510 assignment

75th Anniversary Article

Jonathan Bendor, having earned all

of his degrees in political science at the

University of California, Berkeley, took a

job at the Graduate School of Business

at Stanford University in 1979. He is now

Walter and Elise Haas Professor of Political

Economics and Organizations there and

professor of political science, by courtesy.

He is a member of the American Academy

of Arts and Sciences and was a fellow of the

Center for Advanced Study in the Behavioral

Sciences in 1999–2000 and 2004–05.

E-mail: [email protected]

194 Public Administration Review • March | April 2015

Public Administration Review,

Vol. 75, Iss. 2, pp. 194–205. © 2015 by

The American Society for Public Administration.

DOI: 10.1111/puar.12333.

Jonathan Bendor Stanford University

Woodhouse and Collingwood’s claim is accurate, but Bendor’s is badly off the mark in an important respect: although disjointed incrementalism no longer exists as a clearly identifi ed approach to policy making, its components, especially the “Big Th ree” (local search, iterative adaptation, and the distributed intelligence of multiple minds) are fl ourishing, especially in applied fi elds such as computer science and operations research.

Th is is relevant for an assessment of “Muddling Th rough” because that paper is a work of applied theory. Indeed, virtually all of Lindblom’s work from 1958 to 1963—“Policy Analysis” (1958), “Muddling Th rough” (1959), “Decision-Making in Taxation and Expenditure” (1961), and Strategy of Decision (Braybrooke and Lindblom 1963)—was applied theory, as were signifi cant parts of Th e Intelligence of Democracy (1965) and Th e Policy-Making Process (1968). We have misunderstood what he was about.2 More importantly, we have misunderstood what applied theory is and what its role in public administration might be. Our collective confusion partly explains why this vital part of Lindblom’s work, though much cited, did not produce a vibrant research program.

Incrementalism: Dead yet Flourishing

Editor’s Note: Th is 75th anniversary essay revisits the most cited, reprinted and downloaded article in the history of PAR. Charles Lindblom’s “Th e Science of ‘Muddling Th rough’” looked critically at synoptic decision making and introduced a new strategy, disjointed incrementalism, into the social science lexicon. In an insightful analysis, Professor Jonathan Bendor examines Lindblom’s classic article and the aftermath of the applied theories it advanced.

JLP

Abstract: Charles Lindblom’s 1959 essay “Th e Science of ‘Muddling Th rough’” is best known for the strategy of deci- sion making—disjointed incrementalism—that it recommended. Th at famous paper and Lindblom’s related work also provided two theories: a critique of the conventional method (the synoptic approach) and an argument for using incrementalism instead. Both are applied theories: they are designed to help solve complex policy problems. Lindblom’s negative applied theory has stood the test of time well: the empirical foundations of its main micro-component (cogni- tive constraints of individuals) and its central macro-component (the impact of preference confl ict on policy mak- ing) have grown stronger since 1959. Th e picture regarding the positive applied theory is more mixed. As a coherent decision-making strategy, disjointed incrementalism has almost disappeared. Yet its key elements, the major heuristics identifi ed in “Muddling Th rough,” are thriving in many applied fi elds. Intriguingly, they are often accompanied by subroutines—especially optimization as a choice rule—typically associated with the synoptic approach.

Charles Lindblom’s “Th e Science of ‘Muddling Th rough’” (1959) is one of the most famous papers published in the Public Administration Review. Long a cornerstone of organization theory, it has been reprinted in more than 40 anthologies, and as of March 14, 2014, Google Scholar reported 8,431 citations, which makes it one of the most cited articles in the fi elds of public administration, organization theory, and bureaucracy. Th ough estimating impact is an uncertain business, it seems clear that “Muddling Th rough” has had a substantial infl uence on several generations of scholars.

Today, however, it is a spent intellectual force. Even a casual reading of the literature in public administra- tion and organization theory indicates that the idea of disjointed incrementalism as a coherent strategy of decision making is not discussed much. More than 20 years ago, two sympathetic scholars asserted that incrementalism “has not spawned a lively research tradition leading to cumulative refi nement and amplifi cation of the core concepts (Woodhouse and Collingridge 1993, 131); a few years later, another scholar stated that “incrementalism’s fate has been [to] just fade away” (Bendor 1995, 819).1

Incrementalism: Dead yet Flourishing 195

Indeed, I believe there is a serious mistake in Lindblom’s positive applied theory: it overes- timates how tightly linked are incremental- ism’s diff erent elements. A central thesis of the present article is that this position was wrong—disjointed incrementalism is not an integrated package but a toolkit of loosely connected heuristics7—and that it was the

most signifi cant mistake in the 1959 paper and its expansion, the 1963 book. It is a subtle error. But it was consequential: it resulted in overlooking opportunities for improvements, for example, new ways of optimizing via local search. Th is phrase sounds odd: optimizing is usually seen as central to the synoptic method; local search is part of incrementalism. But the implicit premise underly- ing this reaction is that each method is a tightly integrated system. If they are decomposable (Simon 1962), then we might be able to recombine some of their elements. As Weitzman (1998) and other students of innovation have argued, solving hard problems often involves combining elements from disparate sources.

Th us, the basic elements of incrementalism are widely used in a variety of domains, such as artifi cial intelligence and operations research, although an integrated identifi able strategy of disjointed incrementalism is hard to fi nd. Indeed, it is probably dead, even though its constituent parts are fl ourishing.

Th e rest of the article is organized as follows: Th e next section presents what I regard as the fi ve main theses of “Muddling Th rough.” Th e following sections examine how each thesis has stood up over time. Th e next-to-last section tries to explain why the strat- egy of disjointed incrementalism and its underlying positive applied theory did not become a research program. Th e conclusion follows.

The Main Claims of “Muddling Through” and Its Companion Pieces Because the 1959 paper is so terse, Lindblom understood that his argument needed to be unpacked elsewhere. Strategy of Decision was dedicated to this task. But parts of the argument were also devel- oped in “Policy Analysis” (1958), “Decision-Making in Taxation and Expenditure” (1961), Th e Intelligence of Democracy (1965), Th e Policy-Making Process (1968), and “Still Muddling, Not Yet Th rough” (1979).8 Hence, this article will refer to all these works.9 All of them move back and forth between basic and applied sci- ence, without fl agging the move. Th is need not be a problem. Th e diff erences between basic and applied research are often modest. Consider, for example, a topic related to “Muddling Th rough”: the study of expertise (Ericsson and Lehmann 1996). A claim that experts and novices play chess diff erently can simultaneously be a descriptive statement, hence part of basic research, and also a recom- mendation: “to improve your chess play you should . . .” Similarly,

part of Lindblom’s eff ort in his 1959 paper was to (1) codify what many experienced administrators and policy makers already do and (2) explain why what they are doing is sensible. Is this basic or applied work? Th e answer: it is both.

Nevertheless, the diff erences between trying to solve hard real-world problems versus

What, however, is applied theory? Some scholars might regard this phrase as an oxy- moron: theory and practical research are often contrasted in public administration (Martin 1952, 669; Simon 1947, 248) and other fi elds. But mapping the academic division of labor this way is unproductive. Far from an oxymoron, applied theories can help solve practical problems.3

Some examples might help. Suppose a new fi ghter plane repeatedly crashes during tests. Initially, it is believed that this is attributable to pilot error. However, an aerodynamics engineer, believing there is a design fl aw, gives an alternative explanation: the plane’s shape causes a kind of instability that is diffi cult even for skilled pilots to control. Hence, the plane frequently crashes. Th e engineer’s explanation rests on three elements: a detailed examination of the plane’s design, sim- ulations of that design’s performance under a range of conditions, and a theory of aerodynamics that helps explain the simulations’ output as well as the engineer’s choices during her examination of the plane’s design. Th is theory is an applied one in both senses of the word: on the one hand, practical problems—such as how to design airplanes that will be stable as well as maneuverable—were major stimuli for its development (Anderson 1997; Bloor 2011); on the other hand, much of aerodynamics uses work from basic science (mathematical physics).

Similarly, pedagogical strategies could be based on an applied theory of successful learning (Brown, Roediger, and McDaniel 2014), fi re-and-control strategies in artillery could be based on the applied science of ballistics (Bendor and Shapiro 2014), and contemporary eff orts to fi ght epidemics are based on a stunningly useful idea, the germ theory of disease.4

Th is is what “Muddling Th rough” and its companion pieces from 1958 to 1963 are: applied theory. Th ey are not descriptive theories of, say, budgeting or any other policy domain.5 As in our fi ghter plane example, they have design implications, as Wildavsky (1961) quickly saw regarding budgeting. Applied theories should have design implications: if they do not, then they are not useful, and so they are not good applied theories. Lindblom, following the pragmatic tradition in American social science, meant his theories to be useful.6 From his perspective, the common saying “it works in theory but not in practice” is nonsensical.

Applied theories can be negative or positive. A negative applied the- ory off ers a diagnosis: it explains why, for example, certain airplane designs will not work eff ectively. A positive applied theory explains why certain others will work. (Sometimes a single applied theory does both jobs, but in hard problems, one may understand why a certain class of designs fails without know- ing what will work.) “Muddling Th rough” and Strategy of Decision included both kinds of applied theory. Lindblom’s account of the defi ciencies of the standard procedure is a negative applied theory; his justifi cation of disjointed incrementalism is a positive one. Th e former has stood the test of time better than the latter.

I believe there is a serious mistake in Lindblom’s positive applied theory: it overestimates

how tightly linked are incre- mentalism’s diff erent elements.

Th e diff erences between trying to solve hard real-world problems versus describing and explaining phenomena can help us understand what Lindblom

was doing.

196 Public Administration Review • March | April 2015

under reasonable interpretations of “integrated,” thesis 4 is empiri- cally inaccurate. Intriguingly, Lindblom seems to have had second thoughts about it (see Lindblom 1968).

Th esis 5: Th ere is great value in distributed intelligence.15 Institutions can compensate for cognitively limited and parochial offi cials. Multiple groups and organizations can bring diff erent perspectives to bear on complex problems.

Th esis 5 was central to Intelligence of Democracy (but see also Lindblom 1968).

How have these diff erent claims stood up to the test of time? We tackle this question next.

Thesis 1 In thesis 1, Lindblom was challenging the conventional wisdom about how to make decisions. We all know the guidelines: (1) stipulate your goals; (2) identify all16 the diff erent alternatives for reaching those goals; (3) predict their consequences; (4) evaluate them, based on your goals and your predictions; (5) select the best one. What could be more reasonable?

Th erein lies the problem: the conventional wisdom was so reasonable that it amounted to a defi nition of rational decision making rather than a hypothesis about how to solve problems eff ectively. Th e beauty of a defi nition is that it does not require testing. Data are unnecessary; because steps (1) through (5) are the essence of rational (“good”) deci- sion making, we all should follow them. Deviations are, ipso facto, bad.

Lindblom’s shrewd move was to recategorize the conventional prescription, treating it as a hypothesis about a problem-solving proce- dure. Because this enables us to assess the con- ventional prescription by standard scientifi c methods, it was a crucially important move. One of the scientifi c methods involves assess- ing a technique’s feasibility. Here, basic and applied science often overlap. (For example,

an engineering design that violates the second law of thermodynam- ics—it “describes” a perpetual motion machine—will be swiftly rejected by competent patent offi cers.) Th e conventional method’s feasibility was a key issue for Lindblom: he argued that when policy problems are complex, the method is usually impossible to implement. In such situations, government offi cials typically lack key ingredients, especially (a) political consensus on goals precise enough to guide policy choice; (b) knowledge about consequences of many alternatives, especially those that diff er substantially from the status quo; and (c) time needed to fi x (a) or (b).

Criticizing the synoptic method for infeasibility has a salutary eff ect: it puts empirical issues front and center. A claim of infeasibility is a bold—highly falsifi able—hypothesis. Indeed, sometimes such an assertion can be shown to be false, for example, when high-speed computers make comprehensive search possible in real time. But here it is undeniably true: we can easily fi nd policy problems that are so complex that the synoptic method is infeasible.17 Th en chapter 4 of Strategy—“Matching Practices to Political Contexts”—takes center stage: rather than amounting to a defi nition of rational decision

describing and explaining phenomena can help us understand what Lindblom was doing. Th e fi rst page of “Muddling Th rough” is a brief sketch of two ways of making decisions: two procedures. Procedures are naturally linked with the applied sciences—how to eliminate bacteria from a water system or how to supply military bases in a timely manner—and procedural language pervades the article, as it does Strategy of Decision and the 1979 article as well. Th is is a clue about Lindblom’s intentions.10

Th e procedural language also helps us identify the major theses of “Muddling Th rough” and its off spring. I see fi ve main theses. Th e fi rst two, which pertain to the standard method of decision mak- ing, are negative; the last three, about disjointed incrementalism, are positive.

Th esis 1: Th e synoptic method is much less useful than is usually believed.

Th esis 2: Th esis 1 holds for systematic reasons. Th e synoptic method is not adapted to common and stable features of impor- tant policy problems. Th ese features include macro-properties of political systems—especially confl icting policy preferences—as well as micro-properties, especially the cognitive limits of individ- ual decision makers (Braybrooke and Lindblom 1963; Lindblom 1959).

By itself, thesis 1 is merely an assertion. Th esis 2, however, has the makings of a real theory: it explains why the standard method of decision making has the performance property noted by thesis 1. Further, thesis 2 is not merely a compilation of empirical regulari- ties. It identifi es causal mechanisms: crucial ways in which the synoptic approach is mala- dapted to pervasive cognitive and political constraints.11 Hence, it provides a diagnosis of the standard method’s weaknesses. And because that method is supposed to help us solve problems, this diagnosis indicates that thesis 2 is the core of an applied theory.

Th esis 3: Fortunately, however, even when facing important, complex problems, (a) decision makers can and sometimes do use heuristics, which (b) work much better than is usually believed.12

Th e “can” in part (a) isn’t trivial, largely because of the “even in important, complex problems.” Part (b) is justifi ed by Lindblom’s claim that these heuristics are adapted to the stable diffi cult features of important policy problems. Indeed, he argues that in at least some cases, they ameliorate them.

Th esis 4: Th ese heuristics form an integrated strategy.13

Although most critics have ignored this claim, it is central to his argument. After all, the 1959 article is entitled “Th e Science of Muddling Th rough,” which suggests, as he states in the essay, that there is a method here.14

Th e meaning of the claim, however, is not altogether clear: what are the properties of a method of decision making or of an inte- grated strategy? I will address this question later. I will argue that

Lindblom’s shrewd move was to recategorize the conventional prescription, treating it as a hypothesis about a problem-

solving procedure.

Incrementalism: Dead yet Flourishing 197

Regarding cognitive-informational issues, Lindblom’s judgment was eerily on target. Relevant feedback is generated by quite dif- ferent academic communities. Th eoretical computer scientists have carefully studied what makes some problems computationally dif- fi cult. Indeed, one of the main unsolved prob- lems in this fi eld concerns exactly this issue.19 Problems that exhibit combinatorial explo- sion—chess or the traveling salesman problem,

for example—are very diffi cult. Many theoretical computer scientists believe that what Lindblom called “comprehensive” methods—which would guarantee optimal solutions in even the worst cases—may remain unattainable indefi nitely.20

Work in the cognitive psychology of expertise has reached consistent conclusions. It is now well established, for example, that even grand- masters do not solve hard chess problems by anything remotely like a comprehensive method (Simon and Schaeff er 1992). As Lindblom argued, urging chess players to be more comprehensive is useless advice. It betrays ignorance about a key feature of chess: combinato- rial explosion.

In short, Lindblom’s judgment concerning thesis 2 was uncannily good. Th is, in turn, gives support to thesis 1. Hence, Lindblom’s negative applied theory has stood up very well.

Now we turn to his positive applied theory.

Theses 3(a) and (b) Lindblom said repeatedly that he was codifying what experienced competent offi cials already do. Th at is a modest claim: discovering a new way to make decisions would have been grander. But in his view, this lack of novelty was a plus. If certain heuristics are used, then they are feasible, a property central to Lindblom’s argument. As it turned out, nobody has mounted a serious challenge to this part of his formulation: there is no dispute that offi cials often use these heuristics and (hence) they are feasible.

Th e much more challenging part of this thesis is 3(b), the perform- ance claim. Here, Lindblom in several ways made things too easy on himself. First, he often slipped into arguing that because the conven- tional method is infeasible for hard problems, we have no choice: we must muddle through. But this argument works only if there are just two methods for making decisions: the standard one and disjointed incrementalism. But this claim is obviously false. Hence, one can agree with Lindblom that the conventional method cannot be used in certain domains yet reject his conclusion that we must therefore use disjointed incrementalism. (Th is position implies that we should also reject his assertion that the only choice in such domains is between a thoughtful explicit incrementalism and some ad hoc version of it.)21

Second, the performance claims he off ered were often22 quite cautious: for example, “the choice is between a combination of practical devices that off ers some chance of headway and pedantic prescriptions that will not even succeed in getting the anchor lifted” (Braybrooke and Lindblom 1963, 112).23 Indeed, some were virtu- ally unfalsifi able: “a group of partisan adjusters may generate a great deal more information and analysis than will a central coordinator.

making, the conventional method is embed- ded in an “if … then … .” And though it is grander than deciding when it is appropriate to use other problem-solving methods, such as checklists (Gawande 2011), or queuing theory or linear regression or quasi-experiments, using the conventional method would be disciplined by the same battery of analytical and empirical tools that we deploy on other techniques.

Feasibility is crucial: showing that a method is infeasible for a certain class of problems can end the debate, at least regarding the technique’s usefulness for that class. Defenders of the conventional method, alert to the potency of the criticism, have tried two rebut- tals. Th e fi rst was to assert that Lindblom was critiquing a straw man: the instruction to identify and evaluate all alternatives should not be taken literally. To this, Lindblom retorted, in eff ect, “then what does it mean?” As he pointed out, the conventional method’s vaunted crispness starts to evaporate once this concession is made.

Th e second rebuttal was that although following the standard guidelines literally may be impossible when problems are complex, decision makers should try to follow them. Lindblom realized, how- ever, that this defense is itself a hypothesis (“When the conventional method is infeasible, trying to follow it is the best way to make deci- sions”) and pointed out that it was not empirically substantiated. He sharpened his criticism with a telling analogy: humans cannot fl y, and telling someone to try really hard by fl apping one’s arms really fast will not solve the problem (1979, 518). Th is clever example was buttressed by a sustained analysis of the ways the conventional method is maladapted to diffi culties presented by complex policy problems—the argument linked to thesis 2.

Thesis 2 With thesis 2, Lindblom was building a theory of why a certain method would not work as well as many people thought it would; that is, he was off ering an explanation for why thesis 1 is plausible.

Th esis 2 is of great practical importance. It rejects several common diagnoses—offi cials do not make decisions in the conventionally prescribed “rational” way because they are incompetent, ill trained, or insuffi ciently motivated—and replaces it with a penetrating examination (Braybrooke and Lindblom 1963, 48–56) of a subset of the problems decision makers face (hard ones) and the mismatch between key features of these tasks and the standard prescriptions.

Th esis 2 has been critiqued around the edges, largely on domain- specifi c grounds, as our knowledge and information-processing capabilities advance.18 But its general thrust has never been over- turned. Th is is especially so regarding its political component (pref- erence confl ict). We are, for example, no closer to being able to use Verne Lewis’s classical public fi nance rules for budgeting today than we were in 1952, and a fundamental reason for this is, as Wildavsky (1961) stressed, confl icting preferences. Indeed, recent studies of congressional voting indicate that our parties are more polarized today than they were when Wildavsky off ered his now-classic cri- tique of Lewis et al. Greater preference confl ict makes it even harder to use the standard public fi nance rule of equating marginal values across diff erent programs.

Regarding cognitive-infor- mational issues, Lindblom’s

judgment was eerily on target. Relevant feedback is generated

by quite diff erent academic communities.

198 Public Administration Review • March | April 2015

a serious fl aw: it did not defi ne the idea of an integrated strategy. Unfortunately, neither did Lindblom. Th e concept’s properties were left mysterious.

However the conceptual question is resolved, we should recognize that assessing how tightly linked are some problem-solving methods is an empirical matter. Th erefore, asserting that procedure x cannot be used without y, or that w is more valuable when used in tandem with z, are hypotheses. Hypotheses can be wrong. Maybe the links are not so tight. Or perhaps they form an integrated whole for one class of problems but not for another. Or perhaps they are mutually reinforc- ing but so are other combinations of problem-solving procedures.26

Th at thesis 4 is a hypothesis can be seen by noting an alternative position about incrementalist heuristics that Lindblom articulated several times: these heuristics are merely a set of tools, a grab-bag of “dodges” (1968, 24). Th ere is no attempt in Th e Policy-Making Process to argue that they form an interrelated system. Instead, the criterion for belonging to the list seems to have been a functional property: whether a particular item simplifi es a decision maker’s problem. Th is not only produces considerable variety (e.g., the 1968 list included working on bottlenecks, which the 1959 arti- cle ignores). It also underscores the issue of how the members of the list are related: in particular, if they are functional substitutes, then they probably do not form an integrated package. Typically, it is not substitutes but complements—a car’s brakes, steering wheel, and engine—that form an interdependent whole. Brakes and steering wheel are indeed “mutually reinforcing”: the value of each is much enhanced by the other. In contrast, the value of one substitute typically decreases when another is present.27

Furthermore, even though brakes, steering wheels, and engines strongly complement each other, they can be decoupled and reassembled in new artifacts. (Th e armored tank was designed by borrowing treads from farm tractors, engines from internal combus- tion vehicles, cannon from artillery, and armor plate from ships; see Bendor and Shapiro 2014.) So even if a system’s components are highly interdependent, it does not follow that the system is nonde- composable in a design (or redesign) sense.

All this leads to claim (3), above: perhaps we can combine elements (subroutines) from the standard “method” and from disjointed incrementalism to form new problem-solving procedures. Th is is especially likely if incrementalist subroutines are functional substi- tutes, whereas they would complement conventional subroutines.28 And if we can mix and match across the two decision making meth- ods, then we will have shown that they are not tightly integrated rival procedures; they are toolkits of loosely connected subroutines that can be recombined.

Consider the combination of local search and optimization. I do not mean optimization in a grand sense, as in, “I’ve discovered the best solution to this problem.” Th at is a bold performance claim, especially when made in a practical setting. I mean it in a more limited, procedural sense: the choice process ends with a subrou- tine which involves (a) comparing some alternatives to each other and (b) selecting the one believed to be the best according to some criterion or criteria. (Th is may sound vacuous but it is not: the satisfi cing heuristic skips step (a)—alternatives are compared only to

Again, they will not necessarily do so, but they may” (Lindblom 1965, 174; see also 154–56).

But before we use a Popperian club to beat thesis 3(b) to death, we should remind ourselves of the practical orientation of Lindblom’s argument. He believed that scholars and practitioners overestimate the value of the conventional method (theses 1 and 2) and under- estimate incrementalism (thesis 3b). Th erefore, if defenders of the conventional method believe that incrementalist heuristics are worthless, Lindblom only needs to explain why they are better than that low assessment. Th is does not require showing that disjointed incrementalism works better on average than the conventional method anywhere; one must show only that it is better than most people believe it to be. Th is is easier.

Th us, Lindblom tended to avoid making direct performance com- parisons between the conventional method and incrementalism.24 Th is maneuver produced two undesirable side eff ects. First, incre- mentalists would have been more likely to ask, “what do we need to show now?” either empirically—observational or experimental data on how the two methods perform—or theoretically (mathemati- cal or computational models of the two methods) had conjectures about performance diff erences been put forward. Second and relat- edly, methodological opportunities were missed. Trying to answer questions about the methods’ relative eff ectiveness might have motivated Lindblom or his students to learn modern research tools: mathematical or computational models for theoretical assessments; experiments for empirical ones. As I argue later, that this did not happen was unfortunate.

We now turn to thesis 4, the most problematic of the fi ve.

Thesis 4 Th is section focuses on four points.

1. Lindblom’s formulation signifi cantly overestimates how tightly connected are the heuristics of disjointed incremen- talism. Th ese rules of thumb can be decoupled more often than his work implies.

2. Similarly, the elements of the same conventional method can be decoupled more often than Lindblom maintained.

3. Strikingly, we can mix and match subroutines across the two approaches. Th us, because recombining a few basic elements can yield tremendous variety (Simon 1962; Weitzman 1998), there is not merely a “third way” (Etzioni 1967); there are a great many.

4. Overlooking this combinatorial explosion and the opportu- nities it gives us were Lindblom’s biggest mistakes.

In both his negative and positive analyses, Lindblom juxtaposed two decision-making methods. Th is juxtaposition indicated that he saw them as distinct procedures with integrated sets of components. He buttressed that impression with explicit statements: “Successive limited comparisons is, then, indeed a method or system; it is not a failure of method for which administrators ought to apologize” (1959, 87).25

Th e standard method is often presented as an integrated package; because here Lindblom was following common practice, he did not need to justify the claim. Common practice, however, contained

Incrementalism: Dead yet Flourishing 199

through successive approximation is one such tactic. Others (though they are not always sharply distinguishable) are: crisis decision making … satisfi cing. (Lindblom and Cohen 1979, 67)

Th us it seems that Lindblom, always thoughtful, came to have doubts about thesis 4. But we should be cautious about this claim; he never explicitly disavowed the relevant statements from “Muddling Th rough” or Strategy of Decision. Th e foregoing quotes from Policy-Making and Usable Knowledge are tantalizing clues about how his beliefs might have changed, but they are not defi nitive.

Implications of the Rejection of Thesis 4 for Thesis 3 Th e accuracy of thesis 4 has implications for thesis 3(b). If the former is false, then we should rethink the latter: in particular, we should compare the performance of diff erent combinations of heu- ristics or subroutines. (If there are many diff erent packages, then the value of one particular combination—that identifi ed by “Muddling Th rough”—may be rather unimportant.)

Seen as individual tools, each heuristic exam- ined in “Muddling” is extremely general: none is restricted to a particular policy domain or country. Th is generality may come at a steep price. A regularity in cognitive science, Newell’s Law, says that at any given time, no problem-solving method is both more general and more powerful than all others. Th ere are

trade-off s: powerful methods are specialized; general ones, weak (Newell 1969, 1980).34 Put more dramatically, Newell’s Law says that there is no magical problem-solving method.35

Th e present author knows of no exceptions to Newell’s Law. Enthusiasts sometimes proclaim they have discovered one—ran- domized experiments, cost–benefi t analysis—but experience invariably tempers the enthusiasm: the method cannot be used in certain situations (e.g., moral constraints against experimenting on humans) or in others can only be approximated (e.g., important benefi ts cannot be easily quantifi ed). Th ese experiences, combined with the warning inherent in Newell’s Law, should make us cau- tious. On the one hand, if an enthusiast provides evidence that he or she has discovered a powerful problem-solving tool, then our priors should be that this power is confi ned to a particular domain; identifying the domain’s boundaries is important.36 If, on the other hand, an advocate provides evidence that he or she has discovered (or codifi ed, per “‘Muddling Th rough”) a general tool, then our priors should be that the method is weak in some signifi cant ways; fi guring out how to ameliorate these defi ciencies is important.

Clearly, the elements of disjointed incrementalism are general. Hence, Newell’s Law suggests they are weak.37 Consider, for exam- ple, partisan analysis. True, this type of analysis does not presup- pose agreement on program objectives, which makes it usable when more classical analysis, which does presume agreement, cannot be deployed. Newell’s Law predicts that we will pay a price for this generality. We do not have far to look. Folk wisdom and formal reasoning (noncooperative game theory) agree here: the more decision makers’ preferences confl ict, the less they trust each other, which reduces the amount of credible information transmitted from

an aspiration level; see Simon 1955; Bendor, forthcoming.) Call this computational optimization to distinguish it from the much bolder performance claim of optimization in fact.29

Th e hybrid procedure is simple: of the alternatives generated by local search, select the best one.30 Th is is algorithmically kosher: it is perfectly consistent to select the best of a set of options which all resemble the status quo, even though the two subroutines have been associated with opposed academic camps.31

Not only are these subroutines consistent; combining them makes sense in terms of Lindblom’s own (applied) theories. Restricting search to options similar to the status quo makes the choice problem easier, as Lindblom argued; it may therefore make computational optimization feasible. Th at is, the hybrid procedure is adapted to the information- processing constraints of real problem solvers. And supplementing local search with another heuristic—“just devise a few new options”— puts computational optimization even more within our grasp.

Lindblom’s positive applied theory is largely about making a problem-solving procedure politically and cognitively feasible. Th ere might be several ways of making an overly ambitious procedure respect relevant con- straints. Th ese alternative adaptations are substitutes: consequently, modifying the pro- cedure in one way may make it unnecessary or even harmful to do so in another way.32

We can hybridize still more. Suppose that local search generates too many alternatives: they swamp an organization’s capacities to evaluate and make a selection. Suggestion: combine multicriteria satisfi cing (Simon 1964; Tversky 1972) with computational optimi- zation. Initially, decision makers should discard many alternatives by comparing them to criteria (e.g., does an option violate a budget constraint?), not to each other. A key premise is that comparing an alternative to an external benchmark (“Will the courts uphold this policy?”) is often easier than comparing it to another option, which may entail complex, high-dimensional evaluation.

Concluding Thoughts on Thesis 4 After Strategy of Decision, Lindblom seems to have moved toward the “toolkit” perspective. Th e Policy-Making Process refers to “certain strategies or dodges that man has developed for dealing with very complex problems” (1968, 24), and then lists seven—satisfi cing, the next chance, feedback, remediality, seriality, bottlenecks, and incre- mental search—without arguing that they form a unifi ed method. Instead, he suggests that what unifi es the various “dodges” is their common function: they are “useful devices for stretching man’s analytic capacities” (27)—procedural rationality, in Simon’s terms (1976, 1978).33

In Usable Knowledge, Lindblom returned to the toolkit perspective:

It seems reasonably clear that [the people who do profes- sional social inquiry] … and policy makers and other practical problem solvers … do attack their problems with a variety of invented tactics and that the tactics diff er both from situa- tion to situation and from person to person. Problem solving

Lindblom’s positive applied theory is largely about making a problem-solving procedure

politically and cognitively feasible.

200 Public Administration Review • March | April 2015

55 years ago. Today’s version of “Toward a Th eory of Budgeting” would not be taken seriously by many scholars. Lindblom has convinced us that those recommendations are not very helpful, a conclusion anchored by a hardnosed assessment of the eff ects of power- ful political and cognitive constraints.

However, I have also suggested that several pieces of convergent evidence indicate that the positive applied theory of “Muddling” and Strategy of Decision has not had much impact. Indeed, it and the associated strategy of dis-

jointed incrementalism are now moribund. Why?

If thesis 4 is, in fact, wrong, and specialists recognize this, then that would be a major part of the answer. But it is not a complete one. Research programs in the social sciences have been created out of material much less promising than disjointed incrementalism and its supporting theory. Why did not the 1959 paper and the 1963 book create a research program based on their ideas? Perhaps it would not have lasted long, as scholars came to realize that a cornerstone— thesis 4—is problematic, but that probably would have taken quite some time. It is hard to kill off empirically shaky theories (basic or applied) in the social sciences. Indeed, there seem to be zombie ideas among us (Quiggin 2010). Th e strategy of muddling through and its underlying theory are better ideas than most of these. Yet they did not produce a sustained research program.38 Why?

I believe there were several causes that together made it unlikely that incrementalism would become a research program. First, what, exactly, were professors in schools of public administration or public policy supposed to teach their students? If disjointed incrementalism really is a systematic method for solving problems and making decisions, then we should be able to provide a reason- ably clear description of the procedure. But Kenneth Arrow, for one, thought that such a description was not given in Strategy of Decision (1964, 586).39

In contrast, consider the allied methods of social experimentation and program evaluation. Th ese overlap with incrementalism in signifi cant ways. All three are infused by a spirit of pragmatism: try new policies, then see whether they work. Further, social experi- mentation typically involves rather modest departures from the status quo; although radical experiments are possible—as Dahl and Lindblom noted when pointing out diff erences between incremen- talism and experimentation (1953, 84)—they are the exception, not the rule. And important strands of program evaluation (Scriven 1998) have stressed modest assessments of practice that, as in dis- jointed incrementalism, emphasize empirics over theory.40

But social experimentation and program evaluation also diff er from incrementalism in important ways. In particular, articles and books specifi ed guidelines for how to implement these methods. Th us, probably most teachers of experimentation have heard about the seminal publications of Donald Campbell (1969) and his collabora- tors (Campbell and Stanley 1963). Similarly, program evaluation has become institutionalized, with journals, a professional society, and the other social infrastructure of a profession.

informed to uninformed actors. Hence, there are good domain-specifi c reasons for believ- ing that Newell’s Law holds here: the average value of analysis in highly partisan settings will be lower than that of analysis in less partisan contexts. Partisan analysis may be feasible when classical analysis is not, but its value will often be slender.

Of course, this trade-off is not restricted to incrementalism’s tools. Quasi-experiments can be done where fully randomized experiments cannot; but, as advocates of the latter method have pointed out, we pay a price in the quality of inferences that the former can support. Neither method dominates the other in the generality–power space. Similarly, whereas cost–benefi t analysis requires an input of quantifi ed benefi ts, all measured on the same dimension, cost-eff ectiveness does not; hence, the latter can be used where the former cannot. But the price of this greater generality is weaker (less decisive) output: whereas cost-eff ectiveness generates a set of (undominated) options, cost–benefi t analysis typically identi- fi es a unique alternative—the one that maximizes net benefi ts.

Th us, if thesis 4 is false and Newell’s Law is true, we should regard the major tricks identifi ed by “Muddling Th rough” as general-but- weak metaheuristics (see note 38), elements of a set of undominated problem-solving methods.

Thesis 5 At one time, our fi eld probably underestimated the value of dis- tributed intelligence. Now, however, the argument for redundancy (Landau 1969)—an important kind of distributed intelligence—is well accepted in public administration. In particular, it is widely recognized that when means-end relations are uncertain, having multiple minds work independently on the same problem increases the probability of success.

Several aspects of Landau’s argument have been modifi ed (Bendor 1985; Ting 2003). But the central thrust of thesis 5 is now uncontroversial (see Page 2007 for a detailed contemporary argu- ment for the value of multiple cognitively diverse minds in problem solving).

Why Did Lindblom’s Positive Applied Theory Have So Little Impact? Some Tentative Answers I have argued that Lindblom’s negative applied theory is on target. Furthermore, it seems to have had a major impact on the fi elds of public administration and policy analysis. Admittedly, systematic evidence about an idea’s causal power is usually hard to come by, and this case is no exception. But the data on citations, reprints, and awards noted earlier clearly indicate how highly “Muddling Th rough” is regarded—consistent with the hypothesis that Lindblom’s negative applied theory has had major impact. Further, consider the following thought experiment. Suppose a mental clone of Verne Lewis presented a paper at the American Society for Public Administration conference that restated Lewis’s argument. We can confi dently forecast that it would be received with much skepticism, supplemented by suggestions that its author read a classic essay by one Charles E. Lindblom, published in the fi eld’s main journal

Folk wisdom and formal rea- soning (noncooperative game theory) agree here: the more decision makers’ preferences confl ict, the less they trust

each other, which reduces the amount of credible information transmitted from informed to

uninformed actors.

Incrementalism: Dead yet Flourishing 201

Whatever the explanation, the absence of mathematical or computational mod- els became especially problematic when Lindblom moved into the systemic level of Intelligence of Democracy. Teasing out the implications of complicated institutional processes is diffi cult under any circumstances; without the aid of formal models, this is like

wrestling with a hand tied behind one’s back.43

Finally, it is important to recognize that we are talking about an unusual event: few theories—applied or basic—generate a vibrant research program. Sheer persistence (zombie ideas) is not enough; other scholars must build on the idea. Our fi eld exhibits a striking piece of evidence regarding the diffi culty of this kind of project: the fate of Simon’s Administrative Behavior. Although famous in public administration—in his review of the fi eld’s best-known books Sherwood said that Administrative Behavior “was in a class by itself ” (1990, 254)—there is little evidence that it triggered a research program. A study of the impact of Simon’s volume on textbooks concluded that “Simon did not inspire a new paradigm through Administrative Behavior … No text—not even that by Simon, Smithburg, and Th ompson (1950)—utilizes the paradigm put forward in [Simon’s] book to organize their treatment of the fi eld of public administration” (Dunn 1988, 381).

Perhaps these two giants, Simon and Lindblom, were both too early. As Landau (1960) noted, an applied science presupposes a basic one: without the latter, there is nothing to apply. So both Simon and Lindblom failed in this project for the same reason: there was not enough basic science to apply.44

Conclusions “Muddling Th rough” and its companion pieces are primarily works of applied theory. Of course, they rely, as must any adequate applied

theory, on empirical premises—in this case, about how decision makers think and choose and about durable, causally powerful proper- ties of their work environment, especially preference confl ict. But they go beyond these empirical formulations; they include recom- mendations, about procedures worth trying and those that are not when problems are complex.

Lindblom’s negative applied theory has stood up very well. To the best of my knowledge, no plausible critique has ever been off ered. On the contrary: both its micro-component (cognitive constraints) and its macro-component (confl icting policy objectives) have been strengthened by a great deal of basic research, both theoretical and empirical, since 1959. Of course, boundaries keep shifting—what was computationally infeasible then may be tractable now45—but we also confront extremely complex new problems, for exam- ple, climate change, that were not on anyone’s screen in 1959. Governments today confront a dynamical steady state: some prob- lems transition from very hard to easier, others get harder, and dif- fi cult new problems appear. Hence, the overall thrust of Lindblom’s negative theory is as strong today as it was when his justly celebrated article was published.

Advocates of disjointed incrementalism might dismiss program evaluation and social experimentation as mere techniques—tools we teach our students. Th ey are not a grand general method of decision making, as is dis- jointed incrementalism. But that response is vulnerable to Newell’s Law: if incrementalism is so general, then its problem-solving power is weak. Perhaps Lindblom, for all his brilliant insights, simply was too ambitious, that is, searching for something that does not exist.41 Perhaps there is no grand strategy of decision.

A second, and related, factor: by not availing himself of modern ways of representing theoretical ideas (mathematical and compu- tational models), Lindblom made it diffi cult for followers to build a research program of applied theory. Th ere are no guarantees here—algorithms for building research programs do not exist—but representing theories formally makes it easier for them to be com- municated, criticized, and modifi ed. Th ese activities are the life of theoretically oriented research programs.

In this regard, it is instructive to glance at a line of work closely related to thesis 5: Condorcet’s jury theorem. Born a mathemati- cally expressed idea, the jury theorem has spawned many exten- sions. Is one troubled by Condorcet’s original assumption that decision makers’ errors are independent? No problem: Ladha (1992, 1995) has constructed models that assume correlated errors. Is one bothered by Condorcet’s assumption that decision makers have the same preferences (free the innocent and imprison the guilty)? No problem: in Miller (1986), the actors are voters who have confl ict- ing preferences over which party to elect. And so on (Bendor and Bullock 2008 n. 6).

Admittedly, the procedure studied by the jury theorem(s)—an up-or-down vote on two alternatives—is much simpler than the processes examined by the verbal theories of “Muddling Th rough” and Intelligence of Democracy. Neither representational method dominates the other. Th ere are trade-off s: verbal theories can be richer than formal ones, but the latter make it easier for sub- sequent scholars to criticize and modify the original formulation. Formal models also make it easier for suitably trained schol- ars to understand the logical structure of each other’s arguments—it is clear which claims in the canonical Condorcet jury theorem are premises and which are conclu- sions—which helps create research programs and the invisible colleges that work on those programs.

Th at Lindblom passed up the opportunity of using formal rep- resentational techniques is somewhat ironic, given that he, like Simon, believed that our minds are small compared to the com- plexity of some important problems.42 Human capital theory probably explains this: although a PhD economist, apparently he did not get a strong training in mathematical methods in graduate school—Samuelson’s revolution was yet to come—and retooling in midcareer (he was 42 when “Muddling Th rough” was published) is uncommon.

Th ere are trade-off s: verbal theories can be richer than for- mal ones, but the latter make it easier for subsequent scholars to criticize and modify the original

formulation.

Advocates of disjointed incremen- talism might dismiss program

evaluation and social experimen- tation as mere techniques—tools

we teach our students.

202 Public Administration Review • March | April 2015

4. For an examination of the intricate relations between basic and applied research in the fi elds of bureaucracy and public administration, see Bendor 1994.

5. Hence, this article will not examine the eff ect of “Muddling Th rough” on the empirical study of budgeting. Obviously, Lindblom’s work had a signifi cant impact on Wildavsky’s seminal Politics of the Budgetary Process (1964), which, in turn, infl uenced generations of political scientists. Indeed, the biggest academic impact of “Muddling Th rough” has probably been on the descriptive fi eld of budgetary politics. Here, however, I want to focus on Lindblom’s main project: applied science.

6. How much Lindblom was infl uenced by the pragmatist movement per se is unclear. “Muddling Th rough” refers neither to pragmatism nor its major fi gures. John Dewey is mentioned several times in Strategy of Decision. However, although Braybrooke and Lindblom acknowledge that their perspective “is, in broad outlines at least, very much like the spirit that animated John Dewey’s notions of evaluation and choice” (1963, 18), they also maintain that their points are “complementary but independent” (18–19). But then, with their characteristic honesty, they ask, “Yet can the results be independent, when Dewey has done so much to aff ect the climate of social science, at least in America?” (19).

7. As we will see, Lindblom came to perceive disjointed incrementalism in a dif- ferent way. But in the early infl uential publications, his position clearly was that incrementalism is an integrated package.

8. Indeed, his early book, Politics, Economics, and Welfare (1953, 82–85), coau- thored with Robert Dahl, provided a clear explication of incrementalism. Lindblom devoted considerable time to these ideas in the 15 years between the publications of Politics and Th e Policy-Making Process (1968). Th e last element in his project, “Still Muddling” (1979), was an afterthought.

9. For convenience, my wording will sometimes incorrectly suggest that Lindblom was the sole author of all these works. Th is is meant as no disrespect to David Braybrooke, a fi ne scholar and a kind person.

10. A concern with improving the world was a marked characteristic of Lindblom, as his daughter Susan indicated in her informal biography of him (Friedman 2010).

11. Moreover, Lindblom makes clear that these causal mechanisms are not spatially confi ned (e.g., to the United States), another indication that his formulation should be considered a theory.

12. Th esis 3 was anticipated by Karl Popper, as Lindblom acknowledged: “Incrementalism has much in common with what Karl Popper calls ‘piecemeal social engineering.’ Karl Popper, Th e Open Society and Its Enemies … 1945, Vol. 1, Pp. 139–144” (Dahl and Lindblom 1953, 82).

13. For example, “Successive limited comparisons is, then, indeed a method or system; it is not a failure of method for which administrators ought to apologize” (Lindblom 1959, 87). Perhaps the strongest statement is in Strategy of Decision: “we have not at all forgotten our intention of showing that these attributes interrelate in such a way as to constitute a systematic, understandable, and defensible strategy for problem solving generally and for evaluation in particular” (1963, 91).

14. Th ere is again the basic versus applied science ambiguity here—the “science” of muddling through could mean a systematic description of a chaotic process— but the text, which is respectful of administrators working on hard problems, makes clear that that is not the argument.

15. Th esis 5 was anticipated by Friedrich Hayek. Lindblom does not cite Hayek in Intelligence, but he studied under Frank Knight at the University of Chicago, so he must have read Hayek. See Adelstein (1992) for a discussion of the similari- ties between Hayek’s thinking and Lindblom’s.

16. Here a key word, much used in 1959 and 1963, appears: the method is “compre- hensive” (Lindblom 1959, 80–81; 1963, 37–39).

17. For example, suppose that a logistics offi cer wants to fi nd the shortest path for 40 supply depots. Th is problem exhibits combinatorial explosion: there are 40! possible paths—more than 1047. Hence, although problem (a) does not

Th e assessment of his positive applied theory is more mixed. Th esis 3(a) is on solid ground: there is no doubt that in complex situations even very experienced offi cials use simplifying heuristics, includ- ing the major ones identifi ed by “Muddling Th rough.” Th e jury is still out on thesis 3(b): researchers in a variety of fi elds continue to investigate how well these heuristics perform. Th esis 5, the benefi ts of multiple organized minds, is a hot topic in fi elds from organiza- tion theory to computer science, and it seems to be in good shape.

But thesis 4, the claim that disjointed incrementalism is an inte- grated strategy, is shaky. A strong version of the thesis is almost certainly wrong. Although disjointed incrementalism’s major ele- ments—the key heuristics—are fl ourishing today46 in a diverse array of applied fi elds, it is no longer an integrated and identifi able body of ideas. Indeed, contrary to a central view of “Muddling”—that disjointed incrementalism and the synoptic method are diametri- cally opposed packages of tightly intertwined elements—ingenious eff orts at recombination have shown that subroutines from each approach can cross-breed and produce viable off spring. Applied researchers will continue to generate these hybrids.

We should not regret the disappearance of the strategy of incremen- talism. Regarding it as a coherent approach to decision making or problem solving was a mistake. It is more accurate and more useful to see it as a toolkit of heuristics that can be deployed separately and combined in various ways. Indeed, this set of clever tricks is, as Lindblom noted, only part of a much larger toolkit; hence, its major elements can be combined and recombined with a diverse array of subroutines.47

We have a much richer set of options for solving hard problems and making diffi cult choices than “Muddling Th rough” suggested. Th at is good news.

Acknowledgments I thank Holke Brammer for valuable research assistance, Ned Woodhouse for enlightening insights into Lindblom’s thinking, Steve Callander and Ken Shotts for helpful comments on the rela- tion between game theory and partisan analysis, Richard Adelstein for sending me his thoughtful paper on Lindblom’s ideas, and Ed Lindblom for several stimulating points about applied research.

Notes 1. I will often use Lindblom’s term—“disjointed incrementalism”—as the name

for his decision strategy. Sometimes, however, I follow common practice and abbreviate it to “incrementalism.”

2. Certainly, the present author did. For a long time, I did not see the diff erence between the applied nature of Lindblom’s work and the basic science character- istics of Simon’s. I was not alone: many scholars in public administration have regarded Lindblom and Simon as doing essentially the same kind of work.

3. I am using “basic” and “applied” in conventional ways. Basic sciences (e.g., chemistry, physiology, political science) are oriented toward knowledge for its own sake—to satisfy our curiosity about the physical, biological, or social world. A particular body of knowledge may be useless (solves no practical problem); that does not aff ect its standing in the relevant basic science. In contrast, research in applied sciences (e.g., engineering, medicine, public administration) focuses on projects that could help us build planes, cure diseases, or teach children how to read. Hence, “applied theory” refers to theories that are supposed to help us solve practical problems.

Incrementalism: Dead yet Flourishing 203

(Bendor 2003, 435), especially when thinking about computational optimization, but a common one.

32. Corresponding descriptive theories of bounded rationality might predict, prior to the adaptation, that the constraints will become manifest somewhere (Bendor 2003): Simon’s scissors will bind at some stage. Th is, however, is consistent with the possibility that simplifying the problem in one respect suffi ces.

33. Clearly, they were thinking along similar lines in the 1950s and 1960s. Consider Lindblom’s crisp statement of the fundamental problem that Simon tackled throughout his career: “Th e piecemealing, remedial incrementalist or satisfi cer may not look like an heroic fi gure. He is nevertheless a shrewd, resourceful problem-solver who is wrestling bravely with a universe that he is wise enough to know is too big for him” (1968, 27).

34. Innovations in problem-solving methods can make some methods obsolete— dominated by new ones that are both more powerful and more general. But tinkering will produce specialized improvements that re-create a set of undomi- nated methods.

35. A corollary: there are no magicians. Th is corollary has signifi cant implications for the value of distributed intelligence (thesis 5).

36. Th is will lead to “if … then . . .” rules (a mark of a well-developed applied sci- ence) rather than simplistic unconditional injunctions of “Do this!”

37. In computer science, ideas such as local search are considered metaheuristics (Blum and Roli 2003), a term that emphasizes their domain generality.

38. Clearly, Wildavsky started a research program in budgetary politics. Th is, however, was mainly a basic science program, and its links to Lindblom’s applied project have weakened over time.

39. Replies of “it’s all tacit knowledge” will not do, for several reasons. First, Lindblom argued that he was codifying what administrators already do (1959, 81). Codifying makes explicit what had been implicit. Second, codifi cation signifi cantly increases the chance of establishing a research program in today’s professional schools. Th eir students also acquire craft knowledge, sometimes informally transmitted. But textbooks matter.

40. It is reasonable to see social experimentation and program evaluation as specifi c ways of implementing incrementalism’s general heuristics. (Th at Lindblom accepted the job of director of Yale’s Institution of Social and Policy Studies, which does policy analysis and evaluation, is a relevant biographical fact.)

41. One should not overstate this. Lindblom was too hardheaded to believe in magical methods of problem solving. Th e comments in the text are addressed to someone who strongly believes, even now, in the power of disjointed incrementalism.

42. “[T]here is the plain fact that, like Winnie-the-Pooh, [man] is an animal of very little brain. Th e number of alternatives man would need to consider in order to act rationally is very often far beyond his limited mental capacity” (Dahl and Lindblom 1953, 60). Adelstein (1992) alerted me to this statement.

43. Lindblom also passed up the opportunity to use a modern empirical method— experiments—that is well suited for probing thesis 3(b). Contemporaneously, Simon, Newell and colleagues at Carnegie Mellon University were demonstrat- ing the power of this method to study cognitively constrained problem solving. But few social scientists were doing experiments in the sixties; that Lindblom did not try this method is not surprising.

44. It is signifi cant that Simon turned toward basic science after fi nishing Administrative Behavior. Even the textbook Public Administration “is concerned chiefl y with how people behave in organizations rather than with specifi c recom- mendations for ‘better’ organization” (Simon, Smithburg, and Th ompson 1950, 22). Organizations (March and Simon 1958) is explicitly a work of basic social science. His joint work with Allen Newell, which culminated in Human Problem Solving (Newell and Simon 1972), focused on fundamental research questions. Th us, whereas Lindblom plugged away at applied theory, Simon took what was in some ways an easier path. (Th at worked out well: he and Newell did establish a research program, but in basic science—the cognitive psychology of problem solving—not applied.)

arise—the goals defi ning the problem are clear and consensual—implement- ing the conventional method is impossible even with high-speed computers. (Lindblom was probably unaware of theoretical work in computer science on “hard” problems, but his hunches here were excellent.)

18. Braybrooke and Lindblom use checkers as an example of an extremely hard problem (1963, 250 n. 18). Yet, thanks to progress in software and hardware, checkers has been solved: it is a draw (Schaeff er et al. 2007). But we should not play “gotcha” here; see their insightful remarks on the moving boundary of computational infeasibility (246–47 n. 8).

19. Th e issue is whether a class of diffi cult problems, called NP-hard, can be reduced to an easier class, called P problems because they can be solved in polynomial time. For an accessible introduction to this question, see Fortnow (2009).

20. Although it is an open question whether NP-hard problems can be reduced to P problems, many experts in theoretical computer science believe that they cannot (Fortnow 2009, 2).

21. Lindblom later said that disjointed incrementalism was but one type of what he called “strategic analysis” (1979, 518). Unfortunately, this concept was never precisely defi ned.

22. Not always, however: his seminal article states that muddling through “will be superior to any other decision-making method available for complex problems in many circumstances, certainly superior to a futile attempt at superhuman comprehensiveness” (1959, 88).

23. To be fair, it is sensible to be cautious about heuristics that one believes will work “pretty well.”

24. He even asserted that “it is not a reasonable objection to mutual adjustment that I cannot show directly that its outcomes are superior to other outcomes. As with the most social processes, we argue the superiority of the outcome from the proc- ess, not of the process from the outcome” (Lindblom 1961, 323). Th is position is dangerous: it reduces an applied theory’s empirical accountability.

25. Lindblom’s beliefs about the integrated nature of disjointed incrementalism also infl uenced his empirics: “If the adaptive methods are to be seen combined as a systematic strategy, we must show how they interconnect in single pieces of policy analysis; hence we are precluded from relying for illustration solely on scattered examples from diff erent pieces of analysis” (Braybrooke and Lindblom 1963, 83).

26. Th is possibility indicates why Lindblom’s modal empirical approach—describing examples in which multiple incrementalist components were used and appar- ently worked well together—cannot provide strong evidence for the claim that these heuristics are mutually reinforcing: if we looked for examples in which they worked well with nonincrementalist procedures, perhaps we would fi nd them, too.

27. For example, Lindblom hypothesized that evaluating alternatives similar to the status quo improves the estimates of their value. A benefi t of multiple minds is to make policy evaluation more accurate (Condorcet’s jury theorem). Hence, the heuristics of local search and distributed intelligence are partly substitutes, and it has been shown that if there are suffi ciently many decision makers, nonlocal search is better than local (Bendor 1995, 829).

28. Th at conventional and incrementalist subroutines might complement each other makes sense: the former are attuned to a decision maker’s external environment (a problem’s objective “shape”); the latter, to her internal environment (Simon 1969).

29. Stating that a decision maker used a procedure of computational optimization by itself implies nothing about that actor’s cognitive constraints (bounded rational- ity). Th ese issues, though related, are not equivalent.

30. It is highly regarded by computer scientists and applied mathematicians who design search algorithms for practical problems: typing “local search” together with “optimization” yields 5,210,000 results in Google (June 5, 2014).

31. Is this truly hybrid? Some might say no: Lindblom did not specify a choice rule in “Muddling Th rough.” But the choice rule has always been important in the debate between partisans of rational choice and of bounded rationality. Indeed, the debate is often framed as optimizing versus satisfi cing—a big mistake

204 Public Administration Review • March | April 2015

Campbell, Donald, and Julian Stanley. 1963. Experimental and Quasi-Experimental Designs for Research in Teaching. In Handbook of Research on Teaching, edited by N. L. Gage, 171–246. Chicago: Rand McNally.

Dahl, Robert A., and Charles E. Lindblom. 1953. Politics, Economics, and Welfare: Planning and Politico-Economic Systems Resolved into Basic Social Processes. New York: Harper.

Dunn, Delmer D. 1988. Th e Impact of Administrative Behavior on Public Administration Textbooks. Public Administration Quarterly 12(3): 369–84.

Ericsson, K. A., and A. C. Lehmann. 1996. Expert and Exceptional Performance: Evidence on Maximal Adaptations on Task Constraints. Annual Review of Psychology 47: 273–305.

Etzioni, Amitai. 1967. Mixed-Scanning: A “Th ird” Approach to Decision-Making. Public Administration Review 27(5): 385–92.

Fortnow, Lance. 2009. Th e Status of the P versus NP Problem. Communications of the ACM 52(9): 78–86.

Friedman, Susan Lindblom. 2010. Th e Story of Charles Edward Lindblom. Santa Fe, NM: Lulu Press.

Gawande, Atul. 2011. Th e Checklist Manifesto: How to Get Th ings Right. New York: Picador.

Ladha, Krishna. 1992. Th e Condorcet Jury Th eorem, Free Speech, and Correlated Votes. American Journal of Political Science 36(3): 597–634.

———. 1995. Information Pooling through Majority-Rule Voting: Condorcet’s Jury Th eorem with Correlated Voting. Journal of Economic Behavior and Organization 26(3): 353–72.

Landau, Martin. 1960. Th e Concept of Decision-Making in the Field of Public Administration. In Concepts and Issues in Administrative Behavior, edited by Sidney Mailick and Edward H. Van Ness, 1–28. Englewood Cliff s, NJ: Prentice Hall.

———. 1969. Redundancy, Rationality, and the Problem of Duplication and Overlap. Public Administration Review 29(4): 346–58.

Landemore, Hélène. 2012. Democratic Reason: Politics, Collective Intelligence, and the Rule of the Many. Princeton, NJ: Princeton University Press.

Lewis, Verne B. 1952. Toward a Th eory of Budgeting. Public Administration Review 12(1): 42–54.

Lindblom, Charles E. 1958. Policy Analysis. American Economic Review 48(3): 298–312. ———. 1959. Th e Science of “Muddling Th rough.” Public Administration Review

19(2): 79–88. ———. 1961. Decision-Making in Taxation and Expenditure. In Public Finance:

Needs, Sources, and Utilization, National Bureau of Economic Research, 295–336. Princeton, NJ: Princeton University Press.

———. 1965. Th e Intelligence of Democracy: Decision Making through Mutual Adjustment. New York: Free Press.

———. 1968. Th e Policy-Making Process. Englewood Cliff s, NJ: Prentice Hall. ———. 1979. Still Muddling, Not Yet Th rough. Public Administration Review 39(6):

517–26. Lindblom, Charles E., and David K. Cohen. 1979. Usable Knowledge: Social Science

and Social Problem Solving. New Haven, CT: Yale University Press. March, James G., and Herbert A. Simon. 1958. Organizations. New York: Wiley. Martin, Roscoe C. 1952. Political Science and Public Administration. American

Political Science Review 46(3): 660–76. Merton, Robert K. 1961. Singletons and Multiples in Scientifi c Discovery: A

Chapter in the Sociology of Science. Proceedings of the American Philosophical Society 105(5): 470–86.

Miller, Nicholas. 1986. Information, Electorates, and Democracy: Some Extensions and Interpretations of the Condorcet Jury Th eorem. In Information Pooling and Group Decision Making, edited by Bernard Grofman and Guillermo Owen, 173–92. Greenwich, CT: JAI Press.

Newell, Allen. 1969. Heuristic Programming: Ill-Structured Problems. In Progress in Operations Research, vol. 3, edited by Julius Aronofsky, 360–414. New York: Wiley.

45. A prescient footnote in Strategy of Decision (1968, 246–47) discusses this issue in detail.

46. “Muddling Th rough” was widely read partly because Lindblom had superb judgment about which heuristics were worth analyzing. Local search, seriality, and distributed intelligence are vital tools for hard problems. Th ey will be used indefi nitely in many fi elds.

47. I am not stating that Lindblom’s positive applied theory has vanished because “we are all incrementalists now.” Th at would be a gracious way to honor one of our most distinguished scholars, but it is bad intellectual history. It makes two mis- takes. First, it is inaccurate about incrementalism in public administration: if being an incrementalist means embracing the package of heuristics as the way to make public policy, then today many public administration scholars are not incremental- ists. Nor should they be: asserting that the incrementalist package dominates all others is surely wrong. (Among other things, it violates Newell’s Law.) Second, it wildly overestimates Lindblom’s impact on other fi elds. Th ere is no evidence that computer scientists or applied mathematicians working on optimization via local search have read anything he wrote. Th ey have their own intellectual traditions, and local search is such a useful and natural heuristic that it was independently invented in several scientifi c communities—another case of “multiples” in scientifi c discovery (Merton 1961). Th e same holds for the other key heuristics of disjointed incrementalism. (See Landemore 2012 for a lucid description of several intellectual streams that have contributed to the idea of distributed intelligence.)

References Adelstein, Richard. 1992. Charles E. Lindblom. In New Horizons in Economic

Th ought: Appraisals of Leading Economists, edited by Warren J. Samuels, 202–26. Cheltenham, UK: Edward Elgar.

Anderson, John D., Jr. 1997. A History of Aerodynamics and Its Impact on Flying Machines. New York: Cambridge University Press.

Arrow, Kenneth J. 1964. Review of A Strategy of Decision: Policy Evaluation as a Social Process, by David Braybrooke and Charles E. Lindblom. Political Science Quarterly 79(4): 584–88.

Bendor, Jonathan. 1985. Parallel Systems: Redundancy in Government. Berkeley: University of California Press.

———. 1994. Th e Fields of Bureaucracy and Public Administration: Basic and Applied Research. Journal of Public Administration Research and Th eory 4(1): 27–39.

———. 1995. A Model of Muddling Th rough. American Political Science Review 89(4): 819–40.

———. 2003. Herbert A. Simon: Political Scientist. Annual Review of Political Science 6: 433–71.

———. Forthcoming. Aspiration-Based Models of Politics. In New Explorations in Political Psychology, edited by Jon Krosnick. New York: Psychology Press.

Bendor, Jonathan, and John G. Bullock. 2008. Lethal Incompetence: Voters, Offi cials, and Systems. Critical Review 20(1–2): 1–24.

Bendor, Jonathan, and Jacob Shapiro. 2014. Th e Evolution of Modern Military Problem-Solving. Unpublished manuscript.

Bloor, David. 2011. Th e Enigma of the Aerofoil: Rival Th eories in Aerodynamics, 1909–1930. Chicago: University of Chicago Press.

Blum, Christian, and Andrea Roli. 2003. Metaheuristics in Combinatorial Optimization: Overview and Conceptual Comparison. ACM Computing Surveys 35(3): 268–308.

Braybrooke, David, and Charles E. Lindblom. 1963. Th e Strategy of Decision: Policy Evaluation as a Social Process. New York: Free Press.

Brown, Peter C., Henry L. Roediger, and Mark A. McDaniel. 2014. Make It Stick: Th e Science of Successful Learning. Cambridge, MA: Belknap Press of Harvard University Press.

Campbell, Donald T. 1969. Reforms as Experiments. American Psychologist 24(4): 409–29.

Incrementalism: Dead yet Flourishing 205

———. 1969. Sciences of the Artifi cial. Cambridge, MA: MIT Press. ———. 1976. From Substantive to Procedural Rationality. In Method and Appraisal

in Economics, edited by Spiro Latsis, 129–48. Cambridge, UK: Cambridge University Press.

———. 1978. Rationality as Process and Product of Th ought. American Economic Review Papers and Proceedings 68(2): 1–16.

Simon, Herbert A., and Jonathan Schaeff er. 1992. Th e Game of Chess. In Handbook of Game Th eory with Economic Applications, edited by Robert J. Aumann and Sergiu Hart, vol. 1, 1–17. Amsterdam: North-Holland.

Simon, Herbert A., Donald W. Smithburg, and Victor A. Th ompson. 1950. Public Administration. New York: Alfred A. Knopf.

Ting, Michael M. 2003. A Strategic Th eory of Bureaucratic Redundancy. American Journal of Political Science 47(2): 274–92.

Tversky, Amos. 1972. Elimination by Aspects. Psychological Review 79(4): 281–99. Weitzman, Martin. 1998. Recombinant Growth. Quarterly Journal of Economics

113(2): 331–60. Wildavsky, Aaron. 1961. Political Implications of Budgetary Reform. Public

Administration Review 21(4): 183–90. ———. 1964. Th e Politics of the Budgetary Process. Boston: Little, Brown. Woodhouse, Edward, and David Collingridge. 1993. Incrementalism, Intelligent

Trial and Error, and the Future of Political Decision Th eory. In An Heretical Heir of the Enlightenment: Politics, Policy, and Science in the Work of Charles E. Lindblom, edited by Harry Redner, 131–54. Boulder CO: Westview Press.

———. 1980. One Final Word. In Problem Solving in Education: Issues in Teaching and Research, edited by D. T. Tuma and F. Reif, 175–89. Hillsdale, NJ: Lawrence Erlbaum.

Newell, Allen, and Herbert A. Simon. 1972. Human Problem Solving. Englewood Cliff s, NJ: Prentice Hall.

Page, Scott E. 2007. Th e Diff erence: How the Power of Diversity Creates Better Groups, Firms, Schools, and Societies. Princeton, NJ: Princeton University Press.

Quiggin, John. 2010. Zombie Economics: How Dead Ideas Still Walk among Us. Princeton, NJ: Princeton University Press.

Schaeff er, Jonathan, Neil Burch, Yngvi Björnsson, Akihiro Kishimoto, Martin Müller, Robert Lake, Paul Lu, and Steve Sutphen. 2007. Checkers Is Solved. Science 317(5844): 1518–22.

Scriven, Michael. 1998. Minimalist Th eory: Th e Least Th eory Th at Practice Requires. American Journal of Evaluation 19(1): 57–70.

Sherwood, Frank. 1990. Th e Half-Century’s “Great Books” in Public Administration. Public Administration Review 50(2): 249–64.

Simon, Herbert A. 1947. Administrative Behavior. New York: Macmillan. ———. 1955. A Behavioral Model of Rational Choice. Quarterly Journal of

Economics 69(1): 99–118. ———. 1962. Th e Architecture of Complexity. Proceedings of the American

Philosophical Society 106(6): 467–82. ———. 1964. On the Concept of Organizational Goal. Administrative Science

Quarterly 9(1): 1–22.

Public Administration Review

is a professional journal for practitioners and scholars.

If you are a practitioner who is interested in writing a commentary for an article,

please email [email protected]. Also check out the information for contributors pdf

available under PAR at www.aspanet.org.

Copyright of Public Administration Review is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.