10—
Conclusion

Three Criteria for Theories of Organization

This book has been guided by three main criteria for theories of organization: (1) they should explain how organizations can be more rational than individuals (though of course they are not always); (2) they should be social structural and thus able to explain, for example, why 10 percent of an organization's parts can be absent on a given day and the organization still work the same way; and (3) they should explain variance deep inside organizations by variations in what the individual parts have to do—and by extension, they should explain variations among organizations as well as why some parts of a single organization are so different from others. After the general argument involving these three criteria together is discussed, each criterion will be presented in more detail with illustrations from the argument. Finally, the implications for research strategy in organizational sociology will conclude the book.

The first criterion is that the theory should provide the basis for explaining why societies that can organize large-scale complex organizations hardly ever leave anything important to individuals. Our explanation is that organizations ordinarily do all the things we have been analyzing in this book better than individuals do: they teach the advances in science better (Chapter 9); they produce both explosives and paints better (Chapter 4), drill for oil a couple of miles under the North Sea better (Chapter 3), and organize work better (better than we usually manage in our households) so that the people who do it have the appropriate skills (Chapter 7). The theory, then, has to explain why organizations are more rational than individuals.

If, as we will argue below, the variables we want to use in the theory are descriptions of uncertainties and of the social structures of those parts, and the units of analysis are the parts of the organization, the mech-

anism that connects them is a postulate of organizational rationality. In this book, the postulate of rationality is what connects conditions in the environment to the structure of parts of organizations.

But the concrete substance of rationality depends on what part of the total uncertainty a particular part of the organization has to deal with. Thus, rather than postulating one rationality for the whole organization (such as "maximize net present financial value") and assuming that the rest of the problems will be dealt with by an opaque "production function," we postulate many different rationalities depending on what uncertainty has to be dealt with to get a given subtask done. One of the many (and often the hierarchically dominant) rationalities of an organization is to maximize net present value. But that cannot be done, for example, without preventing blowouts and hence without dealing rationally with the risk of high-pressure gas coming into the drill hole.

Further, we do not connect the postulate of rationality to decisions (a mass of ethnography of decisions shows that they are not very rational), but to social structures that gather information and, more or less routinely, make decisions. The postulate of rationality, then, is much more a guiding principle of growth of the structure than it is a notion that a single utility is everywhere maximized in each decision. And this means that the tradeoffs "between rationalities" are ordinarily to be conceived of as competing principles of growth of different parts of the decision-making system.

For example, the structure for recruiting, retaining, and promoting faculty in a university is organized to deal with the uncertainty of whether people will do good research in the future, and grows into a baroque system of many levels of committees writing many letters to the outside "peers" to ask what decision the organization should make. The structure for assigning teaching is oriented to a structure of requirements for awarding degrees, for assigning course grades, for setting formal teaching loads and negotiating exceptions, often department by department, so as to get the teaching job done. They are interdependent structures because the tuition one can charge for the teaching is determined by university reputation for scholarship and because the research that builds the reputation takes time that might otherwise be used to teach. But the two structures do not grow at the same speeds and with the same authority in different colleges and universities, and the tradeoff between the values of teaching and research in a particular university is an outcome of the relative growth of two structures of rationality.

In some sense, this implicit choice is a long-run choice of utilities, and to some extent the tradeoff is determined by the hierarchically superior

levels of the organization. But the rationalities built into the American semester-hours course-credit system that Thorstein Veblen so liked to make fun of are not the same as the rationalities built into the use of citation counts to measure reputation that old-fashioned faculty who think one ought to read a person's work to judge them (that includes me) like to make fun of. As this example shows, it matters a good deal whether the rationality is located theoretically in the organization as a whole or in its subparts—a point we deal with as the last of our three criteria. If it is located in the subparts, then relative rates of growth of the subparts are the dynamic component in the choice of utilities, and the utility function can be expected to change over time. But the rationality of decisions is also determined by the social structure in which those decisions are made, by how far that structure has grown to be able to bring good information about the kind of uncertainty that mainly matters and to organize the decision process so that that information can be used when it comes in.

The problem of whether one can trust the source of information is pervasive in organizations, because the timely information about subparts of the uncertainty tends to be concentrated among people whose special responsibility it is to deal with it—which means that they need not report the best information available if so doing is to their disadvantage. So the problem of the appropriateness of the social structure in which an uncertainty is dealt with has two main parts: whether the information collected is good, and whether the people who have the information have incentives to use it for the benefit of the organization or motives to misrepresent it so that they look better. Performance measurement systems in particular are shot through with problems of correctly measuring the environment confronted by the organization in a particular area of its functioning so that they can evaluate performance relative to that (see, e.g., Eccles and White 1988, on the difficulty of getting the appropriate "market price"—or "cost," which in economic theory is supposed to be the same—at which divisions of a multidivisional company sell each other their products).

The information system on a particular area of the environment, pervaded by a particular kind of uncertainty, "needs," then, to be constructed in such a way as to solve the main difficulties, substantive and motivational, for collecting information on that sort of uncertainty. The basic assumption of our theorizing to explain social structure variations within and among organizations is that we can explain much of the variation in different parts of different organizations by those needs. We hope that this approach can give us systematic answers about why we feel, when we come to the end of Olsen's analysis (in March and Olsen 1976)

of how a nameless university chose a successor to a nameless dean, that we have not only a case study of a particular university but also a general case of an organization trying to hire on the basis of scholarly distinction (in spite of hiring also for an administrative job). We get that feeling of generality, we argue, because we know we are not studying a particular decision that just happened to come out the way it did; rather, we are studying a social structure shaped by the nature of the decision problem, which is shaped in turn by the kind of problem it is to get reliable information on scholarly distinction. And this brings us to our second criterion for theories of organization—that they explain variations in social structure rather than variations in the outcomes of particular decisions.

The big difficulty with approaching organizational life through decision making, as in the Simon and March tradition (Simon [1947] 1976; March and Simon 1958; Cohen, March, and Olsen 1972), is that the fundamental fact of the matter is defined, in the first instance, as what happens in a particular decision. The aura of generality about the best work in that tradition (e.g., March and Olsen 1976) comes from tracing what happened in a particular case back to something we recognize as general. For example, the ease with which a search committee for a new dean collects information about candidates who have written articles in books edited by the former dean would clearly be similar in the social structure of other universities, as all professors would recognize. The reason economists have trouble assimilating the results of such studies into their theories is exactly that their idea of generality is not of a social structure built into the organization's mode of proceeding, but of the human tendency to maximize self-interest narrowly conceived. Our basic argument here is that the things worth explaining about decisions in organizations are those that are due to continuing features of social structure. That is, we believe that one should follow one's intuition about what is explicable and general in Olsen's study of selecting a dean—which is not that people make mistakes about scholars' reputations because they do not have time and energy to get everything right, but that universities by their nature collect information by the method of scholars following out their own contacts to get good evaluations of the research that candidates are doing.

But of course, to be sources of variance in behavior between organizations and their parts, such features of the social structure have to be variables. Social structures of parts of organizations have enough constancy to be worth explaining, and enough variation between organizations and between their parts to give something to explain sociologically. Our purpose in this book has been to say what those social structural variables are, and why they are systematic in many different organizations. Our

answer to the why question is that organizations grow social structures to handle decisions by creeping toward information about the main sources of uncertainty that matter of the organization. Of course, those sources of uncertainty are different for different parts of the organization.

The big trouble with some versions of the contingency theory of organizations (the classic source is J. Thompson 1967, but Thompson is also interested in internal variations) is that it tries to define one big uncertainty for the whole organization. But if an organization is drilling for oil in the North Sea, one part of the organization faces the uncertainty of whether high-pressure gas will blow a mile of pipe and other equipment out of the hole; another part faces the uncertainty of whether OPEC will fall apart (no longer uncertain, though economic theory taught that OPEC would come apart long before any practical people could assume it actually would), in which case only the rich fields would pay; another part faces the uncertainty of whether the marine insurance industry will be able to put together a coalition willing and able to insure the enormous investment in installations (Heimer 1958b, c); another part faces the uncertainty of whether the socialist government will let them keep the profits their contract with that government gives them; and so on. This is then not one big uncertainty, but lots of different uncertainties, and one collects the latest news about them in very different ways. Consequently, one has to build social structures to keep tabs on different parts of the environment, to respond on the very different time scales that these different uncertainties are uncertain on, to be competent to evaluate the risks as the news comes in, and so on.

The and so on s in the previous paragraphs are meant to suggest that an organization has to be built to respond better to each of the various contingencies it is faced with than the average individual could do, because it would be a very unusual individual who could monitor the information about exactly when high-pressure gas might come into a drill hole and what coalition of bureaucrats and politicians might convince the socialist government to be a reliable partner in its contracts with world capitalist enterprises. The standard operating procedure of the various parts of the organization, or the "program for decisions" in the Simon and March language (March and Simon 1958), is not only a system for taking discretion away from smart individuals who might have guessed better in a particular situation; it is also a system for collecting information so that the ordinary people who are likely to be at the point of decision at the crucial time are moderately likely to take the right decision both about blowout protection and about what tack to take to defend oil industry profits in Norwegian politics. The standard operating procedure that will do one

of these things well is very unlikely to do the other well. A rubber stamp with Bullshit! on it (see Chapter 3) would destroy the delicate political balance, while it is likely to remind drillers to pay attention to what other drillers say, rather than to the papers sent out by people onshore, if they want to stay alive.

According to our third criterion—that the units of analysis should be parts of organizations—a theory that explains why, say, the labor contract for professors in a university is so different from the labor contract for engineers in a manufacturing firm (see Chapter 9; also Schroeder and Finlay 1986; Heimer 1984) should also explain why professors' contracts are so different from accountants' contracts within the same university. A casual inspection will show that less than half of the people in a university have jobs that depend on scholarly competence and reputation outside the university. The fact that most organizations have none, rather than less than half, has to be explained, to be sure; but so does the fact that the universities most devoted to scholarship have most of their people devoted to something else. A theory that is good for explaining the differences between organizations should be good for explaining differences between parts of a given organization.

This criterion means that the theory developed by Williamson (1975) to explain why some activities are organized hierarchically while others are organized in a market, for example, needs to be extended in two ways. In the first place, it should be able to predict when activities organized in the market itself will have more hierarchical elements in their contracts (as we tried to do in Chapter 6 and Williamson has attempted in his later work [1985]). But second, it should be able to predict which parts of an organization will be organized more hierarchically, and which more like a market (as we followed Chandler doing in Chapter 4, discussed for universities in Chapter 9, and is discussed in relation to the particular problem of internal pricing in Eccles and White 1988). The crucial points here are (1) that the transaction is too small a unit to shape a social structure, while a long series of transactions resolved with similar sources of news about similar uncertainties can shape durable parts of the social system, and (2) that the organization is too big a unit because the uncertainties it faces are too many and too various to have one dominant outcome.

At the opposite end of the variation in units of analysis (ranging from transactions to whole organizations) is the theory developed in organizational ecology analyses of failure rates of whole organizations (e.g., Aldrich 1979; Carroll 1987). The theory relates such rates to whole organizations' structural features and to features of their environments. Naturally, this misses a lot of failures that did not destroy the organiza-

tions because in other respects the organizations were successful (NASA still existed after the Challenger disaster). But more important from our point of view, it does not provide the opportunity to use nuanced measures of social structure—of whether, for example, the accounting department is so set up and so related to the rest of the organization as to give prior warning of failure of parts (see the discussion of Du Pont's use of accounting indicators to locate failing parts in Chapter 4), thus allowing structural changes to be made in the failing parts before the whole organization is brought down.

Everyone knows from experience with their own organization that there are enormous variations from place to place in the organization's social structure, in rights and incentives, in hours, in style of dress, in access to expense accounts, in autonomy, in responsibility for crisis management, in career prospects, and so on. If, as we have argued above, our central variables ought to be social structural, then we are required to be able to analyze these internal variations. They appear to the newcomer as part of the natural order of things; the newcomer to a university, for example, learns as a natural fact that professors are allowed to do pretty much as they please and be prima donnas, but that the rest of us are subject to the authority of our hierarchical superiors. But these differences persist and are part of the natural order of things for the next cohort of newcomers: they are part of the social structure.

Organizational Rationality

People who work for organizations are held to a higher level of responsibility to be rational than are people answering only to their individual desires needing to be maximized. Further, people who design information and decision systems in organizations are held responsible for both their own rationality and that of others who will be making decisions in the system. Consequently, the ideal of calculating all the tradeoffs among all the costs and benefits, the ideal of collecting all the information needed to make a rational decision (if only to cover one's ass), the ideal of considering benefits some distance in the future on roughly the same basis as benefits at the next moment—all are more nearly achieved in organizations than in individual lives. I take the "postulate" that organizations are more rational than individuals as an observation, and consequently as something we need to explain, as well as taking it as the basis for the reasoning in this book.

The empirical assertions of the arguments we have posed here can be turned on their heads to explain the connection between organization

and rationalization. We can say of the argument about divisionalization in Chapter 4, for example, that the more we observe that organizations are divisionalized when they are in multiple markets (in the sense defined in Chapter 4), the more support there is for the proposition that organizations are sources of rationality. Similarly, when organizations use seniority as a certificate for selecting workers for moderately stable interdependent productive systems, but use more peer judgment of skill for certificates of workers' ability to deal with a variable flow of contingencies (as suggested in Chapter 6), we have further evidence for organizations as sources of rationality.

This book, however, has been written the other way around, with the rationality of organizations being postulated. Note that this is not a postulate that organizations do not make errors. Instead it is a postulate that one can explain variations in the social structure of different parts of organizations on the presumption that the structure of the information and decision system will grow toward those sources of uncertainty whose rational management will matter most to the organization. The evolutionary principle here is that the organs of an organization grow better adapted to their environment over time—in other words, that the rationality of an organization improves (see Carroll 1987 for evidence that new organizations fail more often than older ones).

At a particular time, for example, an organization may be trying to run the administration of a product innovation in the same way they run the production of housepaint. But this will create such tensions, such clear irrationalities, that if the organization keeps up its commitment to the innovation it will evolve a tighter integration of engineering, marketing, and production within the innovation's administration, will reward skill more strongly in the innovating subpart of the organization, will orient its marketing staff toward marketing add on innovations through regular contacts with former buyers rather than toward low price and steady supplies, and so on. Similarly, if cost reduction is a primary organizational imperative, we will expect the cost accounting system to grow more rapidly in the direction of a more detailed system of cost codes, of a more flexible system of analysis of costs, and of a system of financial categories that are more unique to the organization rather than dictated by the financial and tax institutions of the society. In short, the postulate is not that organizations are rational, but rather that as organizations become better organized, they will tend to improve their information and decision systems and so become more rational.

But our fundamental postulate here is that the core of organizational rationality is efficiency in news processing, not efficiency in maximiza-

tion. The reason drillers are oriented mainly to other drillers is not that other drillers have maximization schemes that come to better answers over the long run, but that other drillers, especially those in the same fields, have news about where a drill crew is likely to get into trouble. What makes internal communication such a high priority is that only those other drillers will have news about what is two miles down the hole, not that only they will know what to do about a pocket of high-pressure gas or a stratum that tends to collapse. If they can find out where the high-pressure gas is likely to be, drillers' skills will allow them to minimize risks. And they get their skills from their colleagues only in the long run: in the short run they get news from them.

Likewise, the observation that universities wait for outside offers to find out how much their faculty members are worth (J. Thompson 1967) does not mean that the faculty and administration of other universities are so much smarter than our own. (For one thing, this could not be true for all universities.) Rather, it means that faculty and administrators at other universities have no reason to overvalue the reputations they are buying, whereas the best experts on those reputations within the first university—namely, the owners of the reputations—want to exaggerate them. News about reputations of one's own faculty has to be gathered outside the university partly because outside opinion is precisely what reputation consists of, but mainly because the news from inside is so easily and routinely corrupted.

Starting with the postulate that rationality for an organization consists mainly of reacting accurately to recent uncorrupted news, then, is a different theoretical orientation than imagining that the main source of rationality is the ability to maximize. For example, parts of one's own organization are systematically motivated to provide distorted news, as when one asks scholars how high their reputation is in order to determine their salary. The argument of this book is that one should react to this not only by asking what coalition in the organization will control what is actually maximized; one should also ask how the organization might grow its information and decision structures toward sources of uncorrupted news, so it can in the long run maximize what it set out to maximize.

One could, of course, use the same approach to explain the problem of individual rationality—that individuals grow competences and information-collection strategies which enable them to deal with those uncertainties that are most important for maximizing their utilities. But a whole life is too complicated to develop such personal competences throughout the range of all important uncertainties. In contrast, the problem of man-

aging the uncertainties of a department within an organization is often sufficiently limited that specialized growth of social structural competence can pay off. Thus, within the department we will expect to find that individual competences for collecting and reacting to specialized uncertainties (i.e., skills) will grow (Chapter 2), and that specialized information and decision systems in that department will develop to detect, hire, improve, and retain those skills (Chapter 7). We will also expect specialized departmental information-collection systems to grow (Chapter 3) and authority to integrate information about a market with distinctive uncertainties collected by different departments to be integrated by divisional organization (Chapter 4).

All this complexity in the information and decision system, all the files, the specialized computer programs, the cultural elaboration of enclosed departmental systems, and the like, is much more than could be crammed into an individual. And it is all about a part of the environment—how to drill for oil at the bottom of the North Sea (Stinchcombe 1985e), how to produce and market passenger airplanes (Newhouse 1982)—that is much more specialized than the whole of an individual's life. If one imagines that what people and organizations are doing to be rational is solving a maximization problem with given data, there is no reason to believe that an organization should do this job better than an individual. If one imagines instead that rationality is efficient processing of news about uncertainties, it is obvious that organizations are likely to do it better than individuals, and that individuals are likely to do it better in their organizational roles than in their lives. Thus, the idea that organizations are more rational than individuals is intimately tied up with the conception of rationality as processing and reacting to news about uncertainties.

What Makes the Dependent Variables Social Structural?

The catch phrase we have used to denote the general class of phenomena we are trying to explain is "information and decision system." When we described the difference between cost accounting systems that collect data for use in improving a productive system and financial accounting systems, we were not trying to explain where all the numbers come from that enter into a particular estimate of how much a given machine will save, or how those are translated into an "expected time to payback" using assumptions of a stable future market. Instead we were interested

in the fact that the cost accounting system will be used, among other things, for a series of ever-changing projects that involve buying many different machines and that need different sorts of data, often unique to the plant. Financial accounts have to answer the standard questions that tax authorities and capital markets routinely ask in a way that those markets can trust (e.g., in a way that is auditable). What is accidental and contingent about cost accounting numbers is what happens in distorting the numbers about a particular machine and resolving the residual uncertainties in a particular machine-buying decision; what is of the essence in accounting is the different characteristics of the systems responsive to the uncertainties of a series of cost-reducing "projects," as contrasted with systems oriented to satisfying the authorities and the capital market that they will get their share of responsibly and honestly calculated returns.

We have in general chosen to analyze social structural systems rather than cultural systems, the social relations across which information is transmitted and modified rather than the category systems with which it is analyzed. Gaye Tuchman (1978), for example, analyzes the social structure of information processing in the news business—the way story assignments create flexibility, say, or why the names of news sources are defined as individual newspeople's private property. But she also analyzes the culture that newspeople use to define the nature of a story: news stories may be spot news (a fire), soft news (the problem of battered children), continuing news (a bill's course through Congress), or "what a story" (the assassination of a president).

Tuchman shows that these categories do not exactly characterize what is going on in the world or what the final story will look like; instead they reflect the problems of scheduling the work that the flow of information about a story creates, and in that sense are an example of the kind of analysis the theory here would generate. The structural analysis shows why news organizations would need a quick system of classifying stories by what sorts of work-scheduling problems they will create. The agency problem of how to get newspeople to develop a network of news sources and the requirement that people be reassignable to stories outside their specialty, sometimes to become subordinate to lower-status people who were on the scene first, are things of the sort we have been building theories of. But the classification scheme developed in the informal culture of newspeople is not the sort of variable that enters much into our analysis here. The content of the culture used to define the uncertainties that, in our theory, create the structure, the content that sets the information and decision structure into appropriate motion, we have left to one side. Instead we want to know which different sorts of information for making

which different decisions the structure has to be ready for, not which decisions it does in fact make when various uncertainties come in.

The reason for leaving the culture that is used to analyze uncertainties out of our theory is that that culture gets its stability only from the structure of stability of the social organization of which it is a part. For example, it is only because the decision about what goes into the newspaper has to be made in time to print it before it is delivered that the news collection problem is so pervaded by the problem of temporal organization of information collection. Consequently, the fact so ingeniously demonstrated by Tuchman, that the concepts used to classify news stories have to do with the temporal characteristics of the work that has to be done rather than anything about the resulting story or the nature of the event, is from our point of view to be explained by the social structure she also describes. There is of course no contradiction in adding a set of cultural variables to be explained by those social structural variables we have tried to explain here. Our only point is that this is a different job from the one we have tried to analyze here.

To put it another way, we do not analyze why the structures we have outlined here sometimes make mistakes, why with the best drilling crew in the world sometimes one stills gets a blowout, why a firm that stabilizes labor relations by playing the good-sportsmanship incentive game with its machinists goes bankrupt even when the system works fine, why no one was ready to cover John F. Kennedy's assassination (Tuchman 1978). These things must be dealt with in the classic studies of decisions, such as Allison's analysis of the Cuban missile crisis (1971), because when one studies a particular decision one has to explain why it came out the way it did. We have explained not the outcome of particular decisions but the structure because we did not have a theory of error, a theory of why even when the structure was well designed to deal with the relevant uncertainty the world still managed to fool it.

The advantage of such a strategy is that the social structure stays in place to be studied. When Olsen studies the biases of an academic search process, as part of a diagnosis of why universities are "organized anarchies" (March and Olsen 1976, 82–139), the interest is focused on the particular causes that made the process deviate from the ideal set forth in the university's decision-making procedure for faculty appointments. But it is very unusual for an organization to have the exact ideal from which this particular process deviates; most organizations do not collect reputational information from other organizations for hiring decisions. While it is true that faculty, for example, often do not show up for the meetings called to take crucial decisions, that is largely because universities do not

put them on the relevant committees unless they have a good deal of other research work, graduate student supervision, and cosmopolitan responsibilities that clutter their schedules and take them out of town. Why should one have an ideal that causes it to be impossible to live up to it?

Our explanation of why universities are organized anarchies is given in Chapter 9; our argument would be only that organized anarchy is a diagnosis of the state of the dependent variable: what kind of information and decision system is appropriate for universities. Briefly organized anarchies are found in universities (but much less in teaching colleges or secondary schools) because only outsiders, "peers," can be trusted to give uncorrupted information on the crucial decisions of a research university and on the "reputation" of people it might hire or promote. Its information system, then, has to start with a person's peers inside the university, who will know how to collect information from the relevant peers outside.

Since a university can charge a good deal more for its services if it has a good reputation, it has to collect its information through the people who carry the main work load at the lowest level—its faculty. This inability to detach people from their other obligations to do crucial administrative tasks means in turn that universities must make their decisions in committees with only two-thirds of their members present, who are at best devoting only 5 percent of their working time to collecting the crucial information for an optimal decision, and who think the administrators responsible for the decision are an irritating and more or less useless excrescence on the body of scholars. Other organizations, the ones that can get full-time administrators to come to meetings for crucial decisions, do not have to get the information collected at the lowest level by people of the highest reputation as scholars. Thus, the shape of the errors made in organized anarchies, their general inability to invade the research space of departments and centers, and the complications of their personnel systems that bet on future research reputations in a wide variety of fields are determined by the nature of universities' uncertainties and by the difficulties of getting unbiased information within the organization on those uncertainties. These shape the kind of social structures needed to process decisions, which are more or less necessarily organized anarchies.

We have chosen here to theorize about the social structure rather than about the decisions because without the social structural context the errors appear merely as avoidable mistakes, weaknesses of the human beings rather than "by-products of the best we can do under the circumstances." We have provided the basis for analyzing the types of errors that may be distinctive of different sorts of structures. Just as we expect decisions to be made in universities with only two-thirds of the relevant

people at the meetings, so we will expect drillers to pay too little attention to how to save money on spare parts, marketers of software innovations written for a given innovative computer to pay too little attention to clients who have bought a different computer than the computer they are trying to preserve the monopoly of, and the like.

Similarly, we have (Chapter 5) given grounds to expect, for example, that the ways in which the information and decision system for product innovations will differ from the larger organization in which it is embedded in similar ways in most organizations. Further, we have given grounds to expect that when this is not true, the innovations have even less likelihood of success than innovations generally do. Because the success of innovations is very uncertain in the best of circumstances, we will expect there to be a lot of errors in decisions even when the innovations are ideally administered. Our purpose, however, is to explain nearly universal differences between innovation management and other management, so we have concentrated on explaining the social structural requirements of innovation management.

We expect there to be similar tensions in all manufacturing firms between building an accounting department to support cost reduction planning and building one to certify profitability for the capital market, as analyzed in Chapter 3. These tensions, we argue, will tend to be a feature of the social structures of accounting departments in manufacturing firms, given that requirements for using the same data for different routine purposes are generally incompatible.

The variables to be explained, then, are social structural features of information and decision systems, because these can be detached from the particular decisions being made and because it is easier to generalize across organizations, and across subparts of organizations, when the value of the variable one is explaining is of the essence of the organization, rather than an accident of the decision.

Units of Analysis and Variations to Be Explained within Organizations

Each substantive chapter of the book can be thought of as about a main type of social structure useful for dealing with particular kinds of uncertainty: skills in Chapter 2, manufacturing subsystems in Chapter 3, autonomous divisions in Chapter 4, specialized structures for introducing innovations in Chapter 5, interorganizational contracts in Chapter 6, personnel and labor market systems in Chapter 7, categories of workers subject to uniform incentives and output measurement systems in Chap-

ter 8, and specialized scholarly reputational systems in Chapter 9. That is, in each chapter a broad class of social structures is presented that responds to a distinctive amount and kind of uncertainty.

These different sorts of structures are partly autonomous, as they have to be to do significantly different jobs for the organization. But they can also operate simultaneously and together in determining the details of a given social system. Thus, for example, a drilling supervisor has a skilled job (as we analyzed in Chapter 2) because he must respond to many contingencies and must have a repertoire of responses ready for instant use. He is in turn embedded in a system that gets most of its information from within the drilling operation in that particular field, as discussed in Chapter 3. To explain the complete social structure of a drilling crew, we need to use the causes and structural adaptations to those causes that were outlined in the two chapters: why drillers are skilled in a way similar to machinists (Chapter 2) and why they are at the same time very different from machinists because the news about the uncertainties of their job does not come, as a machinist's does, from other parts of the organization than the drilling crew (Chapter 3).

The cost accountant who specializes in negotiations within the same firm about which costs should be allocated to the drillers, which to manufacturing operations, also needs to be skilled, as machinists and drillers are: to remember the precedents and the general policies on the matter, to understand the general purpose behind the policies so that discretion can be guided by those purposes, to know whom to consult, to make sure all the trails of paper about a given expense end up in the place corresponding to the cost allocation, with the correct signatures, and so on. But the general purposes behind the accountant's allocation are oriented to taxation law and to capital accounting practice and law, to the measurement of executive performance in a fair way for both drilling and operations, to trails of paper that lead from invoices from outside the organization through to the final cost allocation, and in general to information systems that have to be uniform both within the organization and, to some extent, across organizations (see Chapter 3). So the same basic types of features of jobs that make drillers, machinists, and accountants skilled are embedded in very different types of manufacturing information systems, because the uncertainties the skills respond to come from different places. One kind of skilled person listens mainly to other drillers; another listens to the requirements of the people using the machined part; the third listens to standards that spread throughout the organization, throughout the society, or, in the extreme, throughout world capitalism.

Some themes recur throughout many types of information systems.

For example, when some set of activities cannot easily be decoupled because the performance of one depends on the performance of another, because they cannot easily be separately measured, and because they have to be carried out under time pressure, the information for the activities needs to be processed by the same authority system and the activities need to be subject to common incentives, because they have common measures. I have called this, in one particular context, the "decoupling principle"—that activities should be subject to separate contracts only when they can be decoupled, in the sense of low interdependence, separate measurement of performance, or loose scheduling so that late and imperfect performance of one does not lead to late or inefficient performance of the other (Stinchcombe 1985e, 68–71). This decoupling principle applies not only to contracting practices for construction, but also to which activities should be administered in a centralized fashion by a division of a multidivisional company, which should be administered separately by the separate divisions of that company (see Chapter 4); to why innovations often have to be administered as minidivisions (Chapter 5); to why systems of subcontracts often include hierarchical elements (Chapter 6); to why the main decisions that depend on knowledge of the reputation of a given scholar in his or her research specialty have to be initiated in the scholar's own university department as supervised by the chair rather than elsewhere in the university (Chapter 9).

The form of the book suggests that such basic general principles are not sufficient for a substantive analysis. They will not tell us how integrated systems responsive to distinctive types of news about uncertainties are formed. The centralization of divisions analyzed in Chapter 4 is analyzed there in the light of the need to coordinate different parts of the organization to respond to a given market. The substantive generalization concerns the number of markets and their degree of uncertainty; the number of uncertain markets is used to predict the decentralization to divisions, because that is the way uncertainty affects divisionalization. In Chapter 5, in contrast, we are interested in a certain type of market—the market for an innovation—and how that market is connected to engineering, marketing, and manufacturing in a distinctive way in many different kinds of organizations. The argument of Chapter 5 concerns the common ways that different organizations are likely to respond to the problems of the uncertainty about how to develop an innovation and to manufacture it cheaply so that it will fit its developing market. Paint at Du Pont, while not an innovation, required a separate division that was internally centralized because its market was different from that of explosives, which were also not an innovation. Since the substance of the uncertainty of an innovation leads to a peculiar dependence of engineer-

ing on marketing and manufacturing on engineering, it generally requires centralization within the innovating subpart of the organization and differentiation from the rest of the organization.

The point here, then, is that centralization of authority in subparts of organizations often forms a part of the response to uncertainty when uncertainties are interdependent. But it plays quite a different role in determining the resulting structure when the uncertainty comes from, say, the difference between package goods and tonnage goods than when it comes from administering an innovation in a firm most of whose products are not innovations. So although the decoupling principle may be a fundamental one in explaining where we find centralization, exactly what its implications are for any concrete setting varies with the details of that setting. One has to be very careful in making the mapping from the components of the theory, such as the decoupling principle, to the organization of response to a particular type of uncertainty, such as differentiation of markets, innovation, or estimation of scholarly reputation.

Similarly, skill pervades our analysis. In Chapter 2 the basic description of what skill is ties it to uncertainties that have to be dealt with at the level of the workers and to the qualifications required of workers to produce that response. It appears again in Chapter 5 in the explanation of why innovating parts of an organization have higher skill levels. In Chapter 7, a large part of the analysis of why firms participate the way they do in the labor market has to do with their problem of finding, certifying, rewarding, and motivating skill use and skill acquisition. But the personnel system uncertainty in Chapter 7 about who is skilled is different from the uncertainty in Chapter 2 that requires skill; it has to be solved with a different structure of information processing and decision making, one that extends outside the organization and structures segments in the labor market (Pfeffer 1977).

Skill levels appear again as a basis for the similarity of labor contracts of artisans and other manufacturing and construction workers in Chapter 8, and consequently as a basis for class consciousness. The uncertainty that artisans and craft workers have about who is on their side, who on the other side, and how that analysis is influenced simultaneously by the information they get from the political system and the information they get from the incentive systems that are built into their craft labor contracts is neither the same uncertainty as creates the need for their skill nor the same as shapes the employer's problem of how to decide whom to hire. Of course, the incentive systems that employers design are influenced by both skill requirements and hiring uncertainties, so these can be regarded as ultimate determinants of artisans' class consciousness.

Thus, each chapter of the book can be thought of as specifying an in-

terrelated set of variations in a complex of interrelated dependent variables, things to be explained about why some parts of an organization are different from other parts of the same organization, and also why some organizations are different from other organizations. We have picked out the dependent variable complexes according to our argument about what sort of uncertainty they are responsive to.

Restructuring Research on Organizations

The conception outlined above about what organizational analysis ought to be about has some implications for the practice of organizational research. Let me summarize them under three headings: (1) units of analysis, (2) types of variables, and (3) forms of theory.

Units of Analysis

By units of analysis I mean those units on which data are typically collected. That is, should one collect data on individuals, departments, transactions, organizations, systems of organizations, or what? The central criterion for determining appropriate units of analysis is that, with respect to the theory at stake, they should be strongly causally connected internally. The units of analysis have to be chosen so that they are the place where causes are connected to effects—functions to needs, decisions to situations, flows to stocks, structures to environments. The central argument of this book is that units of analysis should be subparts of organizations that deal with distinctive sorts of uncertainties, that are responsible for securing effective responses to different sorts of news. Whole organizations may sometimes be proxies for the right units of analysis when they are dominated by a single type of information-processing system, as when universities are dominated by departments in various branches of scholarship. But even in this case one gets stronger and cleaner results if one distinguishes faculties from building and grounds departments.

Ordinarily organizations themselves, because they have had to build different substructures to deal with different sorts of uncertainty, will themselves provide guides to the right units. Universities will tend to locate building and grounds far from the faculty in the administrative system, but will tend to locate sociology moderately near to physics, because estimating the reputation of a physicist is the same general sort of problem as estimating the reputation of a sociologist. One has to go to a much higher level to find a common superior for building and grounds and sociology than to find a common superior for sociology and physics.

Quite often, whatever is called a "department" in an organization will be the right unit of analysis; but if the uncertainty in question has to do with investments or other matters with a longer time horizon, divisions or subsidiaries may be the right units. But since we have been arguing that features of the social organization of news processing are to be the main variables, we do not want to have units that cannot have social organization with a high degree of continuity through time. We therefore do not want to have either individuals or decisions as the central units of analysis.

Since in the theory a social structure relates decisions to news about uncertainties, the units chosen have to be those that tie news to decisions. Consequently, we need to so choose the units that they have authority (with review, to be sure) over a class of decisions that is served by a common news-collecting structure. Since authority is very generally divided in much the same way specialized departments are distinguished from each other (we rarely find departments that do not make—or at least recommend—important decisions), this criterion usually coincides with the one that tells us to choose as units of analysis specialized subunits. But it draws attention to the volume of the flow of information, short periods between transmissions of information, and the like, as a criterion for what is a unit of analysis. Thus, if it is true that some drug companies have a "vice president in charge of going to jail," who is responsible for everything affecting the quality of drugs but gets very little information about those things and makes very few decisions about them (John Braithewaite, personal communication), we would not want the different departments for which he or she is "responsible" to be a unit of analysis for our theory. His job is to go to jail when low quality is to be punished by the courts, not to unify quality control.

Units of analysis are fundamental aspects of the strategy of scientific inquiry and scientific theorizing. If people do not typically look at the units that connect a given kind of uncertainty to decisions, their theory is unlikely to be about such connections. Since our units do not have the natural skins that serve as boundaries around the units of analysis in much of psychology, and do not necessarily have legal existence as separate organizations as the units in ecological organizational theory have, we have to tell how to recognize those units in some detail. That makes the definition of units of analysis itself into a theoretical question. Our suggestion that one normally will not go too far wrong by using whatever is called a "department" as a unit of analysis is merely an empirical convenience. But that empirical convenience comes about because of an observation that connects it to the core of our theory—that usually departments connect news about a distinctive sort of uncertainty to the authority required to

make decisions, just as a thing with a skin is a unit in psychology because it happens that things with skins are what connect motives to individual actions.

Types of Variables

The argument of this book implies that we need to look at two broad types of variables, one describing the variations in the kinds of uncertainty that affect the units of analysis, another describing variations in the information and decision structure of the parts of the organization that form the units. For example, the service provided by a university faculty member is more valuable and has a higher price if the university has a good reputation, a fact that describes one type of uncertainty, about how to estimate reputation, while customer satisfaction and the capacity to get the job done right are more important in a service like that of barbers or beauticians, a fact that describes a different sort of uncertainty, about adequacy of competence.

It is variations in this uncertainty that gets universities into the problem of estimating both the present and the future scholarly reputation in contracting for their faculty, while no such problems arise in contracting for barbers or beauticians. The complicated structure of peer evaluation, and the system of paying faculty members for research work that is monitored not inside the university but in the researcher's own field, reflect (or so we have argued) the fact that it is uncertainty of reputation that determines the price a university can charge for its teaching. We therefore expect peer review when such renting of reputations is rational, because news of reputations cannot reliably be collected by other means (of course, it is not all that reliable when collected by these means either). Similarly, hospitals that depend on referrals to their physicians by primary-care physicians or by other hospitals should be expected to use peer review and investment in future reputations of their staff in much the way universities do. Of course, such hospitals are often university teaching hospitals and so find it easy to adopt variations of usual university personnel procedures. But, to return to the point made above about units of analysis, we will not expect much peer review in the hiring of nurses in that hospital, or in the building and grounds department of the university, because neither service becomes much more valuable by increasing the reputation of the service givers.

We have given many examples of such variables: whether or not the structure of the program governing a worker's handling of uncertainty is analogous to a batch program or to an interactive program, and the corresponding skill level of the worker (higher for interactive, lower for

batch; see Chapter 2); whether the accounts have to provide for comparisons among firms by outsiders and honesty of reporting to those outsiders rather than materials to analyze cost reduction projects, and the corresponding standard codes versus specialized detail and rigidity versus flexibility of the accounting structures (Chapter 3); whether a firm is in several markets or a single market (in a sense described in considerable detail in Chapter 4), and the corresponding divisionalization; whether the subpart of the organization is trying to maintain the monopoly position derived from having introduced a product innovation, and the corresponding network structure connecting users to the marketing department and the close intergration of engineering with marketing and manufacturing (Chapter 5). These examples suggest the generativity or fruitfulness of the theoretical approach. Many more examples are scattered throughout the book.

Forms of Theory

The definition of the variables and units of analysis involve a simultaneous orientation of the theory to three main domains of fact: the sources of uncertainty outside the organization; the organizational objectives that make the uncertainty important to the organization; and the volume, error, and bias of the flow of information about the uncertainty. The theory connects the larger social structure and the available productive technology to the microstructure of organizations. But it does not connect every part of the larger structure to every micropart of the organizational structure.

The crucial indicator that some uncertainty outside the organization is shaping a part of an organization is that information or news about that uncertainty is flowing through that part and being reshaped by that part into such a form that it can serve as the basis of a decision. The citations to a faculty member's work in the scholarly literature are processed into an overall ranking of the impact that scholar has on his or her field, as required by the dean's office, that is, departments must demonstrate the distinction of a given faculty member, as compared with alternative candidates who might be hired instead. This indicates that the information in the scientific or scholarly community shapes the personnel process of the university. Similarly, the fact that the well plan for a given well is shaped in detail by the well reports for neighboring wells is what indicates that drillers care a lot about the information from neighboring drill crews and drilling engineers, and makes it less surprising that they stamp Bullshit! on information from purchasing on how they might save money buying spare parts.

The theory, then, is about a flow of interactions between the environment and the subpart of the organization. Further, the argument is that the main thing that is going on in those interactions, the part that matters, is what will shape the main outlines of the subpart's structure. Sometimes the theory will be about variations in what matters in the interaction. For example, the argument of Chapter 5 is that the things that matter when one is trying to preserve the monopoly advantage got from a product innovation are quite different from those that matter when one is selling competitive products in a straightforward way. Sometimes the theory will be about variations in the temporal aspects of the interaction flow, as when the information for the general office of a multidivisional firm is abstracted into longer-period measures of performance, because the temporal structure of investment and return is slower than the temporal structure of market variations and response (see Chapter 4). Sometimes the theory will be about variations in the degree to which one has unanalyzed information indicating that everything seems to be going all right, as when personnel systems make great use of seniority criteria (Chapter 7).

But the basic presumption of all the subvarieties of the theory is that it is massive flows of interaction between an organizational part and the uncertainties of the environment that shape organizational structure. All the theoretical structures, then, are not so much about critical events as they are about flows of interaction, flows of information about uncertainty, flows of impacts of uncertainty on important continuing objectives, flows of outcomes that show that things are (or are not) being handled by the system. The overall theory of this book, then, is about all the different sorts of things a flow of interaction between an organization and its environment brings in, and how that flow of interaction affects the structure of work flow in different subparts of the organization.