Bounded Rationality and Politics (Wildavsky Forum Series): Volume 6 - Hardcover

Bendor, Jonathan

Hardcover

ISBN 10: 0520259467 ISBN 13: 9780520259461

Publisher: University of California Press, 2010

Synopsis

In Bounded Rationality and Politics, Jonathan Bendor considers two schools of behavioral economics-the first guided by Tversky and Kahneman's work on heuristics and biases, which focuses on the mistakes people make in judgment and choice; the second as described by Gerd Gigerenzer's program on fast and frugal heuristics, which emphasizes the effectiveness of simple rules of thumb. Finding each of these radically incomplete, Bendor's illuminating analysis proposes Herbert Simon's pathbreaking work on bounded rationality as a way to reconcile the inconsistencies between the two camps. Bendor shows that Simon's theory turns on the interplay between the cognitive constraints of decision makers and the complexity of their tasks.

About this title

About the Author

Jonathan Bendor is Walter and Elise Haas Professor of Political Economics at the Graduate School of Business, Stanford University.

From the Back Cover

"Bendor's Bounded Rationality and Politics provides an adept and illuminating critique of existing theories while also introducing new models and concepts that are sure to remain part of the conversation for generations to come. This book will reinvigorate the field of political science."--Daniel P. Carpenter, Harvard University

"Bendor's scholarship is top drawer. Excellent. These essays are not only intellectually deep, but also engaging and powerful."--Scott Page, University of Michigan

From the Inside Flap

"Bendor's Bounded Rationality and Politics provides an adept and illuminating critique of existing theories while also introducing new models and concepts that are sure to remain part of the conversation for generations to come. This book will reinvigorate the field of political science." Daniel P. Carpenter, Harvard University

"Bendor's scholarship is top drawer. Excellent. These essays are not only intellectually deep, but also engaging and powerful." Scott Page, University of Michigan

Excerpt. © Reprinted by permission. All rights reserved.

Bounded Rationality and Politics

By Jonathan Bendor

UNIVERSITY OF CALIFORNIA PRESS

Copyright © 2010 The Regents of the University of California
All rights reserved.

ISBN: 978-0-520-25946-1

Contents

List of Figures, ix, 
Preface, xi, 
1. Introduction Jonathan Bendor, 1, 
2. Herbert A. Simson: Political Scientist Jonathan Bendor, 11, 
3. Satisficing: A Pretty Good Heuristic Jonathan Bendor, Sunil Kumar, and David A. Siegel, 48, 
4. A Model of Muddling Through Jonathan Bendor, 61, 
5. The Perfect Is the Enemy of the Best: Adaptive versus Optimal Organizational Reliability Jonathan Bendor and Sunil Kumar, 93, 
6. Garbage Can Theory Jonathan Bendor, Terry Moe, and Ken Shotts, 119, 
7. Institutions and Individuals Jonathan Bendor, 163, 
Notes, 183, 
References, 207, 
Index, 225, 

CHAPTER 1

Introduction

JONATHAN BENDOR

There are two main orientations toward bounded rationality (BR) in political science. The first orientation sees the glass as half full, emphasizing that decision makers often manage to do "reasonably well"—even in complex tasks—despite their cognitive limitations. Virtually all of Simon's work and also the theory of "muddling through" (Lindblom 1959; Braybrooke and Lindblom 1963) belong to this branch, which we can call the problem-solving approach. In the second orientation the glass is half empty: the emphasis is on how people make mistakes even in simple tasks. Most of the research on heuristics and biases, following Tversky and Kahneman's pioneering work (1974), belongs here.

Prominent early use of the problem-solving approach can be found in Aaron Wildavsky's studies of budgeting. In, for example, The Politics of the Budgetary Process, he devotes much space to showing how and why making resource allocation decisions in the federal government is beset by complexities and how the professionals cope with their difficult tasks: "It [is] necessary to develop mechanisms, however imperfect, for helping men make decisions that are in some sense meaningful in a complicated world" (1964, p. 11). One might argue that his orientation was due simply to the time paths of these different intellectual currents: Simon and Lindblom had launched the problem-solving branch before Aaron wrote his pioneering book on budgeting, whereas the Tversky-Kahneman branch didn't get started until nearly a decade later. But there is a deeper explanation. Aaron did field research on federal budgeting, including 160 interviews with "agency heads, budget officers, Budget Bureau staff, appropriations committee staff, and Congressmen" (1964, p. v). He was not interested in how experimental subjects committed errors of judgment or choice in laboratory settings; he was interested in how real decision makers tackled problems of great complexity. Hence, he was intrigued by how they managed to do this extremely difficult task reasonably well. One sees in the book a respect for the decision makers, arising in large measure from an appreciation of the difficulty of the tasks they confronted.

Indeed, I suspect that the seriousness with which Aaron thought about the tasks of budgetary officials was part of a long-standing theme of his professional life: a passionate interest in the real-world problems confronting government officials in a modern society. (Helping to found Berkeley's Graduate School of Public Policy was another reflection of this theme.)

This is more than biographical detail. It also illustrates an important—though neglected—part of the problem-solving approach to bounded rationality: a close examination of decision makers' tasks. In Simon's pioneering formulation, the focus was always on a comparison between a decision maker's mental abilities and the complexity of the problem he or she faces: for example, "the capacity of the human mind for formulating and solving complex problems is very small compared with the size of the problems whose solution is required for objectively rational behavior in the real world—or even for a reasonable approximations to such objective rationality" (1957, p. 198). Thus, for Simon, as for Wildavsky, the idea of bounded rationality is not a claim about the brilliance or stupidity of human beings, independent of their task environments. Many social scientists miss this central point and reify the idea of BR into an assertion about the absolute capacities of human beings. The fundamental notion here is that of cognitive limits; and, as for any constraint, if cognitive constraints do not bind in a given choice situation, then they will not affect the outcome. And whether they bind depends vitally on the information-processing demands placed on the decision makers by the problem at hand. More vividly, Simon has called the joint effects of "the structure of task environments and the computational capacities of the actor ... a scissors [with] two blades" (1990, p. 7): Theories of BR have cutting power—especially when compared to theories of (fully) rational choice—only when both blades operate. Thus, any analysis that purports to fall into this branch of the research program yet examines only the agent's properties is badly incomplete.

Thus, Wildavsky not only belonged squarely in the problem-solving branch of the BR program; his intellectual propensities—his interest in how real officials tackle real problems of great complexity—predisposed him to use both blades of Simon's scissors. That was unusual. It was also productive: many of his insights about budgeting flowed from his effective use of Simon's scissors.

Of course, every research method focuses our attention on some scholarly questions in the domain at hand and deemphasizes others in that same domain. (Lindblom's warnings [1959] about the utopian folly of trying to be comprehensive apply to academics as well as to government officials.) So it is not surprising that Wildavky's research methods led him to ignore certain topics. In particular, his interest in applying the basic ideas of bounded rationality to the study of real-world budgeting steered him away from analyzing the foundations of BR theory. That simply was not part of his intellectual agenda. But a serious focus on those foundations is long overdue. Brilliant as they were, neither Simon nor Lindblom said it all. We political scientists—particularly those of us who work on the behavioral (bounded rationality) side—have done too much quoting and too little reworking. I believe that we will see vigorous scientific competition between rational choice (RC) theories of policy making and behavioral theories only if behavioralists take the foundations of their theories as seriously as RC theorists take theirs. Further, I think that this entails transforming verbal theories into mathematical models. (For an argument on this point in the context of incrementalism, see chapter 4.)

The next section surveys a family of theories that has been central to the problem-solving branch of the BR program: those that use the idea of aspiration levels as a major concept.

THEORIES OF ASPIRATION-BASED PROBLEM REPRESENTATION AND CHOICE

The main claim I offer in this section is that the idea of aspiration-based choice constitutes a major family of theories in the bounded rationality research program. The word family matters: I think it is a serious mistake to view satisficing per se as an alternative to theories of optimization. As careful scholars working in the optimization tradition have often pointed out, there is no single RC theory of (e.g.) electoral competition (see, e.g., Roemer's comparison [2001] of Downsian theory to Wittman's), much less just one RC theory of politics. Similarly, satisficing is a theory of search. It is not the Behavioral Theory of Everything. Moreover, a key part of satisficing—the idea of aspiration levels—is shared by several other important behavioral theories: theories of reinforcement learning (Bush and Mosteller 1955) and prospect theory (Kahneman and Tversky 1979). So I first argue that a "family" of theories is a significant grouping that fits into the more conventional hierarchy of research program, theories, and models, and that substantively we can gain some insight by focusing our attention on this common feature of aspirations.

(A reason that it is methodologically important to identify this family of theories is that the size of this set is probably indefinite. That is, an indefinitely long list of choice problems may be representable via aspiration levels. I see no reason why this should not hold. All that is required of the choice problem is that there be more than two feasible payoffs, but the agent simplifies the problem by reducing that complex set into two simple equivalence classes. Importantly, there is no restriction on the substantive type of problem, or the task of the decision maker, which can be represented this way.)

Technically, an aspiration level is a threshold in an agent's set of feasible payoffs. This threshold partitions all possible payoffs into two disjoint sets: those below the threshold and those that are greater than or equal to the threshold. In all aspiration-based theories, this dichotomous coding matters: that is, important further implications flow from this representation of the choice problem. The details of these implications vary, for they naturally depend on the substantive nature of the theory at hand, as we will see shortly. This parallels how the details of optimal strategies vary, depending on the type of problem confronting the decision maker. Strategies of optimal candidate location in a policy space look quite different from strategies of optimal nuclear deterrence. But they share a common core, that of optimal choice. Similarly, psychological theories of learning look quite different from prospect theory, but they too have a common core. Interestingly, we will see that early on, scholars working on certain members of this family of theories were not even aware that their particular theories required, as a necessary part of their conceptual apparatus, the idea of aspirations; they backed into this idea. Let me now briefly describe several important members of this family of theories: satisficing and search, reinforcement learning, and prospect theory.

Theories of Aspiration-Based Behavior

Search Search was Simon's original context for satisficing. The idea is simple. Simon posited that when an agent looked for, say, a new job or house, he or she had an aspiration level that partitioned candidates into satisfactory options and unsatisfactory ones. As soon as the decision maker encountered a satisfactory one, search ended. The verbal theory suggests that the aspiration level adjusts to experience, falling in bad times (when one searches without success) and rising in good times (swiftly encountering something that far exceeds the aspiration level), but this was not represented by the formal model. (Cyert and March [1963] allow aspirations to adjust to experience in their computational model.)

There is, however, a rival formulation: optimal search theory. The basic idea is that agents assess both the expected marginal gains and the expected marginal costs from searching further and set an optimal stopping rule that equates the two. Some behavioralists (e.g., Schwartz et al. 2002) are completely unaware that this rival exists. This is a pity. The ignorance allows such behavioralists to have scholarly aspirations that are too low: they think satisficing easily beats RC theories because the latter predict that decision makers search all options exhaustively before making a choice. Since this is obviously false in many (most?) domains, satisficing (hence BR, etc.) comes out on top in this scientific competition. But this horse race was bogus: optimal search theory does not, in general, predict exhaustive search. Indeed, in many environments the optimal stopping rule takes the form of a cutoff: if the agent stumbles on an option worth at least v, stop searching; otherwise, continue. Significantly, the main qualitative features of this prediction are exactly the same as those of the satisficing theory: the set of feasible options is partitioned into two subsets, and the searcher stops looking upon finding something belonging to the better subset. Clearly, then, some additional cleverness is required in order to derive different predictions from RC and BR theories of search. It's doable, but we can't stop with the obvious predictions.

Now that the notion of optimal search has been spelled out, it is clear that in most search problems the stopping rule could be either suboptimally low or suboptimally high. If the costs of search are sufficiently low, then one should keep searching until one has found the highest-quality option, but this will rarely be the case in policy making, particularly not when decision makers are busy (Behn and Vaupel 1982).

Interestingly, however, the possibility that uncalculated aspirations may be too high has gone almost unnoticed by the behavioral literature. Indeed, an auxiliary premise has been smuggled into the concept of satisficing: it is virtually defined as search with a suboptimally low threshold (as in the phrase that probably most of us have heard, "merely satisfice").

The problem is not with the definition per se—one can stipulate a technical term as one pleases—but with its uses. This implicit smuggling of an important property into the heart of satisficing helps us to overlook the possibility of excessively high aspirations. It reinforces the mistake of equating "optimal" with "best quality," or worse, assuming that optimal equals perfect. (Chapters 3 and 5 analyze the implications of equating optimal with perfect.) We thereby neglect some of the empirical content of aspiration-based models of adaptation.

Learning In experiments on learning, psychologists often give subjects a set of options that they can repeatedly try. In so-called bandit problems, every option either pays off some fixed amount v > 0 or yields nothing. Some options are better than others—pay v with a higher probability—but subjects aren't told which. Instead, they must learn which options are better.

Originally, these experiments were part of the behaviorist research program in psychology, which eschewed mentalistic concepts such as aspirations. However, learning theorists in psychology discovered (the hard way) that they needed this concept to explain the behavior of subjects.

This issue becomes more pressing in choice situations where there are more than two payoffs. Given only two payoffs, it is quite natural to hypothesize that subjects will regard getting something as a success while getting nothing is a failure. But many choice situations do not provide such an obvious coding. In, for example, the two-person prisoner's dilemma, is the payoff to mutual cooperation reinforcing? How about the payoff to mutual defection?

Prospect Theory Perhaps the best-known postulate of prospect theory (Kahneman and Tversky 1979) is that decision makers are risk averse regarding gains but risk seeking regarding losses. This is not, however, a good way to remember the theory. Its fundamental axiom—its most important departure from classical utility theory—is that people evaluate outcomes relative to a reference point. Indeed, under the classical theory, the claim that people are, for example, risk averse about gains makes no sense: the idea of a "gain" has no place in the conceptual apparatus in standard utility theory. Decision makers simply have preferences over baskets of consequences; that's all that matters. They do not compare baskets to an internal standard of goodness or acceptability.

A reference point is, in effect, an aspiration level. Of course, aspirations in prospect theory have a different function than they do in satisficing-and-search theory. (Prospect theory has not been applied to search problems, as far as I know.) Rather than serving as a stopping rule, aspirations in the context divide the set of feasible outcomes into those coded as gains and those coded as losses. But again we see a dichotomizing of payoffs into two qualitatively different subsets.

An Important Problem: The Empirical Content of Aspiration-Based Theories

Although I think that aspiration-based theories form a tremendously important family in the BR research program, no set of theories in the social sciences is free of problems. (At least, I have not been lucky enough to encounter such a set!) And since I completely agree with Martin Landau's view that scientific progress depends tightly on criticism, I think it is vital for scholars who work on these theories to detect their weaknesses and work on them. I do not think that aspiration-based theories of choice are perfect. That would be a reflexively bizarre claim.

(Continues...)

Excerpted from Bounded Rationality and Politics by Jonathan Bendor. Copyright © 2010 The Regents of the University of California. Excerpted by permission of UNIVERSITY OF CALIFORNIA PRESS.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.

Edition details

• Publisher: University of California Press
• Publication date: 2010
• Language: English
• ISBN 10: 0520259467
• ISBN 13: 9780520259461
• Binding: Hardcover
• Number of pages: 240