You are on page 1of 56

Fit, Equifinality, and Organizational Effectiveness: A Test of Two Configurational Theories Author(s): D. Harold Doty, William H.

Glick, George P. Huber Reviewed work(s): Source: The Academy of Management Journal, Vol. 36, No. 6 (Dec., 1993), pp. 1196-1250 Published by: Academy of Management Stable URL: http://www.jstor.org/stable/256810 . Accessed: 03/02/2012 06:54
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Academy of Management is collaborating with JSTOR to digitize, preserve and extend access to The Academy of Management Journal.

http://www.jstor.org

? Academy of Management Journal 1993, Vol. 36, No. 6, 1196-1250

FIT, EQUIFINALITY, AND ORGANIZATIONAL EFFECTIVENESS: A TEST OF TWO CONFIGURATIONAL THEORIES


D. HAROLD DOTY University of Arkansas WILLIAM H. GLICK GEORGE P. HUBER University of Texas at Austin
Mintzberg'sand Miles and Snow's configurational theories have both received widespread attention. Most researchers, however, have interpreted these theories in terms of categories of organizations rather than organizational configurations based on ideal types. We explicated the logical structure of configurationaltheories and developed a set of configurational fit models that are congruentwith alternative assumptions of equifinality, which is the premise that multiple organizationalforms are equally effective. Then the two theories were formalized with these models and tested empirically. Contraryto our expectations, the results do not support Mintzberg'stheory that organizations will be more effective to the extent that they resemble his five ideal types. In contrast with these null results, configurational fit based on Miles and Snow's theory predicted 24 percent of the variance in overall organizational effectiveness.

Configurational theories compose a growing body of organizational literature. At the organizational level of analysis, configurational theories typically posit higher effectiveness for organizations that resemble one of the ideal types defined in the theory. The increased effectiveness is attributed to the internal consistency, or fit, among the patterns of relevant contextual, structural, and strategic factors. Two excellent examples of configurational theories that have enjoyed widespread popularity but mixed or limited empirical support are Mintzberg's (1979, 1983) theory of organizational structure and Miles and Snow's (1978) theory of strategy, structure, and process.

This research was supported in part by Army Research Institute grant #MDA-903-85-0404, by the College of Business Administration, University of Arkansas, and by the Charles and Elizabeth Prothro Regents Chair in Business Administration. We gratefully acknowledge the data collection assistance of Kenneth L. Bettenhausen, Joan N. Boothe, Geoffrey R. Brooks, Kim S. Cameron, Richard L. Daft, Sarah J. Freeman, James B. Goes, Alan D. Meyer, C. Chet Miller, Aneil K. Mishra, Charles A. O'Reilly, Richard C. Snyder, Kathleen M. Sutcliffe, and Beverly B. Tyler and the expert ratings provided by Raymond E. Miles, Charles E. Snow, and Alan D. Meyer. 1196

1993

Doty, Glick, and Huber

1197

Mintzberg's (1979, 1983) theory identifies five ideal types of organization: simple structure, machine bureaucracy, professional bureaucracy, divisionalized form, and adhocracy. An organization that approximates one of these ideal types is hypothesized to be more effective than other organizations, especially when its context fits the ideal type. For example, Mintzberg's (1979) reasoning suggests that young organizations with simple but dynamic environments and simple technologies will be more effective if they rely on simple structure and internal coordination via direct supervision. These young organizations will be less effective if they rely on bureaucratic structure and internal coordination via formalization or standardization. The widespread attention given to Mintzberg's (1979, 1983) typology reflects its strong intuitive appeal, but because the typology and the underlying theory have received little or no systematic empirical examination in large-scale comparative studies, there is little support for the theory. Indeed, few theories have received so much attention in management textbooks and organizational science journals with such meager empirical support. A second example of a configurational theory, developed by Miles and Snow (1978), identifies four ideal types of organization: the prospector, the analyzer, the defender, and the reactor. Each of these ideal types is richly described as a unique configuration of contextual, structural, and strategic factors. Miles and Snow posited that at least three of these ideal types-the effective forms of organization. prospector, analyzer, and defender-were Miles and Snow's theory has stimulated numerous empirical investigations (for a review, see Zahra and Pearce [1990]). Several studies have concluded that organizations classified as prospectors, analyzers, and defenders are generally more effective than reactors and that the relative effectiveness of the types varies with context. For example, Snow and Hrebiniak (1980) found that in most environments organizations classified as analyzers were generally the most effective and that in highly regulated environments reactors were more effective than prospectors or defenders. Smith, Guthrie, and Chen (1989) found that the effectiveness of the different categories was contingent on firm size. Firms classified as defenders performed better when they were relatively small, but firms classified as prospectors performed better when they were relatively large. Hambrick (1983) found that the relative performance of organizations classified as prospectors and defenders varied according to environment and the performance measure by which they were evaluated, but that both could be effective. On balance, the empirical work seems to provide moderate support for Miles and Snow's theory. Firms that resemble the prospector, defender, or analyzer type appear to be more effective than firms resembling the reactor type. Zahra and Pearce (1990) noted, however, that there have been a variety of inconsistencies in the translations of the theory for empirical testing. For example, researchers have included different numbers of ideal types, and the analyzer has been interpreted in two different ways, sometimes as a distinct ideal type and sometimes as a combination of the prospector and the

1198

Academy of Management Journal

December

defender. Further, most studies have drawn on only one construct or on a few constructs to characterize the configurations described in the theory. Such theoretical and methodological deficiencies may explain the variation in the reported levels of support. Although Mintzberg's (1979, 1983) and Miles and Snow's (1978) theories are clearly distinct, the difficulties encountered when testing the two theories are similar and common to many configurational theories. Despite the apparent clarity and simplicity of the individual arguments in each theory, the overall theories are very complex. For example, Mintzberg posited that in order to be maximally effective, organizations must have design configurations that are internally consistent and fit multiple contextual dimensions, such as age, size, environment, and technology. Therefore, valid tests of the theory must assess the relationship between fit and effectiveness. Given the complexity of configurational theories, however, simple empirical approaches are inadequate. Unless the conceptual and analytical models are consistent with the logical structure of a theory, the test of the theory will not be valid (Blalock, 1969, 1989; Schoonhoven, 1981; Venkatraman, 1989). The complex fit assertions cannot be tested with simple bivariate or interactive conceptualizations of fit. Thus, the appropriate conceptualization of fit and the corresponding analytical models must be identified before fiteffectiveness assertions can be tested empirically (Van de Ven & Drazin, 1985). The deficiencies in the empirical literature concerning Miles and Snow's (1978) theory highlight additional problems encountered when testing configurational theories. Because they have treated the organizational configurations as categories rather than as the ideal types posited in the theory, most empirical investigations have failed to test the core thesis of this theory. Most researchers have categorized organizations in their samples as belonging to one of the four ideal types and then have compared the mean levels of effectiveness among the categories (Hambrick, 1983; Smith et al., 1989; Snow & Hrebiniak, 1980; Zajac & Shortell, 1989). The appropriateness of this approach is questionable for four reasons. First, the organizational configurations identified in the theory have typically been interpreted as categories of organizations rather than as ideal types of organizations. Categorical systems and typologies based on ideal types should not be confused; they are very distinct forms of theory building. When the configurations are treated as categories, marginal members of the categories are predicted to be as effective as their central members. When the configurations are treated as ideal types, however, organizations that marginally resemble the types are predicted to be much less effective than organizations that closely resemble them. Second, most studies have not modeled fit as the consistency among the relevant contextual, structural, and strategic factors and thus have not directly tested the fit-effectiveness predictions of the theory (Venkatraman and Prescott [1990] is an exception). Thus, much of the rich description of the ideal types provided in Miles and Snow (1978) has been ignored in empirical tests of the theory. Third, past studies have failed to explicitly

1993

Doty, Glick, and Huber

1199

model the equifinality assumption-the assumption that there are multiple in the theory. Fourth, effective organizational forms-incorporated equally the categorical approach to testing this theory has not supported any direct tests of several competing interpretations of the theory. The primary purpose of this study was to advance the fields of organizational science and strategic management by clearly articulating the logical structure of configurational theories. We used a configurational approach to develop quantitative models that more accurately translate the logical arguments incorporated in configurational theories. We attempted to clarify the logic of configurational theories by explicitly modeling the logical structures with precise multivariate definitions of the ideal types identified by configurational theorists and by articulating the relevant fit and equifinality assumptions in configurational theories. A second purpose of the research was to demonstrate that the models derived from the configurational approach can be used to conduct valid tests of configurational theories. We applied this configurational approach first to Mintzberg's theory and then to Miles and Snow's theory. In study 1, we extended Mintzberg's (1979, 1983) theory by (1) developing a priori hypotheses that are consistent both with the typology and the underlying theory, (2) identifying the models of configurational fit that are most closely tied to Mintzberg's arguments, and (3) testing those hypotheses both crosssectionally and longitudinally. In study 2, we first used the configurational approach to clarify two unresolved theoretical issues in Miles and Snow's (1978) theory: (1) the status of the reactor as a unique ideal type or as a residual category for ineffective organizations; and (2) the appropriateness of treating the analyzer as a midpoint on the continuum from prospector to defender. We then used the newly developed models of configurational fit to test the hypothesized fit-effectiveness assertions. The results from both studies have implications concerning the use of the configurational approach in future theory building in organizational science and strategic management. THE LOGICAL STRUCTURE OF CONFIGURATIONAL THEORIES Developing quantitative models of a theory is a necessary step in the theory development process (Blalock, 1969, 1989). Ideally, the quantitative modeling process is merely a translation exercise (Blalock, 1969). In practice, however, ambiguous assumptions are often identified in the modeling process. Such identification forces the model builder to develop new theory to clarify the ambiguous assumptions. Thus, the modeling process is an important component of the theory-building process because it clarifies and refines the logic underlying a theory. The quantitative model then serves as a more precise statement of the theory that can be more unambiguously falsified. When the quantitative model is not an accurate translation of the logical arguments in the theory, however, the model neither validly represents the theory nor adequately tests it (Blalock, 1969; Venkatraman, 1989).

1200

Academy of ManagementJournal

December

Thus, the critical issue is not to develop any quantitative model of a theory, but rather to develop a valid model that accurately represents its logical structure. Developing valid quantitative models of configurational theories involves three steps. First, the organizational configurations identified in a theory must be conceptualized and modeled as ideal types. Second, a model of fit that is consistent with the fit assertions in the theory must be developed. Third, the assumption of equifinality must be interpreted and integrated with the model of fit because most configurational theories identify multiple effective ideal types of organizations. Modeling Ideal Types One of the fundamental conceptual issues that arise when modeling configurational theories is the interpretation of the organizational configurations they identify. Many researchers misinterpret the organizational forms identified in configurational theories as nominal groups rather than as ideal-type organizations. For example, most researchers have interpreted Miles and Snow's work as developing a set of nominal categories (e.g., Conant, Mowak, & Varadarajan, 1990; Segev, 1987; Shortell & Zajac, 1990; Zajac & Shortell, 1989). A more accurate interpretation is that Miles and Snow have developed a theory that is based on a fundamentally different theory-building device, the ideal-type construct. An ideal type is a theoretical construct (McKinney, 1966) that can be used to represent a holistic configuration of organizational factors. The ideal type serves as "an abstract model so that deviation from the type can be noted and explained" (Blalock, 1969: 32). Miles and Snow supported this alternative interpretation when they stated explicitly that they were presenting the "pure" form of each configuration (1978: 30). Similarly, Mintzberg explicitly stated that his organizational forms composed "a typology of ideal or pure types" (1979: 473). Interpreting the organizational forms as ideal-type organizations rather than as categories of organizations has at least two implications for formally modeling and testing configurational theories. First, each ideal type is a singular and discrete phenomenon rather than a nominal category of organizations. Consequently, any formal statement or empirical test of a configurational theory should not rest on simple classification but should involve a much richer, multivariate approach to defining the ideal types. Second, each real organization in a sample need not be classified into one of the nominal groups identified in the theory. Instead, the degree of deviation between each real organization and the ideal types is measured. This deviation measure can then be used to predict organizational effectiveness in a more valid test of the configurational theory than is possible with a categorical interpretation of a theory that only requires classifying each organization into one of the nominal groups.

1993

Doty, Glick, and Huber

1201

Modeling Fit According to most configurational theories, effectiveness is highest in the ideal types of organization identified in the theory because the fit among contextual, structural, and strategic factors is at a maximum in those configurations. Thus, the second step in modeling a configurational theory is selecting a conceptual model of fit that is consistent with the conceptualization of fit used in the theory. Many different conceptualizations of fit appear in the organizational literature (e.g., Fry & Slocum, 1984; Mohr, 1982). Van de Ven and Drazin (1985; Drazin & Van de Ven, 1985) organized many of these conceptualizations into the interaction, selection, and systems approaches to fit. The interaction approach to fit characterizes many traditional theories, especially "contingency" theories, which typically define fit as the statistical interaction of two variables (Schoonhoven, 1981; Venkatraman, 1989). The selection approach to fit is adopted by theorists who develop organizational taxonomies (e.g., McKelvey, 1982) and by population ecologists. Despite their other merits, however, neither of these two approaches is consistent with the complex fit assertions in configurational theories. The conceptualization of fit that is most consistent with the logical arguments of configurational theories is the systems approach to fit, which Drazin and Van de Ven identified as the most complex and promising for future research (Drazin & Van de Ven, 1985; Van de Ven & Drazin, 1985). The systems approach defines fit in terms of consistency across multiple dimensions of organizational design and context. Fit is high to the extent that an organization is similar to an ideal type along multiple dimensions. "Deviations from ideal type designs should result in lower performance" (Van de Ven & Drazin, 1985: 335). Assessing fit as conceptualized in the systems approach requires measuring the deviations of a real organization from one or more ideal-type organization. The ideal types are represented by multivariate ideal profiles that provide the correspondence between the verbal descriptions of the ideal types and the measures used to assess real organizations. Real organizations' deviations from ideal types can be assessed with profile analysis (cf. Cattell, 1949; Miller, 1978), which is described more fully in Appendix A. Modeling Equifinality The third step in modeling a configurational theory is to state explicitly the relevant interpretation of the equifinality assumption, of which Katz and Kahn wrote that "a system can reach the same final state [e.g., the same level of organizational effectiveness] from differing initial conditions and by a variety of paths" (1978: 30). This assumption allows a "feasible set of equally effective, internally consistent patterns of context and structure" (Van de Ven & Drazin, 1985: 335, emphasis added). The assumption of equifinality is implicit in configurational theories because they identify multiple ideal types of organization that maximize fit

1202

Academy of ManagementJournal

December

and effectiveness. For example, each of Mintzberg's five ideal types is posited to be maximally effective. Thus, a researcher's task is not to determine if equifinality should be modeled, but rather to determine what interpretation of the assumption should be modeled. Different plausible interpretations of equifinality can be modeled by identifying constraints on the set of ideal types that an organization can adopt and be effective. One constraint on the feasible set of equally effective organizational forms is the number of ideal types identified in a theory. All configurational theories initially identify a discrete set of ideal types. For example, Mintzberg initially identified five ideal types, and Miles and Snow (1978) initially identified four. Many configurational theories are complicated, however, by the inclusion of an assertion that hybrids might be effective. Asserting that hybrids are effective allows for effective organizational forms that are combinations of two or more of the initial ideal types. Mintzberg implied that hybrid types existed when he argued that there may be multiple combinations of his five ideal types (1979: 474) and explained that some organizations may be forced to adopt hybrid forms when they must respond simultaneously to conflicting contingencies. Thus, configurational theories may be interpreted either as restricted to the initial ideal types posited or as allowing hybridization among the ideal types. A second constraint on the set of effective organizational forms is the presence of contingency factors that determine the ideal type of organization that a real organization must resemble to be maximally effective. When contingency factors are not identified, the organization may adopt any of the ideal types defined by the theory and remain effective. When important contingency factors are identified, however, the form that an organization can adopt to be maximally effective may be restricted to a single ideal type. For example, Mintzberg (1979) argued that an organization's context may constrain its choice of structure. Thus, configurations of contextual factors may restrict the selection of structures, strategies, or both. Four Configurational Interpretations of Fit and Equifinality Four plausible interpretations of the equifinality assumption incorporated in configurational theories can be specified by crossing the two constraints described above. Each of these different interpretations of equifinality can be integrated with the systems approach to fit. This integration results in four models of configurational fit, each of which may be consistent with a different substantive interpretation of specific configurational theories. Ordered from least to most complex, these four models are: ideal types fit, contingent ideal types fit, contingent hybrid types fit, and hybrid types fit. A summary of the models follows; Appendix B gives formulas for them. Ideal types fit. In the ideal types model of configurational fit, a theory posits a small, finite set of equally effective ideal types of organizations. It is assumed that (1) an organization is free to adopt any one of these ideal types and (2) the only effective types are the posited ideal types. Hybridization is not allowed. In this model, fit is conceptualized as consistency across the

1993

Doty, Glick, and Huber

1203

relevant dimensions and modeled as the lack of deviation from the one ideal type that an organization most closely resembles. Thus, an organization will be effective if it closely resembles any one of the ideal types in the theory. The underlying logic of this model of configurational fit is that organizational effectiveness results from the consistency of the relevant contextual, structural, and strategic factors. For example, Miles and Snow (1978) argued that an organization can mimic any of the ideal types that they identified because organizations can enact the contextual, structural, and strategic alignment necessary for effectiveness. Contingent ideal types fit. A second model of configurational fit is the contingent ideal types model. This is the most restrictive model of configurational fit. First, the feasible set of effective organizational forms is restricted to the ideal types explicitly identified in a theory because hybridization is not allowed. Second, contingencies constrain the number of forms that an organization can adopt. A theory positing contingent ideal types defines a finite number of ideal types of contexts and a single ideal type of organization structure or strategy that is appropriate for each ideal-type context. Thus, an organization must mimic the one ideal type that is most congruent with the contingencies facing it. This model of configurational fit is appropriate for testing theories positing that contingency factors constrain an organization's choice of form. Typically, an assumption that contextual factors are the contingencies that organizations face underlies such theories, as well as an assumption that existing organizations have little or no ability to influence their contexts. For example, Mintzberg (1979, 1983) argued that an organization's context may determine the form that the organization must adopt to remain effective. Such theories imply that organizational effectiveness results from both the consistency of structural and strategic factors and the congruence of the structural and strategic factors with contextual factors. Contingent hybrid types fit. The contingent hybrid types model represents a third possible definition of configurational fit. Unlike the two previous models, this model allows hybridization among the initial ideal types and shifts from positing a finite number of ideal types to positing continua of ideal types formed by hybrids among the specified ideal types. This model of fit defines continua of contexts and a single hybrid type of organization design that is appropriate for each context. Thus, there are an infinite number of hybrid contexts paired with an infinite number of hybrid designs, but the one hybrid type that an organization must mimic to remain effective is determined by the unique contingencies that the organization faces. The assumption of an infinite number of hybrid types distinguishes the contingent hybrid types model from the contingent ideal types model, the latter being limited to a finite number of explicitly defined ideal types. The contingent hybrid configuration model posits that performance will be higher for an organization when the deviation between the organization's design configuration and each ideal type of design configuration is directly proportional to the deviation between the organization's context and the contextual configurations defined for each ideal type of design.

1204

Academy of Management Journal

December

The contingent hybrid configurations fit model is appropriate for testing theories such as Mintzberg's (1979, 1983) and Miles and Snow's (1978) that link specific organizational forms with specific types of contexts and hypothesize that some hybrid forms may also be effective. This model of configurational fit allows organizations to deviate from the ideal types explicitly identified in a theory. To maximize effectiveness, an organization should most closely resemble the one hybrid type that its context determines. Hybrid types fit. The hybrid types model of configurational fit is the least restrictive of the four models. The hybrid types model implies that many different effective hybrid combinations can arise from the congruence between a constructed environment and a chosen organizational configuration. The feasible set of effective organizational forms includes an infinite number of hybrids among the ideal types identified in a theory, and the organization's choice among these types is not constrained by contingency factors. Thus, an organization is free to mimic any hybrid of the initial ideal types and remain effective. The implications of the hybrid types model of configurational fit appear to run counter to configurational theories that posit a small set of ideal-type organizational forms that maximize effectiveness (Miller, 1981, 1987; Miller & Friesen, 1984; Miller & Mintzberg, 1983). However, when configurational theories allow hybridization among the ideal types and do not limit the pattern of permissible hybridization or constrain an organization's choice among hybrid types, they are implicitly adopting the assumptions depicted in the hybrid types model. The models as simultaneous processes. The presentation of the four models of configurational fit seems to imply that the models represent mutually exclusive processes. It is possible, however, that the different models represent complementary processes. For example, the ideal types described in a theory may be the most effective forms of organization. Hybrid configurations, however, may be the most effective forms that a particular organization can adopt given the organization's history and other constraints on its ability to mimic the ideal types (Gresov, 1989). In this case, the best alternative for an organization is to adopt a form that is simultaneously closer to both an ideal type and some of the hybrids. In this example, organizational effectiveness depends on both (1) the similarity between the organization and the appropriate hybrid type and (2) the similarity between the organization and the appropriate ideal type. Thus, models of configurational fit may best be used in combination to test configurational theories. The remainder of this article presents two studies that apply these configurational fit models to two different configurational theories and discusses synergies between the two studies. STUDY 1: THE STRUCTURING OF ORGANIZATIONS Mintzberg (1979, 1983) presented both a typology and a theory. This distinction is important because the two are not the same (Bacharach, 1989; Blalock, 1969; Scott, 1981). A typology is a device that can be used to de-

1993

Doty, Glick, and Huber

1205

scribe phenomena such as organizations, structures, strategies, and environments. As a typology, Mintzberg's work provides a rich descriptive tool that identifies five potentially effective configurations of design and contextual factors. In contrast, a theory is a set of formal logical arguments that specify relationships among constructs or variables (Bacharach, 1989; Dubin, 1969; Kerlinger, 1986). As a theory, Mintzberg's work presents a series of logical arguments that result in specific predictions about organizational effectiveness as a function of the degree of similarity between a real organization and one or more of the ideal types. We develop hypotheses derived from the internal logic of the typology in the next section and hypotheses based on the theory in a subsequent section. Mintzberg's Typology Mintzberg (1979, 1983) presented his typology in three parts: (1) a set of design factors that can be used to characterize an organization's structure, (2) a set of contingency factors that can be used to characterize an organization's context, and (3) five ideal types of organizations described in terms of the design and contingency factors. Design factors. Mintzberg (1979, 1983) discussed primary and secondary design factors. Three primary design factors critical to each of the ideal configurations are the key coordinating mechanism, the key part of the organization, and the type and degree of centralization. The secondary design factors include bureaucratic characteristics such as formalization, specialization, and hierarchy. Although each of the secondary design factors is less important than each of the primary design factors, in combination these bureaucratic characteristics compose an important set of dimensions for defining the ideal configurations. Coordinating mechanisms are processes used to coordinate activities throughout an organization. Five unique processes are included in this primary design factor. Direct supervision occurs when one individual takes responsibility for supervising and coordinating the work of others. Standardization of work is accomplished by routinizing the various tasks performed by operating personnels. Standardization of skills achieves coordination by insuring that employees have the proper mix of skills and the required level of competency. Standardization of outputs is realized by specifying acceptable levels of outputs and quality. Finally, mutual adjustment refers to coordination through informal communication processes among employees. The second key design factor is which part of an organization is dominant. Mintzberg identified five organizational parts. The strategic apex includes a firm's chief executive officer (CEO), its top management team, and the employees who provide direct support to the CEO and the top management team. The middle line includes managers below the top management team down to and including first-line supervisors. The operating core includes the employees who perform the basic work in the organization. The technostructure includes analysts and engineers who effect standardization

1206

Academy of Management Journal

December

in the organization. Finally, the support staff includes employees who provide services such as legal counsel and personnel services to the rest of the organization. The final key design factor is centralization. Mintzberg (1979) identified four patterns of decentralization. Vertical decentralization occurs when decision-making authority appears low in the organizational hierarchy. Horizontal decentralization is a concentration of decision-making authority among members of the technostructure or the support staff rather than among line managers. Selective decentralization occurs when the authority to make selected decisions is decentralized, but authority over other decisions is retained by the strategic apex. Finally, parallel decentralization occurs when the authority to make many kinds of decisions rests in the same level of the organization (such as middle management). Contextual factors. Multiple factors define the contextual configurations in Mintzberg's (1979, 1983) theory, including an organization's age and size and attributes of its environment and technology. Mintzberg conceptualized environment along the two dimensions proposed by Duncan (1972): simple versus complex, and stable versus dynamic. Two dimensions of techand sophistication. Regulation is denology are most relevant-regulation fined as the level of control exerted on employees by the equipment needed to perform their work and the structure of the work. As the control of the equipment over the operator increases, technological regulation increases and employee control decreases. Sophistication is defined as the level of knowledge needed to understand the technical system, but not to the level of knowledge required to operate the equipment. As the knowledge needed to understand the system increases, the level of sophistication increases. Five ideal types. Mintzberg (1979, 1983) synthesized the preceding design and contextual factors into the five ideal types of organizations: simple structure, machine bureaucracy, professional bureaucracy, divisionalized form, and adhocracy. Each of the five ideal configurations is a unique combination of organizational design and contextual factors. For example, organizations with simple structures have very active entrepreneurial top management teams that maintain centralized control of all operations and coordinate activities through direct supervision. The ideal context for a simple structure is an unsophisticated, nonregulating technology in a simple and dynamic environment, and the firm should be young and small. Machine bureaucracies tend to be older, larger organizations operating with routine technologies in relatively stable environments. A machine bureaucracy is very high on standardization of work and moderately high on other forms of standardization, and it is dominated by its technostructure, limited in overall decentralization, moderate in terms of selective decentralization, and very bureaucratic. Professional bureaucracies, such as universities and large accounting firms, also operate in stable environments and rely on standardization of skills and extensive training and indoctrination. The divisionalized form is a corporate suprastructure imposed over a number of business units that tend to be internally structured as machine bureaucracies. Finally,

1993

Doty, Glick, and Huber

1207

adhocracies tend to be very organic professionalized organizations that emphasize coordination by mutual adjustment and liaison devices. Adhocracies typically operate in turbulent environments. The internal logic of Mintzberg's typology. The primary hypothesis associated with Mintzberg's typology concerns the logical relationships between the design and contextual configurations associated with each ideal type of organization. Throughout his 1979 book, Mintzberg argued that each design configuration corresponds to a unique contextual configuration. Thus, Hypothesis la: Each of Mintzberg's five ideal-type configurations will be associated with a unique contextual configuration. Mintzberg noted, however, that organizational design changes may follow changes in an organization's context (1979: 223). If a large lag effect exists, a strong relationship between the design configurations and the contextual configurations would not be expected to emerge in a cross-sectional sample unless both design and context are highly stable over time. Instead, we would expect that an organization's design at any given time will be related to its context at some earlier time. Mintzberg did not specify the length of this lag period, and we know of no systematically established estimate. Our opinion is that one year should be sufficient to allow most small to moderate-sized organizations to respond to contextual changes. Thus, we specified a one-year time lag between two measurement points (time 1 and time 2) and formulated the following: Hypothesis lb: Each of the five design configurations measured at time 2 will be associated with a unique contextual configuration measured at time 1. Mintzberg's Theory Another theme that permeates Mintzberg's (1979, 1983) work is that the typology is one of effective organizational forms. Therefore, we might not find the relationships specified in the previous hypotheses unless we restricted the analysis to highly effective organizations because less effective organizations might dilute the hypothesized relationships. However, we could draw on Mintzberg's theory and predict that organizations with the correct pairing of design and contextual configurations would be more effective than organizations with an incorrect pairing of design and contextual configurations. Thus, Hypothesis 2: An organization with the correct pairing between its design and contextual configuration will be more effective than an organization with an incorrect design-context pairing.

1208

Academy of ManagementJournal

December

Because there may be a lag between the time that an organization achieves a correct design-context pairing and any noticeable impact on organizational effectiveness, this effectiveness hypothesis was also tested longitudinally.1 In addition to simple statements about the relative effectiveness of organizations, Mintzberg's (1979, 1983) theory contains arguments about the complex processes that lead to organizational effectiveness. These arguments lead to two major theoretical assertions: (1) organizational effectiveness is a function of the fit across multiple dimensions of organizational design and context and (2) fit is at a maximum in the five ideal types of organization the theory identifies. These fit assertions can be tested with the configurational fit models developed in this article. Three of these models, the ideal types model, the contingent ideal types model, and the contingent hybrid types model (see Appendix B), can be used to test Mintzberg's theory. The fourth model, the hybrid types model, is not consistent with the theory because Mintzberg (1979) argued that the only reason for organizations to adopt hybrid forms is to respond to conflicting contextual forces. Ideal types fit. The ideal types model of fit can be restricted to design factors and used to represent the configuration hypothesis in Mintzberg's theory, in which he stated that "effective structuring requires an internal consistency among the design factors" (1979: 219). This hypothesis posits higher effectiveness in organizations possessing internally consistent configurations of design characteristics, with the most internally consistent design configurations defined by the five ideal types. The more similar an organization's design is to any one of the ideal designs, the more effective the organization is expected to be. Thus, we hypothesize that the ideal types model of fit will predict organizational effectiveness. Since the degree of fit achieved in an organization may have either an immediate or a delayed effect on measures of organizational effectiveness, this hypothesis was tested both cross-sectionally and longitudinally. Hypothesis 3: The greater an organization's ideal types fit, the greater the organization's effectiveness. Contingent ideal types. The contingent ideal types model of fit is consistent with the extended configuration hypothesis in Mintzberg's theory, in which he stated that "effective structuring requires a consistency among the design parameters and the contingency factors" (Mintzberg, 1979: 220). Though Mintzberg did acknowledge that organizations have some ability to control their contexts, he limited this assumption of strategic choice (Astley & Van de Ven, 1983; Child, 1972) to "certain aspects of its [the organization's] situation" (Mintzberg, 1979: 220). Thus, we assumed that the ideal-

1 To conserve space, the longitudinal versions of this and subsequenteffectiveness hypotheses are not stated explicitly, but they were explicitly tested, and we discuss those results.

1993

Doty, Glick, and Huber

1209

type configuration appropriate for an organization is dependent on its context and hypothesized that a contingent ideal types model of configurational fit will predict organizational effectiveness either concurrently or longitudinally. Hypothesis 4: The greater an organization's contingent ideal types fit, the greater the organization's effectiveness. Contingent hybrid types. The contingent hybrid types model of fit is consistent with the arguments in the final chapter of Mintzberg's 1979 book. In this chapter, Mintzberg relaxed the importance of the five ideal types and suggested that some hybrid forms of organization may be effective. Specifically, he argued that "the set of five configurations can be treated as the basis for describing structural hybrids" and that "nothing precludes a combination of the characteristics of three or more configurations" (Mintzberg, 1979: 474-475). Organizations are expected to be more effective when they adopt these hybrid forms in response to conflicting contingencies. Thus, we also hypothesized that a contingent hybrid types model of configurational fit will predict organizational effectiveness. Hypothesis 5: The -greater an organization's contingent hybrid types fit, the greater the organization's effectiveness. Simultaneous processes. In our presentation of the models of configurational fit, we discussed the possibility that the models of fit might represent simultaneous rather than mutually exclusive processes. Thus, the models should be used together to predict organizational effectiveness. As a consequence, Hypothesis 6: The greater an organization's ideal types, contingent ideal types, and contingent hybrid types fit, the greater the organization's effectiveness. STUDY 1: METHODS Organizations The organizations in study 1 were contacted as part of a larger longitudinal project investigating changes in organizational design and effectiveness (for a complete description, see Glick, Huber, Miller, Doty, & Sutcliffe, [1990]). The larger study required that organizations participate in four rounds of interviews and two rounds of surveys. A diverse set of organizations, including independent firms, strategic business units of corporations, medium-sized governmental units, and others, was selected from the health care, manufacturing, education, financial, and retail industries in four regions of the United States. Organizations were included in the study only if they (1) had at least two levels of management, (2) had external constituencies other than a parent organization, (3) retained primary control over their

1210

Academy of ManagementJournal

December

strategies and designs, and (4) initially agreed to participate in all rounds of the data collection. In many cases, an organization was initially selected because one or more members of the research team had a relationship with a top manager. Though this approach did not generate a random sample from a theoretically defined population of organizations, it did allow inclusion of a diverse set of industries and subsets of relatively homogeneous organizations within each industry. Further, this approach greatly increased the likelihood that the organizations would participate throughout the study. The data were collected via two surveys separated by approximately one year. We mailed the surveys to each organization's top manager or, in a few cases, to a member of the top management team the top manager identified as more knowledgeable about the organization's design. (This typically occurred in medical organizations formally headed by a part-time physician who deferred to a chief operating officer on organizational issues.) In all cases, during a subsequent interview the appropriate individual reported that he or she had completed the survey. By collecting data from the top managers or other designated top administrators rather than from other individuals, we reduced the effects of informational and motivational biases associated with hierarchical levels (Hambrick, 1981; Ireland, Hitt, Bettis, & DePorras, 1987) and functional differences (Dearborn & Simon, 1958; Zajac & Wolfe, 1966). Using only a single key informant provides four advantages (Glick et al., 1990): (1) high probability that the most knowledgeable informant in each organization is providing data, (2) low variation in informational and motivational biases across organizations that result from the use of informants occupying positions that do not exist in all organizations (Seidler, 1974), (3) high probability that organizations will agree and actually continue to participate in the study, and (4) a high number of organizations that can be included within the constraints of a fixed budget. Although there are also disadvantages associated with the single informant methodology, the tradeoffs were reasonable in the context of the overall study (cf. Glick et al., 1990). A total of 146 and 112 completed surveys were returned at time 1 and time 2, respectively. We eliminated surveys that were missing responses on any of the constructs included in this study from the analysis. This process left 128 organizations providing data at time 1, 104 organizations at time 2, and 85 organizations in the longitudinal analyses. The mean and median sizes of the organizations were 949 and 226 employees at time 1, 1,187 and 513 employees at time 2, and 1,248 and 500 employees in the longitudinal study. For-profit organizations constituted 70 percent of the study group at time 1, 57 percent at time 2, and 63 percent in the longitudinal data. The mean and median ages of the organizations were 42 and 33 years at time 1; 51 and 43 years at time 2; and 49 and 43 years in the longitudinal study. Many of the changes across time occurred because a number of young, small firms in the electronics industry dropped out of the study between times 1 and 2.

1993

Doty, Glick, and Huber

1211

Measures Organizational effectiveness and the specific dimensions used to characterize Mintzberg's (1979, 1983) ideal types were assessed with questionnaire measures adopted or developed from previous studies. Whenever possible, data from the current study were factor-analyzed (Mulaik, 1972) to examine support for the a priori scales. In a few cases, items were dropped from the scales for one or more of the following reasons: cross-loading on inappropriate factors following oblique rotation, low communalities, instability of factor loadings between time 1 and time 2, and low interitem reliabilities. The measures described below met the standard criteria for scale development (DeVellis, 1991; Mulaik, 1972). Organizational effectiveness was assessed along five distinct dimensions adapted from Quinn and Rohrbaugh's (1983) model of organizational effectiveness: resource acquisition, efficiency, human relations, quality, and costs. Multiple conceptualizations of effectiveness were required because organizational effectiveness is multidimensional (cf. Cameron & Whetten, 1983), the organizations in the study were diverse, and the ideal types identified by Mintzberg differ (1979). Each effectiveness dimension was measured with two to four Likert-type items. A complete list of the questionnaire items for study 1 is available from the authors. Age and size were assessed with open-ended questions about the year of an organization's founding and its number of full-time employees. We applied natural logarithmic transformations to both age and size to normalize the distributions and make the measures more consistent with the theory. Environmental turbulence was assessed with a six-item scale that captured turbulence in three environmental sectors. Because the a priori items had low communalities and weak factor loadings, environmental complexity was measured with a single item. Analyzability and the number of exceptions to routine operations in the technology were each measured with two pairs of items drawn from the scales of Withey, Daft, and Cooper (1983). Analyzability and the number of exceptions were used as proxies of regulation and sophistication, respectively. These two dimensions parallel Perrow's (1967) characterization of technology, which has dominated organizational research on technology and structure (Fry, 1982; Van de Ven & Ferry, 1980). As regulation increases, the frequency of exceptions decreases. As the sophistication of a system increases, analyzability decreases. Thus, analyzability and the number of exceptions were used in place of sophistication and regulation. The five key coordinating mechanisms were measured with one to three items each. The stem of the main question asked an informant, "To what extent does your organization coordinate activities through. ..." The dominance of each part of an organization was assessed with a measure of the percentage of employees in each of the five organizational parts. Although Mintzberg defined domination in terms of organizational influence rather than size, theories of power in organizations (cf. Pfeffer, 1981) suggest that

1212

Academy of Management Journal

December

the number of employees in a given part is likely to be higher if that part is dominant. Vertical and selective decentralization were captured by questions asking about the extent of delegation for each of seven types of decisions. The mean response on these items reflects vertical decentralization, and the interitem variance reflects selective decentralization. Rating of Ideal Types The verbal descriptions of the ideal types that appear in most theories are rich but not well suited for comparing the types on specific dimensions. Researchers should quantitatively specify ideal profiles that correspond to the qualitative descriptions of the ideal types and do so in terms of the measures used to assess real organizations. Doty (1990) argued that the best approach for defining ideal profiles for theories that develop a priori ideal types is the method of theoretical specification, which relies on ratings by experts who are very familiar with the descriptions of the ideal types provided by the original theorist. The expert raters determine the level of each construct that best represents each ideal type and then assign the appropriate value on the measures that correspond to each construct. The level of a construct in the ideal profile for a given ideal type is then defined as the mean of the values that the raters assigned to the corresponding measures. The two senior authors of the present study rated each of Mintzberg's (1979) ideal types on each questionnaire measure. Both raters had read multiple articles and books by Mintzberg and predicted that this study would support his theory. Although it is usually preferable to have naive raters, this rating task required a thorough understanding of the typology. After reviewing Mintzberg's (1979) description of each ideal type, the raters independently estimated the value for each measure that would be selected by the top manager of an organization conforming to an ideal type. On measures that were anchored in absolute values, such as age and size, the raters consulted the observed frequency distributions from the set of organizations to anchor Mintzberg's labels of young, large, and so forth, with specific values consistent with the organizations included in this study. Each rater also reviewed the distribution of his own ratings across the ideal types to increase their internal consistency. Initial interrater reliabilities were computed using an intraclass correlation formula [ICC(3,k)] (Shrout & Fleiss, 1979) and were reviewed for major disagreements. After discussing the rationales for the ratings, the raters made a few adjustments and recomputed the interrater reliabilities. Initial and final interrater reliabilities averaged across scales were .74 and .87, respectively. Table 1 reports the final interrater reliabilities for each design and contextual variable, along with the mean of the final ratings on each construct for each of the ideal types. The mean across the two raters of the final ratings provides the operational definition of each ideal type as a configuration of design and contextual factors. Deviation Analyses The design of this study required assessing the similarity of real organizations to Mintzberg's ideal types along multiple dimensions. One ap-

TABLE 1 Reliabilities and Rater Estimates of Mintzberg's Ideal Configur


Rater Estimates of Ideal Configurations Variables Context Environmental complexity Environmental turbulence Analyzability Number of exceptionsc Age Number of employees Design Vertical decentralization Selective decentralizationd Direct supervision Standardization of work Standardization of skills Standardization of outputs Mutual adjustment Formalization Hierarchy of authority Specialization Operating core Middle line Technostructure Support staff Strategic apex
a

Scale Reliabilitiesa

Simple Structure 1.50 6.75 6.50 3.00 3.50 50.25 1.79 0.79 7.00 1.75 3.00 2.67 5.83 1.83 0.00 1.50 85.50 3.00 0.00 1.50 10.00

Machine Bureaucracy 2.00 3.08 6.50 6.75 77.00 1,100.00 4.29 1.71 4.50 6.75 4.50 5.00 2.16 6.67 4.00 7.00 74.00 10.00 9.00 5.00 2.00

Professional Bureaucracy 7.00 3.08 5.25 3.00 24.50 285.25 5.64 1.60 2.50 4.75 7.00 4.33 4.83 5.00 1.50 6.00 86.00 3.50 0.50 9.00 1.00

Divisionalized Form 5.00 3.83 5.75 6.00 87.00 5,800.00 5.36 1.48 3.00 5.50 4.50 6.67 2.50 6.00 5.00 4.50 67.00 20.00 5.50 6.50 1.00

0.65 0.75 0.71

0.90

0.79 0.60 0.67 0.84

b We calculated reliability using ICC(3,K) following review for major disagreement (Shrout & Fleiss, 197

Statistics are Cronbach's alphas. Variables for which reliabilities do not appear used single-item scales

c This variable was reverse-scored. d Scale reliability is not appropriate as the measure is not additive.

1214

Academy of ManagementJournal

December

proach that is appropriate for assessing the similarity of two objects that are described in terms of multiple dimensions is deviation, or profile, analysis. The smaller the measured distance between the objects, the greater the similarity of the objects. Typically, deviation analysis uses a weighted Euclidean distance formula to assess the similarity between objects. Thus, the specification of the weighting matrix is a critical concern because the weighting matrix determines the relative importance of the variables of interest (Doty, 1990; Van de Ven & Drazin, 1985). The weighting scheme used in this study controlled for the different number of scales used to measure the four contextual factors and the four design factors. We weighted each of the scales used to measure a design or contextual factor by the inverse of the number of scales included for the factor, resulting in a unit weighting for each factor. For example, technology and environment were each measured with two scales. Thus, each of the scales associated with these factors was weighted by .5 so that the technology and environment factors each received a total weight of 1. Each of two remaining contextual factors, age and size, received a weight of 1 because each factor was assessed with a single measure. In terms of theoretical importance, this weighting scheme insures that the importance of each design and contextual factor is equal and that the theoretical importance of the set of design factors is equal to the theoretical importance of the set of contextual factors. Deviation scores were calculated to assess the distance between each organization's design and contextual configuration and the design and contextual configurations of each of the five ideal types of organization Mintzberg identified. We used the deviation scores in two ways: first, to classify organizations' designs and contexts in terms of Mintzberg's ideal types, and second, as part of the measures of configurational fit. Prior to conducting the deviation analysis, we standardized all measures of the independent variables to remove effects resulting from differences in scale variances. Classifying organization designs and contexts. The first two sets of hypotheses involved classifying each organization into a categorical definition of the typology. Thus, the organization's design was classified as the design of the ideal type with the smallest deviation score. Similarly, the context of each organization was classified as the ideal context that was most similar to the organization's context. Calculation of configurational fit. Fit is conceptualized in terms of lack of deviation between the multidimensional context and design configurations of the ideal type and the real organization. We calculated the three configurational fit measures using the mathematical formulations of the models presented in Appendix B. Ideal types fit was measured by the additive inverse of the deviation between an organization's design and the most similar of the ideal-type configurations. Ideal types fit is high to the extent that an organization's design is similar to the design of any one of Mintzberg's ideal types. According to the contingent ideal types model, an organization's design

1993

Doty, Glick, and Huber

1215

is contingent on its context. We calculated this fit in two stages. First, the appropriate design was determined by finding the minimum deviation between the organization's contextual configuration and the five ideal-type contextual configurations. Then, the contingent ideal types fit score was computed as the additive inverse of the deviation between the organization's design and context and the indicated ideal-type design and context. Calculation of the contingent hybrid types measure of fit also required a two-step process. First, the appropriate hybrid type for each organization was identified by calculating the deviation of the organization's context from each of the ideal contexts. The deviations from the ideal contexts were then used to determine the hybrid type with a context most similar to the organization's context. The contingent hybrid types fit score was then computed as the additive inverse of the organization's deviation from the appropriate hybrid type. STUDY 1: RESULTS Table 2 presents the scale reliabilities and correlations among the effectiveness dimensions.2 The mean Cronbach's alpha for the set of effectiveness measures was .77 at time 1 and .80 at time 2. All the cross-sectional and longitudinal correlations among the effectiveness dimensions are small to moderate. These results indicate that (1) the effectiveness measures are related but that each dimension of effectiveness represents a different aspect of organizational effectiveness and (2) each dimension of effectiveness varies across time. Tests of Hypotheses Derived from the Typology The tests of the hypotheses derived from Mintzberg's (1979, 1983) typology involve the simple relationships between classifications of the organizations' designs and contexts. Table 3, a contingency table, shows the crossclassification of each organization at time 1 by the ideal types that are most

2 The means and standard deviations of all questionnaire measures and the cross-sectional correlations among the measures used to compute the fit indexes are available upon request. Interrater reliabilities could not be assessed with a single informant. Three subsequent studies, however, used selected items and effectiveness scales from the current questionnaire. Two of the studies (Glick, Miller, & Huber, 1993; Sutcliffe, 1991) employed the top manager and the top management team as multiple informants and reported generally acceptable (x = .70) interrater reliability coefficients (Glick, 1985; Shrout & Fleiss, 1979). These results support the convergent validity of the current effectiveness measures. In the third study, managers who had firsthand knowledge of the strategic planning process but who were themselves not top managers proved to be unreliable as multiple informants about a firm's effectiveness and environmental turbulence (Ogilvie & Glick, 1990). This study of strategic planners reminds us to remain cautious in interpreting single informant data, but we believe that the use of an organization's top manager rather than a mixture of managers from different levels strengthened the validity of our measures.

TABLE 2 Correlations Among the Effectiveness Measures in Study


Time 1 Effectiveness Dimensions Time 1 Cost Efficiency Quality Human relations Resource acquisition Time 2 Cost Efficiency Quality Human relations Resource acquisition
aCronbach's

Cost .73 .29** .06 .07 .02 .45** .10 -.08 .10 .00

Efficiency

Quality

Human Relations

Resource Acquisition

Cost

Efficiency

.79 .34** .31** .06 .37** .37** .15 .28** .03

.79 .40** .06 .18 .33** .46** .29** .14

.83 .08 .09 .05 .27 .52** -.06

.69 -.05 -.01 -.06 -.09 .00 .65 .25** -.01 .17 -.00

.87 .46** .35** .22*

alphas appear on the diagonal. * p < .05, two-tailed test ** p < .01, two-tailed test

TABLE 3 Similarities to Mintzberg's Ideal Types


Contextual Configurations Design Configurations Simple structure Machine bureaucracy Professional bureaucracy Divisionalized form Adhocracy Total for contextual configurations Simple Structure 3 1 2 O 7 13 Machine Bureaucracy 1 5 1 0 1 8 Professional Bureaucracy 6 10 11 0 19 46 Divisionalized Form 2 21 16 10 4 53

1218

Academy of Management Journal

December

similar in terms of the contextual and design configurations.3 For example, the observed design configuration of a total of 38 organizations is most similar to the ideal configuration of the machine bureaucracy. However, only 5 of these organizations have an observed contextual configuration that is most similar to the ideal contextual configuration of the machine bureaucracy. Hypotheses la and lb, which are derived from Mintzberg's typology, predict that organizations will tend to fall on the diagonal of the contingency tables. Clearly, these data do not even approximate such a pattern. The results of a maximum likelihood "log linear" analysis (Kennedy, 1983) failed to refute the null hypothesis of independence between the designs and the contexts at both time 1 (X2 = 51.54, p < .01) and time 2 (X2 = 40.69, p < .01). At time 1, 74 percent of the organizations adopted ideal-type configurations that were inconsistent with their contextual configurations. At time 2, 65 percent of the organizations had inconsistent designs. Thus, results at neither time 1 nor time 2 supported Hypothesis la. Hypothesis lb predicts that an organization's design at time 2 will be related to the organization's context at time 1. Once again, the results of a maximum likelihood log linear analysis failed to refute the null hypothesis (X2 = 36.67, p < .01). In the longitudinal analysis, 73 percent of the organizations had designs at time 2 that were not consistent with their contexts at time 1. Thus, Hypothesis lb was not supported. Tests of Hypotheses Consistent with the Theory Effectiveness of alignment. The second set of hypotheses assessed the power of classification of organizations by Mintzberg's (1979, 1983) typology to differentiate more effective organizations from less effective organizations. Mintzberg clearly predicted that organizations adopting the design appropriate for their context would be more effective than organizations adopting inappropriate designs. We tested this prediction by classifying organizations into one of two groups, one for organizations with the appropriate designs and one for organizations with the inappropriate designs. Ten t-tests (Table 4) revealed no significant differences between the mean effectiveness of the two groups on any of the five effectiveness dimensions at time 1 or at time 2. Thus, Hypothesis 2 was not supported. Further, five additional tests revealed no significant differences in any of the effectiveness measures taken at time 2 between organizations with correct and incorrect contextdesign pairings at time 1. Thus, results did not support the longitudinal version of Hypothesis 2. Effectiveness of alternate models of fit. The remaining hypotheses derived from Mintzberg's (1979, 1983) theory contain predictions about an organization's degree of fit and effectiveness. We tested the three sets of
To conserve space, we have omitted the contingency tables for time 2 and the longitudinal analysis.
3

1993

Doty, Glick, and Huber

1219

TABLE 4 Results of Tests of Effectiveness as a Function of Alignment


Effectiveness Dimensions Time 1 Cost Efficiency Quality Human relations Resource acquisition Time 2 Cost Efficiency Quality Human relations Resource acquisition Longitudinal Cost Efficiency Quality Human relations Resource acquisition t -1.12 - 0.04 -0.17 0.51 1.29 0.89 0.89 0.14 -0.12 1.73 1.31 1.01 -0.27 0.09 0.64 df 126 126 126 126 126 102 102 102 102 102 83 83 83 83 83 p .26 .96 .85 .60 .20 .37 .37 .88 .90 .08 .19 .31 .78 .92 .52

hypotheses based on individual measures of configurational fit (Hypotheses 3-5) by calculating the zero-order correlations between the relevant configurational fit measures and the effectiveness measures. We tested Hypothesis 6 concerning the simultaneous process interpretation of the theory by calculating the canonical correlation between the relevant set of configurational fit measures and the set of effectiveness measures. In combination, those procedures allowed Mintzberg's theory to be tested with both a limited and a global conceptualization of organizational effectiveness. The tests based on the zero-order correlations assess the power of the theory to predict limited conceptualizations of effectiveness, and the tests based on the canonical correlations assess the power of the theory to predict a global notion of effectiveness. Table 5 presents the correlations among different measures of configurational fit. Ideal types fit and contingent ideal types fit are positively related both at time 1 (r = .64) and time 2 (r = .56). The two models were, however, based on different ideal-type configurations, as indicated in the test of Hypothesis la. Contingent hybrid types fit is positively related to both the other measures of fit at both times, but the correlations are moderate. All the correlations among the fit measures across time are significant but lower than the correlations within a single time period. These results support the assertion that the three models represent different interpretations of Mintzberg's theory and that the level of fit achieved in an organization can vary across time. Hypothesis 3 posits a positive relationship between the ideal types mea-

1220

Academy of Management Journal

December

TABLE 5 Correlations Among Fit Measures for Mintzberg's Ideal Types


Time 1 Ideal Types Contingent Ideal Types Contingent Hybrid Types Ideal Types Time 2 Contingent Ideal Types Contingent Hybrid Types

Types of Fit Time 1 Ideal types Contingent ideal types Contingent hybrid types Time 2 Ideal types Contingent ideal types Contingent hybrid types

.64** .60** .44** .29** .32** .61** .36** .27* .25* .43** .32** .54** .56** .54** .52*

* p < .05, two-tailed test ** p < .01, two-tailed test

sure of fit and the individual measures of effectiveness in cross-sectional data. None of the ten relevant correlations presented in Table 6 are significant. Thus, results do not support the cross-sectional version of Hypothesis 3. The longitudinal version of Hypothesis 3 posits a positive relationship between the ideal types measure of fit at time 1 and the effectiveness measures at time 2. Only the correlation with the efficiency measure of effectiveness is significant (r = .22, p < .05), and this correlation is relatively small. Although this one significant correlation should not be completely ignored, 80 percent of the hypothesized longitudinal relationships were not significant. Further, the omnibus multivariate test of the relationship between the ideal types measure of fit and the set of effectiveness measures is not significant (adjusted r2 = .03, n.s.). Thus, the preponderance of the evidence indicates that Hypothesis 3 should not be accepted. Hypothesis 4 predicts positive relationships between the contingent ideal types measure of fit and organizational effectiveness in cross-sectional and longitudinal data. Only one relationship is significant both at time 1 (r = -.21, p < .05) and time 2 (r = -.21, p < .05), but the direction of these significant correlations is opposite the predicted direction (see Table 6). The theory fares no better in the longitudinal test. None of the hypothesized relationships are significant. Thus, Hypothesis 4 was not supported crosssectionally or longitudinally. The tests of Hypothesis 5 are based on the correlations between the contingent hybrid types measure of fit and the measures of organizational effectiveness. The results in Table 6 show that only two of the cross-sectional correlations are significant (r's = - .18 and -.21, p < .05), but the direction

1993

Doty, Glick, and Huber

1221

TABLE 6 Correlations Between Fit and Effectiveness Measures Testing Mintzberg's Configurational Theory
Time 1 Effectiveness Dimensions Time 1 Cost Efficiency Quality Human relations Resource acquisition Time 2 Cost Efficiency Quality Human relations Resource acquisition Ideal Types .03 .08 -.02 - .05 - .06 .10 .22* -.06 .06 .02 Contingent Ideal Types - .01 -.21* -.01 - .06 .09
.14a

Time 2 Contingent Hybrid Types -.18* -.21* -.05 - .03 .08 .00 - .02 -.15 - .05 -.04 .02 .03 -.03 - .04 -.10 -.02 - .07 -.08 -.21* - .03 .08 .09 -.03 -.02 - .02 Ideal Types Contingent Ideal Types Contingent Hybrid Types

.08 .01 - .03 - .04

a This value becomes significant (p < .11, one-tailed test) when the alpha level is relaxed

in the post hoc power test. * < .05, one-tailed test p ** p < .01, one-tailed test

of both correlations is opposite prediction. None of the longitudinal correlations are significant. These results do not support Hypothesis 5.4 Effectiveness of simultaneous models of fit. The final test of Mintzberg's (1979) theory focuses on the possibility that the different models of fit represent simultaneous rather than mutually exclusive processes. Thus, we tested Hypothesis 6 by calculating the canonical correlations between the relevant sets of fit measures and the set of five effectiveness measures. For example, Hypothesis 6 was tested at time 1 by calculating the canonical correlation between the ideal types, contingent ideal types, and contingent hybrid types measures of fit at time 1 and the five effectiveness measures at
Another potential challenge to the validity of the effectiveness measures suggested by an anonymous reviewer is that not-for-profit organizations might pursue effectiveness goals that are not included in our measures. To examine this possibility, we excluded the not-for-profit organizations from the analyses. The new results are consistent with the previous results. More specifically, with 90, 59, and 55 organizations at time 1, time 2, and longitudinally, respectively, none of the 45 correlations in Hypotheses 3-5 are significant and in the predicted direction, and two of the correlations are significant and opposite the predicted direction. Thus, the null results reported in this study do not appear to be attributable to inappropriate effectiveness measures for not-for-profit organizations.
4

1222

Academy of Management Journal

December

TABLE 7 Canonical Correlations Between Sets of Fit Measures and the Effectiveness Measures Testing Mintzberg's Configurational Theory
Time Fit Assessed 1 2 1 Time Effectiveness Assessed 1 2 2 First Canonical Correlation .31 .26 .36 Wilks's Lambda p > .22 p > .82 p > .39

Adjusted rc2 .05 .02 .06

time 1. The results of the canonical analyses (see Table 7) revealed that neither of the cross-sectional relationships (adjusted rc2 = .05 and .02 at time 1 and time 2, respectively) nor the longitudinal relationship (adjusted rc2 = .06) was significant. Thus, Hypothesis 6 was not supported. STUDY 1: DISCUSSION The empirical results of this study are disappointing given the widespread acceptance and intuitive appeal of Mintzberg's (1979, 1983) work. We selected Mintzberg's theory for this research because we firmly expected to find strong support for it. Our inability to detect the specified relationship between design and contextual configurations in the contingency analysis suggests that Mintzberg's reasoning is not reflected in the organizational designs put in place by organizations in this study. Organizations do not appear to adopt design configurations that Mintzberg's typology identifies as most appropriate for their contexts. Although the theory has intuitive appeal, the results of this empirical test suggest that it requires revision, indicating that classifying organizations by their designs and contexts on the basis of Mintzberg's (1979, 1983) typology was not useful for making predictions about the relative effectiveness of the organizations in this study. Further, in contrast with the main hypotheses derived from Mintzberg's theory, the continuous measures of fit used in this study failed to allow prediction of any of the effectiveness measures. Given the preponderance of null results in this study, a critical concern is our confidence in interpreting them. To support any interpretation of these null results as refuting, rather than merely failing to support, Mintzberg's (1979, 1983) theory we (1) document the statistical power of the results, (2) address threats to the validity of the measures, (3) discuss other methodological issues, and (4) provide an alternative theory to replace Mintzberg's. Statistical Power The null results are not meaningful if a type II error (falsely accepting the null hypothesis) has been committed. To guard against type II error, we calculated the minimum true correlation that would lead to correctly reject-

1993

Doty, Glick, and Huber

1223

ing the null hypothesis at least 95 percent of the time (Cohen, 1977) for each cross-sectional analysis and for the longitudinal study. The 95 percent level was selected to ensure use of the same level of confidence in accepting and rejecting the null hypothesis. With 128 observations at time 1, a beta set equal to .05, and an alpha set equal to .05 for a one-tailed test of the directional hypotheses, the statistical power analysis supports the conclusion that there is less than a 5 percent chance that we falsely accepted the null hypothesis for strong or moderate (greater than .30) true correlations between fit and effectiveness. If the true correlations are less than .30, then our type II error is greater than .05, but we can safely conclude that all the true correlations between fit and the effectiveness dimensions are small or negligible (less than .30). To maintain the same sensitivity to moderate or strong true correlations (those greater than .30) and the same level of protection against type II error for the smaller number of observations at time 2 (N = 104) and in the longitudinal data (N = 85), the power calculations must relax the alpha-that is, the type I error rate-to .10 and .11, respectively. Relaxing the protection against type I error did not change our conclusions about the significance of any of the cross-sectional analyses and did change our conclusion about only one longitudinal effect: the positive correlation between time 1 contingent ideal types and cost effectiveness is significant (r = .14, p < .11, B < .05) with a one-tailed directional test. Thus, when we maintained a type II error rate of .05 and adjusted the type I error rate to maintain sensitivity to any true correlations greater than .30, 2 of 45 correlations were significant and in the predicted positive direction, and 4 correlations were negative (and significant with a two-tailed test) despite clearly directional hypotheses from the theory. Thus, we can be confident that at the .95 level the results do not support the hypotheses consistent with Mintzberg's (1979, 1983) theory. In other words, there is a 95 percent chance that the true correlations are small or negligible (less than .30), and may approximate 0.00. Validity The statistical power tests indicate that the null results are probably not the result of falsely accepting the null hypotheses (type II error) but do not address the possibility that the results are a methodological artifact. The use of perceptual measures from a single informant opens the study to the criticism that the results may be attributable to low construct validity. As noted above, all of the scales used in this study have undergone substantial psychometric evaluation. Further, the effectiveness measures have been subsequently validated in multiinformant studies, as described in footnote 2, and the technology scales have been shown to have good discriminant and convergent validity in several previous studies (Withey et al., 1983). The measures of environmental turbulence and complexity, however, do not have such rich historical validation and are discussed further below. The nomological validity of the turbulence measure was supported with post hoc analyses. Industry classification was used to predict environmental

1224

Academy of Management Journal

December

turbulence in the set of organizations that provided longitudinal data. The results of the one-way analysis of variance (ANOVAs) indicate that industry classification explained a significant amount of the variation in environmental turbulence (p < .01 and .05, time 1 and time 2, respectively). These results support the validity of the turbulence measure and indicate that there was meaningful variation across the set of industries included in this study. Two post hoc analyses were conducted to assess the validity of the environmental complexity measure. First, we predicted that not-for-profit organizations would face a more complex environment than for-profit organizations. At first this prediction may seem counterintuitive. Consider, however, two hospitals with different profit-seeking status. The not-for-profit hospital must deal with all the elements in the for-profit hospital's environment. In addition, the not-for-profit hospital has to deal with elements not found in the for-profit hospital's environment, such as city or county governmental agencies. These additional elements increase the relative complexity of the not-for-profit hospital's environment. The results of t-tests supported this prediction at both time 1 (t = 2.02, p < .05) and time 2 (t = 2.02, p < .05). Second, we predicted that health care organizations would face a more complex environment than manufacturing organizations. Results supported our prediction at time 1 (t = 2.19, p < .05) and marginally supported our prediction at time 2 (t = 1.80, p < .10). In combination, these post hoc results provide some support for the predictive validity of the environmental complexity measure. Methodological Issues One potential methodological limitation of this research concerns possible inadequacies in measuring the typology and assessing the design and contextual configurations. Any improvements in the validity and reliability of the measures would improve the probability of observing true effects. Some design and contextual measures were single items, and a few components of Mintzberg's (1979, 1983) model were not measured. A more complete description of the ideal types might lead to different results. Further, the two raters may have misinterpreted Mintzberg's typology systematically as they assigned values to the ideal-type configurations (see Table 1 for a complete list of the values assigned). Another group of organizational researchers might interpret Mintzberg's typology differently and find different results from the current data. A second possible limitation is the use of a one-year time lag for the longitudinal analyses. As noted in the development of the hypotheses, Mintzberg (1979, 1983) did not specify the appropriate lag between changes in contextual factors and realignment of design configurations. Although a one-year lag may be a reasonable estimate, it is preferable to specify an appropriate time lag that is based on strong theoretical specification and empirical validation. Given that multiple causal processes are involved and that each of these causal processes may unfold at a different pace, empirical validation of the time lags would require multiple waves of data collected at

1993

Doty, Glick, and Huber

1225

relatively short intervals (Monge, 1990). Such data collection was not possible. Thus, we can only conclude that Mintzberg's (1979, 1983) theory failed to receive support for cross-sectional analyses and for longitudinal analyses with a one-year lag. Future researchers should investigate alternative lags. An Alternative Theory The final step to interpreting the null results was to supply an alternative theory and test it with the current data. Support for the alternative theory would reduce or eliminate many of the methodological concerns previously discussed and significantly enhance our confidence in interpreting the null results. We viewed Miles and Snow's (1978) theory as an alternative and tested it in study 2. STUDY 2: STRATEGY, STRUCTURE, AND PROCESS Miles and Snow's (1978) theory seemed a good alternative to Mintzberg's theory for three reasons. First, although the theories differ in several respects, their logical structures are very similar. Both theories contain the argument that the way in which organizational attributes are configured determines effectiveness. Second, the fit-effectiveness assertions in Miles and Snow's theory are subject to multiple interpretations, like those in Mintzberg's theory. Third, the set of ideal types in Miles and Snow's theory differ from Mintzberg's ideal types both in the number identified and in how they are described. Thus, support for Miles and Snow's theory (1) would strengthen our confidence in the interpretation of the previous results as refuting Mintzberg's theory and (2) might inform us about potential theoretical limitations in Mintzberg's theory. Identifying inadequacies in Mintzberg's theory should help future configurational theorists develop better theories. Miles and Snow's Typology Miles and Snow (1978) identified four ideal types of organization: the prospector, the analyzer, the defender, and the reactor. We summarize the rich descriptions provided for the four types. The prospector is the most dynamic of the organizational forms Miles and Snow described. This type of organization operates in an environment characterized by rapid and unpredictable changes. Prospectors adapt to this turbulent environment by using high levels of environmental scanning (Daft & Weick, 1984) to identify opportunities for developing the new products or markets that are critical to their success. Because of the rapid rate of product development, flexible, nonroutine technologies are characteristically used. These technologies are typically associated with low levels of specialization and moderate to high levels of interdependence. Structurally, prospectors are very organic, with low levels of formalization and specialization and high levels of decentralization. Prospectors also possess relatively few hierarchical levels.

1226

Academy of Management Journal

December

The defender is a less dynamic form of organization operating in an environment that is more stable and predictable than that of the prospector. This more stable environment allows defenders to engage in less environmental scanning and more long-range forecasting and planning than do prospectors. The key to the defender's success is a focus on efficiency. Defenders compete by producing low-cost goods or services and obtain efficiency by relying on routine technologies and economies of scale gained from largeness. Defenders have much more mechanistic structures than prospectors and achieve coordination through formalization, centralization, specialization, and vertical differentiation. This bureaucratization tends to reduce the level of interdependence in defenders. The analyzer is a "unique combination of the Prospector and Defender types" (Miles & Snow, 1978: 68). Analyzers adopt some characteristics of prospectors and some characteristics of defenders, seeking effectiveness through both efficiency and new products or markets. This dual focus may result in increased size because the organizations must engage in both mass production and research and development. Analyzers may also exhibit higher levels of interdependence than either prospectors or the defenders. The fourth ideal type, the reactor, is characterized as an unstable organizational form that lacks a consistent context-structure-strategy alignment. A critical difference between the reactor and the other ideal types is that the reactor lacks a consistent, clearly articulated strategy. Thus, levels of product or market development and efficiency that vary across time characterize reactors. This lack of a consistent, clearly articulated strategy is due in part to their interpreting their environments as highly unstable and unpredictable. This interpretation may also explain the low levels of environmental scanning and long-range forecasting and planning that characterize reactors. Like Mintzberg, Miles and Snow (1978) developed a configurational theory that is based on a set of ideal types. Unlike Mintzberg, however, Miles and Snow did not make strong arguments about links between naturally occurring contextual configurations and structural configurations. Furthermore, although Mintzberg consistently defined his typology, the interpretation of Miles and Snow's organizational forms is more controversial. Thus, rather than developing specific hypotheses based on Miles and Snow's typology that are parallel to Hypotheses la and lb, we began study 2 by clarifying two unresolved theoretical issues in Miles and Snow's typology: (1) ambiguities about the status of the reactor as a residual category and (2) ambiguities about the relationships among the other three types. The reactor as a residual category. Miles and Snow's theory contains conflicting assertions about the reactor type. Some arguments suggest that the reactor is a fourth ideal type. For example, an entire chapter of the 1978 book is devoted to a description of the type. In other passages, however, the authors describe the reactor as "a 'residual' type of behavior in that organizations are forced into this response mode when they are unable to pursue one of the three stable strategies" (Miles & Snow, 1978: 93). An organization may be classified as a reactor when "management fails to align strategy,

1993

Doty, Glick, and Huber

1227

structure, and [context] in a consistent fashion" (Miles & Snow, 1978: 12). Thus, the reactor may be a residual category for organizations that are not effective rather than a fourth ideal type. The status of the reactor as a residual category has been addressed both by assumption and by empirical investigation. Some authors have simply assumed that the reactor is a residual category (e.g., Hambrick, 1981, 1983; Zajac & Shortell, 1989). Others have assumed that it is a unique category of organizations that can be effective (e.g., Snow & Hrebiniak, 1980). Only a few studies, however, have empirically addressed the status of the reactor type. For example, Smith and colleagues (1989) used cluster analytic techniques to identify a group of organizations that closely resembled the reactor. Ginn (1987) used discriminant analysis to differentiate four clusters of organizational forms, in one of which the organizations resembled the description of the reactor. Segev (1989) concluded that the four ideal types in the theory are unique types that are similar to, but distinct from, the ideal types Porter (1980, 1985) identified. Overall, the theoretical arguments of Miles and Snow and the current empirical evidence suggest that the reactor should be treated as a unique ideal type. Miles and Snow also made conflicting assertions about the effectiveness of reactors. Most of their arguments suggest that the reactor is not an effective type because fit is not achieved. However, the authors also suggested that the reactor can be effective when the proper alignment of contextual, structural, and strategic factors is achieved. Specifically, the reactor may be effective in benign environments (Miles & Snow, 1978: 154). The empirical evidence about the effectiveness of the reactor is mixed. Smith and colleagues (1989) concluded that organizations classified as reactors were not effective. Snow and Hrebiniak (1980), however, found that organizations classified as reactors were more effective than other organizations in highly regulated environments. These results indicate that the reactor is a viable organizational form and confirm the predictions of Miles and Snow that the reactor can be an effective ideal type. Relationships among the ideal types. An additional source of ambiguity in Miles and Snow's theory concerns the relationships among three of the ideal types. Many authors have interpreted Miles and Snow's theory to imply that three of the ideal types define a continuum in which the prospector and the defender are endpoints and the analyzer is the midpoint (Ginn, 1987; Hambrick, 1981, 1983; Zajac & Shortell, 1989). There is considerable justification for this assumption. Throughout the book, Miles and Snow drew on the continuum notion to facilitate the discussion. For example, the authors "believe that the Defender and the Prospector reside at opposite
ends of a continuum. .. . Between these two extremes we have observed a

third type of organization called the Analyzer" (1978: 68). Other passages of the 1978 book, however, can be interpreted as suggesting that the continuum notion may be intended as a heuristic device and should not be interpreted literally. For example, on some constructs the analyzer is not positioned between the prospector and the defender. Specif-

1228

Academy of Management Journal

December

ically, the authors describe the analyzer as exhibiting higher levels of differentiation and complexity than either of the other types (Miles & Snow, 1978: 76). Other passages are less specific but also suggest that a strict continuum is not intended. For example, the analyzer is a "unique combination" of the two types that represents a "viable alternative" to the other ideal types (Miles & Snow, 1978: 68). The analyzer has a dual technology, with one part resembling the technology of the prospector and the other part resembling the technology of the defender. Thus, the continuum notion may be a heuristic device that should not be interpreted literally. Adopting the assumption that three of the ideal types define a continuum may threaten the validity of a study for at least two reasons. First, assuming that three types define a continuum implies that the theory hypothesizes a single pattern of interdependence among constructs and that these three ideal types can be characterized by a single underlying dimension. Thus, the continuum assumption may not be consistent with the rich descriptions that Miles and Snow (1978) provided for each ideal type. Second, assuming that three of the types define a continuum seems to imply that the reactor is a residual category of ineffective organizations because it is not on the continuum of effective organizational forms. As discussed in the previous section of this article, treating the reactor as a residual category may not be consistent with the original theory, and it is not consistent with the findings of Snow and Hrebiniak (1980). Miles and Snow's Theory The configurational approach developed in this article appears to be appropriate for formally stating and empirically testing Miles and Snow's theory. The theory is based on the idea that there are ideal types of organizations and contains configurational arguments that draw on the notions of fit and equifinality. Thus, the theory is best tested by assessing the deviations between real organizations and the ideal types and then using those deviations to predict organizational effectiveness. Different arguments in the theory suggest that the theory's fit-effectiveness assertions should be tested with different models of configurational fit. Miles and Snow first observed that in some situations organizations are free to adopt any one of the ideal types defined in the theory (1978: 5-7). This line of argument, which draws on the assumption of strategic choice (Astley & Van de Ven, 1983; Child, 1972), implies that organizations are relatively free to create their contexts, structures, and strategies, and that organizational effectiveness results from the consistency among these factors. This implication of Miles and Snow's theory should be tested with the ideal types model of configurational fit. Miles and Snow also noted that strategic choice is sometimes constrained and that organizations may have limited ability to control their contexts (1978: 7-8). As a consequence, organizations may not be free to adopt any one of the ideal types identified in the theory and remain effective. Instead, the contingencies facing an organization may determine the form it

1993

Doty, Glick, and Huber

1229

must adopt to be effective. Thus, the number of forms that an organization can adopt is limited to the number of effective ideal types identified in the theory and constrained further by the presence of contingency factors. This interpretation of the theory is more closely aligned with an external control perspective and should be tested with the contingent ideal types model of configurational fit described earlier. Miles and Snow also raised the issue of hybridization. Their book initially proposes that there are four ideal types. Other arguments in the book, however, suggest that hybrid forms may be effective. For example, chapter 9 of the book, "Mixed Strategies and Structures," explains why some organizations may be forced to adopt hybrid forms. Thus, the theory may be interpreted either as restricted to the initial set of ideal types or as allowing hybridization. Acknowledging hybridization implies that the theory must also be tested with the contingent hybrid types and the hybrid types models of configurational fit. The arguments above suggest that the four models of configurational fit are consistent with different plausible interpretations of Miles and Snow's theory. Thus, Hypothesis 7: The greater an organization's ideal types fit, the greater the organization's effectiveness. Hypothesis 8: The greater an organization's contingent ideal types fit, the greater the organization's effectiveness. Hypothesis 9: The greater an organization's contingent hybrid types fit, the greater the organization's effectiveness. Hypothesis 10: The greater an organization's hybrid types fit, the greater the organization's effectiveness. In addition, we have argued that the models of fit may represent simultaneous processes rather than mutually exclusive processes. Thus, Hypothesis 11: The greater an organization's ideal types, contingent ideal types, contingent hybrid types, and hybrid types fit, the greater the organization's effectiveness. STUDY 2: METHODS Organizations The set of organizations used in study 2 is the same as the set of organizations used for study 1 at time 2.5 The number of organizations included in
5 Three of the five strategy measures were revised significantly between the time 1 and time 2 surveys because of measurement problems at time 1. For each of these three measures, scope, futurity, and strategic clarity, only a single item was common to the time 1 and time 2 surveys. Given the importance of these measures to the test of Miles and Snow's theory and the measurement problems in the time 1 data, we do not consider our data to support an adequate longitudinal test of Miles and Snow's theory.

1230

Academy of ManagementJournal

December

study 2 is slightly larger than that in study 1, however, (109 versus 104) because fewer organizations were eliminated because of missing data. Developing Ideal Profiles As in study 1, the ideal profiles used in study 2 were developed by expert raters. Unlike the raters for study 1, however, those in study 2 were three of the authors of the original theory. Each rater described each of the four ideal types on a separate questionnaire and then reviewed his ratings to ensure that each ideal type was positioned correctly relative to the other ideal types. Because values were missing, some constructs were specified as the mean of two raters' assignments, and two of the constructs were specified as the value reported by a single rater. Measures This study drew on a large set of constructs selected to meet two criteria. First, each construct was relevant to the descriptions of the ideal types presented by Miles and Snow (1978). Second, we expected each construct to vary between two or more of the ideal types of organization Miles and Snow identified. These constructs can be organized into contextual, structural, and strategic factors and dimensions of organizational effectiveness. We pretested all measures with the time 1 survey of study 1. On the basis of factor and reliability analyses, several measures were modified for the final data collection, which occurred approximately one year later, with the time 2 survey of study 1. The individual questionnaire items associated with each measure are available from the authors. Contextual constructs. The contexts in which organizations operated were assessed with five constructs. We used a three-item scale to measure environmental turbulence and six items for environmental predictability. Technology was conceptualized in two ways. Technological routinization (Perrow, 1967) was measured via a four-item scale. Four patterns of technological interdependence (pooled, sequential, reciprocal, and team interdependence) were each assessed with two items. These four patterns were combined into a single measure of overall interdependence in which each successive pattern of interdependence was weighted more heavily (pooled, 1; sequential, 2; reciprocal, 3; and team, 4). This weighting scheme is consistent with Thompson's (1967) arguments that each successive pattern of interdependence results in a greater level of overall interdependence in an organization. Finally, organizational size was defined as an organization's number of full-time employees and was assessed with a single open-ended question. We performed a logarithmic transformation prior to the analyses to normalize the distribution of organizational sizes and to make the measures more consistent with the theory. Structural constructs. Five structural constructs were assessed. Decentralization, formalization, specialization, and environmental scanning (Daft

1993

Doty, Glick, and Huber

1231

& Weick, 1984; Hambrick, 1981) were each assessed with a two- to four-item scale. Vertical differentiation, defined as number of hierarchical levels, was assessed with a single open-ended question. Strategic constructs. An organization's strategy was assessed along five dimensions: product-market development, focus on efficiency, scope (Porter, 1985), strategic clarity, and futurity (Miller & Friesen, 1984). Each construct was assessed with a two- to four-item scale. Organizational effectiveness. The effectiveness measures used in this study were identical to those in study 1. Analysis Strategy The analysis strategy in study 2 paralleled that of study 1: psychometric validation of the measures, calculation of fit measures based on the formulas in Appendix B, and comparison of a series of canonical correlations between the effectiveness measures and the set of four fit measures associated with each interpretation of the ideal types-four ideal types, three independent ideal types, and three ideal types that form a continuum. In study 2, the canonical correlations were used first to determine which interpretation of the ideal types was most empirically valid. Then a series of hierarchical canonical correlations was used to determine (1) which model or models of configurational fit resulted in the strongest support for Miles and Snow's (1978) theory and (2) how well the theory predicted individual effectiveness constructs. STUDY 2: RESULTS AND DISCUSSION Table 8 presents the mean value and standard deviation of each construct across the set of organizations. Table 8 also presents the value of each construct for each ideal type and the interrater reliability coefficient (Shrout & Fleiss, 1979) across the set of types. The reliabilities assess the means of the raters; the mean interrater reliability across the set of constructs was .83. Table 8 also presents the interitem reliability coefficients of the individual construct measures. The mean interitem reliability across the set of contextual, structural, and strategic constructs is .73. The interitem reliability of each of the effectiveness measures is .65 or higher, and the mean across the set of effectiveness measures is .80. The correlations among the measures are available from the authors. Testing Alternative Interpretations of Miles and Snow's Typology The theoretical controversies involving the number of ideal types and the relationships among the types suggested three possible interpretations of Miles and Snow's (1978) typology: (1) four ideal types, (2) three independent ideal types, and (3) three ideal types that form a single continuum. To re-

TABLE 8 and Standard Deviations of Constructs Across Organizations a Reliabilities, Means, and Snow's Ideal Types
Construct Context Environmental turbulence Environmental predictability Technological routinization Technological interdependence Size Structure Decentralization Formalization Specialization Vertical differentiation Environmental scanning Strategy Product-market development Focus on efficiency Scope Futurity Strategic clarity Interitem Reliabilitiesa .77 .73 .88 .77 Means 4.74 4.75 5.26 3.97 6.23 4.40 4.36 4.18 4.07 5.28 4.80 5.13 3.82 4.77 5.05 s.d. 1.25 1.40 1.15 0.91 1.58 1.01 1.28 1.11 2.96 1.13 1.06 1.00 1.05 1.34 1.06

Prospector 5.78 5.56 3.00 4.78 6.94 6.13 2.33 4.56 3.00 6.83 5.67 4.50 2.50 3.38 6.11

Ideal Profiles Ideal Profiles De Analyzer 4.83 4.89 4.12 5.12 7.34 4.73 4.50 4.83 5.00 5.50 4.41 4.91 4.00 5.16 5.33

.63 .88 .61 .83 .78 .72 .52 .68 .72

a Cronbach's alphas are shown. Where values are lacking, the measure was a single-item scale. b We calculated reliability using ICC(3,k) following review for major disagreement (Shrout & Fleiss, 1979 rater provided a rating.

1993

Doty, Glick, and Huber

1233

solve these controversies, we considered three criteria: the a priori arguments of Miles and Snow, the variance in effectiveness explained by the different interpretations of the theory, and the parsimony of the interpretation. Thus, for each of the three interpretations of the typology we examined the empirical relationships between the set of four fit measures and the set of effectiveness measures with canonical correlations. The dependent variables in each analysis were the effectiveness measures. The independent variables were four measures of fit representing the four interpretations of equifinality. Each of the canonical analyses adopted a different interpretation of the typology. We represented the interpretation of the theory suggesting that the reactor is a fourth effective ideal type by formulating each of the fit models with a set of four ideal profiles and represented the interpretation of the reactor as a residual category by formulating each of the fit models with three independent ideal types. The interpretation of the theory suggesting three ideal types that form a continuum was represented by formulating each of the fit models with three ideal types in which the profile for the analyzer was specified as the arithmetic means of the values for the prospector and the defender. The results of the three canonical analyses indicate that the set of fit measures associated with each of the different interpretations of the ideal types is significantly related to organizational effectiveness (see Table 9). The models of fit calculated with four ideal types predict 8 percent (p < .05) of the variance in effectiveness; the models of fit calculated with three independent ideal types predict 17 percent (p < .01) of the variance in effectiveness; and the models of fit calculated with a continuum of types predict 24 percent (p < .01) of the variance in effectiveness. The results of the canonical analyses, when interpreted in accordance with the criteria presented above, indicate that interpreting the theory as identifying three ideal types that define a continuum is preferable. First, the findings validate the a priori arguments of Miles and Snow that "the Defender and the Prospector reside at opposite ends of a continuum" and that TABLE 9 Canonical Correlations Between the Measures of Fit Formulated with Different Interpretations of Miles and Snow's Ideal Types and the Effectiveness Measures
Ideal Types Prospector, analyzer, defender, and reactor Prospector, analyzer, and defender Prospector, analyzer, and defender * p < .05, Wilks's lambda **p < .01, Wilks's lambda Relationship Among the Types Independent Independent Continuum First Canonical Correlation .391* .472** .537** Adjusted r,2 .077 .168 .242

1234

Academy of Management Journal

December

the analyzer lies "between these two extremes" (Miles & Snow, 1978: 68). Second, the continuum-of-types interpretation predicts 7 percent more of the variance in effectiveness than does the three-independent-types interpretation and 16 percent more of the variance than the four-independenttypes interpretation. Third, the assumption that three of the ideal types define a continuum enhances the parsimony of the theory. When a continuum of types is assumed, the theory is fully defined with only two of the ideal types, the prospector and the defender, because the third ideal type, the analyzer, is defined as a point equally distant from each of the other two types. Thus, the continuum-of-types interpretation of the theory is superior on each of the three evaluation criteria suggested above: consistency with a priori arguments of Miles and Snow, predicted variance in effectiveness, and parsimony. The converging support based on the latter two criteria is particularly strong given that more parsimonious theories are often less accurate (Thorngate, 1976; Weick, 1979). Consequently, we concluded the continuum-of-types interpretation of the typology should be adopted for a valid test of the theory. Testing Miles and Snow's Theory Which of the four fit models in the continuum-of-types analysis provides the best model of the assumptions of equifinality in Miles and Snow's (1978) theory? Closer examination of the previous continuum-of-types canonical correlation, particularly the relative contributions of the four fit models, best addresses this question. Specifically, we first examined the underlying correlations and then proceeded with a stepwise canonical correlation analysis based on the continuum-of-types interpretation. We used the same evaluation criteria used in the first study 2 analysis: consistency, predictability, and parsimony. Independent models of fit. Table 10 presents the correlations among the measures of fit that result when the four models of fit are calculated with the continuum-of-types interpretation. The mean correlation among the fit measures is .73. These high intercorrelations among the fit measures are not surprising. Based on the continuum-of-types interpretation of the theory, each of the fit measures assesses the distance between an organization and a point on a line running through multidimensional space. The endpoints of this line are defined by the ideal profiles for the prospector and defender types, and the midpoint of this line defines the ideal profile for the analyzer. Thus, we expected substantial correlations among the fit measures. Table 10 also presents the correlations between the measures of fit and the effectiveness dimensions. In general, the correlations are small, and the only significant ones are opposite the predicted direction. Thus, results do not appear to support Hypotheses 7 through 10. However, the negative correlations between hybrid types fit and the contingent hybrid types fit and human relations effectiveness must be interpreted in light of the strong positive correlations among the different types of fit and the near zero correlations between most types of fit and the effectiveness dimensions. The overall

1993

Doty, Glick, and Huber

1235

TABLE 10 Means, Standard Deviations, and Correlations Among Effectiveness Measures and the Measures of Fit Formulated with Miles and Snow's Three Ideal Types on a Continuum
Ideal Types Contingent Ideal Types Contingent Hybrid Types Hybrid Types

Variables Fit measures Ideal types Contingent ideal types Contingent hybrid types Hybrid types Effectiveness measures Cost Efficiency Human relations Quality Resource acquisition p < .05 * p < .01

Means -4.24 -4.80 -4.51 -3.40 3.37 5.38 4.89 4.20 4.80

s.d. 1.02 1.10 1.60 1.08 1.27 1.15 0.99 1.59 1.26

.81** .67** .85** .06 .02 .01 .04 .17

.57** .71** .13 .02 .03 -.02 .10

.78** .13 -.15 -.26** -.01 .06 .09 -.06 -.25** .02 .06

pattern of correlations seems to suggest that the two negative correlations result either from an unusual random occurrence or from a suppression effect. We examine the possibility of a suppression effect below in the stepwise canonical correlation analyses used to test Hypothesis 11. Simultaneous processes. We tested Hypothesis 11, which posits that the models of fit might represent simultaneous processes rather than mutually exclusive processes, by computing a series of hierarchical canonical correlations. A "step down" hierarchical analysis approach analogous to a backward searching stepwise regression (Thompson, 1982, 1991) was appropriate because it (1) allowed us to test Hypothesis 11 as originally stated and (2) provided a mechanism for assessing the relative importance of the different models of configurational fit. Thus, the four measures of fit associated with the continuum-of-types interpretation of the theory were first used simultaneously to predict organizational effectiveness. In each successive step of the analysis, we eliminated the measure of fit that contributed least to the level of predicted variation from the set of independent variables. The step down process was stopped when eliminating an additional measure of fit would have resulted in a significant decrease in the level of predicted variation. Table 11 presents the results from the step down hierarchical canonical analyses. In the first step of the stepwise analysis, we computed the canonical correlation between the four measures of fit associated with the continuum-of-types interpretation of the theory and the five measures of organizational effectiveness. The level of predicted variation in step 1 was 24 percent (p < .01), supporting Hypothesis 11. In the second step, the contingent ideal types measure of fit was eliminated and the level of predicted variation

1236

Academy of Management Journal

December

TABLE 11 Canonical Correlations Between the Fit and Effectiveness Stepwise Measures for Testing Miles and Snow's Configurational Theory
Step 1 Measures of Fit Ideal types, contingent ideal types, contingent hybrid types, and hybrid types Ideal types, contingent hybrid types, and hybrid types Ideal types and hybrid types p < .05, Wilks's lambda ** p < .01, Wilks's lambda * Canonical Correlation Adjusted rc2 Change in Adjusted rc2

.537**

.242

.537** .521**

.250 .239

.008 .011

increased from 24 percent to 25 percent. This trivial increase in the level of predicted variation was possible because the degrees of freedom used to calculate the adjusted canonical correlation decreased between step 1 and step 2. In the third step, we eliminated the contingent hybrid types measure of fit from the set of independent variables. The level of predicted variation declined insignificantly from 25 percent to 24 percent. We stopped the hierarchical analysis after this third step because eliminating either of two remaining measures of fit resulted in a significant reduction in the level of predicted variation. When used alone, the ideal types measure of fit explained only 1 percent (p > .05) of the variance in effectiveness, and the hybrid types model explained only 7 percent (p < .05) of the variance. Thus, when used in conjunction with Miles and Snow's (1978) theory, a composite of the ideal types and the hybrid types models of fit predicted 24 percent (p < .01) of the overall variance in organizational effectiveness. The strength of the canonical correlation when both ideal types fit and hybrid types fit are included in the canonical analysis suggests that the negative correlations reported in Table 10 can be explained as a suppression effect. For example, the correlation between hybrid types fit and human relations effectiveness is -.25 (p < .01). This negative relationship appears to suggest that the more closely an organization resembles a hybrid type, the less effective the organization is. Recall, however, that hybrid types fit is strongly correlated with ideal types fit (r = .85, p < .01) and that ideal types fit appears to be uncorrelated with human relations effectiveness (r = .01, n.s.). In combination, these two correlations indicate that the relationship between hybrid types fit and human relations effectiveness results from the unique variance in hybrid types fit that is independent of ideal types fit. This unique variance is a function of the dissimilarity between the hybrid type and the relevant ideal type. Thus, rather than indicating that hybrid types are not effective, the negative correlation between hybrid types fit and human relations effectiveness indicates that greater dissimilarity between the

1993

Doty, Glick, and Huber

1237

hybrid type and the closest ideal type reduces the effectiveness of the hybrid type. In other words, we can conclude that (1) the more closely an organization resembles a hybrid type, the greater the organization's effectiveness and (2) the more closely a hybrid type resembles an ideal type, the greater the hybrid type's effectiveness. This line of reasoning is consistent with Gresov's (1989) argument that hybrid types may be less effective than ideal types but may be the most effective form possible for some organizations. By extension, this line of reasoning may also explain the negative correlation between contingent hybrid types fit and other dimensions of effectiveness.6 One implication of the above findings is that the internal consistency, or fit, among contextual, structural, and strategic factors is an important determinant of organizational effectiveness for the Miles and Snow (1978) theory. The most consistent configurations of these factors are described by three of the ideal types: the prospector, the analyzer, and the defender. Thus, the greater the correspondence between a real organization and one of the ideal types, the greater the organization's effectiveness. In many organizations, however, external constraints, such as the directives from a parent organization, make it impossible for managers to configure the organization to closely resemble any one of the ideal types. When such external constraints exist, the hybrid types model implies that managers should configure an organization as a hybrid type because greater correspondence between a real organization and one of the hybrid types results in greater effectiveness. Further, the results indicate that greater similarity between this hybrid type and one of the ideal types will lead to a dramatic increase in effectiveness. This finding suggests that the ideal types are the most effective form of organization and that hybrid types are also effective, but less effective than the ideal types. The finding that a composite of the ideal types and hybrid types models of fit is the best predictor of organizational effectiveness also seems to imply that contextual factors do not constrain an organization's choice among the types. Neither of the two models of fit included in the composite model bounds the assumption of equifinality with contingency factors. Given such results, we might conclude that the assumption of strategic choice (Astley & Van de Ven, 1983; Child, 1972) was supported. However, it is possible that organizations must adopt the form that is appropriate for their contexts to survive and that all the organizations studied here had adopted such a form before this research was conducted. In this case, we would not expect to find
6 Two post hoc analyses were conducted to validate the interpretation of the negative correlations as a suppression effect. First, we correlated the distance between the hybrid type most similar to an organization and the most similar ideal type with human relations effectiveness. The results indicate that the divergence between the hybrid type and the relevant ideal type is negatively related to human relations effectiveness (r = -.38, p < .01). Second, holding constant the divergence between the hybrid type and the relevant ideal type resulted in a positive partial correlation (partial r = .20, p < .05) between hybrid types fit and human relations effectiveness. Thus, the post hoc analyses provide empirical validation for our arguments.

1238

Academy of Management Journal

December

TABLE 12 Canonical Correlations Between the Effectiveness Measures and the Ideal Types and Hybrid Types Measures of Fit for Testing Miles and Snow's Configurational Theory
Canonical Correlations Canonical Variate Function 1 Function 2 Canonical Correlation .521 .191 Adjusted R2 .239 .036 Wilks's Lambda .001 .423

strong contingency effects in a cross-sectional sample of existing organizations. Thus, contextual contingencies should not be omitted from tests of Miles and Snow's (1978) theory until these results can be replicated in a longitudinal design. Further validation of these results is also necessary because the measures of fit we used are highly correlated (see Table 10). This multicollinearity may contribute to instability of the results concerning the attribution of predicted variance among the different models. Future research might find more support for the contingent ideal types and the contingent hybrid types models. Individual dimensions of effectiveness. The predictive power of the composite model with respect to each of the five individual dimensions of effectiveness can be assessed by examining more fully the results of the canonical analysis (Tables 12 and 13). The human relations measure of effectiveness had the strongest relationship with the canonical variate and thus with the composite model of fit. This finding suggests that Miles and Snow's theory is a good predictor of the extent to which an organization is successful in managing human resources and supports Zahra and Pearce's TABLE 13 Coefficients of the Canonical Variates Associated with the Fit and Effectiveness Measuresa
Correlation with Canonical Variate of Fit Measures -.001 -.524 -.085 .142 .152 .488 .032 Correlation with Canonical Variate of Effectiveness Measures -.001 -.273 -.163 .272 .291 .936 .060

Measures Fit Ideal types model Hybrid types model Effectiveness Cost Quality Efficiency Human relations Resource acquisition
a

Standardized Canonical Weight 1.627 -1.911 -.362 -.123 .073 1.017 .028

Only the first canonical variate is reported because it was the only canonical variate that was both significant and interpretable.

1993

Doty, Glick, and Huber

1239

(1990) observation that the managerial philosophies of an organization with respect to human resources are important and should not be overlooked in future tests of Miles and Snow's theory. Some caution, however, is necessary in interpreting these canonical variates because of the moderate correlations among the effectiveness measures. Multicollinearity may have contributed to instability in the canonical variates. Thus, this last finding also should be validated with further research. CONCLUSIONS In combination, the results from study 1 and study 2 do not support Mintzberg's (1979, 1983) theory but do support Miles and Snow's (1978) theory. To make the best use of the two studies, we first discuss the implications from the Miles and Snow study and then suggest how the Miles and Snow results can be informative for the Mintzberg study. Study 2 One of the most striking findings of study 2 was the level of support generated for Miles and Snow's theory. When the theory was tested with quantitative models of fit that are consistent with the fit and equifinality arguments in the theory, it predicted 24 percent of the overall variation in organizational effectiveness. Thus, a major conclusion from study 2 is that fit among contextual, structural, and strategic factors is a relatively powerful predictor of organizational effectiveness. The finding that Miles and Snow's theory predicts organizational effectiveness is not unique to this research. Four points, however, distinguish the current research from previous studies. First, study 2 relies on theoretically specified ideal profiles developed by the authors of the theory rather than on treating the ideal types as if they were categories (cf. Conant et al. 1990; Venkatraman & Prescott, 1990; Zajac & Shortell, 1989). Thus, the methodology more accurately defines the ideal types than the categorical approaches used in previous studies. Second, this study suggests that Miles and Snow's theory is a more powerful predictor of organizational effectiveness than has previously been reported. For example, in the only other study to report a level of predicted variation in effectiveness, Hambrick (1983) predicted 13 percent of the variance in cash flow on investment and 10 percent of the variance in return on investment. The current work, by contrast, predicted 24 percent of the variation in self-reported overall effectiveness. Third, this study provides the first direct test of the fit and equifinality assertions in the theory. Given the arguments regarding the logical structure of configurational theories, we suggest that the increased explanatory power in this study results from the increased accuracy of the newly developed theoretical and analytical models used to test Miles and Snow's theory. Fourth, this is the first study to systematically assess all of the extant plausible interpretations of Miles and Snow's theory. The results indicate that the most parsimonious interpretation of the theory-that there are three ideal types de-

1240

Academy of ManagementJournal

December

in the models of fit with the greatest predicfining a continuum-resulted tive validity. Although the configurational test of Miles and Snow's theory has generated stronger support for the theory than has been previously reported, the current work has a number of limitations. First, it was based on a crosssectional analysis and thus cannot be interpreted as a test of the adaptive cycle Miles and Snow proposed (Zahra & Pearce, 1990). Second, the empirical tests in this study were based on canonical correlations that maximize the predicted variance in the dependent variables. Thus, the current study reports an upper bound on the level of variation in effectiveness that Miles and Snow's theory can explain in the current data. A third potential limitation is that the data in this study were collected from a single informant in each organization, raising the familiar threat that common methods variance might have inflated the observed relationships (Cote & Buckley, 1987; Doty & Glick, 1989; Glick, Jenkins, & Gupta, 1986; Phillips, 1981). Common methods variance creates a problem whenever the same informants (or method) provide data for both the independent and the dependent variables in a study and a pattern of responses on the independent variables obviously and logically implies a pattern of responses on the dependent variable. In the current studies, all the data on each organization were collected from a single informant. Given the complexity of the fit calculations, it seems unlikely that informants could structure their responses to the effectiveness measures on the basis of the patterns of their responses to the items that were used to generate the fit indexes. Doing so would have required an informant to understand the details of both Miles and Snow's (1978) theory and the mathematical formulations of the fit measures. We doubt that many informants maintained such a complex mental model. Further, the strong results in study 2 were generated using the same methodology with the same informants in the same organizations that produced consistent null results in study 1. Thus, it is unlikely that the significant results in study 2 are a function of common methods variance. A fourth potential limitation is that organizations were selected to represent a variety of contexts, structures, and strategies rather than a random sample. Thus, we must be cautious in generalizing from this data set. Finally, the current work did not investigate other aspects of the theory, such as the human resource policies of the different ideal types or the characteristics of the top management team. Future research that investigates such issues may provide more and stronger support for the theory. Despite such limitations, the current work appears to present the most complete test of the theory conducted to date. Study 1 The study 2 results strengthen our ability to draw conclusions about the null results of study 1. The former suggest that the null results from study 1 are not a function of (1) the complex fit indexes used to test Mintzberg's

1993

Doty, Glick, and Huber

1241

theory,7 (2) the limited set of effectiveness measures used, (3) the set of organizations included, or (4) the use of a single informant methodology. To the contrary, given the relatively high statistical power of this research, the lack of significant results in study 1, the appropriateness of the current approach for testing the configurational hypotheses derived from Mintzberg's (1979, 1983) theory, the inclusion of multiple hypotheses to capture alternative interpretations of fit that are implicit in the theory, Mintzberg's inclusion of the same kinds of organizations we observed (profit and notfor-profit, divisions and headquarters), and the presence of an alternative theory to replace Mintzberg's theory, we can only conclude that the theory was refuted for the current data. Several characteristics of Mintzberg's theory provide plausible explanations for the nonsignificant results and opportunities for improving the theory. First, we examined only the five ideal configurations and the hybrid forms of the five ideal types. Mintzberg (1979, 1983) clearly stated that other ideal types may exist-and briefly mentions a missionary form-but he did not fully define these additional ideal types. Other theories may identify ideal configurations that are more strongly related to effectiveness than those suggested by Mintzberg (1979). For example, Mintzberg (1989) described missionary and political organizations as additional ideal types. These types were not included in the current test of Mintzberg's earlier (1979, 1983) model. Revising the theory to include additional ideal types and including them in future tests of the theory might result in support for a revised theory. The theory also might be revised to include additional configurations of contextual factors. Post hoc examination of the rater estimates of ideal configurations and the correlations among the deviation scores for the contextual configurations suggests that the contexts Mintzberg (1979) described are relatively similar to each other. The range of contexts facing organizations in the current study extends significantly beyond Mintzberg's contextual configurations. Adding new configurations of contextual factors or reconceptualizing the patterns of contextual factors for each of Mintzberg's current ideal types might reveal stronger relationships between contingent ideal types fit and organizational effectiveness. Comparing the results of the two studies suggests a third potential extension of Mintzberg's theory: the addition of more factors in the description of the ideal types. As one example, a key difference between the two theories is that Mintzberg does not describe his ideal types in terms of strategic attributes. Thus, an additional revision that might improve the predictive power of Mintzberg's theory involves expanding the descriptions of the ideal types to include strategic factors. As a second example, a temporal dimension might be added to the theory. One of the implications of Miller's research is that "managers may have to perform their adaptive tasks sequen-

7 Drazin and Van de Ven (1985) and Govindarajan (1988) also found significant results using a systems conceptualization of fit.

1242

Academy of Management Journal

December

tially striving for a harmonious alignment among their internal variables in order to achieve smooth functioning, but periodically disrupting this harmony to adapt to a changing environment" (1992: 159). As a final example, the type of leadership might be added as a component of the organizational configuration, as Miller (1987) suggested. Mintzberg's description of the hybrid types also may benefit from revision. Hybrid types are suggested in Mintzberg's (1979) final chapter, but the form of hybridization is not clearly articulated. In the current test of the theory, we assumed that hybrids lie along the continua among the five ideal types. Although this assumption is parsimonious and consistent with Mintzberg (1979), alternative formulations of hybridization might be based on explicit tradeoffs between specific contextual or design characteristics. Specifying a more precise set of hybrid types in this theory and in other theories that allow hybridization among the initial ideal types should result in better configurational theories. Mintzberg's (1979, 1983) theory might also benefit from careful specification of its boundary conditions. We interpreted Mintzberg's theory as a grand theory of organizations intended to apply across the population of formal organizations. Thus, we incorporated a very diverse set of organizations in our test. The diversity of these organizations precluded testing the theory within any single industry.8 Future researchers, however, might attempt to specify important factors that might limit the range of Mintzberg's theory and then retest it with restricted samples. Mintzberg's theory may be valid in some industries and not in others, or in some sectors of the economy and not in others, or only among mature organizations, or only among truly autonomous firms rather than among semiautonomous strategic business units. Such research might show that Mintzberg's theory is a powerful predictor of organizational effectiveness when it is interpreted as a middle range theory (Pinder & Moore, 1979; Weick, 1974) rather than a grand theory. Further, future configurational theorists should carefully specify the boundary conditions of their theories so that valid tests of the theories can be conducted. Although the potential methodological limitations of study 1 cannot be wholly discounted, the current test appears to be the most valid and complete test of Mintzberg's (1979, 1983) theory conducted to date. The evidence for revising or rejecting Mintzberg's typology and theory is strong. Organizations do not appear to structure their designs to be consistent with their contexts in terms of that typology. Further, real organizations' similarity to any of Mintzberg's (1979) ideal types or hybrids of those types does not
8 Although we tested Hypotheses 3-5 within each industry group, the numbers of firms within each industry were too small to allow us to claim adequate statistical power. Despite this caveat, it is interesting to note that the results were very consistent with the results reported earlier. A maximum of 1 of the 45 correlations was significant (p < .05) and in the predicted direction in any industry. Because of the low statistical power, however, these results do not provide an adequate test of the theory within any of the industry groups.

1993

Doty, Glick, and Huber

1243

appear to be associated with organizational effectiveness. These results suggest either that those ideal types do not maximize fit, or that Mintzberg's notions of fit do not lead to organizational effectiveness. In sum, until other researchers can provide empirical support for Mintzberg's work, we are unable to conclude that either the typology or the theory is valid. Contributions from the Combined Studies This research also has important implications beyond the domain of the two theories tested. First, in the theory development portion of this article we explored one of the primary differences between configurational theories and more traditional theories relying on bivariate or simple multiplicative relations among a limited set of variables or categories. Configurational theories are developed using an ideal-type construct to represent complex synergistic effects that arise from specific configurations of organizational attributes. Researchers attempting to test configurational theories must recognize and accurately model the ideal types incorporated in configurational theories to ensure that their tests of the theories are valid. Further, future theorists should include a complete set of factors in their descriptions of ideal types. At a minimum, ideal types should be described in terms of the imperatives that drive organizations toward certain configurations (Miller, 1978). Second, we have elaborated and clarified the notion of fit. We conceptualized fit as the consistency among contextual, structural, and strategic factors, a conceptualization consistent with many organizational theories, especially those that identify a typology of effective organizational configurations (e.g., Burns & Stalker, 1961; Mintzberg, 1979, 1983; Porter, 1980, 1985; Weber, 1946). Fit was operationally defined with a composite of the ideal types and hybrid types models of configurational fit. The results from testing one theory with this composite model indicated that fit is higher in the ideal types than in the hybrid types and higher in the hybrid types than in other organizational forms. Third, we have clarified the assumption of equifinality. Ashmos and Huber (1987) identified the investigation of equifinality as one of the "missed opportunities" in organizational research. The current research identified four plausible interpretations of equifinality that result in different substantive interpretations of a configurational theory. We integrated these four interpretations of equifinality with the systems conceptualization of fit in a set of configurational fit models that provide quantitative representations of the complex assumptions incorporated in many configurational theories. A fourth contribution of this research was to demonstrate that the configurational approach developed in this article is viable for testing configurational theories. The quantitative models associated with the configurational approach can be used to represent complex fit and equifinality assertions such as those incorporated in Miles and Snow's (1978) and Mintzberg's (1979, 1983) theories. The dramatically different outcomes of the two em-

1244

Academy of Management Journal

December

null results from study 1 and the strongly supportive pirical studies-the results from study 2-suggest that the configurational fit models developed in this article are adequate for conducting valid tests of configurational theories. Future researchers can now adapt these quantitative models to test other configurational theories. Starbuck noted that "Francis Bacon, John R. Platt (1964), and Karl R. Popper (1959) have argued persuasively that science makes progress mainly by showing some hypotheses to be incorrect, not by showing that other hypotheses might be correct. . . . More dependable progress comes from eliminating poor hypotheses than from sustaining plausible ones" (1993: 3). Although it may be impossible to show that any hypothesis is unequivocally incorrect, a final contribution of this article is that it raises significant doubts about some plausible interpretations of two very important theories. Null results were observed for all the hypotheses derived from Mintzberg's (1979, 1983) theory, and alternative interpretations of Miles and Snow's (1978) theory were disconfirmed. The null results were supported by statistical power calculations, supplemental evidence about the validity of the measures and research design, and strong support for alternate a priori hypotheses using the same methodology with the same informants in the same organizations. At the very least, this pattern of results should stimulate some revisions and further research for both theories. REFERENCES
Ashmos, D. P., & Huber, G. P. 1987. The systems paradigm in organization theory: Correcting the record and suggesting the future. Academy of Management Review, 12: 607-621. Astley, W. G., & Van de Ven, A. H. 1983. Central perspectives and debates in organization theory. Administrative Science Quarterly, 28: 245-273. Bacharach, S. B. 1989. Organization theories: Some criteria for evaluation. Academy of Management Review, 14: 495-515. Blalock, H. M., Jr. 1969. Theory construction: From verbal to mathematical formulations. Englewood Cliffs, NJ: Prentice-Hall. Blalock, H. M., Jr. 1989. The real and unrealized contributions of quantitative sociology. American Sociological Review, 54: 447-460. Burns, T., & Stalker, G. M. 1961. The management of innovation. New York: Barnes & Noble. Cameron, K. S., & Whetten, D. A. 1983. Organizational effectiveness: A comparison of multiple models. San Diego: Academic Press. Cattell, R. B. 1949. rp and other coefficients of pattern similarity. Psychometrika, 14: 279-298. Child, J. 1972. Organization structure, environment, and performance: The role of strategic choice. Sociology, 6: 1-22. Cohen, J. 1977. Statistical power analysis for the behavioral sciences. New York: Academic Press. Conant, J. S., Mowak, M. P., & Varadarajan, P. R. 1990. Strategic types, distinctive competencies and organizational performance: A multiple measures-based study. Strategic Management Journal, 11: 365-383. Cote, J. A., & Buckley, M. R. 1987. Estimating trait, method, and error variance: Generalizing across 70 construct validation studies. Journal of Marketing Research, 24: 315-318.

1993

Doty, Glick, and Huber

1245 Bul-

Cronbach, L. J., & Gleser, G. C. 1953. Assessing similarity between profiles. Psychological letin, 50: 456-473.

Daft, R. L., & Weick, K. E. 1984. Toward a model of organizations as interpretation systems. Academy of Management Review, 9: 284-295. Dearborn, D. C., & Simon, H. A. 1958. Selective perception: A note on the departmental identifications of executives. Sociometry, 21: 140-144. DeVellis, R. F. 1991. Scale development: Theory and applications. Newbury Park, CA: Sage. Doty, D. H. 1990. Context, structure, and strategy: A configurational approach to organizational effectiveness. Unpublished doctoral dissertation, University of Texas at Austin. Doty, D. H., & Glick, W. H. 1989. Construct validity in organizational research: Good newsbad news. Paper presented at the national meeting of the Academy of Management, Washington, DC. Drazin, R., & Van de Ven, A. H. 1985. Alternate forms of fit in contingency theory. Administrative Science Quarterly, 30: 514-539. Dubin, R. 1969. Theory building. New York: Free Press. Duncan, R. B. 1972. Characteristics of organizational environments and perceived environmental uncertainty. Administrative Science Quarterly, 17: 313-327. Fry, L. W. 1982. Technology-structure research: Three critical issues. Academy of Management Journal, 25: 532-552. Fry, L. W., & Slocum, J. W., Jr. 1984. Technology, structure, and work group effectiveness: A test of a contingency model. Academy of Management Journal, 27: 221-246. Ginn, G. 0. 1987. Strategic adaptation in the hospital industry: A study of selected Texas hospitals from 1976 through 1985. Unpublished doctoral dissertation, University of Texas at Austin. Glick, W. H. 1985. Conceptualizing and measuring organizational and psychological climate: Pitfalls in multilevel research. Academy of Management Review, 10: 601-616. Glick, W. H., Huber, G. P., Miller, C. C., Doty, D. H., & Sutcliffe, K. M. 1990. Studying changes in organizational design and effectiveness: Retrospective even histories and periodic assessments. Organization Science, 1: 293-312. Glick, W. H., Jenkins, G. D., Jr., & Gupta, N. 1986. Method versus substance: How strong are underlying relationships between job characteristics and attitudinal outcomes? Academy of Management Journal, 29: 441-464. Glick, W. H., Miller, C. C., & Huber, G. P. 1993. The impact of upper-echelon diversity on organizational performance. In G. P. Huber & W. H. Glick (Eds.), Organizational change and redesign: 176-214. New York: Oxford University Press. Govindarajan, V. 1988. A contingency approach to strategy implementation at the business-unit level: Integrating administrative mechanisms with strategy. Academy of Management Journal, 31: 828-853. Gresov, C. 1989. Exploring fit and misfit with multiple contingencies. Administrative Quarterly, 34: 431-453. Science

Hambrick, D. C. 1981. Environment, strategy, and power within top management teams. Administrative Science Quarterly, 26: 253-276. Hambrick, D. C. 1983. Some tests of the effectiveness and functional attributes of Miles and Snow's strategic types. Academy of Management Journal, 26: 2-26. Ireland, R. D., Hitt, M. A., Bettis, R. A., dePorras, D. A., 1987. Strategy formulation processes: Differences in perceptions of strength and weakness indicators and environmental uncertainty by managerial level. Strategic Management Journal, 8: 467-485.

1246

Academy of Management Journal

December

Katz, D., & Kahn, R. L. 1978. The social psychology of organizations (2d ed.). New York: Wiley. Kennedy, J. J. 1983. Analyzing qualitative data: Introductory log-linear analysis for behavioral research. New York: Praeger. Kerlinger, F. N. 1986. Foundations of behavioral research. New York: Holt, Rinehart & Winston. McKelvey, B. 1982. Organizational systematics: Taxonomy, evolution, classification. ley: University of California Press. Berke-

McKinney, J. C. 1966. Constructive typology and social theory. New York: Appleton Century Crofts. Miles, R. E., & Snow, C. C. 1978. Organization strategy, structure, and process. New York: McGraw-Hill. Miller, D. 1978. The role of multivariate "Q-techniques" in the study of organizations. Academy of Management Review, 3: 515-531. Miller, D. 1981. Toward a new contingency theory: The search for organizational gestalts. Journal of Management Studies, 18: 1-26. Miller, D. 1987. The genesis of configuration. Journal of Management Studies, 12: 686-701. Miller, D. 1992. Environmental fit versus internal fit. Organizational Science, 3: 159-178. Miller, D., & Friesen, P. H. 1984. Organizations: A quantum view. Englewood Cliffs, NJ: Prentice-Hall. Miller, D., & Mintzberg, H. 1983. The case for configuration. In G. Morgan (Ed.), Beyond method: Strategies for social research: 57-73. Beverly Hills, CA: Sage. Mintzberg, H. T. 1979. The structuring of organizations. Englewood Cliffs, NJ: Prentice-Hall. Mintzberg, H. T. 1983. Structure in fives: Designing Cliffs, NJ: Prentice-Hall. effective organizations. Englewood

Mintzberg, H. T. 1989. Mintzberg on management: Inside our strange world of organizations. New York: Free Press. Mohr, L. B. 1982. Explaining organization behavior. San Francisco: Jossey-Bass. Monge, P. R. 1990. Theoretical and analytical issues in studying organizational processes. Organizational Science, 1: 406-430. Mulaik, S. A. 1972. The foundations of factor analysis. New York: McGraw-Hill. Ogilvie, D. T., & Glick, W. H. 1990. A study of strategic planning processes and how information systems are used in strategic planning. Working paper, University of Texas at Austin. Overall, J. E. 1964. Note on the multivariate methods for profile analysis. Psychological letin, 61: 195-198. Overall, J. E., & Klett, C. J. 1972. Applied multivariate analysis. New York: McGraw-Hill. Bul-

Perrow, C. 1967. A framework for the comparative analysis of organizations. American Sociological Review, 32: 194-208. Pfeffer, J. 1981. Power in organizations. Marshfield, MA: Pitman. Phillips, L. W. 1981. Assessing measurement error in key informant reports: A methodological note on organizational analysis in marketing. Journal of Marketing Research, 28: 395415. Pinder, C. C., & Moore, L. F. 1979. The resurrection of taxonomy to aid the development of middle range theories of organizational behavior. Administrative Science Quarterly, 24: 99-118.

1993

Doty, Glick, and Huber

1247

Porter, M. E. 1980. Competitive strategy: Techniques for analyzing industries and competitors. New York: Free Press. Porter, M. E. 1985. Competitive advantage: New York: Free Press. Creating and sustaining superior performance.

Quinn, R. E., & Rohrbaugh, J. 1983. A spatial model of effectiveness criteria: Towards a competing values approach to organizational analysis. Management Science, 29: 363-377. Schoonhoven, C. B. 1981. Problems with contingency theory: Testing assumptions hidden within the language of contingency theory. Administrative Science Quarterly, 26: 349377. Scott, W. R. 1981. Organizations: Rational, natural, and open systems. Englewood Cliffs, NJ: Prentice-Hall. Segev, E. 1987. Strategy, strategy-making, and performance in a business game. Strategic Management Journal, 8: 565-577. Segev, E. 1989. A systematic comparative analysis and synthesis of two business-level strategic typologies. Strategic Management Journal, 10: 487-505. Seidler, J. 1974. On using informants: A technique for collecting quantitative data and controlling measurement error in organizational analysis. American Sociological Review, 39: 816-831. Shortell, S. M., & Zajac, E. J. 1990. Perceptual and archival measures of Miles and Snow's strategic types: A comprehensive assessment of reliability and validity. Academy of Management Journal, 33: 817-832. Shrout, P. E., & Fleiss, J. L. 1979. Interclass correlations: Uses in assessing rater reliability. Psychological Bulletin, 86: 420-428. Smith, G. S., Guthrie, J. P., & Chen, M. 1989. Strategy, size, and performance. Organization Studies, 10: 63-81. Snow, C. C., & Hrebiniak, L. G. 1980. Strategy, distinctive competence, and organizational performance. Administrative Science Quarterly, 25: 317-336. Starbuck, B. 1993. Why we should accept null hypotheses. OMT Digest, 2(60): 3. Sutcliffe, K. M. 1991. Determinants and outcomes of top managers' perceptions and interpretations of the environment. Unpublished doctoral dissertation, University of Texas at Austin. Thompson, B. 1982. A logic for stepwise canonical correlation analysis. Perceptual and Motor Skills, 54: 879-882. Thompson, B. 1991. Hierarchical canonical correlation analysis. Academy of Management Research Methods Newsletter, 7(1): 8. Thompson, J. D. 1967. Organizations in action. New York: McGraw-Hill. Thorngate, W. 1976. Possible limits on a science of social behavior. In L. H. Strickland, F. E. Aboud, & K. J. Gergen (Eds.), Social psychology in transition: 121-139. New York: Plenum. Van de Ven, A. H., & Drazin, R. 1985. The concept of fit in contingency theory. In L. L. Cummings & B. M. Staw (Eds.), Research in organizational behavior, vol. 7: 333-365. Greenwich, CT: JAI Press. Van de Ven, A. H., & Ferry, D. L. 1980. Measuring and assessing organizations. New York: Wiley. Venkatraman, N. 1989. The concept of fit in strategy research: Toward verbal and statistical correspondence. Academy of Management Review, 14: 423-444. Venkatraman, N., & Prescott, J. E. 1990. Environment-structure coalignment: An empirical test of its performance implications. Strategic Management Journal, 11: 1-23.

1248

Academy of Management Journal

December

Weber, M. 1946. Bureaucracy. In H. H. Gerth & C. W. Mills (Eds. & trans.) From Max Weber: Essays in sociology: 196-245. New York: Oxford University Press. Weick, K. E. 1974. Middle range theories of social systems. Behavioral Science, 21: 1-19. Weick, K. E. 1979. The social psychology Wesley. of organizing (2d ed.). Reading, MA: Addison-

Withey, M., Daft, R. L., & Cooper, W. H. 1983. Measures of Perrow's work unit technology: An empirical assessment and a new scale. Academy of Management Review, 26: 45-63. Zahra, S. A., & Pearce, J. A., II. 1990. Research evidence on the Miles-Snow typology. Journal of Management, 16: 751-768. Zajac, E. J., & Shortell, S. M. 1989. Changing generic strategies: Likelihood, direction, and performance implications. Strategic Management Journal, 10: 413-430. Zajac, R., & Wolfe, D. 1966. Cognitive consequences of a person's position in a formal organization. Human Relations, 19: 139-150.

APPENDIX A Assessing Profile Similarity


Techniques for assessing profile similarity are well developed (Cattell, 1949; Cronbach & Gleser, 1953; Miller, 1978; Overall, 1964). Most of these techniques assess deviation with some form of this weighted Euclidean distance formula: Dio = V(Xi - Xo)W(Xi - X)', where Di, = the distance between ideal type i and organization o, Xi = a 1 x j vector that represents the value of ideal type i on attribute j, X0 = a 1 x j vector that represents the value of organization o on attribute j. and W = a j x j diagonal matrix that represents the theoretical importance of attribute j to ideal type i. (1.0)

APPENDIX B Assessing Fit


Ideal Types Fit Under this model an organization can adopt any one of the ideal types identified in the theory and remain effective. This model of fit is appropriate for testing theories in which organizational effectiveness results from the consistency of the relevant contextual, structural, or strategic constructs: FitIT = -min Here, Dio is determined by equation 1.0 (Appendix A) and Dio (2.0)

min Dio) takes the minimum value of Dio across all ideal types i for organization o.
i=1

1993 Contingent Ideal Types Fit

Doty, Glick, and Huber

1249

In this model, fit is conceptualized as the lack of deviation from the one ideal type that is most congruent with the contingencies that an organization faces.
FitCIT

-Dio,

(3.0)

where i is determined by

min D
i=l

(3.1)

and Do = the distance between the contextual attributes of ideal type i and the contextual attributes of organization o.

min Di takes the minimum deviation on all contextual attributes across all ideal types i.
i=1

Contingent Hybrid Types Fit This model of configurational fit retains the assumption that a set of contingencies constrains the forms that an organization can adopt and remain effective but adds the possibility of hybridization of the ideal types to form effective hybrid types. Thus, there are an infinite number of effective hybrid types, but the one hybrid type that an organization must adopt to remain effective is determined by the unique contingencies that the organization faces.
FitcHT = -/(Xh - Xo)W(Xh - Xo)', (4.0)

where Xh = a 1 x j vector that represents the value of hybrid type h on attribute j, Xo = a 1 x j vector that represents the value of organization o on attribute j, and W = a j x j diagonal weighting matrix that contains the theoretical importance of attribute j. In addition, only one hybrid type is defined for each organization, as described below. To maximize effectiveness, the extent to which a real organization deviates from the ideal types of organizations identified in the theory is determined by the deviation between the real organization's contingencies and the contingencies specified for the ideal types of organizations. Thus, complete hybridization among the ideal types is permitted, but an organization's context determines the one hybrid type that an organization must adopt to be maximally effective. This hybrid type can be identified by assessing the organization's context and then predicting the structural and strategic configuration that is most congruent with that context. This hybrid type can be identified using Equations 4.1 and 4.2. Equations 4.1 and 5.1 are adapted from Overall and Klett's (1972) discussion of linear typal analysis. Equations 4.2 and 5.2 are adapted from the least squares regression formula, p = (X'X)-1(X'Y). The adaptation is necessary to produce a beta matrix with the correct dimensions. Xh = Po1Xo + oz2Xo+ .. . PoiXo, where, (4.1)

1250

Academy of Management Journal

December

Xh = a 1 x j vector that represents the value of hybrid type h on attribute j,

Xo = a 1 x j vector that represents the value of organization o on attribute j, and ,oi = a scalar that represents the extent to which organization o resembles ideal type i. Po = (XcXC')(Xcc) - , where Po = a 1 x i vector that represents the extent to which organization o resembles ideal type i, xc = a 1 x j vector that represents the value of real organization o on contextual attribute j, and Xc = an i x j matrix that represents the value of ideal type i on contextual attribute j. Hybrid Types Fit This model is like the prior model except that it does not include the assumption about the importance of contingencies. An organization is free to adopt any hybrid type of the ideal types identified in the theory and remain effective. Thus, the equation is similar to Equations 4.0 and 4.1:
FitHT = -\(Xh - Xo)W(Xh - Xj', (5.0)

(4.2)

and Xh = Bo1Xo+ o02Xo+


. PoiXo.

(5.1)

The two models differ, however, on how the appropriate hybrid type is identified. Unlike the contingent hybrid types model, the hybrid types model does not limit the identification of the appropriate hybrid type to contextual factors. Thus, modeling the hybrid types model requires that Equation 4.2 be modified to the form shown in Equation 5.2: Po = (XoX')(XX')-1, where Bo = a 1 x i vector that represents the extent to which organization o resembles ideal type i, Xo = a 1 x j vector that represents the value of real organization o on attribute j, and X = an i x j matrix that represents the value of ideal type i on attribute j. D. Harold Doty is an assistant professor of management at the University of Arkansas, Fayetteville. He received his Ph.D. degree from the University of Texas at Austin in management. His current research is in the areas of organizational design, configurational theories, and business-unit effectiveness. William H. Glick is an associate professor of management of the University of Texas at Austin. He received his Ph.D. degree from the University of California at Berkeley in business administration. His current research is in the areas of job and organizational design and business process improvement. George P. Huber is the Charles and Elizabeth Prothro Regents Chair in Business Administration at the University of Texas at Austin. His research interests focus on organizational learning and adaptation, organizational design, and organizational decision making. (5.2)

You might also like