Comput Math Organ Theory (2008) 14: 84–119 DOI 10.1007/s10588-008-9022-6
Putting the organization back into computational organization theory: a complex Perrowian model of organizational action Brian W. Kulik · Timothy Baker
Published online: 27 February 2008 © Springer Science+Business Media, LLC 2008
Abstract At best, computational models that study organizations incorporate only one perspective of how organizations are known to act within their environments. Such single-perspective models are limited in their generalizability and applicability to the real world and allow for researcher bias. This work develops a multi-agent simulation using eight different well-known organizational perspectives: Strategic choice, contingency theory, behavioral decision theory, enactment, resource dependence, institutional theory, population ecology, and transaction cost economics. A literature review of each field is applied to the construction of algorithms which, when combined with techniques derived from a literature review of computational modeling of organizations, was applied to the construction of a series of algorithms describing a multi-perspective computational model. Computer code was written based on the algorithms and run across different types of environments. Results were statistically analyzed to both validate the model and to generate contingency-oriented hypotheses. Conclusions were made with regard to the expected behavior of organizations and the model’s applicability toward further research. Keywords Computational organization theory · Agent-based simulation · Organizational complexity · Organization theory
B.W. Kulik () Department of Management, College of Business, Central Washington University, 400 E. University Way, Ellensburg, WA 98926, USA e-mail:
[email protected] T. Baker Department of Management and Operations, Washington State University, 2710 University Drive, CIC 125N, Richland, WA 99352-1671, USA e-mail:
[email protected]
Putting the organization back into computational organization theory
85
1 Introduction A central question in organization theory (OT) is: How do and should organizations behave within their environments? Despite its simplicity, answers are fleeting, in part because the delineations of organization-organization, environment-environment, and organization-environment boundaries are difficult draw clearly given real-world conditions (Starbuck 1976); however, this difficulty only points to the larger reality that organizations are undeniably and unavoidably complex (Perrow 1986), and as such our question is difficult to answer with any reliability and generality. Therefore, extant research on organizations and their environments has not progressed beyond basic, simple prescriptions for managers, such as establishing department-level fit between environmental uncertainty and departmental flexibility (Thompson 1967), and the idea that organizations enact their environments so organizational leaders should be aware of their own biases (Weick 1979). These general perspectives, with their general recommendations, are analytically narrow, and as Perrow (1986) rather systematically pointed out, are easily criticized for their avoidance of the complexity issue and unrealistic assumptions: Barnard (1938) and the human relations school “neglected power as a variable” (p. 264); bounded rationality “inadequately considered the question of how the power of matters is enhanced by their capacity to set premises” (p. 264); evolutionary theory “generally ignores state models” (p. 277); economics models “single out a concept such as markets versus hierarchies, . . . assume self-interest as the motive, [and] sweep aside reality” (p. 256); and institutional perspectives fail to “see society as adaptive to organizations” (p. 173). Perrow (1986) often pitted different perspectives against each other as contradictions for each. For example, economics perspectives contradicted the human relations view of organizations, and institutional and ecological perspectives at times contradicted each other. Indeed, an argument might be made that the formulation of each organizational perspective was motivated from a criticism of at least one of the other perspectives. The result of the proliferation of organizational perspectives might therefore be considered a set of independent perspectives of organizations that, when taken in total, constitute a relatively complete and realistic view of how organizations act within their environments, assuming that each perspective has something to contribute at the organizational level of analysis. Since the development and proliferation of the perspectives discussed by Perrow, progress has been made along Kuhn’s (1996) prediction that once the research paradigm has been set, work proceeds toward filling out that paradigm (in the social sciences, Kuhn allowed for simultaneous existence of multiple paradigms), so that today most researchers will claim an “expertise” in only one perspective, such as institutional theory, agency theory, population ecology, decision theory, or resource dependence. Conferences, journals, and textbooks follow suit and are organized along the separation of organizational perspectives. Yet this progression into numerous fields of specialization ignores Perrow’s (1986) criticism that organizations are unavoidably complex, and that any one perspective taken by itself is taken out of context. An alternative to the above analytical approach is to simultaneously accommodate for multiple views of organizations in a simulated environment. However, while modern computers have this capability, the field of computational organization theory (COT) has tended to retain the analytical, one-perspective style that is prevalent
86
B.W. Kulik, T. Baker
in OT, with studies focused on behavioral decision theory (Cyert and March 1963; Masuch and LaPotin 1989; Prietula and Watson 2000), contingency theory (Burton and Obel 2004; Levinthal 2000; Lin 1998; Masuch and Huang 1996), and organizational ecology (Lomi and Larsen 1998; Sastry 2000); do not environmental conditions influence decision-making, and ecological conditions influence environmental conditions? It is our position that COT has much more to contribute to OT in that the complexity of multiple perspectives can be directly modeled with a computer simulation, which can be used for experimentation to hypothesize about real-world relationships. In other words, we argue that while a “good model should reflect the central features of the phenomena of interest” and “suppress less central details” (Levinthal 2000, p. 331), this does not imply that the entire model should consist of only the “phenomena of interest”, but also should include the modeling of all phenomena that are expected to affect those phenomena of interest. Thus, we agree with Ashworth and Carley’s (2007, p. 101) note that with regard to modeling of organizational behavior, “there is much ground to cover in integrating existing theory using computational approaches”, albeit at the organizational level of analysis in the present case. Furthermore, as such a level of complexity is too difficult to model mathematically, this leaves computer simulation as the most appropriate methodology to explore complex organizational phenomena: on the one hand, simulation can conduct experiments that are impossible to carry out in the real world, while on the other hand, simulation can readily accommodate the required level of complexity (Naveh and Sun 2006; Law and Kelton 2000). If computer simulation, as a research methodology, is to be used as intended by Law and Kelton (2000), then an integrated, multi-perspective architecture, also called a canonical representation (Levinthal 2000), of complex organizations should be created and provided to researchers, with each researcher asking a unique question and extending the architecture accordingly. If these simultaneous views are reasonably sufficient, and if they have been reasonably validated, then given some properly framed question, a sufficiently modeled extension of the canonical representation should offer some predictions for real-world phenomena which could not have been derived from either real-world or mathematical methodologies. In an effort to guide research on organizations toward a more complex and integrated approach, this paper proposes such an architecture of organizational behavior at the organizational level of analysis. As such, this study is expected to contribute to extant theory several ways. First, it constitutes a novel approach toward accommodating issues of organizational complexity and thus advances the field of organization science; second, it advances computational organization theory (COT) by combining OT perspectives into a single, necessarily complex model; third, it provides the organizational researcher with a tool that can be applied toward the development of new organizational theories; finally, the results generated herein attempt to extend contingency theory at the organizational level of analysis by identifying organizational characteristics significantly associated with longevity and identifying characteristics that are contingent on different environments for organizational survival. This work is organized as follows. First, we develop and integrate perspectives in the OT literature and simulation tools commonly used in the COT literature to formulate a set of algorithms that describe a multi-agent computer simulation. Second,
Putting the organization back into computational organization theory
87
code generation methodology is described and the validity of the model, through the analysis of the simulation data, is discussed. Third, we illustrate the usefulness of our simulation by developing contingency-oriented hypotheses, and finally we conclude with some final remarks as to how the simulation might be extended to study currently intractable or problematic areas of research.
2 Literature Our strategy here is to investigate the literature in order to arrive at a basic set of rules for our simulation. The intent here is to develop rules that are simple and unambiguous, yet clearly reflect the intent of each perspective. This approach is consistent with Epstein and Axtell (1996) seminal work, and like Epstein and Axtell (1996) work, the algorithm has the benefit of face validity (Anastasi 1988). 2.1 An industrial-organization context In order to provide context at the industry level of analysis, Porter’s (1980) description of industry characteristics was applied to the general framework for our simulation. In this classic work that champions the strategic choice perspective, Porter applied the industrial-organization economics perspective toward the development of analysis tools for corporate executives. He saw industries as consisting of groups of organizations along industry-specific relevant critical dimensions (cost position, product quality, technology, leverage, vertical integration, service, etc.). Thus, organizations could be “mapped,” as if on a landscape, along an industry’s two most important dimensions (in our simulation, these “maps” resemble “chessboards” of 22 squares to a side; on each square can reside one and only one agent, and on each square exists a defined amount of resources). Once mapped, an organization’s executives could position their firms within a high-profitability strategic group (or at least to begin to overcome barriers to entry to the group), or find an unoccupied area on the strategic landscape (a niche) to inhabit that could be relatively free from the potentially detrimental reactions of competitors (i.e., neighbors on the landscape). We model two additional industrial-organizational concepts. First, we modeled a basic three-member vertical supply chain, with harvesting-type agents that collect raw materials/resources directly from the landscape, manufacturing-type agents that purchase supplies from harvesters and transform the supplies to finished goods at some cost, and retailers that buy finished goods from manufacturers and sell to a simulated demand function. Second, we assume the production and sale of a single product that is commoditized in every way except for the two landscape dimensions. This assumption allows us to assume that production and operations are standardized and well-known in such a way that we expect choice of operations strategy (Boyer et al. 2005) to make an insignificant impact on cost variation among simulated organizations. We apply Porter’s theory to a simulation of organizations because it is able to accommodate transaction cost economics, resource dependence (since “resources” could be identified as a critical dimension for any industry), and contingency theory
88
B.W. Kulik, T. Baker
at least in terms of strategy formulation (strategies are based on a fit between industry conditions and the organization’s strengths) and uncertainty (which serves as an entry barrier if high). In addition, Aldrich (1979) incorporated the strategic choice perspective in his modified population ecological perspective, and the idea of ‘adaptation by choice’ has been addressed by COT studies (Carley 1995, 2002; Carley and Svoboda 1996; Handley and Levis 2001; Lin and Hui 1997; Mild and Taudes 2007; Perdu and Levis 1998; Turner et al. 2002). Since Porter’s perspective appears to be both so pervasive and accommodating, it was adopted for the framework of our simulation of organizations: a multi-agent simulation, with each firm represented by an agent (simulating the characteristics of a top management team) and located on a two-dimensional landscape (simulating an industry), not unlike Epstein and Axtell’s (1996) individual-level landscape-based simulation of human societies, where agents were allowed to “move” incrementally across their landscapes in each simulation iteration. Our sugarscape-type model is consistent with the industrial-organization perspective because agents in our simulation move away from (known) nearby competitors by calculating a niche vector and selecting a vacant location in that direction. Furthermore, within the same strategic choice perspective, Miles and Snow (1978), in a view consistent with Bettis and Prahalad (1995) dominant logic and Hambrick and Mason’s (1984) upper echelons theory, saw firms as harboring certain distinct characteristics that differed from other firms in the same industry. They identified four characteristics in their work (defenders, prospectors, analyzers, and reactors), but left open the exact number of firm types. To model this insight, each agent was randomly assigned values that were associated with magnitudes of certain characteristics. We must reiterate here that our simulation model is not a study of strategic management per se; it merely uses the strategic management/industrial-organization paradigm as an overall framework in order to make our simulation more realistic. The alternative would be to ignore that industry members are aligned into supply chains, that organizations make strategic choices that are relevant and relative to their competitors, and that organizations make choices that are aligned with their internal characteristics. The alternative would therefore ignore reality. However, these prescriptions for a realistic simulation are not explicitly identified by the OT literature, where a single organization is the typical focus of analysis. We note that the present study is no closer to the study of strategic management than Epstein and Axtell’s (1996) sugarscape model was to the study of sugar farming, although we admit below in our discussion that the presently developed architecture could be adapted for the pursuit of strategic management questions. Our focus in this study remains on the organization, not the industry, with the integration and validation of multiple OT perspectives (discussed next) for the purpose of study in the fields of OT and COT. 2.2 OT perspectives At a minimum, any reasonably comprehensive simulation of organizations must be constructed from insights made with regard to how organizations have been observed to operate at the organizational level of analysis. In this section, we briefly explore the perspectives as outlined by Perrow (1986) with an interest in how each might provide simulation guidelines. Table 1 lists the claims of these perspectives, with
Putting the organization back into computational organization theory
89
Table 1 Eight fundamental organizational perspectives Theory/approach
Description/claim
Relevant citations
Strategic choice
Though constrained, the variance of managerial choice across organizations is nonetheless considerable. Managers choose strategies that guide their organizations’ behavior during competition with other organizations.
Porter 1980; Miles and Snow 1978; Aldrich and Pfeffer 1976; Child 1975; Child 1972; Ansoff 1965
Contingency theory
An organization’s internal configuration (structure) is “contingent” on (i.e. correlated with or caused by, depending on the particular author) conditions of environmental uncertainty.
Venkatraman 1989; Aldrich and Pfeffer 1976; Staw and Szwajkowski 1975; Jurkovich 1974; Tosi et al. 1973; Duncan 1972; Hickson et al. 1971; Lawrence and Lorsch 1967; Thompson 1967; Emery and Trist 1965; Burns and Stalker 1961; Woodward 1958; Dill 1958
Behavioral decision theory
Managerial choice is limited due to bounded rationality which results in decisions based on feasibility rather than optimality, past experience in making similar decisions, and routines that develop to solve recurring problems.
Cyert and March 1963; March and Simon 1958; Simon 1945
Enactment
Managers scanning the environment are biased, so these scans result in self-confirmation or pre-conceived notions for the organization; meanwhile, the environment is influenced by the organization’s attentional biases.
Weick 1979
Resource dependence
Organizations depend mostly on important, scarce resources to survive. Therefore, much attention in organizations is given to the management of these resources, resulting in firm behavior that would otherwise appear irrational.
Pfeffer and Salancik 1978; Aldrich and Pfeffer 1976
Institutional theory
Pressures from government, society, professional organizations, and other organizations in the task environment constrain managerial choice in ways that are both irrational (institutionalization) and rational (neo-institutionalization).
DiMaggio and Powell 1983; Hirsch 1975; Blau and Meyer 1971; Selznick 1957; Parsons 1956; Blau 1956
Population ecology
Organizational environments are analogous to biological systems in that variation, selection, and retention mechanisms determine population composition and firm survival, thus limiting the relevance of managerial choice.
Aldrich 1979; Hannan and Freeman 1977; Aldrich and Pfeffer 1976; Alchain 1950; Boulding 1950
Transaction cost economics
The efficiency of transactions between organizations determines an organization’s size in such a way that an size is always the result of the most efficient configuration of transactions.
Perrow 1986; Williamson 1975
90
B.W. Kulik, T. Baker
relevant literature cited, while Tables 2 (classic literature) and 3 (current literature) show how specific citations were converted into prescriptions toward a rule set for our simulation. While space constrains a full discussion of Tables 2 and 3 here, a far more detailed discussion of this literature may be found in Kulik (2006). Contingency theory claims that there is no best way for organizations to organize; rather, organizations depend on environmental conditions (Burns and Stalker 1961; Lawrence and Lorsch 1967; Thompson 1967). A literature review of what those “conditions” might be is summarized in Table 4, where three environmental dimensions were identified: munificence, dynamism, and complexity. Our simulation therefore included 27 different simulated industries, each environment a different combination of low, medium and high values for each dimension. Furthermore, since the path to success depends on environmental conditions, it stands to reason that different organizational characteristics will be successful in different environments. For example, Burns and Stalker (1961) described organization-environment interaction by categorizing two rational responses that organizations develop toward two different types of environments. In the first type of environment, characterized by stability, individuals within the organization develop a mechanistic management system, characterized by a bureaucratized, functional structure, a focus on the “technical improvement of means,” (p. 119), the precise definition of work tasks, vertical communication rather than horizontal, the assumption of omniscience imputed to the head of the organization, and other individual-level behavior. In the second type of environment, characterized by instability, individuals develop an organic management system as characterized by a network structure of control, the spread of commitment beyond technical concerns, imprecise definition of work tasks, horizontal communication rather than vertical, the absence of the assumption of omniscience imputed to the leader, etc. Furthermore, these two organizational forms represent extreme ends of a continuous scale, and any organization may contain elements of both types. Their point was to emphasize that there is no “one best way” of organizing, but in some situations a mechanistic management system may be appropriate (i.e. for a stable environment). Organization-environment fit was seen as a determinant of the specific level of uncertainty in a specific environment. Our approach here is to generalize this perspective by creating characteristics for organizations as suggested by the other perspectives, and then to test whether these characteristics differ between those agents that are successful (i.e., are profitable and survive) and those that are not. Therefore, Burns and Stalker’s (1961) list of characteristics is not explicitly modeled, but their contingency idea is applied to other organizational characteristics as developed below and listed in Table 4. Behavioral decision theory claims that managerial decisions are rational, but limited and adaptive (Simon 1945; March and Simon 1958; Cyert and March 1963), and has been the foundation of numerous COT studies (Carley 1995; Huberman 2001; Malyankar and Findler 1998; Rakotobe-Joel et al. 2002). This view suggests that agents should be limited in their knowledge of the environment (i.e., search is limited to the agent’s immediate vicinity), and that decisions should be made relatively simply. This perspective outlines the decision process for buying and selling: a buyer will choose to buy supplies from a list of known sellers based on one of several simple preferences; thus, sellers are sorted a certain way (based on selling price, status,
Putting the organization back into computational organization theory
91
Table 2 Simulation guidelines derived from extant OT literature Citation(s)
Topic
Conclusions
Simulation guideline
Simon 1945; March and Simon 1958; Cyert and March 1963 Blau 1956
Bounded and Adaptive Rationality
Mgrs are limited in their ability to scan the environment; Managers make decisions sequentially
Agent scans only adjacent cells and makes decisions of “fit” based on info from that scan
Powerful vs. Powerless
Powerful orgs influence their environ’s; powerless orgs are influenced
Selznick 1957
Role Taking
Orgs take on roles in order to “fit” in society
March and Simon 1958
Organization Studies
Cyert and March 1963
Environmental Munificence
Org studies can be split into org-env interaction and internal org’n studies Orgs shore up resources during periods of relative munificence
Highest-profitable and highest-market share agents are copied by troubled and small firms Agents move around a landscape and transact with nearby agents Simulate organizations as autonomous agents
Computer Simulation Burns and Stalker 1961; Thompson 1967 Thompson 1967 Lawrence and Lorsch 1967
Contingency Theory
Firm behavior can be translated into simple decision rules Firms are successful by different means, depending on environmental conditions
Environments
Focused on task environments
Environmental Uncertainty
Magnitude of difference of uncertainty between subunits
Weick 1979
Environmental Enactment
Hickson et al. 1971
Subunit Power and Uncertainty Perceived Uncertainty
Mgrs are influenced by their biases when influencing, then making sense of, their environ’s Subunits which have more power are those that process the most uncertainty Uncertainty variables operate under 2 diff. dimensions: perceived and actual Perceived uncertainty is reasonably homogeneous within subunits In high-mun.-env., firms concentrate on perf.; low munif. ⇒ survival
Duncan 1972
Environmental Munificence
Agents measure their munificence and “save” resources accordingly Computer simulation of firm behavior may be simplified without loss of generality Discrete environ’s can be simulated by varying degrees of (multidimen’l) uncertainty A baseline simulation of task environments is sufficient Assign a coordination cost to each agent when input and outcome uncertainty differences observed Local scanning of information biased by agent’s profile/background Give priorities to some decisions over others to simulate power asymmetry Keep track of and measure both perceived and actual uncertainty Perceived uncertainty of agent can represent subunit uncertainty Success measure contingent on level of perceived munificence
92
B.W. Kulik, T. Baker
Table 2 (continued) Citation(s)
Topic
Conclusions
Simulation guideline
Staw and Szwajkowski 1975
Environmental Munificence
Scarce environment: survival may be more important than performance
Measure both survival rates and performance
Williamson 1975
TCE
Conditions of bounded rationality, small numbers, environmental uncertainty, vert. integration
These conditions must exist simultaneously in order for firms to grow beyond single unit
Aldrich and Pfeffer 1976
Nat. Selection and Resource Dependence
Incorporate planned variation and possibility of organizations influencing their environments
Planned variation modeled by randomization of profile at initialization and each iteration; successful org’s influence others
Empirical Studies
Should be longitudinal to account for processes in nat. selection and res. dep
Time should be modeled with simulation iterations
Levels of Analysis
Within-organization decision level and aggregate industry level measures
Incorporate aggregate measures for the study of industries
Penrose 1952; Starbuck 1976
Organizational Boundaries
Must be clearly defined
Clearly define organizational and environmental boundaries
Miles and Snow 1978
Fit strategies
Firms make decisions based on a profile of characteristics
Program agents with a wide variety of characteristics at initialization and observe outcomes
Pfeffer and Salancik 1978
Environmental Uncertainty
The more organizations in the task environment, the more uncertainty
Environmental density is an element of env’l uncertainty
The more connected task env. members are, the more the uncertainty
Network tie density is an element of env’l uncertainty
Decision Making
Decisions about present and future demands are based on past data
Resource, demand, and production needs are based on calculations of historical data
Contingency Theory
Nonmonotonic contingencies uncovered
Suggests a new method of hypothesis generation for org’s which are complex systems
Population Ecology
Selection need not be accidental
Constrained choice: limited landscape movement and transact only with nearby agents
Environment: 5 Dimensions
Degrees of homogeneity, stability, concentration, domain consensus, turbulence
Domain consensus: one agent per spot; landscape resource gatherers may/may not access resources from adjacent spots
Fine-grained/ Coarsegrained
Many short-term changes in fine-grained environment
Additional environ’l variable: landscape reset frequency
Schoonhoven 1981; Aldrich 1979
Putting the organization back into computational organization theory
93
Table 2 (continued) Citation(s)
Topic
Conclusions
Simulation guideline
Randolph and Dess 1984; Porter 1980; Dess and Beard 1984; Bourgeois 1985
Strategic Process Uncertainty: 3 Dimensions
Strategic and minimum strategic processes identified Munificence, complexity and market dynamism (empirically validated) Successful managers’ perceptions of uncertainty close to actual
Basic and extended decision processes for simulated agents Provides a way to model and measure environ’l uncertainty
Tushman and Romanelli 1985
Punctuated Equilibrium
Prahalad and Bettis 1986; Bettis and Prahalad 1995
Dominant Logic
Org’s encounter long periods of stability punctuated by short periods of instability and change Organizations make decisions based on a narrow conceptualization of the world which resists change
Agents are modeled with profiles that enable them to make decisions in certain ways; re-creation and reorganization represented by randomized replacement of profile
Hambrick and Mason 1984; Jensen and Zajac 2004
Upper Echelons Theory
A CEO’s life experience creates bias in taking action and making decisions
Agents have simulated CEOs with “background” in either buying, production, or sales
Blau 1956; DiMaggio and Powell 1983
Institutional Theory
Coercive, mimetic, and normative isomorphism
Agents replace elements of their profiles with those of nearby and best agents
Perceptual Acuity
Need to measure both and test hypothesis in simulation
or willingness to set up a long-term supply relationship, as discussed below and indicated as BUYSELLPREF in Table 5), and the buyer selects the seller at the top of the sorted list. However, the list of potential sellers only comprises agents that are located near the agent that the buying agent has actually identified with a search process. Also, since managers are boundedly rational in that they cannot know future events with certainty, agents average the previous two periods of demand to determine the supplies needed for the next production period. Finally, since no agent holds a perfect knowledge of its environment, it moves across its landscape (i.e., selects a strategic position) based on its scan of its vicinity, with more extensive scans costing more in currency and the extent to which an agent can move across its landscape. Enactment, as described by Weick (1979), claims that individuals learn about their environments along pre-conceived biases, and in so doing influence other industry members to adopt the same biases. An example might be Lee Iacocca who, likely because of his background as salesman at Ford, saw marketing as Chrysler’s basic solution to its problems when he became CEO of the company in the 1980s. Chrysler’s subsequent turnaround and success under Iacocca’s leadership in turn influenced other industry members to also focus on their own marketing efforts. For our simulation, enactment suggests that agents in our simulation should make decisions based on biased notions of the actual (simulated) environment. Therefore, when funds are scarce, agents with a bias toward sales (modeled in the simulation as money spent on searching the environment) satisfy the demands of the sales department (i.e.
94
B.W. Kulik, T. Baker
Table 3 Simulation guidelines derived from recent OT literature Perspective
Citation(s)
Topic
Simulation guideline
Institutional Theory
Deephouse 1996
Legitimacy
Legitimacy points are added and subtracted based on performance and whether copied by others
Suchman 1995
Dimensions of legitimacy
Useful iteration range of study will be an intermediate death rate period
Four organizational archetypes
Follow one archetype identified and model commodity producers
Behavioral Decision Theory
Enactment
Contingency Theory
Oliver 1992
Deinstitutionalization Organizations change by shifting from institution to institution over time
Oliver 1997
Institutionalization can increase organizational performance
Rational decision rule for buyers can be legitimacy-based, in addition to competition-based and supply-chain efficiency based
Dacin et al. 2002
Institutional change
Mimetic isomorphism is consistent: copy profile elements from the same agent until fully isomorphic
Washington and Zajac 2005
Status
Status points determined by resource possession and contagion
Ireland and Miller 2004; Nutt 2004; Ketchen et al. 2004
Decision search ∼ speed
Four alternative stopping rules for agent search; transaction decisions are sequenced in tiers according to speed; extent of search is expensed
Shimizu and Hitt 2004
Strategic flexibility
Profile changes are reversed if threshold poor performance observed
Ketchen et al. 2004
Strategic positioning Response time to trespassing treated as a random variable; agent movement constrained by nearness to highly competitive agents
Janney and Dess 2004
Real Options
Before entering, agents purchase the right to enter an unoccupied area
Sadler-Smith and Shefy 2004
Intuition
Use of blacklists and whitelists; copy movement of other identified agents in complex environments according to a variable threshold amount
Danneels 2003
Creeping commitment
Limit agents’ movement across landscape; option to not move at all, accompanied by entrenchment
Gibson and Birkinshaw 2004; He and Wong 2004
Ambidexterity
Rate of entrenchment as a random variable in agent’s profile
Aragon-Correa and Sharma 2003
Environmental variables as mediators
Adopt this perspective rather than strategic-management (moderators) perspective
Hough and White 2003
Scanning pervasiveness ∼ decision quality
Movement & transaction decisions based only on regions of env actually scanned
Putting the organization back into computational organization theory
95
Table 3 (continued) Perspective
Citation(s)
Topic
Simulation guideline
Hough and White 2004
Pervasivenessenviron’l dynamism
Baseline simulation validation tools
Evolutionary inefficiency
Simulated agents programmed for inefficiency
Levinthal 1994
Survival in Schumpeterian environments
Financial resources: vary failure time to death; learning capacity; vary elements retained in black/whitelists and move in and out by FILO
Mezias and Lant 1994; Usher and Evans 1996; Bruderer and Singh 1996
Ecological, Model as simultaneous coexistence institutional and learning mechanisms
Organizational March (1994) Ecology
Table 4 Environmental dimensions after Dess and Beard (1984) and Randolph and Dess (1984) Dimension
Variables
Definition
Supporting citations
Quoted from Randolph and Dess 1984, p. 121 Munificence
Growth in: Sales, price-cost margin, total empl., value added, # of establish’s
The extent to which the industry can support present organizations, enable the present organizations to grow and prosper, and enable new organizations to gain entrance into the industry
Aldrich 1979; Hirsch 1975; Child 1975; Staw and Szwajkowski 1975; Scherer 1971
Dynamism
Specialization ratio; Instability in: Sales, value added, total employment, # of establishments
The degree of change that characterizes environmental activities relevant to an organization’s operations
Tung 1979; Child 1974; Tosi et al. 1973; Child 1972; Duncan 1972; Thompson 1967
Complexity
Geographical concentration: Sales, value added, total employment, # of establishments
3 major components: 1. frequency of changes in relevant environmental activities 2. degree of differences 3. degree of irregularity in the overall pattern of change; the variability of change The heterogeneity of and range of environmental activities that are relevant to an organization’s activities.
Tung 1979; Jurkovich 1974; Child 1972; Duncan 1972; Thompson 1967; Dill 1958
the search function) before distributing money to buying and manufacturing departments. Alternatively, those agents with a bias toward buying fully meet the needs of the buying department, and those agents with a bias toward manufacturing fully meet
96
B.W. Kulik, T. Baker
the needs of that function, before funds are allocated toward the needs of the other two functions. This bias is indicated as variable PROPBIAS in Table 5, and its change frequency is BIASCHANGE, also described in Table 5. The resource dependence perspective posits that surviving organizations are those that are able to obtain the necessary scarce resources to enable it to function. In the simulation, each occupiable space of the landscape is given a resource value, and like Epstein and Axtell (1996), resources vary across the landscape resembling peaks and valleys of resources. To model resource dependence, agents in the simulation are allowed to move toward landscape areas of higher resources based on resource values identified during its search function. Therefore, agents are dependent on one another, as in Conte and Sichman (2002). While only harvesters can directly take advantage of the landscape’s resources, it is expected that any type of agent should benefit by moving toward higher resources through the setup of supply chains with those relatively wealthy harvesters. To allow for this characteristic to co-exist with the strategic management perspective, the variable AVERSION was created (see Table 5), where if AVERSION = 0, the agent moves away from competitors in support of the strategic management perspective, and if AVERSION = 1, the agent moves toward higher known resources in support of the resource dependence perspective. At simulation initialization, agents are randomly assigned a 0 or 1 value of this variable with an equal probability. Note that this is a key modeling strategy: model countervailing perspectives by randomly and evenly assigning countervailing characteristics across all agents. This strategy was also adopted by Naveh and Sun (2006) in their CLARION-based model of cooperative and competitive aspects of the advancement of scientific fields. In this study, it is presumed that as long as there are not too many countervailing perspectives (there are never more than three), the effects of both/all will be strongly detected in subsequent regression analyses. Institutional theory attempts to explain why there are similarities across organizations. Essentially, organizations in the same environment copy each other either because they are forced to (government regulation) or to gain legitimacy from the perspective of other same-environment members by copying more legitimate members or by normative pressures (DiMaggio and Powell 1983). Under this perspective, the most legitimate organizations are expected to survive the longest. In our simulation, mimetic isomorphism was modeled by first identifying the five highest-profit agents in an environment, and then allowing each agent in the simulation to copy a characteristic from one of these five most-profitable agents (chosen at random)—a behavior not unlike Teitelbaum and Dowlatabadi’s (2000) agents. To accommodate this activity, the agent characteristic STATUS was created. Agents could increase their STATUS by occupying a high-resource spot on the landscape, by becoming profitable, and by transacting with other agents of higher STATUS. Conversely, agents lost STATUS by moving to lower-resource locations, realizing losses, and by transacting with other agents of lower STATUS. Furthermore, as noted above, some buyers were programmed through the agent characteristic BUYSELLPREF to prefer to buy from highest-status sellers (with ties decided on lowest price). Agents with high STATUS are not only copied, but they receive real benefits: some agents prefer to buy from highest-STATUS agents regardless of selling price (when BUYSELLPREF = 2, as described in Table 5), and higher-status agents move across the landscape before lower-status agents.
Putting the organization back into computational organization theory
97
Table 5 List of variables used in validation study Variable
Description
Initial condition
PROFIT
Total amount of capital accumulated by an agent.
0
Longevity
Total life of the agent. If Longevity = 50, then the agent has survived until the end of the simulation.
0
STATUS
Status at time of death or at the last iteration. Agents with higher status are given priority in interaction and movement.
U (0, 5)
DIE
Number of contiguous periods of losses before the agent dies. It is generally a characteristic of the tolerance for poor performance.
0
TYPE
Agent is harvester (TYPE = 0), manufacturer (TYPE = 1), retailer (TYPE = 2).
DU (0, 2)
BUYSELLPREF
Agents choose which suppliers to prefer to buy from, sorting by supply-chain orientation (BUYSELLPREF = 0), lowest-price (BUYSELLPREF = 1) or highest STATUS (BUYSELLPREF = 2).
DU (0, 2)
AGENTMEM
Number of agents that an agent remembers; i.e. the length of an agent’s whitelist, blacklist, INTERACTWITH list, and on an agent’s map.
DU (8, 32)
PROPBIAS
CEO’s background bias: buying = 0; production = 1; selling = 2. Used as a sorting rule for the disbursement of capital needs. The CEO will satisfy capital needs from his/her own background first.
DU (0, 2)
BIASCHANGE
A number between 0 and 1. If closer to 1.0, the CEO’s background bias will change more often, thus representing a more even-handed (& less biased) CEO with regard to the disbursement of capital needs.
U (0, 1)
Markup
Amount that the agent marks up its price for sale after transforming the unit to a sellable form: sellprice = buyprice + (2 × Markup).
U (1.5, 10)
ReactTime*Mag *Prob
Reaction time before an agent attacks a competitor times the magnitude of the attack times the likelihood of attack per iteration. The higher this score, the more aggressive toward competitors the agent is expected to be.
DU(1, 5)∗ U (1.1, 10)∗ U (0, 1)
TERRITSETUP
If this value is nonzero, it indicates that the agent has set up a stronghold with number equal to the size of the territory.
0
AVERSION
If AVERSION = 0, the agent moves away from known competitors; if AVERSION = 1, the agent moves toward higher known resources.
DU(0, 1)
MOVE adj
Same as MOVE, but with 0 and 1 switched: when MOVEadj = 0, the agent does not move and tries to set up a stronghold; when MOVEadj = 1, the agent moves a maximum of one square per iteration; when MOVEadj = 2, the agent moves a maximum of two squares per iteration.
DU(0, 1, 2)
Change
Probability of change given that # loss periods > THRESHREORG.
U (0, 1)
SEARCHLAND
Number of squares searched per iteration, if sufficient capital allocated.
DU(8, 32)
THRESHFLEX
Flexibility threshold: after interaction with agent, if profits > THRESHFLEX, place interacting agents on WHITELIST; if losses are > THRESHFLEX, place interacting agents on BLACKLIST.
U (0.1, 8.0)
Note: DU(x, y) = discrete uniform distribution; U (x, y) = continuous uniform distribution, where x = minimum and y = maximum
98
B.W. Kulik, T. Baker
Fig. 1 Decision hierarchy as derived from Tushman and Romanelli (1985) and Prahalad and Bettis (1986)
Population ecology argues that, as environments change, some organizations are selected to survive while others die off, as long as there is sufficient variation across the population and the birth and death rates are sufficiently high (Aldrich 1979). In the simulation, agent “birth” occurs at program initialization, while agents are allowed to die if they do show profitability for a variable number of contiguous iterations (for relevant COT studies, see Vermeulen and Bruggeman 2001; Bruggeman and Nualláin 2000, and Rakotobe-Joel et al. 2002). The agent characteristic DIE in Table 5 models this behavior, while the characteristic Longevity records when and if an agent has died during the simulation run. Unlike Naveh and Sun’s (2006) agents, those here are not replaced after they die, and there are no births after the program has initialized. Also, to maximize variation across agents, values for agent characteristics (see Table 5) were assigned with the aid of a uniform distribution (for example, an agent had an equal chance of having its PROPBIAS value set to 0, 1, or 2). Transaction cost economics provides a justification for vertical integration (a diversification strategy) when transaction costs across a supply chain are high. For independent supply-chain members, it also implies that the more transactions that are made between the same two entities, the lower the transaction cost will be. Thus, there should be a transaction cost discount based on the number or prior transactions between the same pair of entities. This discount is modeled explicitly in the simulation. We also allow agents seeking low transaction costs to find each other: when BUYSELLPREF = 0 (supply-chain preference), buyers prefer to buy from sellers with the same BUYSELLPREF setting of 0 (then sorted by selling price). Ideas on organizational change were also modeled in the simulation. Tushman and Romanelli (1985) and Prahalad and Bettis (1986), among others, identified a decision process by which either no change, incremental change, or “recreation” or transformational change, limited by organizational intertia, was made as a result of the process, similar to an internal decision choice of incremental or incremental innovations allowed by Teitelbaum and Dowlatabadi’s (2000) agents, although in our simulation, agents were not allowed to choose a mixture of the two. A summary of this process is shown in Fig. 1. This decision process was modeled in our simulation by allowing agents to choose no change, incremental change, or transformation, based on each agent’s (randomly assigned) characteristic of tolerance for contiguous iterations of (financial) loss,
Putting the organization back into computational organization theory
99
where a transformational change subjected the agent to a re-initialization of all of the agent’s randomly assigned characteristic values. Furthermore, inertia (or “creeping commitment” as termed by (Danneels 2003) under the environmental enactment perspective) was modeled by restricting a firm from moving across its landscape (which models an organization’s competitive repositioning) to only two spaces per iteration. Ideas on institutional change were also modeled. Editors for the special issue of the Academy of Management Review on institutional change (Dacin et al. 2002) observed that organizations do not change incrementally but from institution to institution; this observation was incorporated by first allowing for firms to copy characteristics of the most profitable agents in their landscapes, and then once the agent begins copying that successful agent’s characteristics, to “lock in” the agent’s copying activity until all characteristics of that agent has been copied (at a rate of one characteristic per simulation iteration) before allowing the agent to copy from another successful agents. In conclusion, our review of the literature has provided a surprisingly narrow and detailed direction toward how a simulation of organizations might be carried out. Basically, the OT (and industrial-organization) literature argues for the representation of organizations as supply-chain connected, autonomous agents, each having unique characteristics, which move and otherwise make decisions based on a scan of nearby organizations on a resource-variable landscape. The status of each agent is based on each agent’s success (as defined by its profitability) which controls the order of movement and agent transactions, and can be used by agents to prefer transacting with higher-status agents rather than lower-status agents. Also, agents have the option of developing long-term “relationships” with other agents, realizing transaction-cost efficiencies in the process. 2.3 COT tools and Bonini’s paradox In addition to the above literature review, a review of the COT literature at the organizational level of analysis was conducted in an effort to apply these tools to the current study. A small number of these articles, included for illustrative purposes, along with tools that each was identified to have employed, is shown in Table 6. The most widespread of these were the tools of simplification (deliberate reduction of the known real-world situation) and limited scope (design the model to address only one or two questions in a single study); we intend to apply 7 of the 8 tools identified. Some insight into the importance of simulation as a methodology and the issues involved can be gained from a consideration of how Dutton and Starbuck (1971a) regarded simulation as a research tool. In the introduction to their book, editors Dutton and Starbuck (1971b) outlined a number of advantages and criticisms of simulation. On the positive side, simulation had, by 1965, become a “widely used methodology” in “all of the social sciences” (p. 3), with advantages including intelligible results compared to mathematical modeling, at least a modest degree of logical rigor, and the freedom of repeatable experimentation using variation impossible in the real world. On the negative side, it was observed that it may not be possible to replicate human thought or the “complex essence” of human beings by using computers, and even if such a thing were ever approached
100
B.W. Kulik, T. Baker
with a computer simulation, it would be nevertheless the too complex to understand the causal relationships that led to the simulation’s particular output. The authors termed the extreme case of this negative aspect of simulation “Bonini’s paradox” (p. 4) after Bonini (1963), who concluded after his simulation of a firm: “We cannot explain completely the reasons why the firm behaves in a specific fashion. Our model of the firm is highly complex, and it is not possible to trace out the behavior pattern throughout the firm. . . . Therefore, we cannot pinpoint the explicit causal mechanism in the model” (Bonini 1963, p. 136). It appeared evident to Dutton and Starbuck (1971b) that, since Bonini (1963) made this conclusion after his effort to analyze the output of a simulation of a firm, such a subject for simulation can be an ipso facto reason for avoiding the application of simulation toward the inner workings of a firm or organization. A critical question for Dutton and Starbuck (1971b, p. 5) was, “How complex must a model be to portray the behavior?” Simplifying techniques such as linear programming could be used to model firm activity in some applications, but the behaviors of interest in this and other COT studies must be at a sufficiently superficial level in order to avoid Bonini’s paradox. In this study, we expect to overcome Bonini’s paradox by modeling each OT perspective simply rather than completely, by analyzing results as one might analyse the results of a real population of organizations with regression and difference test statistical methodologies, and by using the simulation to generate hypotheses about the real world that should subsequently be tested by real-world studies (Carley 1999). Other tools listed in Table 6 can also be applied for the purpose of avoiding Bonini’s paradox. For example, the application of pseudorandom numbers in the randomization of landscape and agent characteristics, where random numbers are generated based on a defined seed number so that every repeat of the same simulation yields exactly the same results, can be used to set up a simulation experiment: run a baseline simulation (or a “canonical representation” after Levinthal 2000), and then an extension, with the same random number seed; any observed differences in results must therefore be the result of changes between baseline and extension. In this way, causation can be inferred, at least between baseline and extension. With regard to causal inference within one simulation run, the tool of “survival and evolution” may be applied, wherein successful and unsuccessful agents can be identified by allowing for agents to “die” during the simulation after a certain number of sufficient iterations. Undesirable characteristics can then be associated with early dyers with characteristics held in common; similarly, desirable or successful characteristics can be identified among the survivors.
3 Methodology The results of our literature reviews, summarized in Tables 1 through 6 and discussed above, resulted in the creation of a detailed set of algorithms that describe in reasonable detail how an OT-based simulation might be constructed. The following is a summary algorithm:
Prietula and Watson (2000)
Sastry (2000)
Malerba et al. (2000)
Levinthal (2000)
Miller (2000)
Lomi and Larsen (2000)
Krackhardt (2000)
Barron (2000)
Bothner and White (2000)
Prietula (2000)
Loch et al. (2000)
Macy and Strang (2000)
Carley and Hill (2000)
Harrison and Carroll (2000)
Carley (2000)
Lin (1998)
Carley and Prietula (1998)
Axelrod (1980)
This Study
Article
×
× × × × × × × × × × × × × × × × × × × × ×
×
× ×
×
Hypothesis Development
Simplification
Simulation tools
Table 6 Comparison of COT simulation tools
×
Pseudorandom # control
× × × ×
× × × × × × × × × × × × × ×
Limited Scope
× ×
×
×
×
Survival & Evolution
×
×
×
×
×
Equilibrium
× ×
× × × ×
×
× ×
× ×
Extreme Conditions
×
×
Data Reproduction
Putting the organization back into computational organization theory 101
102
B.W. Kulik, T. Baker
1. Initialize the 18 landscapes with peaks and valleys of resources a. For munificence i = 1, 2, 3: i. Number of peaks = 5 · (i 2 ) ii. Maximum peak value = 5 · I ; minimum peak value = i b. For dynamism j = 1, 2: landscape reset frequency = (4 − j )3 c. For complexity k = 1, 2, 3: number of agents = 0.3 · 222 , 0.6 · 222 , and 0.9 · 222 , respectively d. Randomly locate peaks on each landscape and fill in fill in gradients between peaks and valleys 2. Initialize characteristic values for each agent (see Table 5); each agent begins with 15 units of capital 3. Internal activity a. Allocate capital to input, transformation, and output activities, meeting the expected needs of each function after the favored function is first satisfied (according to PROPBIAS) b. Randomly regenerate PROPBIAS if U (0, 1) < BIASCHANGE i. Number of peaks = 5 · (i 2 ) c. Conduct transformation activity (turning inputs to outputs) up to resource allocation 4. External activity a. Search number of squares in immediate environment up to SEARCHLAND squares, as affordable under output activities allocation b. Add newly discovered agents to known lists of suppliers and competitors c. In order of STATUS per landscape, buy from suppliers on order of BUYSELLPREF d. Adjust STATUS allocations based on resources on landscape spot occupied and transaction partners 5. Territory setup and retaliation a. If no competitor is located on an adjacent spot for 5 iterations in a row, set up a “territory”; MOVE = 1 b. If a competitor is located in another agent’s territory, agent “attacks” that agent with REACTMAG of available capital, if willing to do so (determined by NOTREACT and REACTPROB) c. Attacked agent loses REACTMAG2 of capital 6. Organizational transformation and death a. Calculate PROFIT for each agent i. If PROFIT > 0 1. adjust STATUS upward 2. if PROFIT > THRESHFLEX, place on WHITELIST up to AGENTMEM agents on list ii. If PROFIT < 0 1. adjust STATUS downward 2. if PROFIT < THRESHFLEX, place on BLACKLIST up to AGENTMEM agents on list b. If the agent has a loss and the number of consecutive losses > DIE, then DIED = the iteration number (= agent’s Longevity; if DIED = 0 by the final iteration 50, Longevity is set to 50)
Putting the organization back into computational organization theory
103
c. If DIED = 0, and the number of consecutive losses > THRESHREORG, then i. If CHANGEPREF = 0 and U (0, 1) < TRANSFINCREM, then reorganize incrementally by copying one agent characteristic from a top-5 status landscape member ii. If CHANGEPREF = 1 and U (0, 1) < TRANSFREORG, then reorganize by re-initializing all of the agent’s characteristics 7. Movement a. If MOVE = 0, or MOVE = 2 and SEARCHLAND > 15, move one square across the landscape i. If AVERSION = 1, move away from competitors ii. If AVERSION = 0, move toward higher resources b. If Move = 2 and SEARCHLAND < 15, move two squares across the landscape i. If AVERSION = 1, move away from competitors ii. If AVERSION = 0, move toward higher resources 8. Miscellaneous and collect data a. Collect data on total industry resources and number of agents per landscape b. Reset searched array and lists of suppliers and competitors c. Re-initialize landscape according to landscape reset frequency 9. Repeat #3 until the completion of 50 iterations The simulations were run using version 7.0.1 of S-Plus with code converted from a more detailed version of the above algorithm. The effort to reduce the considerable runtime of the simulations to 200 hours total included setting the landscape sizes to only 22 × 22, while only one repetition and 50 iterations were run for each simulation. However, as the relatively “small” landscape size resulted in the use of 484 spaces per landscape and the generation of 7,840 agents, it was nevertheless considered sufficient for the present exploratory study, as real-world studies have been conducted on the order of 2,000 companies or less. Furthermore, as the 7,840 agents are small enough in number, the entire population of 7,840 agents was used in the data analysis; no sampling was necessary. After each simulation was run for the specified 50 iterations, each variable was printed on-screen by S-Plus and then copied over to MS Excel for the appropriate graphs and ANOVA tests; S-Plus was used for multiple regression analysis and output (importing the data back from Excel) because of the restrictive 16-variable limit in Excel’s regression analysis tool; S-Plus was also used for the Wilcoxon rank sum test, while Excel was used for t-tests, both used in testing differences between groups. Finally, model assumptions were checked in MS Excel with the aid of some PHStat2 functions such as normal probability plots; the regression assumptions were checked from within the S-Plus regression function. We reiterate here that regression, and generally a treatment of the results as if they were real-world data, is appropriate here because the simulation outputs are highly complex, full of countervailing effects and complicated behavior. Because of the model’s high level of complexity, we run into Bonini’s paradox in that we cannot determine causes and effects, and we are left only with associations, as in the real world. Therefore, we apply statistical tools of analysis, such as regression, as, say,
104
B.W. Kulik, T. Baker
a sociological study might analyze census data. While we lose some sense of causality with this analysis, we still retain some detectability of cause-and-effect through the tool of pseudo-random number generation, where the baseline data (shown here) can be compared to some modified simulation run (shown in future studies) that asks some question, with each run, baseline and modified, is initialized using the same random number seed.
4 Results The agent characteristics that were recorded for each agent and used for analysis are listed and described in Table 5, with the convention that variables listed in all-capital letters were directly obtained from the simulation results, while title-case variables were modified somewhat in MS Excel before being subjected to analysis. Two main foci of interest here are (1) the strength of association between the first two variables, PROFIT and Longevity, and the rest; in other words, how well do the variables STATUS, DIE, TYPE, BUYSELLPREF, etc. predict PROFIT and Longevity; and (2) any identifiable differences in characteristics between the list of variables in Table 5 between profitable and unprofitable (and also between early-dead and survived) agents. A table of correlations is shown in Table 7. As desired, it indicates independence between each variable studied under regression, which suggests that multicollinearity was not significant for this data set. Only one correlation appeared that was greater than 0.40, that between “MOVE adj” and “TERRITSETUP”. Some correlation between these variables was expected, however, as TERRITSETUP is only nonzero when the agent has set up a territory, and in this case, MOVE adj is zero, by definition. It is interesting to note the low correlation between PROFIT and Longevity of 0.26. This low correlation suggests that those agents which made a high profit were not necessarily the agents that survived the longest—an important principle which institutional theory attempts to explain (survival is through the most legitimate, not the most profitable), as does population ecology (survival is decided by the environment, which selects only the “fittest” organizations). In the simulation, an agent must do more than merely generate a profit to survive; it must position itself as part of a profitable supply chain which has access to ample resources, achieve sufficiently high enough status in order to receive priority in choosing the best agents to transact with and in order to move to more desirable landscape locations before others do, replace its less-profitable characteristics with more profitable ones, cope with its biases, etc. Thus, the issues of profitability and legitimacy, and what cause them, are two largely separate issues both in real life (from the perspective of institutional theory) and in this simulation’s results. 4.1 Expected results: model validation Our construction of organizations as definable and discrete compared to the realworld situation (our justification for embarking on a computer simulation in the first place) lends to the model’s construct validity (Davis et al. 2007; Shadish et al. 2002).
−0.01
−0.03
0.00
0.05
0.02
−0.03
−0.07
DIE
BUYSELLPREF
AGENTMEM
PROPBIAS
−0.02
−0.09
0.00
0.10
−0.01
0.00
−0.04
ReactTime*Mag*Prob
0.03
−0.06
0.04
0.02
−0.05
0.03
MOVE adj
Change
SEARCHLAND
THRESHFLEX
0.01
−0.03
0.00
AVERSION
0.00
0.00 −0.01
0.00
−0.01
0.01
0.02
−0.03 −0.01 −0.02
0.00
0.01
0.02 −0.03 −0.01 −0.02 0.00
0.00
0.00
0.00
0.02
0.01
0.04
0.01
0.01
0.02 −0.02 0.01
0.02 0.02
0.01
0.00
0.00
1.00 −0.03
0.02
0.00
1.00
0.02
0.02
0.01
−0.01
0.01
0.03
1.00 1.00
0.01
0.01 −0.01
−0.02 −0.24
−0.10
1.00 0.01 −0.01
−0.01 −0.01
−0.47
0.02
1.00
0.00
0.02
1.00 −0.02
1.00 1.00
Markup React- TERRIT- AVER- MOVE- Change SEARCH- THRESHTime*- SETUP SION adj LAND FLEX Mag*Prob
0.01
0.00 0.00
0.01 −0.01
0.08 −0.01 −0.02
0.00
−0.01
1.00
0.04 −0.01 −0.02
0.03 −0.05
0.01
0.08
0.00 0.00
0.00
1.00 0.01
0.00 −0.01
0.11
TERRITSETUP
0.00
1.00
BUY- AGENT- PROP- BIASSELL- MEM BIAS CHANGE PREF
−0.01 −0.01 −0.01
−0.01 −0.01
Markup
0.17
1.00
DIE
0.01 −0.01
0.00
0.05
1.00
0.16
BIASCHANGE
0.33
0.33
0.12
STATUS
1.00
1.00
0.26
PROFIT
Longevity
PROFIT Longe- STAvity TUS
Table 7 Table of correlations
Putting the organization back into computational organization theory 105
106
B.W. Kulik, T. Baker
Our strategy for model validation is (1) to construct a simple rule set from OT literature and in so doing establish a degree of face validity, as discussed above, (2) conduct a degree of internal validity during code creation, and (3) check conceptual, emergent validity of our model by finding evidence of each perspective through statistical analysis of our results. We do not pursue external validity, arguably the strongest type of validity, as discussed by Davis et al. (2007) and Carley (1996) because, as Carley (1996) noted, external validity is a multi-year, teamwork effort that should be left to others who did not develop the model. In any case, internal validity appears to be the usual validation path in COT studies (Gilbert and Troitzsch 2002; Saoud and Mark 2007), with detection of emergent phenomena also conducted for the purpose of validity in this field (Naveh and Sun 2006). With regard to internal validity, we checked for errors in the landscape construction by printing out a landscape with peaks and valleys with and without agents, across multiple iterations to verify that every agent persisted, moved across the landscape with no more than one agent occupied any particular landscape location in any iteration, conducted transactions with other agents, realized gains and losses, and sometimes died. We also verified that an equilibrium state was reasonably achieved with the 50 iterations we used for the simulation duration by plotting curves of the number of agents left alive at the end of each iteration (the number of agents left alive over the final iterations of the simulation run was reasonably stable). Our conceptual, emergent validation effort consisted of finding emergent evidence of our OT perspectives in our statistical analyses. An outline of the validation strategy is shown in Table 8, and results of regressions and difference tests are shown in Tables 9, 10 and 11. Note that the analyses in Table 8 are designed to discover “emergent” (after Naveh and Sun 2006) evidence and are therefore not circular with the simulation model’s design. For example, while resource dependence was modeled by allowing agents to move toward areas of (expected) higher resources, validation of the resourcedependence perspective was made (in part) through the significant positive association of PROPBIAS = 0 with the dependent variables Longevity and PROFIT. The strategic choice perspective was generally verified by results in Table 8 in that the regression models involving PROFIT were significant: the choice of some characteristics apparently mattered with respect to profitability. Organizational ecology was verified by the significance of variables in Table 10. Institutional theory was strongly verified by the strong positive significance of STATUS in Tables 9, 10, and 11, and also by the significance of independent variables unrelated with profitability in Table 9 when Longevity was the dependent variable. Behavioral decision theory was verified by the maximization of THRESHFLEX (to a value of 0.1843) in Table 9 for survived agents (with PROFIT as the dependent variable) and the significance of SEARCHLAND in Table 10 where early dead agents exhibited a significantly higher average than those that survived. Enactment was verified by the significance of PROPBIAS in Table 9 (some biases were found to be less profitable than others) and the strong, positive significance of BIASCHANGE in Table 10 (less biased agents survived longer), while the negative sign of the coefficient of this variable in Table 10 verified the resource dependence perspective (agents which biased their allocation of capital toward the acquisition of resources tended to be more profitable). The contin-
Putting the organization back into computational organization theory
107
Table 8 Validation analysis Analysis method
Intended to confirm
Chart of number alive across iterations
Equilibrium state arrived at (number of iterations is sufficient); separation of complexity across simulations; Resource Dependence
Significant differences among variables in Table 5 between surviving agents and those agents that died within the first 30 iterations
Population Ecology
Regression Dependent
PROFIT
Strategic Choice
Variables:
Longevity
Institutional Theory
Independent
STATUS
Institutional Theory
Variables:
AGENTMEM, SearchLand, MOVE, THRESHFLEX
Behavioral Decision Theory; Strategic Choice
Choice
Strategic Choice and Institutional Theory
BIASCHANGE
Strategic Choice or Institutional Theory
Markup, ReactTime*Mag*Prob, AVERSION
Strategic Choice
PROPBIAS, BUYSELLPREF
Resource Dependence (when PROPBIAS = 0), Enactment
All Together
Significance of simulation model
Significant differences among variables in Table 5 between agents in least and most harsh landscapes
Contingency Theory
2-Way ANOVA on Landscapes Main Factor: Munificence
Compare End-of Simulation total resources
Main Factor: Dynamism
Compare End-of Simulation total resources
Main Factor: Complexity
Compare End-of Simulation total alive
Contingency Theory; Resource Dependence
Note: TCE efficiency is confirmed in the diversification extension.
gency theory perspective was verified with the significance of differences in population deaths between landscapes with different populations (results not shown), and by the finding of significantly different averages in variables shown in Table 11 (some agents’ characteristics were different in different environments). The transaction cost economics (TCE) perspective was not validated with the use of the results, but it was assumed that the TCE perspective was valid since agents were given a discount upon repeated transactions with the same agent in the computer code.
108
B.W. Kulik, T. Baker
Table 9 Results of regression analyses Variable
STATUS
PROFIT
Longevity
All agents
Survived
All agents
0.004***
0.001*
0.05***
DIE
0.02*
−0.04***
BUYSELLPREF
0.08†
0.12*
−0.12
AGENTMEM
0.01
0.00
−0.01
AGENTMEM2
0.82***
0.00
0.00
0.00
PROPBIAS
−0.25***
−0.21***
0.07
BIASCHANGE
−0.06
−0.17
2.69***
0.47***
0.49***
−0.03***
−0.04***
−0.27
ReactTime*Mag*Prob
0.02
0.03†
−0.02
ReactTime*Mag*Prob2
0.00
−0.001*
−0.00
−0.02*
−0.03***
Markup Markup2
TERRITSETUP
3.60***
0.21***
AVERSION
0.07
0.14
0.24
MOVE adj
0.07
0.09
0.32 1.41**
0.10
−0.02
−0.02
−0.02
SEARCHLAND2
0.00
0.00
−0.004†
THRESHFLEX
0.10
0.22**
−0.16
Change SEARCHLAND
THRESHFLEX2 Multiple R 2
−0.01 0.04***
−0.01† 0.03***
0.14
0.02 0.24***
† p < 0.05 (one-tailed test) * p < 0.05 (two-tailed test) ** p < 0.01 (two-tailed test) *** p < 0.001 (two-tailed test)
Finally, the landscapes were found to be distinct in our ANOVA analysis (not shown), with more agents dying under progressively harsh conditions. We encountered a problem with the dynamism dimension: under the condition of high dynamism, where landscapes were re-initialized with peaks and subsequent valleys were most frequently randomly re-located, relatively few deaths occurred (apparently, agents on high-resource locations could ‘wait out’ the period of low-resource occupation upon landscape re-initialization until a subsequent re-initialization returned the landscape location to higher resources). Rather than re-program the dynamism condition, we simply ignored the high-dynamism condition in our subsequent analysis. Thus, we analyzed 18 different landscapes (3 × 3 × 2) rather than the full 27 (3 × 3 × 3). 4.2 Contingency results Our above discussion of Burns and Stalker (1961) led us to ask: Are there agent characteristics that are common among surviving agents in most harsh environments
Putting the organization back into computational organization theory
109
Table 10 Summary of two-sample difference tests between early dead and survived agents Variable
Average, early dead
Average, survived
p-value of ANOVA test
(Died within 1st 30 iterations) 0.00E+00*
PROFIT
−0.511
STATUS
66.7
122.0
0.00E+00*
DIE
14.6
17.5
0.00E+00*
1.27
TYPE
0.582
1.56
0.00E+00*
BUYSELLPREF
0.985
0.984
0.8744**
AGENTMEM
20.31
19.88
0.0163*
PROPBIAS
0.970
0.970
0.9583**
BIASCHANGE
0.485
0.517
6.35E−06*
Markup
4.56
5.66
0.00E+00*
ReactTime*Mag*Prob
8.14
8.10
0.6833*
TERRITSETUP
1.34
2.56
0.0021*
AVERSION
0.500
0.484
0.2331**
MOVE adj
1.030
1.004
2.24E−08**
Change
0.546
0.579
0.0002*
SEARCHLAND THRESHFLEX
22.33 4.028
21.26 4.067
1.30E−07* 0.4802*
* p-value for Wilcoxon Rank Sum Test ** p-value for a Pearson χ 2 contingency table test
that are significantly different from survivors in least harsh environments? Results in Table 11 indicate that surviving agents in most harsh environments were significantly lower in agent characteristics STATUS, DIE and TERRITSETUP, but higher in movement across the landscape. Our above discussion of institutional theory and population ecology led us to ask: Which agent characteristics are positively associated with agent longevity and survival? Results in Tables 9 (with Longevity as the dependent variable) and 10 show the following variables as significantly higher for high-longevity agents in both tables: STATUS, DIE, BIASCHANGE, Markup, TERRITSETUP, and Change. Additionally, the difference test for PROFIT shown in Table 10 was significantly higher among survived agents. TYPE was also found to be higher among survived agents, but this finding was regarded as an artifact of the simulation code, as it suggested that less harvesters were needed to support more retailers. 5 Discussion It has been the intent until now to carefully refer to any observed results as restricted to the findings in the simulation, but the purpose of the simulation was to be able to make predictions about real-world contingencies. The caution with approaching the real world is emphasized in this discussion by formulating hypotheses about the real world; thus the assumption is made that any findings in the simulation must first be verified with real-world data before it can be considered a contribution to the advancement of the field of OT. However, because of the complexity of the simulation,
110
B.W. Kulik, T. Baker
Table 11 Two-sample difference tests: surviving agents in most vs. least harsh environments Variable PROFIT STATUS DIE
Average, least harsh 1.18
Average, most harsh 1.24
p-value of ANOVA test 0.6761*
148.9
70.7
0.00E+00*
17.9
17.3
0.0260*
TYPE
1.46
1.56
0.0684**
BUYSELLPREF
0.96
0.98
0.4309**
20.22
19.98
0.5925*
PROPBIAS
0.92
0.93
0.9517**
BIASCHANGE
0.53
0.52
0.3627*
Markup
5.35
5.66
0.0308*
ReactTime*Mag*Prob
8.40
7.86
0.2694*
TERRITSETUP
4.49
1.63
0.00E+00*
AVERSION
0.52
0.48
0.2369**
MOVE adj
0.95
1.03
5.97E−08*
Change
0.57
0.58
0.8215*
SEARCHLAND
20.54
21.22
0.1603*
THRESHFLEX
4.12
4.07
0.7259*
AGENTMEM
* p-value for Wilcoxon rank sum test ** p-value for a Pearson χ 2 contingency table test
more can be presumed about the real world than mere propositions, which are losely based on a theorist’s understanding and synthesis of progress in one or more fields to that point. Thus, our simulation allows us to be more confident than a theorist that makes mere propositions, but somewhat less confident than the scientist who tests hypotheses using real-world data. 5.1 Organizational characteristics While not the focus of this study, the analysis conducted for the purpose of validation can also be used to hypothesize some organizational characteristics. For example, hypercompetition (D’Aveni 1994) claims that environments are increasingly becoming more harsh, at an increasing rate. One might, then, ask what effect hypercompetition has on agent characteristics. To address this question, in the simulation results, one might observe changes in agent characteristics when an agent changes from least to most harsh conditions (Table 11). While this might not indicate all of the conditions of hypercompetition, which indeed could and should be studied by its own simulation extension with which would allow for agents to switch between arenas of competition, one can at least note the effects of increasing competition on agent characteristics and generalize the observations to say something about the real world: Hypothesis 1 Organizations in environments with increasing harshness will, on average, exhibit lower reputations, be less tolerant of loss periods, and be associated with
Putting the organization back into computational organization theory
111
higher price markups for same-value products, less territoriality, and more agility in altering their competitive positions. Hypothesis 1 might be operational in the real world for the following reasons. First, in a more competitive environment, attacks and counterattacks will be more frequent; each attack is more likely to contain an assault on the competitor’s reputation, thus reducing the reputation of, perhaps, both participants. Second, periods of contiguous losses will be tolerated less because investors will be less tolerable in a more risky environment. Third, organizations will move across their parameters of competition faster to find a niche in the competitive landscape, following Porter’s (1980) method of avoiding direct competition, but once that niche is found, it creates a micro-monopoly of sorts which allows the firm to mark up its prices more than otherwise, albeit for a shorter period of time; but since there are more firms behaving in this manner, markups industry-wide will increase somewhat. Note that this last argument runs counter to the popular economics idea that competition increases industry efficiency and reduces profit margins; however, this economics view does not take into account industry dynamics and regional inefficiencies created by Porter’s micromonopoly creation strategy. In the real world, no “perfect competition” really exists, but while increasing competition may root out organizations with poor characteristics and result in overall average efficiency, increased competition may also result in decreased efficiencies with regard to pricing markups. One way the above hypothesis might be used in a practical manner might be as a way to identify more and less harsh environments, assuming that firms exhibiting higher risk are generally those firms found in more harsh environments. For example, if an organization finds itself in a competitive environment and wants to diversify its risk by acquiring an unrelated organization in a less harsh environment, it might measure reputations, loss tolerance, price markups, territoriality, and agility in its own industry and compare the measures with those of other industries. Acquisition candidates might then be chosen from industries with lower average scores when compared with the firm’s industry. Alternatively, one could develop a risk-controlled portfolio of stocks based on designing stock purchases which would result in a wide variance of industry average scores. In other words, the characteristics in Hypothesis 1 might be used to construct a sort of measure of industry risk. The concept of the liability of newness in OT certainly involves finer-grained elements, but it has until now received a black-box sort of treatment; much more can and should be done in real-world studies to identify more detailed characteristics of this liability, while COT studies have been limited to rates of obsolescence, rather than organizational characteristics, as antecedents (Schulz 1998). Certainly, the success rate of entrepreneurial ventures is characteristically low, and on one hand this is because risk has been encouraged by state laws, in part evidenced by the growing popularity of both the state-by-state availability of limited-liability corporations (LLCs) which protects the property of proprietors from debt reclamation, and on the other hand because not enough entrepreneurs engage in well-deliberated strategic plans. Suppose, however, that an entrepreneur has set up an LLC and has expertly written a business plan which she fully intends to carry out, what then? What characteristics, at that point, could her organization exhibit which might reduce the firm’s liability of newness? Hypothesis 2, derived from Table 10, offers some suggestions:
112
B.W. Kulik, T. Baker
Hypothesis 2 Given that strategic choice matters, organizations will live longer if they are more profitable, have a higher reputation, tolerate longer periods of losses, search their environments somewhat less, change in characteristics more frequently, markup prices higher, and set up territorial areas for launching attacks against competitors. Of course, the “finding” that more profitable organizations live longer is simply a repeat of the strategic choice perspective of organizations, and the idea that higher-reputation organizations live longer is an observation central to institutional theory—an input to the simulation code. However, in the present study, it must be noted that these findings, while seemingly circular, were at least falsifiable in that survived agents need not have exhibited these different characteristics. It therefore stands as further evidence for both the strategic choice and institutional theory perspectives. The prediction that a longer-lasting organization will tolerate longer periods of losses is also an unsurprising result, as for example it is widely known that, given time, an organization will uncover and exploit a niche, and perhaps a propitious niche which will grow at the rate of the company’s capability for growth, but that it has until it burns through its present store of cash to unearth this situation. More cash buys more time for an organization to find a profitable, survivable place in its environment because, during the time that it is burning through its cash stores and spending it on operations, it is operating at a loss. Finally, firms having shorter “memories” are willing to forgive blacklisted interaction agents and be more skeptical of whitelisted agents, while organizations which search their environments somewhat less extensively are able to make faster, cheaper decisions, and those firms which set up territorial areas and attack intruders are more apt to live longer because of their established revenue stream (in effect, they have found a reliable niche). Thus, firms oriented toward longer life are not merely those which make higher profits and are able to garner higher reputations. It could be argued that a lower risk is associated with a firm which lives longer, on average. Therefore, Hypothesis 2, like Hypothesis 1, could be used to aid the investment or acquisition decision, but in this case, the organization’s score could be compared its industry’s average. When Hypotheses 1 and 2 are combined, they might be especially useful for investors/acquirers interested in finding, say, a relatively lowerrisk organization in a lower-risk industry, or hedging one’s investment in a high-risk industry by investing in/acquiring a lower-risk organization in that industry. Perhaps an important contribution of this study is toward the advancement of contingency theory itself. The traditional independent variables of age, size, change in size (Baker and Cullen 1993), as well as structural characteristics such as formalism, centralization, and specialization (Blau and Shoenherr 1971), may be the tip of the proverbial iceberg with regard to the number of actual organization-level characteristics that may be contingent on environmental conditions. Other fields have advanced this additional-characteristics idea. For example, operations management research has extended the idea of contingency toward the success of flexibility (Ketokivi 2006) and integration (Koufteros et al. 2005) strategies, negotiation strategies have been considered contingent on the nature of an international conflict in the field of public relations (Zhang et al. 2004), and predicting whether CEOs have chief op-
Putting the organization back into computational organization theory
113
erating officers was found to be contingent on the CEO’s experience, but lower performance ensues (Hambrick and Cannella 2004). In management theory, the use of additional independent and dependent variables toward the further development of contingency theory has been surprisingly sparse. Recently, post-bankruptcy strategic change was found to be contingent on the type of CEO successor choice (Brockmann et al. 2006), and the success of structural adaptation was theorized to be contingent on rhetorical congruence (Sillince 2005). Certainly, organization theories themselves suggest many more basic contingency variables, and a few of them have been proposed here. Perhaps this study will influence the further development of contingency theory as the association between specific organizational characteristics and specific environmental conditions. 5.2 Limitations A key limitation of the model developed in this work is that it is based in only eight perspectives of how organizations are expected to act. Surely, there are more perspectives within organization theory and strategic choice that could be incorporated and modeled, such as multiple arenas of competition (D’Aveni 1994), negotiations (Denis et al. 1999), organizational structures (Galbraith 1973), organizational climate and culture, innovation, etc. While the objective was to develop a basic model of what virtually all organizational researchers might accept as how organizations actually function, the conclusions of this study must be qualified within the context of the eight perspectives modeled. With regard to the eight perspectives, then, there is the possibility that they have been incompletely or incorrectly modeled. This limitation is due to the fact that realworld data was not used as input into the simulation. However, as detailed in the COT literature review above, the approach taken in this study follows the norm of the field rather than the exception; thus, it is a “weakness” of virtually all COT studies that have been published. The proper way to curtail the conclusion based on this limitation is to develop hypotheses that are intended as input to real-world studies (Carley 1999), as done in this work. The simulation, as modeled, makes at least two assumptions with regard to operations research issues. First, manufacturing process time is not modeled because the model considers the “product” to be a commodity with basically equal manufacturing times for each simulated organization. In other words, manufacturing equipment can be readily acquired off-the-shelf and relatively costlessly integrated into an existing manufacturing line. Manufacturing times were made constant because this study was intended to be concerned with behavior at a higher level of analysis; it is not a study on operations, so certain operations factors may be held constant (such as manufacturing time) without a significant loss in generality. Note that operations research studies usually do just the opposite in that they observe operations conditions while holding constant higher-level factors. A second operations research issue that is assumed for this study is that no advance orders are placed (and no backorders accumulated) with manufacturers (or harvesters) for their finished inventories. This, again, is a simplification of the real-world operations condition, which can admittedly be very complicated, but is allowed here because, again, operations was not an area of focus for this study.
114
B.W. Kulik, T. Baker
5.3 Extensions It is not difficult to imagine issues toward which this baseline can be extended to contribute to extant research; all one needs to do is look for a stream of organizationlevel research which exhibits contradictory findings and complex conditions. In the field of strategic management, the research areas of the benefits and/or drawbacks of different types of diversification (Hoskisson and Hitt 1990; Reed and Luffman 1986), the conditions of hypercompetition (D’Aveni 1994; Wiggins and Ruefli 2005), and the environmental conditions for appropriate executive behavior and compensation (Kulik 2005). The latter area is especially relevant because access to the boardroom and executives’ offices is often not available to researchers, making the analysis of real-world data unusually scarce. The simulation model could also be extended to further study organizational characteristics. One could model more or all of the environmental differences as described in Table 4, rather than modeling just one element of each environmental dimension as conducted in the simulation. For example, the effect of sales concentration could be modeled by superimposing a sales matrix that varies geographically, analogous to the resource landscape, to further model the effects of environmental complexity. It may well be that different aspects of the same environmental dimension may have different, if not opposing, effects on agent profitability and longevity. In other words, it may well be that “customer dependence” may oppose, or alter, the effects of resource dependence.
6 Summary and conclusion This paper contributes to a number of fields in a number of ways. First, it contributes to the field of COT by suggesting that the modeling of organizations in this field have been needlessly simple, and that computer simulation can contribute more to the real world when these simulations are more complex, even if the rules programmed remain simple. Further, it demonstrates that studies rooted in COT can contribute to OT once the analysis is made simple and relevant to issues prevalent in OT. Third, this study contributes to the field of OT by broadening and extending the concept of contingency to include any organizational characteristic, by identifying certain trends in organizational characteristic under more harsh environments, and by proposing a number of antecedents to organizational longevity. This work suggests a direction of future research that combines the fruit of computational studies with that of real-world studies. It contends that in some areas of scientific research, the scientific method cannot proceed readily enough without the combined efforts of computational and real-world researchers, especially when realworld findings are contradictory or scarce. It is hoped that this work revitalizes the potential complementary relationship between computational and real-world research which, in our view, has been exploited far too sparsely in the fields of strategic management and organization theory. Acknowledgements We would like to thank the John Cullen, Jonathan Arthurs, and Len Trevino for their helpful comments at various stages of this paper’s development.
Putting the organization back into computational organization theory
115
References Alchain AA (1950) Uncertainty, evolution, and economic theory. J Polit Econ 58(6):211–221 Aldrich HE (1979) Organizations and environments. Prentice-Hall, Englewood Cliffs Aldrich HE, Pfeffer J (1976) Environments of organizations. Ann Rev Sociol 2:79–105 Anastasi A (1988) Psychological testing. Macmillan, New York Ansoff I (1965) Corporate strategy. McGraw Hill, New York Aragon-Correa JA, Sharma S (2003) A contingent resource-based view of proactive corporate environmental strategy. Acad Manag Rev 28(1):71–88 Ashworth MJ, Carley KM (2007) Can tools help unify organization theory? Perspectives on the state of computational modeling. Comput Math Organ Theory 13(1):89–111 Axelrod R (1980) Effective choice in the prisoner’s dilemma. J Confl Resol 24(1):3–25 Baker DD, Cullen JB (1993) Administrative reorganization and configurational context: the contingent effects of age, size, and change in size. Acad Manag J 36(6):1251–1277 Barnard CI (1938) The functions of the executive. Harvard University Press, Cambridge Barron DN (2000) Simulating the dynamics of organizational populations: a comparison of three models of organizational entry, exit, and growth. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 209–242 Bettis RA, Prahalad CK (1995) The dominant logic: retrospective and extension. Strateg Manag J 16(1):5– 14 Blau PM (1956) Bureaucracy in modern society. Random House, New York Blau PM, Meyer MW (1971) Bureaucracy in modern society, 3rd edn. Random House, New York Blau PM, Shoenherr RA (1971) The structure of organizations. Basic Books, New York Bonini CP (1963) Simulation of information and decision systems in the firm. Prentice-Hall, Englewood Cliffs Bothner MS, White HC (2000) Market orientation and monopoly power. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 181–208 Boulding KE (1950) A reconstruction of economics. Wiley, New York Bourgeois LJI (1985) Strategic goals, perceived uncertainty, and economic performance in volatile environments. Acad Manag J 28(3):548–573 Boyer KK, Swink M, Rosenzweig ED (2005) Operations strategy research in the POMS journal. Prod Oper Manag 14(4):442–449 Brockmann EN, Hoffman JJ, Dawley DD (2006) A contingency theory of CEO successor choice and post-bankruptcy strategic change. J Manag Issues 18(2):213–233 Bruderer E, Singh JV (1996) Organizational evolution, learning, and selection: a genetic algorithm-based model. Acad Manag J 39(5):1322–1349 Bruggeman J, Nualláin B (2000) A niche width model of optimal specialization. Comput Math Organ Theory 6(2):161–170 Burns T, Stalker GM (1961) The management of innovation. Oxford University Press, Oxford, England Burton RM, Obel B (2004) Strategic organizational diagnosis and design, 3rd edn. Kluwer, New York Carley KM (1995) Computational and mathematical organization theory: perspective and directions. Comput Math Organ Theory 1(1):39–56 Carley KM (1996) Validating computer models (working paper). Available at http://www.casos.cs.cmu. edu/bios/carley/working_papers.php Carley KM (1999) On generating hypotheses using computer simulations. Syst Eng 2(2):69–77 Carley KM (2000) Organizational adaptation in volatile environments. In: Ilgen DR, Huin CL (eds) Computational modeling of behavior in organizations: the third scientific discipline. American Psychological Association, Washington, pp 241–268 Carley KM (2002) Intraorganizational complexity and computation. In: Baum JAC (ed) The Blackwell companion to organizations. Blackwell, Oxford, pp 208–232 Carley KM, Hill V (2000) Structural change and learning within organizations. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 61–92 Carley KM, Prietula MJ (1998) WebBots, trust, and organizational science. In: Prietula MJ, Carley KM, Gasser L (eds) Simulating organizations: computational models of institutions and groups. AAAI Press/MIT Press, Menlo Park, pp 3–22 Carley KM, Svoboda DM (1996) Modeling organization adaptation as a simulated annealing process. Sociol Methods Res 25(1):138–168
116
B.W. Kulik, T. Baker
Child J (1972) Organizational structure, environment, and performance: the role of strategic choice. Sociology 6:2–22 Child J (1974) Managerial and organizational factors associated with company performance, part I. J Manag Stud 11:175–189 Child J (1975) Managerial and organizational factors associated with company performance, part II. J Manag Stud 12:12–27 Conte R, Sichman JS (2002) Dependence graphs: dependence within and between groups. Comput Math Organ Theory 8(2):87–112 Cyert RM, March JG (1963) A behavioral theory of the firm. Prentice-Hall, Englewood Cliffs Davis JP, Eisenhardt KM, Bingham CB (2007) Developing theory through simulation methods. Acad Manag Rev 32(2):480–499 D’Aveni RA (1994) Hypercompetitive rivalries: competing in highly dynamic environments. Free Press, New York Dacin TM, Goodstein J, Scott RW (2002) Institutional theory and institutional change: introduction to the special research forum. Acad Manag J 45(1):45–57 Danneels E (2003) Tight-loose coupling with customers: the enactment of customer orientation. Strateg Manag J 24(6):559–576 Deephouse DL (1996) Does isomorphism legitimate? Acad Manag J 39(4):1024–1039 Denis DD, Denis DK, Sarin A (1999) Agency theory and the influence of equity ownership structure on corporate diversification strategies. Strateg Manag J 20(11):1071–1076 Dess GG, Beard DW (1984) Dimensions of organizational task environments. Adm Sci Q 29(1):52–73 Dill WR (1958) Environments as an influence on managerial autonomy. Adm Sci Q 2(4):409–443 DiMaggio PJ, Powell WW (1983) The iron cage revisited: institutional isomorphism and collective rationality in organizational fields. Am Sociol Rev 48:147–160 Duncan RB (1972) Characteristics of organizational environments and perceived environmental uncertainty. Adm Sci Q 17(3):313–327 Dutton JM, Starbuck WH (1971a) Computer simulation of human behavior. Wiley, New York Dutton JM, Starbuck WH (1971b) The plan of the book. In: Dutton JM, Starbuck WH (eds) Computer simulation of human behavior. Wiley, New York, pp 3–8 Emery FE, Trist E (1965) The causal texture of organizational environments. Hum Relat 18:21–31 Epstein JM, Axtell R (1996) Growing artificial societies: social science from the bottom up. Brookings Institution Press, Washington Galbraith J (1973) Designing complex organizations. Addison-Wesley, Reading Gibson CB, Birkinshaw J (2004) The antecedents, consequences, and mediating role of organizational ambidexterity. Acad Manag J 47(2):209–226 Gilbert N, Troitzsch KG (2002) Simulation for the social scientist. Open University Press, Philadelphia Handley HAH, Levis AH (2001) A model to evaluate the effect of organizational adaptation. Comput Math Organ Theory 7(1):5–44 Hambrick DC, Cannella AA (2004) CEOs who have COOs: contingency analysis of an unexplored structural form. Strateg Manag J 25(10):959–979 Hambrick DC, Mason PA (1984) Upper echelons: the organization as a reflection of its top managers. Acad Manag Rev 9(2):193–206 Hannan MT, Freeman JH (1977) The population ecology of organizations. Am Sociol Rev 82:929–964 Harrison JR, Carroll GR (2000) Modeling culture in organizations: formulation and extension to ecological issues. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 37–62 He Z-L, Wong P-K (2004) Exploration vs. exploitation: an empirical test of the ambidexterity hypothesis. Organ Sci 15(4):481–494 Hickson DJ, Hinings CR, Lee CA, Schneck RE, Pennings JM (1971) A strategic contingencies theory of intraorganizational power. Adm Sci Q 16(2):216–229 Hirsch PM (1975) Organizational effectiveness and the institutional environment. Adm Sci Q 20(3):327– 344 Hoskisson RE, Hitt MA (1990) Antecedents and performance outcomes of diversification: a review and critique of theoretical perspectives. J Manag 16(2):461–509 Hough JR, White MA (2003) Environmental dynamism and strategic decision-making rationality: an examination at the decision level. Strateg Manag J 24(5):481–489 Hough JR, White MA (2004) Scanning actions and environmental dynamism: gathering information for strategic decision making. Manag Decis 42(6):781–793
Putting the organization back into computational organization theory
117
Huberman BA (2001) The dynamics of organizational learning. Comput Math Organ Theory 7(2):145– 153 Ireland RD, Miller CC (2004) Decision-making and firm success. Acad Manag Exec 18(4):8–12 Janney JJ, Dess GG (2004) Can real-options analysis improve decision-making? Promises and pitfalls. Acad Manag Exec 18(4):60–75 Jensen M, Zajac EJ (2004) Corporate elites and corporate strategy: how demographic preferences and structural position shape the scope of the firm. Strateg Manag J 25(6):507–524 Jurkovich R (1974) A core typology of organizational environments. Adm Sci Q 19(3):380–394 Ketchen DJ, Snow CC, Street VL (2004) Improving firm performance by matching strategic decisionmaking processes to competitive dynamics. Acad Manag Exec 18(4):29–43 Ketokivi M (2006) Elaborating the contingency theory of organizations: the case of manufacturing flexibility strategies. Prod Oper Manag 15(2):215–228 Krackhardt D (2000) Viscosity models and the diffusion of controversial innovation. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 243–267 Kulik BW (2005) Agency theory, reasoning, and culture: in search of a solution. J Bus Ethics 59(4):347– 360 Kulik BW (2006) Strategic action and executive behavior: an agent-based simulation. PhD dissertation, Washington State University, Department of Management and Operations Koufteros X, Vonderembse M, Jayaram J (2005) Internal and external integration for product development: the contingency effects of uncertainty, equivocality, and platform strategy. Decis Sci 36(1):97–133 Kuhn TS (1996) The structure of scientific revolutions, 3rd edn. The University of Chicago Press, Chicago Law AM, Kelton WD (2000) Simulation modeling and analysis, 3rd edn. McGraw Hill, New York Lawrence PR, Lorsch JW (1967) Organization and environment: managing differentiation and integration. Harvard Business School Press, Boston Levinthal DA (1994) Surviving Schumpeterian environments: an evolutionary perspective. In: Baum JAC, Singh JV (eds) The evolution of evolution. Oxford University Press, New York, pp 167–178 Levinthal DA (2000) Modeling adaptation on rugged landscapes. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 329–348 Lin Z (1998) The choice between accuracy and errors: a contingency analysis of external conditions and organizational decision making performance. In: Prietula MJ, Carley KM, Gasser L (eds) Simulating organizations: computational models of institutions and groups. AAAI Press/MIT Press, Menlo Park, pp 67–87 Lin Z, Hui C (1997) Adapting to the changing environment: a theoretical comparison of decision making proficiency of lean and mass organization systems. Comput Math Organ Theory 3(2):113–142 Loch CH, Huberman BA, Ulku S (2000) Multi-dimensional status competition and group performance. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 119–140 , Lomi A, Larsen ER (1998) Density delay and organizational survival: computational models and empirical comparisons. Comput Math Organ Theory 3(4):219–247 Lomi A, Larsen ER (2000) Failure as a structural concept: a computational perspective on age dependence in organizational mortality rates. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 269–303 Macy MW, Strang D (2000) Dedicated followers of success: a computational model of fashinable innovation. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 93–117 Malerba F, Nelson R, Orsenigo L, Winter S (2000) Product diversification in a “history friendly” model of the evolution of computer industry. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 350– 375 Malyankar RM, Findler NV (1998) A methodology for modeling coordination in intelligent agent societies. Comput Math Organ Theory 4(4):317–345 March JG (1994) The evolution of evolution. In: Baum JAC, Singh JV (eds) Evolutionary dynamics of organizations. Oxford University Press, New York, pp 39–49 March JG, Simon H (1958) Organizations. Blackwell, Cambridge Masuch M, Huang Z (1996) A case study in logical deconstruction: formalizing J.D. Thompson’s organizations in action in a multi-agent action logic. Comput Math Organ Theory 2(2):71–113
118
B.W. Kulik, T. Baker
Masuch M, LaPotin P (1989) Beyond garbage cans: an AI model of organizational choice. Adm Sci Q 34(1):38–67 Mezias SJ, Lant TK (1994) In: Baum JAC, Singh JV (eds) Mimetic learning and the evolution of organizational populations. Oxford University Press, New York, pp 179–198 Mild A, Taudes A (2007) An agent-based investigation into the new product development capability. Comput Math Organ Theory 13(3):315–331 Miles RE, Snow CC (1978) Organizational strategy, structure, and process. Stanford University Press, Stanford Miller JH (2000) Evolving information processing organizations. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 307–327 Naveh I, Sun R (2006) A cognitively based simulation of academic science. Comput Math Organ Theory 12(3):13–337 Nutt PC (2004) Expanding the search for alternatives during strategic decision-making. Acad Manag Exec 18(4):13–28 Oliver C (1992) The antecedents of deinstitutionalization. Organ Stud 13(4):563–588 Oliver C (1997) The influence of institutional task environment relationships on organizational performance: the Canadian construction industry. J Manag Stud 34(1):99–124 Parsons T (1956) Suggestions for a sociological approach to the theory of organizations. Adm Sci Q 1(1):63–85 Penrose ET (1952) Biological analogies in the theory of the firm. Am Econ Rev 62(5):804–819 Perdu DM, Levis AH (1998) Adaptation as a morphing process: a methodology for the design and evaluation of adaptive organizational structures. Comput Math Organ Theory 4(1):5–41 Perrow C (1986) Complex organizations: a critical essay, 3rd edn. McGraw-Hill, New York Pfeffer J, Salancik GR (1978) The external control of organizations: a resource dependence perspective. Stanford University Press, Stanford Porter ME (1980) Competitive strategy: techniques for analyzing industries and competitors. Free Press, New York Prahalad CK, Bettis RA (1986) The dominant logic: a new linkage between diversity and performance. Strateg Manag J 7(6):485–501 Prietula MJ (2000) Advice, trust and gossip among artificial agents. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 141–177 Prietula MJ, Watson HS (2000) Extending the Cyert-March duopoly model: organizational and economic insights. Organ Sci 11(5):565–585 Rakotobe-Joel T, McCarthy IP, Tranfield D (2002) A structural and evolutionary approach to change management. Comput Math Organ Theory 8(4):337–364 Randolph WA, Dess GG (1984) The congruence perspective of organization design: a conceptual model and multivariate research approach. Acad Manag Rev 9(1):114–127 Reed R, Luffman GA (1986) Diversification: the growing confusion. Strateg Manag J 7(1):29–35 Sadler-Smith E, Shefy E (2004) The intuitive executive: understanding and applying ‘gut feel’ in decisionmaking. Acad Manag Exec 18(4):76–91 Saoud NB-B, Mark G (2007) Complexity theory and collaboration: an agent-based simulator for a space mission design team. Comput Math Organ Theory 13(2):113–146 Sastry MA (2000) Understanding dynamic complexity in organizational evolution. In: Lomi A, Larsen ER (eds) Dynamics of organizations: computational modeling and organization theories. AAAI Press/MIT Press, Menlo Park, pp 378–404 Scherer FM (1971) Industrial market structure and economic performance. Rand McNally, Chicago Schoonhoven CB (1981) Problems with contingency theory: testing assumptions hidden within the language of contingency “theory”. Adm Sci Q 26(3):349–377 Schulz M (1998) A model of obsolescence of organizational rules. Comput Math Organ Theory 4(3):241– 266 Selznick P (1957) Leadership in administration: a sociological interpretation. University of California Press, Berkeley Shadish WR, Cook TD, Campbell DT (2002) Experimental and quasi-experimental designs for generalized causal inference. Houghton Mifflin, Boston Shimizu K, Hitt MA (2004) Strategic flexibility: organizational preparedness to reverse ineffective strategic decisions. Acad Manag Exec 18(4):44–59
Putting the organization back into computational organization theory
119
Sillince JAA (2005) A contingency theory of rhetorical congruence. Acad Manag Rev 30(3):608–621 Simon HA (1945) Administrative behavior, 4th edn. Free Press, New York Starbuck WH (1976) Organizations and their environments. In: Dunnette MD (ed) Handbook of industrial and organizational psychology. Rand-McNally, Chicago, pp 1069–1123 Staw BM, Szwajkowski E (1975) The scarcity-munificence of organizational environments and the commission of illegal acts. Adm Sci Q 20(3):345–354 Suchman MC (1995) Managing legitimacy: strategic and institutional approaches. Acad Manag Rev 20(3):571–609 Teitelbaum D, Dowlatabadi H (2000) A computational model of technological innovation at the firm level. Comput Math Organ Theory 6(3):227–247 Thompson JD (1967) Organizations in action. McGraw-Hill, New Brunswick Tosi H, Aldag R, Storey R (1973) On the measurement of the environment: an assessment of the Lawrence and Lorsch environmental uncertainty subscale. Adm Sci Q 18(1):27–36 Tung RL (1979) Dimensions of organizational environments: an exploratory study of their impact on organization structure. Acad Manag J 22(4):672–693 Tushman ML, Romanelli E (1985) Organizational evolution: a metamorphosis model of convergence and reorientation. In: Staw BM, Cummings LL (eds) Research in organizational behavior, vol VII. JAI Press, Greenwich, pp 171–222 Turner SF, Bettis RA, Burton RM (2002) Exploring depth versus breadth in knowledge management strategies. Comput Math Organ Theory 8(1):49–73 Usher JM, Evans MT (1996) Life and death along gasoline alley: Darwinian and Lamarckian processes in a differentiating population. Acad Manag J 39(5):1428–1466 Venkatraman N (1989) The concept of fit in strategy research: toward verbal and statistical correspondence. Acad Manag Rev 14(3):423–444 Vermeulen I, Bruggeman J (2001) The logic of organizational markets: thinking through resource partitioning theory. Comput Math Organ Theory 7(2):87–111 Washington M, Zajac EJ (2005) Status evolution and competition: theory and evidence. Acad Manag J 48(2):282–296 Weick K (1979) The social psychology of organizing, 2nd edn. Addison-Wesley, Reading Wiggins RR, Ruefli TW (2005) Schumpeter’s ghost: is hypercompetition making the best of times shorter? Strateg Manag J 26(10):887–911 Williamson O (1975) Markets and hierarchies: analysis and antitrust implications. Free Press, New York Woodward J (1958) Management and technology. Her Majesty’s Stationery Office, London Zhang J, Qiu Q, Cameron GT (2004) A contingency approach to the Sino-U.S. conflict resolution. Public Relat Rev 30(4):391–399
Brian W. Kulik is a professor at Central Washington University where he conducts research in organization theory, organizational research methods, and strategic management. His specific areas of research involve the integration of organization theories, organizational behavior, diversity theory, strategic management theories, and economics toward the application of research in executive behavior and compensation. Additionally, his work in research methods applies work in Monte Carlo simulation toward factor analysis, management education, life-cycle costing, and marketing surveys. Journals in which his research appeared include Journal of Business Ethics and Organizational Analysis. Dr. Kulik has an undergraduate degree in mechanical and materials engineering from Vanderbilt University, a masters degree in materials science and engineering from the University of Cincinnati, an MBA from the University of Denver, a masters degree in statistics from Washington State University, and a Ph.D. in business administration, with concentrations in organization theory and strategic management, from Washington State University. Timothy Baker is a professor at Washington State University where he conducts research in decision science and operations management. He is conducts research in service operations, especially auction theory and pricing in general. Some of his pricing work focuses on large-scale price optimization systems for industries such as airline, hotel, and rental car. He is also interested in sustainable supply chain management, such as understanding economical means for recycling and reuse of electronic components. Journals in which his research appeared include Decision Sciences, IEEE Transactions in Engineering Management, and Productions and Operations Management. Dr. Baker has an undergraduate degree in mathematics and economics from Claremont McKenna College, a masters degree in operations research from the University of North Carolina at Chapel Hill, and a Ph.D. in operations management from Ohio State University.