4—
Market Uncertainty and Divisionalization: Alfred D. Chandler's Strategy and Structure

Introduction

A central theme in organizational sociology for the last couple of decades has been the growth and distribution of "decentralized" management. Broadly speaking this has divided into two branches: the "contracts" branch that says that decentralization is often achieved by autonomous firms tying themselves together for specific purposes, and the "decentralized administration" branch started by Alfred D. Chandler (1962) and Peter F. Drucker (1946). We will treat the contracts tradition more extensively in Chapter 6. In this chapter we treat Chandler's argument. Since the richest source of data on the development of the multi-divisional structure is still Chandler's own, this chapter is in effect a close reading of the theoretical aspects of Chandler's great historical monograph and of the relation of the theory to the data Chandler himself presents.

Our first task is to describe as exactly as we can what the dependent variable, the thing to be explained, is. It turns out that according to Chandler the central trouble with a centralized administration was not that it was centralized, but that it made centralization impossible where it was needed. Chandler's argument, especially in the treatment of Du Pont, is that in order to centralize the response in each of several markets that required markedly different tactics, one had to decentralize the firm as a whole. The main difficulties he points to that kept people at organizational reform derive from the firms' lack of centralized response to particular markets in which they were losing money. The reason they did not respond in a centralized way to those markets was that the firm had centralized at the firm level, which implied that the various functions (e.g., manufacturing, marketing, and engineering) that needed to be coordinated to respond to a particular market were each centralized above

the level at which response to a particular market had to be organized. A central manufacturing executive would coordinate all manufacturing together (explosives and dyes, for example), rather than explosives manufacturing with explosives marketing and dyes manufacturing with dyes marketing; similarly for a central marketing executive and a central engineering executive.

This part of the argument, and the historical analysis of Du Pont, is also the place where Chandler best shows the connection between the firm-level problem of inadequate response to the various markets the firm was in and the resulting solution of administrative reorganization. In particular he treats in considerable detail how individuals in the firm came to see that they had a problem of market response, how they came to see that reorganization of the administration was a solution, and how they stuck to administrative reform until they came up with the solution: centralized administration in each market, decentralized administration in the firm as a whole. Consequently, it is in this analysis that we can best see how functional problems in large firms come to be posed as problems in the work life of executives, the problem that has come to be called "methodological individualism."

Briefly, methodological individualism is the principle that explanations at the social level are not adequate unless the individual actions that are required for the social explanation to be true can themselves be explained adequately. This idea has been treated in the literature primarily as a question of the structure of theory. We will show in the first section below that Chandler solved the problem of methodological individualism for this part of his argument without having a theory of the general individual rationality kind that methodological individualists generally demand.

The argument for the functionality of the general office, detached from operating responsibilities and responsible for investment policy, is made most intensively in Chandler's chapter 3, in connection with Sloan's reorganization of General Motors. The basic idea is that only if the firm has better knowledge than the stock market does of the market projections and cost picture of each of its divisions can it do better than a completely decentralized, holding company structure. What is lacking in Chandler here, first, is a demonstration that such general offices do in fact do better than the stock market as a whole in predicting the future value of investments in their "subfirms." Are the profit rates of divisionalized firms with a general office in fact greater than those of holding companies with investments in various firms, each of which must respond separately to stockholder interests in profits?

Second, we need to be shown that this low predictability of the rates of return on investments in the subparts of firms without a general office was posed as a set of problems to the executives, who then set up the general office working with abstracted projections of demand, costs, and capital improvement proposals, and that those executives had the power to institute their proposals. In the second part of the chapter we will try to show that Chandler's analysis of General Motors almost satisfies those requirements. General Motors in fact expanded a great deal and remained the most profitable firm in the industry for many years after its reorganization into a multidivisional structure with a general office. The individual-level analysis required turns out to be, in large measure, a biography of Sloan himself, based largely on Sloan's own accounts. It is clear that Sloan had an adequate theory of his own behavior.

We will have shown in these first two analyses that the explanation of why a company will invent divisions is quite different from the explanation of why a company with divisions will invent a general office. The divisions are to facilitate different responses in different commodity markets; the general office is to provide information to satisfy bankers and other investors, to provide them with better guarantees of continued profitability than they would have from a holding company without such a general office. Thus Chandler himself has two different dependent variables here, which have two different explanations.

In the third section we treat Chandler's analysis of Sears. This is a particularly useful case to look at in connection with the problem posed for General Motors. Sears in fact competed with a structure of wholesalers and retailers who were organized as autonomous firms, with no centralized office and no centralized buying. We can pose the problem Sears had when it tried to use a mail order firm's buying structure to supply an appropriate inventory to a dispersed set of department stores in automobile-oriented shopping centers, then, as one of building a bureaucratic structure that could do what a market-organized structure of wholesalers and retailers was already doing very well. This is not a functional problem peculiar to Sears, because during the period Chandler analyzes here chain stores with multiple commodity lines were coming to prominence in the novelty trade (the "dime stores"), groceries ("supermarkets"), auto accessories, and eventually hardware and furniture, as well as in department store retailing.

Our analysis of Chandler's study of Sears, then, involves a systematic treatment of what functions the structure of wholesalers in various commodity lines nationwide in fact served and still serves, which parts of those functions were already served by the extremely decentralized struc-

ture of buyers in the nineteen or so commodity lines that characterized the Sears mail order regime before the reorganization Chandler discusses (described at that time as "a coalition of merchants"), and where in the new Sears structure the remaining, unfulfilled, functions were located.

Our argument here is that the decentralization into commodity lines, already present in the old regime, remained the central functional structure in the new regime as well. It was apparently the addition of detail to the inventory control system in commodity lines, rather than, as Chandler argues, decentralization into autonomous regional divisions, that made Sears (and other chain stores) able to do the jobs that specialized regional autonomous wholesalers were already doing.

We argue that Sears had to begin with an organization in many different commodity-line markets, that had already created "divisions"—the autonomous buyers for the mail order business—to respond to those markets. This divisionalized organization already had a general office, the one that decided to invest in department stores rather than expand the mail order business. In short, it started with the very structure to be explained, and had that structure for the reasons that the theory says it should have had such a structure. It entered very few new commodity lines. So what it needed was regional offices as branches of the general office to decide on investments in stores, and it also needed new modifications of its commodity-line information systems to feed store inventory and sales data into the offices of the commodity-line divisions run by buyers.

We argue, in short, Chandler in fact has little to explain—and mistakes what there is to explain—because he believes the regional divisions of Sears are "operating" divisions. He thus has not interpreted his theory of the first two cases correctly in analyzing Sears.

In the final part of the chapter we try to unpack the empirical content of Chandler's main independent variable, "being in several different markets." The overall functional argument in Chandler's book is that a multidivisional administrative structure is required if a firm is in several different markets. We provide a detailed definition of a distinct market as market uncertainty ramifying into the manufacturing and engineering of a commodity line in a way different for other commodity lines. In Chandler's chapter 7 he analyzes a large number of manufacturing industries in terms of his overall functional theory. If Chandler's book had been a sociological article, say in Administrative Science Quarterly , chapter 7 would be the article, and a page or two about the histories of Du Pont, General Motors, Jersey Standard, and Sears might have been in the

original draft as motivation, to be cut by the editors as not science but anecdote. But I believe it is more useful to treat the chapter as a careful analysis of what the phenomenon, as perceived by the corporation, of "being in several diverse markets" looks like as an "objective" variable spanning a number of industries.

We will treat the variable of being in several markets as a matter of the social organization of uncertainty, of the organization of information necessary to adapt to that uncertainty, and of the amount of uncertainty. This will constitute the explication of the implicit analysis in Chandler's chapter 7 of the conditions under which market uncertainty ramifies into the manufacturing and engineering of a commodity line, so forcing marketing, manufacturing, and engineering together into a division. We will find that the most successful parts of the case studies, as well as the cross-sectional analysis of industries in chapter 7, are neatly summarized by five variables describing how uncertainty of demand is organized in different sets of commodity lines.

The Concepts of Centralization and Decentralization

The first big conceptual trouble in understanding Chandler's argument at the structural level is that "centralization" and "decentralization" as Chandler uses them are really two different kinds of centralization: (1) centralization for administering one product (or a small group of related products) for one market, with market control (losses and profits) over the central authority; and (2) centralization for administering one product for one market in a division of the company , with both market and administrative controls over that division (the administrative controls are by a central office that also controls over divisions).

To see what is going on, let us start with the Du Pont firm making explosives or U.S. Steel making rails or Ford Motor Company making Model T's. In each firm there was a marketing or sales department—Chandler tells about the development of the marketing department in Du Pont, with explosives experts advising clients (for example, mines or tunneling projects) on efficient and safe blasting. In each firm there was some sort of engineering or chemistry department (or both). And of course there was a manufacturing department.

For any one product, the firm needed the engineering force to design the product to fit the tastes of the market and to design the production line so it could make the product cheaply, yet so production could be adjusted to variations in demand; the engineering had to be adapted to

both marketing and manufacturing. The firm needed to arrange for feedback from marketing to manufacturing so that manufacturing would not make too much and get committed to an enormous inventory, or would not have delays in the delivery that could kill the market, or so manufacturing could produce the right mix of sizes and colors. And obviously the firm needed engineering to design a line that manufacturing could in fact keep going, work on bottlenecks in the line, or redesign the product so as, say, to eliminate a weld by stamping the piece differently or otherwise contribute to cheap or flexible production.

A big problem of all industrial administration is to get each of these three functional departments to work together, to do what the others need instead of what they are inclined to do. Manufacturing can always save money by making long runs at a steady speed, with inventory taking up the fluctuations in demand. But if it turns out that a market fluctuation is not just a fluctuation, if clients are really stopping building railroads with the firm's steel or if folks are really poor in the depression and cannot buy a Model A, then the firm is stuck with the inventory. Then (as Ford did) the firm could force dealers to take the inventory and pay for it, and the dealers (the smart ones) would switch to General Motors so they would not go bankrupt—and Ford would have screwed up its marketing.

Marketing wants each order treated as a special case: stop production and make us this thing for this good customer. But when that happens, manufacturing gets disorganized, costs go out of sight, and the firm cannot sell (and it is "manufacturing's fault"). That is, there is a true dilemma between the requirements of cheap manufacturing and those of effective marketing, a dilemma someone has to solve in the interest of overall profits on that line of goods. Some things that sell cars better make manufacturing more expensive; some things that make manufacturing cheap really ruin the market.

Since engineering (e.g., design for ease of maintenance versus design for engineering efficiency) has to depend on what the market wants, on how much the dealers complain about the costs of warranty service, and so on, engineering has to be coordinated with marketing. Similarly, the tradeoff between "getting it out the door" and making it "state of the art" determines engineering delays, and those delays are reflected in manufacturing costs.

There were, then, in such centralized firms true conflicts among departments, because there were true technical and economic tradeoffs between what one department and another were supposed to maximize. And this meant that somebody (or some group—but in fact, Chandler says, a firm needs somebody) had to tell engineering to do what man-

ufacturing said instead of what they wanted to do, to tell manufacturing not to treat each customer as a special case but to control their production by an inventory depletion time criterion, and so on.

Note that the right decision about which department ought to give way on what questions depended on the product. Eventually Du Pont would sell enough blasting powder to use up a shipload of nitrate, but they might not sell enough of that awful artificial leather Chandler talks about to use up a shipload of wood pulp. The decisions Du Pont needed to make for a standard manufacturing ingredient like blasting powder and an ingredient of novelty goods like artificial leather were very different. There was a different combination of the risk of being stuck with an inventory the company could not sell (low with blasting powder, high with artificial leather), manufacturing cheapness (very important for blasting powder, not so important—as long as it is cheaper than leather—for artificial leather), marketing flexibility (not important for blasting powder, of the essence for artificial leather), and so on. That is, the contradictions between marketing, manufacturing, and engineering had to be resolved in different ways to make money on different products.

The centralized firm, the structure with which Du Pont started after Pierre reorganized it to look like a steel plant, puts one man with a staff of accountants, marketing specialists, production line managers, and so forth in charge of all three departments. He (it is always a he in Chandler's book, so this "he" is a historical fact) needed expert engineering managers, sales managers, and manufacturing executives under him, because it was hard enough to do engineering, marketing, and manufacturing in the first place. But he also needed to be able to order them to give up their own professional standards in the interest of one of the others. And he had to do it in the light of whether his firm was selling blasting powder or artificial leather, Model A's or Cadillacs, steel rails or "special shapes."

Centralized administration is represented by diagram A in Fig. 3; decentralized by diagram B. The decentralized firm looked exactly like the centralized one, except the thing in the decentralized firm that looks like the centralized firm is the division—and there was, in addition to the market controls over the division, also administrative control by the general office. The reasons for the similar organization for administering a single product were the same. For a given product, the engineering and the manufacturing boss had to be forced to resolve their different priorities in the light of how one really made money on that product , and the same held for marketing and the other two. Forcing departments to resolve their conflicts in a way appropriate to the product requires autono-

Fig. 3.
Comparison of the Organizational Form of Centralized and Decentralized Firms in Chandler

mous operating divisions. Before we use Chandler's materials to develop our argument, it will be useful to consider a methodological issue: the role of individuals in organization-level functional arguments.

A Definition of Methodological Individualism

By "methodological individualism" we mean that every individual action that enters into some collective pattern must itself have an adequate explanation. The collective pattern may be, for example, that corporations that enter multiple markets with markedly different contracting patterns,

spans of temporal adjustment, quality or service requirements, and so forth will tend to develop differentiated divisions to respond separately to those markets, and a higher-order integration mechanism (a "general office") to coordinate this divisional structure, because such a decentralized divisional structure is more efficient in responding to multiple markets. This is a functional connection at the level of the corporation: corporations in multiple markets "need" separate responses to the separate markets, and a decentralized divisional structure "fills that need." This is a way of stating the central argument of Chandler's Strategy and Structure .

The principle of methodological individualism, specified to this functional argument, says (1) that the actions of individuals in corporations that create the new structure should be adequately explained, in the sense that they should ultimately be shown to be a response to multiple markets; (2) that those actions should continue and aggregate in a way that ultimately creates the multidivisional structure; and (3) that such individual actions arise because the separate responsiveness to markets was "needed."

It has frequently been alleged that many functional explanations do not—and, in general, cannot—satisfy the requirements of methodological individualism (e.g., Homans and Schneider 1955; and Elster, e.g. 1983, 55–68). But obviously, if the explanations are true they must satisfy those requirements, since all social action is carried out by individual human beings.

Often the plea for methodological individualism is coupled with a preference for rational actor explanations: it is a concealed plea for utilitarianism. There is, however, no inherent reason why the model of individual action in a methodologically individualist account must be a rational one. For example, John Padgett has offered an account of how the Department of Housing and Urban Development (HUD) budget-making process "serves the function" of adjusting the departmental budget to the macroeconomic objective of the president's office, as transmitted through the Office of Management and Budget (OMB) (Padgett 1981, 85–92). The models of individuals in his treatment are specifically designed to take account of limitations on individual rationality, and have many random and nonoptimizing elements in them. The central point, though, is that the requirements set by the president's office enter as variables and constraints in the nonoptimal problem solving of the bureaucrats in HUD. Such models of a collective process satisfy the requirement of methodological individualism but are the opposite of any appeal to utilitarianism or to rational actor models.

Chandler's book is an especially interesting case in this regard, because he has followed some part of the strategy of deriving predictions

from the collective-level theory and documenting them. His chapter 7 (1962, 324–382) tries to show empirically that in industries that function in single markets the multidivisional structure has not been adopted, while in those that function in multiple markets it has been. But Chandler thought it necessary for his purposes to write some three hundred—odd pages of history, mainly of four firms (Du Pont, General Motors, Jersey Standard, and Sears), to see how and why the multidivisional structure was invented and adopted.

As is typical for historians, many proper names of individuals appear in the accounts. Clearly Chandler does not believe the functional argument at the collective level is adequate with only collective-level evidence, for he includes additional evidence involving a large number of named individuals as well. If these people were optimizing, they took a long time about it and lost a lot of money in the meantime, and still disagreed about what was optimal when they adopted the new structure. That is, the Chandler book looks like a case of a functional argument with individual-level explanations, which is not the same as a simple rational actor model at the individual level. We will next examine Chandler's explanation of the origin of the multidivisional structure in Du Pont, with some side glances to his own theoretical statements and to the more formal models in Padgett for guidance on the logical and theoretical role that statements about named individuals in organizations can have (Chandler 1962, 52–113 on Du Pont; and Padgett 1980, 1981).

Individuals in Du Pont: Organizing Information Flows

In Chandler's discussion of the old-regime centralized administration at Du Pont, he emphasizes the systematic effort to create reliable flows of information, especially by Pierre du Pont. Pierre du Pont thought of the legal and financial changes that incorporated previously independent businesses into E. I. du Pont de Nemours Powder Co. as "only necessary preliminaries" to reorganizing the administration of all of them (44; all page numbers cited in the following with no further specification are to Chandler 1962). In that reorganization a central point was "to establish a system of costs in order that an economical manufacture could be installed throughout the business." Cost accounting introduced first and foremost in manufacturing (55, 58). Once sales were taken from jobbers and agents and done by salaried technical sales people, cost statistics were also created in sales, as were market share statistics (59–60). The treasurer's department, headed by Pierre du Pont, a member of the top management, came to develop a system of cost, performance, and sales

statistics and projections (60, 66, 101–102). Since Pierre du Pont later became the power in the board of the company, the place of cost and performance statistics in defining the realities that confronted the company was secure.

At several points in Chandler's narrative, locating the difficulties facing the company depended exactly on the flow of differentiated cost, profit, and other performance measures (86, 92, 95, 100, 104). That is, a flow of differentiated statistics determined, in a general way, where people thought problems lay at various points throughout the administrative evolution. In one case, the fact that the statistics were inadequate to prevent problems of overstocking of supplies or of manufactured products (which were partly detected and diagnosed by statistics that came in later) was a crucial argument for reorganization (102).

Thus the environment in which administrative discussion went forward was shaped by information-collection technologies that could, for example, show that the problems after diversification were concentrated in lines where package goods were sold directly to the ultimate consumer, rather than in lines where tonnage goods were sold for further processing (92, 95, 100). In short, the problems people thought they had, and the fact that those problems persisted even after the company had reorganized to solve them (104), were features of an organizational intelligence system created largely by Pierre du Pont, and the results of that system were sufficiently persuasive that a negative decision on reorganization by the president, Pierre's brother Irénée, after Pierre had gone to General Motors, was overturned by the data (99–100 for the rejection in 1920; 110–113 for its acceptance in 1921), despite Irénée's continued skepticism.

The flow of detailed performance measurement information thus shaped the perception of the problem by individuals. In particular, it connected a systematic pattern of failure to individual perceptions in top management of that failure in a way that kept management looking for a solution even after rejecting the one they finally adopted (they obviously cannot have been optimizing in both 1920, when they rejected decentralization, and 1921, when they adopted it). The statistics to facilitate centralized administration thus ultimately helped cause decentralized administration. They also then became the basis of the general office's capacity to assess the work of the divisions they helped to create.

Individuals in Du Pont: Organizational Theory

In order for a problem, perceived through an organizational data collection system, to induce an organizational solution, it must also be ana-

lyzed by an organizational theory. The standard theory at the time of the innovations at Du Pont was one that ordinarily gave centralization and specialization as its answer (100 for Irénée's statement of these principles). But at least the theory's concrete manifestation at Du Pont was one that could diagnose a pattern in the data (e.g., the fact that other companies were making money in lines where Du Pont was losing money: 95) as requiring an organizational solution.

The most general guide to division of responsibility that could give rise to operational divisions specializing in a product line as stated by Harry Haskell (then vice president in charge of the high explosives department, but detached to head a subcommittee on organization): "The most efficient results are obtained at the least expense when we coordinate related effort and segregate unrelated effort" (69). He had already used that notion as an argument "to carry on the dye business as a separate entity. I think it would [be better] because it is a developing, unstandardized industry and should merit independent attention just as the Parlin chemical mixtures business was better by itself until standardized—when it was merged with the regular sales and operating departments" (68). The dye business needed "one individual in control of both production and sales, because the relation of the product and its qualities is so mixed up with the demands of the market for the product that to divorce them . . . would be detrimental to the business" (70).

This analysis meant that later evidence of interdependence of marketing, manufacturing, and raw materials purchasing (as shown, for example, in projections by manufacturing that were uncoordinated with marketing experience, resulting in inventory losses on raw materials or on manufactured goods in "merchandise goods" lines) was an argument for integration of those three functions. Further, it meant that the sales manager, Frederick W. Pickard, could explain the problems with the sale of package goods to the ultimate consumer by saying that he had products with "no logical sales connection with one another" (101). In analyzing these sales problems further, a subcommittee (which included Pickard) of the executive committee pointed out that (successful) competitors had one person in charge of both manufacturing and sales of each line and responsible for the profits of that line (96). Further interdependence provided a theoretical category that made the experience with "minidivisional" management of several new products (dyes, artificial silk) relevant to these problems of interdependence of sales and manufacturing on larger lines of products (96).

As is usual when evidence requiring a "paradigm shift" accumulates, the facts were at first explained away as a series of individual mistakes in a system of functional departments and centralized administration

that had functioned brilliantly in producing munitions (96–100) and was still making much money on tonnage goods (104). In particular, "[we] have carried excessive stock . . . [have made] several unfortunate guesses on the purchase of raw materials . . . have been working against orders rather than setting up appropriate stocks. . . . Rearrangements, repairs, renewals, and replacements of plants have been high. . . . [There have been] miscellaneous shortcomings in factory operations such as poor routing, and inefficient piece-work, pay schedules, short-runs, etc." (98).

But the theoretical category of interdependence, and the proposed general solution in the theory that problems arising out of interdependence should be put under a single authority, maintained cognitive focus on the difficulty with the reigning centralized organizational paradigm. The category existed in the organizational theory of the old regime, and this made it possible to codify the experience to which it was relevant, so both the category and the experience would be available to diagnose the new problems in multiple markets.

Individuals in Du Pont: Responsibility for Inventing and Adopting a Remedy

The process outlined above provided the environment in which individuals could see functional need, which Chandler identifies at the collective level, as a problem whose answer was probably organizational. But individuals become aware of thousands of problems over the course of a couple of years (say, 1918–1920 in the Du Pont company). These problems have to become the responsibility of someone, of some committee, or of some routine process. Committees were central in Du Pont's higher management.

By the time Chandler began his detailed story, the Du Pont company had an overlapping set of problem-allocation mechanisms, by status and by committee jurisdictions. The status distinction was that each department had two chief executives, the higher of whom (a "vice president") was responsible for long-range policy matters, the lower (a "director") in charge of day-to-day or operational matters (57). There were both standing committees and task force committees (usually created as subcommittees of a standing committee). The committees had a shifting membership whose most active members were usually the vice presidents responsible for long-term planning. The main work of writing proposals for reorganization, writing critiques of those proposals, and passing or

vetoing the final proposals was done by committees, though often one member of the committee was fairly clearly the author.

The mechanisms for creating committees, determining the charge to the committee (or the jurisdiction of standing committees), and specifying the membership are not clear in Chandler, probably because they were not clear to the people in Du Pont. There seems to have been more "constitutional" thought given to the composition and jurisdiction of standing committees with some decision-making authority, and much more ad hoc arrangements to give a charge and a range of authority to more specialized and temporary committees (58, 65, 93–96, 98–99, 103, 105, 113). Decision making was also apparently much influenced by rank; when Irénée du Pont, the company president, was being overruled on the issue of divisional organization, the executive committee was expanded to include the financial committee as well as Pierre du Pont and Raskob (who came back from General Motors to attend). In deference to the opinion of his colleagues, Irénée du Pont did finally vote for reorganization (110).

Two things are central to explaining how individuals got responsibility for devising solutions to the problems identified by the data flow and the existing organizational theory. The first is that the structure allowed for cutting the flow of "problems of subordinates" to the high executives, so that the latter could be detached from day-to-day problems to study, and restudy, problems of organization (303–304; see also Stinchcombe 1974, 24–25, 81–93). An executive directly in charge of a group of subordinates has to divert his attention to their problems when they need it, and so is not a deployable resource for analyzing long-run problems. The second factor is the existence of a flexible structure to tie together responsibility for solving problems and authority to implement the solution at the top levels of the organization. The top executives have to be not only free to study reorganization but also authorized to do so and permitted to put the resulting proposals on the agenda of a decision-making committee . The right to have one's proposals put on the agenda is the core of what we mean by "authorizing" a committee to study a problem, though of course the authorized committee also gets access to the intelligence apparatus of the top executives (58).

Although we do not get the details from Chandler of how this worked in Du Pont (the structure at the top seems to resemble that described by Cohen, March, and Olsen 1972), responsibility for the analysis clearly ended up being pinned on particular individuals (sometimes the individuals were drafting a committee consensus), and the proposals then usually went onto a decision-making committee agenda. (For a tempo-

rary hiatus in the second, see 99, where the first recommendation of a multidivisional structure was deleted from a subsubcommittee report by the subcommittee; this hiatus was only overcome in the face of a company financial crisis.) What we have, then, is not a description of the structure that authorized individuals to propose solutions to problems, but evidence that such a structure effectively existed in the top management at Du Pont—a sort of incomplete methodological individualism. It is incomplete because the collective-level solution is located in the behavior of specific individuals operating in a given structure, with their individual actions in fact linked in such a way as to create the collective-level pattern. We are not told how they came to be so linked, or why those individuals rather than others were authorized to act.

The submission by minorities to the decisions of the majority of the (expanded) executive committee is also not discussed in Chandler, though the lucky fact that one of the minority was the president, Irénée du Pont, gives us some evidence on the question. Aside from the uninformative phrase "in deference to the strong opinion of the majority of his colleagues," we are also told that he was appointed to the new executive committee, whose structure he had objected to. In addition, we are offered the speculation (in the final report to the board of directors of the corporation) that, after having experienced the operational running of market segment–oriented divisions by divisional councils (one member from sales, one from manufacturing, one from purchasing), managerial morale will improve by fixing responsibilities—which may be an explanation for minority acceptance. The failure to explain the submission of minorities is not unique to Chandler; the explanation of the acceptance of decisions made by authorized bodies is a weak part of social science generally.

Individuals in Du Pont and HUD: How Decentralization Works

What we mean by decentralization, presumably, is that the controls of a superior level of the organization over an inferior level are highly abstract. The performance measures that the general office gets about the inferior level in a decentralized organization carry few details that would enable them to penetrate the veil shielding the division manager from close inspection of his or her subordinates. If the general office knows too much about the performance of subordinates of the divisional manager, the knowledge that the general office is ultimately in command of

their careers will lead those subordinates to try to make a good impression on the higher management rather than on the divisional manager. The goals that the general office specifies for the division must thus be somewhat abstract, so as to leave discretion with the division manager. Padgett describes the supervision of OMB by presidential-level bureaucratic politics as follows:

Presidential decision making about fiscal policy and defense may be exceedingly complicated and convoluted. However, from the point of view of OMB, all that is produced from this within-level conflict is a single number—either of the form "your overall expenditure target is 500 billion" or of the form "cut 2 billion." The very top echelon of OMB may have some idea where this number came from, but whether they do or not is largely irrelevant to their task. Politics has been compressed into a single piece of information. (Padgett 1981, 80)

Similarly, Chandler points to the general office working mainly on "estimates and forecasts of anticipated conditions rather than on past or even current performance" (292). Clearly they cannot inspect many details of subordinate-unit behavior at any considerable distance into the future. They have to focus on one or a few numbers measuring performance (e.g., costs, market share, profits) and to supervise with a few numbers measuring constraints or goals (especially capital allocations and the description of which measures of performance will be used).

Thus the key feature of decentralization is the degree of abstractness of the information flow between the divisions and the general office. When that flow is, in Padgett's words, "compressed into a single piece of information," decentralization is maximal. It is the responsibility of the divisional manager to combine that one piece (or a few pieces) of information with the details about the world brought in by his or her subordinates and by the information-collection system of his or her division.

Padgett has shown that simple models of individual behavior in budget making at each hierarchical level are adequate to explain, on the one hand, the statistical dispersion of cuts across programs within HUD and, on the other, how the cuts come to add up to the totals specified by OMB, and therefore (more hypothetically) how OMB itself allocates cuts across cabinet departments so that they add up to presidential fiscal and macroeconomic objectives (Padgett 1980, 1981).

By analogy, we might construct a model of problem solving within di -

visions of Du Pont which could coordinate flows of information about purchasing, manufacturing, and sales of a particular product line from mines in Chile to the ultimate consumer. We could incorporate in this model investment and cost constraints imposed by the general office and specify the performance measures that office will use as objectives for the divisional manager to try to maximize. For our model to be workable, flows of information about what the problem is and flows of control communication about what the solution has to look like must be correctly segregated by degree of abstractness, so that each level has the right information to solve its level of problems. For example, in the abstraction process that turns divisional information into a flow to the general office, much information relevant to measuring subordinate performance within the division must be deleted, but information relevant to making market size, price, and cost projections must be retained.

The rather ordinary men at the general office deal with large time spans easily because information about daily and weekly time spans has been deleted from the information flow to them and because they allocate capital authorizations over large time spans rather than approving purchase orders. The rather ordinary division managers likewise have a manageable number of subordinates, given how often they have to confer with each or to force them to confer with each other, and the directions they get from the general office are simple enough that these can be integrated with the flow of communications with subordinates. A key indication of how well decentralization is working is the proportion of all communication of top executives that occurs in committee meetings (Stinchcombe 1974, 79–84); if they are talking to individual subordinates rather than to committees (often made up of the subordinates of the committee chair), they are probably not making long run policy with sufficient attention to data abstracted to reflect long-run trends.

To explain how abstraction and corresponding decentralization add up to the effective functioning in the U.S. federal government, Padgett uses the image of differentiated ecological control systems.

Each level of organizational aggregation [of lower-level budget data] is embedded in a distinctive cultural context of "ecological control" premises which reflect, in highly compressed form, the historical residues of past political struggles—[at the lower level], program controllability [i.e., the degree to which future budgets are subject to executive discretion rather than fixed], whose roots lie in the legal structure; organizational priorities [of particular government departments], whose roots lie

in institutional roles and constituency relations; and presidential fiscal targets, whose roots lie in macroeconomic and defense issues. Such ecological control premises change on different historical time frames. (1981, 121–122)

The point here is that through segregated levels of abstraction and the time frame of responsiveness, different types of problems become problems to different hierarchical levels of the government. The same applies in Chandler to Du Pont and General Motors. The solutions to the problems at each level go to the other levels in highly abstract form. For example, the presidential level sends out a target cut and gets back a proposal to cut certain programs by certain amounts; it need never know how controllability and constituency relations were balanced in arriving at these cuts.

Similarly, Du Pont's new general office need not know how much of the capital and operating expenses of a division went into making manufacturing respond more quickly to market changes, how much into producing the product more cheaply; they need only know whether whatever the division did worked to preserve market share. But they need to know this over the course of years, not weeks, as one does if one must keep the inventory of unsold goods of many types within bounds. The statistical techniques for abstracting performance records to the appropriate level to detect problems for the general office that require for their remedy yearly shifts of investment are different from those for detecting problems with too high production of one type of paint, too low of another. The hierarchical segregation of information is essential if individuals are to be able to respond to shifts in the environment by manipulating resources under their control.

This sketch of Chandler's theory of what decentralized administration does (considerably sharpened by a reading of Padgett) shows that his main functional connection can also be disaggregated into individual behavior. Through decentralization from the central office and recentralization at the divisional level , authority to respond on different time scales is coordinated with information adequate to detect problems on those time scales, so each person's job is doable and yet the organization responds correctly enough at all time scales. The sketch also shows that individual rationality depends on appropriate decentralization. Only if a general office executive gets information appropriately abstracted for making projections, but not for promoting a division chief's subordinates, can he or she respond strategically in making investment decisions.

Is It Still Sociology?

The historical chapters of Chandler's book are convincing partly because they disaggregate the functional connection to the individual level. But they are also convincing because they show that even if we do not always understand exactly why the individuals acted the way they did, they in fact acted in such a way as to create a multidivisional structure. So far his discussion could be either psychology or history, but nothing at the collective level, the level of the corporation, shows that the connections are social. Chandler's chapter 7 furnishes the evidence, in a form familiar to sociologists, that the histories are not merely unique paths leading from wherever it was to wherever the company happened to get to, as illuminated by what the individuals said they were trying to do.

Chapter 7 shows that there is a strong correlation between being in a single market and having a centralized structure and between being in multiple differentiated markets and having centralized divisions—but divisions decentralized from the central office. A sociologist would not ordinarily be satisfied with the four administrative histories in the first part of the book because there are more variables and contingencies than cases. Chapter 7 convinces that, almost invariably, firms that succeed after having entered many markets have a multidivisional structure, while those that stay in a single market have a centralized structure. This, then, is a functional connection that is decomposable into individual actions, yet it is clearly not the outcome of any simple aggregation of individual rationalities. If decentralization is an aggregation of individual rationalities to all, the majority of top Du Pont management who were against it for a year when they were losing a lot of money by not having it shows that it is a more complex rationality than we generally meet in microeconomics.

If a centralized firm, without changing structure, has one central executive coordinating engineering, manufacturing, and marketing of both explosive powder and artificial leather, the executive's job will be within his or her capacities only if he or she does the job by rule. If the executive is a powder person, he or she will say to engineering: make it cheap; to manufacturing: build up inventory, there will be rock to break up forever; to marketing: teach the customer how to do it right—get in there and help them run their blasting operation. But those rules will make the firm go broke in artificial leather for novelty goods.

The solution is to invent subordinate rule makers when the firm goes into many businesses: the bosses of divisions. The central management have to teach themselves not to meddle because they do not know artificial

leather, only powder. Creating these subordinate dispute-resolving and rule-making authorities—the division heads—and getting the general office head of engineering, manufacturing, or marketing not to interfere are the first big achievements that Chandler talks about in Du Pont.

The Causes of Divisionalization

We have used our analysis of the details of Chandler's history of Du Pont to explicate his theory of the causes of divisionalization. The dependent variable in this story, then, is the centralization of the management of product lines in distinct markets . The first part of Chandler's explanatory problem is to explain why centralization at the divisional level, rather than at the firm level, is required for manufacturing in lines with distinguishable markets.

We have not explained why Du Pont wanted to maintain itself as a single firm, nor does Chandler (but see Chandler's later work [1977] for an explanation). Consequently we, and Chandler, have left aside the question of why there needed to be a general office at Du Pont. We have said only that if they had one, it needed to have abstract nonoperating information fed to it by the divisions.

Since Du Pont was already centralized, and since this fact did not change, we get no information from Chandler's history as to what brought about a general office, or what brought about firm unity when unity was not demanded by operational unity. To explain the multidivisional form, we need to explain the causes both of divisionalization and of a general office. But the second set of causes is better explored by looking at the creation of a general office than by looking at the creation of divisions. In Du Pont that process can be explained by what we now require the general office executives to do, now that we can no longer let them run operations.

To guess what general office executives would do after decentralization, one has to realize that they were the ones who controlled investments and had to answer for their profitability. But investment is a future-oriented activity, which means that an investment center like the general office has to form a picture of the future; this, then, will tell them returns from an investment made now in powder or in artificial leather. The general true answer they ought to be giving, of course, is, "Damned if I know what the future will be like!" But they cannot run a business on that truth, because they have to put their money somewhere or they will get no profit. So what the general office does is make projections—of market demand, of Japanese competition, of costs, of competitor effi-

ciency. Then they calculate how these projections would influence profitability if, say, more money were to be put into artificial leather than into blasting powder, and vice versa. Then they decide where to expand, and how much.

The advertisements for Lotus 1-2-3 or Visicalc give an idea of what general office executives are doing. What the firm usually gets is an "exact" calculation (Lotus 1-2-3 does not give standard errors) of expected profit with a margin of error a mile wide; then it takes a decision. On average, the decision taken is the wrong one. Apparently the general office does not make its investment decisions any better than the stock market as a whole does, for the profits of multidivisional firms are not ordinarily any bigger than the average return in the stock market. Many firms with general offices that are very expert in highly abstract projections of future demands and costs go bankrupt. Chandler would argue that if the central management were not engaged in making such projections, they would be meddling in the business of the divisions and so would do worse than the stock market as a whole. But constructing an argument about what a general office is for, when, as in the case of Du Pont, it might well be a fossil left from the days when the company was a single centralized firm, is not a satisfactory way to proceed. Let us turn, then, to General Motors and its creation of a general office.

General Motors Creates a Multidivisional Structure by Centralizing

In Chandler's account of what happened to General Motors, the basic structure of William C. Durant's carriage business was just what Sabel (1982) says is the alternative mode to mass production, to "fordism," in organizing industrialism—what in Chapter 2 we called the "interactive" program structure. But Chandler does not describe how the skills of the workers in, say, the wheel maker's firm were taught and organized.

By subcontracting for the parts, Durant himself (or rather, Durant's company itself—Durant was off in New York talking to bankers) managed assembly, and especially marketing, directly. That way the carriage firm that preceded General Motors could respond rapidly to shifts in fashion. In particular, the whole structure could more or less turn around and make cars instead of buggies, using more or less the same marketing structure and the same set of subcontractors, but with different technical details, different volumes, and so on. The flexible structure in Flint, Michigan, was similar to that described by Sabel for flexible manufacturing in many classical industrial cities and was sufficiently adaptable to

switch from routines for producing carriages to routines for producing automobiles.

If all these independent craft firms were doing fine, why did Durant buy out his suppliers to form General Motors? First, he didn't always. But when he did, the reason seems to have been risk management and capital flow problems; these led to vertical integration—to assemblers buying out suppliers or marketing firms. Small business owners could not afford to build to ten times their previous size, going deeply in debt, unless they knew they would have contracts. But the only game in Flint for these big contracts was (and still is) General Motors. Durant himself moved on the New York capital market, not in Flint. To borrow large amounts to finance a wheel plant's expansion, Durant could commit General Motors to buying from "GM's own" wheel plant (or from the Fisher Body plant after GM bought out Fisher Body). If Durant acquired a firm, he could afford to take the risk of making it ten times as big—which the owner could not—because he could guarantee the market for ten times as many wheels. Essentially, General Motors played "insurer" as its larger assets and secure market position guaranteed the loan or stock issue; in addition, it made the returns of the small business (say, the wheel maker or Fisher Body) insurable by providing a "guaranteed" market.

The firm becoming a subsidiary sold its human and organizational capital to General Motors, had that spread over a larger capital stock, and was given a secure access to the market (as secure as General Motors', that is). There was then no reason for Durant to intervene in the administration of the subsidiary unless it lost money, had too little capacity, or priced itself out of the market. The "private" wealth of the subsidiary's old owners was freed up (by having General Motors' purchase price in the bank), instead of further embedded in their parts firm. Especially, they did not have their wealth "leveraged" with a large debt that made the leverage very risky for them (which would have happened had they expanded with capital borrowed in their own name rather than letting Durant do the expansion) just because General Motors needed more wheels than they had capital to produce. General Motors needed the new, much larger capacity, so General Motors should take the risk, and not some little wheelwright who would have had to mortgage his house (or his race horses) to expand.

So what made this extreme decentralization work? In fact it is the thing that Sabel says makes fordism work, namely, a mass market for identical goods, organized into a bunch of reliable dealerships providing parts and service as well as selling cars and spread out around the coun-

try. This was what made Durant able to organize General Motors; he had made the carriage business work the same way. But what produced the money to invest in expanding the parts suppliers by, say, ten times was mass marketing. Durant's own innovations were not in fordism, but in mass marketing supported by craft skills; and that mass marketing provided backing for capital flows that eventually allowed parts of General Motors to fordize their production.

The basic generalization in Sabel (1962) is that mass marketing encourages the development of technical routines (see above, Chapter 2). But here we see that mass marketing first produced innovative technical firms tied together by subcontracts, with skilled, flexible workers producing whatever parts are needed this year. In short, it caused expanded craft production.

Perhaps the key is that at that time producing automobiles was technically new. A mass market in a technically new system may produce a flourishing "small-firm skilled worker" system. Only when the firm has mastered the technical problems at the operative level can the new knowledge of how to make, say, a million wheels a year filter up so that engineers can routinize the knowledge of how to make a million wheels. It is important to realize that a flexible system held together by flexible contracts and occasional vertical integration to redistribute the risks could compete successfully in the same market that the Ford Motor Company itself was producing in. Subcontracting can compete with fordism on fordism's own turf.

The Centralization Revolution at General Motors

To understand what the Sloan revolution at General Motors was, we have to note more specifically what functions it was that were created as new functions of the central administration (now the "general office"), what functions were not decentralized to the division, and what functions for the divisions and for the investors the general office had.

Coordinating Projected Demand

The crises that created the Sloan revolution at General Motors were centered in the marketing system. The basic problem seems to have been that the parts suppliers, being far back in the system from the dealers who have direct information from the clients, could not tell whether a slackening of demand was temporary or long term. (It is indicative that Sloan himself had experience in the parts production part of General Motors.) To keep their own production efficient, they needed to keep

their trained technical workers on, to use their capital to the fullest, and generally to work up to capacity. So they had motives not to mistake a temporary downturn for a shift in demand. The main reason General Motors needed a general office, then, was to organize, and then universalize, an authoritative system for projecting demand, with projections revised often enough to control production and inventory policy far back into the parts and materials supply system.

Centralized Differentiation of General Motors' Markets

The second big thing Durant had developed was a class-differentiated marketing system, with the makes he had bought up into slots in a market organized from cheap to expensive. The cheapest line was Ford's, and it was a long time before Durant and Sloan could figure out how to produce a car to beat Ford in the low end of the market. Their short-run adaptation was to continue to produce and market the distinctive cars of the companies they had bought. Clearly, though, one should not encourage a successful Olds division, say, to expand to competitively damage the Buick division, when maybe no one else in the market was in a position to damage Buick.

As General Motors centralized, then, they did not want to go in a completely fordist direction, producing only one car. Diversified mass marketing was already a corporate strategy, an implicit result of acquisitions and letting the firms acquired continue to produce what they had been producing. In any case, decentralization as a way to deal with the administrative problems of producing several lines of cars was already institutionalized. What was new with Sloan and the general office was a deliberate plan to cover the spectrum of markets, to fill in (invest in) areas of the class system they did not cover, and to run down (disinvest or convert investments) in situations where subcorporations competed with each other for a class segment of the market. So the firm needed parallel feedback—parallel demand projections—to control production in the several lines differentiated by class .

Coordinating the Service Market with the Car Market

The third general problem confronting Sloan, apparently solved in the operating divisions rather than by the general office, is that marketing cars ultimately depends on a reliable supply of repair parts. Producers of parts for, say, the Buick division had to cooperate as well with Buick dealers all over the country in supplying the repair parts needed to keep Buicks running. Having lots of Buicks in local junkyards for lack of parts ultimately depresses sales. The corporation thus had to have the market-

ing information collected from Buick dealers available for planning repair parts production.

Still, General Motors did not want Buick parts suppliers using monopoly power (e.g., being the only producer of Buick air filters) to jack up the price of repair parts. So while GM needed to supply information to the Buick air filter manufacturer, it needed also to keep that producer from exploiting his position as the only one who could supply the dealers. Sometimes such a problem was resolved by buying the parts supplier and putting it administratively into the Buick division, but such decisions would have been strategic ones made by the general office on recommendation of the Buick division.

Centralized Functions Not Delegated to the Divisions under Sloan

It is important to note that some things were not left to divisional decision. The biggest example was purchasing. There are all sorts of reasons why people generally do not want purchasing to be under divisional (or any decentralized) control: corruption, price competition between divisions, integration of orders to get longer production runs for suppliers, relevance of purchasing information to the decision of whether or not to acquire a supplier, and so on. In any event, a crucial part of the administration of divisional supplier relations was not under divisional control. Similarly, standards of cost and profit accounting, calculation of investment returns, and the like were not decentralized. That is, mechanisms central to the integrity of general office control of the divisions (for the aspects that they mainly wanted to control) were not left in the division's hands.

The General Function of a General Office

The main point of the specificity of our description of what was centralized is that when a firm already has a decentralized administration, imposing a new centralized structure on top of it threatens to centralize decisions that had better be left decentralized. Decentralization was the situation General Motors was left in by the acquisition strategy of Durant. When a firm like Du Pont has a centralized structure and then goes into various businesses that require different sorts of coordination among, say, production, engineering, and marketing, the firm is likely to end up running one of the businesses by a strategy inappropriate to that business, though it may work very well for another one—thereby losing money. Such a firm thus has to learn not to make so many decisions centrally. Chandler's point is that both firms—Du Pont and General Motors—ended up with pretty much the same structure, with coordina-

tion of the different specialties involved in a manufacturing operation going on at the divisional level and projections (and investment strategy based on those projections) being done at the general office level.

The functions outlined above for the general office are especially those in which information from one division, parts subsidiary, or marketing branch was useful for assessing the investment value of another division, parts subsidiary, or marketing branch. In such circumstances the general office has an interest in securing the following features in that information:

1. Financial integrity. The chains of information that link expenditures in a business to capital markets are more effective if they guarantee financial integrity. Centralized monitoring of purchasing, firmwide accounting standards, and coordinated projections all tend to make the standards for investment in each part of a decentralized firm as good as the standards for the whole firm. One function of the general office, needed to justify to an investment bank a loan of money to expand a wheel maker tenfold, is to guarantee the financial integrity of the information both on the wheel maker and on the buying division.

2. Guarantees of markets. If General Motors gives a long-term contract to its wheel-making subsidiary, this guarantees (within limits) a market for the wheels, which in turn guarantees (within limits) the profitability of the investment. Similarly, arranging that Olds shall not invade Buick markets, and vice versa, does not make either division secure from competition; but it does reduce the risks that completely autonomous divisions would autonomously invade each others' markets. A general office can make competitive destruction of the value of an investment less likely.

3. Reliable external economies. Reliable repair parts supplies and available retail installment credit are profitable businesses in their own right, but in addition, their reliability and availability are selling points for car marketing. Yet a completely autonomous firm supplying repair parts or installment credit might not find it profitable to supply exactly what a dealer needs. Being able to arrange reliable supplies of such externally generated selling points stabilizes the value of marketing investments, and hence of the manufacturing investments behind them. In turn, of course, stable marketing arrangements provide reliable opportunities to repair

parts suppliers and installment credit suppliers. Coordinated provision of complementary goods and services makes a more profitable portfolio of investments.

Thus, the functions we see performed by the newly organized general office in General Motors provide information crucial to capital markets (cf. Zetka 1988). They also developed out of functions performed by Durant as an individual, before Sloan created a committee version of Durant. In this sense, Chandler's analysis of the causes of the general office reduces to one cause, of supplying information of use to investors, rather than of supplying coordination of response in commodity-line markets. There is no general reason to believe that general offices will be required wherever divisions are required.

The Theoretical Problem of Sears

This disjunction of the causes of divisionalization and the causes of general offices creates a problem for Chandler's analysis of Sears, for we cannot, I believe, understand the administrative reform of Sears without a careful analysis of Sears's competition. That competition was a structure of small and medium wholesalers in distinct commodity lines, with no national commodity-line buyers and certainly no national general office. Sears had to compete with this extremely decentralized system, and could only do so if it could coordinate decentralized demands better than that competition, through national "divisions" run by commodity-line buyers. But I will argue that the central administrative innovation discussed by Chandler—Sears's regional divisions—could not do that work.

The problem Sears had was not to create commodity-line divisions, as at Du Pont, because it already had them. Nor was it to create a general office to govern investment, because it already had that too. Instead, it was to build new structures of flows of information from retail stores to autonomous buying divisions, inherited from the mail order business. The regional divisions, I will argue, were not part of that flow.

The central argument, then, will be that neither the causes of divisionalization by commodity lines, nor the flows of information justifying investments, changed. Hence the causes in Chandler's theory as developed above cannot explain what happened at Sears.

Regional Information in Merchant Wholesaling and Sears

Let us take a problem of marketing administration from Chandler's analysis of Sears. We will try to illustrate again how Chandler argues, and in

particular how a functional inadequacy of a structure creates a flow of problems for individuals (or for small groups, more often) and how structural change in a functional direction is caused by the solution of individual-level problems. When we inspect Chandler's argument about Sears from this point of view, several inadequacies appear. We will look at the problem of adjusting the inventory displayed and on hand in a Sears store to the seasons and to the regions of the country. Obviously adjustments all the way back to the buyer (the person who arranges the contract between Sears and the factories that produce the goods) are needed if the overall "functional" problem of having summer goods gone and back-to-school goods on display by August, and of having different back-to-school clothes in Phoenix than in Seattle, is to be solved.

"The market" solves this problem by thousands of wholesalers stocking what local retailers are going to want to buy and sending travelers (traveling salesmen) around to the stores in their area with catalogs and samples, which results then in a large flow of small orders from wholesalers to manufacturers. If Sears was to have an advantage in this trade, it had to use the experience of, and the opportunity for, volume buying that it had developed in the mail order business.

But buyers for a mail order house need not time deliveries and displays for each store by the season and the region. They need to know how many raincoats and how many cowboy boots children whose parents order from Sears will want this year and to place both large orders rationally, thus allowing manufacturers to run their plants year-round and to process only one order for raincoats or one for cowboy boots. Sears did not need to build the buyers' information system to be attuned to seasonal and regional variations when it was a mail order business, and the buyer did not necessarily notice that the children's cowboy boot orders in August came from Phoenix and the children's raincoat orders from Seattle.

After going into the retail shopping center business, then, Sears had the "structural" problem that store-level problems were not the same problems their buyers were good at solving, namely minimizing manufacturing and wholesaling costs. This was because the functions of wholesalers—which cost so much—of adjusting inventories to seasonal and regional variations had no place to be solved in a structure created by "adding stores to a mail order house."

Chandler shows that when the aggregate performance statistics of the store division were not satisfactory, many people were set to work to find out what was wrong. They located many different kinds of problems—for example, a consultant with experience in cutting costs of government administration, found problems of cost minimization that Sears could

work on. The structural strain of having a buyer information system that is not adequate to the new coordination problem does not necessarily mean the problem will be posed correctly. Many reforms may be instituted which do not solve the problem that Chandler (and we) locate by hindsight, that the organizational environment of Sears structures the buyers' problems in such a way that the buyers do not solve the stores' inventory adjustment problems as well as wholesalers in the open market solve competitive stores' problems.

If Chandler provides lots of evidence that Sears's people did not "rationally" solve their problem (instead, they "rationally" addressed a whole lot of problems in searching for what their problem really was), he still has to explain why "structure follows strategy," that is, why a functional result came about in the long run. His basic proposition is an "absorbing Markov chain" proposition: the organization will continue to throw up problems to individuals until they correctly identify the source of the functional inadequacy and build a structure that remedies it.

Commodity Line Rationality Versus Store Inventory Rationality

Note that the buyer versus store problem will recur in many different lines. In Arizona, Sears should not stock as many gardening tools in the spring as in Illinois; above the glacier line, Sears probably should stock more fishing and hunting gear in the sports department; in the Southwest, the automobile lines should be adjusted for the large population of pickups. Buyers specialize in commodity lines because efficient buying requires a detailed knowledge of the characteristics and prices of competitive goods and competitive producers. Individual-level adjustments to problems of different store inventories in different parts of the country by particular buyers specializing in commodity lines is an inefficient strategy, because too many buyers in too many lines have to learn too much about regional variations. Individual-level rationality by buyers in each merchandise line is thus not a very viable "functional alternative" to reorganizing the connection between stores and buyers on a regional basis. Sears needed to divorce some of this flow of store problems to buyers (after being reanalyzed in regional inventory terms by a regional management) from the mail order flow to buyers, because the mail order flow has to solve a different structure of problems.

There are two kinds of connections from the functional outcome (adjustment of store inventories to season and region without losing all the

advantages of massed buying) to the structural thing to be explained (operative divisions controlling wholesale buying and some kinds of regional administration of retail stores). The first is that individuals posed with problems of inadequate competitive performance in stores hunt for solutions. The second is that until they find a patterning of overall activity that solves the functional problem, people keep on being posed with individual problems that keep them looking for solutions.

So the basic problem Sears had, according to Chandler, was where to locate informational complexity in the organizational structure. The old regime, that of the mail order house, had something like nineteen decentralized divisions in the merchandise division (the number varied from time to time): nineteen lines in which the core executive—the buyer—had to keep track of what sorts of garden tractors were selling this year, of whether they could sell heavy goods like fertilizer competitively when they had to send it through the mail at a high transportation cost (they could not, when they were selling by mail order, buy carload lots locally, as a garden and lawn supply place in each town might do), of what manufacturer of law mowers could reliably produce a machine that Sears could guarantee customer satisfaction on without going broke, and so on.

The old Sears structure could be described as a "coalition of merchants" in these different lines, each functioning essentially as an autonomous buyer and "selling to" the mail order division, which in turn sold retail for the coalition. There was much decentralization of decisions to operative (i.e., merchandise-line) levels in which the big decisions about the sizes and contents of merchandise flows and inventories were made essentially by buyers in the different lines. This is one part of the essential information a wholesaler must have to be successful in the market; the other part is knowledge of local and regional peculiarities of the seasonal variations in how lines of goods move.

Sears's advantage was that they had a lot better buyers than commodity-line wholesalers for supplying a national market and for arranging for cheap production and response to changing fashions nationally in commodity lines. They had a big statistical base on which to estimate future demand and a developed routine for translating the necessary information from the buyer into a commodity-line section of the seasonal mail order catalogs.

The code word for this system in the mail order business is "quantity buying," by which, with that information and the big flow of goods through the mail order house, the buyer in a given commodity line could make it possible for the manufacturer to smooth out production of a

good over the seasons and have long production runs, and so make it cheaper. The manufacturer could plan a smooth production schedule in much bigger batches before a changeover, reduce the amount of clerical work on a given quantity of goods because it was all for one buyer, and would ordinarily have no inventory risk because he or she would be producing to Sears's order rather than to what he hoped a lot of wholesalers in different parts of the country would decide to do.

Wholesalers both had and created for the manufacturer a much bigger burden of clerical work than the wholesale part of the Sears mail order business did, and they had and created a bigger inventory risk as well. This was because decentralized wholesalers' information was not as reliable and their expertise not as well organized to take advantage of big flows of goods. Wholesalers were more likely to have to write off or sell at cut prices if a line did not sell to retailers, for instance, because a competitor wholesaler had a slightly better model. Sears or Penney's might guess wrong about what would sell at the retail level, but at least as a wholesaler for their own retail outlets they did not have to mark down a lot of goods because of a few big retailer decisions to go with a competitor. The Sears buyer did not have to have salespeople out convincing each department head or buyer in each department store in many different metropolitan areas and small market towns to buy from her or him, as the small wholesaler did.

The clerical, inventory, and sales costs of wholesaling and manufacturing are cut down by chain stores generally, and it is these costs that are especially cut by a joint buying operation of a huge mail order house with a very big chain store. But what Sears had been sacrificing in order to make that saving (because the sacrifice was not important in the mail order business) was all the information about local peculiarities of demand that were built into the thousands of wholesalers adapting their buying and sales strategies to local markets throughout the country.

Now that Sears was going into the local department store business, they had to try to perform the functions that all those wholesaler clerical and sales people and all those autonomous buyers in each retail store did in the "free market" structure. The free market structure was very expensive, but it did that work of relating local demands to centralized "commodity-line" production by manufacturers who sold on national markets (Beniger 1986). Their competition, unlike General Motors', was already decentralized a lot more than Sears would ever be, because the competition was many autonomous firms scattered throughout the country. The average wholesaler in the United States even nowadays has about twelve employees (Statistical Abstract of the United States, 1989 , 761). Sears had to compete against a structure that used a lot of clerks, a

lot of salespeople, a lot of special ordering from factories, a lot of experienced small entrepreneurs in particular lines of wholesale trade in particular parts of the country, a lot of folks taking inventory risks because they knew their local market—a structure that was doing a very good job at something Sears had not had to do when they were only in the mail order business. But Sears had the advantage of knowing a lot better than that decentralized apparatus of wholesalers how to use the power and efficiency of buying on a large scale and of advertising nationally through catalogs.

The structure Sears had to compete with would never deliver swimming suits in midwinter in Maine or skis to a department store in New Orleans. What we see in Sears in the period Chandler studied is Sears trying to improve their accuracy and flexibility in wholesaling up to the level that their competition was already at, without losing their cost advantages from large orders placed well in advance.

Note that except for a very few comments on how to get good store managers and department heads in retailing, almost all of Chandler's talk about administrative reorganization is about the wholesaling end of things: how to relate the buying and supplying of mail order with that of stores (the long-run solution was to have separate inventory and quantity controls but to use the same buyers); how to adjust overall supplies to regional market variations; how to put controls on the buyers by forecasting demand so that Sears would not be caught (again) with excessive inventories; how to get the departments in the merchandise department (buying) down to a reasonable number; how to decide who was going to coordinate transport and select suppliers for different regions to minimize transport; how much Sears was going to intervene in giving technical specifications to manufacturers rather than just picking what was offered; and so on.

Sometimes (always, really) these decisions had retail implications. But these decisions were, in the competitive structure of small wholesalers spread around the country, all wholesaler decisions. Mass or chain store merchandising is mainly a reorganization of wholesaling by substituting bureaucratic or administrative control of wholesaling decisions for market control. Sears's decentralization had to cope with problems that the already decentralized wholesaling system had not had, and that Sears had not had when they had a terrifically decentralized system of buyers operating as more or less independent merchants but selling nationally through the catalog in a mail-order-only business. To see what was wrong with Sears's "buyer divisions" for coordinating inventories at many local departments stores, we need to analyze what Sears's competition, the system of tiny wholesaling firms, accomplished.

The General Problem of Wholesaling

Where does the problem that wholesalers have to deal with come from? The analysis has to start with a contradiction between what one needs at the retail store level and what the factories can produce. Each factory makes a restricted line of goods, goods related by a common technology. What a store like a Sears branch has to have is a flow of goods into the store inventory that is shaped by the consumption bundle that customers are looking for when they go to a "men's" department store, what a given kind of person or family buys on their monthly (or so) trip to a big store.

The job of wholesaling is to take the set of flows of homogeneous goods from a set of factories and to turn it into a set of flows of heterogeneous goods through a whole bunch of stores that is shaped to the current buying habits of each store's clientele. Price, style, quantity, performance criteria, and the like for the mixed flow at the store end has to come from folks who know what is moving, has to be adjusted for class and regional variations in consumption habits, and has to vary with the seasons and with what the seasons mean locally. That is where the ultimate information comes from, from the individual Sears big store responding to local peculiarities in consumption bundles that families buy on their once-a-month shopping trips.

But to turn all these dribbles of information on how many sofas we sell in the spring in each store into a large order for a factory, the information has to be aggregated into one decision about how many sofas to order of what styles and prices several months in advance—and it has to be aggregated separately for factories in different parts of the country, since shipping sofas is expensive. That aggregation is how a chain store gets its price break. It is what the economics of chain store wholesaling consists of. The chain store has to transmit that information to its wholesale buyer in a given commodity line in such a way that, when it looks as if things are maybe not moving as fast this year, the buyer can cut down the flow volume without completely disrupting the manufacturer and the traffic department in a way that will cost the chain a lot in the long run. The chain wants to make the adjustment before it has a cubic mile of extra unsold sofas.

It is because this problem can be solved by forecasting, by using expert buyers in each line of goods, by having regional executives manage wholesaling in distinct areas of the country, that we have chain stores that can sell at lower prices. It is because the competition of independent department stores and independent wholesalers spread out all over the landscape can do that adjustment, at a cost of more clerical and sales work

and more inventory risk takers, that a chain store has to keep reorganizing until it gets a responsive system that solves the inventory problem of each local store and at the same time solves the manufacturer's problem of more expensive production if it cannot get larger orders far in advance.

Actually, the multiple-regional-divisional structure that Chandler focuses on was only a small part of the administrative innovation he describes. The main innovations had to be increased flows of market information from the various departments of the many stores to the wholesale buyers in specialized lines—information about what would move, how much of it would move, and at what price, style, and seasonal mix it would move. Regional top executives, even with larger staffs than they had, just could not do it. There was too much knowledge of too many lines of goods for them to administer it.

In fact, Chandler says as much, but he does not emphasize this point. He talks about the reforms undertaken by Barker, the vice president of retail administration and personnel, in the 1930s. As Huber says in his critique of Chandler's analysis of Sears, "Despite the centralized retail administration in place, Barker developed a policy of 'unit control' and other 'statistical procedures to ensure . . . that merchandise would be in stock in the right amounts at the right times' (p. 264 of Chandler). By 1934, according to Chandler, Barker had created a 'cohesive and capable retail organization,' even though it was to be more or less this same organization which would later be reorganized on territorial lines" (Huber 1985, 3). It may be that the regional divisional structure helped fine-tune this overall system for feeding buyers the adequate aggregated information that ultimately came from what was moving in the retail stores. But Chandler does not actually show the interconnection between the development of the information system that did the aggregating which functionally had to be done and the decentralization to regions as well as to buyers in each commodity line.

Another weakness of Chandler's evidence (Huber 1985, 3–4) is that Chandler does not actually show regional diversity. For example, the Pacific Coast territory must have stretched from Seattle to San Francisco, at least. But if one considers climatic variation, the wet mountainous market stretching out from Seattle and the desert market stretching behind San Francisco are probably no more alike than those of Seattle and Chicago. Instead, what may have been going on is that as Sears got a lot more stores in total, they needed to have lots of intermediate offices to handle the sheer quantity of decisions that are involved in running stores in San Francisco and Seattle, and then in Eureka, Portland, and Pullman. While Sears could run two stores on the West Coast with those stores'

big decisions being made in Chicago, it was difficult to run twenty Pacific Coast stores that way simply because there would be ten times as many decisions to review and execute.

The same was particularly true of decisions involving investments in local stores. Manufacturing investments are in commodity lines; retail investments are in locations. The investment arm of the general office needed to know that stores in the Pacific region went in the right places. Once they were in the right places, the store-to-buyer information flow could adjust the store inventory. It therefore seems likely that regional offices were evaluating investments in stores in particular locations—a general office function—rather than aggregating inventory statistics—an operating division function. The more such localized investments had to be made far from Chicago, the more a regional division was required to advise on them.

Organizational Problems of the Service Sector

All of this illustrates why it is that the service industries tend to have a lot of the labor in the economy nowadays. The service sector as a whole has many clerks, salespeople, commodity specialists like the buyers and department heads in department stores, and the like. These people know about the prices, quantities to be sold, qualities, fashionableness, and so on of a line of detailed commodities for a local area or a particular store. In particular, they need to know what will be wanted in one department or boutique of this product from Janzen, this one from Dior, or of this product from AC Sparkplug, this one from Black and Decker.

Then all that information has to be translated into orders to a wholesaler for particular quantities at particular prices, shipping orders and bills of lading, addresses on the hundreds of pieces of mail that keep things in order so the mail carriers know where to deliver them, and so forth. A very large share of the work in the service sector, especially in wholesale and retail trade, transportation, and communications, is the processing of enormously detailed pieces of paper about how much of thousands of different commodities go to this or that store, when, and at what price. It is in this mass of clerical and routine sales work that the feminization of the service industry has progressed furthest; only about one out of five lawyers is a women (about one out of three new law school graduates is a woman), whereas something like nineteen out of twenty secretaries or clerks in retail and wholesale trade are women (Statistical Abstract of the United States, 1987 , 386).

Sears's administrative problem emphasizes that what makes mass production possible is the processing and aggregating of this enormous mass of detailed information about commodity flows for thousands of commodities by all these women clerical and sales workers in the service sector. They are the ones who solve the problem of turning the particular demands of the clientele of a particular store into the factory's yearly plan of production, the plan that allows the factory to have efficient long runs of production of the same commodity.

The problem that Sears had to solve was this same one: how to make mass production possible when the information was concentrated originally in what was happening to this and that stock of lawn mowers in Seattle and in Phoenix, and in how much less demand for lawn mowers there was in Tucson than in Phoenix (because more folks in Tucson have desert plants on their "lawns," but people still have grass lawns in Phoenix). This is why Chandler's analysis of regional divisions is not adequate to explain all the changes made in information-processing structures using store data. Regional divisions did not usually have buyers to arrange for the production of goods with factories. What is problematic, what shapes the uncertainty, is the relation between store inventories and "quantity buying" by a decentralized "coalition of merchants," each in a single commodity line. For Sears to succeed, it had to coordinate its local store-inventory information so commodity-line quantity buyers could use it. Much of that information processing, as Chandler reports it, completely bypassed the regional divisions he thought were central.

But Sears did not gain its competitive advantage in wholesaling by regional organization. It gained it by using its mail order commodity-line buying divisions, by using information aggregated by commodity line from the stores. They used less sales and clerical labor, and created less trouble for factories, not because they had regional divisions, but because they had national commodity-line operating divisions. Thus, Chandler's main cause of divisionalization did not change. The divisions needed new information from commodity-line departments in stores, but their uncertainties were not substantially different from what they were in the mail order business.

Sears may have made its general office more competent to invest in retail stores by having regional divisions, but it already had a general office coordinating investment in commodity-line divisions. The causes for having a general office also did not change. Chandler's theory is more useful for explaining the structure Sears started with than for analyzing the administrative changes, including regional divisions, of the period he studied.

What Is Chandler's Independent Variable?

Chandler's analysis in Strategy and Structure is mainly about divisionalization; in The Visible Hand (1977) it is more directly about coordination and centralization. Consequently, his argument is clearest in Strategy when he argues about the causes for divisions becoming separated and internally centralized. We have only hinted above at these causes, as "being in several markets."

We have proceeded so far as if it were obvious whether a company was in several different markets. But since Chandler's argument as a whole depends on markets being a source of uncertainty that has to be adapted to (and since this book is about that part of his argument), we have to define markets such that they are separable sources of uncertainty. That uncertainty in turn has to ramify through the several functional departments (which make up the multifunction divisions whose origins Chandler is trying to analyze) in such a way that a coordinated response of the different functions to a market's particular uncertainties is necessary.

In this section we will be more specific about what "being in several markets" means. We will argue that in Chandler's argument being in more than one market had five elements: (1) that market uncertainties ramify into manufacturing and engineering; (2) that information on market uncertainties come from segregated social sources; (3) that those segregated social sources carry distinct information; (4) that the inventory risks that reflect the ramifying of market uncertainty are managed in similar ways within a market, perhaps in different ways in different markets; and (5) that total uncertainty is large. If all five conditions are met, then market uncertainty ramifies into the administration of manufacturing and engineering in such a way as to create divisions.

The Transmission of Market Uncertainty to Manufacturing

Consider, for example, Chandler's description of the steel industry (331–337), which ends with the statement, "If the market, apparently, has not been different enough [for different products] to bring about the growth of autonomous, integrated divisions, neither have strategic decisions been numerous or complex enough to demand the building of a general administrative office with general executives and staff specialists" (337). Yet this is in spite of the fact that "the big integrated steel works produce a vastly greater amount of a larger variety of products for more types of industries and businesses than do the copper smelters, refineries, and fabricating mills" (331). The key reason why vastly larger and more dif-

ferentiated markets do not lead to differentiated multifunctional divisions (i.e., decentralization as defined above) is that

the complex demands of scheduling and of coordinating departments require close centralized administrative supervision if the great integrated mills, the mines, ore boats, and distributing facilities are to be operated at a fairly even capacity and if loss and waste are to be avoided by piling up inventory in one department or another or through having temporarily idle facilities and other resources. The continuance and growth of the multidepartmental division may have been checked by the fact that the demands of a single market, whether it is in one industry or one region, is not enough to keep a single great mill operating at full capacity. Therefore such mills must draw on the larger national market. (336)

That is, for steel production in great mills, (1) there are great economies in producing all the stages up through molten steel (iron ore mining and transportation; mining and transportation of coal; coke production; reduction of the ore to iron; and burning off the carbon together with other alloying and chemical changes to make steel) for all the various products in one operation. There are also large economies of scale in combined iron and steel production (see, e.g., Scherer's and others' estimates in Scherer 1980, 96). (2) There are varying but substantial economies in producing many of the heavier steel products (plate, construction shapes, rails, large tubes) from semimolten steel at the same site. (3) The users of these products are willing to pay, in some delay between their orders and their delivery, for the cheaper products.

It has paid the great integrated steel mills to turn many markets in a given large region of the country into one market by administrative means. That is, they cut the connection between the order coming in and any special speedup or slowdown of the steel-making process. Instead they integrate all the orders for different products into a gross quantity of steel needed from the plant and govern the first phases of the process up through molten steel by that aggregated quantity. Then the flow of steel from the integrated part of the plant is sent to fabricating installations (especially rolling mills, but some to foundries), where it is differentiated to meet the needs of the different markets.

This means that the differentiation of the market does not result in differentiation of the main part of the manufacturing organization—mining, transportation, coke production, ore reduction, and steel production. The technology is not dedicated to serving a particular market,

but instead is a general-purpose technology that produces a raw material for a wide variety of markets. But for a large share of steel products manufacturing in which that raw material is used, it pays to have the fabrication and marketing of that product near the furnaces. Fabrication near the furnaces in particular reduces transportation costs as well as energy costs for reheating.

When the market does get differentiated in what kind of steel it wants, so that smaller (usually electrical) furnaces must be dedicated to producing it (as in the steel minimills that have been expanding recently while the great integrated mills have been losing customers), then there are great advantages in producer autonomy and integration of molten steel manufacturing with the marketing of the products of that steel. So far, that autonomy has mostly been achieved by independent firms producing specialty steel products rather than by having autonomous divisions in the integrated mills.

The first requirement for there to be "differentiated markets" in Chandler's theory, then, is that the manufacturing apparatus serving the market be a dedicated resource , one of little use for servicing any other market with a different pattern and source of uncertainty. Otherwise the uncertainty from the market will not in general be transmitted to the manufacturing operation, except after being aggregated with the uncertainty that comes from other markets. Sears did not have regionally differentiated manufacturing, so its regional divisions did not amount to much. The differentiation of contact with manufacturing was already there in the nineteen commodity-line "divisions" of the mail order house.

When there are differentiated markets served by common, general-purpose manufacturing, then we will expect to find differentiation of the marketing structure and considerable autonomy of each marketing department. Of course, separate marketing departments may have to be hooked onto the administration of the last phase of fabrication (e.g., the rolling mill) in such a way that uncertainties from the market can give information on what exact products need to be produced in that last phase, how much inventory needs to be carried, what engineering improvements in the product might be required (for example, for fabricated steel products, the recent development of X-ray and acoustic inspection of the finished product has allowed the detection of many more flaws, resulting in much-changed quality control procedures in the last phases of production as an adaptation to customer demands; see Stinchcombe 1985b, 157). There will then be pressures in steel to have innovations that correspond structurally to the multifunctional division only for a late stage of the manufacturing process. But these pressures will not tend

to penetrate to the management of the whole manufacturing operation—as happens when a given chemical plant produces only paint, another only explosives—and so will not actually result in multifunctional divisions.

The first point, then, in defining Chandler's independent variable—in defining when a firm is in several different markets—is this: whether the market of one manufacturing operation is different from that of another one depends on whether manufacturing operations are technically differentiated, so that the same general-purpose technical operation cannot serve now one market, now another. Even when there are reasons to administratively connect a general-purpose technical system—such as iron ore reduction and steel production furnaces—with technical operations serving different markets—such as rolling mills for producing plate (mostly sold to motor vehicle manufacturers), tubes (mostly sold to the petroleum industry), rails, and construction shapes (mostly sold through wholesalers and brokers to a dispersed clientele)—the differentiation of these markets tends not to be transmitted backward into the general-purpose technology that serves all of them.

More briefly, several markets with different sources and kinds of uncertainty may be transformed into one market by administrative means, if that seems wise and profitable. It is wise and profitable if a general-purpose technology can serve all the markets, especially when there are large economies of scale in that technology. (See 352–353 for Chandler's analysis of the same general pattern as is found in the steel industry for the early history of the petroleum industry, and 349 for similar considerations in the tobacco and meat-packing industries and for United Fruit and American Sugar.)

The Segregation of Product Information

One of the ways manufacturing and engineering are connected to the market is that engineered and manufactured features like design, color or finishing treatment, chemical composition, quality control and reliability, maintainability and service cost, taste, and complementarity or compatibility with other products available in the market (e.g., software) largely determine a product's success in the market. Engineers and the manufacturing branch thus need to get market information in order to do their design, manufacturing, inspecting, or finishing work better. The question of whether some improvement in the product is worth its cost, then, is decided jointly by what the market will pay for it and what the engineering and manufacturing cost of the improvement is.

The social location of that joint market and manufacturing information for a particular product depends on the social organization of the

selling process. The fact that information about what fertilizer sells in what sorts of packages to homeowners is concentrated in the same place as knowledge of what features sell garden tractors, while information on what fertilizer sells to farmers is concentrated in the same place as knowledge of what features sell gigantic field machines, is not determined by the technical interdependencies of fertilizer and tractor production. It may well be that the same machine manufacturers produce garden tractors and field machines, while one chemical plant produces both packaged fertilizers for the home gardener and carload lots for the farmer.

A commodity line that includes home fertilizers and garden tractors will therefore tend to develop information on the gardening market this year in a given Sears department in a given store, and that information has to be aggregated properly for the gardening buyer who deals with both the machine factory and the chemical factory. Similarly, outside Sears there will be a feed store commodity line with a farm implement department, which provides information on this year's demand for fertilizer in carload lots in a given part of the country and on the availability of credit for new field machines, and that information has to be aggregated by wholesalers and brokers for those same chemical plants and machine manufacturers.

The markets are different both because features that sell the products to homeowners and farmers are different and because the differentiated information is available in two different social structures. Of course, there is not much commonality in the concrete material features that sell a bag of fertilizer versus a garden tractor, and in that sense they address "different markets." But the gardening buyer in Sears (after appropriate reorganization of the Sears structure, as analyzed by Chandler) gets the information about the two products' features from the same flow of retail information, as aggregated by the appropriate inventory system and by informal transmission of sales know-how. The manufacturing commonality between field and garden tractors, or between a carload and a bag of fertilizer, does not mean they have common markets, because the information about what causes them to sell is not located in the same social place: the Sears buyer never learns what sells either carloads of fertilizer or great field machines, nor does the feed dealer learn what sells bags of fertilizer or garden tractors.

Chandler often talks as if selling more different products leads to differentiation into multifunctional divisions. For instance, in discussing why the chemical industry almost uniformly adopted the multidivisional structure, he starts with a quotation from Williams Haynes, American Chemical Industry—A History:

Production of chemicals by the petroleum industry appeared to be economically and technically sound, but most petroleum executives could not see what appeared to them to be a tiny market for a multitude of chemicals produced by a complexity of operations and sold to a long and diversified list of customers, tasks for which they had neither the technical nor the sales staffs. (358)

Chandler's basic idea here is that the more products one sells, the more different markets one is in, and so the more pressure there will be for multidivisional organization.

Presumably this generalization is approximately true for the way diversity in chemical products is produced. In chemicals, diversity is due to the knowledge needed to produce one material being useful in producing a chemically related material whose uses may be very distinct. When the uses are very distinct, then the knowledge of product characteristics that help to sell the product is likely to be organized into different social structures.

This principle applies even to the same chemical product. The knowledge of what makes nylon bearings and runners sellable to the furniture industry, for instance, is not found in the same places as knowledge of what makes ladies' nylon hose sellable in supermarkets and department stores. Similarly, one could not learn in the department store that a bearing for a drawer ought to be white, whereas one could learn that nylon for hose ought to be "nude" and that nowadays "nude" is not the color of Rubens's nudes. When the information needed to perfect the manufacturing process or the product design is located in different social structures, the manufacturing process has to be organized to respond to information coming from those social structures separately.

This connection between segregation of information and divisionization is shown by what happened in cases where diversity of products was not generated by research or manufacturing commonality, but by the social organization of the market itself. In most such cases, multidivisional structures did not develop. For example, Firestone "took on automobile parts, rubber and plastic items, and other lines [other than tires] that used its marketing organization more than its production facilities" (352) but did not develop a multidivisional structure. Similarly, a General Foods memo is quoted:

A single selling organization would be used in marketing the several product lines, since "the demonstrated economics of selling a line of products through a single

sales organization would be increased by the number of items handled by each salesman" [footnote to the original source of this quotation omitted]. To administer the other activities involved in producing such a large line, the company promptly set up a centralized, functionally departmentalized structure. (347)

Earlier, Pittsburgh Plate Glass, "to make fuller use of its marketing resources . . ., turned after 1900 to developing a line of paints, brushes, and related items that could be sold through the same channels as window and plate glass" (342) without developing multifunctional divisional structures.

The example of many commodities using a single merchandising division is related to the steel one, except that in this case it is the marketing information resources that are general purpose and the engineering and manufacturing operations that are specialized and dedicated to a given product line. General-purpose marketing resources (rather than dedicated marketing resources) do not tend to produce multidivisional structures any more than general-purpose manufacturing resources do. Hence, it was not necessary for Pittsburgh Plate Glass to integrate its paint sales information and its paint production group separately from its integration of sales information about window and plate glass and its glass production group, since the sales organization could transmit product information about both from an integrated structure back to the differentiated manufacturing operations. Later, some of the separate production organizations at Pittsburgh Plate Glass developed industrial markets, and the sales organizations marketing to these customers were integrated with the manufacturing function in a divisional organization. Nevertheless, "merchandise division" sells the products of all these manufacturing divisions "to independent dealers, jobbers, distributors, and directly to small customers. Since the chemical unit sells only industrial goods, it has long been almost completely autonomous" (342).

Chandler restates his own hypothesis about the number of products as follows: "Those that sold a larger variety of one major line of products in much higher volume to a greater number of industries and businesses have consistently centralized the control of their activities through developing and rationalizing their functionally departmentalized structures; while only those making and selling quite different lines for increasingly differentiated groups of customers turned to the new multidivisional form" (343; Chandler's emphasis). Our point is to define more exactly just what makes a group of products into "one major line." In short, a group of products becomes one line when a single marketing social struc-

ture provides engineering and manufacturing with information on what characteristics help sell the products.

Socially Organized Market Segments Carry Different Information

The problem we pointed out above in our critique of Chandler's analysis of Sears's regional multifunctional divisions reappears in his chapter 7. In a number of cases regional divisions have disappeared, to be replaced sometimes by product line divisions, sometimes by centralized structures (334–335 for U.S. Steel; 341 for American Can; 348 for National Dairy Products; 354 for petroleum, ambiguously; 355–356 for Indiana Standard in particular, less ambiguously).

The basic fact here seems to be that although regional marketing systems are socially differentiated, serving different sets of customers through distinct staffs, the flows of information about product characteristics or product mix demanded are not distinct. Once one knows what the ratio of chocolate ice cream demand to vanilla ice cream demand is in one section of the country, one knows the ratio in other sections. Tin cans for tomatoes and applesauce have common characteristics, even if they are sold in different parts of the country, and milk cartons substitute for milk bottles at about the same rate in California as in New York. Only when distinct social structures carry distinct information do they have to be integrated separately with manufacturing and engineering. A difference that makes no difference is no difference. If my analysis above of whether Sears's regional divisions carry distinct information about product qualities is correct, we might expect to see the regional divisions wither away there as well.

Consumer Products Versus Industrial Products

Du Pont started to differentiate its divisions when it for the first time entered the "package goods" market on a large scale, especially in paint (see above), while retaining its market for "tonnage goods," especially in explosives. Pittsburgh Plate Glass was already in the paint market, and it differentiated when various of its parts started to sell to industrial markets. The multifunctionality of Pittsburgh Plate Glass's divisions consisted in the fact that their industrial sales were administered together with their manufacturing, while their package good sales were still managed through the merchandise division. What is common to both histories is that selling to industrial markets and coordinating manufacturing with such selling is quite different from selling to consumer markets and coordinating manufacturing with that.

Part of the problem here is that the social structures that collect and aggregate information about desirable product characteristics are different for the two types of market (see above). Yet much of the problem is instead that the producing firm's job of projecting demand and the associated job of inventory control differ greatly for the two types of goods. With industrial goods, the customer typically predicts its own future demand, signs a contract for delivery in the future, and gives some warning if it is going to change that future demand. This procedure of course minimizes the inventory risk of the manufacturer. At the other extreme, the human capital resource of a regional wholesaler in a line of consumer goods is an informed projection of what will be sold in that wholesaler's area of the country; the wholesaler can then take the inventory risk involved in buying goods from the factory in hopes of a sale to local retailers. That is, the projection is not by the customer but by the seller, and the inventory risk associated with changes in demand is likewise typically borne by the seller.

There are, of course, intermediate forms. For example, building materials, though often sold to "industrial" consumers—construction firms—are not typically sold on long-term contracts signed far in advance, and so lumber yards and construction steel wholesalers carry large and risky inventories. And while Ford made the Ford dealers carry the inventory created in the depression by his continuing to produce at a higher rate than sales justified, other automobile companies took a good deal of the inventory loss themselves. Within a given industry, who does the projection, who carries the inventory risk, varies with administrative and contractual arrangements.

Nevertheless, the broad distinction is whether the buyer or the seller has to deal with the uncertainty of future demand, both in the sense of projecting that demand and in the sense of taking the risk. When selling consumer products, the manufacturer (or others on the seller's side of the market) has to do the projections and take the inventory risk itself. So manufacturers had better adjust their projection methods and their production response to fluctuations in the market—otherwise they, not the customer, will pay for projection and inventory mistakes.

The central rationalization of the information flow Chandler talks about in General Motors (146) is a set of running predictions of demand and of running corrections of those projections (projections for four months, updated and extended to a new month each month). The system of "unit control" introduced into Sears retail stores seems to have been similar, but with much-simplified projections based on recent sales (and resulting inventory changes). They had to be simplified because there were thousands of demands that had to be projected.

The pressure to form autonomous divisions tends to be higher where market knowledge derived from a large number of small flows of goods needs to be immediately translated into manufacturing and engineering decisions. This pressure is in general higher for goods in which the seller makes the projections of demand and takes the inventory risk. So in general, the integration of sales and manufacturing is higher in "package goods" than in "tonnage goods."

Where the demand is quite unpredictable at the level of the retailer or wholesaler, and so requires very rapid adjustment of product mixes and quantities (as in fashion goods in the apparel trade, novelties in dime stores, some building materials for buildings that are highly sensitive to the business cycle, and goods that are sold to secondary manufacturers in the fashion or other volatile trades, such as dyestuffs and artificial leather, information has to flow very rapidly back to the manufacturers (most of these examples come from Chandler's chapter 7; some are from Vernon 1960). No one can afford to take much inventory risk, but here, since sales are still basically from the retail inventory, the risk is on the sellers' side. The great firms whose administration Chandler studies generally stay away from these businesses, because such small flows are most efficiently managed only in entirely autonomous small and tiny firm structures, which rely a great deal on subcontracting.

Where there are large economies of scale in production, as in plate glass, or where specialized manufacturing knowledge is required, as in the manufacture of dyestuffs or artificial leather, the general rule excluding large firms sometimes breaks down. In such cases, however, sales of the specialized product has to be tightly integrated at least with manufacturing, and usually with some parts of engineering as well.

In commodity lines in which product differentiation is the basic form of competition, and consequently where small differences in the qualities of the product (real or merely advertised) can make large differences in market success, product design needs to be tightly integrated with market knowledge. In the sorts of enterprises Chandler mostly studied, product design (with attendant production-line design considerations) was in the hands of engineers. In such cases, engineers will also tend to be tightly integrated with sales (as well as with manufacturing, of course) giving rise to a typical integrated product division.

In the publishing trade, however, the relation of product design to the market is managed by authors and (especially) by acquisitions editors (Powell 1985), so it is acquisitions editors rather than engineers who tend to be tightly integrated with market information networks (often acquisitions editors are former book salespeople) and with manufacturing. There are no "engineers" designing books to integrate with marketing. Simi-

larly, in the apparel trade the central design experts who respond to the market are artistic and style professionals, so it is they rather than engineers who are tightly integrated with market and manufacturing people in the same enterprise.

The general point, then, is that manufacturing units oriented toward retail markets and consumer goods, or other goods with a dispersed and variable market (e.g., industrial goods for agriculture or the building industry), are pushed to integrate market information into the administration of manufacturing and engineering. Hence, when incorporated into large diversified firms, they need autonomous divisions. This is more true when competition by product differentiation requires rapid shifts in design or product mixes to the latest marketing information, as for example in selling industrial goods to the apparel trades.

In particular, these sorts of retail product competition businesses cannot in general be run by an administration attuned to a market in which the customers themselves project their own demand and take the consequent inventory risks (and reduce them as best they can), as is typical in selling tonnage goods or other industrial products for concentrated manufacturing operations. If a company has manufacturing operations oriented to such industrial markets, it is very unlikely to go into consumer markets without organizing autonomous divisions to connect market information intimately with manufacturing and engineering (or the equivalent of engineering, such as acquisitions editors or dress designers).

Sources of Total Amounts of Uncertainty in Markets

The more total uncertainty there is in the markets to which a firm is oriented, the more total information has to flow to govern manufacturing quantities and product mix or to institute shifts in product characteristics. The more whatever market information there is has to be coordinated with uncertainties of other kinds in the environment of production or to meet technical changes in products or production processes, the more total information has to flow from manufacturing and engineering to the people who coordinate those two functions with market information. (If costs or technical characteristics of products are uncertain, that is in effect an uncertainty of markets: costs determine prices, and technical characteristics determine desirability.)

That is, the total uncertainty of a given market can be analyzed as in Chapter 2, when we were predicting routinization of production. We "specified four variables that produce uncertainty in a production process, and have argued that the more uncertainty there is, the less the production can be routinized, and so the less one can turn skilled work into

semiskilled work by many different workers with different tasks (a 'collective skilled worker with semiskilled parts'). We have also argued that the most opportunity for routinization will exist when there is less uncertainty from (1) unstable markets, (2) unstable specifications, (3) unstable technologies, and (4) unstable natural conditions or unstandardized parts that affect the process of production." The same four variables should apply here to predict decentralization to divisions as are there applied to predict decentralization to skilled workers or professionals, except that they should be modified by the variables of linkage of manufacturing to markets, social structuring of markets, different information in different markets, and inventory risk by the sellers in the market.

In fact, we generally predicted in Chapter 2 that very high pressures for detailed response to vagaries of the market would produce so much decentralization that small firms of skilled workers would dominate a market, and there would be no decentralization of large bureaucracies to predict because there would be no large bureaucracies. Whenever large bureaucracies get into such uncertain markets, however, there would be pressure toward decentralization. Our question here is when that decentralization would tend to take the form of multidivisional organization.

The first four indicators of when firms are, according to Chandler, "in different markets" should tend to produce their effects more strongly, the more market or other uncertainty there is in a given commodity line. Chandler often talks about the degree of routinization in decision making as a determinant of how complex the coordination problem of operations is, or how complex the decisions about investment and long-run firm expansion are. Cases where he mentions routinization as a variable that depresses the tendency to form autonomous divisions include steel (337: "The selling of steel as well as the obtaining of ore and its production and fabrication still follows a fairly long-established routine pattern"); ambiguously petroleum, in its main markets (362, gasoline, fuel oil, and lubricants) and as a variable whose absence tends to increase divisionalization in the chemical industry (358–362, petrochemicals; 374–378, other chemicals); and the electrical and electronics industry (363–370).

There is also often an implicit argument in Chandler's analysis that corresponds to the argument in the Du Pont dyestuffs business, that more total uncertainty obtains in a new business (including a new business for the company), where closer coordination between marketing, manufacturing, and engineering is required than in more routine businesses. For instance, a proposal to incorporate the Stainless and Strip Steel multifunctional division into Jones & Laughlin's centralized admin-

istrative structure (333) seems to imply such a developmental routinization, after the innovation of building a stainless production and marketing operation in an autonomous division, leading to recentralization.

Multifunctional divisions responding to uncertainties of the environment rather than of the market probably include the extensive occurrence of divisions in the petroleum industry centered on regional crude production. To produce crude, large risks have to be taken in exploration, and large investments have to be made in drilling, crude stabilizing installations, and transportation to refineries. Both exploration and drilling involve great uncertainties about what is under the ground from three hundred feet to as far as three to five miles down. Separate crude production divisions are found in Jersey Standard (the Humble Oil Company, 173); in Standard Oil (Indiana) (the Stanolind group, 355); and in Mobil (Magnolia and General Petroleum, 356).

Considerable autonomy is generally conceded to foreign raw materials–producing operations as well as to foreign marketing subsidiaries in metals, petroleum, and some agricultural products companies. Sometimes it is not quite clear how far the presence of both marketing and producing subparts in a foreign operations division is really an integrated operation—sometimes, evidently, the foreign raw materials operations also ship to the United States, and the foreign sales divisions also sell goods produced in the United States. The uncertainty that ties separate production and marketing organizations together administratively, then, seems to come from the political foreignness of their common environment, rather than from the uncertainties of dedicated manufacturing operations being tied to the uncertainties of the foreign markets they are dedicated to.

Conclusion

The core variable to be explained in this chapter is the degree to which the information-processing and decision-making systems in manufacturing, marketing, and engineering have to be authoritatively coordinated by creating product-line divisions. The basic explanation has been that when distinct market uncertainties ramify the manufacturing and engineering of distinct products, then divisionalization and decentralization are necessary. But this statement is too vague to be of much use to a student of organizations. By reworking Chandler's data and argument, we have tried to give a more concrete version of this explanation.

Let us start at the back end of the chapter, with the social organization of information about market uncertainty. We argued that Chandler's im-

plicit definition of a firm as "in several distinct markets" was of distinct social structures bringing in information about market risks that affect manufacturing and engineering. Chandler's real independent variable might be conceived to be (1) the total uncertainty in a given market, including certain types of uncertainty of the environment of production and of technical change; modified by (2) a factor that describes how tied manufacturing uncertainties are to uncertainties in the market; modified by (3) the extent to which the information for adapting to the uncertainties is located in several different social organizations in the marketplace itself; modified by (4) the extent to which these different social structures actually carry different information, i.e., are not mere administrative conveniences such as regional arms responding to geographical distance; modified by (5) a factor that measures whether or not the projection of demand and the inventory risk are taken by the seller rather than the buyer in that market.

More formally,

where MD stands for the degree of multidivisional organization, then each variable in parentheses describes a flow of commodities and is a matter of degree, varying between 0 and 1 for convenience.

This description gives concreteness to the explanatory side of the general argument. The argument now says that whenever all five of these conditions obtain to a high degree, then the information-processing task of an organization breaks into separable divisional parts. Consequently, these five conditions predict divisions with distinct authority and that such divisions will be units in the profit, investment, and performance measurement systems of the general office, in the case of multidivisional organizations, or of the stock market, in the case of divisions becoming separate firms.

Only if divisions do not become separate firms will we need to explain the existence of a general office and of "decentralization" to operating divisions. The theory of market uncertainty and market information segregation does not predict anything about the general office, except that if it exists it will tend to restrict itself to investment, profit assessment, and performance measurement functions, much as the stock market or investment bankers do.

In sum, the variable to be explained in Chandler's central argument, as we discussed at the beginning of this chapter, is really centralization

within product lines . This only becomes an organizational problem when manufacturing, engineering, and marketing might not be coordinated within product lines—for example, when they are coordinated at a corporation level and the corporation is producing and marketing for several types of markets, as at Du Pont. When this is the case, internal processes may take place in the firm to produce decentralization. Information flows that focus attention, organizational theory that facilitates correct diagnosis, a structure of detachable high executives to study long-range organizational problems, a shifting committee structure allocating responsibility for solving these episodic structural problems—these are some of the conditions that facilitated such an internal evolution at Du Pont.

When the problem is instead how to manage investments centrally for a firm that is already very decentralized, as at General Motors, an evolution of more abstract information flows to a general office will tend to be instituted. We argued that the way Durant built General Motors showed that he (and General Motors as a corporation) functioned as an investment service for decentralized car parts suppliers, car assemblers, car sales finance companies, and car dealers. To continue to serve that function, Sloan (and the Du Pont financial crew who supported him) thought they had to build a system of more abstract projections to evaluate internal investments without destroying the operating efficiency of the autonomous divisions already in place.

Sears posed a still different problem for Chandler. This company was, in the mail order days, already decentralized (a "coalition of [nineteen or so] merchants") and already had a unified structure to coordinate the commodity-line "divisions." Sears's task, then, was not to build a Du Pont–General Motors–type of structure, because they already had one. Instead, it was to attach an information system that translated the behavior of store departmental (and specific commodity) inventories into information that the quantity-buyer divisional structure could use. Because Chandler misidentified the core of that structure (because in turn he was not looking very hard for the crucial links that ought to have existed, according to his own theory, between the regional divisions and the buyer divisions), he failed to interpret much of the administrative reform he reported at Sears.

His central trouble here was not applying a principle he often used in his chapter 7 analysis of manufacturing firms, having to do with whether the distinct regional markets really carried much distinct information to the manufacturing installations or buyers. Our argument here is that commodity line–supplier relations at Sears remained distinct with dis-

tinct buyers but that the regional markets by and large conveyed the same information to those buyers.

When Chandler's vague connection between ramifying market uncertainty and divisional organization is theoretically elaborated, it helps us to identify difficulties in his treatment. General Motors did not have the same reasons to develop investment expertise in the general office as Du Pont had to develop divisions. General Motors already had divisions. Sears's commodity-line divisions and general office were already in place; their regional divisions did not substantially change this structure, because they did not transmit much information that buyers needed.

What Chandler, modified, adds to our theory that formal structure is to be explained as information structures growing toward central sources of uncertainty, then, is a specification of the conditions under which operating information systems will tend to be integrated into divisions. He also gives a few hints about when investment decisions will be taken by a general office that is fed by very abstract information flows from the divisions. We have not had much success in developing a theory of why such decisions are sometimes taken by the stock market and investment bankers, sometimes by an internal corporate general office. That problem is left as an exercise for the reader.

In the next chapter we study in detail a particular kind of force that tends to produce divisionalization, namely the introduction of major product innovations. We treat this problem by analyzing how the findings of Chandler on making bureaucracies flexible fundamentally change the antibureaucratic premises on which Schumpeter (e.g., 1942) predicted that innovations would always tend to be introduced by autonomous entrepreneurs.