Centralized Versus Decentralized Computing: Organizational Considerations and Management Options

Centralized versus Decentralized Computing: Organizational Considerations and Management Options

JOHN LESLIE KING Department of Information and Computer Science, University of Cahfornia, Irvine, Irvine, California 92717

The long.standing debate over whether to centralize or decentralize computing is examined in terms of the fundamental organizational and economic factors at stake. The traditional debate is evaluated, and found to focus predominantly on issues of efficiency versus effectiveness, with solutions based on a rationalistic strategy of optimizing in this trade-off. A behavioral assessment suggests that the driving issues in the debate are the politics of organization and resources, centering on the issue of control. The economics of computing depolyment decisions are presented as an important issue, but one that often serves as a field of argument in which political concerns are dealt with. The debate in this light appears to be unresolvable in the long run, although effective strategies can be developed when the larger issues are recognized. The current situation facing managers of computing, given the advent of small and comparatively inexpensive computers, is examined in detail, and a set of management options for dealing with this persistent issue is presented. Categories and Subject Descriptors: K.6.0 [Management of Computing and Information Systems]: General; K.6.1 [Management of Computing and Information Systems]: Project and People Management; K.6.4 [Management of Computing and Information Systems]: System Management General Terms: Economics, Management Additional Key Words and Phrases: Centralization/decentralization, management of computing/information systems, structural arrangements for computing

INTRODUCTION usually prompted by technological changes that affect the efficiencies of existing ar- Managers of computing have confronted rangements [EDPIDR 1979; Bernard 1979; decisions about centralizing or decentral- Breslin and Tashenberg 1978; Bucci and izing computing ever since the computer Streeter 1979; Buchanan and Linowes proved to be more than just another piece 1980a, 19801o; Chervany et al. 1978; Demb of office equipment. The debate has flour- 1975; D'Oliveria 1977; Ein-Dor and Segev ished for nearly twenty years in the infor- 1978; McKenney and McFarlan 1982; mation systems community. The goal has Metres 1981; Nolan 1979; Reynolds 1977; been to determine an appropriate arrange- Rockhart et al. 1979; Withington 1980]. ment for the deployment of computing re- The terrain in which to make centrali- sources in organizations, given user needs zation decisions has been continually and the desire of management to control changing. Predictions about the "comput- both costs and uses. A universally appro- ing arrangement of the future" range from priate arrangement has never been found. the conservative to the revolutionary: from Nevertheless, there has been a steady flow the deployment of small, special-purpose of advice on how to deal with this question, computers in user departments, to net-

Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given that copying is by permission of the Association for Computing Machinery. To copy otherwise, or to republish, requires a fee and/or specific permission. © 1984 ACM 0360-0300/83/1200-0319 $00.75

ComputingSurveys, Vol. 15, No. 4, December1983 320 • John Leslie King CONTENTS in perspective. It provides a working definition of these concepts and outlines the basic pros and cons of each. The traditional arguments are then discussed, focusing on INTRODUCTION the customary assessment of trade-offs be- 1. THE DEBATE IN PERSPECTIVE tween efficiency and effectiveness. A de- 1 1 A Working Defimtion tailed examination of the problems with 1.2 Pros and Cons of the Alternatwes this traditional perspective concentrates on 2, THE TRADITIONAL DEBATE 3. ORGANIZATIONAL CONSIDERATIONS the critical importance of political and or- IN THE DEBATE TODAY ganizational factors, and notes the ways in 3.1 Politics of Orgamzation which simple economic analyses of the al- and Resources ternatives can result in misleading conclu- 3 2 The Economic Dynamics of Deployment Decisions sions. The examination reveals that control 3.3 The Interaction of Pohtical is the major factor in centralization/decen- and Economic Considerations tralization decisions, and that other aspects 4 MANAGEMENT OPTIONS IN of the debate must be seen in light of this CENTRALIZATION VERSUS DECENTRALIZATION fact. The paper concludes with discussion 4.1 A No-Option Option of the options available to management in 4.2 The Centralization Continuum dealing with the centralization issue, given 4.3 Mixed Strategies the political and economic factors that un- 4.4 Choosing an Option derlie the debate.2 5. CONCLUSION ACKNOWLEDGMENTS REFERENCES 1. THE DEBATE IN PERSPECTIVE 1.1 A Working Definition A v Most conceptualizations of centralization worked distributed systems deployed and decentralization rely on some concept throughout the organization, and even to of distance: the distance between plant lo- remote intelligent terminals in employees' cations, the distance between organiza- homes enabling the work force to "telecom- tional levels or operations, or the distance mute" instead of coming in to the office. 1 (physical or organizational) between where This rapid pace of change helps sustain the decisions are made and where they are en- centralization debate. But as we shall see, acted.3 the debate is rooted in more basic questions of organizational structure and behavior. 2 This paper deals specifically with computer applica- Changing technology merely rearranges the tion to administrative data processing. It does not address the special cases such as process control ap- forum in which these issues are debated. At plications or computer-aided design and manufactur- a time of rapid technological change, with ing that typically do not require intraorganizational concomitant pressure to decide how "best" data sharing. However, many of the issues discussed to organize the use of computing technol- here apply also to such applications. ogy, a review of these basic issues will help 3 This paper addresses centralization as a policy issue (i.e., what centralization/decentralization policies make sense of a confusing subject. should be followed under certain circumstances). It This paper puts the issue of centraliza- does not deal with centralization as an artifact of other tion versus decentralization of computing forces. See Robey [1981] for a discussion oftbe impact of computing on organizational structure (i.e. comput- 1 The predictions made about the potential impact of ing use as it results in greater or lesser organizational computing on organizations can be drawn from a wide centralization). The definition of centralization pro- variety of sources, including the popular press. Among vided in this paper draws upon an established tradition the better grounded predictions and assessments of research in organizations. The reader is encouraged are those by Buchanan and Linowes [1980a, 1980b], to review work by Blau [1970], Burns and Stalker Dolotta et al. [1976], EDP Analyzer [1980a, 1980b, [1961], Child [1973], Cyert and March [1963], Downs 1980c], Infotech International [1977a, 1977b], Jenkins [1967], Lawrence and Lorsch [1969], Meyer [1972], and Santos [1982], King [1982b], Kraemer [1982], Moore [1962, 1967], Perrow [1979, 1982], Pfeffer Kraemer and King [1982], McKenney and McFarlan [1982], Simon et al. [1954], Wagner [1966], Weber [1982], Mertes [1981], Olson [1983], Rockart et al. [1947, 1952] and Zannetos [1965]. These provide both [1979], Salerno [1981], and Statland [1979]. overviews and detailed definitions of the concept of

Computing Surveys, Vol. 15, No.' 4, December 1983 Centralized versus Decentralized Computing • 321

There are three separate aspects to the performance. It also encourages lower level centralization issue. Centralization versus managers to exploit innovative opportuni- decentralization of control concerns the lo- ties that improve unit-level performance. cus of decision-making activity in the or- Decentralization of control can create prob- ganization. Centralization implies the con- lems if lower level managers are incompe- centration of decision-making power in a tent, are not appropriately held to account single person or small group; decentraliza- for their decisions, or make decisions that tion implies that decisions are made at result in problems for other organizational various levels in the organizational hier- units or for higher management. archy. Centralization versus decentraliza- Centralization of physical location capi- tion of physical location concerns the siting talizes on economies of scale and preserves of facilities. Centralized physical location organizational integrity in operations. The has all facilitiesin one place; decentralized economies of scale arise from exploiting the location spreads facilitiesaround the region full potential of technologies that cause or the country, or even internationally. output to increase more rapidly than costs. Centralization versus decentralization of The costs of duplicating overhead and fa- function refers to the position of an activity cilities can be avoided, and organizational or responsibility within the structure of the protocols are easier to enforce. However, organization. For example, centralized ac- these advantages can be outweighed by counting and control would require all de- costs for organizational communications partments and units to report financial (including travel costs), transportation of data to a single unit, whereas decentrali- raw materials and finished goods, and zation might establish a number of profit maintaining close ties to customers and and cost centers with their own accounting clients. In special cases, such as military activities and require that only aggregated deployment or location of fire stations, the data be passed up to the corporate head- need for rapid response to unexpected sit- quarters. uations also dictates the need for physical decentralization. 1.2 Pros and Cons of the Alternatives Centralization of organizational functions keeps performance in line with Centralization of control preserves top organizational protocols and standards, management prerogatives in most deci- smoothes work flow on highly discrete sions, whereas decentralization allows tasks, constrains labor cost escalation by lower level managers discretion in choosing reducing the need for new employees, and among options. The former strategy pro- allows close monitoring and adjustment of motes continuity in organizational opera- work activities to better correspond with tions, but separates the making of decisions overall organizational operations. Decen- from their environment. If decisions are tralization of functions is advantageous misguided owing to poor top-level under- when the functions being performed require standing of the problem, or are subverted close cooperation with other units, when owing to poor enforcement at the lower the tasks being done require great worker levels, centralization can be disadvanta- discretion and less central guidance, or geous. Decentralization of control forces when regular interaction with members of lower level managers to take responsibility other organizational units requires too for their decisions, possibly improving their much "commuting" by individuals, either from the centralized functional department to the other departments, or vice versa. centralization and decentralization from the policy and artifact perspectives, and in particular illustrate The basic questions, then, revolve the importance of the concept of "distance" in most around tailoring organizational arrange- uses of the terms. It should be noted that there m ments to meet the constraints of organiza- considerable ambiguity in the meanings ascribed to tional size, the nature of the technology the terms centrahzation and decentralization as they are used in common discourse, and this ambiguity involved in organizational operations, and carries over into the research environment. See Pfeffer the needs of organizational clients and cus- [1982] for a useful discussion of this matter. tomers. These differences set the stage for

ComputingSurveys, Vol, 15, No. 4, December 1983 322 * John Leslie King the discussion of centralization and decen- 1970a, 1970b; Bernard 1979; Demb 1975; tralization that follows. D'Oliveria 1977; Glaser 1970; Golub 1975; Joplin 1967; Solomon 1970b; Statland 1979; 2. THE TRADITIONAL DEBATE Streeter 1973]. There has been a preference There is some disagreement in the litera- for consolidation until problems of geo- ture about the driving forces behind cen- graphic dispersal and increasing size of op- tralization decisions related to computing. erations forced decentralization. The ad- Some hold that prevailing organizational vantages of centralized location (mainly structures dictate computing system ar- hardware deployment) seemed especially rangements [Allen 1982; Brandon 1970; compelling, given Grosch's law, that com- Buchanan and Linowes 1980a, 1980b; puting power could be obtained at a func- Butch and Strater 1974; Burlingame 1961; tion equal to the square of the cost of the Dearden 1965; Dearden et al. 1971; Demb computer [Cale et al. 1979; Grosch 1953, 1975; D'Oliveria 1977; Ein-Dor and Segev 1975, 1979; Littrel 1974; Oldehoeft and 1978; Forest 1977; Glaser 1970; Golub 1975; Halstead 1972; Selwyn 1970; Sharpe 1969; Hannan and Fried 1977; Jenkins and San- Solomon 1966, 1970b]. This law, borne out tos 1982; King 1978, 1980; Long 1982; Mar- in subsequent tests, offered a powerful in- kus 1981; McFarlan 1972; Nolan 1973; Or- centive to centralize facilities in a single licky 1969; Patrick 1976; Perlman 1965; location, which frequently resulted in cen- Phister 1975; D. Price 1965; H. Price 1969; tralized computing control and functions as Roark 1971; Robey 1981; Rockart et al. well. Until the late 1970s, the authors of 1979; Scott-Morton 1975; Sprague and articles on computing centralization were Carlson 1982; Statland 1979; Streeter 1973; nearly unanimous in the conclusion that, Wagner 1966; Weiss 1975; Withington regardless of its other impacts, centraliza- 1973, 1980]. Organizations with centralized tion saves money [Bernard 1979; Glaser control and/or location of most activities 1970; Golub 1975; D. Price 1965; Roark are likely to have centralized computing. 1971; Selwyn 1970; Solomon 1970a; Stat- Yet, computing policies are sometimes set land 1979; Weiss 1975; Zannetos 1965]. without regard for other organizational Arguments favoring decentralization practices, or even in efforts to change those tended to focus not on economies, but on practices [Chervany et al. 1978; Danziger improved computing service for users 1979; Danziger et al. 1982; Kling 1978, [Demb 1975; D'Oliveria 1977; Glaser 1970; 1980; Kling and Scacchi 1979, 1982; Krae- Golub 1975; King 1978]. The central ques- mer and King 1979; Lucas 1982, 1984; Mar- tion was control over use of the technology kus 1981; Rockart et al. 1979; Scacchi 1981; and physical access to computing facilities. Wagner 1966]. The traditional debate over A common admonition was that the de- computing centralization does not provide partment that controlled computing would a clear statement of the issues because com- dominate its use [Berman 1970a, 1970b; puting has often been treated as a unique Bernard 1979]. 4 Users who could not get organizational resource that must be con- access to the technology would not use it, sidered separately from other organiza- thereby foregoing possible benefits of com- tional activities. Thus there is a question puting use. Users close to those in control about the appropriateness of various op- and to the location of the technology, and tions for organizing computing activities in who could interact directly with those pro- the context of other organizational arrangements. The logic of various options 4 This discussion usually centered on the finance and must be drawn from an assessment of the accounting departments, where the first computeri- arguments provided by procentralization zation of administrative data processing usually took and prodecentralization sides in the debate. place. This is a general comment, however. The first The most common arguments in favor of uses of administrative data processing in many uni- centralized computing have focused on lo- versitiestook place in computing centers used primarily for research and instruction, while in engineering cation and function: that is, whether to companies such uses often began on the machines centralize facilities and/or service [Berman normally used for engineering calculation.

ComputingSurveys, VoL 15, No. 4, December 1983 Centralized versus Decentralized Computing • 323 viding computing service, would make 3. ORGANIZATIONAL CONSIDERATIONS greater use of the technology. Similarly, IN THE DEBATE TODAY users with control or possession of their If the new technologies have made decen- own facilities would utilize the technology tralization affordable, why not decentral- more efficiently and effectively than under ize? The answer is twofold. For one thing, centralized conditions, thereby increasing although it is now possible to buy powerful the benefits that they derived from com- computer processors for a fraction of what puting use [Berman 1970a, 1970b; Bernard equivalent machines would have cost a dec- 1979; King 1978]. ade ago, the costs of computing entail more The centralization debate has tended to- than just the procurement of processors. ward trade-offs, in which the organizational For another, computing has grown in size, advantages of centralized control, uniform complexity, and centrality to many facets operations, and economies of scale have of organizational life. Decentralization of been pitted against user department needs computing often means decentralization of for ready access to computing and oppor- important organizational activities. It is tunity for fitting computing capabilities to necessary, therefore, to look beyond the department requirements [Demb 1975; traditional to the new and more compli- D'Oliveria 1977; King 1978; Zannetos cated factors that make centralization ver- 1965]. The trade-off can be reduced to one sus decentralization such a potent and per- of efficiency versus effectiveness [King sistent issue. We identify two factors that 1978]. The proponents of centralization appear to be of special importance: the have argued that centralized computing en- politics of organization and resources and sures efficiency and permits effective serv- the economic dynamics of deployment de- ice to users as long as good communications cisions. are maintained between the providers and the users. Centralization has been oriented 3.1 The Politics of Organization toward top-down control: control of com- and Resources puting costs, control of computing uses, and in some cases control over the information The politics of organization and resources being processed. The proponents of decen- refers to those formal and informal means tralization have argued that properly de- by which decisions are made as to how veloped, decentralized computing arrange- different organizational units are treated in ments are profitable, even if somewhat terms of resources, influence, and auton- more costly, because they improve the pro- omy [Arrow 1974; Benson 1983; Child 1973; ductivity of computer use [Wagner 1976]. Cyert and March 1963; Danziger et al. 1982; Decentralization has been oriented toward Downs 1967; Kraemer and King 1976; Law- bottom-up productivity improvement: im- rence and Lorsch 1969; Lucas 1984; Markus proved exploitation of computing for de- 1981; Meyer 1972; Moore 1967; Niskansen partmental tasks and improved system de- 1973; Perrow 1979; Wildavsky 1976; Yin sign to meet user needs. 1979]. Often the roles taken by specific The high cost of computers caused most units dictate their organizational power organizations to adopt relatively central- and resources. But sometimes there is con- ized computing policies, and proponents of siderable uncertainty about which roles decentralization usually had to fight an various units should assume in the "best" uphill battle. The advent of smaller, less interests of the organization, and disagree- expensive computers has greatly changed ments must be resolved through political the dynamics of the debate. User depart- processes involving the different interests ments can claim that decentralization is at stake. affordable, and maybe even less expensive 3.1.1 Consensus versus Divergence than centralization. This change has cre- in Goal Setting ated new challenges for managers faced with the responsibility for selecting a com- The centralization debate is fueled by dis- puting strategy for their organization. agreements over goals and the means for

Computing Surveys, VoL 15, No. 4, December 1983

• ~ :~.~.~r'~ ~ ...... ~ J, ...... 324 • John Leslie King accomplishing them. Much of the prescrip- Scacchi 1979, 1982; Markus 1981]. In fact, tive management literature assumes that there are important differences among fac- the ends are agreed on: that computing is a tions within most large and complex orga- tool to be utilized in the "best" interests of nizations that suggest the presence of con- the organization [e.g., Allen 1982; Axelrod flict and disagreement over organizational 1982; Dearden et al. 1971; Ein-Dor and goals and the means for meeting them. A Segev 1978; Gibson and Nolan 1974; Golub behavioral view of organizations suggests 1975; McKenney and McFarlan 1982; No- that individuals value their personal opin- lan 1973, 1977, 1979, 1982; Orlicky 1969; ions and the needs of their own depart- Rockart et al. 1979; Scott-Morton 1975; ments more highly than they do those of Sprague and Carlson 1982]. According to the organization at large. 5 Computing is this literature, the goal of computing poli- seen as package consisting of both technol- cies is to provide services at the most effec- ogies and the intentions behind their use. tive level given costs, to maximize the or- It is used to further the goals of specific ganization's profitability and performance organizational .actors (e.g., top manage- through use of computing, and to improve ment, financial controllers, data processing information flow throughout the organiza- professionals) in ways that might or might tion to expedite operations and manage- not improve organizational performance or ment. This goal-oriented view sees the or- help meet organizational goals. ganization as a system of interrelated tasks, The behavioral perspective has been con- staffed by employees who are primarily spicuously absent from the discussion of concerned with maintaining and improving centralization and computing. Disagree- organizational performance [Danziger et al. ments over computing policies have been 1982; Kling and Scacchi 1979, 1982; Scac- attributed to misunderstanding of either chi 1981]. In this rationalistic framework, the facts or the goals. In this rationalistic computing systems are instruments (or in interpretation, the solution to disagree- more elaborated settings, "environments") ments is to conduct fact-finding studies that, when properly managed, help the or- (e.g., cost-benefit analyses) and to set or ganization to meet its goals, adapt to its clarify goals through discussion (e.g., user surroundings, and improve the perform- committees). It is seldom suggested that ance of its employees. the facts are simply elusive, or that disa- Computing policy in this context seeks greements on goals are intractable. to deploy computing resources in a manner Although the behavioral complexion of that best facilitates their productive use computing policy making has been ne- and maintains managerial control over im- glected in discussion of centralization is- portant organizational information. The task of management is to ensure that this 5This section relies on behavioral research in organi- tool is made available to users at the lowest zations, little of which deals directly with computing feasible cost, taking into account factors use in organizations, but all of which is relevant. In such as geographical remoteness, special- particular, it reflectsthe workof Allison [1971], Arrow ized user needs, or high concentrations of [1974], Burns and Stalker [1961], Child [1973], Cyert and March [1963], Downs [1967], Lawrence and demand for service that might warrant Lorsch [1969], Meyer [1972], Moore [1962, 1967], costly solutions. The hallmark of a ration- Niskansen [1973], Perrow [1979, 1982], Wildavsky alistic approach is the attempt to balance [1976], and Yin [1979]. Of special importance to this efficiency and effectiveness. The design of analysis is the observation that the idea of genuine "organizational goals" can be illusory [Allison 1971; policy concentrates on the overall goals of Cyert and March 1963; Downs 1967; Niskansen 1973; the organization. Perrow 1982]. Organizations may appear to follow The weakness of the rationalistic ap- coherent policies, but in fact goals are often generated proach lies in its assumption of consonance by internal organizational conflict and undergo fre- between the goals of individual organiza- quent change. In studies of computing this has been noted by Danziger [1979], Danziger et al. [1982], King tional actors and the stated goals of orga- [1978, 1982a, 1982b], Kling [1980], Kling and Scacchi nizational leadership [Danziger 1979; Dan- [1979, 1982], Kraemer and Dutton [1979], Kraemer et ziger et al. 1982; Kling 1980; Kling and al. [1981], Markus [1981], and Scacchi [1981].

ComputingSurveys, Vol 15, No. 4, December1983 Centralized versus Decentralized Computing • 325 sues, many studies indicate that the pri- (5) The need to meet information require- mary factors in setting policy are the inten- ments of management. tions behind computing use, not the nature (6) The need to provide computing services of the technology itself [Danziger et al. in a reliable, professional, and techni- 1982; King 1982b; King and Kraemer 1984; cally competent manner. Kling 1980; Kraemer et al. 1981; Lucas (7) The need to allow organizational units 1982; Markus 1981; PPRO 1980]. In small, sufficient autonomy in the conduct of narrowly focused organizations, the inter- their tasks to optimize creativity and ests of all the actors in the organization performance at the unit level. might coincide. But such organizations are (8) The need to preserve autonomy among usually too small to be facing the centrali- organizational units, and if possible, to zation issue. Larger, more complex organi- increase their importance and influ- zations usually have many and diverse so- ence within the larger organization. cial groupings', multiple organizational (9) The need, wherever possible, to make tasks and objectives,~ and more decentral- the work of employees enjoyable as well ized decision-making structures. A single as productive.6 decision-making group that speaks for all interests is unlikely [Kling 1980]. Disagree- These nine objectives can be viewed from ments over how best to use computing are a rationalist perspective, with the goal in therefore endemic to such organizations. formulating policy being to balance each Students of organizational change sug- against the others so as to optimize the gest that in many complex organizations, overall result. The problem with such an polices swing back and forth between cen- approach is that, depending on to whom tralization and decentralization [Lawrence one talks, a different set of priorities for and Lorsch 1969; Moore 1962, 1967]. Stud- these considerations will emerge. More- ies and discussion of organizational goals over, in some cases, they contradict one are important in setting computing policies, another. For example, the need to maintain and genuine consensus is sometimes pos- integrity in operations across the organi- sible, but the task requires sensitivity to a zation may conflict with the desires of or- wider array of organizational interests than ganizational units for autonomy. The way those represented by the dominant decision in which these different factors interact can makers. be seen in the following illustrative history, which briefly recounts the evolution of pol- 3.1.2 The Driving Factors in the Debate e The importance of entertainment value in the exten- From the behavioral viewpoint, there are sive adoption of computing technology has been over- nine organizational objectives that drive looked in most of the research, but the success of the centralization/decentralization debate computer-based entertainment products and the use over how to manage computing: of games as major marketing tools by computer vendors suggests that this is an important factor. It is (1) The need to provide computing capa- clearly a major component of the widespread acquisi- bility to all organizational units that tlon of computers for home use [Vitalariand Venka- tesh 1984],and a largenumber of fascinatingcomputer legitimately require it. games have been developed (and are activelyplayed) (2) The need to contain the capital and at umversitiesand other organizations.The entertain- operations costs in provision of comment potential of computing plays an important role puting services within the organization. in encouraging experimentation and learning among (3) The need to satisfy special computing those who develop and use computer systems, and lessens the apprehensions that new users might oth- needs of user departments. erwise have about getting involved with computers. (4) The need to maintain organizational The author's experience indicates that most people integrity in operations that are depend- are curious about computers and desireto experiment ent on computing {i.e., avoid mis- wlth them and use them, glven the chance. Resistance to use of computing seems to persistonly as long as matches in operations among depart- there is uncertaintyabout the impact of computer use ments}. on one'sjob and statuswithin an organization.

Computing Surveys,Vol. 15, No. 4,December 1983 326 • John Leslie King icies governing administrative data pro- to buffer the finance/accounting depart- cessing over the past two decades.7 ment from objections to specific reporting requirements imposed on other units. 8 3.1.3 An Illustrative History This exclusive relationship between the finance/accounting unit and the computer Computers were first applied to tasks that did not last very long. Other organizational were easily rationalized and programmed. units began to see possible applications of In organizational data processing these the computer, and to explore means for were usually in areas such as payroll prep- exploiting the technology. These new users aration and accounting that had well-de- could either acquire their own computers veloped procedures, which allowed them to or use the services of the finance/account- be automated easily, and large processing ing unit. This posed a dilemma. Because volumes, which made automation attrac- computers were large, stand-alone ma- tive. Such tasks frequently were already chines, their high capital costs and semiautomated on unit record equipment. Grosch's law suggested that other users The conversion of these tasks to digital should share the finance/accounting com- computing was a natural progression. At puter. The finance/accounting unit's ex- the same time, there was a trend toward perience with computing helped ensure that centralization of the financial control func- computing services would be provided in a tion in many organizations for other rea- competent manner. And the close ties be- sons [Simon et al. 1954]. Centralized com- tween finance/accounting and top manage- puting in the finance/accounting unit was ment ensured that the latter's interests a logical step. Capital and operations costs would be served. However, sharing meant related to computing could be easily man- that the new users would lose some depart- aged through existing decision processes mental autonomy by placing their own data within the finance/accounting unit, espe- processing tasks (which in some cases were cially the normal budgetary process semiautomated or computerized) in the whereby the department negotiated for its hands of another organizational unit. This share of organizational resources. The fi- had two disturbing results for the operating nance/accounting unit controlled the com- units: It moved an important function out puting system and could tailor the system to meet its needs. Well-established com- of its home unit to another unit, and it provided the finance/accounting unit with munications channels between finance/ac- a powerful rationale for increasing its own counting and top management facilitated budget and staff,and thereby its power and upward flow of important financial infor- influence in the organization. New users mation. The finance/accounting unit could found that they were required to follow the justify increased organizational invest- finance/accounting department's guide- ments in computing on grounds that they lines for computing use, and that their jobs would serve the "whole organization." This would generally receive lower priority. Fi- centralization of computing began the ev- nally, the finance/accounting unit main- olution of a new form of bureaucracy based tained an exclusive hold over the exciting on the special skills of computer techni- and status-raising technology of comput- cians [Danziger 1979]. The newness and ing. mystique surrounding computers gave re- Complaints from operating units brought ports printed by computers authority that this situation to top management's atten- other reports did not have, while the tech- tion, and the response to this dilemma was nical complexities of computing were used usually an edict aimed at balancing the high procurement costs of computing with the 7 This history is a hypothetical synthesis based on the needs of various user groups. Decentralized comments of many who experienced and wrote about the evolution of computing use in organizations over the past two decades (as found in the references). s Excellent accounts of the influence that computer- Many organizations never had the experiences re- generated reports can have are found in Danziger et ported here. al. [1982] and Kling [1980].

ComputingSurveys, Vol. 15, No. 4, December 1983 Centralized versus Decentralized Computing • 327 computing centers were established in or- agreements between the finance/account- ganizations where geographical considera- ing unit and other users, who found them- tions required them, or when operating selves increasingly dependent on an impor- units had sufficient influence to overcome tant resource controlled by a "service the centralist arguments of economies of agency" with little understanding of their scale and integrity of use. But in most needs or little inclination to take their organizations the centralist arguments pre- problems seriously. They were also becom- vailed, usually on economic grounds. The ing increasingly dependent on the technical computer center remained in the finance/ data processing specialists who worked for accounting unit, and procedures were set the finance/accounting unit (as were the up to allow access by other users [Gibson finance/accounting users also). and Nolan 1974; Nolan 1979]. In many organizations these pressures Neither the centralized nor the decen- resulted in the creation of independent data tralized strategy proved to be perfect. processing departments, usually under di- Where multiple centers were established, rect control of top management. This re- there arose criticism of "proliferation" of form was designed to preserve the benefits expensive computers and lack of techni- of centralization while reducing the inter- cally competent computing operations, as departmental disagreements about access well as difficulty in meeting the needs of to services. It also moved computing closer top management for information. Costs for to top management by elevating the status computing did indeed rise rapidly, partly of the data processing professionals. Com- because of growing applications of the tech- puting services were to be run like a busi- nology and partly because of the exploita- ness, providing high-quality services to all tion of the budgetary leverage computing users while achieving maximum efficiencies provided to the departments that had their and effectiveness in the computing opera- own computers.9 The professional cadres of tion. To overcome disagreements among data processing specialists that had user departments about priorities and qual- emerged were able to consolidate their ity of service, and to improve accountability power around their professional knowledge, to top management, these independent buffering themselves from demands from data processing units established manage- both their clients and top management rial mechanisms such as structured needs [Danziger 1979; Danziger et al. 1982]. In assessment procedures, complete with response, some organizations centralized cost-benefit analyses, to assess user re- (or recentralized) computing to gain or re- quests for new systems or rewrites of old gain control of the increasingly important systems. They also established "charge- and complicated data processing function. out" policies to impose pseudomarket con- In other cases, control over computing straints on use of computing services. tended to devolve to the decentralized fa- Training of managers and users in use of cilities as part of a larger decentralization computing was increased. Finally, user pol- of organizational activity. icy boards and steering committees were There were also problems with central- established to help determine organiza- ized installations. Centralization helped tional needs for computing and the means top management contain costs and retain to meet them. In short, these reforms were control over the growth and use of the part of a concerted effort to find facts and technology, but also resulted in serious dis- establish consensus on goals, in keeping with the rationalist view. w 9 Budgetary leverage refers to the role computer use Unfortunately, the upward movement of can play in justifying increases in departmental budg- data processing in the organizational hier- ets. In this respect, computing is like many other organizational activities (particularly those with bureaucratic characteristics) that justify and enhance ~o Policies of this kind are discussed by Danziger et al. the organizational positions of the units that carry [1982], Ein-Dor and Segev [1978], Gibson and Nolan them out [Danzlger 1979; Kling 1980; Kraemer and [1974], King and Kraemer [1984], Kraemer et al. King 1976; Markus 1981; Wildavsky 1976]. [1981], and Nolan [1973, 1977, 1979, 1982].

ComputingSurveys, Vol. 15, No. 4, December1983

..... ~ ~ ~_~? ...... ~~ ~ ~ 328 • John Leslie King archy removed it even further from the questioned, the establishment of minicom- operating needs of user departments. The puter operations in user departments could establishment of an independent data pro- proceed. In many organizations a number cessing unit under the control of top man- of satellite or stand-alone computing cen- agement made it difficult for some users to ters built around minicomputers emerged. negotiate favorable terms for service, since In the late 1970s the microprocessor the independent unit was there to serve again cut the cost of basic computer equip- "all" users. In fact, the user departments ment. By 1980 a computer system with an that made the heaviest demands on com- 8-bit processor, 48K main memory, 5- puting services, such as the finance/ac- megabyte hard disk drive, operating sys- counting unit, immediately became the tem, terminal, and printer could be pur- most important "customers" of the new chased for less than $6,000. Expenditures independent data processing department, this low are almost insignificant in the since they provided the bulk of the data budgets of major departments in large or- processing department's business. But even ganizations, and could be easily approved. the major users faced loss of autonomy in A new era of"proliferation" was under way. their operations, inflexibility in exploiting 3.1.4 The Current Situation the productivity-improving potential of the technology, and lack of opportunity to ex- Small and inexpensive minicomputers and periment with computing. They now had microcomputers have radically changed to accept the standards of the data process- the centralization/decentralization debate. ing department. The creation of the in- User departments can now obtain their own dependent data processing department computing capability, in some cases with- accommodated several of the major consid- out the knowledge of top management or erations in the centralization/decentrali- the central data processing department, zation debate, but by no means all of them. and in other cases by arguing that the cost The centralized, independent data pro- is so low that the economic benefits in favor cessing shop dominated data processing of centralization no longer apply. Wide- from the mid-1960s to the mid-1970s. Dur- spread use of small computers can provide ing this time the economic advantages of highly individualistic service to all the decentralization prevailed, but new techno- partments needing computing, allow users logical capabilities such as time sharing to establish and maintain autonomy in over remote terminals and use of job entry/ their operations using their own equip- output sites provided users with more direct ment, and provide users with hands-on op- access to the computing resource. This dec- portunity to enjoy computing use while im- ade of stability began to give way in the proving departmental productivity. The mid-1970s, however, as technologies such prodecentralization forces in the debate can as the minicomputer arrived. The minicom- now argue that Objectives 1, 3, 7, 8, and 9 puter could provide considerable computing of those listed above are met. Concern for power but at a much lower price than large costs (Objective 2), they argue, is no longer mainframes. Minicomputers could do many an issue since these machines are so inex- of the smaller jobs then being done on the pensive to procure, and off-the-shelf soft- large mainframes, and allowed acquisition ware for such systems makes it possible to of computing capability in small pieces. keep operations costs down. Departments that had depended for service Assuming for the moment that these on the centralized computing installation arguments are correct, there remain several could assemble a computer system in a problems. First, the use of computer sys- number of inexpensive, incremental pur- tems by users not familiar with the broader chases. Individual purchase approvals requirements of system management might could be made at much lower levels in the compromise the quality of computing activ- organization hierarchy than those required ity in the organization (Objective 6). Many for purchase of large and expensive main- practices that the centralized data process- frames. As long as these purchases were not ing shops have learned over the years, often

ComputingSurveys, Vol. 15, No. 4, December 1983 Centralized versus Decentralized Computing * 329 at considerable cost, will not be known to a question of where the computer proces- the new users. These include methods for sors are located or how they are acquired. forecasting system requirements and costs, Rather, the issue is control over computing: development of backup and fail-safe pro- who does it, what they do with it, and how cedures, adoption and enforcement of doc- they do it. Control must reside someplace. umentation and maintenance standards, It cannot be shared equally among different and effective methods for dealing with ven- groups of different opinions. The basic dors and suppliers. Individual installations question has never been, "Which way is might reduce the impact of problems in best?" It is usually, "Who's way is it going these areas, but the aggregate of such prob- to be?" The advent of small computers with lems throughout the organization could be low purchase prices does not change this. serious. It merely alters the bases on which the Second, giving user departments carte various sides take their positions and con- blanche to establish their own computing struct their arguments. operations increases the likelihood of dis- The issues involved in centralization/de- integration in interdepartmental opera- centralization decisions are deeply tied to tions (Objective 4). This is especially true organizational behavior, and the conse- if incompatible systems are adopted quences of centralization/decentralization through which interdepartmental informa- politics become increasingly important as tion must flow, but it applies even to situ- organizational investments in and depend- ations in which small systems are compat- ency on computers increase. ible with one another at the hardware and software levels. Unless the whole package 3.2 The Economic Dynamics of procedures and protocols required for of Deployment Decisions smooth use of the technology is standard- ized throughout the organization, there Economic opportunities or constraints are arises the opportunity for serious mis- often the most extensively discussed crite- matches from one department to another. ria in the political process of deciding The dilemma of deciding between organi- whether to centralize or decentralize. zational standardization and departmental Changing economic conditions keep alter- autonomy persists. ing the economic rationales behind either Third, the devolvement of data process- course of action. As long as the economies ing activities to the departmental level can inherent in different deployment strategies increase the difficulty of obtaining data for are undergoing change, there can be no top management use (Objective 5). For permanent resolution to the centralization years a major goal of the data processing question on economic grounds. To under- profession has been to enhance the provi- stand the economic dynamics of computing sion of information to top-level decision as they relate to centralization/decentrali- makers. But this is difficult even with cen- zation decisions, the issue can be structured tralized operations. The problems are not in terms of the costs and benefits and how so much in the technology, but in determin- they interact. ing what information to provide and how to provide it. The adoption of differing 3.2.1 Computing Costs departmental standards and protocols makes uniform collection of data for up- The cost dynamics of computing have ward reporting more difficult, whereas the changed substantially over the past two imposition of organization-wide standards decades. Nowhere has this change been again diminishes departmental autonomy. more dramatic than in the relative costs of Should computing be centralized or de- hardware and software. Boehm [1979, centralized? Unfortunately, there is no easy 1981] claims that in 1955 computer hard- answer. The fundamental considerations in ware costs dominated software costs 7:1, the debate over centralization persist, re- but by 1985 software costs are expected to gardless of the technology. It is not so much dominate hardware costs 9:1. This is a dra-

ComputingSurveys, Vol. 15, No. 4, December1983 330 • John Leslie King matic reversal, with equally dramatic ef- more than offset by higher price tags for fects on perceptions about the costs of com- facilities, software procurement, software puting generally. Hardware is usually ac- maintenance, data management, data com- quired before software, and so this shift has munications, and computing management. reduced the entry costs of computing.~1 These costs are rising for five reasons. Start-up has become comparatively less The first is growth in application. As com- costly than successfully implementing computing use grows and the number of appli- puting systems that meet organizational cations in operation increases, so does the needs. Computing now appears to many demand for both technical and managerial decision makers as inexpensive, but a closer people to develop these systems and main- look reveals that this is not so. tain them. And demand for programmers, The little comprehensive research on the systems analysts, and managers skilled in costs of computing that exists suggests that data processing administration has kept they remain substantial, and are often well ahead of supply, bidding up the price higher than they are estimated to be [King of their labor. This condition will continue 1982b; King and Schrems 1978]. TM There as long as the need for new talent is not are many hidden costs, such as expendi- offset by an increase in the supply and/or tures for computing-related staff in user productivity of such labor. 13 departments not accounted for in comput- Second is increasing maintenance costs. ing budgets [Morrisey and Wu 1979], staff System maintenance has been estimated to time of users and upper management who consume most of programmers' and ana- must deal with intrusions and disruptions lysts' time [Boehm 1979, 1981; Ledbetter arising from computing [Leintz and Swan- 1980]. Kommunedata, a publicly owned son 1974; Leintz et al. 1978], and "lost" service bureau in Denmark that provides productivity owing to employees who spend computing to all 276 municipalities and their time "playing" with computing [Kling counties in the country, estimates that it and Scacchi 1982; Scacchi 1981]. Such costs devotes approximately 20 percent of system can rightly be attributed to many cross- development manpower per year to main- departmental functions, but they can be tenance of each system that it develops significant. [Romer 1979]. Boehm [1979, 1981] esti- Computing costs are not only high, but mates that by 1985, 70 percent of all soft- appear to be going up [Buss 1981; Dolotta ware expenses will be for maintenance. De- et al. 1976; King 1982b; Nolan 1979; Phister partment of Defense software maintenance 1979], a paradoxical situation, given the costs run between $2.5 and $4 billion per decreasing costs of computers. Dramatic year, and this figure is going up. 14 As the reductions in price/performance ratios for number of systems increases, so do the computer processor hardware have been carrying costs of maintenance. Mainte- nance also requires high-quality software 11 The choice of hardware first, followed by choice of documentation, which is expensive to pro- software, is a practice that might be changing. This is especially true in the area of smaller computers, where 13 There has been considerable discussion about alle- availability of certain software packages directs the viating this problem, the most common proposal being customer to equipment that supports those packages. that of improving the productivity of individuals who This phenomenon is clearly evident in the case of the develop software [Boehm 1979, 1981; Chrysler 1978; rapid migration of personal computer users to the IBM DeRoze and Nyman 1978; Fisher 1974; Freeman 1983; PC, which almost instantly upon release became the Ledbetter 1980; Leintz and Swanson 1974; Mornsey standard around which personal computer software and Wu 1979; Phister 1979; Scacchi 1984; Zelkowitz would be written. Nevertheless, the computer itself 1978]. Whether such methods and tools will make up remains the most expensive single purchase when first the difference is unclear. If they do not, equilibrium obtaining a computer system, and its price (with nec- will be achieved when the pace of systems development essary peripherals and operating systems software) slows. constitutes the "entry price" of computing. u This figure is obtained by multiplying estimated ~2 This account is of necessity constrained by a lack of DoD software expenditures of $5 billion per year detailed, empirical assessment of the economic Impact [Boehm 1979; Fisher 1974] by estimates of mainte- of computing. Useful references to this subject include nance as a percentage of overall costs for military King and Schrems [1978], Phister [1979] and Roark software, thought to be between 50 and 80 percent [1971]. [Boehm 1981; Leintz et al. 1978].

ComputingSurveys, Vol. 15, No. 4, December1983 Centralized versus Decentralized Computing • 331 duce [Boehm 1979; DeRoze and Nyman chi 1984]. To the extent that system inte- 1978; Fisher 1974; Leintz and Swanson gration does occur, it comes by linking to- 1974; Leintz et al. 1978], and is a low- gether otherwise stand-alone applications, priority task in most software environ- so that the output of one component serves ments. Because of the high turnover rate as input to another. Integration makes sys- among programmers, those who design systems more complicated, which increases tems seldom stay around long enough to system costs, and also requires successful help maintain them through the system and timely performance of organizational cycle. Large and complex systems are built units that use the systems. Integration by teams of programmers in which no single among unintegrated systems is usually co- analyst or programmer understands the ordinated through policies and protocols whole system, and so the task of maintain- that allow for slippage along the bounda- ing them is often experimental, time con- ries: delays and problems can be taken care suming, and expensive. of within the system or unit before it is A third factor is the growing complexity necessary to interact with other systems or of systems, and the disruptions caused units. Integrated systems make organiza- when they are implemented and when they tional units more interdependent in real malfunction [King and Kraemer 1981; time, so problems in one system or unit can Kraemer and King 1979, 1982; Nolan literally stop progress in others simply by 1979]. Simple, routine applications (e.g., a disruption of the process of interaction. billing program for an electric utility) are A final factor contributing to the growing basically automated versions of previously cost of computing is the fact that many manual operations. Their scale and com- existing systems will have to be rewritten plexity are relatively low. However, as ef- in the next decade [Boehm and Standish forts are made to improve these systems by 1982]. A large number of major systems adding new features (e.g., providing cus- were developed during the 1960s and 1970s, tomers with comparisons of this year's elec- with several features in common: They tricity use to last year's), their implemen- were very large systems to do large routine tation requires considerably more prepara- tasks such as accounting and record keep- tion, training, and time, all of which are ing; they were applied to important tasks very expensive. Complex systems are more so that they could not simply be abandoned; likely to malfunction and are more difficult and they were developed under obsolete to fix because of their interdependency in hardware and software technologies. Many operation [Kling and Scacchi 1982; Scacchi such systems are still operating today, run- 1981]. Most fixes are like "patches" sewn ning under emulation on modern main- on the fabric of the system. Often, they do frames. But they are becoming exceedingly not solve the problem, but "work around" costly to maintain, and in many cases, unit. In complex systems, patches frequently reliable in operation. The costs that will be generate new problems, which then must incurred in rewriting these systems are sub- be patched as well. Eventually, the design stantial. Rewrites must take place apart integrity of the original system is destroyed, from the existing systems (which must still and systems often must be completely re- continue in operation), and on top of other built before their expected useful life is over demands such as maintenance of existing [Boehm and Standish 1982; Kling and systems and development of entirely new Scacchi 1979, 1982; Ledbetter 1980; Leintz systems. Rewrites will also incorporate and Swanson 1974; Leintz et al. 1978]. modern software engineering techniques in A fourth factor is the growth of inte- order to keep long-run maintenance costs grated systems on which organizations de- down, but these will require major up-front pend. Integration means the interconnec- development investments. tion of different systems or subsystems to In summary, the overall cost of comput- form larger systems. Large efforts to build ing is rising rapidly as new systems are integrated systems from the ground up have implemented, as the price of technical tal- been attempted, but most have not been ent increases, as maintenance costs of ex- successful [Kraemer and King 1979; Scac- isting systems mount, as the complexity of

ComputingSurveys, Vol. 15, No. 4, December 1983 332 • John Leslie King

computing systems increases, as previously successful systems are accrued by way of stand-alone automated tasks are linked to- the cost overruns incurred in development gether in complex, integrated systems, and of unsuccessful or marginally successful as the need to rebuild older systems be- systems. Despite structured design and comes more acute. The only factors that other techniques for "deliberately" produc- might alter these cost dynamics are a dra- ing successful systems, most development matic increase in the productivity of the is a trial and error process. The failure rate technical specialists who build and main- is just as high as it was 15 years ago [Buss tain such systems (brought about through 1981]. implementation of new methods and/or The economic benefits claimed for most technologies), or a curtailment in the systems are not based on the amount of growth of computing application causing money saved, but on the differences in the the demand for such resources to fall back character of the tasks performed with com- in line with supply. puting, which seem beneficial to the using organization. 15 Often, these benefits are in- 3.2.2 The Question of Benefits tangible, particularly those dealing with "improved information." This problem is Computing is used because it has demon- illustrated by experience with another in- strable benefits. These benefits do not ac- formation technology, the office photoco- crue across the board, but are primarily pier. This technology has increased both concentrated in those applications of the the volume and velocity of paper flow. But technology that assist in conducting rou- what is the "benefit" of having copies of tine, well-understood tasks [King and memos go personally to five people instead Kraemer 1984; King and Schrems 1978]. posting a copy in a central location where Benefits predicted as the most significant all can see it? Improvements in organiza- economically--cost reductions and staff re- tional information flow and communica- ductions-have not appeared as expected, tion are extremely hard to measure. Often, and are now seldom promised in systems no one knows what the information flow or proposals [Edelman 1981]. The primary the quality of communication was in the benefits of computing have been three: im- first place, and it is difficult to put a value proved speed and accuracy for some tasks, on the change even if it can be firmly avoiding the need to hire new staff; quali- measured. tative improvements in operations (e.g., re- It is argued that the economic benefits of duced error rates, more flexibility in re- computing are demonstrated by the fact porting and data manipulation, a greater that so many organizations use it. 16 This range of analyses available); and increased argument makes two critical assumptions: capabilities for performing structured but that the true costs, short run and long, are complex tasks (e.g., airline reservation sys- known to the organization when it makes tems). its decisions, and that estimates of the ben- Benefits from computing application to efits it will receive are reasonable and not more complex and uncertain tasks such as exaggerated. If these two assumptions are management decision making are more dif- in error, outcomes can be drastically differ- ficult to ascertain. Most claims that appli- ent from expectations. The benefits of cations such as decision support systems "save money" weaken considerably on close ~5 The fact that little rigorous research has been done examination because extensive develop- does not mean that tangible benefits do not accrue ment costs are usually excluded from the from advanced applications. Edelman [1981] presents data suggesting that direct economic benefits from cost-benefit equation [King and Schrems cost savings and avoidance do sometimes result from 1978]. There is a propensity for measurable such applications. See also Axelrod [1982]. costs to outdistance measurable benefits, ~e This is based on the assumption in theories of which is why "runaway costs" and "cost welfare economics that households are the best deter- miners of their own welfare. There are obvious excep- overruns" are familiar terms, whereas "run- tions in which intervention is required from outside away benefits" and "benefit overruns" are agents (e.g., experts), but as a general rule this as- not. The learning costs that go into building sumption seems reasonable.

ComputingSurveys, Voi. 15, No. 4, December 1983 Centralized versus Decentralized Computing • 333 adopting computing systems are not nec- (5) Economic benefits from advanced ap- essarily economic; some adoption decisions plications such as decision support systems are made on strict economic grounds, but and planning systems are more difficult to most are influenced to some degree by other identify, especially in relation to their costs, organizational and political factors (e.g., although recent research suggests that they department managers' strategy to become do accrue in some circumstances [Edelman a computer user to build their budgetary 1981]. claims, or simply the desire to have their (6) Claims of economic benefit are usu- own computers). ally made to justify proposed computing A final problem in assessing the benefits developments, but other organizational and of computing across organizations rests in political factors figure prominently in mo- the choice of the applications evaluated. tivations to engage in computing. Successful systems demonstrate what is (7) Regardless of the potentials for com- possible, but not what is likely to happen puting benefits demonstrated by advanced in most instances. Predicting industry-wide user organizations, most organizations will potential of computers on the basis of ex- take longer to realize such benefits if they periences of a small number of highly tal- in fact do so at all. ented organizations is unwise. Sophisti- These findings do not suggest that com- cated applications and innovations appear puting is unbeneficial in economic terms. sparsely across the population of organiza- Rather, they imply that there are othel' tions [King 1982a; Kraemer et al. 1981]. A kinds of benefits that play a role in orga- given organization might have one or two nizational decisions about computing, and significant and successful innovations, but the rest are either rather routine applica- that the hoped-for economic benefits of computing systems often do not accrue actions or are failed examples of more ambi- cording to plan. tious efforts. The benefits of such innovations are often discussed in terms of their demonstrated potential in a few special 3.2.3 Computing Costs and Decentralization cases, and not on their probable performance in wide deployment. This review of computing costs and benefits A summary of what is known about the provides a base from which to analyze the benefits of computing in organizations impact of decentralization. The experience yields seven findings: in large organizations over the past two decades suggests that decentralization en- (1) Benefits seem to be difficult to pin- tails organizational changes that are likely point in strict, economic terms, although to prove costly for two primary reasons. the fact that computing has been and con- The first factor is the expansion of com- tinues to be heavily adopted and used sug- puting activity as users gain control of com- gests that organizations believe that the puting resources. If computing activities benefits are there. were undertaken only to improve efficiency (2) Direct economic benefits, such as and effectiveness, there would be no prob- staff reductions and cost savings, seem not lem. But computing lures users for other to have materialized in most applications reasons, not the least being its attractive- of computing to administrative data pro- ness as an exciting and potentially useful cessing. technology. Faith and delight in computing (3) Indirect economic benefits, such as is a strong motivator for adoption, and improved capabilities and cost avoidances often overcomes the need for "demon- or displacement, do seem to have accrued strated benefits" in the decision to adopt as a result of computerization of adminis- [Danziger et al. 1982; Kraemer and Dutton trative data processing. 1979; Kraemer et al. 1981; Turkle 1979]. (4) Most of the measurable economic This phenomenon is not new. Technologi- benefits from computing appear to come cal advances in such fields as automobiles, from fairly routine applications (e.g., ac- photographic equipment, home entertain- counting, record keeping). ment equipment, household appliances,

ComputingSurveys, Vol. 15, No. 4, December 1983 334 • John Leslie King

and medical technology have led to major user departments had already spent over increases in consumer expenditure,iv As $1 million on microcomputer and mini- new capabilities emerge, the perceived computer equipment in the previous twelve needs of users increase. months, without the knowledge of the data This seems likely to happen in the case processing department. Had the organiza- of new, small computer systems. When a tion instituted and maintained strict cen- user department acquires an inexpensive tralization of acquisition control, it might system to take care of simple departmental have been able to review and guide these needs, the needs often begin to grow, as do procurement decisions. But that would investments in the system as new and more have entailed planning costs to develop enticing equipment becomes available. Be- protocols, and increased management costs fore long, the overall investment has grown for monitoring and evaluating the decisions far beyond expectations. of subunits. A case recently observed by the author Aside from the question of whether sen- illustrates this. A university's financial ad- sible expansions of computing use take ministrators were dissatisfied with the place, there is the further probabability service that they received from the campus that in time the small, decentralized sys- computer center, and bought a fairly pow- tems will become true satellites of the larger erful minicomputer to do their own com- centralized systems. This will occur in cases puting jobs. They hired ten people to staff where users with their own minicomputer the enterprise. Within three years they had or microcomputer systems desire to share two minicomputers, were buying a third, data or establish electronic communica- and had a computing staff of forty. The tions links with the central resource and computer center they left had also grown with each other. Under this arrangement, bigger. These users had the best of inten- the new systems become additional "ter- tions when they got their own system, but minals" connected to the mainframe. When they did not know what the computer staff users begin to make substantial demands had learned over the years: that computing for data, capabilities for uploading and is a very expensive business. To accomplish downloading files, and establishment of their goals they continually had to increase electronic mail and other office automation their investment. And their investment was capabilities, the central resource will have the university's investment. In very few to be upgraded to deal with the demand. cases does a computing installation, cen- Thus, instead of decentralization through tralized or decentralized, get smaller and use of minicomputers and microcomputers cheaper over time. signaling the demise of the mainframe era When control over computing procure- [Breslin and Tashenberg 1978], it could ment and system development decisions signal the beginning of an era in which both devolves to users, unwise investments can the centralized and decentralized comput- be made, sometimes without the knowledge ing activities of the organization grow dra- of top executives. A situation like this came matically. In a sense, each new microcom- to light recently in a computer printer man- puter or minicomputer installed in a user ufacturing company that was deliberating department can be thought of as either a over whether to make a $300,000 invest- potential data processing center or another ment in new equipment for the data proc- terminal to the mainframe; in some cases essing center. ~8 Someone suggested that the they will be both. money might be better spent for micro- The second major factor suggesting that processors for users. A study revealed that decentralization of computing will increase computing costs arises from disruptions in organizational operations that often ac- 17 Systemic research on the adoption and use of com- company decentralization [Utta11982]. De- puters in the home is relatively new. See Vitalari and Venkatesh [1984]. centralization requires change, which must is The author thanks Suzanne Iacono for assistance in be carried out coherently throughout the complhng this information. organization. Too often, decentralization

ComputingSurveys, Vol. 15, No. 4, December 1983 Centralized versus Decentralized Computing • 335 occurs by default as beleaguered data pro- 3.2.4 Decentralization Benefits: cessing departments simply give users per- Possibility and Reality mission to get their own systems, or users get their systems without asking whether We have discussed the hoped-for benefits anyone else thinks they should or not. In to users from "stand-alone" decentraliza- such cases there are no comprehensive tion (i.e., small, independent computer ac- plans. Implementation of completely new tivities in user departments): easier access systems in these decentralized operations to the technology, increased user involve- will expand the number of applications the ment in system design and modification, organization must support, whereas dis- increased sophistication in the use of the placement of services provided by central technology, and the opportunity for users data processing can duplicate existing sys- to decide for themselves how computing tems. In either case, costs rise. best can be of service to them. However, More important is the likelihood that these benefits will only accrue to users (and decentralized users will develop systems thereby to the organization overall) if the that clash with current tasks and interac- uses themselves are beneficial. Users must tions with other departments. Systems and know how to distinguish wasteful from pro- their operation will require constant ad- ductive applications. User involvement in justment to bring everything into harmony. design will produce systems of greater util- If left alone long enough, things might "sort ity only if users are sufficiently knowledg- themselves out" through natural processes. able about computing to design in the most But this can take a long time and exact a productive features while leaving out fancy considerable toll in frustration and re- but costly "wish list" features. Users almost sources. certainly will learn more about computing, The dilemma for the organization, how- but this knowledge must be comprehensive ever, is that there are high costs from either enough to engender a sophisticated under- careful control or no control. A coherent standing of computing and its role in the decentralization plan will be expensive be- organization. cause the technical details and the interests Where users will acquire these knowledge of the different parties involved make the skills is unclear. A shortage of high-quality task complicated, and there are considera- data processing personnel already exists, ble costs for implementing even well-devel- and so hiring complete staffs of specialists oped plans. Careful control requires a com- for user departments will be prohibitively mitment to explore the options, work out expensive. Vendor-provided training might the compromises, and make the considera- be technically valuable, but is unlikely to ble up-front investment in planning nec- teach the more subtle skills of evaluating essary to execute a coherent change. There and judging the worth of systems needed to must be some incentive for top managers, make users discriminating in their assess- data processing professionals, and users to ments. seek the creation of a plan. Often the in- Additional considerations arise with net- centives are not present, or at least not worked, as opposed to stand-alone, decen- present in equal measure for all, so that tralization. There have been a number of gradual evolution toward decentralization advantages anticipated from combining de- appears to be the easier course. Unfortu- centralization strategies with telecommun- nately, this approach is often more costly ications and networking technologies. in the end, as changes must be made to Users will be able to "share" machines, "reintegrate" computing operations with thereby avoiding the loss of large available top management desires and to restore co- capacity from centralized arrangements. herence in computing uses and activities. They will be able to tap into network-wide Whether or not the investment that decen- databases, reducing data redundancy and tralization requires of the organization is integrating their work with others in the warranted depends on the benefits to be organization. They will also be able to use derived. their own and other machine or network

ComputingSurveys, Vol. 15, No. 4, December 1983 336 • John Leslie King resources and data interactively and more fore users) in the computing environment, efficiently. Networking requires interde- will be difficult to accomplish as long as the pendency of equipment, and will reduce the various equipment manufacturers are un- tendency of decentralized units to adopt decided about whether universal standards equipment that is incompatible with other should be adopted. TM These problems may units. Finally, networking will improve re- be solved in time, but as yet networks are lations among units, facilitate organiza- not available in the same sense that main- tion-wide controls and computing manage- frames or terminals are. Beyond these tech- ment, and reduce problems of maintaining nical problems there are also unresolved top management control common to decen- management concerns, such as who will be tralized, stand-alone systems. in control of the networks and who will be These benefits are technically feasible, allowed to connect to them. Networking is but they are even less likely to occur than still experimental, and organizations that are the benefits from stand-alone decen- adopt networking must accept the attend- tralization. They depend on unproven tech- ant risks of experimentation. nological capabilities and uncommon or- The problems of data sharing engendered ganizational behaviors. The ability to share by decentralization also persist in a net- computing resources among networked ma- worked environment. Information provides chines is limited at this time. The major users with power in proportion to the de- experimental networks (e.g., ARPANET sirability of the information to others, and and CSNET) do not allow actual linking of few organizational units are comfortable machine resources, but merely allow users giving other units or higher management to communicate with other network users open and easy access to their data [Petti- or to move to the machine of their choice grew 1972]. Centralized data processing to conduct their work. Computing power is forced departments to relinquish and cen- not shared among machines, but among tralize their data. But once users have their users and host organizations, using a heav- own systems, there will be no centralized ily subsidized recharge system that radi- authority to enforce data sharing through cally distorts the cost picture perceived by direct sanctions on computing use. Formal both hosts and users. rules governing access to data are weak There are more serious limitations with mechanisms of enforcement, since there are the emerging network technologies, espe- many ways for users to make the actual cially local-area networks designed to allow accessing of the data costly to those trying machine-to-machine communications at to get it. Users with their own computing high data rates [Paul 1982]. The goal of capability can be difficult to hold in com- designing such networks is to enhance com- pliance with organization-wide rules. Net- munications among users and allow them worked decentralization will not necessar- to use different resources available through ily alter this situation. the network. Such networks could, in the- Increased interaction among users might ory, provide a means for having both cen- not be facilitated by networked systems, tralized and decentralized computing si- either. It is dangerous to infer too much multaneously. Shared functions (e.g., large from experimental networks such as AR- processors, databases, special machines, or PANET and CSNET, because they involve peripherals) could be provided from one technically sophisticated users (academic, location but be available to all through the research, and professional) with the needs network. Local users with their own pro- 19 The question of standardization has a long tradition cessor and storage capabilities would also in the computing field. The most notable split in be able to use the shared resources through standards has been in the differences between IBM the net. The primary technical problems and IBM-compatible equipment and those of most arise from the lack of standards for com- other manufacturers. But there has been a lack of munications protocols, file structures, da- standardization in basic communications protocols even within given vendors' product lines. The whole tabases, and operating systems. Moreover, issue of communications and networking standards is the basic goal of networking, which is to tie in turmoil at this time, and how it will be resolved (or together different components (and there- even whether it will be) remains an open question.

Computing Surveys, Voi. 15, No. 4, December 1983 Centralized versus Decentralized Computing • 337 and skills to communicate with others on is seeking to reestablish control, or when a their highly subsidized nets. For most or- new managerial strategy is being imple- ganizations the integration of work through mented (e.g., a move from highly central networked computing will take a long time corporate control to divisional control). In to evolve and will entail substantial costs other cases, the economic issues can turn as users learn what the networks facilitate undercurrents of change into serious op- and what they do not. tions, or even force changes in order to take Compatibility of equipment will not be advantage of new opportunities. As we have ensured by networking unless highly cen- seen, the high costs of computer processors tralized control of procurement is main- once required many organizations to move tained. Networked arrangements are sub- departmental data processing activities to ject to the same pressures that create centralized units, but the declining costs of incompatibility in other organizational computer processors have recently enabled contexts. In extreme cases, decentralized a movement of such activities back to de- organizational computing centers adopt partments. Other economic factors, such as the strategy of "maximum feasible incom- the rising costs of software development patibility" in computing equipment and and maintenance, the costs of networking, operating systems procurement to make it and the problem of supply and demand in difficult and costly for other centers to ab- the data processing labor market have sorb them if the decision to decentralize equally important effects. computing is reversed. Neither the political nor the economic Nor will networking decentralized com- factors can be considered universally dom- puting establishments necessarily facilitate inant. However, the fact that many orga- managerial control, because possession of nizations have chosen and stayed with less computing capability is nine-tenths of the economical arrangements suggests that po- law in control of the technology. Decen- litical factors often are paramount. This tralized units will seek to build up their observation requires some qualification. capability to meet their own needs, possibly The question of what is economical might weakening the ties to the network and re- entail more than the obvious costs and ducing managerial control options. With- benefits: An expensive organizational strat- out serious top management restrictions at egy may be pursued because it is expected the unit level, and real control at some to yield long-run benefits. Nevertheless, central node under direct managerial con- the politics of organization and resources trol, there is only the budgetary process remains the fundamental factor in the and broad-brush, top-down policy to en- centralization debate. Economic issues are force management expectations on user be- frequently weapons in the discussion over havior. Such weak enforcements are often policy that serve political ends. This does more costly and less effective than the more not mean that the economic issues are un- direct control of access to computing re- important. Rather, it means that the larger sources that a central computer center pro- set of factors behind the call for either vides. course of action must be considered. We shall now address the basic management 3.3 The Interaction of Political options for centralization/decentralization and Economic Considerations policy with these factors in mind. The importance of political versus eco- 4. MANAGEMENT OPTIONS nomic considerations in the centralization IN CENTRALIZATION debate depends on the situation at hand. VERSUS DECENTRALIZATION When there is considerable organizational upheaval under way, political considera- The fundamental question, when one looks tions can overshadow economic factors en- carefully at the issue of whether to central- tirely. This can happen when departments ize or decentralize computing, is who will are competing with one another for re- have control over procurement, use, and sources to expand, when top management management? Traditional studies suggest

Computing Surveys, Vol. 15, No. 4, December 1983 338 • John Leslie King that centralization generally is less costly 4.2 The Centralization Continuum to the organization. As the assessment of economic dynamics above implies, this is Table 1 presents the major options man- likely to remain the case despite falling agers have in arranging for the three major entry costs for computing. But factors be- aspects of computing: control, location, and yond economics are involved, and the eco- function. Each is presented as a continuum nomics themselves are often complicated. between extreme centralization and decen- The challenge facing managers is to find tralization strategies. The intermediate ar- an arrangement for computing that meets rangements noted do not capture all possi- user needs as well as providing them with ble points between the extremes, but they an opportunity to experiment with and do illustrate how one might arrange a com- learn about the technology, without writing promise. Note also that even extreme a blank check for computing or creating decentralization does not preclude some problems for management control and or- organization-wide controls, facilities, or ganizational operations. We shall summa- functions. There still might be a need for rize the major management options for centralized arrangements to take care of dealing with this challenge, and suggest a organization-wide tasks such as computing strategy for finding the appropriate option personnel management. The key factor in given organizational conditions. extreme decentralization is that user departments are free to acquire their own 4.1 A No-Option Option computing capabilities and build their own computing operations. The intermediate Managers do not have the option of letting arrangement has certain aspects of control, the issue resolve itself in a convenient and location, and function reserved for the cen- appropriate manner. There are two reasons ter, while other aspects of these dimensions why this option is foreclosed. One is the are devolved to user departments. The pri- continuing presence of organizational fac- mary discriminator between centralized tors that keep the debate alive, regardless versus decentralized arrangements in this of the strategy followed. The question is case is corporate versus user department not whether the issue must be addressed, needs. but when and how often. The other is the As we have seen, each alternative has disappearance of automatic managerial advantages and drawbacks. Extreme cen- control over computing growth resulting tralization keeps computing activity and from the declining significance of computer growth under control of the center. It can procurement decisions. Unless managers provide economies of scale, and substantial want to issue categorical directives govern- overhead justification for procurement of ing procurement and use, which in itself expensive but specialized capabilities. It al- raises difficulties (e.g., Is a sophisticated lows management to control adherence to programmable calculator a computer?), organizational standards in system design users that want small computers will prob- and quality, and keeps track of computing ably find ways to acquire them. And even activity. On the other hand, extreme cen- categorical directives can be disobeyed, tralization can result in the creation of an subverted, or simply ignored.2° Managers insensitive bureaucracy that fails to meet now must face and deal with a much more user needs, and that is difficult for users to complex set of decisions than in the past. interact with. In some cases, centralized operations grow lazy and fail to exploit new ~o The author is familiar with one military organization that circumvented federal regulations centralizing opportunities that computing might offer. procurement of computers by buying microprocessor- Moreover, centralized service providers based diagnostic equipment not covered by the regu- must account for their expenses, which lations, removing the microprocessors from the equip- tend to be significant and organizationally ment, and throwing the rest away. Most administra- obvious. If computing is provided as an tive rules have loopholes. For a more detailed assessment of the effectiveness of policies governing com- overhead item, top management often will puter procurement see General Accounting Office wonder whether good use is being made of [1977]. this expensive function. If computing serv-

ComputingSurveys, Vol. 15, No. 4, December 1983 Centralized versus Decentralized Computing ° 339 ices are charged back to users, the users agement, extreme decentralization can often feel they pay too much for the services make it very difficult to keep computing they receive. This is exacerbated if there activities in line with organizational goals. are frictions between the central computing Once user facilities and computing opera- service and user departments. When back- tions are in place, the costs of change to logs for development of new systems grow conform to new organizational guidelines long, which they do when maintenance de- can be prohibitive. Giving away control is mands of existing systems grow large, users often easier than getting it back. begin to wonder whether their best inter- Intermediate arrangements appear to be ests are served by the centralized facilities. a promising compromise solution. It can be The lure of low entry costs for getting their very effective to retain centralized control, own computing capability provides a strong facilities, and functions in cases where this incentive for departments to lobby for de- is necessary or desirable, while allowing centralization, backed up by claims of poor limited decentralization where the payoffs service and high costs from the centralized are likely to be high. Thus top management facility. might reserve for the center those activities Extreme decentralization, on the other and functions upon which the organization hand, provides much greater flexibility for as a whole directly depends, while allowing users in exploiting computing technology users to take advantage of the technology and adapting it to their needs. It gives them within guidelines ranging from strict to re- the opportunity to learn about computing laxed, depending on circumstances. This in much greater detail, which can eventu- strategy will probably characterize most ally make them more effective consumers computer-using organizations in the future, of this technology. Properly handled, it can but intermediate arrangements have im- build up the computing talent of the overall portant drawbacks. The question of which organization by creating a "grass roots" compromise arrangement to pursue is dif- computing culture. It can enhance the will- ficult to answer, and in its own way embod- ingness of departments to accept changes ies the problems of the overall debate. It that might be beneficial to the organization. is sometimes impossible to differentiate It can also improve productivity when sen- between computing activities that are or- sible applications of the technology take ganization wide and those that are de- place at the departmental level that could partment specific because some serve both not or would not be provided by centralized purposes. The question remains whether service. The drawbacks of extreme decen- control should be left to the center or trans- tralization are that overall organizational ferred to the outlying units. costs of computing are likely to rise. This Intermediate strategies also require ex- is acceptable if productivity improvements tensive attention from both top manage- rise commensurately, but this will not hap- ment and departmental management. pen unless the departmental uses of com- Planning and executing an intermediate puting are well planned, well executed, and arrangement requires creation of protocols well maintained. New users have yet to to govern who is responsible for what and learn the lessons their more experienced in which cases. Central management must counterparts have learned, and the price find a way of relinquishing some control to for this learning can be dear. Moreover, departmental management while encour- departmental personnel who become com- aging conformance to the goals of the over- petent in computing soon discover the mar- all organization. Once arrangements are in ketability of their new talents, and either place, they must be maintained and en- demand wage increases or leave for better forced. The creation of a compromise does opportunities (sometimes within the larger not eradicate the root causes of centraliza- organization). If they stay, they may begin tion disputes; it merely finds a way of ac- to view themselves more in terms of their knowledging the different interests in- computing activities than their functional volved and providing them with at least responsibilities, causing personnel prob- some of what they want so they can get on lems. Perhaps most important for top man- with the organization's business. The same

Computing Surveys, Vol. 15, No. 4, December 1983 ..d~"~ ~ ~ ~ ~'~ "~ ~ ~ ~ ~..~ ~ ~d ~-J

Q. c.)

~ C~ ~D ~ c~ ~ ~ c~ ~,~ ~. ~ ~" ~

~, ~ ~ ~) c~ ,.~ ~ r~ C~ C ~.~~

._N

~;..~ ~ ~)o ~o ~ ~ ~ ~:~ ~ a~

c~ ~ ~ C~ ~i~ --~ ~> ~'1~ ~,-~ ~l~ c~ ~]~ 4~ o~

CL~ ~ _0 .D

c~ ,j::: ~ a3

q~ c..~ ~'~ ~ ~ '~l-~ ~ ~ ~ ~ ~~

0 ~ 342 • John Leslie King changes that affect the traditional central- the function of the service from the top ization debate will affect any compromise.21 down. Second, most arrangements of this kind (including Kommunedata's) soon be- 4.3 Mixed Strategies gin to feel the tugs of centrifugal force as The discussion thus far has focused on the members of the federation want to pull out three points along the centralization/de- and develop their own capabilities. If the centralization continuum. It is not neces- independent units do retain control, this is sary, however, to maintain the same degree an option they can pursue. If they cannot of centralization across the board. For ex- pull out, they have relinquished control. ample, some organizations maintain cen- This leads us back to the issue of the tralized control and facilities, while allow- relationship between location and control, ing users to acquire their own applications and the influence of location over function. programmers or other functional capabili- As we noted earlier, those who control cen- ties. These mixed strategies can work well, tralized facilities can direct the actions of but there are some restrictions. Theoreti- users of the facilities by simply witholding cally, one could choose any cell from each or restricting service to user departments column in Table 1 when determining com- that fail to comply with facility manage- puting arrangements. In practice, mixes ment. Centralized facilities will usually be can only be made by choosing cells across responsive to highly centralized, top man- or downward to the right of the table. Thus, agement controls, but they need not be for example, it would be possible to have responsive to a consortium of users. highly centralized control, somewhat de- Location (or facilities, in the case of com- centralized facilities, and either somewhat puting) influences function in the same way or widely decentralized functions. It is not that control influences facilities. Functions feasible to choose mixes by moving upward that are dependent on computing are tied to the right, because of the critical influence to the facilities they utilize, and must con- of control over location and function, and form to their demands. Moreover, when that of location over function. functional resources such as programmers This does not imply that a federation of are decentralized to user departments, split independent users cannot form a consor- loyalties can result. Analysts and program- tium and pool their resources to establish mers serve their departments, and may a centralized computing facility with cen- even depend on departmental evaluations tralized functions. This can and does hap- for their employment and advancement, pen, as when the federation of local govern- but they are part of the culture of comput- ments in Denmark created Kommunedata ing. They need access to the facility's com- to provide computing service to nearly puting resources, and they share common every local government in the country bonds of knowledge and career with its [Romer 1979]. But such examples do not staff. Yet it is possible to have centralized undermine our argument for several rea- facilities and decentralized functions as sons. In the first place, the practical effect long as the responsibilities of the decen- of the federation's action is to place some- tralized specialists to their operating de- body (i.e., the central facility's leadership) partments are clear. in control. A board of directors represent- When facilities are decentralized but ing the federation might provide policy di- functional personnel are not, computing rection and hire and fire the director of the personnel find themselves facing the prob- service, but they cannot practically control lem of employees in "matrix organizations": Their administrative home is in the central 21 Intermediate arrangements such as those suggested department or pool, but they work in the here have been the subject of many recent articles on decentralized departments of facilities. The organizing the use of computing and information tech- people who evaluate them for advancement nology in organizations. See in particular Alien within the pool do not have much contact [1982], Buchanan and Linowes [1980a, 1980b], Jen- kins and Santos [1982], Lucas [1982], McKenney and with the work they actually do. Eventually, McFarlan [1982], Rockart et al. [1979], Sprague and these functional specialists tend to move Carlson [1982], and Withington [1980]. their positions to the facilities and depart-

Computing Surveys, Vol. 15, No. 4, December 1983 Centralized versus Decentralized Computing * 343 ments that they serve. Thus decentralized resource used for many different kinds of location encourages decentralized func- applications at the department level, and tions. departmental managers have considerable The practical options for arranging con- autonomy in how they run their operations, trol, location, and function in computing some decentralization of control to these therefore tend to flow downward from con- managers might be appropriate. The goal trol arrangements. This limits the options is to ensure that the arrangements for con- suggested by Table 1, but is advantageous trolling computing are not wildly out of in that it reduces the number of alterna- keeping with other organizational prac- tives that management must consider. As tices. 22 Computing should not be thought long as control is seen as the crucial dimen- of as a tool by which the basic structures sion of the centralization/decentralization and behaviors of organizations can be debate, and the arrangements for location changed. and function reflect the behavioral realities Second, the issue of control must ulti- that arise from control decisions, sensible mately be decided by central management, arrangements are possible. It should also who retain responsibility for overall orga- be noted that there might be more than one nizational operation and performance and stable configuration for an organization cannot avoid being judged for its decisions. within this framework. This remains true whether there is a delib- erate or an unconscious policy regarding 4.4 Choosing an Option control of computing. Central managers should remember that decentralization of The final question to consider is which option is appropriate for a given organiza- 22 Committees are a popular means for incorporating tion. There is not sufficient space here to the ideas of outlying units into direction provided from describe all the cases in which various con- the center. This approach has recently been recom- figurations might be appropriate, but here mended as a means of dealing with the new era of are some guidelines for the decision procentralization/decentralization issues [Nolan 1982]. Research into the effectiveness of user committees cess. suggests that they are not very effective at solving First, control must be recognized as the major political problems in the management of com- most important issue in making centrali- puting [King and Kraemer 1984; Kraemer et al. 1981; zation/decentralization decisions. The pre- Kraemer and King 1981]. It seems that the basic problems of control remain. The growing literature on vailing norms of the organization can pro- the subject of citizen participation provides a useful vide guidance for dealing with control over analogy to the problems faced by incorporating users computing. If the organization is highly in political decision processes. For example, Arnstein's centralized in most of its operations, a [1969] "ladder of participation" suggests that there highly centralized control arrangement for are eight "rungs" representing levels of actual power conferred on participants: citizen control; delegated computing is possible and probably sensi- power; partnership; placation; consultation; inform- ble. Similarly, if the organization follows ing; therapy; and manipulation. Only the top three highly decentralized control policies, such confer genmne power on committee members, and as establishment of operating units as these of course require that actual power be relinquished from the center. The middle three rungs pro- profit centers, decentralization of control vide for some opportunity to assess the desires and might be necessary and desirable. Most or- frustrations of committee members, but action on ganizations have a range of control ar- these remains the perogative of the existing elite. The rangements, depending on what is being bottom two rungs can actually result in deterioration controlled. Decisions about control over of performance because the committee can serve as a shield behind which inadequately performing central computing should parallel those organiza- actors can hide while maintaining the appearance of tional arrangements governing the areas in sensitivity to users. Perhaps the most useful role that which it is applied. Thus, if computing is committees can serve is to help improve the sensitivity applied mainly to centralized financial ac- of both data processing specialists and users to one anothers' needs and problems, and facilitate what tivities, centralized control of computing politzcal scientists call "mobilization of bias" among should be appropriate. On the other hand, participants around genuine problems that can be if computing tends to be a general-purpose solved by collective action.

Computing Surveys, Vol. 15, No. 4, December 1983 344 • John Leslie King control can be profitable in some circum- tions about new possibilities. Emerging stances, but it can also be a source of many data communication technologies, by al- problems if not done in a manner that lowing networking, expand the options for ensures benefits for the organization. Re- deploying computer equipment. However, centralization of control can be difficult or this technology is still in its infancy. Only even impossible, and will usually be expen- the most technically advanced organiza- sive. 23 tions will succeed in the endeavor to install Third, managers should be cognizant of sophisticated network systems. If past the ramifications of computing location de- trends in diffusion of new computing decisions, which should be delayed until the velopments hold true, the majority of or- issue of control is settled. Decentralization ganizations will not be able to adopt this of computing location often has the prac- technology for at least five years. More tical effect of decentralizing control. Cen- important, networking does not alleviate tralized control and decentralized location the problems of control raised by decen- are possible and perhaps desirable, but the tralization of location, and in some cases it arrangements must be thoughtfully worked can create new problems, as we have seen. out, implemented, and enforced. At the very least, networking is an inte- Fourth, location decisions should be grating technology, and as such brings with based on a careful assessment of the actual it the difficulties associated with integrated uses the organization makes of computing. systems noted above. New technologies do If those uses depend on computing capabil- not provide a simple fix for the problem of ities that are only provided by large and centralized versus decentralized facilities. expensive processors or other costly re- Sixth, current arrangements should be sources, it might be infeasible to decentral- evaluated carefully before a change is im- ize. Many organizations have discovered plemented. Too often, problems with pres- too late that the small machines purchased ent arrangements stimulate demands for by departments are incapable of handling major change when these problems might certain major applications now being done be attenuated by minor changes. Some- (i.e., the large central systems must be re- times, increasing the resources of the com- tained), or that the portfolio of applications puting center can meet these user needs at desired by the departments will soon out- lower cost than establishing decentralized strip the small initial systems. Many facilities and functions. Conversely, if de- smaller computer systems have limited ex- centralized arrangements are causing trou- pansion potential. Failure to review present ble for integrating applications and meeting and future applications can result in naive organizational guidelines for compatibility expectations about which computing re- of equipment, stricter procurement proto- sources will be required. cols and centralized approval for procure- Fifth, location decisions should be based ments might be the answer. on current technologies, not on expecta- Seventh, it should be recognized that there is a drive toward decentralization of computing among users. This drive is likely 23 It is frequently suggested in prescriptive literature on the management of computing and data processing to grow stronger as entry costs for comput- that top management should be actively involved in ing decrease, for reasons that we have dis- decision making. Recent research suggests that inten- cussed at length. The development of ap- sive involvement of top management in such decisions propriate computing arrangements requires is associated with higher levels of computing problems, although why this is the case in not clear from the a careful assessment of the factors moti- data [King and Kraemer 1984; Kraemer and King vating the proposal to decentralize. Im- 1981]. More often, what data processing managers provements in effectiveness and better user need is not the involvement of top management, but departmental uses of computing are likely their support. A lack of top management input to to be cited, whereas a desire to gain new decision making still allows for appropriate decisions to be made in many cases, but a lack of top manage- resources, increased budgetary leverage, or ment support for the data processing managements' the entertainment value of computing will decisions can easdy cripple their implementation. probably be left out. All of these factors can

Computing Surveys,Vol. 15, No. 4, December1983 Centralized versus Decentralized Computing • 345 play a role in proposals to decentralize com- AXELROD, C. W. 1982. Dynamic planning and con- puting, and it is sometimes difficult to de- trol of the net value of data processing. In The termine what is really at issue. Economics of Information Processing, vol. 2, R. Goldberg and H. Lorin, Eds. Wiley, New York, 39-46. 5. CONCLUSION BENSON, D. H. 1983. A field study of end-user computing: Findings and issues. Manage. Inf. Syst. Q. The debate over centralized versus decen- 7, 4 (Dec.), 33-45. tralized computing has been around for a BERMAN, P. 1970a. A vote against centralized staff. long time, and will be around for a long Datamation 16, 5 (May), 289-290. time to come. Changes in the technology BERMAN,P. 1970b. Decentralized again. Datamation will not resolve the issue because the most 16, 13 (Oct.), 141-142. important factors in the debate are BERNARD, D. 1979. Managementissues in coopera- grounded in constant reassessment of tive computing.ACM Comput. Surv. 11, 1 (Mar.), 3-17. where control of organizational activities BLAU, P. M. 1970. A formal theoryofdifferentiation ought to reside. Changes in technology in organizations. Am. Sociol. Rev. 35, 201-218. merely alter the options available and the BOEHM, B. 1979. Software engineering: R & D economies surrounding them. Neverthe- trends and defense needs. In Research Directions less, decisions must be made at least for the m Software Technology, P. Wagner, Ed. MIT short run. Computing capability must be Press, Cambridge, Mass. provided, preferably in a manner that serv- BOEHM, B. W. 1981. Software Engineering Econom- ws. Prentice-Hall, New York. ices the subunits of the organization and BOEHM, B. W., AND STANDISH,T. A. 1982. Software the organization as a whole. There is no technology in the 1990s. Department of Infor- universal "best" solution, and so each or- mation and Computer Science, University of Cal- ganization must find its own. With proper ifornia, Irvine. attention to the endemic organizational is- BRANDON, D. H. 1970. Management Planning for sues surrounding the debate, the economics Data Processing. Brandon/Systems Press, of various arrangements, and the prevailing Princeton, N. J. norms and goals of the organization, it is BRESLIN, J., AND TASHENBERG, C. B. 1978. DIS- tributcd Processing Systems: End of the Main- possible to construct an arrangement that frame Era? AMACOM, New York. will serve until new technological or orga- BvccI, G., AND STREETER, D. N. 1979. A method- nizational developments force a reconsider- ology for the design of distributed information ation. systems. Commun. ACM 22, 4 (Apr.), 233-245. BUCHANAN, J. R., AND LINOWES, R. G. 1980a. ACKNOWLEDGMENTS Making distributed data processing work. Har- vard Bus. Rev 58, 5 (Sept.-Oct.), 143-161. This research was supported in part by grants from BUCHANAN,J. R., AND LINOWES,R. G. 1980b. Un- the Natmnal Science Foundatmn. The author ac- derstanding distributed data processing. Harvard knowledges helpful contributions from Sheila Bow- Bus. Rev. 58, 4 (July-Aug.), 143-154. man, Julian Feldman, Kenneth Kraemer, Kathleen BURLINGAME, J. F. 1961. Information technology King, Herb Schwartz, Nicholas Vitalari, and two and decentralization. Harvard Bus. Rev. 39, 6, anonymous reviewers. 121-126. BURNS, T., AND STALKER, M. 1961. The Manage- REFERENCES ment of Innovatmn. Tavistock, London. ALLEN,B. R. 1982. Computerstrategy: A philosophy Buss, M. D. 1981. Penny-wise approach to data for managing reformation processing resources. processing. Harvard Bus. Rev. 59, 4 (July-Aug.), In The Economws of Information Processing, 111-117. vol. 1, R. Goldberg and H. Lorin, Eds. Wiley, New CALE, E. G., GREMILLION,L. L., AND MCKENNEY,J. York, pp. 7-18. L. 1979. Price/performance patterns of U.S. ALLISON,G. T. 1971. Essence of Decision: Explaining computer systems. Commun. ACM 22, 4 (Apr.), the Cuban Missile Crtsis. Little, Brown, Boston. 225-233. ARNSTEIN, S. R. 1969. Eight rungs on the ladder of CHERVANY, N. C., DICKSON,G. W., AND NAUMANN, citizen participation. J. Am. Inst. Planners 35, 4 J. 1978. Distributed data processing for the (July), 216-232. State of Minnesota. College of Business Admin- ARROW, K. J. 1974. On the agenda of organization. istration, University of Minnesota, Minneapolis, In The Corporate Society, R. Marris, Ed. Mac- Nov. millan, London. CHILD, J. 1973. Strategies of control and organiza-

ComputingSurveys, Vol. 15, No. 4, December1983 346 • John Leslie King

tional behavior. Admin. Sci. Q. 18, 1 (Mar.), 1- GENERAL ACCOUNTING OFFICE 1977. Problems 17. found with Government Acquisition of and Use CHRYSLER, E. 1978. Some basic determinants of of Computers: 1965-December 1976. General Ac- computer programming productivity. Commun. counting Office, Washington, D.C., FGMSD-77, ACM 21, 6 (June), 472-483. Mar. 15. CYERT, R. M., AND MARCH, J. 1963. A Behavioral GIBSON, C, AND NOLAN, R. L. 1974. Managing the Theory of the Firm. Prentice-Hall, New York. four stages of EDP growth. Harvard Bus Rev. 52, DANZIGER, J. N. 1979. The skill bureaucracy and 1 (Jan.-Feb.), 76-84. intraorganizational control. Sociol Work Occup. GLASER, G. 1970. The centralization vs. decentrali- 6, 204-226. zation issue: Arguments, alternatives and guide- DANZIGER, J. N., DUTTON, W. H., KLING, K., AND lines. Data Base 2, 3 (Mar.), 1-7. KRAEMER, K. L. 1982. Computers and Politics: GOLUB, H. 1975. Organizing information system re- High Technology and American Local Govern- sources: Centralization vs. decentralization. In ments. Columbia University Press, New York. The Information Systems Handbook. F. W. DEARDEN, J. 1965. How to organize information sys- McFarlan and R. L. Nolan, Eds. Dow Jones- tems. Harvard Bus. Rev. 43, 2, 65-73. Irwin, Homewood, Ill. GROSCH, H. A. 1953. High speed arithmetic: The DEARDEN, J., MCFARLAN, F. W., AND ZANI, W. M. 1971. Managing Computer-Based Information digital computer as a research tool. J. Opt. Soc. Systems. Irwin, Homewood, Ill. Am. 43, 4 (Apr.), 11-16. DEMB, A. B. 1975. Centralized vs. decentralized GROSCH, H. A. 1975. Grosch's Law revisited. Com- computer systems: A new approach to organiza- puterworld 8, 16, 24. tional impacts. CISR Report 12, Center for Infor- GROSCH, H. A. 1977. On costs. Comput. Eur. 5, 26 mation Systems Research, Sloan School, Massa- (June 30), 16. chusetts Institute of Technology, Cambridge, GROSCH, H. A. 1979. On Grosch's Law. Commun. Mass. ACM 22, 6 (June), 376. DEROZE, B. C., AND NYMAN, T. H. 1978. The soft- HANNAN,J., AND FRIED, L. 1977. Should you decen- ware lifecycle: A management and technological tralize? Comput. Decis. 9, 2 (Feb.), 40-42. challenge in the Department of Defense. IEEE INFOTECH INTERNATIONAL1977a. Distributed Pro- Trans. Softw. Eng. SE-4, 4 (July), 309-313. cessing 1: Analysis and bibliography. Infotech D'OLlVERIA, C. R. 1977. An analysis of computer International, Maidenhead, England. decentralization. AD-A045 526, National Tech- INFOTECH INTERNATIONAL1977b. Distributed Pro- nical Information Service, Springfield, Va., Oct. cessing 2: Invited papers. Infotech International, DOLOTTA, T. A., BERNSTEIN, M. I., DICKSON, R. S., Maidenhead, England. FRANCE, N. A., ROSENBLATT, B. A., SMITH, D. JENKINS, J. M., AND SANTOS, R. S. 1982. Cen- M., AND STEEL, T. B. 1976. Data Processing in tralization vs. decentralization of data processing 1980-1985. Wiley, New York. functions. In The Economics o/Information Proc- DOWNS, A. 1967. Inswle Bureaucracy. Little, Brown, essing, vol. 2, R. Goldberg and H. Lorin, Eds. Boston. Wiley, New York, pp. 62-67. EDELMAN, F. 1981. Managers, computer systems, JOPLIN, B. 1967. Centralizing data processing facil- and productivity. Manage. Inf. Syst. Q. 3, 3 ities--A case study. Municip. Fmanc. 39, 4, 161- (Sept.), 1-20. 166. EDP ANALYZER1980a. Educating executives on new KING, J. L. 1978. Centralization vs. Decentrahzation technology. EDPAnai. 18, 11 (Nov.). of Computing: An Empirical Assessment in City EDP ANALYZER1980b. Risk assessment for distrib- Governments. Public Policy Research Organiza- uted systems. EDPAnal. 18, 4 (Apr.). tion, University of California, Irvine. EDP ANALYZER 1980C. Tools for building distrib- KING, J. L. 1980. Centralized and Decentralized Is- uted systems. EDPAnal. 18, 10 (Oct.). sues for Computing Management Public Policy EDPIDR 1979. The trend to decentralization, Part Research Organization, University of California, 2. EDP In-Depth Rep. 8, 5, 1-13. Irvine. EIN-DOR, P., AND SEGEV, E. 1978. Managing Man- KING, J. L. 1982a. Local government use of infor- agement Information Systems Lexington Books, mation technology: The next decade. Public Adm. Lexington, Mass. Rev. 42, 1, 25-36. FISHER, D. A. 1974. Automatic data processing costs KING, J. L. 1982b. Organizatzonal cost considera- in the defense department. Paper 1046, Institute tions in centralized vs. decentralized computing for Defense Analysis, Oct. operations. In The Economics of Information FOREST, R. B. 1977. Centralization vs. distributed Processing, vol. 2, R. Goldberg and H. Lorin, Eds. processing vs. anarchy. DP Dig 23, 1 (Jan.), 1-2. Wiley, New York, pp. 68-81. FREEMAN, P. 1983. Reusable software engineering: KING, J. L., AND KRAEMER, K. L. 1981. Cost of a Concepts and research directions. Dept. of Infor- social impact of information technologies. In mation and Computer Science, Univ. of Califor- Telecommunwaaons and Productivity, M. Moss, nia, Irvine. Ed. Addison-Wesley, Reading, Mass.

ComputingSurveys, Vol. 15, No. 4, December 1983 Centralized versus Decentralized Computing • 347

KING, J. L., AND KRAEMER,K. L. 1984. The Dynam- LUCAS, H. C. 1982. Alternative structures for the ics of Computing. Public Policy Research Orga- management of information processing. In The nization, University of California, Irvine. Economics of Information Processing, vol. 2, R. KING, J. L., AND SCHREMS, E. L. 1978. Cost-benefit Goldberg and H. Lorin, Eds. Wiley, New York, analysis for information systems development pp. 55-61. and operation. ACM Comput. Surv. 10, 1 (Mar.), LUCAS, H. C., JR. 1984. Organizational power and 19-34. the information services department. Commun. KLING, R. 1978. Automated welfare client-tracking ACM 27, 1 (Jan.), 58-65. and service integration: The political economy of MARKUS, M. L. 1981. The politics of implementa- computing. Commun. ACM 21, 6 (June), 484-493. tion: Top management support and user involve- KLING, R. 1980. Socml analyses of computing: The- ment. Syst. Objectives Solutions 1, 4, 203-216. oretical perspectives in recent empirical research. MCFARLAN, F. W. 1972. Problems in planning the ACM Comput. Surv. 12, 1 (Mar.), 61-110. information system. Harvard Bus. Rev. 49, 2 KLING, R., AND SCACCHI,W. 1979. Recurrent dilem- (Mar.-Apr.), 75-89. mas of computer use in complex organizations. MCKENNEY, J. E., AND MCFARLAN,F. W. 1982. The In Proceedings of the AFIPS 1979 Conference, information archipelago: Maps and bridges. Har- vol. 48. AFIPS Press, Reston, Va., pp. 107-115. vard Bus. Rev. 60, 5 (Sept.-Oct.), 109-119. KLING, R., AND SCACCHI,W. 1982. The web of com- MERTES, L. H. 1981. Doing your office over elec- puting: Computer technology at social organiza- tronically. Harvard Bus. Rev. 59, 2 (Mar.-Apr.), tion. In Advances in Computers, vol. 21, M. Yov- 127-135. its, Ed. Academic Press, New York. MEYER, M. W. 1972. Size and structure of organi- KRAEMER, K. L. 1982. Telecommunications/trans- zations: A causal analysis. Am. Sociol. Rev. 37, 4 portation substitution and energy conservation, (Aug.), 434-440. Part 1. Telecommun. Policy 6, 1, 39-59. MOLDOW, B. D. 1982. The economic impact of net- KRAEMER, K. L., AND DUT'FON, W. H. 1979. The working. In The Economics of Information Pro- interests served by technological reform: The case cessing, vol. 1, R. Goldberg and H. Lorin, Eds. of computing. Adm Soc II, 1, 80-106. Wiley, New York, pp. 44-51. KRAEMER, K. L., AND KING, J. L. 1976. Computers, MOORE, W. E. 1962. The Conduct of the Corporation. power and urban management. Sage Publ., Bev- Random House, New York. erly Hills, Calif. MOORE, W. E. 1967. Order and Change. Wiley, New KRAEMER, K. L., AND KING, J. L. 1979. A requiem York. for USAC. Pohcy Anal 5, 3, 313-349. MORRISEY, J. H., AND WU, L. S. Y. 1979. Software KRAEMER, K. L., AND KING, J. L. 1981. Computing policies and problems: A stage theory approach. engineering--An economic perspective. In Pro- Telecommun. Policy 5, 3, 198-215. ceedings of the 4th International Conference on Software Engineering, 79CH1479-5C. IEEE, New KRAEMER, K. L., AND KING J. L. 1982. Telecom- York, pp. 412-422. munications/transportationsubstitution and en- NISKANSEN, W. A. 1973. Bureaucracy: Servant or ergy conservation, Part 2. Telecommun. Policy 6, Master~ Institute of Economic Affairs, London. 2, 87-99. KRAEMER, K. L., DUTTON, W. H., AND NORTHRUP, NOLAN, R. L. 1973. Managing the computer re- A. 1981. The Management of Informatton Sys- source: A stage hypothesis. Commun. ACM. 16, 7 tems. Columbia University Press, New York. (July), 399-405. LAWRENCE, R., AND LORSCH, J. W. 1969. Or- NOLAN, R. L. 1977. Management Accounting and ganization and Environment" Managing Differ- Control of Data Processing. National Association entiatlon and Integration. Irwin, Homewood, Ill. of Accountants, New York. LEDBETTER L. E. 1980. The Software Life Cycle NOLAN, R. L. 1979. Managing the crises in data Model: Implicattons for Program Development processing. Harvard Bus. Rev. 59, 2 (Mar.-Apr.), Support Systems. Schlumberger-Doll Res., Ridge- 115-126. field, Conn. NOLAN, R. L. 1982. Managing information systems LEINTZ, B. P., AND SWANSON, E. B. 1974. Software by committee. Harvard Bus. Rev. 60, 4 (July- maintenance: A user-management tug of war. Aug.), 71-79. Data Manage. 4, 26-30. NOLAN, R. L., MCFARLAN, F. W., AND NORTON, D. LEINTZ, B. P., SWANSON,E. B., AND THOMPKINS, G. 1973. Information Systems Administration. E. 1978. Characteristics of application software Holt, New York. maintenance. Commun. ACM 21, 6 (June), 466- OLDEHOEFT, A. E., AND HALSTEAD, M. H. 1972. 471. Maximum computing power and cost factors in LITTREL, R. F. 1974. Economies of scale in the IBM the centralization problem. Commun. ACM 15, 2 360 and 370. Datamatzon 20, 3 (Mar.), 33-36. (Feb.), 94-96. LONG, L. E. 1982. Design and strategy for corporate OLSON, M. H. 1983. Remote office work: Changing information systems: MIS long-range planning. patterns in space and time. Commun. ACM 26, 3 Prentice-Hall, New York. (Mar.), 182-188.

ComputingSurveys, Vol. 15, No. 4, December 1983 348 • John Leslie King

ORLICKY, J. 1969. The Successful Computer System: Computer Science, University of California, Ir- Its Planning, Development, and Management in a vine. Business Enterprise. McGraw-Hill, New York. SCACCHI, W. S. 1984. Managing software projects: a PATRICK, R. L. 1976. Decentralizing hardware and social analysis. IEEE Trans. Sofiw. Eng. SE-IO, dispersing responsibility. Datamatmn 5, 79-84. 1 (Jan.), 49-59. PAUL, L. 1982. Micro-CPU link eluding DPers. Com- SCOTT-MORTON, M. S. 1975. Organizing the infor- puterworld (Nov. 8), 1. mation function for effectiveness as well as effi- PSRLMAN, J. A. 1965. Centralization vs. decentrali- ciency. CISR Report 39, Center for Information zation. Datamation 11, 9, 24-29. Systems Research, MIT Sloan School, Cam- PERROW, C. 1979. Complex Organizations: A Crittcal bridge, Mass. Essay, 2nd ed. Scott-Foresman, Glenview, Ill. SELWYN,L. L. 1970. Economies of scale in computer PERROW, C. 1982. Not risk but power. A review of use: Initial tests and implications for the com- Schwing, R. C., and Albers, W. A., Jr. Societal puter utility. Project MAC Technical Report TR- R~sk Assessment. How Safe Is Enough? Contemp. 68, Massachusetts Institute of Technology, Cam- Sociol. 11, 3, 298-299. bridge, Mass. PETTIGRSW, A. M. 1972. Information control as a SHARPS, W. F. 1969. The Economics of Computers. power resource. Sociology 6, 179-205. Columbia University Press, New York. PFSFFSR, J. 1982. Organizations and Organization SIMON, H. A., KOMETSKY, G., GUETZKOW,H., AND TYNDALL, G. 1954. Centrahzation vs. Decen- Theory. Pittman, Marshfield, Mass. tralizatwn m Organizmg the Controller's Depart- PHISTSR, M., JR. 1975. Shared systems vs. stand ment. The Controllership Foundation, New York. alone systems. Harvard University Program on SOLOMON, M. B., JR. 1966. Economies of scale and Information Technology and Public Policy working paper 75-1, Boston, Mass. the IBM System/360. Commun. ACM 9, 6 (June), 435-440. PHISTER, M., JR. 1979. Data Processing Technology and Economics. Digital Press, Bedford, Mass. SOLOMON, M. B. 1970a. Economies of scale and computer personnel. Datamatmn 16, 3 (Mar.), PRICE, D. G. 1965. The development of computer 107-110. centralization. In Economics of Automatic Data Processing. A. B. Frielink, Ed. Elsevier North- SOLOMON, M. B. 1970b. Economies of scale de- Holland, New York. fended. Datarnation 16, 6 (July), 293-294. PRICE, H. 1969. Centralized versus decentralized SPRAGUE,R. H., AND CARLSON,E. D. 1982. Building data processing in municipal government. Data Effective Decision Support Systems. Prentice- Process. 14, 389-402. Hall, New York. PUBLIC POLICY RESEARCH ORGANIZATION(PPRO) STATLAND,N. 1979. Computer systems: Centralized 1980. Orange County Faciliaes Management. or dispersed? Adm. Manage. 39, 3, 57. Public Policy Research Organization, University STREETER D. N. 1973. Centralization or dispersion of California, Irvine. of computing facilities. IBM Syst. J 12, 3, 183- REYNOLDS, C. H. 1977. Issues in centralization. Da- 301. tamation 23, 3, 91-100. TURKLE, S. 1979. Comments on the reasons behind ROARK, A. 1971. Centralization vs. decentralization. user procurement of personal computer systems, In Proceedings of the 2nd Conference of the Society made at the Irvine Conference on Social Issues for Management Information Systems. Society for and Impacts of Computing {Lake Arrowhead, Calif., Aug.). Management Information Systems, Chicago. UTTAL, B. 1982. What's detaining the office of the ROBSINS, C. 1975. Distributed data processing. Data future. Fortune (May 3), 176-196. Manage. 13, 6, 71-75. VITALARI, N. P., AND VENKATESH, A. 1984. Shifts ROBEY, D. 1981. Computer informationsystems and m the allocation of time: The case of computing in organization structure. Commun. ACM 24, 10 the home. Public Policy Research Organization, (Oct.), 679-687. University of California, Irvine. ROCKART J. F., BULLSN, C. V., AND KOGAN, J. N. WAGNER, F. V. 1976. Is decentralization inevitable? 1979. The management of distributed process- Datamatmn 22, 11, 86-97. ing. CISR Report 39, Center for Informatmn Sys- tems Research, MIT Sloan School, Cambridge, WAGNER, L. G. 1966. Computers, decentralizations, Mass. and corporate control. Calif Manage. ReD. 9, 2, ROMER, M. 1979. Comments on the costs of software 25-32. maintenance made at the 1979 annual conference WEBER, M. 1947. The Theory of Social and Economic of the Urban and Regional Information Systems Organization Oxford University Press, New Association (San Diego Calif., Aug.). York. SALERNO, L. M. 1981. Catching up with the com- WEBER, M. 1952. The essentials of bureaucratic or- puter revolution. Keeping informed. Harvard ganization: An ideal-type construction. In Reader Bus. ReD. 59, 6 (Nov.-Dec.), 8-24. in Bureaucracy, R. Merton, Ed. Free Press, Glen- SCACCHI, W. S. 1981. The Process of Innovatwn in coe Ill. Computing. Department of Information and WEISS, E. A. 1975. Centralization or decentraliza-

ComputingSurveys, Voi. 15, No. 4, December1983 Centralized versus Decentralized Computing * 349

tion: Which is best? Computerworld (Sept. 24), YIN, R. 1979. Production efficiency vs. bureaucratic 29. self-interest: Two innovative processes? Policy WlLDAVSKY,A. B. 1976. The Politics of the Budget- Sct 8, 381-399. ary Process Little, Brown, Boston. WITHINGTON, F. G. 1969. Data processing's envolv- ZANNETOS, Z. 1965. On the theory of divisional ing place in the organization. Datamation 15, 6 structures: Some aspects of centralization and (June), 58-68. decentralization of control and decisionmaking. WITmN6TON, F. G. 1973. Crystal bailing: Trends in Management Sc~. 12, 4, B49-B67. EDP management. Infosysterns 20, 1, 20-21. WlTHIN6TON, F. G. 1980. Coping with computer ZELKOWITZ, M. V. 1978. Perspectives on software proliferation. Harvard Bus. Rev. 58, 3 (May- engineering. ACM Comput. Surv. 10, 2 (June), June}, 152-164. 197-216.

Rece=ved October 1983; final revision accepted February 1984.

Computing Surveys,Vol. 15, No. 4, December 1983