Bend or and Hammond 1992 A PSR
Bend or and Hammond 1992 A PSR
Bend or and Hammond 1992 A PSR
net/publication/259427305
CITATIONS READS
209 3,844
2 authors:
All content following this page was uploaded by Thomas H. Hammond on 01 September 2015.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact [email protected].
American Political Science Association is collaborating with JSTOR to digitize, preserve and extend access to The American
Political Science Review.
http://www.jstor.org
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2 June 1992
MODELS
ALLISON'S
RETHINKING
JONATHAN BENDOR Stanford University
THOMAS H. HAMMOND Michigan State University
T he ideas in Graham Allison's Essence of Decision (1971) have had an enormous impact on
the study and teaching of bureaucracy and foreign policy making. While Allison's work has
received considerable critical attention, there has been surprisingly little examination of the
content and internal logic of his models. We subject each of Allison's three models to a systematic
critical analysis. Our conclusion is that the models require substantial reformulation.
G raham Allison's study of the 1962 Cuban each as he could, and to develop alternative explanations
missile crisis, published as "Conceptual Mod- for this event, which had the heuristic effect of leading
els and the Cuban Missile Crisis" (1969) and him to look for possibilities that had not occurred to
as Essence of Decision: Explaining the Cuban Missile analysts more nearly the prisoners of a single, and poorly
articulated, set of assumptions. These are, I think,
Crisis (1971), is a landmark in our discipline's analysis
among the important benefits to be derived from explicit
of bureaucracy's role in foreign policy making. Prior theory construction, and Allison's example is therefore
to Allison's work, most studies of bureaucracy and one well worth taking seriously. (pp. 446-47)
foreign policy, though rich and informative, were
largely descriptive and rather unfocused theoreti- As predicted in the reviews, Allison's approach has
cally. They seldom presented an explicit analytical indeed had a substantial impact on both the study
perspective, and readers were often left to infer what and the teaching of bureaucracy. His three models
the authors believed were the fundamental proper-
have been widely applied, and other scholars have
ties of bureaucratic policymaking.
followed his lead by developing additional models to
Allison's work demonstrated that a more self-
account for aspects of policymaking left unexplained
consciously theoretical approach to the study of bu-
by Allison's analyses. Moreover, Allison's book is
reaucracy and foreign policy was both feasible and
desirable. The earlier literature had rarely developed still widely used in courses on bureaucracy, public
alternative explanations for events and decisions. But administration, and organization theory. We have
in formulating his three models-the rational actor, frequently used it in our own classes; one of us
organizationalprocess, and governmentalpolitics models, (Hammond) is alone responsible for the sale of sev-
labeled I, II, and III, respectively-Allison showed eral hundred copies to students in the past 15 years.
how to make explicit the explanations of events that As Holsti's review suggested, the book is so useful
had only been implicit in many of these earlier because it shows students how to formulate and
studies. In so doing, Allison helped to place the study evaluate alternative explanations for political events
of bureaucracy's influence on foreign policy on a (1972, 137). Indeed, for courses dealing with bureauc-
more scientific foundation. racy and foreign policy there are few alternatives to
Allison's work received high praise in reviews by a Essence of Decision. Even now, over 20 years later, the
wide range of scholars such as Ole Holsti (1972), book remains heavily cited by scholars in a remark-
Francis Rourke (1972), and Harrison Wagner (1974). ably wide range of disciplines.2
For example, as Wagner remarked in his review of Allison's work did attract considerable critical at-
Essence of Decision: tention from the outset (e.g., Art 1973; Ball 1974;
Bobrow 1972; Caldwell 1977; Cornford 1974; Freed-
It may, in fact, be the most generally persuasive Perlmutter 1974; and
man 1976; Krasner 1972;
attempt ever made to show the relevanceof theorizingto
the treatment of what are ordinarilythought to be the Yanarella 1975). Even those praising the book (e.g.,
main problems of the study of internationalpolitics. Its Holsti 1972; Rourke 1972; and Wagner 1974) also
general appeal is primarily due to the effectiveness of expressed reservations. However, none of these early
Allison's theoretical arguments in his analysis of the criticisms comprehensively examined the internal
Cubanmissile crisis, which contains a numberof striking logic of all three of Allison's models, and it is pre-
insights and some information that no one else had cisely the logical structure of his work that is partic-
thought to look for. The results are a much more complex ularly beginning to show its age.
analysis of that event than anyone else has offered and a With the advantage of two decades of hindsight,
radicallydifferentunderstandingof what happened from much more is known about the properties of models
the one most people had previously adhered to. To the
extent that these considerablevirtues are the directresult of bureaucracy and policymaking. It is now clear that
of Allison's reflections on the theoreticalliteraturemen- alternative versions of Allison's models could and
tioned, they seem to be the product of his effortsto make should be developed to highlight somewhat more
these three paradigms as explicit as possible, enabling fundamental factors in policymaking. Since Allison's
him to derive as much insight from the assumptions of work still has wide currency in our discipline, the
301
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
time is ripe for a critical analysis of his models and most part, we will not address empirical issues raised
arguments. by Allison's analysis.
We make five arguments about Allison's work.
First, while a key purpose of modeling is to force the
analyst to clarify the assumptions on which the A TYPOLOGY OF MODELS
analysis will be based, it is often difficult to determine
for Allison's models just what the assumptions are. If Before plunging into a detailed analysis of Allison's
one cannot determine what is driving a model, its three models, it is advisable to distinguish one from
explanatory power cannot be accurately assessed. another by defining their key properties. As several
Second, the hypotheses drawn from a model should, early critics of Allison's work noted, however, this is
ideally, be logically derived from the initial assump- more difficult than one might think (see, e.g., Corn-
tions. But many of Allison's hypotheses seem, at ford 1974, 241; Wagner 1974). It seemed particularly
best, only loosely related to the models' initial as- difficult to disentangle Models II and III, Wagner
sumptions. Third, whatever the logical status of arguing that "it is not entirely clear whether Model III
Allison's propositions (are they rigorously derived is independent of Model II or an extension of it;
from his assumptions, or are they freestanding con- certainly bureaucratic bargaining seems to be con-
jectures?), it is possible to demonstrate, on strictly strained by many of the factors discussed under
logical grounds, that several key propositions in Model II, and many of Allison's readers seem to
Model II are simply incorrect. mingle the two together in speaking of the lessons of
his book" (1974, 448). Hence, we will begin by asking
Fourth, a well-crafted model must strike a balance
a very general and deceptively simple question about
between simplicity and complexity. Too simple a
Allison's three models: On what assumptions is each
model misses key aspects of the problem one is trying
one based?
to understand; too complex a model is analytically
Four different kinds of assumptions appear to be
intractable and yields few testable hypotheses. Of important, and they lead to a typology of models of
course, what is too simple and what is too complex is policymaking useful for characterizing Allison's three
partly a matter of what analytical technology is avail- models. The first class of assumptions concerns the
able. Nonetheless, even given what was available 20 number of actors. The classical approach to the study
years ago, we think that it is fair to say that Allison's of international relations postulates that the govern-
Model I is (and was) much too simple. On the other ment of a sovereign state behaves as if it were a
hand, we will also argue that Model III is, in some unitary actor. Allison's contrasting approach, exem-
ways, too complex. Finally, Allison's Models I, II, plified by his Models II and III, is to decompose the
and III are based on three bodies of literature (rational government into multiple actors. Note, however, that
choice theory, organization theory, and bureaucratic a shift to multiple actors immediately requires speci-
politics, respectively) that only specialists in these fication of whether the multiple actors have the same
fields know well. Specialists in other fields, such as or conflicting goals. Very different models-and very
foreign policy making, have often used Essence of different understandings of policymaking-follow
Decision as an introduction and guide to these litera- from different assumptions here. Though Allison is
tures. Unfortunately, Allison misinterpreted parts of less explicit here than he might be, he appears to
each of these three fields. In some cases, the misin- assume in both Models II and II that the actors' goals
terpretations were, in our judgment, quite substan- are in conflict. But even if actors have the same
tial. Thus, unsuspecting readers may be led into a objectives, this does not mean that policymaking is a
double error-believing that the models are better trivial problem. As any student of bureaucracy
constructed than they are and that they accurately knows, even when there is no serious conflict over
reflect the literatures on which they are based. goals, coordinating the actions of a large number of
We examine the various sets of assumptions on executive branch actors is no easy task. Thus, a model
which the models are based. These sets of assump- could logically-and quite meaningfully-stipulate
tions will be used to develop a typology of policy- multiple actors whose objectives are the same.
making models, thus providing a context for analyz- Hence, when we consider the number of actors
ing the models in Essence of Decision. We then probe posited by a model, a second type of assumption,
the logic of the three models and conclude by dis- namely, whether or not the actors' goals are the
cussing some further implications of our arguments. same, is necessarily involved. If a central goal of
We should emphasize that our general purpose Essenceof Decision is to understand the relative virtues
here is to examine the theoreticalaspects of Allison's of government-as-single-actor versus government-as-
work. There is, to be sure, a large and growing multiple-actors models, the results of the comparison
empirical literature on the Cuban missile crisis. Some will be misleading if this second kind of variable is
of this literature is quite new, drawing on evidence neglected.
only recently made available to the public (see Allyn, The third class of assumptions concerns the degree
Blight, and Welch 1989/90; Blight, Nye, and Welch of rationality attributed to decision makers. As Alli-
1987; Bouchard 1991, ch. 4; Brugioni 1991; Garthoff son stresses, the classical approach assumed that
1988, 1989; Hampson 1984/85; Lukas 1987; Trachten- decision makers are rational. In contrast, students of
berg 1985; and Welch and Blight 1987/88). But for the behavioral organization theory like Herbert Simon
302
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
A Typology of Policymaking
DECISIONMAKERS
la lb 2a 2b
Single Decisionmaker complete ' incomplete complete incomplete
information % information information information
3a ' 3b 4a 4b
Many Decisionmakers,
same goals complete ' incomplete
information ' information
complete
information
: incomplete
information
5a 5b 6a 6b
Many Decisionmakers,
conflictinggoals complete incomplete complete incomplete
information ' information information information
and James March, as well as political psychologists With this typology we can now attempt to charac-
like Robert Jervis and John Steinbruner, maintain that terize Allison's three models. Model I, the rational
all agents have limits to their cognitive capacities. actor model, clearly belongs in cell 1. Indeed, the
Though there is a continuum of possibilities here, we overwhelming bulk of Allison's discussion in Essence
can simplify by dichotomizing. A model could stipu- of Decision suggests placement in cell la, with its
late either that agents are perfectly or imperfectly assumption of a single, perfectly rational actor with
rational. complete information. We will soon argue, however,
Finally, a model may assume that decision makers that the complete information assumption neglects
are perfectly rational yet also postulate that they are what international relations theorists have identified
imperfectly informed, either about which outcomes as a central aspect of their field, namely, actors'
will occur (prospective uncertainty) or about what pervasive uncertainty about the capabilities, motives,
has already transpired (retrospective uncertainty).3 It and actions of others. Considerable insight, in other
seems intuitively obvious-and here explicit theoriz- words, is to be gained by extending Model I into cell
ing supports intuition-that either informational lb, the realm of incomplete information. Further
problem can strongly affect the predictions of models. insight into the workings of single-actor models could
Thus, the fourth kind of assumption concerns the be obtained by relaxing the assumption of perfect
amount of information attributed to decision makers. rationality, thus moving into cell 2. As examples we
We simplify by assuming that with "complete infor- might cite psychological analyses of individual
mation" there is neither retrospective nor prospective bounded rationality in foreign policy decision making
uncertainty. Thus, all players know the structure of (e.g., George 1980, chaps. 2-3; Jervis 1976).
the game (including payoffs), who has done what in At first glance, it might seem that Model II, the
the past, and the outcome that will result from any organizational process model, belongs in cell 4b.
set of moves. With "incomplete information" there After all, boundedly rational actors, facing uncertain
can be either retrospective or prospective uncer- environments and using simple adaptive strategies,
tainty, or both. are the heart of this model. However, at several
These four classes of assumptions can be combined places in his discussion of Model II Allison mentions
to yield the typology of models in Figure 1. The conflicting goals. For example, he refers to "parochial
typology will help us identify various possible mod- priorities," the "quasi-resolution of conflict," and
els of the policy process. To serve this purpose, a organizational "imperialism" (1971, 81-82, 93). He
typology's categories should be mutually exclusive even refers to agenda manipulation: "The short list of
(overlapping categories make identification uncer- alternatives [generated by an organization] reflects
tain) and collectively exhaustive, so that any candi- not only the cost of alternative generation but, more
date can be "typed" as an instance of some class. A important, each organization's interest in controlling,
quick inspection of this typology reveals that the cells rather than presenting, choices-for example, by
are indeed mutually exclusive and collectively ex- serving up one real alternative framed by two ex-
haustive. tremes" (p. 90). One can also find, in his discussion
303
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
of Cyert and March (1963), references to organiza- pie, Allison refers to foul-ups ("e.g., choices that are
tions as a "coalition of participants" and "bargaining not made because they are not recognized or are
among potential coalition members" (p. 76). Thus, raised too late, misunderstandings, etc." [pp. 145-
Model II might appear more properly to belong 46]); the limited attentional capacities of decision
somewhere else, perhaps in cell 6. makers (pp. 176-78); and problems of misperception,
As an empirical observation, an assumption of misexpectation, and miscommunication (pp. 178-79).
conflicting goals in an organizational process model is Of incrementalism in decision making, he writes
unassailable. For most important issues (virtually by "Analysis is drastically limited. Important policy out-
definition of important), organizations and their lead- comes are neglected.... By proceeding incremen-
ers disagree about objectives. However, we do think tally and comparing the results of each new policy
that some interesting hypotheses-particularly about with the old, actors reduce or eliminate reliance on
the kinds of problems involving bureaucratic coordi- theory" (p. 154). All these characteristics bear a
nation that Allison discusses in regard to Model strong family resemblance to the decision processes
II-could be generated by "withdrawing" into cell 3. of ordinary, imperfectly rational humans.
This class of models is ideal for examining pure In terms of our typology, then, Model II (which
coordination problems among perfectly rational ac- initially might seem to belong in cell 4, probably 4b)
tors. Anyone who thinks that pure coordination can also be classified in cell 6, perhaps even 6b.
problems are trivial should consider Schelling's ex- Model III, which might seem to belong somewhere in
amples, which only scratch the surface of coordina- cell 5 (probably 5b) can also be classified in cell 6,
tion problems (1960, 54-58). If different members of a perhaps even 6b. Thus, we find that Models II and
team observe different aspects of the environment III, which have different intellectual pedigrees and
and these environmental signals are noisy, formida- which Allison presumably intended to be distinct,
ble communication and division-of-labor problems apparently share much of the same analytical turf.
typically arise.4 Even in the context of goal consensus We can see at the outset, then, that our simple
and perfect rationality (i.e., each agent, like a perfect typology raises some serious questions about just
Bayesian statistician, extracts the maximal amount of what is driving each of Allison's models. We do
information from each noisy observation), there re- believe that Allison's initial instincts were on the right
main structural issues of who should communicate track. As our typology suggests, there is room for
what to whom.5 If we make the problem dynamic several different classes of models, and it is an essen-
(e.g., involving optimal adaptation to a changing tial enterprise to assess the relative explanatory and
environment), the complexities multiply-even for interpretative power of models in each of these
actors with the same goals. Understanding of these different classes. It is necessary, however, to specify
particular issues is probably hindered, not helped, by much more precisely what is assumed for each of
construction of models involving conflicting goals. these models. For this reason, Allison's subsequent
The heart of Model III is multiple decision makers enterprise of combining Models II and III into one
with conflicting goals-hence the third line of the overarching bureaucraticpolitics model (see Allison
typology. However, as Harrison Wagner pointed out, and Halperin 1972) is probably a step in the wrong
it is unclear whether this model assumes perfect or direction. Combining two theories before clarifying
bounded rationality: "Model III . . . contains no dis- their defining properties can easily produce inconsis-
cussion of behavioral assumptions at all. Yet one tent claims (Landau 1972, 227). Hence, we are in-
could presumably construct a theory of bureaucratic clined to follow Cornford's (1974) lead in arguing that
bargaining based on the decision theorists' assump- there are virtues in separating Models II and III and in
tions, and another based on Herbert Simon's" (1974, clarifying their internal logic prior to synthesizing.
448). We close by noting that identifying a particular
Allison's own summary of the core of Model III model as belonging, say, to cell la does not necessar-
(1971, 162) emphasizes bargaining and other obvi- ily yield a detailed prediction about what policy
ously political processes as central to policymaking outcomes to expect from the model.7 Such an identi-
and deemphasizes issues involving information proc- fication does have certain implications-for example,
essing. Moreover, his discussion of power involves that the relevant government will not encounter
actors who are apparently able to make complex surprises (by virtue of the complete information as-
strategic calculations about how to maximize their sumption) and will exhibit transitive preferences (the
influence (pp. 168-69). These actors thus seem to unitary actor and rationality assumptions). But such
have the cognitive capacity usually assumed by ra- predictions fall well short of the kind of policy-
tional choice models of policymaking, so that readers relevant claims naturally sought by students of for-
might conclude that classical rationality is presumed eign policy. The burden of generating falsifiable pre-
for Model III, leading to its placement in cell 5.6 dictions lies on the models, not on the typology.
Moreover, since Allison explicitly states that bureau-
cratic politics games "are not played under conditions
of perfect information" (p. 178), Model III seems to ANALYSIS OF MODEL I
belong in cell 5b.
However, other of Allison's statements about Allison's analysis of rational decision making is the
Model III suggest a different classification. For exam- most coherent and most easily understood of his
304
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
models. A rational actor model in international rela- why the Soviet Union placed the missiles in Cuba in
tions should have two essential components: a deci- the first place, why the United States responded with
sion-theoretic one and a game-theoretic one. The a naval quarantine, and why the Soviet Union with-
former covers attributes of the decision maker(s), the drew its missiles.
latter, strategic interactions. We have two major Allison's rational actor model is as simple a version
points to make about Allison's rational actor model. of rational choice as one could imagine. His purpose
First, the decision-theoretic component is too simple: in formulating this model was to show how some-
one can easily construct a far more complex rational thing much like it underlies many different bodies of
actor that better reflects the complexities of real thought in international relations. But in this partic-
decision makers and the choice situations they con- ular enterprise lies what we see as the model's major
front. Second, the model neglects strategic interac- fault. It is too simple, for it omits several significant
tions. properties of rational actors and important problems
Before examining the details of Model I, we must they face. The model can be enriched in several
discuss one difficulty with analyzing Allison's treat- different ways.
ment of the model: his tendency to distance himself
from his own presentation. As he notes "Since the Goals. An ambiguous aspect of Model I is whether
purpose of the Model I account is to present a strong, rational actors are presumed to have just one goal.
typical explanation, it seems fair to let the account Before explicating the model, Allison notes that the
reflect only conventional evidence plus additional Model I theorist "presents an argument for one objec-
facts that the Model itself would naturally uncover" tive that permits interpretation of the details of Soviet
(1971, 248). This raises the problem whether one behavior as a value-maximizing choice" (1971, 11;
should critique (1) the classical version of single-actor emphasis added). Later he discusses how this analyst
rational choice that scholars have traditionally used, "produced an argument for one goal (rectifying the
(2) Allison's interpretation of this standard model, or nuclear balance) that made the Soviet emplacement
(3) what Allison himself thinks this model should plausible" and "proceeded as if his assignment had
look like if (given its premises) one developed and been: make a powerful argument for one objectivethat
applied it "properly." We are inclined to think that permits the reader to see how, given the strategic
approach 1 or 2 best characterizes Allison's treatment; problem, if he had been playing the Soviet hand, he
Allison never tells the reader what he thinks Model I would have chosen that action" (pp. 246, 253; em-
would be like if properly developed. This matters phasis added). And in applying Model I to the
because we think a more fully developed version of Soviets' decision to install the missiles, Allison's
Model I would have substantially more explanatory discussion appears largely oriented toward inferring
power than Allison's Model I. For simplicity, we will which one of the five possible goals best explained the
treat the model as Allison's own and criticize it as Soviets' decision. For example, in ending his discus-
such, understanding that it has a different status in sion of Cuban defense as a possible motive, he
Allison's thought than his Models II and III.8 concludes, "Cuban defense might have been a sub-
sidiary effect of the Soviet gamble, but not its over-
riding objective" (p. 50).
Decision-theoretic Aspects
Yet in discussing the international relations theo-
Model I employs two key decision-theoretic assump- rists Hans Morgenthau and Raymond Aron, Allison
tions about nation-states, postulates found in most notes in particular, "Criticizing the attempts of theo-
canonical formulations in international relations. rists such as Morgenthau to explain national action by
First, the nation can be treated as a single actor reference to a single goal, Aron argues that govern-
endowed with a set of goals and the capacity for ments pursue a spectrum of goals, tempered by 'the
pursuing these goals. Second, this single actor pur- risk of war [that] obliges [them] to calculate forces or
sues these goals rationally. Though it is easy to means' " (1971, 27). And in his presentation of Model
criticize the assumption that the nation is a single I, Allison seemingly allows the possibility that a
actor, many abstract analyses require treating an rational actor might have multiple goals: "National
aggregation of human beings as a single unit. For security and national interests are the principal cate-
some purposes such aggregation is useful and appro- gories in which strategic goals are conceived. Nations
priate. We will, instead, focus our attention on Al- seek security and a range of other objectives" (p. 33).
lison's treatment of rationality. Although Allison does not explicitly discuss this
In common parlance and in Allison's usage, ra- matter of multiple goals, it raises important issues in
tional decision making consists of four key steps: understanding and applying Model I. Nothing what-
define one's goals, list all options for achieving these soever in the theory of rational action requires an
goals, evaluate each option in terms of the extent to actor to have just one goal. This is a significant
which it achieves the goals, and choose the option feature of the theory, for empirically, an actor with
that best achieves the goals. As Allison points out, only one goal is odd indeed. In fact, we normally use
students of international politics and foreign policy the term monomaniacalto refer to the irrationality of
have long used this model (if only implicitly). He people motivated by only one objective. Typically,
demonstrates how the model might be used to make we expect rational actors to have multiple goals.
sense of three major decisions in the missile crisis: Indeed, nothing in the axioms of rational action states
305
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
that it is nonsensical to harbor multiple goals. The joyed long after the moment of choice? Note that
goals may conflict. But in a rational choice model a actors with identical goals but different time prefer-
utility function is postulated to resolve all intraper- ences would value the same option differently, due to
sonal conflict.9 Thus, rationality lies in the considered their different degrees of impatience.
and consistent pursuit of whatever goals one hap-
pens to have. Uncertainty. Allison's rational actor model also ig-
We can support our arguments by reference to the nores the problem of uncertainty. While Allison
Soviet Union's possible motives for placing nuclear obliquely considers uncertainty in his discussion of
missiles in Cuba. Presumably there was, as Allison Model II, the problem is completely neglected in
concluded, a strong concern for the strategic balance. Model I. This is a striking omission, since the tradi-
The recent U.S.-Soviet conferences tend to bear this tional literature in international relations, which Al-
out (Blight and Welch 1989). However, it is not lison cites as having implicitly used rational actor
unreasonable to think that the Soviets were also models for centuries, emphasizes how a state's un-
concerned about defending Cuba. The Bay of Pigs certainty about other states' goals and capabilities
invasion had given the Soviets (and the Cubans) shapes its own choices.
cause for concern. Moreover, the Kremlin would There do exist formal models of rational choice that
have been pleased if placing missiles in Cuba would show how an actor should make optimal choices
have diminished the prestige of the United States in under uncertainty. Some of these models were for-
the eyes of West Europeans and nonaligned coun- mulated well before Allison conducted his study.
tries. (Indeed, he cites relevant texts on statistical decision
The Soviets' concern for the strategic balance may theory and game theory [1971, 285, n. 87].) Thus,
well have been of overriding importance; other con- several different types of uncertainty were analyzed
cerns may have been of negligible significance in their and well understood by rational choice theorists
minds. Nonetheless, it is reasonable to think that the before Allison's article or book was written. One kind
Soviets would have been less inclined to place the of uncertainty that was well understood was a non-
missiles in Cuba if this action would have clearly strategic, prospective type, involving exogenous and
impaired these other goals. Thus, these other goals unpredictable acts of nature. Though probably less
cannot be considered irrelevant to an understanding important than the strategic aspects considered be-
of the Soviets' motives and choices. low, they were not irrelevant in the Cuban missile
For each of these other goals, Allison may be crisis. Bad weather, for example, would have made a
correct in arguing that there was some other action surgical air strike more difficult. Unexpected equip-
better than placing nuclear missiles in Cuba. How- ment problems, such as those that plagued U.S.
ever, the bare fact that the chosen action is not optimal helicopters in the aborted Iranian hostage operation,
for the other goals (such as defending Cuba) does not also fall under the domain of statistical decision
imply that these other goals were completely irrele- theory. No rational military planner would assume
vant to the decision. that all his equipment will work with certainty. And
these problems can be sensibly addressed within the
Time. Allison's rational actor model is characterized confines of the single-actor assumption. Indeed, sta-
by a single-time-period orientation. Thus, the rational tistical decision theory classically developed under
actor considers a problem occurring at a point in time; this assumption, as did the related probabilistic meth-
and once an alternative is chosen and implemented, ods in operations research.1
the matter is over. Allison himself mentions this Observe that just as introducing multiple time
feature as a limitation of the model, labeling it "static periods raises the issue of decision makers' time
selection." preference, introducing uncertainty raises the issue
One might argue that a single-time-period model is of decision makers' attitudes toward risk. Whether
appropriate for crises. However, even in a crisis, the they are risk-averse, risk-neutral, or risk-seeking
effects of options may be spread over multiple time must be specified in order to complete a rational actor
periods. Consider, for example, President Kennedy's analysis. However, it must be recognized that these
concern for the United States' international reputa- issues, though analyzed in more recent international
tion for firmness and resolve: such reputational con- relations theories (e.g., Bueno de Mesquita 1981,
sequences may endure for years. Thus, in estimating 1985), were not salient in rational choice theorizing in
the effects of any option, a rational agent must political science prior to Essence of Decision.
aggregate, into a single measure, costs and benefits In addition to uncertainty about states of nature, it
that may be spread over a long time. Aggregating is empirically clear that U.S. decision makers were
streams of costs and benefits is a rather different kind very uncertain about the Soviets' motives for placing
of choice problem than aggregating costs and benefits missiles in Cuba. Allison himself pointed up this
that all occur at the same time. problem, describing the markedly different U.S. in-
Incorporating multiple time periods in a model of terpretations of Soviet behavior (1971, 40-56).1" These
rational choice leads directly to the issue of time interpretations placed the Cuban adventure in the
preferences. Is the rational actor patient, valuing context of larger Soviet strategies, and the magni-
distant benefits nearly as much as current ones, or tudes of these strategies' payoffs were hotly debated
impatient, sharply discounting benefits that are en- by U.S. policymakers.
306
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
307
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
3). There are two Nash equilibria in chicken: (aggress, answer to the question, "Why did this outcome
concede) and (concede, aggress). In the first outcome, occur?" has the form, "Because actor so-and-so
player 1 gets its most preferred outcome, and player wanted it."12 Such explanations do not come close to
2 is stuck with its third best. In the second equilib- tapping the potential of rational choice analyses.
rium, matters are reversed. The presence of these two They completely overlook the complexities arising
equilibria, with their markedly different payoffs, both from strategic interaction and from the institu-
points up another important difference between de- tional context of the game. Indeed, by slighting such
cision- and game-theoretic analyses. If there is only a complexities, these uses of rational choice theories
single decision maker, then there is either a unique degradetheir value; much of the intellectual benefit of
alternative that is value-maximizing or several alter- rational choice analyses is that by black-boxing inner
natives. Neither case presents a problem; for even if mental processes, we can concentrate our attention
there are several such alternatives, they must by on relations among decision makers.
definition yield the same payoff and are therefore
equivalent. But in a multiperson situation, the equi- Endogenous Uncertainty in Games of Complete Informa-
libria need not be equivalent; they can differ sharply tion. The preceding remarks about the game of
in both relative and absolute senses. Accordingly, the chicken informally suggest that game theory might
players in chicken are not indifferent between the two view some interactions as being fundamentally inde-
equilibria. terminate. We will now sharpen this point. It has
Hence, once again, there is no simple relation been well known for decades that uncertainty can
between the game-theoretic notion of equilibrium arise endogenouslyin strategic situations. To illustrate
outcomes and the decision-theoretic idea of value the point squarely in the matter at hand, consider the
maximization. It is true that if player 1 aggresses, representation of one aspect of the Cuban missile
player 2's value-maximizing choice is to concede. crisis (Figure 4).
But player 2 would itself be better off aggressing-so In the situation shown in Figure 4, if the United
long as player 1 concedes. Thus, contrary to Allison's States carries out intense inspections, the USSR
claim that "in modern statistical decision theory and would prefer not to put the missiles in Cuba (1 > -1).
game theory, the rational decision problem is re- However, if the USSR holds back, then the United
duced to a simple matter of selecting among a set States would rather not incur the cost of intense
of given alternatives" (1971, 29; emphasis added), inspections (2 > -1). But if the United States con-
chicken exhibits an intricate strategic structure. In- ducts only cursory inspections, then the USSR does
deed, due to its complexity-the potential for mutual want to install the missiles (1 > 0). This, however,
disaster if each side tries to get its maximal payoff or takes us full circle, since if the USSR is going to install
mutual accomodation if both sides try to avert disas- the missiles, the United States wants to inspect
ter-predicting the outcome of chicken is by no closely (1 > -1). Hence, in this kind of situation,
means trivial. Any of the four outcomes could plau- there is no equilibrium in pure strategies; but there is
sibly occur. Whether this is a defect of game theory in an equilibrium if each player uses mixed strategies,
not making a falsifiable claim for this game or an that is, plays probabilistically.13 Thus, uncertainty
indeterminacy inherent in the situation, we cannot will arise endogenously from the strategic logic of the
say. What we can say is that just assuming rationality situation. The only equilibrium in this circumstance is
does not, in itself, reduce the decision problem to a where each side keeps the other guessing about its
"simple matter." course of action, just as in the child's game of
In our experience teaching models of politics, most stone-scissors-paper.14
students begin a course believing that preferences It is evident, therefore, that strategic uncertainty
and outcomes are tightly linked in rational choice was relevant in the crisis and that some theoretical
models and that the paradigmatic rational choice tools were available to aid analysis. (For a recent
308
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
analysis of the Cuban missile crisis using a model of Major lines of organizational action are straight-i.e.,
strategic uncertainty, see Wagner 1989.) behavior at one time, t, is marginally different from
behavior at t - 1. Simple-minded predictions work best:
Many readers of Essence of Decision, we suspect, behavior at t + 1 will be marginally different from
behavior at the present time.
conclude that models of states as unitary rational
actors provide an inadequate account of this crisis. The preeminent feature of organizational activity is its
Such a conclusion is not well founded. Allison's version programmed character: the extent to which behavior in
of a rational action explanation is so oversimplified that any particular case is an enactment of preestablished
routines.
it should not have been expectedto perform well.
The fact that the fixed programs (equipment, men, and
routines that exist at the particular time) exhaust the
ANALYSIS OF MODEL II range of buttons that leaders can push is not always
perceived by these leaders. (1971, pp. 91, 81, and 79,
respectively)
Allison's discussion of Model II is one of his strongest
sections. He extracts several insights from organiza- This image resonates strongly with our intuitive
tion theory and applies them to the missile crisis in conceptions. We are all familiar with the stereotype of
interesting ways. For example, though the Soviets an inertial bureaucracy mindlessly following its pro-
may have had sound engineering reasons for build- cedures regardless of consequences. As with many
ing the missile sites in their characteristic pattern, popular images, this one is not completely wrong-
Model II suggests an interesting alternative explana- far from it. The picture of simple, rule-governed
tion-it was simply a bureaucratic routine being action provides important insights into organiza-
enacted. In the study of foreign policy making and tional behavior. However, precisely because the im-
implementation, Allison's chapters on bureaucratic age is so intuitive, we must be careful. Familiarity in
routines were pioneering efforts. model building breeds laziness, if not contempt. The
Nonetheless, we have some reservations about his notion that simple rules generate simple behavior is
analysis. The core of Model II is the idea, developed so commonsensical that it is easy not to question it.
by Herbert Simon and James March, that individual Yet the link between rules and behavior is more
decision makers are boundedly rational (see Allison subtle than is captured by either Model II or its
1971, 71). We have two main concerns with Allison's associated popular image. Complex behavior can
interpretation of this theoretical core. First, Model II emerge from simple rules. We now describe six ways
assumes that imperfectly rational agents would usu- this can happen.
ally use simple decision rules (p. 83). In itself, this
assumption is consistent with much of the writing of Varying, and Possibly Stochastic, Inputs. Many choice
the Carnegie School.15 But Model II goes beyond this rules work on inputs-information, people, objects-
premise by suggesting that these simple rules, or that vary significantly. This variability, in tandem
standard operating procedures, sharply limit and con- with even a simple rule, can produce complex behav-
strain behavior, that is, that simple rules generate ior. A good example of this is the behavior of one of
simple, predictable behavior (pp. 78-79, 83, 87-91). the simplest strategies in the prisoner's dilemma, Tit
We believe that this conclusion greatly underes- for Tat (TFT). This strategy-cooperate on the first
timates how complex behavior-of even a single round and thereafter play what one's partner did on
decision maker-can arise out of the use of simple the previous round-is only a tad more complicated
rules. Second, suppose we grant that individual deci- than the simplest possible rules (always cooperate or
sion makers are sharply constrained in their cognitive always defect (all-D)). However, whereas those un-
abilities. What does this premise imply about how conditional strategies generate behavior as simple as
constrained organizations are? Consistent with the themselves (regardless of their strategic environ-
viewpoint of Simon's Administrative Behavior, we ar- ment), in variegated environments TFT will behave in
gue that organizations can significantly ease the con- correspondingly complex ways. Against all-D, TFT
straints on information processing that confront indi- cooperates once and never again; against a nice
viduals (1947, 79-80). Taking these two points partner TFT cooperates forever; against a sneaky
together, our position is that the use of simple deci- nasty strategy, TFT will cooperate as long as its
sion rules by individual decision makers does not partner does, retaliating against a defection but then
imply that the behavior of an organization will be returning to cooperation if its sneaky partner apolo-
simple, unsophisticated, or predictable. Thus, we gizes. Moreover, if the environment is noisy (unob-
agree with March and Simon, who argue that "com- served random disturbances affect the players' pay-
plex processes can be aggregated from simple ele- offs), TFT produces variable behavior even when
ments" (1958, 178). playing an equally simple strategy-itself (Bendor
1987).
Because simple rules can be activated by random
Decision Rules Versus Behavior
shocks in an organization's environment, its behavior
Allison's basic image of organizations in Model II is can be, in a certain sense, surprising, as well as
that of simple, predictable behavior generated by complex. For example, disaster relief could be highly
simple rules: routinized and preprogrammed, but the organiza-
309
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
tion's behavior would not be unconditionally predict- The parameter a represents the rate of learning and is
able. For these organizations, it need not be true that between 0 and 1. If x failed, the administrator's
the best predictor of the organization's behavior at propensity to try it weakens: pi+1(x) = bpi(x), with b
time t is its behavior at t - 1.16 also between 0 and 1. The equations for alternative y
have the same form. Note that these equations are
RecursiveRules. Many public agencies seem to follow linear.
a simple history-dependent rule of the form, decision However, whereas an expected utility maximizer
t + 1 = f(decision t). Such rules are called recursive.To will behave simply in the long run-choosing x
figure out what the decision will be in period t + k, in always-a decision maker using this linear adaptive
terms of the decision made in period t, one just rule will never settle down on either alternative.
reapplies the rule f(Q)over and over again. Thus, Instead, the adaptive decision maker will use alter-
decision t + 2 = f(decision t + 1) = f[f(decision t)], native x in the long run (1 - q)/[(1 - p) + (1 - q)]
and so on. Do such recursive choice rules always proportion of the time and y the remainder. Thus, the
generate simple behavior? They do not. Interesting simple linear adaptive rule produces more complex
developments over the last 10 years in a field of behavior than does the optimal strategy.
mathematics known as chaos theory have shown that
if a recursive rule is nonlinear, it can create a pat- CombinatorialEffects. Several times in Essence of Deci-
tern that is so complex as to appear random, even sion, Allison uses the metaphor of chess when dis-
though the rule itself is completely deterministic. For cussing how leaders are constrained in their choice of
example, consider the simple recursive rule xi+, = options. This is odd, because chess is the paradig-
k- xi(1 - xi), where k is a constant between 0 and 4. matic example of a choice situation that involves only
Let the initial value xl be between 0 and 1. How will a handful of basic rules yet exhibits truly Byzantine
x behave over time? It turns out that if k is close to 4, strategic complexity. There are only six distinct pieces
in chess; each moves in only a few ways. (We may
the behavior of x is "chaotic," wandering all over the
consider each piece, with its legal moves, as equiva-
[0, 1] interval and never settling down. For example,
lent to a standard operating procedure.) Therefore,
let k = 3.7 and xl = .4. Then the next 10 values of xi
each player is "limited" to shuffling and reshuffling
(rounded to four places) are .888, .368, .8605, .4441,
this combination of standard operating procedures
.9134, .2926, .7658, .6636, .8259, and .5319.
over the 64 possible locations.
Merely inspecting these values of xi, one would But the quantitative impression thereby conveyed,
probably attribute them to a stochastic process. Yet namely, that the limited number of procedures cre-
the underlying equation is not only deterministic but ates a roughly equal (even in respect to order-of-
also very simple, being only one degree higher than magnitude) number of moves is grotesquely inaccu-
linear. Moreover, it is not a historically rich rule, for it rate. As is well known and as Allison himself noted,
depends only on the decision in the previous period. the number of possible sequences of play (i.e., behav-
Real precedent-governed agencies have a much more ior) in chess is staggeringly large (1971, 286, n. 93). A
complex memory structure. common estimate is 1012, a number so large as to be
equivalent, for all practical purposes, to infinity. Hu-
Simple Linear (ThoughProbabilistic)Rules. Since Model man beings can never exhaust the richness of chess.
II repeatedly refers to the linear nature of organiza- This is pure combinatorial explosion: the rules are
tional behavior, one may object to the previous deterministic. Nor does the complexity depend upon
example because it is nonlinear. We respond in two stochastic inputs. It is simply a matter of stringing
ways. First, linearity is an empirical hypothesis; we together the handful of rules to generate new positions,
have no a priori guarantee that organizations typi- whence one reapplies the basic rules, and so on. The
cally behave linearly. Second, we now demonstrate, example of chess suggests that our intuitions grossly
via another example, that behavior can be surpris- underestimate combinatorial complexity.
ingly complex even if linearity is retained. Moreover, when we compare chess to the strategic
Consider the following model of experiential learn- maneuverings of two real military forces (to return to
ing. Suppose an administrator has two options, x and Essence of Decision's empirical domain), the odds are
y, and knows that neither one is sure to work or good that chess is simpler. No matter how one counts
doomed to fail. Objectively, the probability that x distinct "pieces" in, say, a navy, there are surely
succeeds is p, the chance that y works is q, where 0 < more than seven! No matter how crudely one wants
q < p < 1. (They need not sum to one.) These to count the basic standard operating procedures of,
probabilities are constant and independent over time. say, a destroyer, it has more than the handful avail-
In the first period, the decision maker has a propen- able to even the most complex chess piece. And the
sity to try x denoted pl(x) and a propensity to try y "board" of any moderately large battle is more var-
denoted P1(Y)= 1 - pl(x). Over time, the administra- iegated than the (nontopographical) eight-by-eight
tor adapts via a simple learning rule. If in period i x chessboard. Therefore, chess, the paradigmatic
was tried successfully, the administrator's propensity choice environment of behavioral decision theory,
to try x again is strengthened: provides a lower bound for the complexity of behavior
one would expect to see in a clash between two
pi+ l(x) = pj(x) + a[1 - pi(x)]. governments. Thus, in this instance Essence of Deci-
310
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
sion seriously misapplied the lessons of the Carnegie sion may not enable one to predict behavior, because
School. one does not know which of the many legal prece-
dents the judge will select. Again, behavior is unpre-
Hierarchies of Procedures. In a sense the linking of dictable despite the stable set of rules. The test is not,
procedures into a larger unit (creating a "program," Did the decision follow from an organizational rule?
in Allison's terminology) is just a type of combinato- The test is, Could one have predictedwhich organiza-
rial complexity. In this sense the remarks just made tional rule would be evoked?
apply. However, hierarchies of rules create another It should be understood that we are not here
way for complex behavior to arise from simple rules. advancing empirical hypotheses. We are not assert-
Suppose the Department of Defense used the simple ing that any particular set of officials or agencies use
budgeting rule, "Every year we allocate 10% of our any of the rules described in our first five points. Our
budget to research and development." Naturally, this argument is theoretical. The relation, implicitly as-
rule would be accompanied by a host of procedures sumed by Model II, between simple rules and simple
concerning for example, how the designs of weapon behavior is weaker than one might think. The exam-
systems should be specified. For our purpose, we ples illustrate that rules can generate relatively com-
focus on the relation between this one rule and plex behavior even when the rule is deterministic
consequent behavior."7 (points 1 and 2) or linear (point 3) or the game is
Again, the rule's simplicity is deceptive. The allo- isolated from stochastic shocks (point 4). Accord-
cation of a portion of the Defense Department's ingly, the suspicion grows that simple rules will
resources to research and development has created a produce simple organizational behavior only if (1) the
historically unprecedented number of new weapons, rules are deterministic and (2) the rules are linear and
introduced at an equally unprecedented rate. Each (3) the game is isolated from stochastic shocks and (4)
new weapon is accompanied by its own standard individual choice rules are not aggregated in a com-
operating procedures, as well as the procedures "fro- binatorially rich way and (5) individual choice rules
zen into" the hardware itself. Thus simple high-order
are not grouped hierarchically so that higher-level
rules can generate new lower-order ones. (For a
rules generate new lower-order ones (and new ensu-
discussion of this point, see March and Simon 1958,
ing behavior). These conjunctions indicate a shrink-
150, 170.)
ing of the parametric space in which the postulated
LargeSets of Rules. Our first three points showed how simple behavior will occur. In terms of our prior
a single rule could produce complex behavior. The beliefs about bureaucratic behavior, one may feel
next two points showed how a few simple elemental confident in asserting, say, that individual choice
rules could aggregate into complex behavior. Here rules are linear, or that they are deterministic, or the
we wish to underscore an obvious empirical point: like. But would many scholars in this field confi-
even if we set aside combinatorial complexity, some dently assert all five conditions? We doubt it.
agencies have a great many rules. Consider the
Internal Revenue Service code. The gradual, barna-
clelike accumulation of exemption piled upon exemp- Bounded Rationality: Individual and
tion, modification upon modification, has created a Organizational
code of rules that is awesomely complex.
Very large sets of rules create problems for schol- A second problem of Model II concerns the relation
ars, as well as citizens. An observer may find it between individual and organizational constraints on
difficult to predict organizational behavior because of rationality. Organization theorists have often moved
not knowing which rule will be evoked out of the set "rather cavalierly from theories of individual cogni-
of all possible rules. There are two related reasons for tion and choice to theories of organizational cognition
this. First, large sets of rules increase the possibility and choice" (March and Shapira 1982, 11). This
that two or more rules may conflict: more than one tradition has sometimes been carried to the point of
rule will apply to any one situation, and it will be anthropomorphizing organizations, positing that or-
unclear which has priority. Second, though many ganizational constraints on information processing
organizational rules have an if/then structure like simply mirror individual constraints.
computer programs ("If condition x obtains, apply This quick passage from individual to organiza-
procedure y"), the conditional part of the rule-the tional limits on rationality is evident in Essence of
description of condition x-is often much less precise Decision: "Simon and the Carnegie School focus on
than in computer programs. This also makes it more the bounded character of human capabilities. Firms
difficult to predict which rule will be evoked. are physically unable to possess full information,
The classic illustration of both problems-conflict- generate all alternatives" (Allison 1971, 174). Again,
ing rules and imprecise conditionals-is the legal "The physical and psychological limits of man's
system. If one regards the judiciary as an organiza- capacity as alternative generator, information proces-
tional system, the model of rule-governed behavior sor, and problem solver constrain the decision-
seems apt. A judge's behavior is, after all, supposed making processes of individuals and organizations"
to be informed by legal rules. But even knowing the (p. 71).1
whole set of rules that could guide the judge's deci- Undoubtedly the bounded rationality of single de-
311
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
cision makers constrains the information-processing normative principles of scientific inference. Further-
capacities of agencies in some fashion. However, we more. . ., no single person could have solved all the
maintain that the relation between the two is more problems involved in such necessarily collective efforts
complex than is portrayed by either Model II, in as space exploration. Getting to the moon was a joint
particular, or much of organization theory, more project, if not of idiot savants, at least of savants whose
generally. For normative reasons-and because the individual areas of expertise were extremely limited-
literature has focused heavily on the tendency of .... Finally, those savants included people who be-
institutions to be dumber than their members (via, lieved that redheads are hot-tempered, and who at the
advice of an acquaintance's brother-in-law bought their
e.g., conformity pressures)-we are especially inter-
last car at the cocktail-party. (1980, 250)
ested in how organizations can be smarter than the
individuals who compose them. We therefore con-
Reliable Systems, Unreliable Components. In his book on
sider three reasons why organizations can process
information more effectively than individuals can. the Polaris missile system, Sapolsky (1972) observed
that several teams worked independently to develop
From Serial to Parallel InformationProcessing. In a neg- a key component. The Navy's Special Projects Office
lected passage of Organizations, March and Simon deliberately established this redundant structure to
note an important difference between individual and increase the odds that the essential component would
collective cognition: "The individual possesses only a be ready by a specified date. By doing so, the Special
single focus of attention, hence can only deal with Projects Office loosened the connection between the
one aspect of the problem at a time. In organizations, reliability of an organizational subunit (one of the
however, there is no limit to the number of attention problem solving teams) and the larger organiza-
centers that can be allocated to the parts of a prob- tion-in contrast to the message of Model II.
lem" (1958, 193). They go on to note the advantages More generally, a properly designed agency can
of parallel processing. Clearly, if one large problem display large disparities between the reliability of
can be decomposed into many subproblems, an different organizational levels (Bendor 1985; Grof-
agency can greatly speed up its information process- man, Owen, and Feld 1983; Landau 1969). Consider a
ing bX assigning a different official to each subprob- representation of the Polaris project. There are n
lem.1 Of course, there is no guarantee that any given teams trying to develop the same component. The
division of labor will be an effective decomposition. probability that any one of the teams will succeed in
The trick, as many students of organizational design the specified time is p, where 0 < p < 1. If the
have noted, is to figure out a scheme of specialization performance of the teams is independent, the prob-
that "carves nature at the joints" (see Gulick 1937; ability that the organization will succeed equals p(at
March and Simon 1958; Simon 1947; Simon, Smith- least one of the teams succeeds), which equals 1 -
burg, and Thompson 1950). When this can be done, [(1 - p)n]. This probability increases steadily toward
the serial constraint can be substantially eased.
one as n increases, so long as p exceeds zero. Thus, in
This organizational advantage is obvious. Just
this simple setting, the performance of subunits and
mentioning the phrase, division of labor, helps us
units need not be closely connected.
recognize an organization's advantage over an indi-
vidual. In the simplest model of problem decompo-
From Tunnel Vision to Innovation. Model II's image of
sition (an additive one) information processing is
essentially the same as performing a physical task, organizational change is of sluggish entities domi-
like Frederick Taylor's ([1911] 1947) pig iron loading. nated by a single way of thinking. Yet modern
Obviously, the more people, the more pig iron is bureaucracies are often populated by different kinds
loaded or the faster information is processed. Other- of professionals with distinctive mindsets. What is
wise, why bother to hire the additional people? Yet in commonplace to one professional culture may be
the organizational decision theory literature, this ob- dramatically new-even bizarre-when introduced
vious point is sometimes overlooked. to an agency dominated by another profession. Ob-
A second, more subtle advantage of specialization serve, for example, the clash between civil engineers
is that it enables decision makers to become experts in and environmental analysts in the Corps of Engineers
their domains. Becoming an expert means relying (S. Taylor 1984). The questions raised by new envi-
less on ordinary folk heuristics, with their attendant ronmental analysts about water projects, though
biases and more on scientifically based inferences, alien to the engineers and the Corps, were standard
with their lower rates of error (Hogarth 1987; Nisbett in the community of environmental analysts. Or
and Ross 1980). As Nisbett and Ross replied to a consider the idea of vouchers for schools. To a
colleague who, having read a draft of Human Infer- professional with a background in educational ad-
ence, asked, "If we're so dumb, how come we made it ministration, vouchers are a dramatic departure from
to the moon?"- the status quo. But to an economist specializing in
Humans did not "make it to the moon" by trusting the education policy, they are an obvious extension of the
availabilityand representativenessheuristics or by rely- basic principle of market competition. In both exam-
ing on the vagaries of informal data collection and ples, innovation was based on straightforward appli-
interpretation. On the contrary, these triumphs were cations of professional expertise; heroic mental efforts
achieved by the use of formalresearchmethodology and were not required.
312
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
Scholars typically understand the main lesson of Hence, chapter 4 of Essenceof Decision inverts Simon's
Model II to be that routines and standardized scenar- view. Instead of organizations boosting individual
ios constrain and rigidify organizational behavior rationality, the bureaucracy's rioid conduct nearly
(e.g., Art 1973, 476-79; Gallucci 1975, 145; Jefferies nullified Kennedy's best efforts.'
1977, 231-32; Krasner 1972, 169-75; Levy 1986; Peter- Why do Simon's and Allison's books assess stan-
son 1976, 113; Scott 1981, 6; Thompson 1980, 27; dard operating procedures in particular and organi-
Williamson 1979, 140). (For a partial exception to this zations in general so differently? We believe that
pattern, see Posen 1984, 46-47.) True, Allison occa- different baselines of evaluation are at work here. In
sionally mentions the possibility of organizational Administrative Behavior, Simon began with the
innovation and at times portrays standard operating premise that individuals are boundedly rational. With
procedures as efficient ways of handling standard this as a starting point, it is not surprising that he
problems.20 But most of chapter 3's theoretical expo- emphasized the enabling aspects of standard operat-
sition emphasizes the negative, constraining effects ing procedures and specialization. Given the real
of organizational routines. And this pattern is strik- limits on individual cognition, technically demanding
ingly reinforced in chapter 4's empirical application of projects such as going to the moon can be handled
Model II to the missile crisis. Virtually without excep- only by collective endeavor. In contrast, Essence of
tion, the episodes covered there describe how the Decision begins with Model I, which presumes per-
permanent bureaucracies of both countries fouled fectly rational individuals. Whether Allison intended
things up. Toward the end of chapter 4, Allison unboundedly rational decision makers to be the stan-
pointedly asks, "Were the organizations on top of dard for evaluating Model II actors is not completely
which the President was trying to sit going to drag clear; but the book's structure and the frequent use of
the country over the nuclear cliff in spite of all his Model I as an analytical benchmark suggest that this
efforts?" (1971, 141). The heroes of this chapter are was the implicit standard.23 With that baseline, it is
clearly the unbureaucratic president and his advisors. not surprising that Model II organizations, staffed by
Therefore, the common interpretation that Model II's boundedly rational individuals, appear more the
main lesson concerns the rigidifying, maladaptive problem than the solution. But we would ask, How
effects of bureaucracy is well founded. well could the United States have gathered informa-
This is an irony of intellectual history. As Allison tion about the Soviet missiles in Cuba, conducted the
noted, the central ideas of Model II are rooted in the blockade, and carried President Kennedy's message
Carnegie School. Yet a founding volume of that to the world had there been no routines to rely on?
tradition, Simon's Administrative Behavior, views rou- How effective would the U.S. response been if it had
tines far more positively: had to invent all procedures on the spot? Had Allison
disentangled the different effects of bounded rational-
Habitperformsan extremelyimportanttask in purposive
behavior, for it permits similarstimuli or situations to be ity and organizational action by inserting a model of
met with similar responses or reactions, without the a unitary-actor-with-imperfect-rationality in between
need for a conscious rethinking of the decision to bring Models I and II, his evaluation of the net effects of
about the proper action. Habit permits attention to be organization-holding constant the degree of indi-
devoted to the novel aspects of a situation requiring vidual rationality-might have changed consider-
decision. A large part of the trainingthat goes to make a ably.
championshipfootballteam, crew, armybattalion,or fire
company is devoted to developing habitual responses
that will permit immediate reactionsto rapidly changing ANALYSIS OF MODEL III
situations.
Habit, like memory, has an artificial organization
counterpart, which has been termed by Stene "organi- Allison's central argument in Model III is easily
zation routine." (1947, 88)21 stated: "The name of the game is politics: bargaining
along regularized circuits among players positioned
Indeed, Simon goes much further than this in hierarchically within the government. Government
praising the general role of organizations in amplify- behavior can thus be understood according to a third
ing individual rationality: conceptual model, not as organizational outputs but
It is impossible for the behavior of a single, isolated as results of these bargaining games" (1971, 144). It
individual to reach any high degree of rationality. The proved difficult, however, to build a clear and coher-
number of alternativeshe must explore is so great, the ent model around this claim. Some of the fault lies
informationhe would need to evaluate them so vast that with the literature on which Allison had to rely. For
even an approximationto objectiverationalityis hard to the rational actor model, he could draw on well-
conceive.... One function that organization performs specified, axiomatically constructed theories. The org-
is to place the organizationmembers in a psychological anizational process model is based on works that
environment that will adapt their decisions to the orga- advance propositions that are thematically (though
nization objectives, and will provide them with the
information needed to make these decision correct- not deductively) connected. In contrast, the literature
ly.... In the course of this discussion it will begin to on bureaucratic politics is more discursive and far less
appear that organization permits the individual to ap- explicitly theoretical. Allison himself remarked on
proach reasonably near to objective rationality. (pp. this difficulty: "Unfortunately, the interests of these
79-80). analysts have not led them to invest much labor in
313
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
squeezing their insights into propositions. Conse- they identify with his beliefs to a greater extent than
quently, the paradigm. . . can be no more than a would a randomly chosen group of candidates"
tentative formalization" (p. 147). Even his tentative (1972, 166).
formalization, however, attracted far more critical Moreover, Art (1973) argues that shared "mind-
attention from reviewers of Essence of Decision than sets" often influence foreign policymaking. He notes,
either of his other models. for example, that in the decisions to intervene in
Criticisms of Model III can be grouped into four Korea and the Dominican Republic, "there was una-
categories. First, many critics argued that Model III nimity on the need to use military force, once the
misconstrues the nature of executive branch policy- principals involved realized that nothing short of that
making. In particular, we will argue, the central claim would achieve American objectives" and that for
(that policymaking necessarily involves bargaining Cuba and Vietnam, "the need to use force was
among executive branch decision makers) must be questioned by one or two principal advisors; but they
questioned. Second, we will point out that Model III were clearly out of the mainstream of the President's
said surprisingly little about the fact that executive thinking and his advisors' outlooks" (p. 481). The
branch policymaking takes place within a hierarchy. implication of Art's point is clear: "If shared images
Hierarchy's impact on policymaking is something dominate senior players' outlooks and if they are
that both Allison and many of his critics overlooked. truly shared, then what is the merit in asserting that
Third, we will argue (as did several critics) that Model governmental actions are the resultants of pulling,
III is the least precisely formulated of Allison's ap- hauling, and bargaining?" (p. 476). To the extent that
proaches. Not only are its fundamental assumptions Model III assumes that actors have conflicting goals,
much less clear than those of Models I and II, but these examples suggest some clear limits to the
many of the propositions seem to be ad hoc observa- model's empirical domain.
tions not derived in any rigorous sense from the
assumptions. Finally, while many critics questioned Does Conflict Imply That Policymakers Have Different
Model III's premise of policymaking-as-bargaining Goals? When executive branch conflict does occur,
and some pointed out that Model III's ambiguities understanding its causes requires that we distinguish
raise questions about its status as a model, few critics between differences over goals and differences over
made an argument about Model III that we think is beliefs about how to achieve the goals. The relative
equally important, namely, that the model is simply absence of conflict in Art's four cases might suggest
too complex. There are so many different assump- that the key decision makers did share objectives.
tions, variables, and relationships in Model III that it However, even if actors' goals are identical, they may
is almost impossible to determine the role and impact disagree about how to achieve those ends, and these
of any one of them. For Model III to be useful for conflicting beliefs might foment arguments. While
systematic analysis, it must be simplified consider- Allison does mention the importance of differing
ably. beliefs, he does not adequately distinguish between
conflicts caused by differing beliefs and conflicts
caused by differing goals.
Does Policymaking in the Executive Branch What kind of policymaking should we expect
Proceed Via Bargaining? among actors with conflicting beliefs? Thompson and
Tuden (1959) suggested that when there is agreement
It is not hard to distill from the bureaucratic politics on beliefs about causation but disagreement on goals,
literature the claim that policymaking in the executive then bargaining will be observed, as in Model III. But
branch proceeds via bargaining. Yet this claim is not where there is agreement on goals but disagreement
a completely accurate reading of the literature. And on beliefs, they suggest that "collegial" judgment will
whatever the literature says, the claim turns out to be prevail. Collegial problem solving involves efforts to
vulnerable to several different kinds of criticisms. We ascertain the justification for differing beliefs and to
will examine these difficulties by posing a series of change the beliefs of other actors. If agreement on
questions about the properties of Model III. beliefs remains elusive, more overtly political activi-
ties might occur, though it seems likely that actors
Do Executive Branch Decision Makers Always Have Dif- would try these only as a last resort. Why yield
ferent Goals? Given our typology of models of policy- something by bargaining when one might persuade
making in Figure 1, the assumptions of Model III others of the empirical correctness of one's position?
would clearly seem to place it on row 3: there are When stakes are high, outcomes uncertain, and be-
many decision makers and they have conflicting liefs deeply held, debates over how to reach a com-
goals. However, row 2 of the typology raises the mon end may become rancorous; passionate dis-
possibility that these multiple decision makers may agreements need not indicate goal conflict. But such
sometimes have the same goals. Indeed, Krasner debates differ significantly from those caused by
suggests that considerable similarity in policymakers' conflicting objectives.
views should be expected: "The President chooses As an empirical question, Did President Kennedy
most of the important players and sets the rules. He and the various members of the ExCom have conflict-
selects the men who head the large bureaucracies. ing goals? Decisions were certainly made in a partisan
These individuals must share his values. Certainly context. Recall the criticisms by congressional Repub-
314
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
licans and the administration's nervousness about by Allison for Model II) note that such bargaining has
the upcoming congressional elections. But by Al- some "potentially disruptive consequences":
lison's own account this was more a matter between Bargainingalmost necessarilyplaces strainson the status
the president and his critics outside the administration and power systems in the organization.If those who are
than among ExCom members (1971, 187-200). Cer- formally more powerful prevail, this results in a more
tainly, administration officials argued over which forcefulperceptionof status and power differencesin the
action to take; but these disagreements could have organization (generally dysfunctional in our culture). If
stemmed mainly from honest differences in beliefs they do not prevail, their position is weakened. Further-
about the effects of various options. While Allison more, bargaining acknowledges and legitimizes hetero-
geneity of goals in the organization.Such a legitimation
criticizes Sorensen (1963) and Schlesinger (1965) for removes a possible technique of control availableto the
presenting "the efforts of the ExCom in the Cuban organizationalhierarchy. (1958, 131).
missile crisis as essentially rational deliberation
among a unified group of equals" (p. 146), it is not Thus, bargaining is something the president would
apparent from Allison's own account whether differ- seek to avoid, if at all possible. The question is
ences in personal and institutional goals (hence, whether he can avoid it.
"politics") or differences in beliefs (hence, "colle- Model III, of course, is based on the bureaucratic
gial," and possibly "rational," decision making) were politics literature, and this literature reveals ample
more important in the ExCom debate.24 support for an argument that the president often can
avoid having to bargain with subordinates. For exam-
ple, Huntington's Common Defense has a substantial
Why Would a President Bargain with Other Executive
section analyzing when policymaking is legislative in
BranchOfficials? One of Model III's great ambiguities
character and when it is executive (1961, 146-59). It is
concerns the nature of the president's relations with
interesting to note that Allison quotes a passage from
other officials in his administration. At one point, for this section (Essenceof Decision, 156). Indeed, he even
example, Allison approvingly comments on Neu- cautions that Huntington's analysis "overemphasizes
stadt's picture of the president's role in policymaking: participant equality as opposed to the hierarchy that
"Sometimes Neustadt's 'President-in-sneakers' is the structures the game" (p. 162). Hence, it is all the more
central political gamesman. Sometimes the President puzzling that Allison nowhere actually uses Hunting-
merely observes the pulling and hauling among var- ton's arguments. Instead, he just assumes that exec-
ious groups within the government. Sometimes the utive branch policymaking will always be legislative
President is the target of the tactics of a bureaucratic in nature; the possibility of executive policymaking is
group" (1971, 158). Later, Allison also states, "Where simply ignored.25
an outcome was for the most part the triumph of an Thus, if presidential power is a variable, as Hunt-
individual (e.g., the President) or group (e.g., the ington suggests, when are presidents powerful and
President's men or a cabal) this model attempts to when not? The literatures on bureaucratic politics and
specify the details of the game that made the victory organization theory reveal two basic reasons why a
possible" (p. 173). It would seem, then, that the president may sometimes be in a disadvantageous
extent of the president's involvement in policymak- position vis-a-vis members of his own administra-
ing-and his ultimate success-can vary from case to tion. The first is that subordinates may have sources
case. of political support outside the executive branch. The
Yet the central thrust of Model III is not this bland second turns on Weber's classic argument about
point that the president's influence varies across informational asymmetries between a superior and
issues. Instead, it is the much bolder hypothesis that his subordinates. Allison's presentation of Model III
in order to make policy the president must bargain does mention these two sources of power (1971,
with members of his own administration. As Allison 168-69); but his model does little to explore the role of
himself put it, "The primary source of the paradigm either factor in policymaking.
is the model implicit in Neustadt's work, though his
concentration on Presidential action has been gener- Do Subordinates Have Political Support outside the
alized to a concern with action as a resultant of ExecutiveBranch? While the president has much more
political bargaining among a number of independent formal authority than his appointees, his authority in
players, the President being only a 'superpower' the larger political system is far from absolute. Be-
among many lesser but considerable powers" (1971, cause Congress has an important constitutional role
162). This raises an obvious question: Since the pres- in policymaking, agencies in the executive branch
ident has personally appointed the top officials in his find it useful to cultivate support in Congress. While
administration and can dismiss them at any time and the president can often order his political appointees
since he has substantial formal authority (especially and their respective bureaucracies to do his bidding,
on foreign policy issues) to order them to do what he they may be able to hurt him politically if they
wants, why must the president bargain with them? disagree with his choices and make their disagree-
The literature on which Allison based Essence of ments known to outside supporters. Hence, the
Decision certainly does not lend unqualified support president may end up bargaining with subordinates
to the idea that superiors will bargain with their not because they are intrinsically powerful but be-
subordinates. For example, March and Simon (cited cause their outside supporters can make life difficult
315
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
for him. Thus even though the most obvious mani- dent and his top aides and the coherence of their
festations of bureaucratic politics may take place ideas emerged as key variables. Bureaucratic politics
inside the executive branch, what is driving the flourished largely when the president and aides paid
politics may be happening outside.26 little attention to an issue or lacked clear policy
Of course, when the most important actors in the preferences about it.
environment support the president, he need not Finally, even if informational asymmetries persist,
bargain with any subordinates who are allied with the subordinates' influence depends on the president
them. Moreover, even if influential decision makers, not realizing that he is ignorant. The subordinates'
whether inside or outside the executive branch, have influence is maximized when the president is un-
diverse views and even if some of them are hostile to aware of what constitutes good advice or that a
the president, the president may still be left relatively problem exists or that implementation should be
free to select who will be his allies. As Art suggests, monitored. Even here, however, we think their influ-
"When senior executive players are split on their ence would not make itself felt through bargaining. If
policy stances, the President, by virtue of the divi- the president does not realize that he is ignorant, the
sion, has considerable leeway to choose that which very fact that his subordinates try to bargain with him
he wishes to do, or that which he thinks he ought to should alert him that something odd is going on. To
do, or that which he reasons he must do" (1973, 475). influence the president, the subordinates would have
It is primarily when most of the environment is to use their informational advantage in more subtle
hostile and the president lacks countervailing sup- ways.
port that he must bargain with his nominal subordi-
nates. In sum, support outside the executive branch Was U.S. Policy in the Cuban Missile Crisis Determined
does not always mean influence inside it. by Bargaining? While our purpose is largely theoreti-
cal, it is of interest to ask whether Model III accurately
Do the President's Subordinates Have Informational describes U.S. policymaking during the missile crisis.
Advantages? Even a subordinate who lacks support It seems to us that the most obvious manifestations of
outside the executive branch may still enjoy an infor- politics-coalition formation, bargaining, logrolling,
mational advantage over the president. Nonetheless, splitting the difference, leaks to the press-were
it is not obvious that bargaining will be the empirical largely absent from the ExCom deliberations.28 In-
manifestation of informationally based influence. stead, our reading of Essence of Decision is that poli-
Three categories of informational asymmetries can be cymaking revolved around President Kennedy and
distinguished. For a president to exercise his author- that decision making lay largely in his hands. Al-
ity over the executive branch, he must become aware lison's own evidence reveals that the participants
of what problem needs to be addressed, decide what themselves believed that one person was in charge.
to do about it, and have his choice implemented. For Recall, for example, his observation that Dean
each of these three tasks, subordinates may have Acheson, emerging from an unsuccessful effort to
more information or expertise than the president.27 convince the president to approve an air strike, "left
This puts them in a position where they can manip- with no question in his mind about where the buck
ulate what problems he is aware of, what advice he stopped" (1971, 207). Allison even quotes Robert
receives, and how his final choice is implemented. Kennedy's observation, as the crisis reached its peak,
However, the standard works on bureaucratic pol- "It was now up to one single man. No committee was
itics in the era of Essenceof Decision (e.g., Downs 1967; going to make this decision" (p. 208).
Halperin 1974; Rourke 1969; Tullock 1965), as well as Moreover, a key tenet of Model III is that policy
the more recent literature on incentives (e.g., Bendor, outcomes are resultants-unintended by any one ac-
Taylor, and Van Gaalen 1987), suggest that there are tor-that emerge out of the dynamics of bargaining.
ways the president can alleviate these informational But if policy outcomes are unintended resultants and
problems. A wide-ranging communication network if Model III accurately describes the ExCom delibera-
can alert him to a problem; and if he then attends tions, one must then conclude that the choice of the
closely to the problem, he can learn a good deal about blockade was unintended by President Kennedy. But
it, thereby reducing his subordinates' informationally Allison provides no evidence that the president pre-
based influence. And once he knows what he is ferred some other option. On the contrary, his story
doing, he is not without weapons, as Krasner ob- suggests that the blockade was the option most
served: "The Chief Executive involves himself in favored by the president and his closest associates; it
those areas which he determines to be important. was not chosen by the president as part of a bargain
When the President does devote time and attention or political compromise with other ExCom mem-
to an issue, he can compel the bureaucracy to present bers.29
him with alternatives.... Even when Presidential
attention is totally absent, bureaus are sensitive to his
values. Policies which violate Presidential objectives
A Model of Bureaucratic Politics Needs a
may bring Presidential wrath" (1972, 168-69). Kohl
Model of Hierarchy
(1975) provides systematic support for this argument:
in an empirical test of several models of president- Model III was intended to synthesize the bureaucratic
subordinate relationships, the attention of the presi- politics studies of Hilsman, Huntington, Neustadt,
316
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
Schilling, and others. What is odd about the mod- can, via its impact on the sequence of decisions, have
el-a model of politics in a hierarchy-is that it says a similar impact on which policies are ultimately
almost nothing about how the hierarchy affects the adopted.30
politics. References are made to "players in posi- Allison apparently did judge that hierarchy was
tions," "action-channels," and "the rules of the relatively unimportant in the missile crisis. He ob-
game" (1971, 164-66, 169-71); but it is almost impos- served at one point, for example, that the ExCom
sible to infer from Allison's brief mention of these "functioned with minimal reliance on the standard
variables how they are expected to affect outcomes. channels down into the second or third levels of the
This neglect of the impact of hierarchy is surprising, government, causing no little pain to the players left
since Schilling (1962)-whom Allison discusses at out of the action" (1971, 215). Hence, it may have
length (pp. 154-56)-offered some useful insights seemed unnecessary to explore the impact of hierar-
about hierarchy's impact on policymaking. In his chy. But what is at issue, we would argue, is whether
study of the 1950 defense budget, Schilling argued the hierarchy's particularconfigurationinfluenced, for
that the shape of the military hierarchy affected who example, the CIA's information processing (which
handled which issues. Schilling described several Model II treats as an apolitical concern) or the partic-
post-World War II plans to reorganize the defense ular policy proposals advanced by the bureaucratic
establishment, noting claims that particular plans chiefs represented on the ExCom. Without consider-
would reduce the interservice rivalry over the mili- ably more attention to hierarchy as a variable, Allison
tary doctrines justifying particular allocations of mis- could not have determined whether the bureaucratic
sions and budgets. One proposal was to reorganize hierarchy actually had a negligible impact on policy-
the land, sea, and air forces into unified functional making.
commands: one was for conducting limited war, one
was for all-out war, and one was for continental
Logical Problems with Model III's Propositions
defense. Such reshuffling, Schilling argued, would
not end the key conflicts: Thus far, our argument has been that Model III's
conceptual development is inadequate. Some of the
Some of the issues which had formerlybeen subject to
inter-serviceargumentand determinationwould become model's premises are ambiguous, and it is difficult to
matters for intracommandargument and determination determine from Allison's discussion when bargaining
(for example, the Air Force-Navy dispute over the rela- will characterize policymaking and who will have to
tive merits of sea as comparedto land-launchedvehicles bargain with whom about what. These issues-cen-
for the delivery of a nuclear strikewould now take place tral to a theory of bureaucratic politics-are left un-
within the confines of the All-Out War Command). resolved by Model III. Model III's propositions,
Conversely,other issues formerlysubjectsof intraservice which define how it should be applied to data, are no
conflict would now become matters of intercommand more satisfactory.
dispute (for example, the argumentwithin the Air Force A proposition, in Allison's usage, is an empirically
over the allocation of resources between tactical and testable hypothesis that can shed light on the validity
strategic air forces would now be joined between the
All-Out War Command and the Limited War Com- of the underlying theory. But it is difficult to know
mand). (p. 228) what to conclude from a careful examination of some
of his propositions. For example, consider the prop-
This rearrangement of the issues could affect how osition, "Where you stand depends on where you
the conflicts were settled. As Schilling put it, "What sit" (1971, 176). Several critiques noted that this
reorganization would change, and it would be a proposition contradicts other aspects of Model III (Art
consequential change, would be the political condi- 1973, 472-73; Ball 1974, 77; Caldwell 1977, 94; Krasner
tions under which these issues would be argued and 1972, 165); Allison himself mentions many other
resolved" (1962, 228). For example, since an organi- influences, besides bureaucratic position, on the
zation's design thrusts some issues to the top while stances of participants. Moreover, empirical evalua-
pushing others down and since top decision makers tion of the proposition is difficult, since, as these
care more about some issues than others, different critics observed, some key participants do not "sit"
choices might be made in different structures. anywhere. And even where the proposition can be
Only recently have Schilling's ideas been used to empirically scrutinized and is found to lack support
develop a more explicit theory of hierarchy's impact (as when, in the missile crisis, the secretary of de-
on policymaking. The essential ideas in one set of fense initially took a relatively dovish stand despite
studies, for example, are that policymaking involves the hawkish views of many of his uniformed subor-
making comparisons (of pieces of information, of dinates), it is not clear what conclusions we should
policy options, or of proposals for implementation), draw about Model III more generally.
and that an organization's structure affects who com- In fact, some of the propositions seem to be only ad
pares what with what, so that different structures can hoc generalizations lacking clear derivation from the
produce different policy outcomes (Hammond 1986, underlying model.3" For example, it is not at all
n.d.; Hammond and Thomas 1989). It is well known apparent how the claim "In a nuclear crisis, the
that a legislature's agenda, which specifies the se- central decisions will be hammered out not in the
quence of voting on amendments, can greatly influ- formal forums, e.g., the National Security Council,
ence the outcome. An organization's formal structure but rather by an ad hoc group that includes the
317
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
318
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
ence to, and use of, Essence of Decision has a rather drawn from Model II-that simple rules lead to
troubling character. Aside from its important meth- simple, predictable behavior-is almost surely
odological points, what scholars and students have wrong; and Model III is so complicated that virtually
been learning from the models may lack foundation no propositions can be rigorously derived from it at
and, in some cases, may simply be incorrect. We all.
conclude by emphasizing five of these problems. Fifth, if propositions are not rigorously derived
First, modeling generally begins with some from a model yet receive some evidential support, it
hunches of the theorist-or with the claims of the is difficult to know what one should learn from the
academic literature-about how the world works. empirical corroboration. We may learn something
The model should then reflect these judgments. But about the propositions; but lacking any logical rela-
none of Allison's models are, in our view, sufficiently tionship between them and the model, the empirical
accurate renditions of the literatures that he himself test teaches us little about the model or the hunch
cites in justification for the models. The traditional that originally generated it.
literature on war, crisis, and international politics has Some of the inadequacies of Allison's models stem
emphasized uncertainty and state-versus-state gam- merely from the fact that there has been considerable
ing. Model I almost ignores these properties. The progress in the understanding of game theory and
organizational literature of Simon and March por- rational choice models, in the appreciation of the role
trayed standard operating procedures and routines as of uncertainty and incomplete information, and in
enabling choice and action. Model II depicts them as development of insights about organizations, hierar-
constraining choice and action. The understanding of chy, and bureaucratic politics. Indeed, we have been
politics-in-hierarchies, available in the bureaucratic at pains to point out what literature was available to
politics literature by the late 1960s, is not accurately Allison-and what was not-when he wrote his
reflected by Model III's analysis of bureaucratic bar- book. Yet many of the flaws we have pointed out in
gaining. The role of hierarchy, which presumably Essenceof Decision do not stem from advances in these
structures the entire game of bureaucratic politics, is fields: they were there from the beginning, and a
neglected-and the critical interplay between author- reasonable assessment of the book must recognize
ity and expertise in a bureaucracy, a major topic of this fact. Since the book still has considerable cur-
discussion in organization theory since Weber, re- rency in political science and in a wide range of other
ceives almost no attention at all. disciplines, continued use is thus likely to lead to the
Second, in laying out his three models of policy- widespread perpetuation of major misunderstand-
making, Allison has generally been credited with ings about the nature of bureaucracy and governmen-
advancing the important point that intellectual prog- tal policymaking. If the academic community is not
ress comes from formulating, testing, and evaluating made aware of these flaws, reliance on Allison's
alternative models. Yet his presentation of Model I
models is as likely to lead to error as it is to inform.
suggests that he saw it, at least in part, as something Nonetheless, Essence of Decision richly deserved its
to be set up in order to be knocked down. This may two decades of intellectual prominence. The fact that
explain Model I's inadequate development, even the study of bureaucracy, organization, and foreign
given the relatively crude state of the art of game
policy making has gone beyond it detracts in no way
theory and rational choice models when Allison was from its remarkable accomplishments.
writing. Nonetheless, one cannot seriously evaluate
the relative performance of two new models (II and
III) if the baseline model is almost destined to fail. For
this reason, what is often taken to be a major lesson
of Essence of Decision-that rational choice explana-
Notes
tions of policymaking are weaker than those of Mod-
We would like to thank Michael Barzelay, Michael Cohen,
els II or III-lacks foundation. Whatever their ulti- Scott Gates, David Jones, Roderick Kramer, James Morrow,
mate worth, rational choice models were not given a Scott Sagan, George Tsebelis, Harrison Wagner, Carol Weis-
fair test. sert, and James Q. Wilson for helpful comments on earlier
Third, Allison's overall approach-the develop- drafts of the paper.
ment of explicit models of policymaking-was based 1. Perlmutter (1974) distinguished three different models,
as did Snyder and Diesing (1977) and Steinbruner (1974).
on the view that clarifying one's key assumptions is Kurth (1971) analyzed five models, Kohl (1975) six, while
analytically helpful. Yet the assumptions of Models II Caldwell (1977) distinguished 12 models! Allison himself cited
and III were sufficiently ambiguous that it is difficult Steinbruner's contribution of additional models and noted
to discern the models' defining properties. Hence, it that "a number of others are clearly possible" (1971, 255; see
also pp. 276-77).
is unclear whether Models II and III belong to sepa- 2. E.g., in the 1990 Social Science Citation Index, Essence of
rate classes of theories or not. Decision and Allison's 1969 article were cited 84 times in 58
Fourth, one of the purposes of developing an different journals.
explicit model is rigorously to derive the logical 3. Though classical decision theory emphasized prospec-
implications of one's fundamental assumptions. Yet tive uncertainty, retrospective uncertainty is also prevalent in
crises: Did the enemy just launch a first strike, or are we
for neither of the two models (II and III) for which picking up a flock of geese on our radar screen? This example
Essence of Decision is best known are the propositions also points up the connection between the two types of
rigorously derived. The general proposition to be uncertainty: if one is unsure what one's opponent has done,
319
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
one will also probably be unsure how he will react to one's 19. Allison mentions this advantage once (1971, 80). He
own move. does not, however, emphasize it.
4. For just one example (which could have initiated a 20. "If the SOPs are appropriate, average performance-
nuclear exchange), recall the U-2 that blundered into Soviet i.e., performance averaged over the range of cases-is better
air space during the height of the Cuban missile crisis. As than it would be if each instance were approached individu-
President Kennedy was reported to have remarked about this ally (given fixed talent, timing, and resource constraints)"
incident, "There is always some son-of-a-bitch who doesn't (Allison 1971, 89). He immediately adds, however, "But
get the word" (quoted in Allison 1971, 141). specific instances, particularly critical instances that typically
5. Even if individual attention were unconstrained, costly do not have 'standard' characteristics, are- often handled
communication would mandate this issue. As Simon (1973) sluggishly or inappropriately" (ibid).
suggests, however, attention is often a major bottleneck. 21. See also: "An important objective of standardization is
6. Given this assignment, Model I should not be seen as the to widen as far as possible the range of situations that can be
only rational choice model in Essence of Decision. handled by combination and recombination of a relatively
7. There is a tendency to jump to such claims, particularly small number of elementary programs" (March and Simon
regarding unitary rational actor models; but we shall argue 1958, 150).
that these claims oversimplify the models' implicational rich- 22. In a striking essay, Norton Long also differed sharply
ness. with the viewpoint of Essence of Decision, chapter 4, arguing
8. Ball argues that Model III "is the model closest to that it is the bureaucracythat via its "fact-finding [and] policy
Allison's own heart," citing a seminar discussion at Harvard proposal . . . procedures," must discipline the tendency of
in October 1972, in which "Allison himself made this quite politicians to manipulate "a free-wheeling world of rhetoric
clear" (1974, 76). and emotion, . . . a world in which solid facts evaporate"
9. Cognitive analyses (e.g., Steinbruner 1974) recognize (1954, 28). It is interesting to note that though Allison cited
that decision makers may sometimes fail to resolve intraper- this essay in respect to Model III (1971, 315, n. 65), neither
sonal conflict in a consistent manner. Model II nor Model III was influenced by its main theme.
10. In redundancy theory, one often analyzes the optimal 23. Surprisingly, neither chapter 3 nor chapter 4 specifies
amount of duplicate men and materiel to achieve a task when an explicit baseline of evaluation: using routines and standard
implementation is uncertain. These analyses are facilitated by operating procedures constrains organizational behavior com-
assuming a single decision maker whose objectives are to be pared to what? To an organization that though still staffed by
maximized. boundedly rational actors, did not use standard operating
11. Indeed, Blight and Welch (1989) present ample evi- procedures? Or to how an organization with perfectly rational
dence that even latter-day Soviet analysts disagree about members would behave? Or to how a perfectly rational
Khrushchev's motives. individual would behave? Since the answer is unclear, the
12. Allison does challenge this notion, but only in discuss- reader is left to infer the baseline of comparison.
ing Model III in respect to bargaining inside governments. 24. We do not underestimate the difficulty of determining
13. Mixed strategies are covered in game theory texts going whether actors' disagreements are due primarily to conflicting
back to Luce and Raiffa 1957, chap. 4; see also Schelling 1960. goals or to conflicting beliefs. In a debate involving a presi-
Note that in the inspection game and similar two-player dent, insiders whose goals differ from the president's may
games, the mixed strategy equilibrium is replaced by a pure find it wise to pretend their disagreements are about means.
strategy equilibrium if the players move sequentially. The As March and Simon observe about organizations, "Bargain-
indeterminacy of a mixed strategy equilibrium does require ing (when it occurs) will frequently be concealed within an
some exogenously assumed uncertainty (i.e., the imperfect analytic framework" (1958, 131).
information occasioned by simultaneous play). 25. After criticizing what he considers Allison's overem-
14. Clearly, the optimal outcome for the United States is to phasis on bargaining in foreign policy formation, Rourke does
deceive the USSR, convincing the Soviets that we will inspect remark that "in domestic policy, on the other hand, models
closely, but not to do so. A similar point about deception structured around bargaining are much more useful, since
applies to the Soviets. Neither optimal outcome involves a executive organizations here compete with the President on a
mixed strategy. However, this does not invalidate the text's relatively equal footing" (1972, 432).
point, since neither outcome is an equilibrium. 26. While Allison cites Dahl and Lindblom in support of his
15. For example, Cyert and March describe the general point that Huntington overemphasized participant equality
procedure-"Use simple rules"-as one of the "three basic (1971, 315, n. 65), it is interesting to note that Dahl and
principles" of choice (1963, 102). And certainly, the best- Lindblom consistently maintain that outside political support
known rule in this literature, satisficing (searching for alterna- enables subordinates to bargain with the president (1953,
tives until finding one that exceeds a specified threshold) is a 341-44).
very simple procedure (Simon 1957, 204-5, 252-53). 27. In the missile crisis, for example, President Kennedy
16. March and Simon, who pioneered the metaphor of was apparently uncertain whether an air strike would destroy
organizational behavior as governed by computer programs, all the Soviet missiles in Cuba. When he sought advice from
recognized that behavioral flexibility could arise from a pro- the Tactical Air Command, its commander told him that no
gram's sensitivity to inputs: "The term 'program' is not such guarantee was possible. This judgment apparently
intended to connote complete rigidity. The content of the played a major role in Kennedy's ruling out the air strike
program may be adaptive to a large number of characteristics option.
of the stimulus that initiates it. Even in the simple case of the 28. To be sure, there were leaks to the press from inside the
fire gong, the responses depend on the location of the alarm" executive branch before the president and his top advisors
(1958, 142). Again, "At the [programmed] limit, an environ- became convinced that Soviet missiles were indeed in Cuba
mental stimulus may evoke immediately from the organiza- (Allison 1971, 192). Once the ExCom was convened, however,
tion a highly complex and organized set of responses" (p. 141, Allison mentions nothing further about leaks intended to
emphasis added). influence policymaking.
17. What follows is therefore a conservative analysis that 29. Our conclusions here are in line with those of most of
underestimates the complexity of the armed services' rules for Allison's critics; see, esp., Art 1973; Ball 1974; and Krasner
research-and-development and procurement. 1972.
18. This view of Allison's appears to have influenced other 30. We are not suggesting that politics occurs only within
international relations scholars. For example, Keohane wrote formal channels; evidence on the importance of informal
that "If individuals typically satisfice rather than maximize, all organization is too strong to ignore. We do think, however,
the more so do governments and other large organizations that the informal relationships that do emerge will be efforts
(Allison, 1971; Steinbruner, 1974; Snyder and Diesing, 1977)" to overcome the impact of the formal structure. Thus, formal
(1984, 114). structure shapes informal structure.
320
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
American Political Science Review Vol. 86, No. 2
31. Allison himself bemoans "the difficulty of formulating Revised Expected Utility Model." American Political Science
Model III propositions about outcomes" (1971, 173). Review 79:156-77.
32. We are reminded of Lewis Carroll's (1893) parable Caldwell, Dan. 1977. "Bureaucratic Foreign Policy-Making."
about map-making: "That's another thing we've learned from American BehavioralScientist 21:87-110.
your Nation," said Mein Herr, "map-making. But we've Carroll, Lewis. 1893. Sylvie and Bruno Concluded. London:
carried it much further than you. What do you consider the Macmillan.
largest map that would be really useful?" "About six inches to Cornford, J. P. 1974. Review of Essenceof Decision by Graham
the mile." "Only six inches!" exclaimed Mein Herr. "We very Allison. British Journal of Political Science 4:231-43.
soon got to six yards to the mile. Then we tried a hundred Cyert, Richard, and James March. 1963. A BehavioralTheoryof
yards to the mile. And then came the grandest idea of all! We the Firm. Englewood Cliffs: Prentice-Hall.
actually made a map of the country, on the scale of a mile to Dahl, Robert A., and Charles E. Lindblom. 1953. Politics,
the mile!" "Have you used it much?" I enquired. "It has never Economics,and Welfare.New York: Harper & Row.
been spread out, yet," said Mein Herr: "The farmers objected: Downs, Anthony. 1967. Inside Bureaucracy. Boston: Little,
they said it would cover the whole country, and shut out the Brown.
sunlight! So we now use the country itself, as its own map, Freedman, Lawrence. 1976. "Logic, Politics, and Foreign
and I assure you it does nearly as well." Policy Processes: A Critique of the Bureaucratic Politics
33. For a more detailed discussion of these trade-offs and of Model." InternationalAffairs 52:434-49.
the great heuristic value of conceptual frameworks (partially Gallucci, Robert L. 1975. Neither Peace nor Honor: The Politics of
formalized models), see Bendor and Hammond 1989. AmericanMilitary Policy in Viet-Nam. Baltimore: Johns Hop-
34. For evidence that Allison took the historian's job of kins University Press.
explaining the missile crisis seriously, see 1971, 1-2. Here he Garthoff, Raymond. 1988. "Cuban Missile Crisis: The Soviet
asks four central questions. The first three-Why did the Story." Foreign Policy 72:61-80.
Soviet Union place strategic offensive missiles in Cuba? Why Garthoff, Raymond L. 1989. Reflections on the Cuban Missile
did the United States respond with a naval quarantine? Why Crisis. Rev. ed. Washington: Brookings Institution.
were the missiles withdrawn?-concern particular aspects of
George, Alexander. 1980. PresidentialDecisionmakingin Foreign
this particular event. As such, they are indistinguishable from
Policy. Boulder: Westview Press.
those posed by historians.
Grofman, Bernard, Guillermo Owen, and Scott Feld. 1983.
"Thirteen Theorems in Search of the Truth." Theory and
Decision 15:261-78.
Gulick, Luther. 1937. "Notes on the Theory of Organization."
References In Papers on the Science of Administration, ed. Luther Gulick
and Lyndall Urwick. New York: Columbia University.
Halperin, Morton H. 1974. BureaucraticPolitics and Foreign
Allison, Graham. 1969. "Conceptual Models and the Cuban Policy. Washington: Brookings Institution.
Missile Crisis." American Political ScienceReview 63:689-718. Hammond, Thomas H. 1986. "Agenda Control, Organiza-
Allison, Graham. 1971. Essenceof Decision: Explaining the Cuban tional Structure, and Bureaucratic Politics." American Jour-
Missile Crisis. Boston: Little, Brown. nal of Political Science 30:380-420.
Allison, Graham, and Morton Halperin. 1972. "Bureaucratic Hammond, Thomas H. N.d. "Toward a General Theory of
Politics: A Paradigm and Some Policy Implications." In Hierarchy: Books, Bureaucrats, Basketball Tournaments,
Theory and Policy in International Relations, ed. Raymond and the Administrative Structure of the Nation-State."
Tanter and Richard Ullman. Princeton: Princeton Univer- Journal of Public Administration Research and Theory. Forth-
sity Press. coming.
Allyn, Bruce J., James G. Blight, and David A. Welch. 1989/90. Hammond, Thomas H., and Paul A. Thomas. 1989. "The
"Essence of Revision: Moscow, Havana, and the Cuban Impossibility of a Neutral Hierarchy." Journal of Law, Eco-
Missile Crisis." International Security 14:136-172. nomics, and Organization 5:155-84.
Art, Robert. 1973. "Bureaucratic Politics and American For- Hampson, Fen. 1984/85. "The Divided Decision-Maker:
eign Policy: A Critique." Policy Sciences 4:467-90. American Domestic Politics and the Cuban Missile Crisis."
Ball, Desmond. 1974. "The Blind Men and the Elephant: A InternationalSecurity 9:130-65.
Critique of Bureaucratic Politics Theory." Australian Outlook Hogarth, Robin. 1987. Judgmentand Choice. 2d ed. New York:
28:71-92. Wiley.
Bendor, Jonathan. 1985. Parallel Systems: Redundancy in Gov- Holsti, Ole. 1972. Review of Essence of Decision by Graham
ernment. Berkeley: University of California Press. Allison. Western Political Quarterly 25:136-40.
Bendor, Jonathan. 1987. "In Good Times and Bad: Reciprocity Huntington, Samuel. 1961. The Common Defense. New York:
in an Uncertain World." American Journal of Political Science Columbia University Press.
31:531-58. Jefferies, Chris. 1977. "Defense Decisionmaking in the Orga-
Bendor, Jonathan, and Thomas H. Hammond. 1989. "Re- nizational-Bureaucratic Context." In American Defense Pol-
thinking Allison's Models." Presented at the annual meet- icy, 4th ed., ed. John Endicott and Roy Stafford, Jr. Balti-
ing of the American Political Science Association, Atlanta. more: Johns Hopkins University Press.
Bendor, Jonathan, Serge Taylor, and Roland Van Gaalen. Jervis, Robert. 1976. Perception and Misperception in Interna-
1987. "Stacking the Deck: Bureaucratic Missions and Policy tional Politics. Princeton: Princeton University Press.
Design." American Political Science Review 81:873-96. Keohane, Robert. 1984. After Hegemony. Princeton: Princeton
Blight, James, Joseph Nye, Jr., and David Welch. 1987. "The University Press.
Cuban Missile Crisis Revisited." Foreign Affairs 66:170-88. Kohl, Wilfrid. 1975. "The Nixon-Kissinger Foreign Policy
Blight, James, and David Welch. 1989. On the Brink:Americans System and U.S.-European Relations." WorldPolitics28:1-43.
and Soviets Reexaminethe CubanMissile Crisis. New York: Hill Krasner, Steven. 1972. "Are Bureaucracies Important? (Or
& Wang. Allison Wonderland)." Foreign Policy 7:159-79.
Bobrow, Davis. 1972. InternationalRelations. New York: For- Kurth, James. 1971. "The Widening Gyre: The Logic of
eign Policy Association. American Weapons Procurement." Public Policy 19:373-404.
Bouchard, Joseph F. 1991. Commandin Crisis: Four Case Studies. Landau, Martin. 1969. "Redundancy, Rationality, and the
New York: Columbia University Press. Problem of Duplication and Overlap." Public Administration
Brugioni, Dino A. 1991. Eyeballto Eyeball:The Inside Story of the Review 29:346-58.
Cuban Missile Crisis. New York: Random House. Landau, Martin. 1972. Political Scienceand Political Theory.New
Bueno de Mesquita, Bruce. 1981. The War Trap. New Haven: York: Macmillan.
Yale University Press. Levy, Jack. 1986. "Organizational Routines and the Causes of
Bueno de Mesquita, Bruce. 1985. "The War Trap Revisited: A War."InternationalStudies Quarterly 30:193-222.
321
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
All use subject to JSTOR Terms and Conditions
Rethinking Allison's Models June 1992
Long, Norton. 1954. "Public Policy and Administration: The Simon, Herbert A. 1973. "Applying Information Technology
Goals of Rationality and Responsibility." Public Administra- to Organization Design." Public Administration Review 33:
tion Review 14:22-31. 268-78.
Luce, R. Duncan, and Howard Raiffa. 1957. Games and Deci- Simon, Herbert A., Donald Smithburg, and Victor Thomp-
sions. New York: Wiley. son. 1950. Public Administration. New York: Knopf.
Lukas, Anthony. 1987. "Class Reunion: Kennedy's Men Re- Snyder, Glenn H., and Paul Diesing. 1977. Conflict among
live the Cuban Missile Crisis." New YorkTimes Magazine, 30 Nations. Princeton: Princeton University Press.
August. Sorensen, Theodore C. 1965. Kennedy. New York: Harper &
March, James G., and Zur Shapira. 1982. "Behavioral Deci- Row.
sion Theory and Organizational Decision Theory." In Deci- Steinbruner, John. 1974. The Cybernetic Theory of Decision.
sion Making: An InterdisciplinaryInquiry, ed. Gerardo Ung-
Princeton: Princeton University Press.
son and Daniel Braunstein. Belmont, CA: Wadsworth.
Taylor, Frederick W. [1911] 1947. The Principles of Scientific
March, James G., and Herbert A. Simon. 1958. Organizations.
New York: Wiley. Management. Reprint. New York: Harper & Row.
Morrow, James. 1989. "Capabilities, Uncertainty, and Re- Taylor, Serge. 1984. Making BureaucraciesThink: The Environ-
solve: A Limited Information Model of Crisis Bargaining." mental Impact Statement Strategy of Administrative Reform.
AmericanJournal of Political Science 33:941-72. Stanford: Stanford University Press.
Nisbett, Richard and Lee Ross. 1980. Human Inference:Strate- Thompson, James C. 1980. Rolling Thunder: Understanding
gies and Shortcomings of Social Judgment. Englewood Cliffs, Policy and ProgramFailure. Chapel Hill: University of North
NJ: Prentice-Hall. Carolina Press.
Perlmutter, Amos. 1974. "The Presidential Political Center Thompson, James D., and Arthur Tuden. 1959. "Strategies,
and Foreign Policy: A Critique of the Revisionist and Structures, and Processes of Organizational Decision." In
Bureaucratic-Political Orientations." WorldPolitics 27:87-106. ComparativeStudies in Administration, ed. James D. Thomp-
Posen, Barry. 1984. The Sources of Military Doctrine. Ithaca: son et al. Pittsburgh: University of Pittsburgh Press.
Cornell University Press. Trachtenberg, Marc. 1985. "The Influence of Nuclear Weap-
Peterson, Paul E. 1976. School Politics, Chicago Style. Chicago: ons in the Cuban Missile Crisis." International Security
University of Chicago Press. 10:137-203.
Powell, Robert. 1990. Nuclear Deterrence Theory. Cambridge: Tsebelis, George. 1989. "The Abuse of Probability in Political
Cambridge University Press. Analysis: The Robinson Crusoe Fallacy." American Political
Rourke, Francis. 1969. Bureaucracy,Politics, and Public Policy. ScienceReview 83:77-91.
Boston: Little, Brown. Tullock, Gordon. 1965. The Politics of Bureaucracy.Washing-
Rourke, Francis. 1972. Review of Essence of Decision by Gra-
ton: Public Affairs Press.
ham Allison. Administrative Science Quarterly 71:431-33.
Wagner, R. Harrison. 1974. "Dissolving the State: Three
Sapolsky, Harvey. 1972. The Polaris System Development: Bu-
Recent Perspectives on International Relations." Interna-
reaucratic and Programmatic Success in Government. Cam-
tional Organization 28:435-66.
bridge: Harvard University Press.
Schelling, Thomas. 1960. The Strategy of Conflict. Cambridge: Wagner, R. Harrison. 1989. "Uncertainty, Rational Learning,
Harvard University Press. and Bargaining in the Cuban Missile Crisis." In Models of
Schilling, Warner. 1962. "The Politics of National Defense: StrategicChoicein Politics, ed. Peter Ordeshook. Ann Arbor:
Fiscal 1950." In Strategy, Politics, and Defense Budgets, ed. University of Michigan Press.
Warner Schilling, Paul Hammond, and Glenn Snyder. New Welch, David A., and James G. Blight. 1987/88. "The Eleventh
York: Columbia University Press. Hour of the Cuban Missile Crisis: An Introduction to the
Schlesinger, Arthur M., Jr. 1965. A Thousand Days: John F. ExComm Transcripts." InternationalSecurity 12:5-92.
Kennedy in the White House. Boston: Houghton-Mifflin. Williamson, Samuel. 1979. "Theories of Organizational Pro-
Scott, W. Richard. 1981. Organizations:Rational, Natural, and cess and Foreign Policy Outcomes." In Diplomacy: New
Open Systems. 2d ed. Englewood Cliffs: Prentice-Hall. Approachesin History, Theory, and Policy, ed. Paul Lauren.
Simon, Herbert A. 1947. Administrative Behavior. New York: New York: Free Press.
Macmillan. Yanarella, Ernest. 1975. "'Reconstructed Logic' and 'Logic-
Simon, Herbert A. 1957. Models of Man: Social and Rational. in-Use' in Decision-Making Analysis: Graham Allison."
New York: Wiley. Polity 8:156-72.
322
This content downloaded from 35.8.11.2 on Tue, 01 Sep 2015 14:18:59 UTC
View publication stats
All use subject to JSTOR Terms and Conditions