Here we see tension, such as that between sup-ply and demand, which causes entrepreneurs tostart up possibly innovative new enterpris-es—in effect, phase transitions out of the statusquo. Social systems put under tension, throughrecession, poverty, migration, ethnic conflict,and so forth, can also be torn apart by forcesstarting with tiny initiating events. Given theseconditions, the initiating event may add pres-sure to neighboring interdependencies so as totake the system up to, if not over, the edge ofchaos. An analogy is with a fishing net lyingloosely in a pile. Cut one of its cords and nothinghappens. Now stretch it taut and cut one of itscords; the cut of one link transmits tension toneighboring links, propagating a tear across thenet.

Yet interdependencies are raw materials forany kind of organization. Bak (1996) argued that,to survive, a system must be able to stay withinthe melting zone, in a state that precariouslymaintains its effective complexity near the edgeof chaos, which he called “self-organized criti-cality” (SOC). Bak illustrated SOC with a

sandpile. Keep adding grains of sand to a sand-pile, thus increasing the adaptive tension it issubjected to, and at some critical point the slopebecomes steep enough that tiny to large ava-lanches occur that reduce the steepness of theslope and restore stability. At this point thecausal influences generated by the tension be-come scalable and propagate throughout thesandpile in unpredictable ways, influencinggrains far removed from each other. The size-frequency distribution of avalanches in thesandpile follows a power law that Bak claimedto be universal. They also exhibit a fractal struc-ture—they are self-similar across a range ofscales—meaning that their appearance and theunderlying causal dynamics are essentially thesame across multiple scales or hierarchical lev-els (Mandelbrot, 1982).

Andriani and McKelvey (2007, 2009) identifiedsuch connectivity-based outcomes extendingacross thirty-two magnitudes of physical phe-nomena, twenty-seven magnitudes of biologicalphenomena, and eleven magnitudes of socialphenomena. They also showed how pervasivepower laws are in physical, biological, social,and organizational phenomena—listing over100 of the latter. Barabasi (2002) saw power laws,scalability, and fractal structures operating insocial networks. Here, connections are often es-tablished through communication, and their ef-fects, both positive and negative, are amplifiedthrough the operation of feedback loops. A pow-er-law distribution includes many social “lon-ers” at one network extreme and a single highlyconnected “star” at the other. As Brunk ob-served, “Instead of the bulk of the data beingproduced by one process and the ‘outliers’ byanother, all events—both minuscule and thehistorically monumental—are produced by thesame process in an SOC environment” (2002: 36).

Two Ontologies: Friends or Foes?

Our analysis suggests that we do not have tochoose between connectionist and atomistic on-tologies. The high variety and low variety theyengender, pursued respectively by postmodern-ists and modernists, are but transitory momentsin a broader process in which each has its place.Connectionism and atomism are lenses that webring to bear on events for particular purposes.Complexity theory—about the dynamics of con-nectivity and interdependence—provides us


with an overarching conceptual framework thataccommodates both. Within it, Gell-Mann’s con-cept of effective complexity is well placed tofruitfully integrate modernist and postmodernistinsights.

Atomistic and connectionist ontologies arethus complementary and contingent, rather thanalternatives. Under certain circumstances, al-though they do not have to, phenomena canconnect. When this yields extreme events, weaccount for them through a detailed retracing ofcausal connections presented as a historicalnarrative or a case study: the fall of Constanti-nople, the Cuban Missile Crisis, etc. Yet sincethese causal connections and the resultingcausal patterns are improbable, they do not lendthemselves to systematic replication and exper-imentation. The lessons of history are thusrarely unequivocal. The causal components ofan extreme event, taken individually, may lendthemselves to systematic replication and exper-imentation, but the predictions yielded by suchan atomistic approach remain strictly limited inscope, offering little purchase on the more richlyconnected patterns typically covered by histori-cal accounts. All attempts at grand narrativesignore this point (Lyotard, 1984). We know, forexample, that beyond a certain threshold, socialtensions and instability, for good or evil, canthrow up charismatic leaders, but we cannotpredict when or how. Anticipation rather thanprediction is, then, the best that we can hope for.

BRIDGING TO ORGANIZATIONS

To summarize, modernism advances knowl-edge when phenomena are independent of eachother or can be made so via controlled experi-ments. It targets the ordered regime in theAshby Space, one in which phenomena can bepredicted and responded to efficiently. What wehave called an atomistic ontology takes the in-dependence of phenomena as its default as-sumption, allowing them to be described by anormal distribution. In the case of socially pro-duced knowledge, postmodernism takes this as-sumption to be an unwarranted simplification ofrealities that include coercive social processes.It emphasizes the idea that new order creationdraws on the arbitrary—and sometimes illegiti-mate—use of power (Foucault, 1975). Postmod-ernism implicitly builds on a connectionist on-tology and power-law dynamics to argue that

there exists no socially legitimate basis for mov-ing down the Ashby Space. Yet modernists andpostmodernists are like blind people who haveeach seized different parts of the complexityelephant, little realizing that their ontologiescomplement rather than compete with eachother. The challenge is to understand wheneach applies.

Existing Discourse

Modernist discourses seek to maintain a highlevel of generality that becomes increasinglyunsustainable as they travel down the power-law slope of Figure 2, toward ever-smaller sam-ples of ever-larger and more extreme outcomes.In so doing, however, they often impose over-simplified interpretations (i.e., unjustified algo-rithmic compressions) on the data that may ob-scure the effects of power and bias. Seeing this,postmodernists challenge the legitimacy of the-orizing even in those regions of the power-lawslope—the upper-left region of Figure 2—whereGaussian assumptions may actually be war-ranted. However, by arguing that error-eliminat-ing statistical strategies eliminate more thanjust errors—they also eliminate “weak voices”—postmodernists underplay the methodologicalvalue of replicability and explanatory coverage(Mayo, 1996) that makes some theories moreplausible than others.

Since for postmodernists all theory choice is,at base, politically driven, they find no convinc-ing basis for moving down the Ashby Space ofFigure 3. Yet the “infinite conversations” theyadvocate are a luxury that a practical resource-constrained manager can ill afford; they consti-tute counsels of perfection that have little adap-tive potential. Thus, just as the truth claims ofthe atomistic ontology underpinning modernistdiscourse become increasingly suspect whenmade too far down the power-law slope of Fig-ure 2, so the connectionist ontology underpin-ning the postmodernist discourse overreachesitself when it claims that meaning—belongingas it does to the realm of language and socialinteraction effects—remains unconstrained bythe real-world dynamics operating in the fig-ure’s upper-left-hand regions.

Both modernists and postmodernists aim forreliable knowledge, but, holding competing on-tologies, they end up talking right past eachother. Figure 2, however, suggests that there is a


time to be atomistic and a time to be connection-ist and that it is the degree of adaptive tensionpresent in a system—as determined by someorder parameter—that influences the degree ofconnectivity present among phenomena. Theseemingly incompatible ontologies can thus bereconciled. Given connectivity, one has to ac-cept the possibility of power-law-distributed, oc-casionally extreme, and unpredictable out-comes and, hence, be willing to settle for beingroughly right rather than precisely wrong. In theconnectionist world of living systems, the “jus-tification” of knowledge resides primarily in itscontribution to efficacious adaptability and sur-vival rather than to the attainment of a predic-tive law-like truth (Gell-Mann, 2002). Falsifica-tion, Popper’s (1935) criterion of demarcationbetween science and nonscience, remains inforce since “false” knowledge threatens both ad-aptation and survival. Sooner or later, realitykicks back (Popper, 1983).

In pursuit of a stable and predictable order,modernists who find themselves in the chaoticregime of Figure 3 aim at reaching point C,located in a region of the Ashby Space wherecompact statistical representations have pur-chase. Postmodernists finding themselves in thesame regime, in contrast, are drawn towardpoint B, located in a region where the descrip-tion of events is incompressible and only de-tailed narrative is possible. The complexity per-spective, however, identifies point D as morerelevant to organization science. Given adap-tive tension of some kind, intelligent, interde-pendent agents in the Ashby Space constitutecomplex adaptive systems (CASs; Holland, 1988,2002) striving for improved fitness, growth, andsurvival via self-organizing processes that weassociate with the complex regime of Figure 3.Many of their interdependent behaviors giverise to scale-free dynamics and result in powerlaws. In order to economize on scarce energeticand computational resources, for example,many agents typically seek out the ordered re-gime of Figure 3. Yet because of their own col-lective actions and unpredictable events, theyoften find themselves in the chaotic regime.

Organizational researchers study phenomenathat typically fall somewhere within the com-plex regime—that is, they are neither so lackingin structure as to remain stuck in the chaoticregime nor so structured as to end up in theordered regime. The complex regime is the one

in which the power-law distributions of Figure 1(stylized in Figure 2) make their appearance.Here, compact symbolic representations coexistwith more discursive narrative ones. Yet whilethe ebb and flow of adaptive tension causesbehaviors to shift toward the upper left or lowerright along the distribution, modernist thinkingwants to draw organizational research perma-nently down into the ordered regime of Figure 3.This accommodates normally distributed phe-nomena located in the upper-left “Gaussian” re-gion of Figure 2. Postmodernist thinking, on theother hand, believes that the natural home oforganization research is the chaotic regime—the region that occasionally produces theunique and sometimes extreme events locatedin the lower right of Figure 2. Yet since low-to-high variations in adaptive tension often causescalable outcomes to progress from upper left tolower right down the power-law slope, so shouldmanagement and organizational analysis. Trav-eling left up the slope, one deduces observablebehaviors from underlying patterns in causaldynamics. Traveling from upper left down theinverse slope, however, requires more than in-duction. It calls for an inferential strategy thatwe label scalable abduction. Scalability is whatcauses the target phenomenon to spiral out intothe extreme outcomes located on the lower rightof the power-law slope.

Scalable Abduction

According to Peirce, “Abduction . . . consists ofexamining a mass of facts and in allowing thesefacts to suggest a theory” (1935: 205). Abductionseeks inference toward the best explanation,one that turns on the coherence with which anovel or anomalous event can be related to abackground theory (Aliseda, 2006; Thagard, 2006;Thagard & Shelley, 1997). The observed behaviorof workers in the Hawthorne experiments, forexample, was anomalous relative to prevailingbackground theories of worker motivation(Roethlisberger & Dickson, 1964). These theoriesthen had to be either modified or broadened to“explain” the anomaly. Scalable abduction in-fers toward the best scalable explanation. Asoutcomes move down the power-law slope, scal-able abduction calls for explanations based ontheories about causes operating in the samemanner across the multiple levels/scales of asystem (Gell-Mann [2002: 23] called this “middle-


level” theorizing); it focuses on tiny initiatingevents coupled with scale-free causes operatingfrom the very small to the very large so as toexplain infrequently occurring extreme out-comes. Thus, associating a Gaussian epistemol-ogy with the upper-left region of Figure 2 and a“narrative” epistemology with the lower-rightregion, scalable abduction offers an inferentialengine that can travel between them and trackthe dynamics by which certain tiny events getamplified into extreme outcomes.

When applied to distributions of phenomenagoverned by power laws, scalable abduction al-lows one to derive limited but nonetheless use-ful expectations concerning scale-free dynamicsand the causal processes that underpin them.Scale-free dynamics emerge from myriad lower-level “tiny initiating events” (Holland, 2002),some of which propagate out causally and ex-plode into the larger events that make up oneend of the power-law distribution (Andriani &McKelvey, 2007, 2009; Gell-Mann, 2002). The met-aphor is of a butterfly flapping its wings overeastern Brazil and ultimately triggering a tor-nado in Texas—a “butterfly event” (Lorenz,1972). Here, events uncovered at one scale justifysome forms of extrapolation out to less frequent,more extreme events at another.

Lying between idiosyncratic inductions andpredictions based on deductive tests, scalableabduction offers anticipation. Anticipation is“softer” than prediction, bridging between thestrong predictive claims achievable in, say,classical physics and the unpredictable, oftenseemingly chaotic press of singular events con-fronting us daily at the human scale. Both pre-diction and anticipation shape our expectationsand orient our responses. Both draw on evidencefor their justification, although anticipation, of-ten only expressible in a loose, narrative form,achieves less precision than prediction. Whilepredictability is problematic given complexity,anticipation remains fluid with respect tochanging conditions and tensions, thereby facil-itating adaptive action and survival.

IMPLICATIONS FOR MANAGEMENTRESEARCH

Four points emerge from our analysis:

1. The atomistic and connectionist ontologiesthat respectively underpin modernist and

postmodernist positions have been treatedby organizational researchers as being an-tagonistic to each other (McKelvey, 2003).

2. They each occupy different end points of apower-law distribution that reflects com-plex dynamics such as SOC and new ordercreation (McKelvey, 2004).

3. Under adaptive tension, these dynamicsconnect hitherto disconnected small eventsso as to produce ever-larger, more complex,but less frequent outcomes (Andriani &McKelvey, 2007).

4. A power-law distribution thus reconcilesthe two antagonistic ontologies in a singleoverarching ontology that makes the appro-priateness of either modernist or postmod-ernist perspectives contingent on the de-gree of tension and connectivity present inthe system (stylized in Figure 2).

What implications do the above points carryfor organizational researchers? We identify five:

1. Engage with the properties of power-lawdistributions and the different epistemic strate-gies in the Ashby Space that these suggest. Inthis space, for example, the chaotic regime de-scribes the world of Heraclitus, who famouslysaid, “The river where you set your foot just nowis gone. Those waters giving way to this, nowthis” (Haxton, 2001). Frequently, ours is an epis-temically fragile world of unique yet connectedphenomena that unfold unpredictably and canonly be narrated, not analyzed into simplisticformulas. To the extent that living (social) sys-tems exhibit any regularities—that is, phenom-ena that repeat—we can move down into thecomplex regime where connections become con-tingent and some analysis becomes possible.Here we discover the world of power laws—distributions in which small events sometimesscale up into extreme outcomes. Such phenom-ena cannot be summarily summarized by themeans and standard deviations of “normal”Gaussian statistics. Instead of analyses andtheories based on our conventional statisticalmethods, we need scalable abduction andscale-free causal theories (Gell-Mann’s middle-level theories [2002]). In Table 1 we briefly defineeight of the fifteen scale-free theories thatreadily apply to organizations (Andriani & Mc-Kelvey, 2009).

Epistemic robustness may only be achievablein the ordered regime of the Ashby Space, wherenormally distributed phenomena are suffi-ciently similar and disconnected that the statis-ticians’ i.i.d. assumptions apply. They can then


be aggregated into stable classes and their be-havior deductively predicted. Although theseconditions can only be met by some of the phe-nomena that humans encounter as they goabout their business, they do bring better under-standing of behavior when applicable. While nonatural boundaries separate the three re-gimes—they interpenetrate—the first is the nat-ural home of the historian, the second of socialscientists and biologists, and the third of scien-tists who study nonliving phenomena. Needlessto say, since effective representations in eachregime will call for a different mix of narrativeand abstract symbolic resources, epistemic flex-ibility and tolerance are called for.

2. Explore the power-law distribution beforeyou exploit it. Complexity and power-law think-ing offer researchers and practitioners a choiceof strategies. A move toward the world of Hera-clitus leads them to samples of one and epis-temic fragility. Finding themselves in unfamil-iar territory, they are in March’s (1991)exploratory mode of learning and must behavelike hunter-gatherers. A move toward the worldof normal distributions, in contrast, leads themtoward large i.i.d. samples and epistemic ro-

bustness. Here the territory is more familiar, al-lowing them to operate in March’s exploitativemode of learning and to behave like settledfarmers (Hurst, 1995). It is in the world of power-law distributions, however, that managementand organizational researchers operate in fron-tier scientific territory and have to balance outexploration and exploitation as described byMarch—call this “homesteading.”

Good science requires us to deploy a researchstrategy appropriate to our epistemic circum-stances. Hans Reichenbach, a friend of the Vi-enna Circle, claimed that exploration— hecalled this “discovery logic”—was of no interestto the philosophy of science. Only exploitation—“justification logic”—was of relevance (Reichen-bach, 1938). Yet the positivists advocated sostringent a conception of knowledge that neitherthe natural nor the social sciences could satisfyit. But we don’t get to exploit anything unless wehave paid our dues in the coin of exploration.Homesteading precedes farming, and a long pe-riod of hunter-gathering may, in turn, precedehomesteading. Effective research requires us totravel up and down the Ashby Space—and, byimplication, in both directions along the stylized

TABLE 1A Sample of Scale-Free Theories of Naturea

Theory Definition

Phase transition Exogenous energy impositions cause autocatalytic interaction effects such that newinteraction groupings form (Prigogine & Stengers, 1997)

Spontaneous order creation Heterogeneous agents seeking out other agents to copy/learn from so as to improve fitnessgenerate networks; with positive feedback, some networks become groups, and somegroups become larger groups and hierarchies (McKelvey & Lichtenstein, 2007)

Preferential attachment Given newly arriving agents in a system, larger nodes with an enhanced propensity toattract agents will become disproportionately even larger (Barabasi, 2002)

Combination theory Multiple exponential or log-normal distributions or increased complexity of components(subtasks, processes) sets up, which results in a power-law distribution (Newman, 2005;West & Deering, 1995)

Least effort Word frequency is a function of ease of usage by both speaker/writer and listener/reader(Zipf’s [power] Law [1949]), now found to apply to firms and economies in transition(Ishikawa, 2006; Podobnik, Fu, Jagric, Grosse, & Stanley, 2006)

Square-cube law Surfaces absorbing energy grow by the square, but organisms grow by the cube, resultingin an imbalance; fractals emerge to balance surface/volume ratios (Carneiro, 1987)

Connection costs As cell fission occurs by the square, connectivity increases by n(n ! 1)/2, producing animbalance between the gains from fission and the cost of maintaining connectivity;consequently, organisms form modules or cells so as to reduce the cost of connections(Simon, 1962)

Self-organized criticality Under constant tension of some kind (gravity, ecological balance), some systems reach acritical state where they maintain stasis by preservative behaviors, such as Bak’s smallto large sandpile avalanches, which vary in size of effect according to a power law (Bak,1996)

a We use eight out of fifteen scale-free theories discussed in Andriani and McKelvey (2009).


power-law distribution of Figure 2. There isscope for a division of labor since differences inthe cognitive style of researchers will push theminto different regions of the space. Useful knowl-edge creation, however, ultimately requiressuch labor to be coordinated and integrated.

3. Do not privilege one part of the power-lawdistribution at the expense of another. The worldis a dynamic place, subject both to the emer-gence of order and, according to the second lawof thermodynamics, its erosion. The first law ofthermodynamics holds that the conservation ofenergy drives the creation of matter. In neoclas-sical economics the first law fostered an equi-librium-focused mathematics (Colander, 2006;Mirowski, 1989). The second law of thermody-namics holds that ordered energy-based struc-tures eventually deteriorate into random-ness—a process called “entropy production”(Swenson, 1989). While some structures tempo-rarily stabilize, others rapidly disintegrate. Tounderstand organizational phenomena is to un-derstand these opposing processes.

If we view organizations through a networklens (Boisot & Lu, 2007), we see that organiza-tional research studies the regularities that gov-ern the interdependencies among differentnodes in a network—that is, the structure andthe dynamics of their connectivity. Since nodescan be individuals, departments within an orga-nization, or whole organizations, we see thatmany of these regularities are scalable (Bara-basi, 2002). And since connectivity is a variablethat reflects the level of adaptive tension in thenetwork, organizational research must engagewith the power-law distribution as a whole,without privileging one particular region at theexpense of another. It cannot therefore presumethat studies of “average” or “typical” organiza-tions accurately reflect organizational proper-ties ranging across an entire power-law dis-tribution. Just as Axtell (2008) invoked thepower-law distribution of firm size in the UnitedStates to claim that there is no such thing as thetypical firm, so we hypothesize that the “aver-age” organization does not exist.

4. Study the causal dynamics that call for scal-able abduction. Power laws are the signature ofSOC in natural and social systems. By focusingon circumstances under which independentevents and processes connect, scalable abduc-tion becomes the inferential strategy of choicefor studying SOC in particular and organization-

al phenomena in general. Scalable abductionturns out to be the basis of Dilthey’s (1959) Ver-stehen (understanding), a diacritical conceptdistinguishing natural from cultural sciences.While scalable abduction does not necessarilyyield strong or precise predictions (Burrell &Morgan, 1979), it offers a new answer to the oldquestion of whether a science of history is pos-sible. Historicism argues that history is subjectto laws that allow prediction. From a modernistperspective, however, samples of one—uniqueevents—cannot exhibit law-like behavior and,hence, remain beyond the reach of prediction(Popper, 1945).

Yet do we not also hear that those who fail tolearn the lessons of history are condemned torepeat them? Although it does not allow thelevels of prediction achievable in some of thenatural sciences, for living systems like organi-zations, abductive inference offers a useful levelof anticipation, one that can be efficaciouslyadaptive. A key challenge here is to separatethe small events that are likely to remain inde-pendent and random from the small events that,driven by tension-induced connectivities, arelikely to become scalable. Such events don’tusually come with labels attached, so withoutsome understanding of how scalable causal dy-namics arise in organizations (see Table 1), it ishard to distinguish butterfly events ex ante fromrandom, independent ones. Here, by their fine-grained analyses of individual events, the nar-rative strategies advocated by postmodernistscome into their own. They invite us to proceedcautiously and to avoid the premature closurethat overhasty, statistically driven hypothesiz-ing can produce.

5. Link the methodologies available for study-ing different points on the power-law slope. Gra-ham Allison’s (1971) analysis of the Cuban Mis-sile Crisis illustrates some of the issues we arediscussing. As Allison tells it, the crisis camevery close to generating the ultimate extremeevent: a thermonuclear war. Given the raceagainst time, the level of adaptive tension wasextremely high—at certain moments during thecrisis, the slightest mishap could have triggereda nuclear missile exchange between the SovietUnion and the United States. Had Kennedy notengaged with the causal texture of the eventsthat made up the crisis at the appropriate lev-el—had he not been sensitive to the presence of


butterfly events—the world would have plungedinto a nuclear abyss.

In the book Allison develops three differentmodels through which the crisis can be ana-lyzed: Model I, the rational actor; Model II, or-ganizational process; and Model III, governmen-tal politics. Model I treats the state as a stableand independent object whose behavior is ra-tional and predictable, Model II unpacks thestate to reveal a more complex and organizedentity subject to divergent rules and routinesthat undermine some of the rationality imputedto it by Model I, and Model III subordinates thebehavior of the different components that makeup the state to the games played by politicalactors. Each model adds a layer of complexity—and, by implication, of narrative richness—tothe analysis and improves its explanatorypower. The first seeks predictability; the thirdunderstanding.

Model III is the most complex—not to say cha-otic—of all three models. As Allison points out,the information needed by Models II and IIIdwarfs that required by Model I. In fact, to ad-vocates of Model I, the information requirementsof Model III reflect an “undue concern with sub-tlety” (Allison, 1971: 251). Model I is coarsegrained, offering an informative summary oftendencies, whereas model III is fine grained.Allison’s three models complement each other.The best foreign policy analysts weave strandsof the three models into their accounts. The keydifference between Allison’s and our approachis that whereas he produced three different per-spectives that happened to fruitfully comple-ment each other—after all, they may well haveturned out to be based on incommensurate par-adigms—we locate our three ontologies along asingle continuum that theoretically integratesthe different perspectives we have discussedand identifies the inferential conditions for mov-ing along the continuum in either direction.

Historical case-based narratives such as Alli-son’s look at the way events have connected. Buthistory as currently conceived only delivers use-ful lessons if events connect this way again; itthen has predictive value. Heraclitus, however,tells us that events never connect in the sameway twice. Anticipation is both less demandingand more demanding of history. It does not askhow things will connect but how they could con-nect. It is less demanding in that it does not seekpredictive accuracy or precision. It is more de-

manding because it has to explore a muchlarger space of possibilities than prediction re-quires.

Today, high-powered, agent-based simulationmodels make it possible to engage in such ex-plorations. If deduction was the inferential strat-egy of choice of a prestatistical age and induc-tion that of the statistical age (Stigler, 1986), wehypothesize that scalable abduction will be-come the inferential strategy of choice in theage of computational modeling (Epstein & Ax-tell, 1996; North & Macal, 2007; Tesfatsion & Judd,2006). It allows one to move methodically acrosslevels of resolution and analysis and to explorestatistical and narrative data in ways that werenot available to Allison. As he put it in the con-cluding section of his book,

What we need is a new kind of “case study” donewith theoretical alertness to the range of factorsidentified by Models I, II, and III (and others) onthe basis of which to begin refining and testingpropositions and models (Allison, 1971: 273).

CONCLUSION

By accommodating the dynamics of tensionand connectivity, an epistemology based oncomplexity science offers management and or-ganizational researchers a more encompassinglegitimacy than either modernist or postmodern-ist epistemologies on their own—one that iswell aligned with emerging concepts of organi-zational complexity (Allen, Maguire, & Mc-Kelvey, in press; Lewin, 1999; Maguire, Mc-Kelvey, Mirabeau, & Oztas 2006). If effectiveorganizational complexity lies between orderand chaos, then, by implication, so does theeffective legitimacy of management research.This location implies a methodological expan-sion out from the world of stable, normally dis-tributed entities toward the more kaleidoscopicand problematic world captured by power-lawdistributions.

Ours is a plea for a new direction in organi-zation and management research—and morebroadly in the social sciences. The paradigmaticcompetition between modernism and postmod-ernism has not been fruitful. Natural scientistsand neoclassical economists continue to es-pouse a modernist stance, and many social sci-entists continue to espouse that of postmodern-ism (Colander, 2006; Kelso & Engstrøm, 2006;Mirowski, 1989; Ormerod 1994, 1998). Conse-


quently, the legitimacy of management re-search’s would-be truth claims remains stuck inan epistemological quagmire. Morin (1992), how-ever, pointed out that the new complexity sci-ences are now dissolving the distinction be-tween the natural and the social sciences. Thecomplexity perspective suggests that where pre-diction is problematic, anticipation offers use-fully adaptive information and, hence, becomesa legitimate goal for scientific endeavors. Thus,while the criteria of demarcation that separatescience from nonscience need not be aban-doned—as advocated by Feyerabend (1975) andsome postmodernists—they need to be rathermore accommodating than those promulgatedby modernists.

Organizational researchers study interacting,interdependent agents—individuals, depart-ments, firms, etc. These simply do not behavelike a collectivity of autonomous agents. In-formed interdependencies are the stuff of orga-nization and, indeed, of life itself. Postmodernistorganizational researchers are right in thinkingthat the complexity that results is not well cap-tured by the analytical tools forged by modern-ist thinking. They are, however, wrong in think-ing that such complexity is beyond the reach ofany kind of managerially useful analysis.Agent-based simulation modeling, for example,today provides both natural and social scien-tists with tools for studying the complex scal-able processes outlined in this paper (Epstein,2007; Epstein & Axtell, 1996; North & Macal, 2007;Tesfatsion & Judd, 2006). It is ideally suited toexploring the wide range of possible outcomesout of which more probable ones might emerge.Such possibility thinking—Kauffman (2000) callsit the “adjacent possible”—would place organi-zation scholars more firmly in the context ofdiscovery (Reichenbach, 1938) so long shunnedby modernists, without in any way underminingthe case for a subsequent justification. The nar-rative strategies of the postmodernist wouldthen be used by scholars to select the most plau-sible of these possibilities—those that squareabductively with their prior experience. The ap-proach would, in effect, legitimate a more cre-ative, exploratory approach to organizational re-search that prevails in many Ph.D. programs,one that acknowledges the contingent nature ofmany organizational processes even as it seeksa robust understanding of their exploitable reg-ularities.

REFERENCES

Aliseda, A. 2006. Abductive reasoning: Logical investigationsinto discovery and explanation. Dordrecht: Springer.

Allen, P., Maguire, S., & McKelvey, B. In press. The SAGEhandbook of complexity and management. London:Sage.

Allison, G. T. 1971. Essence of decision: Explaining the CubanMissile Crisis. Boston: Little, Brown.

Alvesson, M., & Deetz, S. 1996. Critical theory and postmod-ernism approaches to organizational studies. In S. R.Clegg, C. Hardy, & W. R. Nord (Eds.), Handbook of orga-nization studies: 191–217. London: Sage.

Andriani, P., & McKelvey, B. 2007. Beyond Gaussian aver-ages: Redirecting organization science toward extremeevents and power laws. Journal of International Busi-ness Studies, 38: 1212–1230.

Andriani, P., & McKelvey, B. 2009. From Gaussian to Paretianthinking: Causes and implications of power laws inorganizations. Organization Science, 20: 1053–1071.

Ashby, R. W. 1956. An introduction to cybernetics. London:Methuen.

Axtell, R. L. 2008. Nonexistence of a typical firm in the U.S.economy: Extremely heavy tails in firm size and growth.Paper presented at the Organization Science WinterConference, Squaw Creek, CA.

Bak, P. 1996. How nature works: The science of self-organizedcriticality. New York: Copernicus.

Barabasi, A.-L. 2002. Linked: The new science of networks.Cambridge, MA: Perseus.

Benacerraf, P., & Putnam, H. (Eds.). 1964. Philosophy of math-ematics: Selected readings. Cambridge: CambridgeUniversity Press.

Berger, P., & Luckmann, T. 1966. The social construction ofreality. Middlesex, UK: Penguin Books.

Blackmore, S. 1999. The meme machine. New York: OxfordUniversity Press.

Boisot, M., & Child, J. 1999. Organizations as adaptive sys-tems in complex environments: The case of China. Or-ganization Science, 10: 237–252.

Boisot, M., & Lu, X. 2007. Competing and collaborating innetworks: Is organizing just a game? In M. Gibbert &T. Durand (Eds.), Strategic networks: Learning to com-pete: 151–169. Malden, MA: Blackwell.

Boisot, M., & McKelvey, B. 2007. Extreme events, power laws,and adaptation: Towards an econophysics of organiza-tion. Paper presented at the annual meeting of the Acad-emy of Management, Philadelphia.

Boltzmann, L. 1887. Uber einige Fragen der Kinetische Gas-theorie. Wiener Berichte, 96: 891–918.

Bridgeman, P. 1936. The nature of physical theory. Princeton,NJ: Princeton University Press.

Brunk, G. G. 2002. Why are so many important events unpre-dictable? Self-organized criticality as the “engine of his-tory.” Japanese Journal of Political Science, 3: 25–44.


Burrell, G., & Morgan, G. 1979. Sociological paradigms andorganizational analysis. London: Heinemann.

Calas, M. B., & Smircich, L. 1999. Past postmodernism? Re-flections and tentative directions. Academy of Manage-ment Review, 24: 649–671.

Callon, M. 1986. Some elements of a sociology of translation:Domestication of the scallops and the fishermen of St.Brieuc Bay. In J. Law (Ed.), Power, action and belief:196–229. London: Routledge & Kegan Paul.

Carneiro, R. L. 1987. The evolution of complexity in humansocieties and its mathematical expression. Interna-tional Journal of Comparative Sociology, 28: 111–128.

Carter. P., & Jackson, N. 2004. Gilles Deleuze and FelixGuattari: A “minor” contribution to organization theory.In S. Linstead (Ed.), Organization theory and postmodernthought: 105–126. London: Sage.

Cilliers, P., 1998. Complexity and postmodernism: Under-standing complex systems. London: Routledge.

Colander, D. 2006. Post Walrasian macroeconomics: Beyondthe dynamic stochastic general equilibrium model.Cambridge: Cambridge University Press.

Crozier, M. 1964. Le phenomene bureaucratique. Paris: Edi-tions du Seuil.

Dawkins, R. 1976. The selfish gene. Oxford: Oxford UniversityPress.

Deleuze, G., & Guattari, F. 1984. Anti-Oedipus: Capitalismand schizophrenia. (Translated by R. Hurley, M. Seem, &H. R. Lane.) London: Athlone.

Dennett, D. 1989. The intentional stance. Cambridge, MA:MIT Press.

Derrida, J. 1978. Writing and difference. London: Routledge &Kegan Paul.

Dewey, J. 1925. Experience and nature. Chicago: Open Court.

Dilthey, W. 1959. Gesammelte Schriften. Stuttgart: Teubner.

Dooley, K. J., & Van de Ven, A. H. 1999. Explaining complexorganizational dynamics. Organization Science, 10: 358–372.

Epstein, J. 2007. Generative social science: Studies in agent-based computational modeling. Princeton, NJ: PrincetonUniversity Press.

Epstein, J. M., & Axtell, R. 1996. Growing artificial societies:Social science from the bottom up. Cambridge, MA: MITPress.

Feyerabend, P. K. 1975. Against method. London: New LeftBooks.

Feynman, R. P. 1967. The character of physical law. Cam-bridge, MA: MIT Press.

Fisher, R. A. 1918. The correlation between relatives on thesupposition of Mendelian inheritance. PhilosophicalTransactions of the Royal Society of Edinburgh, 52: 399–433.

Ford, B. J. 1992. Brownian movement in clarkia pollen: Areprise of the first observations. The Microscope, 40: 235–241.

Foucault. M. 1969. L’archeologie du savoir. Paris: Gallimard.

Foucault. M. 1975. Surveiller et punir: Naissance de la prison.Paris: Gallimard.

Friedman, M. 1953. Essays in positive economics. Chicago:University of Chicago Press.

Gell-Mann, M. 1994. The quark and the jaguar. New York:Freeman.

Gell-Mann, M. 2002. What is complexity? In A. Q. Curzio &M. Fortis (Eds.), Complexity and industrial clusters: 13–24. Heidelberg: Physica-Verlag.

Gibbs, J. W. 1902. Elementary principles in statistical me-chanics. New York: Charles Scribner’s Sons.

Glaser, P. 2009. Fitness and inequality in an increasing re-turns world: Applying the tools of complexity economicsto study the changing distribution of US stock marketcapitalizations from 1930 to 2008. Unpublished paper,UCLA Anderson School of Management, Los Angeles.

Golinski, J. 1998. Making natural knowledge. Cambridge:Cambridge University Press.

Greene, W. H. 2002. Econometric analysis (5th ed.). Engle-wood Cliffs, NJ: Prentice-Hall.

Guth, A. 1997. The inflationary universe. Cambridge, MA:Perseus Books.

Habermas, J. 1972. Knowledge and human interests. London:Heinemann.

Hahlweg, K., & Hooker, C. A. (Eds.). 1989. Issues in evolution-ary epistemology. New York: State University of NewYork.

Haken, H. 1983. Synergetics, an introduction (3rd ed.). Berlin:Springer-Verlag.

Haxton, B. 2001. The collected wisdom of Heraclitus. NewYork: Viking Penguin.

Hempel, C. G. 1965. Aspects of scientific explanation. NewYork: Free Press.

Holland, J. H. 1988. The global economy as an adaptivesystem. In P. W. Anderson, K. J. Arrow, & D. Pines (Eds.),The economy as an evolving complex system: 117–124.Reading, MA: Addison-Wesley.

Holland, J. H. 2002. Complex adaptive systems and sponta-neous emergence. In A. Q. Curzio & M. Fortis (Eds.),Complexity and industrial clusters: 24–34. Heidelberg:Physica-Verlag.

Hollinger, R. 1994. Postmodernism and the social sciences: Athematic approach. London: Sage.

Hollis, M. 1982. The social destruction of reality. In M. Hollis& S. Lukes (Eds), Rationality and relativism: 67–86. Ox-ford: Blackwell.

Hurst, D. K. 1995. Crisis and renewal: Meeting the challengeof organizational change. Boston: Harvard BusinessSchool Press.

Ishikawa, A. 2006. Pareto index induced from the scale ofcompanies. Physica A, 363: 367–376.

Israel, J. 2001. Radical enlightenment. Oxford: Oxford Uni-versity Press.

James, W. 1907. Pragmatism: A new name for some old waysof thinking. New York: Longman Green.


Jenks, C., 1992. The post-modern agenda. In C. Jenks (Ed.),The postmodern reader: 10 –39. London: St. Martin’sPress.

Kauffman, S. A. 1993. The origins of order. New York: OxfordUniversity Press.

Kauffman, S. A. 2000. Investigations. New York: Oxford Uni-versity Press.

Kaye, B. 1993. Chaos & complexity. New York: VCH.

Kelso, J. A. S., & Engstrøm, D. A. 2006. The complementarynature. Cambridge, MA: MIT Press.

Koertge, N. 1998. A house built on sand: Exposing postmod-ernist myths about science. New York: Oxford UniversityPress.

Kuhn, T. S. 1962. The structure of scientific revolutions. Chi-cago: University of Chicago Press.

Lakatos, I. 1976. Proofs and refutations: The logic of mathe-matical discovery. (Edited by J. Worrall & E. Zahar.)Cambridge: Cambridge University Press.

Latour, B. 1988. The Pasteurization of France. Cambridge,MA: Harvard University Press.

Lawson, T. 1997. Economics & reality. New York: Routledge.

Lewin, A. Y. (Ed.). 1999. Special issue. Organization Science,10(3).

Lorenz, E. N. 1972. Predictability: Does the flap of a butterfly’swings in Brazil set off a tornado in Texas? Paper pre-sented at the meeting of the American Association forthe Advancement of Science, Washington, DC.

Lyotard, J.-F. 1984. The postmodern condition. Manchester,UK: Manchester University Press.

Maguire, S., McKelvey, B., Mirabeau, L., & Oztas, N. 2006.Complexity science and organizational studies. In S. R.Clegg, C. Hardy, T. Lawrence, & W. Nord (Eds.), Hand-book of organizational studies (2nd ed.): 165–214. Thou-sand Oaks, CA: Sage.

Mainzer, K. 2007. (First published in 2004.) Thinking in com-plexity (5th ed.). New York: Springer-Verlag.

Mandelbrot, B. 1982. The fractal geometry of nature. NewYork: Freeman.

March, J. G. 1991. Exploration and exploitation in organiza-tional learning. Organization Science, 2: 71–87.

March, J. G., Sproull, L. S., & Tamuz, M. 1991. Learning fromsamples of one or fewer. Organization Science, 2: 1–13.

Mayo, D., 1996. Error and the growth of experimental knowl-edge. Chicago: University of Chicago Press.

McKelvey, B. 1999. Toward a Campbellian realist organiza-tion science. In J. A. C. Baum & B. McKelvey (Eds.),Variations in organization science: In honor of Donald T.Campbell: 383–411. Thousand Oaks, CA: Sage.

McKelvey, B. 2001. Energizing order-creating networks of dis-tributed intelligence. International Journal of InnovationManagement, 5: 181–212.

McKelvey, B. 2003. Postmodernism vs. truth in managementtheory. In E. Locke (Ed.), Post modernism and manage-ment: Pros, cons and the alternative. Research in the

sociology of organizations, vol. 21: 113–168. Amsterdam:Elsevier Science.

McKelvey, B. 2004. Toward a 0th law of thermodynamics:Order-creation complexity dynamics from physics and bi-ology to bioeconomics. Journal of Bioeconomics, 6: 65–96.

McKelvey, B. 2008. Emergent strategy via complexity leader-ship: Using complexity science & adaptive tension tobuild distributed intelligence. In M. Uhl-Bien &R. Marion (Eds.), Complexity and leadership. Volume I:Conceptual foundations: 225–268. Charlotte, NC: Infor-mation Age.

McKelvey, B., & Boisot, M. 2009. Redefining strategic fore-sight. In L. Costanzo & B. MacKay (Eds.), Handbook ofresearch on strategy and foresight: 15–47. Cheltenham,UK: Edward Elgar.

McKelvey, B., & Lichtenstein, B. B. 2007. Leadership in fourstages of emergence. In J. K. Hazy, J. Goldstein, & B. B.Lichtenstein (Eds.), Complex systems leadership theory:93–107. Boston: ISCE.

Merton, R. K. 1949. Social theory and social structure. NewYork: Free Press.

Mirowski, P. 1989. More heat than light. Cambridge: Cam-bridge University Press.

Morin, E. 1992. Method: Toward a study of humankind. NewYork: Peter Lang.

Newman, M. E. J. 2005. Power laws, Pareto distributions andZipf’s law. Contemporary Physics, 46: 323–351.

Nicolis, G., & Prigogine, I. 1989. Exploring complexity: Anintroduction. New York: Freeman.

North, M. J., & Macal, C. M. 2007. Managing business com-plexity: Discovering strategic solutions with agent-based modeling and simulation. Oxford: Oxford Univer-sity Press.

Ormerod, P. 1994. The death of economics. New York: Wiley.

Ormerod, P. 1998. Butterfly economics: A new general theoryof social and economic behavior. New York: Pantheon.

Peirce, C. S. 1935. Collected papers, vol. 5. (Edited byC. Hartshorne & P. Weiss.) Cambridge, MA: HarvardUniversity Press.

Podobnik, B., Fu, D., Jagric, T., Grosse, I., & Stanley, H. E. 2006.Fractionally integrated process for transition econom-ics. Physica A, 362: 465–470.

Popper, K. R. 1935. Logik der Forschung. Vienna: SpringerVerlag. Reprinted as Popper, K. R. 1959. The logic ofscientific discovery. New York: Harper & Row.

Popper, K. R. 1945. The open society and its enemies. Volume1: Plato. Volume 2: Hegel and Marx. London: Routledge &Kegan Paul.

Popper, K. R. 1983. Realism and the aim of science. London:Hutchinson.

Prigogine, I., with Stengers, I. 1997. The end of certainty. NewYork: Free Press.

Radnitzky, G., & Bartley, W. W., III. 1987. Evolutionary epis-temology, rationality, and the sociology of knowledge.La Salle, IL: Open Court.


Reichenbach, H. 1938. Experience and prediction. Chicago:University of Chicago Press.

Roethlisberger, F. J., & Dickson, W. J. 1964. Management andthe worker. New York: Wiley.

Rorty, R. 1980. Philosophy and the mirror of nature. Oxford:Blackwell.

Rorty, R. 1989. Contingency, irony, and solidarity. Cam-bridge: Cambridge University Press.

Shapin, S., & Schaffer, S. 1985. Leviathan and the air-pump.Princeton, NJ: Princeton University Press.

Simon, H. A. 1962. The architecture of complexity. Proceed-ings of the American Philosophical Society, 106: 467–482.

Stigler, S. M. 1986. The history of statistics: The measurementof uncertainty before 1900. Cambridge, MA: BelknapPress of Harvard University Press.

Suppe, F. 1977. The structure of scientific theories (2nd ed.).Chicago: University of Chicago Press.

Swenson, R. 1989. Emergent attractors and the law of maxi-

mum entropy production: Foundations to a theory ofgeneral evolution. Systems Research, 6(3): 187–197.

Taleb, N. N. 2007. The black swan. New York: Random House.

Tesfatsion, L., & Judd, K. 2006. The handbook of computa-tional economics, vol. 2. Amsterdam: Elsevier.

Thagard, P. 2006. Hot cognition. Cambridge, MA: MIT Press.

Thagard, P., & Shelley, C. P. 1997. Abductive reasoning:Logic, visual thinking, and coherence. In M. L. D. Chiara,K. Doets, D. Mundici, & J. van Bentham (Eds.), Logic andscientific methods: 413–427. Dordrecht: Kluwer.

Walker, H. W. 1940. Degrees of freedom. Journal of Educa-tional Psychology, 31: 253–269.

Weinberg, S. 1992. Dreams of a final theory. New York: Vin-tage Books.

West, B., & Deering, B. 1995. The lure of modern science:Fractal thinking. Singapore: World Scientific.

Zipf, G. K. 1949. Human behavior and the principle of leasteffort. New York: Hafner.

Max Boisot ([email protected]) is a professor at the ESADE business school inBarcelona; an associate fellow at the Said Business School, University of Oxford; anda fellow of the Snider Center at the Wharton School, University of Pennsylvania. Hereceived his Ph.D. from Imperial College, London University. His current research is onknowledge creation at CERN’s Large Hadron Collider.

Bill McKelvey ([email protected]) is a professor at the UCLA AndersonSchool of Management. He received his Ph.D. from MIT. His current writing focuses onphilosophy of science, organization science, complexity science, agent-based compu-tational modeling, and complexity leadership.


Copyright of Academy of Management Review is the property of Academy of Management and its content maynot be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express writtenpermission. However, users may print, download, or email articles for individual use.

integrating modernist and postmodernist perspectives on

Documents