rational decision making in business organizations

Rational Decision Making inBusiness Organizations

By HERBERT A. SIMON*

In the opening words of his Principles,Alfred Marshall proclaimed economics to bea psychological science:

Political Economy or Economics is astudy of mankind in the ordinary busi-ness of life; it examines that part ofindividual and social action which ismost closely connected with the attain-ment and with the use of the materialrequisites of wellbeing.

Thus it is on the one side a study ofwealth; and on the other, and moreimportant side, a part of the study ofman. For man's character has beenmoulded by his every-day work, and thematerial resources which he therebyprocures, more than by any otherinfluence unless it be that of his religiousideals.

In its actual development, however, eco-nomic science has focused on just one aspectof man's character, his reason, and particu-larly on the application of that reason toproblems of allocation in the face of scarcity.Still, modern definitions of the economicsciences, whether phrased in terms of allocat-ing scarce resources or in terms of rationaldecision making, mark out a vast domain forconquest and settlement. In recent years therehas been considerable exploration by econo-mists even of parts of this domain that werethought traditionally to belong to the disci-plines of political science, sociology, andpsychology.

*Carnegie-Mellon University. This article is thelecture Herbert Simon delivered in Stockholm, Sweden,December 8, 1978, when he received the Nobel Prize inEconomic Science. The article is copyright © the NobelFoundation 1978. It is published here with the permissionof the Nobel Foundation.

The author is indebted to Albert Ando, Otto A. Davis,and Benjamin M. Friedman for valuable comments on anearlier draft of this paper.

L Decision Theory as Ek:ononiic Science

The density of settlement of economistsover the whole empire of economic science isvery uneven, with a few areas of modest sizeholding the bulk of the population. Theeconomic Heartland is the normative study ofthe international and national economies andtheir markets, with its triple main concerns offull employment of resources, the eflicientallocation of resources, and equity in distribu-tion of the economic product. Instead of theambiguous and over-general term "econom-ics," I will use "political economy" to desig-nate this Heartland, and "economic sciences"to denote the whole empire, including its mostremote colonies. Our principal concern in thispaper will be with the important colonialterritory known as decision theory. I will havesomething to say about its normative anddescriptive aspects, and particularly about itsapplications to the theory of the firm. It isthrough the latter topic that the discussionwill be linked back to the Heartland ofpolitical economy.

Underpinning the corpus of policy-orientednormative economics, there is, of course, animpressive body of descriptive or "positive"theory which rivals in its mathematicalbeauty and elegance some of the finest theo-ries in the physical sciences. As examples Ineed only remind you of Walrasian generalequilibrium theories and ' their moderndescendants in the works of Henry Schultz,Samuelson, Hicks, and others; or the subtleand impressive body of theory created byArrow, Hurwicz, Debreu, Malinvaud, andtheir colleagues showing the equivalence,under certain conditions, of competitive equi-librium with Pareto optimality.

The relevance of some of the more refinedparts of this work to the real world can be,and has been, questioned. Perhaps some ofthese intellectual mountains have been

493

494 THE AMERICAN ECONOMIC REVIEW SEPTEMBER 1979

climbed simply because they were there—because of the sheer challenge and joy ofscaling them. That is as it should be in anyhuman scientific or artistic effort. But regard-less of the motives of the climbers, regardlessof real world veridicality, there is no questionbut that positive political economy has beenstrongly shaped by the demands of economicpolicy for advice on basic public issues.

This too is as it should be. It is a vulgarfallacy to suppose that scientific inquirycannot be fundamental if it threatens tobecome useful, or if it arises in response toproblems posed by the everyday world. Thereal world, in fact, is perhaps the most fertileof all sources of good research questionscalling for basic scientific inquiry.

A. Decision Theory in the Service ofPolitical Economy

There is, however, a converse fallacy thatdeserves equal condemnation: the fallacy ofsupposing that fundamental inquiry is worthpursuing only if its relevance to questions ofpolicy is immediate and obvious. In thecontemporary world, this fallacy is perhapsnot widely accepted, at least as far as thenatural sciences are concerned. We have nowlived through three centuries or more ofvigorous and highly successful inquiry intothe laws of nature. Much of that inquiry hasbeen driven by the simple urge to understand,to find the beauty of order hidden in complex-ity. Time and again, we have found the "idle"truths arrived at through the process ofinquiry to be of the greatest moment forpractical human affairs. I need not take timehere to argue the point. Scientists know it,engineers and physicians know it, congress-men and members of parliaments know it, theman on the street knows it.

But I am not sure that this truth is aswidely known in economics as it ought to be. Icannot otherwise explain the rather weak andbackward development of the descriptivetheory of decision making including thetheory of the firm, the sparse and scatteredsettlement of its terrain, and the fact thatmany, if not most, of its investigators aredrawn from outside economics—from sociolo-

gy, from psychology, and from politicalscience. Respected and distinguished figuresin economics—Edward Mason, Fritz Mach-lup, and Milton Friedman, for example—have placed it outside the Pale (more accu-rately, have placed economics outside itsPale), and have offered it full autonomyprovided that it did not claim close kinshipwith genuine economic inquiry.

Thus, Mason, commenting on Papan-dreou's 1952 survey of research on the behav-ioral theory of the firm, mused aloud:

. . . has the contribution of this litera-ture to economic analysis really been alarge one? . . . The writer of this critiquemust confess a lack of confidence in themarked superiority, for purposes ofeconomic analysis, of this newer conceptof the firm, over the older conception ofthe entrepreneur, [pp. 221-22]

And, in a similar vein, Friedman sums uphis celebrated polemic against realism intheory:

Complete "realism" is clearly unat-tainable, and the question whether atheory is realistic "enough" can besettled only by seeing whether it yieldspredictions that are good enough for thepurpose in hand or that are better thanpredictions from alternative theories.

[p. 41, emphasis added]

The "purpose in hand" that is implicit inboth of these quotations is providing decision-theoretic foundations for positive, and thenfor normative, political economy. In the viewsof Mason and Friedman, fundamental inquiryinto rational human behavior in the context ofbusiness organizations is simply not (by defi-nition) economics—that is to say, politicaleconomy—unless it contributes in a majorway to that purpose. This is sometimes eveninterpreted to mean that economic theories ofdecision making are not falsified in any inter-esting or relevant sense when their empiricalpredictions of microphenomena are found tobe grossly incompatible with the observeddata. Such theories, we are told, are stillrealistic "enough" provided that they do notcontradict aggregate observations of concern

VOL. 69 NO. 4 SIMON: RATIONAL DECISION MAKING 495

to political economy. Thus economists whoare zealous in insisting that economic actorsmaximize turn around and become satisficerswhen the evaluation of their own theories isconcerned, They believe that businessmenmaximize, but they know that economic theo-rists satisfice.

The application of the principle of satisfic-ing to theories is sometimes defended as anapplication of Occam's Razor: accept thesimplest theory that works.' But Occam'sRazor has a double edge. Succinctness ofstatement is not the only measure of a theo-ry's simplicity. Occam understood his rule asrecommending theories that make no moreassumptions than necessary to account for thephenomena {Essentia non sunt multiplicandapraeter necessitatem). A theory of profit orutility maximization can be stated morebriefly than a satisficing theory of the sort Ishall discuss later. But the former makesmuch stronger assumptions than the latterabout the human cognitive system. Hence, inthe case before us, the two edges of the razor,cut in opposite directions.

In whichever way we interpret Occam'sprinciple, parsimony can be only a secondaryconsideration in choosing between theories,unless those theories make identical predic-tions. Hence, we must come back to a consid-eration of the phenomena that positive deci-sion theory is supposed to handle. These mayinclude both phenomena at the microscopiclevel of the decision-making agents, or aggre-gative phenomena of concern to politicaleconomy.

'The phrase "that works" refutes, out of hand, Fried-man's celebrated paean of praise for lack of realism inassumptions. Consider his example of falling bodies(pp. 16-19). His valid point is that it is advantageous touse the simple law, ignoring air resistance, when it gives a"good enough" approximation. But of course the condi-tions under which it gives a good approximation are notat all the conditions under which it is unrealistic or a"wildly inaccurate descriptive representation of reality."We can use it to predict the path of a body falling in avacuum, but not the path of one falling through theEarth's atmosphere, I cannot in this brief space mention,much less discuss, all of the numerous logical fallaciesthat can be found in Friedman's 40-page essay. Foradditional criticism, see Simon (1963) and Samuelson(1963),

B. Decision Theory Pursued for itsIntrinsic Interest

Of course the definition of the word "eco-nomics" is not important. Like HumptyDumpty, we can make words mean anythingwe want them to mean. But the professionaltraining and range of concern of economistsdoes have importance. Acceptance of thenarrow view that economics is concerned onlywith the aggregative phenomena of politicaleconomy defines away a whole rich domain ofrational human behavior as inappropriate foreconomic research.

I do not wish to appear to be admitting thatthe behavioral theory of the firm has beenirrelevant to the construction of politicaleconomy. I will have more to say about itsrelevance in a moment. My present argumentis counterfactual in form: even if there wereno present evidence of such relevance, humanbehavior in business firms constitutes a highlyinteresting body of empirical phenomena thatcalls out for explanation as do all bodies ofphenomena. And if we may extrapolate fromthe history of the other sciences, there is everyreason to expect that as explanations emerge,relevance for important areas of practicalapplication will not be long delayed.

It has sometimes been implied (Friedman,p. 14) that the correctness of the assumptionsof rational behavior underlying the classicaltheory of the firm is not merely irrelevant, butis not even empirically testable in any directway, the only valid test being whether theseassumptions lead to tolerably correct predic-tions at the macroscopic level. That would betrue, of course, if we had no microscopes, sothat the micro-level behavior was not directlyobservable. But we do have microscopes.There are many techniques for observingdecision-making behavior, even at second-by-second intervals if that is wanted. In test-ing our economic theories, we do not have todepend on the rough aggregate time-seriesthat are the main grist for the econometricmill, or even upon company financial state-ments.

The classical theories of economic decisionmaking and of the business firm make veryspecific testable predictions about the con-


Crete behavior of decision-making agents.Behavioral theories make quite differentpredictions. Since these predictions can betested directly by observation, either theory(or both) may be falsified as readily whensuch predictions fail as when predictionsabout aggregate phenomena are in error.

C. Aggregative Tests of Decision Theory:Marginalism

If some economists have erroneouslysupposed that micro-economic theory canonly be tested by its predictions of aggregatephenomena, we should avoid the converseerror of supposing that aggregate phenomenaare irrelevant to testing decision theory. Inparticular, are there important, empiricallyverified, aggregate predictions that followfrom the theory of perfect rationality but thatdo not follow from behavioral theories ofrationality?

The classical theory of omniscient rational-ity is strikingly simple and beautiful. More-over, it allows us to predict (correctly or not)human behavior without stirring out of ourarmchairs to observe what such behavior islike. All the predictive power comes fromcharacterizing the shape of the environmentin which the behavior takes place. The envi-ronment, combined with the assumptions ofperfect rationality, fully determines thebehavior. Behavioral theories of rationalchoice—theories of bounded rationality—donot have this kind of simplicity. But, by wayof compensation, their assumptions abouthuman capabilities are far weaker than thoseof the classical theory. Thus, they makemodest and realistic demands on the knowl-edge and computational abilities of thehuman agents, but they also fail to predictthat those agents will equate costs and returnsat the margin.

D. Have the Marginalist PredictionsBeen Tested?

A number of empirical phenomena havebeen cited as providing more or less conclu-sive support for the classical theory of thefirm as against its behavioral competitors (seeDale Jorgensen and Calvin Siebert). But

there are no direct observations that individu-als or firms do actually equate marginal costsand revenues. The empirically verified conse-quences of the classical theory are almostalways weaker than this. Let us look at four ofthe most important of them: the fact thatdemand curves generally have negativeslopes; the fact that fitted Cobb-Douglasfunctions are approximately homogeneous ofthe first degree; the fact of decreasing returnsto scale; and the fact that executive salariesvary with the logarithm of company size. Arethese indeed facts? And does the evidencesupport a maximizing theory against a satis-ficing theory?

Negatively Sloping Demand Curves.Evidence that consumers actually distributetheir purchases in such a way as to maximizetheir utilities, and hence to equate marginalutilities, is nonexistent. What the empiricaldata do confirm is that demand curves gener-ally have negative slopes. (Even this "ob-vious" fact is tricky to verify, as HenrySchultz showed long years ago.) But nega-tively sloping demand curves could resultfrom a wide range of behaviors satisfying theassumptions of bounded rationality ratherthan those of utility maximization. GaryBecker, who can scarcely be regarded as ahostile witness for the classical theory, statesthe case very well:

Economists have long been aware thatsome changes in the feasible or opportu-nity sets of households would lead to thesame response regardless of the decisionrule used. For example, a decrease inreal income necessarily decreases theamount spent on at least one commodi-ty. . . It has seldom been realized, how-ever, that the change in opportunitiesresulting from a change in relative pricesalso tends to produce a systematicresponse, regardless of the decision rule.In particular, the fundamental theoremof traditional theory—that demandcurves are negatively inclined—largelyresults from the change in opportunitiesalone and is largely independent of thedecision rule. [p. 4]

Later, Becker is even more explicit, saying,"Not only utility maximization but also manyother decision rules, incorporating a wide


variety of irrational behavior, lead to nega-tively inclined demand curves because of theeffect of a change in prices on opportunities"(p. 5).^

First-Degree Homogeneity of Pro-duction Functions. Another example of anobserved phenomenon for which the classicalassumptions provide sufficient, but not neces-sary, conditions is the equality betweenlabor's share of product and the exponent ofthe labor factor in fitted Cobb-Douglas pro-duction functions (see Simon and FerdinandLevy). Fitted Cobb-Douglas functions arehomogeneous, generally of degree close tounity and with a labor exponent of about theright magnitude. These findings, however,cannot be taken as strong evidence for theclassical theory, for the identical results canreadily be produced by mistakenly fitting aCobb-Douglas function to data that were infact generated by a linear accounting identity(value of goods equals labor cost plus capitalcost), (see E. H. Phelps-Brown). The samecomment applies to the SMAC productionfunction (see Richard Cyert and Simon).Hence, the empirical findings do not allow usto draw any particular conclusions about therelative plausibility of classical and behav-ioral theories, both of which are equallycompatible with the data.

The Long-Run Cost Curve. Somewhatdifferent is the case of the firm's long-run costcurve, which classical theory requires to beU shaped if competitive equilibrium is to bestable. Theories of bounded rationality do notpredict this—fortunately, for the observeddata make it exceedingly doubtful that thecost curves are in fact generally U shaped.The evidence for many industries shows costsat the high-scale ends of the curves to beessentially constant or even declining (seeAlan Walters). This finding is compatiblewith stochastic models of business firmgrowth and size (see Y. Ijiri and Simon), butnot with the static equilibrium model ofclassical theory.

Executive Salaries. Average salaries of

'In a footnote, Becker indicates that he denotes asirrational "[A]ny deviation from utility maximization."Thus, what I have called "bounded rationality" is "irra-tionality" in Becker's terminology.

top corporate executives grow with the loga-rithm of corporate size (see David Roberts).This finding has been derived from theassumptions of the classical theory of profitmaximization only with the help of veryparticular ad hoc assumptions about thedistribution of managerial ability (see RobertLucas, 1978). The observed relation isimplied by a simple behavioral theory thatassumes only that there is a single, culturallydetermined, parameter which fixes the aver-age ratio of the salaries of managers to thesalaries of their immediate subordinates (seeSimon, 1957). In the case of the executivesalary data, the behavioral model thatexplains the observations is substantiallymore parsimonious (in terms of assumptionsabout exogenous variables) than the classicalmodel that explains the same observations.

Summary: Phenomena that Fail toDiscriminate. It would take a much moreextensive review than is provided here toestablish the point conclusively, but I believeit is the case that specific phenomena requir-ing a theory of utility or profit maximizationfor their explanation rather than a theory ofbounded rationality simply have not beenobserved in aggregate data. In fact, as my lasttwo examples indicate, it is the classicalrather than the behavioral form of the theorythat faces real difficulties in handling some ofthe empirical observations.

Failures of Classical Theory. It maywell be that classical theory can be patchedup sufficiently to handle a wide range ofsituations where uncertainty and outguessingphenomena do not play a central role—thatis, to handle the behavior of economies thatare relatively stable and not too distant from acompetitive equilibrium. However, a strongpositive case for replacing the classical theoryby a model of bounded rationality begins toemerge when we examine situations involvingdecision making under uncertainty andimperfect competition. These situations theclassical theory was never designed to handle,and has never handled satisfactorily. Statisti-cal decision theory employing the idea ofsubjective expected utility, on the one hand,and game theory, on the other, have contrib-uted enormous conceptual clarification tothese kinds of situations without providing


satisfactory descriptions of actual humanbehavior, or even, for most cases, normativetheories that are actually usable in the face ofthe limited computational powers of men andcomputers.

I.shall have more to say later about thepositive case for a descriptive theory ofbounded rationality, but I would like to turnfirst to another territory within economicscience that has gained rapidly in populationsince World War II, the domain of normativedecision theory.

E. Normative Decision Theory

Decision theory can be pursued not only forthe purposes of building foundations for polit-ical economy, or of understanding andexplaining phenomena that are in themselvesintrinsically interesting, but also for thepurpose of offering direct advice to businessand governmental decision makers. Forreasons not clear to me, this territory was verysparsely settled prior to World War II. Suchinhabitants as it had were mainly industrialengineers, students of public administration,and specialists in business functions, none ofwhom especially identified themselves withthe economic sciences. Prominent pioneersincluded the mathematician, Charles Bab-bage, inventor of the digital computer, theengineer, Frederick Taylor, and the adminis-trator, Henri Fayol.

During World War II, this territory,almost abandoned, was rediscovered by scien-tists, mathematicians, and statisticians con-cerned with military management and logis-tics, and was renamed "operations research"or "operations analysis." So remote were theoperations researchers from the social sciencecommunity that economists wishing to enterthe territory had to establish their own colony,which they called "management science."The two professional organizations thusengendered still retain their separate identi-ties, though they are now amicably federatedin a number of common endeavors.

Optimization techniques were transportedinto management science from economics,and new optimization techniques, notablylinear programming, were invented and devel-

oped, the names of Dantzig, Kantorovich, andKoopmaris being prominent in the earlydevelopment of that tool.

Now the salient characteristic of the deci-sion tools employed in management science isthat they have to be capable of actuallymaking or recommending decisions, taking astheir inputs the kinds of empirical data thatare available in the real world, and perform-ing only such computations as can reasonablybe performed by existing desk calculators or,a little later, electronic computers. For thesedomains, idealized models of optimizingentrepreneurs, equipped with complete cer-tainty about the world—or, at worst, havingfull probability distributions for uncertainevents—are of little use. Models have to befashioned with an eye to practical computa-bility, no matter how severe the approxima-tions and simplifications that are therebyimposed on them.

Model construction under these stringentconditions has taken two directions. The firstis to retain optimization, but to simplifysufficiently so that the optimum (in thesimplified world!) is computable. The secondis to construct satisficing models that providegood enough decisions with reasonable costsof computation. By giving up optimization, aricher set of properties of the real world canbe retained in the models. Stated otherwise,decision makers can satisfice either by findingoptimum solutions for a simplified world, orby finding satisfactory solutions for a morerealistic world. Neither approach, in general,dominates the other, and both have continuedto co-exist in the world of managementscience.

Thus, the body of theory that has developedin management science shares with the bodyof theory in descriptive decision theory acentral concern with the ways in which deci-sions are made, and not just with the decisionoutcomes. As I have suggested elsewhere(1978b), these are theories of how to deciderather than theories of what to decide.

Let me cite one example, from work inwhich I participated, of how model buildingin normative economics is shaped by compu-tational considerations (see Charles Holt,Franco Modigliani, John Muth, and Simon).


In face of uncertain and fluctuating pro-duction demands, a company can smooth andstabilize its production and employment levelsat the cost of holding buffer inventories. Whatkind of decision rule will secure a reasonablebalance of costs? Formally, we are faced witha dynamic programming problem, and thesegenerally pose formidable and often intoler-able computational burdens for their solu-tion.

One way out of this difficulty is to seek aspecial case of the problem that will becomputationally tractable. If we assume thecost functions facing the company all to bequadratic in form, the optimal decision rulewill then be a linear function of the decisionvariables, which can readily be computed interms of the cost parameters. Equally impor-tant, under uncertainty about future sales,only the expected values, and not the highermoments, of the probability distributionsenter into the decision rule (Simon, 1956b).Hence the assumption of quadratic costsreduces the original problem to one that isreadily solved. Of course the solution, thoughit provides optimal decisions for the simplifiedworld of our assumptions, provides, at best,satisfactory solutions for the real-world deci-sion problem that the quadratic functionapproximates. In-principle, unattainable op-timization is sacrificed for in-practice, attain-able satisfaction.

If human decision makers are as rational astheir limited computational capabilities andtheir incomplete information permit them tobe, then there will be a close relation betweennormative and descriptive decision theory.Both areas of inquiry are concerned primarilywith procedural rather than substantiverationality (Simon, 1978a). As new mathe-matical tools for computing optimal and satis-factory decisions are discovered, and ascomputers become more and more powerful,the recommendations of normative decisiontheory will change. But as the new recommen-dations are diffused, the actual, observed,practice of decision making in business firmswill change also. And these changes may havemacro-economic consequences. For example,there is some agreement that average inven-tory holdings of American firms have been

reduced significantly by the introduction offormal procedures for calculating reorderpoints and quantities.

II. Characterizing Bounded Rationality

The principal forerunner of a behavioraltheory of the firm is the tradition usuallycalled Institutionalism. It is not clear that allof the writings, European and American,usually lumped under this rubric have muchin common, or that their authors would agreewith each other's views. At best, they share aconviction that economic theory must bereformulated to take account of the social andlegal structures amidst which market transac-tions are carried out. Today, we even find avigorous development within economics thatseeks to achieve institutionalist goals withinthe context of neoclassical price theory. I willhave more to say about that a little later. .

The name of John R. Commons is promi-nent—perhaps the most prominent—amongAmerican Institutionalists. Commons' diffi-cult writings (for example. InstitutionalEconomics) borrow their language heavilyfrom the law, and seek to use the transactionas their basic unit of behavior. I will notundertake to review Commons' ideas here,but simply remark that they provided me withmany insights in my initial studies of organi-zational decision making (see my Adminis-trative Behavior, p. 136).

Commons also had a substantial influenceon the thinking of Chester I. Barnard, anintellectually curious business executive whodistilled from his experience as president ofthe New Jersey Bell Telephone Company,and as executive of other business, govern-mental, and nonprofit organizations, a pro-found book on decision making titled TheFunctions of the Executive. Barnard pro-posed original theories, which have stood upwell under empirical scrutiny, of the nature ofthe authority mechanism in organizations,and of the motivational bases for employeeacceptance of organizational goals (the so-called "inducements-contributions" theory);and he provided a realistic description oforganizational decision making, which hecharacterized as "opportunistic." The numer-


ous references to Barnard's work in Adminis-trative Behavior attest, though inadequately,to the impact he had on my own thinkingabout organizations.

A. In Search of a Descriptive Theory

In 1934-35, in the course of a field study ofthe administration of public recreationalfacilities in Milwaukee, which were managedjointly by the school board and the city publicworks department, I encountered a puzzlingphenomenon. Although the heads of the twoagencies appeared to agree as to the objec-tives of the recreation program, and did notappear to be competing for empire, there wascontinual disagreement and tension betweenthem with respect to the allocation of fundsbetween physical maintenance, on the onehand, and play supervision on the other. Whydid they not, as my economics bookssuggested, simply balance off the marginalreturn of the one activity against that of theother?

Further exploration made it apparent thatthey didn't equate expenditures at the marginbecause, intellectually, they couldn't. Therewas no measurable production function fromwhich quantitative inferences about marginalproductivities could be drawn; and such qual-itative notions of a production function as thetwo managers possessed were mutuallyincompatible. To the public works adminis-trator, a playground was a physical facility,serving as a green oasis in the crowded graycity. To the recreation administrator, a play-ground was a social facility, where childrencould play together with adult help and guid-ance.

How can human beings make rational deci-sions in circumstances like these? How arethey to apply the marginal calculus? Or, if itdoes not apply, what do they substitute forit?

The phenomenon observed in Milwaukee isubiquitous in human decision making. Inorganization theory it is usually referred to assubgoal identification. When the goals of anorganization cannot be connected operation-ally with actions (when the production func-tion can't be formulated in concrete terms).

then decisions will be judged against subordi-nate goals that can be so connected. There isno unique determination of these subordinategoals. Their formulation will depend on theknowledge, experience, and organizationalenvironment of the decision maker. In theface of this ambiguity, the formulation canalso be influenced in subtle, and not so subtle,ways by his self-interest and power drives.

The phenomenon arises as frequently inindividual as in social decision making andproblem solving. Today, under the rubric ofproblem representation, it is a centralresearch interest of cognitive psychology.Given a particular environment of stimuli,and a particular background of previousknowledge, how will a person organize thiscomplex mass of information into a problemformulation that will facilitate his solutionefforts? How did Newton's experience of theapple, if he had one, get represented as aninstance of attraction of apple by Earth?

Phenomena like these provided the centraltheme for Administrative Behavior. Thatstudy represented "an attempt to constructtools useful in my own research in the field ofpublic administration." The product wasactually not so much a theory as prolegomenato a theory, stemming from the conviction"that decision making is the heart of adminis-tration, and that the vocabulary of adminis-trative theory must be derived from the logicand psychology of human choice." It was, ifyou please, an exercise in problem representa-tion.

On examination, the phenomenon ofsubgoal identification proved to be the visibletip of a very large iceberg. The shape of theiceberg is best appreciated by contrasting itwith classical models of rational choice. Theclassical model calls for knowledge of all thealternatives that are open to choice. It callsfor complete knowledge of, or ability tocompute, the consequences that will follow oneach of the alternatives. It calls for certaintyin the decision maker's present and futureevaluation of these consequences. It calls forthe ability to compare consequences, nomatter how diverse and heterogeneous, interms of some consistent measure of utility.The task, then, was to replace the classical


model with one that would describe howdecisions could be (and probably actuallywere) made when the alternatives of searchhad to be sought out, the consequences ofchoosing particular alternatives were onlyvery imperfectly known both because oflimited computational power and because ofuncertainty in the external world, and thedecision maker did not possess a general andconsistent utility function for comparing hete-rogeneous alternatives.

Several procedures of rather general appli-cability and wide use have been discoveredthat transform intractable decision problemsinto tractable ones. One procedure alreadymentioned is to look for satisfactory choicesinstead of optimal ones. Another is to replaceabstract, global goals with tangible subgoals,whose achievement can be observed andmeasured. A third is to divide up the decision-making task among many specialists, coordi-nating their work by means of a structure ofcommunications and authority relations. Allof these, and others, fit the general rubric of"bounded rationality," and it is now clear thatthe elaborate organizations that humanbeings have constructed in the modern worldto carry out the work of production andgovernment can only be understood asmachinery for coping with the limits of man'sabilities to comprehend and compute in theface of complexity and uncertainty.

This rather vague and general initialformulation of the idea of bounded rationalitycalled for elaboration in two directions:greater formalization of the theory, andempirical verification of its main claims.During the decade that followed the publica-tion of Administrative Behavior, substantialprogress was made in both directions, some ofit through the efforts of my colleagues andmyself, much of it by other research groupsthat shared the same Zeitgeist.

B. Empirical Studies

The principal source of empirical dataabout organizational decision making hasbeen straightforward "anthropological" fieldstudy, eliciting descriptions of decision-making procedures and observing the course

of specific decision-making episodes. Exam-ples are my study, with Guetzkow, Kozmet-sky, and Tyndall (1954), of the ways in whichaccounting data were used in decision makingin large corporations; and a series of studies,with Richard Cyert, James March, andothers, of specific nonprogrammed policydecisions in a number of different companies(see Cyert, Simon, and Donald Trow). Thelatter line of work was greatly developed andexpanded by Cyert and March and its theo-retical implications for economics explored intheir important work, A Behavioral Theoryof the Firm.

At about the same time, the fortuitousavailability of some data on businessmen'sperceptions of a problem situation describedin a business policy casebook enabled DeWittDearborn and me to demonstrate empiricallythe cognitive basis for identification withsubgoals, the phenomenon that had soimpressed me in the Milwaukee recreationstudy. The businessmen's perceptions of theprincipal problems facing the companydescribed in the case were mostly determinedby their own business experiences—sales andaccounting executives identified a sales prob-lem, manufacturing executives, a problem ofinternal organization.

Of course there is vastly more to be learnedand tested about organizational decisionmaking than can be dealt with in a handful ofstudies. Although many subsequent studieshave been carried out in Europe and theUnited States, this domain is still grosslyundercultivated (for references, see March,1965; E. Johnsen, 1968; G. Eliasson, 1976).Among the reasons for the relative neglect ofsuch studies, as contrasted, say, with labora-tory experiments in social psychology, is thatthey are extremely costly and time consum-ing, with a high grist-to-grain ratio, the meth-odology for carrying them out is primitive,and satisfactory access to decision-makingbehavior is hard to secure. This part ofeconomics has not yet acquired the habits ofpatience and persistence in the pursuit off"acts that is exemplified in other domains bythe work, say, of Simon Kuznets or of thearchitects of the MIT-SSRC-Penn economet-ric models.


C. Theoretical Inquiries

On the theoretical side, three questionsseemed especially to call for clarification:what are the circumstances under which anemployment relation will be preferred to someother form of contract as the arrangement forsecuring the performance of work; what is therelation between the classical theory of thefirm and theories of organizational equilib-rium first proposed by Chester Barnard; andwhat are the main characteristics of humanrational choice in situations where complexityprecludes omniscience?

The Employment Relation. A funda-mental characteristic of modern industrialsociety is that most work is performed, not byindividuals who produce products for sale, norby individual contractors, but by persons whohave accepted employment in a business firmand the authority relation with the employerthat employment entails. Acceptance ofauthority means willingness to permit one'sbehavior to be determined by the employer, atleast within some zone of indifference oracceptance. What is the advantage of thisarrangement over a contract for specifiedgoods or services? Why is so much of theworld's work performed in large, hierarchicorganizations?

Analysis showed (Simon, 1951) that acombination of two factors could account forpreference for the employment contract overother forms of contracts: uncertainty as towhich future behaviors would be advanta-geous to the employer, and a greater indiffer-ence of the employee as compared with theemployer (within the former's area of accep-tance) as to which of these behaviors hecarried out. When the secretary is hired, theemployer does not know what letters he willwant her to type, and the secretary has nogreat preference for typing one letter ratherthan another. The employment contractpermits the choice to be postponed until theuncertainty is resolved, with little cost to theemployee and great advantage to theemployer. The explanation is closely analo-gous to one Jacob Marschak had proposed forliquidity preference. Under conditions ofuncertainty it is advantageous to holdresources in liquid, flexible form.

Organizational Equilibrium. Barnardhad described the survival of organizations interms of the motivations that make theirparticipants (employees, investors, customers,suppliers) willing to remain in the system. InAdministrative Behavior, I had developedthis notion further into a motivational theoryof the balance between the inducements thatwere provided by organizations to their par-ticipants, and the contributions those partici-pants made to the organizations' resources.

A formalization of this theory (Simon,1952; 1953) showed its close affinity to theclassical theory of the firm, but with animportant and instructive difference. Incomparing the two theories, each inducement-contribution relation became a supply sched-ule for the firm. The survival conditionsbecame the conditions for positive profit. Butwhile the classical theory of the firm assumesthat all profits accrue to a particular set ofparticipants, the owners, the organizationtheory treats the surplus more symmetrically,and does not predict how it will be distributed.Hence the latter theory leaves room, underconditions of monopoly and imperfect compe-tition, for bargaining among the participants(for example, between labor and owners) forthe surplus. The survival conditions—positiveprofits rather than maximum profits—alsopermit a departure from the assumptions ofperfect rationality.

Mechanisms of Bounded Rationality. InAdministrative Behavior, bounded rationalityis largely characterized as a residual cate-gory—rationality is bounded when it fallsshort of omniscience. And the failures ofomniscience are largely failures of knowingall the alternatives, uncertainty about rele-vant exogenous events, and inability to calcu-late consequences. There was needed a morepositive and formal characterization of themechanisms of choice under conditions ofbounded rationality. Two papers (Simon,1955; 1956a) undertook first steps in thatdirection.

Two concepts are central to the characteri-zation: search and satisficing. If the alterna-tives for choice are not given initially to thedecision maker, then he must search for them.Hence, a theory of bounded rationality mustincorporate a theory of search. This idea was


later developed independently by GeorgeStigler in a very influential paper that took asits example of a decision situation thepurchase of a second-hand automobile.Stigler poured the search theory back into theold bottle of classical utility maximization,the cost of search being equated with itsmarginal return. In my 1956 paper, I haddemonstrated the same formal equivalence,using as my example a dynamic programmingformulation of the process of selling ahouse.

But utility maximization, as I showed, wasnot essential to the search scheme—fortu-nately, for it would have required the decisionmaker to be able to estimate the marginalcosts and returns of search in a decisionsituation that was already too complex for theexercise of global rationality. As an alterna-tive, one could postulate that the decisionmaker had formed some aspiration as to howgood an alternative he should find. As soon ashe discovered an alternative for choice meet-ing his level of aspiration, he would terminatethe search and choose that alternative. Icalled this mode of selection satisficing. Ithad its roots in the empirically based psycho-logical theories, due to Lewin and others, ofaspiration levels. As psychological inquiryhad shown, aspiration levels are not static, buttend to rise and fall in consonance withchanging experiences. In a benign environ-ment that provides many good alternatives,aspirations rise; in a harsher enviornment,they fall.

In long-run equilibrium it might even bethe case that choice with dynamically adapt-ing aspiration levels would be equivalent tooptimal choice, taking the costs of search intoaccount. But the important thing about thesearch and satisficing theory is that it showedhow choice could actually be made withreasonable amounts of calculation, and usingvery incomplete information, without theneed of performing the impossible—of carry-ing out this optimizing procedure.

D. Summary

Thus, by the middle 195O's, a theory ofbounded rationality had been proposed as analternative to classical omniscient rationality.

a significant number of empirical studies hadbeen carried out that showed actual businessdecision making to conform reasonably wellwith the assumptions of bounded rationalitybut not with the assumptions of perfectrationality, and key components of the the-ory—the nature of the authority and employ-ment relations, organizational equilibrium,and the mechanisms of search and satisfic-ing—had been elucidated formally. In theremaining parts of this paper, I should like totrace subsequent developments of decision-making theory, including developments com-petitive with the theory of bounded rational-ity, and then to comment on the implications(and potential implications) of the newdescriptive theory of decision for politicaleconomy.

III. The Neoclassical Revival

Peering forward from the late 195O's, itwould not have been unreasonable to predictthat theories of bounded rationality wouldsoon find a large place in the mainstream ofeconomic thought. Substantial progress hadbeen made in providing the theories withsome formal structure, and an increasingbody of empirical evidence showed them toprovide a far more veridical picture of deci-sion making in business organizations thandid the classical concepts of perfect rational-ity.

History has not followed any such simplecourse, even though many aspects of theZeitgeist were favorable to movement in thisdirection. During and after World War II, alarge number of academic economists wereexposed directly to business life, and hadmore or less extensive opportunities to observehow decisions were actually made in businessorganizations. Moreover, those who becameactive in the development of the new manage-ment science were faced with the necessity ofdeveloping decision-making procedures thatcould actually be applied in practical situa-tions. Surely these trends would be conduciveto moving the basic assumptions of economicrationality in the direction of greater real-ism.

But these were not the only things thatwere happening in economics in the postwar


period. First, there was a vigorous reactionthat sought to defend classical theory frombehavioralism on methodological grounds. Ihave already commented on these methodo-logical arguments in the first part of my talk.However deeply one may disagree with them,they were stated persuasively and are stillinfluential among academic economists.

Second, the rapid spread of mathematicalknowledge and competence in the economicsprofession permitted the classical theory,especially when combined with statisticaldecision theory and the theory of games dueto von Neumann and Morgenstern, to developto new heights of sophistication and elegance,and to expand to embrace, albeit in highlystylized form, some of the phenomena ofuncertainty and imperfect information. Theflowering of mathematical economics andeconometrics has provided two generations ofeconomic theorists with a vast garden offormal and technical problems that haveabsorbed their energies and postponedencounters with the inelegancies of the realworld.

If I sound mildly critical of these develop-ments, I should confess that I have also been apart of them, admire them, and would bedecidedly unhappy to return to the premathe-matical world they have replaced. Myconcern is that the economics profession hasexhibited some of the serial one-thing-at-a-time character of human rationality, andhas seemed sometimes to be unable to distri-bute its attention in a balanced fashion amongneoclassical theory, macroeconometrics, anddescriptive decision theory. As a result, not asmuch professional effort has been devoted tothe latter two, and especially the third, as onemight have hoped and expected. The Heart-land is more overpopulated than ever, whilerich lands in other parts of the empire gountilled.

A. Search and Information Transfer

Let me allude to just three of the ways inwhich classical theory has sought to cope withsome of its traditional limitations, and haseven sought to make the development of abehavioral theory, incorporating psychologi-cal assumptions, unnecessary. The first was to

introduce search and information transferexplicitly as economic activities, with asso-ciated costs and outputs, that could beinserted into the classical production func-tion. I have already referred to Stigler's 1961paper on the economics of information, andmy own venture in the same direction in the1956 essay cited earlier.

In theory of this genre, the decision makeris still an individual. A very important newdirection, in which decisions are made bygroups of individuals, in teams or organiza-tions, is the economic theory of teams devel-oped by Jacob Marschak and Roy Radner.Here we see genuine organizational phenom-ena—specialization of decision making as aconsequence of the costs of transmitting infor-mation—emerge from the rational calculus.Because the mathematical difficulties areformidable, the theory remains largely illus-trative and limited to very simple situations inminiature organizations. Nevertheless, it hasgreatly broadened our understanding of theeconomics of information.

In none of these theories—any more thanin statistical decision theory or the theory ofgames—is the assumption of perfect maximi-zation abandoned. Limits and costs of infor-mation are introduced, not as psychologicalcharacteristics of the decision maker, but aspart of his technological environment. Hence,the new theories do nothing to alleviate thecomputational complexities facing the deci-sion maker—do not see him coping with themby heroic approximation, simplifying andsatisficing, but simply magnify and multiplythem. Now he needs to compute not merelythe shapes of his supply and demand curves,but, in addition, the costs and benefits ofcomputing those shapes to greater accuracyas well. Hence, to some extent, the impressionthat these new theories deal with the hithertoignored phenomena of uncertainty and infor-mation transmission is illusory. For manyeconomists, however, the illusion has beenpersuasive.

B. Rational Expectations Theory

A second development in neoclassicaltheory on which I wish to comment is theso-called "rational expectations" theory.


There is a bit of historical irony surroundingits origins. I have already described themanagement science inquiry of Holt, Modig-liani, Muth, and myself that developed adynamic programming algorithm for thespecial (and easily computed) case ofquadratic cost functions. In this case, thedecision rules are linear, and the probabilitydistributions of future events can be replacedby their expected values, which serve ascertainty equivalents (see Simon, 1956; HenriTheil, 1957).

Muth imaginatively saw in this special casea paradigm for rational behavior under uncer-tainty. What to some of us in the HMMSresearch team was an approximating, satisfic-ing simplification, served for him as a majorline of defense for perfect rationality. He saidin his seminal 196i Econometrica article, "Itis sometimes argued that the assumption ofrationality in economics leads to theoriesinconsistent with, or inadequate to explain,observed phenomena, especially changes overt ime. . . Our hypothesis is based on exactlythe opposite point of view: that dynamiceconomic models do not assume enoughrationality" (p. 316).

The new increment of rationality thatMuth proposed was that "expectations, sincethey are informed predictions of futureevents, are essentially the same as the predic-tions of the relevant economic theory"(p. 316). He would cut the Gordian knot.Instead of dealing with uncertainty by elabo-rating the model of the decision process, hewould once and for all—if his hypothesis werecorrect—make process irrelevant. The subse-quent vigorous development of rational expec-tations theory, in the hands of Sargent, Lucas,Prescott, and others, is well known to mostreaders (see, for example, Lucas, 1975).

It is too early to render a final verdict onthe rational expectations theory. The issuewill ultimately be decided, as all scientificdebates should be, by a gradual winnowing ofthe empirical evidence, and that winnowingprocess has just begun. Meanwhile, certaingrave theoretical difficulties have alreadybeen noticed. As Muth himself has pointedout, it is rational (i.e., profit maximizing) touse the "rational expectations" decision ruleif the relevant cost equations are in fact

quadratic. I have suggested elsewhere(1978a) that it might therefore be lessmisleading to call the rule a "consistentexpectations" rule.

Perhaps even more important, Albert Andoand Benjamin Friedman (1978, 1979) haveshown that the policy implications of therational expectations rule are quite differentunder conditions where new informationcontinually becomes available to the system,structural changes occur, and the decisionmaker learns, than they are under steady-state conditions. For example, under the moredynamic conditions, monetary neutrality—which in general holds for the static consistentexpectations models—is no longer guaranteedfor any finite time horizon.

In the recent "revisionist" versions ofconsistent expectations theory, moreover,where account is taken of a changing environ-ment of information, various behavioralassumptions reappear to explain how expecta-tions are formed—what information decisionmakers will consider, and what they willignore. But unless these assumptions are to bemade on a wholly ad hoc and arbitrary basis,they create again the need for an explicit andvalid theory of the decision-making process(see Simon, 1958a; B. Friedman, 1979).

C. Statistical Decision Theory andGame Theory

Statistical decision theory and game theoryare two other important components of theneoclassical revival. The former addressesitself to the question of incorporating uncer-tainty (or more properly, risk) into the deci-sion-making models. It requires heroicassumptions about the information the deci-sion maker has concerning the probabilitydistributions of the relevant variables, andsimply increases by orders of magnitude thecomputational problems he faces.

Game theory addresses itself to the "out-guessing" problem that arises whenever aneconomic actor takes into account the possiblereactions to his own decisions of the otheractors. To my mind, the main product of thevery elegant apparatus of game theory hasbeen to demonstrate quite clearly that it isvirtually impossible to define an unambiguous


criterion of rationality for this class of situa-tions (or, what amounts to the same thing, adefinitive definition of the "solution" of agame). Hence, game theory has not broughtto the theories of oligopoly and imperfectcompetition the relief from their contradic-tions and complexities that was originallyhoped for it. Rather, it has shown that thesedifficulties are ineradicable. We may be ableto reach consensus that a certain criterion ofrationality is appropriate to a particulargame, but if someone challenges the consen-sus, preferring a different criterion, we willhave no logical basis for persuading him thathe is wrong.

D. Conclusion

Perhaps I have said enough about theneoclassical revival to suggest why it has beena highly attractive commodity in competitionwith the behavioral theories. To some econo-mists at least, it has held open the possibilityand hope that important questions that hadbeen troublesome for classical economicscould now be addressed without sacrifice ofthe central assumption of perfect rationality,and hence also with a maximum of a prioriinference and a minimum of tiresome grub-bing with empirical data. I have perhaps saidenough also with respect to the limitations ofthese new constructs to indicate why I do notbelieve that they solve the problems thatmotivated their development.

IV. Advances in the Behavioral Theory

Although they have played a muted role inthe total economic research activity duringthe past two decades, theories of boundedrationality and the behavioral theory of thebusiness firm have undergone steady develop-ment during that period. Since surveying thewhole body of work would be a major under-taking, I shall have to be satisfied here withsuggesting the flavor of the whole by citing afew samples of different kinds of importantresearch falling in this domain. Wheresurveys on particular topics have beenpublished, I will limit myself to references tothem.

First, there has been work in the psycholog-ical laboratory and the field to test whetherpeople in relatively simple choice situationsbehave as statistical decision theory (maximi-zation of expected utilities) say they do.Second, there has been extensive psychologi-cal research, in which Allen Newell and Ihave been heavily involved, to discover theactual microprocesses of human decisionmaking and problem solving. Third, therehave been numerous empirical observations—most of them in the form of "case studies"—of the actual processes of decision making inorganizational and business contexts. Fourth,there have been reformulations and exten-sions of the theory of the firm replacingclassical maximization with behavioral deci-sion postulates.

A. Utility Theory and Human Choice

The axiomatization of utility and probabil-ity after World War II and the revival ofBayesian statistics opened the way to testingempirically whether people behaved in choicesituations so as to maximize subjectiveexpected utility (SEU). In early studies, usingextremely simple choice situations, it ap-peared that perhaps they did. When evensmall complications were introduced into thesituations, wide departures of behavior fromthe predictions of SEU theory soon becameevident. Some of the most dramatic andconvincing empirical refutations of the theoryhave been reported by D. Kahneman and A.Tversky, who showed that under one set ofcircumstances, decision makers gave far toolittle weight to prior knowledge and basedtheir choices almost entirely on new evidence,while in other circumstances new evidencehad little influence on opinions alreadyformed. Equally large and striking departuresfrom the behavior predicted by the SEUtheories were found by Howard Kunreutherand his colleagues in their studies of individ-ual decisions to purchase or not to purchaseflood insurance. On the basis of these andother pieces of evidence, the conclusion seemsunavoidable that the SEU theory does notprovide a good prediction—not even a goodapproximation—of actual behavior.


Notice that the refutation of the theory hasto do with the substance of the decisions, andnot just the process by which they arereached. It is not that people do not gothrough the calculations that would berequired to reach the SEU decision—neoclas-sical thought has never claimed that they did.What has been shown is that they do not evenbehave as if they had carried out thosecalculations, and that result is a direct refuta-tion of the neoclassical assumptions.

B. Psychology of Problem Solving

The evidence on rational decision making islargely negative evidence, evidence of whatpeople do not do. In the past twenty years alarge body of positive evidence has also accu-mulated about the processes that people useto make difficult decisions and solve complexproblems. The body of theory that has beenbuilt up around this evidence is called infor-mation processing psychology, and is usuallyexpressed formally in computer programminglanguages. Newell and I have summed up ourown version of this theory in our book.Human Problem Solving, which is part of alarge and rapidly growing literature thatassumes an information processing frame-work and makes use of computer simulationas a central tool for expressing and testingtheories.

Information processing theories envisageproblem solving as involving very selectivesearch through problem spaces that are oftenimmense. Selectivity, based on rules of thumbor "heuristics," tends to guide the search intopromising regions, so that solutions willgenerally be found after search of only a tinypart of the total space. Satisficing criteriaterminate search when satisfactory problemsolutions have been found. Thus, these theo-ries of problem solving clearly fit within theframework of bounded rationality that I havebeen expounding here.

By now the empirical evidence for thisgeneral picture of the problem solving processis extensive. Most of the evidence pertains torelatively simple, puzzle-like situations of thesort that can be brought into the psychologi-cal laboratory for controlled study, but a

great deal has been learned, also, aboutprofessional level human tasks like makingmedical diagnoses, investing in portfolios ofstocks and bonds, and playing chess. In tasksof these kinds, the general search mechanismsoperate in a rich context of information storedin human long-term memory, but the generalorganization of the process is substantially thesame as for the simpler, more specific tasks.

At the present time, research in informa-tion processing psychology is proceeding inseveral directions. Exploration of profession-al level skills continues. A good deal of effortis now being devoted also to determining howinitial representations for new problems areacquired. Even in simple problem domains,the problem solver has much latitude in theway he formulates the problem space in whichhe will search, a finding that underlines againhow far the actual process is from a search fora uniquely determined optimum (see J. R.Hayes and Simon).

The main import for economic theory ofthe research in information processing psy-chology is to provide rather conclusive empir-ical evidence that the decision-making processin problem situations conforms closely to themodels of bounded rationality describedearlier. This finding implies, in turn, thatchoice is not determined uniquely by theobjective characteristics of the problem situa-tion, but depends also on the particularheuristic process that is used to reach thedecision. It would appear, therefore, that amodel of process is an essential component inany positive theory of decision making thatpurports to describe the real world, and thatthe neoclassical ambition of avoiding thenecessity for such a model is unrealizable(Simon, 1978a).

C. Organizational Decision Making

It would be desirable to have, in addition tothe evidence from the psychological researchjust described, empirical studies of theprocess of decision making in organizationalcontexts. The studies of individual problemsolving and decision making do not touch onthe many social-psychological factors thatenter into the decision process in organiza-


tions. A substantial number of investigationshave been carried out in the past twenty yearsof the decision-making process in organiza-tions, but they are not easily summarized.The difficulty is that most of these investiga-tions have taken the form of case studies ofspecific decisions or particular classes of deci-sions in individual organizations. To the bestof my knowledge, no good review of thisliterature has been published, so that it isdifficult even to locate and identify the studiesthat have been carried out.^ Nor have anysystematic methods been developed andtested for distilling out from these individualcase studies their implications for the generaltheory of the decision-making process.

The case studies of organizational decisionmaking, therefore, represent the naturalhistory stage of scientific inquiry. Theyprovide us with a multitude of facts about thedecision-making process—facts that are al-most uniformly consistent with the kind ofbehavioral model that has been proposedhere. But we do not yet know how to use thesefacts to test the model in any formal way. Nordo we quite know what to do with the obser-vation that the specific decision-makingprocedures used by organizations differ fromone organization to another, and within eachorganization, even from one situation toanother. We must not expect from these datageneralizations as neat and precise as thoseincorporated in neoclassical theory.

Perhaps the closest approach to a methodfor extracting theoretically relevant informa-tion from case studies is computer simulation.By converting empirical evidence about adecision-making process into a computerprogram, a path is opened both for testing theadequacy of the program mechanisms forexplaining the data, and for discovering thekey features of the program that account,qualitatively, for the interesting and impor-tant characteristics of its behavior. Examples

'For leads into the literature, see March and Simon;March; Johnsen; J. M. Dutton and W. H. Starbuck.However, there are large numbers of specific case studies,some of them carried out as thesis projects, someconcerned with particular fields of business application,which have never been recorded in these referencesources (for example, Eliasson, 1976).

of the use of this technique are G.P.E. Clark-son's simulation of the decision making of aninvestment trust officer, Cyert, E. A. Feigen-baum, and March's simulation of the historyof a duopoly, and C. P. Bonini's model of theeffects of accounting information and super-visory pressures in altering employee motiva-tions in a business firm. The simulation meth-odology is discussed from a variety of view-points in Dutton and Starbuck."

D. Theories of the Business Firm

The general features of bounded rational-ity—selective search, satisficing, and so on—have been taken as the starting points for anumber of attempts to build theories of thebusiness firm incorporating behavioral as-sumptions. Examples of such theories wouldinclude the theory of Cyert and March,already mentioned; William Baumol's theoryof sales maximization subject to minimumprofit constraints; Robin Marris' models offirms whose goals are stated in terms of ratesof growth; Harvey Leibenstein's theory of"X-inefficiency" that depresses productionbelow the theoretically attainable; JanosKornai's dichotomy between supply-drivenand demand-driven management; Oliver Wil-liamson's theory of transactional costs; theevolutionary models of Richard Nelson andSidney Winter (1973); Cyert and MorrisDeGroot's (1974) models incorporatingadaptive learning; Radner's (1975a,b) expli-cit satisficing models; and others.

Characterized in this way, there seems tobe little commonality among all of thesetheories and models, except that they departin one way or another from the classicalassumption of perfect rationality in firm deci-sion making. A closer look, however, and amore abstract description of their assump-tions, shows that they share several basiccharacteristics. Most of them depart from theassumption of profit maximization in theshort run, and replace it with an assumption

*In addition to simulations of the firm, there are veryinteresting and potentially important efforts to use simu-lation to build bridges directly from decision theory topolitical economy. See G. Orcutt and R. Caldwell-Wertheimer, and Eliasson (1978).

VOL 69 NO. 4 SIMON: RATIONAL DECISION MAKING 509

of goals defined in terms of targets—that is,they are to greater or lesser degree satisficingtheories. If they do retain maximizingassumptions, they contain some kind of mech-anism that prevents the maximum from beingattained, at least in the short run. In theCyert-March theory, and that of Leibenstein,this mechanism can be viewed as producing"organizational slack," the magnitude ofwhich may itself be a function of motivationaland environmental variables.

Finally, a number of these theories assumethat organizational learning takes place, sothat if the environment were stationary for asufficient length of time, the system equilib-rium would approach closer and closer to theclassical profit-maximizing equilibrium. Ofcourse they generally also assume that theenvironmental disturbances will generally belarge enough to prevent the classical solutionfrom being an adequate approximation to theactual behavior.

The presence of something like organiza-tional slack in a model of the business firmintroduces complexity in the firm's behaviorin the short run. Since the firm may operatevery far from any optimum, the slack servesas a buffer between the environment and thefirm's decisions. Responses to environmentalevents can no longer be predicted simply byanalyzing the "requirements of the situation,"but depend on the specific decision processesthat the firm employs. However well thischaracteristic of a business firm model corre-sponds to reality, it reduces the attractivenessof the model for many economists, who arereluctant to give up the process-independentpredictions of classical theory, and who do notfeel at home with the kind of empiricalinvestigation that is required for disclosingactual real world decision processes.

But there is another side to the matter. If,in the face of identical environmental condi-tions, different decision mechanisms canproduce different firm behaviors, this sensi-tivity of outcomes to process can have impor-tant consequences for analysis at the level ofmarkets and the economy. Political economy,whether descriptive or normative, cannotremain indifferent to this source of variabilityin response. At the very least it demands

that—before we draw policy conclusions fromour theories, and particularly before we act onthose policy conclusions—we carry out sensi-tivity analyses to test how far our conclusionswould be changed if we made differentassumptions about the decision mechanismsat the micro level.

If our conclusions are robust—if they arenot changed materially by substituting one oranother variant of the behavioral model forthe classical model—we will gain confidencein our predictions and recommendations; ifthe conclusions are sensitive to such substitu-tions, we will use them warily until we candetermine which micro theory is the correctone.

As reference to the literature cited earlierin this section will verify, our predictions ofthe operations of markets and of the economyare sensitive to our assumptions about mech-anisms at the level of decision processes.Moreover, the assumptions of the behavioraltheories are almost certainly closer to realitythan those of the classical theory. These twofacts, in combination, constitute a direct refu-tation of the argument that the unrealism ofthe assumptions of the classical theory isharmless. We cannot use the in vacua versionof the law of falling bodies to predict thesinking of a heavy body in molasses. Thepredictions of the classical and neoclassicaltheories and the policy recommendationsderived from them must be treated with thegreatest caution,

V, Conclusion

There is a saying in politics that "you can'tbeat something with nothing." You can'tdefeat a measure or a candidate simply bypointing to defects and inadequacies. Youmust offer an alternative.

The same principle applies to scientifictheory. Once a theory is well entrenched, itwill survive many assaults of empiricalevidence that purports to refute it unless analternative theory, consistent with the evi-dence, stands ready to replace it. Such conser-vative protectiveness of established beliefs is,indeed, not unreasonable. In the first place, inempirical science we aspire only to approxi-


mate truths; we are under no illusion that wecan find a single formula, or even a moder-ately complex one, that captures the wholetruth and nothing else. We are committed to astrategy of successive approximations, andwhen we find discrepancies between theoryand data, our first impulse is to patch ratherthan to rebuild from the foundations.

In the second place, when discrepanciesappear, it is seldom immediately obviouswhere the trouble lies. It may be located inthe fundamental assumptions of the theory,but it may as well be merely a defect in theauxiliary hypotheses and measurement postu-lates we have had to assume in order toconnect theory with observations. Revisions inthese latter parts of the structure may besufficient to save the remainder.

What then is the present status of theclassical theory of the firm? There can nolonger be any doubt that the micro assump-tions of the theory—the assumptions ofperfect rationality—are contrary to fact. It isnot a question of approximation; they do noteven remotely describe the processes thathuman beings use for making decisions incomplex situations.

Moreover, there is an alternative. Ifanything, there is an embarrassing richness ofalternatives. Today, we have a large mass ofdescriptive data, from both laboratory andfield, that show how human problem solvingand decision making actually take place in awide variety of situations. A number of theo-ries have been constructed to account forthese data, and while these theories certainlydo not yet constitute a single coherent whole,there is much in common among them. In oneway or another, they incorporate the notionsof bounded rationality: the need to search fordecision alternatives, the replacement ofoptimization by targets and satisficing goals,and mechanisms of learning and adaptation.If our interest lies in descriptive decisiontheory (or even normative decision theory), itis now entirely clear that the classical andneoclassical theories have been replaced by asuperior alternative that provides us with amuch closer approximation to what is actuallygoing on.

But what if our interest lies primarily innormative political economy rather than in

the more remote regions of the economicsciences? Is there then any reason why weshould give up the familiar theories? Have thenewer concepts of decision making and thefirm shown their superiority "for purposes ofeconomic analysis"?

If the classical and neoclassical theorieswere, as is sometimes argued, simply powerfultools for deriving aggregative consequencesthat held alike for both perfect and boundedrationality, we would have every reason toretain them for this purpose. But we haveseen, on the contrary, that neoclassical theorydoes not always lead to the same conclusionsat the level of aggregate phenomena andpolicy as are implied by the postulate ofbounded rationality, in any of its variants.Hence, we cannot defend an uncritical use ofthese contrary-to-fact assumptions by theargument that their veridicality is unimpor-tant. In many cases, in fact, this veridicalitymay be crucial to reaching correct conclu-sions about the central questions of politicaleconomy. Only a comparison of predictionscan tell us whether a case before us is one ofthese.

The social sciences have been accustomedto look for models in the most spectacularsuccesses of the natural sciences. There is noharm in that, provided that it is not done in aspirit of slavish imitation. In economics, it hasbeen common enough to admire Newtonianmechanics (or, as we have seen, the Law ofFalling Bodies), and to search for theeconomic equivalent of the laws of motion.But this is not the only model for a science,and it seems, indeed, not to be the right onefor our purposes.

Human behavior, even rational humanbehavior, is not to be accounted for by ahandful of invariants. It is certainly not to beaccounted for by assuming perfect adaptationto the environment. Its basic mechanismsmay be relatively simple, and I believe theyare, but that simplicity operates in interactionwith extremely complex boundary conditionsimposed by the environment and by the veryfacts of human long-term memory and of thecapacity of human beings, individually andcollectively, to learn.

If we wish to be guided by a natural sciencemetaphor, I suggest one drawn from biology


rather than physics (see Newell and Simon,1976), Obvious lessons are to be learned fromevolutionary biology, and rather less obviousones from molecular biology. From molecularbiology, in particular, we can glimpse apicture of how a few basic mechanisms—theDN A of the Double Helix, for example, or theenergy transfer mechanisms elucidated soelegantly by Peter Mitchell—can account fora wide range of complex phenomena. We cansee the role in science of laws of qualitativestructure, and the power of qualitative as wellas quantitative explanation.

I am always reluctant to end a talk aboutthe sciences of man in the future tense. Itconveys too much the impression that theseare potential sciences which may some day beactualized, but that do not really exist at thepresent time. Of course that is not the case atall. However much our knowledge of humanbehavior falls short of our need for suchknowledge, still it is enormous. Sometimes wetend to discount it because so many of thephenomena are accessible to us in the veryactivity of living as human beings amonghuman beings that it seems commonplace tous. Moreover, it does not always answer thequestions for which we need answers. Wecannot predict very well the course of thebusiness cycle nor manage the employmentrate. (We cannot, it might be added, predictvery well the time of the next thunderstorm inStockholm, or manage the earth's climates.)

With all these qualifications and reserva-tions, we do understand today many of themechanisms of human rational choice. We doknow how the information processing systemcalled Man, faced with complexity beyond hisken, uses his information processing capaci-ties to seek out alternatives, to calculateconsequences, to resolve uncertainties, andthereby—sometimes, not always—to findways of action that are sufficient unto theday, that satisfice.

REFERENCES

A. A. Alchian, "Uncertainty, Evolution, andEconomic Theory," / . Polit. Econ., June1950,55,211-21,

A. Ando, "On a Theoretical and Empirical

Basis of Macroeconometric Models," paperpresented to the NSF-NBER Conferenceon Macroeconomic Modeling, Ann Arbor,Oct, 1978.

Chester I. Barnard, The Functions of the Exec-utive, Cambridge, Mass. 1938.

William Baumol, Business Behavior, Value andGrowth, New York 1959.

G. S. Becker, "Irrational Behavior andEconomic Theory," J. Polit. Econ., Feb.1962, 70, 1-13.

Charles P. Bonini, Simulation of Informationand Decision Systems in the Eirm, Engle-wood Cliffs 1963.

Alfred Chandler, Strategy and Structure, Cam-bridge, Mass, 1962.

N. C. Churchill, W. W. Cooper, and T. Sainshury,"Laboratory and Field Studies of theBehavioral Effects of Audits," in C, P,Bonini et al,, eds,. Management Controls,New York 1964,

G. P. E. Clarkson, "A Model of the TrustInvestment Process," in E, A, Feigenbaumand J, Feldman, eds.. Computers andThought, New York 1963.

John R. Commons, Institutional Economics,Madison 1934,

R. M. Cyert, E. A. Feigenhaum, and J. G. March,"Models in a Behavioral Theory of theFirm," Behav. Sci., Apr. 1959, 4, 81-95,

and M. H. DeGroot, "Rational Expecta-tions and Bayesian Analysis," J. Polit.Econ., May/June 1974, 82, 521-36,

and "Adaptive Utility," in R, H,Day and T. Groves, eds.. AdaptiveEconomic Models, New York 1975, 233-46,

.and James G. March, A BehavioralTheory of the Firm, Englewood Cliffs1963.

. and H. A. Simon, "Theory of the Firm:Behavioralism and Marginalism," unpub-lished work, paper, Carnegie-Mellon Univ.1971,

., and D. B. Trow, "Observationof a Business Decision," / . Bus., Univ.Chicago, Oct. 1956, 29, 237^8 ,

D. C. Dearhom and H. A. Simon, "SelectivePerception: The Identifications of Execu-tives," Sociometry, 1958, 21, 140-144;reprinted in Administrative Behavior, ch,15, 3d ed,. New York 1976,


J. M. Dutton and W. H. Starbuck, ComputerSimulation of Human Behavior, NewYork 1971.

G. Eliasson, Business Economic Planning,New York 1976.

, A Micro-to-Macro Model of theSwedish Economy, Stockholm 1978.

B. M. Friedman, "Optimal Expectations andthe Extreme Information Assumptions of'Rational Expectations' Macromodels," J.Monet. Econ., Jan. 1979 5, 23-41.

, "A Discussion of the MethodologicalPremises of Professors Lucas and Sargent,"in After the Phillips Curve: The Persis-tence of High Inflation and High Unem-ployment, Boston 1978.

Milton Friedman, Essays in Positive Econom-ics, Chicago 1953.

J. R. Hayes and H. A. Simon, "UnderstandingWritten Problem Instructions," in W.Gregg, ed.. Knowledge and Cognition,Potomac 1974, 167-200.

A. O. Hirschman, Exit, Voice and Loyalty,Cambridge, Mass. 1970.

Charles C. Holt, Franco Modigliani, John F. Muth,and Herbert A. Simon, Planning Production,Inventories and Work Force, EnglewoodCliffs 1960. .

Y. Ijiri and H. A. Simon, Skew Distributionsand the Sizes of Business Firms, Amster-dam 1977.

E. Johnsen, Studies in Multiobjective DecisionModels, Lund 1968.

D. W. Jorgenson and C. D. Siebert, "A Compar-ison of Alternative Theories of CorporateInvestment Behavior," Amer. Econ. Rev.,Sept. 1968,55,681-712.

D. Kahneman and A. Tversky, "On the Psychol-ogy of Prediction," Psychol. Rev., July1973, 80, 237-51.

Janos Komai, Anti-Equilibrium, Amsterdam1971.

Howard Kunreuther et al.. Disaster InsuranceProtection: Public Policy Lessons, NewYork 1978.

Harvey Leihenstein, Beyond Economic Man,Cambridge, Mass. 1976.

J. Lesoume, A Theory of the Individual forEconomic Analysis, Vol. 1, Amsterdam1977.

R. E. Lucas, Jr., "An Equilibrium Model of the

Business Cycle," / . Polit. Econ., Dec. 1975,83, 1113-44.

_, "On the Size Distribution of BusinessFirms," Bell J. Econ., Autumn 1978, 9,508-23.

James G. March, Handbook of Organizations,Chicago 1965.

and H. A. Simon, Organizations, NewYork 1958.

Robin Marris, The Economic Theory of "Man-agerial" Capitalism, London 1964.

Jacob Marschak, "Role of Liquidity underComplete and Incomplete Information,"Amer. Econ. Rev. Proc, May 1949, 39,182-95.

and Roy Radner, Economic Theory ofTeams, New Haven 1972.

Alfred Marshall, Principles of Economics, 8thed.. New York 1920.

E. S. Mason, "Comment," in Bernard T.Haley, ed., A Survey of ContemporaryEconomics, Vol. II, Homewood 1952, 221-22.

J. M. Montias, The Structure of EconomicSystems, New Haven 1976.

J. F. Muth, "Rational Expectations and theTheory of Price Movements," Econometri-ca, July 1961, 29, 315-53.

, "Optimal Properties of ExponentiallyWeighted Forecasts," J. Amer. Statist.Assn., June 1960, 55, 299-306.

R. R. Nelson, and S. Winter, "Toward an Evolu-tionary Theory of Economic Capabilities,"Amer. Econ. Rev. Proc, May 1973, 63,440-49.

and , "Neoclassical vs. Evolu-tionary Theories of Economic Growth,"Econ. J., Dec. 1974, 84, 886-905.

Allen Newell and Herbert A. Simon, HumanProblem Solving, Englewood Cliffs, 1972.

and , "Computer Science asEmpirical Inquiry: Symbols and Search,"Communications of the ACM, Mar. 1976,79,113-26.

G. Orcutt, and R. CaldweUs-Wertheimer II, PolicyExploration through Microanalytic Simu-lation, Washington 1976.

A. Papandreou, "Some Basic Problems in theTheory of the Firm," in Bernard F. Haley,ed., A Survey of Contemporary Econom-ics, Vol. II, Homewood 1952.


E. H. Phelps-Brown, "The Meaning of theFitted Cobb-Douglas Function," Quart. J.Econ., Nov. 1957, 71, 546-60.

R. Radner, (1975a) "A Behavioral Model ofCost Reduction," Bell J. Econ., Spring1975, 6, 196-215.

, (1975b) "Satisficing," / . Math.Econ., June-Sept. 1975, 2, 253-62.

David R. Roberts, Executive Compensation,Glencoe 1959.

P. A. Samuelson, "Discussion: Problems ofMethodology," Amer. Econ. Rev. Proc,May 1963, 55, 231-36.

Henry Schultz, The Theory and Measurementof Demand, Chicago 1938.

Herbert A. Simon, Administrative Behavior,New York 1947; 3d ed. 1976.

, "A Formal Theory of the Employ-ment Relation," Econometrica, July 1951,19, 293-305

'A Comparison of OrganizationTheories," Rev. Econ. Stud., No. 1, 1952,20, 40-48.

, "A Behavioral Model of RationalChoice," Quart. J. Econ., Feb. 1955, 69,99-118.

., "Rational Choice and the Structureof the Environment," Psychol. Rev., Mar.1956, 63, 129-38.

., "Dynamic Programming under Un-certainty with a Quadratic Criterion Func-tion," Econometrica, Jan. 1956, 24, 74-81.

, Models of Man, New York 1957.., "The Compensation of Executives,"

Sociometry, 1957, 20, 32-35.., "Theories of Decision Making in

Econ. Rev., June 1959, 49, 223-83.-, "Discussion: Problems of Methodol-

ogy," Amer. Econ. Rev. Proc, May 1963,55,229-31.

, "From Substantive to ProceduralRationality," in Spiro J. Latsis, ed.. Meth-odological Appraisal in Economics, Cam-bridge 1976.

-, (1978a) "Rationality as Process andas Product of Thought," Amer. Econ. Rev.Proc, May 1978, 68, 1-16.

., (1978b) "On How to Decide What toDo," Bell J. Econ., Autumn 1978, 9, 494-507.

-, G. Kozmetsky, H. Guetzkow, and. G.Tyndall, Centralization vs. Decentralizationin Organizing the Controller's Department,New York 1954; reprinted Houston 1978.

and F. K. Levy, "A Note on the Cobb-

Economics and Behavioral Science," Amer.

Douglas Function," Rev. Econ. Stud., June1963, 30, 93-94.

G. J. Stigler, "The Economics of Information,"J. Polit. Econ., June 1961, 69, 213-15.

H. Tbeil, "A Note on Certainty Equivalence inDynamic Planning," Econometrica, Apr.1957, 25, 346-49.

John von Neumann and Oscar Morgenstem,Theory of Games and Economic Behavior,Princeton 1944.

A. A. Walters, "Production and Cost Func-tions: An Econometric Survey," Economet-rica, Jan.-Apr. 1963, 31, 1-66.

Oliver Williamson, Markets and Hierarchies:Analysis and Antitrust Implications, NewYork 1975.

S. Winter, "Satisficing, Selection, and theInnovating Remnant," Quart. J. Econ.,May 1971,55, 237-61.

rational decision making in business organizations

Documents